Air Pollution Exposure and COVID-19: Evidence and Unanswered Questions

videoType

dynamic-media

videoDmUrl

keyFrameImage

timestamp

00:00 Right on the

00:01 dot at 11 o'clock.

00:02 Here we are.

00:03 All right.

00:04 Um,

00:04 good morning,

00:06 everyone.

00:06 Good morning,

00:06 ladies and gentlemen.

00:07 Good evening,

00:08 ladies and gentlemen,

00:09 wherever you are,

00:10 and good afternoon,

00:11 wherever you might be.

00:13 So,

00:13 welcome to everyone.

00:15 Um,

00:15 today,

00:16 we're going to deal with a really important topic in this series on COVID,

00:20 and it's on the relationship and the link between COVID-19

00:24 and air pollution.

00:26 If there's one thing that we know about COVID,

00:29 it's that we don't know very much about it.

00:32 And one of the great unknowns about COVID too,

00:34 are the links that COVID might have with air pollution.

00:38 So,

00:38 we hope for some clarity and insight from the 3 speakers that we have today

00:43 in our 2 discussions,

00:45 so that I don't interrupt the flow of what we're going to hear today.

00:48 Let me begin by introducing our 3 speakers in the order in which they will appear.

00:53 And our discussions.

00:55 And after,

00:56 after our speakers and discussions have gone,

00:58 we will open the floor for questions.

01:00 You can either,

01:01 you can either

01:02 ask your questions verbally

01:03 or through the chat line

01:05 and I will forward them to,

01:07 uh,

01:07 to,

01:07 to our speakers and our discussions.

01:10 So we have 3 extremely well qualified speakers to address this issue.

01:15 Our first speaker of the day is going to be David Wheeler.

01:18 David

01:19 is very well known to the World Bank and he has a very long CV too.

01:23 Currently,

01:24 David is a senior fellow at WRI which is the World Resources Institute.

01:30 Prior to that,

01:30 he was a lead economist in the World Bank,

01:33 worked in the CGD,

01:35 and prior to that,

01:36 he was a professor at Boston University.

01:40 And one of the marks I've known David for a very long time,

01:42 one of the marks of David's work.

01:44 is that he's always ahead of the curve,

01:46 especially when it comes to data.

01:48 Long ago,

01:49 it was on pollution,

01:50 then it was on forests,

01:52 and now David is on to the whole COVID issue.

01:55 And it'll be wonderful to hear from,

01:57 from,

01:58 from David.

01:59 Then we'll move to Bo Peter Andre,

02:02 who's currently a consultant in the bank who's doing work on crisis analytics.

02:07 Bo has had stints not only at the World Bank,

02:09 but also at the OECD,

02:11 at the EC,

02:12 ADB.

02:14 And at the Dutch government.

02:15 And he has a book,

02:17 The Theory and Applications of Dynamic Spatial Analysis.

02:20 And I downloaded that book last night to read it and I can assure you,

02:24 it is not bedtime reading.

02:26 You really,

02:26 really have to focus

02:27 because you want to understand about the dynamics of spatial analysis.

02:32 Um,

02:32 finally,

02:33 last but not least,

02:34 we'll have Anna Hansel,

02:36 who's professor of environmental epidemiology at the University of Leicester.

02:40 Prior to that,

02:41 she held positions at Imperial College

02:44 and was associate director of the Health Statistics Unit in the UK.

02:48 And then each of our speakers will speak for

02:51 about 14 minutes at about a 13-minute mark.

02:54 I will rudely interrupt you

02:56 and ask you to wind up.

02:57 And then we will give 7 minutes to each of our discussant.

03:01 Our first discussant is going to be ho Mal,

03:03 who's a lead

03:04 columnist in the Urban Development unit

03:06 and a global lead.

03:08 For anyone that knows anything about urban economics,

03:10 you will know and have read Schmick's work.

03:12 He's extremely well published

03:15 and has led a lot of our thinking in the bank

03:17 on the economics of urbanization,

03:19 agglomeration,

03:20 and development.

03:22 And,

03:22 and then we will finally have comments from Urvashi Narain,

03:26 who's who's also a lead economist in the Environmental

03:28 and Natural Resources unit at the World Bank,

03:31 and she's also extremely well known.

03:34 Especially on air pollution where she's doing incredible things

03:38 in countries with deep air pollution problems and is our go to person

03:42 in the World Bank

03:43 on problems of air pollution.

03:46 Wherever,

03:46 she too is widely published in journals like Jean

03:49 and elsewhere.

03:50 So,

03:51 Without further ado,

03:53 I would like to hand the virtual floor over to our first speaker,

03:57 who's David Wheeler.

03:58 So,

03:59 David,

03:59 it's all yours

04:00 and you have 14 minutes and I will interrupt you at the 13-minute mark.

04:05 Thank you so much.

04:06 Thank you,

04:07 Richard,

04:07 and I'm now going to go to share mode here.

04:09 And bring up my slides.

04:14 Uh,

04:15 thanks very much for your kind introduction and welcome to everyone.

04:18 Uh,

04:19 I'm here,

04:19 uh,

04:20 really as a reporter on

04:22 joint work which has recently been undertaken

04:24 by the Development Research Group and the Bank's

04:26 Development Economics vice presidency and the Global Practice Group

04:30 for Social,

04:30 Urban,

04:31 Rural and Resilience.

04:32 And I'm here to provide some introductory context

04:36 by talking about

04:38 An enterprise that we began very recently,

04:40 uh,

04:41 to look at the question of

04:42 whether one might

04:44 be able to model the spread of COVID-19 in a tractable way

04:48 and in a way that would be useful for developing countries.

04:51 And the focus this morning,

04:52 as you can see,

04:53 will be on the US and Philippines,

04:55 uh,

04:55 with this brief,

04:56 uh,

04:56 visual timeline of the spread of the epidemic in the two places,

05:00 but I'll return to that,

05:01 uh,

05:01 momentarily.

05:03 We're trying to respond to a challenge here which is very unusual.

05:06 As we all know,

05:07 COVID-19 is spreading very rapidly.

05:09 Uh,

05:10 countries

05:11 need to anticipate

05:12 how the spread will occur as best they can.

05:15 But at the same time,

05:16 for many countries,

05:17 uh,

05:17 particularly low-income countries,

05:18 time is scarce,

05:20 resources are limited,

05:21 and there,

05:22 there really is a need for some methodologies that will help.

05:25 Uh,

05:25 to identify

05:27 places that may be subject to the spread of COVID that are not currently afflicted.

05:31 Our project then was,

05:32 was a major challenge for us because it has been formed

05:35 and implemented very,

05:36 very rapidly.

05:37 We only began in March.

05:39 And our task has been to develop a model that is conceptually sound,

05:42 but at the same time,

05:44 simply designed,

05:45 uh,

05:45 easy to explain.

05:47 Easy to transport across regions and income levels,

05:49 easy to estimate and robust nevertheless,

05:52 to estimation problems,

05:54 and there are 3 case studies that we're undertaking for the US,

05:57 the Philippines,

05:58 and,

05:58 uh,

05:59 we are now moving into South Africa as well.

06:02 Uh,

06:02 I thought that,

06:03 uh,

06:04 to begin with,

06:04 I would actually prevent you,

06:06 present you a visual timeline of the evolution of COVID quickly,

06:10 uh,

06:10 in the US at continental scale,

06:12 in the US state of Georgia.

06:14 And in Luzon,

06:15 Philippines.

06:15 And

06:16 the reason I'm going to start with this is to give you a sense of the,

06:19 the central problem that we face in modeling,

06:22 which is handling what I'll call multi-polar spread.

06:25 You see on the screen then,

06:27 a picture of the United States

06:29 at the county level,

06:30 the small outlines there are US counties

06:32 in early March before this problem really emerged

06:35 uh to any significant degree.

06:37 And as these slides unfold,

06:38 you're going to see color scaling going from blue

06:41 to green,

06:41 yellow,

06:42 brown into red,

06:43 and finally lavender

06:44 to give you a sense of the severity of

06:47 Uh,

06:48 the infection rate in each county,

06:50 so that what you are going to see now,

06:52 uh,

06:52 is standardized by population to make viewing easy.

06:57 As you see these slides for each case,

07:00 uh,

07:00 it's good to look for three things.

07:01 The first is the appearance of

07:04 poles for the outbreak,

07:05 places where things begin,

07:06 then the extension outward from those poles,

07:08 and finally,

07:09 the joining together

07:11 of geographic extensions into larger

07:13 patterns.

07:14 So here we are on March 3rd.

07:16 We see the beginning of the emergence of some spot on March 17th.

07:19 March 24th shows us clear emergence of some poles,

07:23 which then extend outward by the 31st.

07:26 April 7th,

07:26 you can see some

07:28 severely impacted areas forming.

07:30 Here is April 21st.

07:31 The process continues through May 12th,

07:33 and finally,

07:35 June 2nd,

07:35 and here,

07:36 after the process of emergence and extension,

07:38 you see a joining up

07:40 where huge regions of the United States

07:42 have now been affected.

07:44 Just to reinforce these ideas,

07:46 let me now go to a

07:48 uh regional case.

07:49 This is the US state of Georgia.

07:51 Uh,

07:52 and let me begin,

07:53 um,

07:54 on March 15th,

07:55 which is about two weeks after,

07:57 uh,

07:57 there was a funeral

07:59 in Albany,

07:59 Georgia,

08:00 which is a circled city in the southwest of the diagram.

08:03 Uh,

08:03 lots of people attended,

08:04 they came from various areas.

08:06 On the map,

08:06 you see three circled areas.

08:08 In the northwest is Atlanta.

08:10 Uh,

08:10 in the southwest is Albany,

08:12 and in the middle is Macon.

08:13 Atlanta is,

08:14 of course,

08:14 a huge transportation hub and very densely populated.

08:17 Now,

08:18 watch the evolution and spread of COVID as we go forward

08:21 into March.

08:22 You see what happened after the funeral in the Southwest,

08:24 you see the spread from Atlanta.

08:26 And now you see

08:27 uh the virus beginning to spread out.

08:29 By April 19th becoming more severe.

08:32 Here's May 3rd when join-ups begins between the Atlanta area and the Albany area.

08:37 And by June 1st,

08:38 most of the map has been overwhelmed,

08:40 including the area

08:41 around Macon.

08:43 Now,

08:43 let's move to Luzon,

08:44 Philippines.

08:45 Again,

08:45 we're gonna start on March,

08:47 in early March.

08:48 Here,

08:48 there was hardly any incidents then.

08:50 You see 3 circles again,

08:52 one near Manila,

08:53 2 in other regions which will turn out to be affected quickly.

08:57 We see the beginning spread from Manila.

08:58 We see it moving outwards from that area.

09:01 Then we see a seeding occurring in other areas.

09:04 Two poles in the circle areas then become

09:06 much more severe and there's continued spread,

09:09 uh,

09:09 which then

09:10 Uh,

09:11 simply evolves and progresses.

09:13 So,

09:14 this is a pattern that we see everywhere,

09:15 and I'll refer to it as fractal,

09:17 meaning there's a sort of an eerie replication of this pattern,

09:20 uh,

09:20 at continental scale and at regional scale.

09:23 In all these cases,

09:24 we see

09:25 similar patterns,

09:26 an emergence in several places,

09:29 a non-uniform pattern of spatial propagation outward.

09:32 A link up ultimately the fastest propagating salience in those patterns,

09:36 and finally,

09:37 a spread into the areas that were initially surrounded by fast moving salient.

09:42 This has been the challenge for us.

09:43 How can we model this process in a way that allows us to predict

09:47 what may happen next?

09:48 And

09:49 in this presentation,

09:50 I'm going to avoid almost all technical detail.

09:53 There's a huge amount of it,

09:54 and I'll be happy to go back

09:56 and answer any questions that you may have uh after

09:59 my presentation in the Q&A session.

10:01 But let me just say that there are 3 elements here in trying to model this.

10:05 The first,

10:06 as you can see the spread pattern involves interaction with neighboring areas.

10:09 That's critical.

10:10 This thing moves from one area to neighboring

10:13 areas quite inexorably.

10:14 And in order to model that,

10:16 uh,

10:17 interaction between two areas,

10:18 we've used a variant of the well-known gravity modeling approach,

10:22 familiar from trade theory,

10:23 uh,

10:24 from,

10:24 uh,

10:25 other branches of economics and other disciplines,

10:28 including epidemiology.

10:30 Uh,

10:30 in,

10:31 in many applications.

10:32 And the basic principle is simple.

10:34 Uh,

10:34 in this context,

10:35 interactions between two areas of people who may be infected

10:39 are simply proportional to the product of

10:41 the populations of those areas that are infected

10:43 and,

10:44 uh,

10:44 in,

10:45 inversely related to the distance between the two populations.

10:49 But in this case,

10:50 we're going to innovate a bit,

10:50 not by using distance,

10:52 but by using the travel time,

10:53 which is a much better

10:54 uh measure of proximity.

10:57 So,

10:57 we're going to then include an element

10:59 which involves interactions with neighboring areas.

11:02 Another element which involves interactions within each area,

11:05 and for that,

11:05 we're gonna use a local measure of population density

11:08 as our best proxy.

11:10 And we have to put all this together into a growth model,

11:12 a model that has growth dynamics.

11:14 We're going to use the Gompert's model for that.

11:16 It's a figure model from technology diffusion,

11:19 uh,

11:19 from epidemiology,

11:20 and for population studies.

11:22 I'd like to touch very briefly on two aspects of this that are relevant.

11:26 On the interactions with neighboring areas,

11:28 uh,

11:28 there was a challenge in trying to quantify travel time in this context.

11:32 And without dwelling on this too long,

11:33 I'd like to advertise a bit for a wonderful new resource

11:36 that we use that's available for everyone in the world now for free,

11:40 called the open-source routing machine,

11:42 which allows you to use OpenStreetMaps

11:44 to compute travel times between any two points

11:47 for free

11:48 on a massive scale.

11:50 For this particular application,

11:51 we had to compute

11:52 hundreds of thousands of travel times.

11:55 Uh,

11:55 fortunately,

11:56 for replication of this approach,

11:57 you only have to do this once and we did it all for free.

12:01 The basic idea is for each

12:03 area.

12:04 Uh,

12:04 you take all areas with which it might feasibly interact,

12:07 and we set a radius of 200 miles

12:10 around the area.

12:11 And then you calculate travel times to the centroids of,

12:14 in this case,

12:15 other areas,

12:16 counties in the US,

12:17 uh,

12:18 and,

12:18 and that gives you the domain within which you do measurement.

12:21 Down below,

12:22 you can see,

12:23 uh,

12:24 the travel time radii that are color coded

12:27 from 5 US counties and 2 areas in the Philippines,

12:30 and of course,

12:31 depending on road conditions,

12:32 travel times can vary a lot.

12:34 Uh,

12:34 once you have this,

12:35 then

12:36 you we the infections in each neighboring unit by inverse travel time and add

12:40 all those up for each unit or each county in the US case,

12:43 for example,

12:44 to get a total measure of interaction.

12:46 So,

12:46 that's one of the terms in

12:48 uh the modeling approach.

12:50 On the growth modeling,

12:51 this is my only brush past mathematics,

12:53 and I'll be very quick.

12:55 Just one point I wanted to make.

12:56 The Gompert's function is quite useful in this context.

12:59 It provides a pretty good fit to actual data on the spread of infections.

13:03 You can see on the right there,

13:04 uh,

13:04 an exemplary picture of South Korea.

13:06 And the reason I wanted to bring up the math is that To

13:10 highlight that red alpha,

13:12 which is there in the middle,

13:13 which we will see again in the econometrics.

13:15 The basic idea behind the Gompert's curve is quite simple.

13:17 It says that

13:18 the rate of change of something is proportional to the

13:21 gap between its ultimate destination and the current state.

13:24 So,

13:25 the bigger that gap,

13:25 the faster the growth will be.

13:28 Having said all this by way of introduction,

13:29 then,

13:30 uh,

13:30 I'm gonna pass immediately to some econometric results which we got,

13:35 uh,

13:35 which suggests that in fact,

13:37 this very simple model,

13:38 two modes of interaction and a,

13:40 and a growth,

13:41 a modeling context give you fairly powerful results.

13:44 So,

13:45 this data,

13:46 It involves both the US and the Philippines.

13:48 We're looking at changes from March 17th to April 14th.

13:52 We have the three terms I've already discussed as variables in this model.

13:56 The first is the log of the initial infection rate,

13:58 and the,

13:59 this is basically minus the value of alpha from the mass or Doppert's curve.

14:04 We have the travel time weighted.

14:07 Measure that I had,

14:08 uh,

14:08 discussed.

14:08 And finally,

14:09 we have local population density.

14:11 In the case of the US and Philippines,

14:13 you could see

14:13 by the standard canons of statistics,

14:16 these are really quite powerful results.

14:17 They're very robust.

14:19 Uh,

14:19 in each case,

14:20 the signs are right.

14:21 Uh,

14:21 in the case of maximum population density locally,

14:24 uh,

14:25 the two coefficients are even very close to one another.

14:27 For the others,

14:28 they differ,

14:29 uh,

14:29 and that's something one could discuss.

14:31 But in any case,

14:32 the model seems robust.

14:34 I should point out one.

14:37 Uh,

14:37 side issue here,

14:38 as part of fitting this model,

14:40 we had to use a grid search algorithm

14:42 to fit the parameter for travel time.

14:45 Uh,

14:46 in this relationship by which if you have an infection in an in,

14:49 in,

14:50 in a,

14:50 an interacting area,

14:52 you wait by inverse travel time.

14:53 That isn't necessarily travel time,

14:56 uh,

14:56 to the first power,

14:57 just dividing by T as you can see up above.

14:59 It might be a different power.

15:00 The only reason I'm bringing this up is that it makes a big difference.

15:04 We got an exponent of 0.7 and the green line,

15:07 uh.

15:07 In this graph shows you that in fact,

15:10 that gives you a much broader reach for this infection

15:12 than you would get if you were dividing by travel time alone,

15:15 that's the blue line

15:16 or the square of travel time.

15:18 Uh,

15:18 that's the,

15:19 the reddish brown line.

15:20 So you can see that an implication of our result is that the influence of,

15:24 of places that are fairly far away

15:26 is still significant

15:27 in determining what happens in a particular county.

15:30 All that said,

15:31 let's move to the central question here,

15:33 which is,

15:34 uh,

15:34 what are the predictions look like from this model?

15:36 In order to set this up,

15:38 I predicted outside of the sample in the case of both the US and Philippines.

15:42 By that,

15:42 I mean that we start as if,

15:44 uh,

15:44 we're in April 14th and all we know is what we have learned up until April 14th,

15:49 and we forecast

15:50 28 days ahead.

15:53 Then having done that,

15:54 we compare that with what actually happened to see how well we do.

15:57 In order to illustrate this in a very simple way,

16:00 uh,

16:00 I've divided the predictions into three groups,

16:03 the lowest 10th percentile,

16:04 the middle 80,

16:05 the middle 80th percentile points,

16:07 and

16:08 the highest 10th percentile,

16:09 or the 90th percentile or higher.

16:12 Uh,

16:12 the graph here shows you on the vertical axis,

16:14 the actual infection growth,

16:17 and uh,

16:17 on the horizontal axis,

16:19 we're showing then

16:20 the,

16:20 the,

16:21 uh,

16:22 Predictions offered by the model,

16:25 uh,

16:25 at the lowest,

16:26 the medium,

16:27 and the highest levels.

16:27 And what this says is simply that

16:30 we do an awful lot better.

16:31 Our prediction is that growth is,

16:34 uh,

16:34 much higher in counties that in fact do grow more quickly.

16:38 And as you can see in some cases,

16:39 this is fairly spectacular.

16:41 I've isolated um

16:43 the case on the right,

16:44 because that's uh the most spectacular of all.

16:47 The peas in here are,

16:48 are different population groups,

16:50 which are defined above P1 through P4.

16:52 P1 is the largest cities,

16:54 P4 is the smallest.

16:56 Turn to the Philippines,

16:57 we see the same thing again.

17:00 Uh,

17:01 as we move up through the prediction percentiles,

17:04 we see actual growth going up.

17:06 And again,

17:07 what these results suggest,

17:08 as they do in the case of the US

17:10 is that,

17:11 uh,

17:11 in the highest 10th percentile,

17:13 we actually go a long way toward identifying areas that

17:17 in the next period of time,

17:18 in the month following the modeling exercise,

17:21 we will see rapid growth of the virus.

17:23 So this model seems to be pretty robust.

17:26 Having gone all this distance,

17:27 let me then stop,

17:29 uh,

17:29 and extend a little bit.

17:30 We want to extend the model and look at some other measures that might be relevant.

17:34 Uh,

17:34 and one of those is vulnerability,

17:36 and here I join my colleagues briefly in,

17:38 uh,

17:39 providing a very quick introduction to topics that Bo

17:41 and Anna will then cover in much more detail.

17:43 One of the things you might want to look at is vulnerability.

17:47 And in this particular case,

17:48 life expectancy

17:50 uh might be a measure that you wouldn't want to look at in that context.

17:53 This is a regression result for all the covariants

17:55 of life expectancy that enter into these discussions.

17:58 Uh,

17:59 there's no,

17:59 I don't want to dwell on this.

18:00 I simply want to

18:01 point out that

18:03 many of the covariants that have been identified

18:04 with uh the spread of PM10 and vulnerability

18:08 are here at present for life expectancy,

18:10 and I get extremely strong results

18:12 for PM 2.5 as one of those covariants.

18:15 Now,

18:15 that in itself,

18:16 uh,

18:16 this,

18:17 this,

18:17 uh,

18:17 finding of covariation

18:19 doesn't necessarily mean that we have causation here.

18:22 However,

18:23 I wanted to turn quickly before I close to this one graph.

18:26 This is built up from the census tract level in the US,

18:28 very detailed data

18:29 that shows something which is quite remarkable.

18:32 The red here,

18:33 uh,

18:33 denotes counties which are

18:36 most afflicted by 2.5,

18:38 and this is

18:39 to an almost eerie degree,

18:40 a picture of coal fired.

18:41 power in the US,

18:43 particularly in Georgia,

18:43 there,

18:44 the,

18:44 the red zone areas there are some of the biggest coal-fired power plants in the US.

18:48 So

18:48 it's clear that pollution generally defined

18:51 seems to have

18:53 a,

18:53 a measure of correlation that's fairly significant with life expectancy.

18:57 That said,

18:57 I've gone back to the original model

18:59 and put in the life expectancy begin to wind up,

19:02 David.

19:03 Sure,

19:04 I'm,

19:04 I'm,

19:04 this is effectively wrapping it up.

19:06 Uh,

19:06 and what we find is that they have,

19:08 the results have the expected signs.

19:10 They're not tremendously powerful statistically.

19:13 I don't think they add too much to our predictive power,

19:15 but they certainly suggest that there is

19:18 a correlation here beyond the part that is explained,

19:21 uh,

19:21 by the initial model that I talked about.

19:24 So just to summarize then very quickly.

19:27 The challenge was very rapidly to put together a model of COVID spread.

19:30 We put together elements including a gravity model of interactions,

19:34 local interactions in the highest density areas,

19:37 used the Gomert's formulation

19:39 to uh characterize change.

19:41 We got similar strong fits for both the US and Philippines,

19:44 a similar pattern for the extent to which travel time

19:48 affects

19:49 the influence of infections.

19:51 The,

19:52 the forecasts look promising.

19:53 It looks like they do differentiate subsequent experiences of cities,

19:57 and we're now undertaking some extended experiments with the model.

20:00 You've seen a quick illustration for the US.

20:03 We're working on Philippines and subsequently we'll work

20:06 on South Africa.

20:07 Let me just close there.

20:09 Thank you so much,

20:10 David.

20:10 That was really fascinating.

20:12 Thank you

20:13 as always.

20:14 Um,

20:14 Bo,

20:15 over to you now,

20:16 please.

20:21 Thank

20:21 you so much.

20:22 I'm gonna share my screen as well.

20:25 Here we go.

20:27 All right,

20:28 thanks everybody,

20:29 and thanks,

20:29 uh,

20:30 uh,

20:30 Richard for your introduction.

20:32 Um,

20:32 like you said,

20:33 my,

20:33 uh,

20:34 my background,

20:34 uh,

20:35 is,

20:35 uh,

20:35 is,

20:36 is quite technical.

20:36 I did my PhD in econometric theory,

20:39 mostly

20:40 working with,

20:41 uh,

20:41 processes that evolve over space and over time,

20:44 and that's really the angle,

20:45 uh,

20:45 from which,

20:46 uh,

20:46 I come at this.

20:48 Um.

20:49 I work in a,

20:50 in a unit strategy,

20:51 analytics,

20:51 finance and solutions and knowledge in the World Bank,

20:54 and what we're doing is,

20:55 uh,

20:56 we're working on crisis analytics.

20:58 So whenever there's a crisis somewhere I try and put together,

21:01 uh,

21:01 uh,

21:02 the relevant data set to,

21:03 to come at some,

21:04 some useful policy advice and guide

21:07 through that,

21:08 uh,

21:08 and it's quite unnatural to have a crisis at large,

21:11 but,

21:11 uh,

21:12 still,

21:12 uh,

21:12 it's,

21:13 it's the angle from which I,

21:14 I come at this.

21:15 Um,

21:16 so I started

21:17 thinking about this actually early in February

21:20 and,

21:21 uh,

21:21 there was,

21:21 uh,

21:22 the,

21:22 the first signs of,

21:23 of how this virus spreads were,

21:26 were quite,

21:26 quite a,

21:27 a,

21:27 a positive picture.

21:28 Uh,

21:28 uh,

21:29 the first tense was that

21:31 this virus was not airborne,

21:33 uh,

21:33 and that airborne transmission was not believed to,

21:36 uh,

21:36 to explain a substantial part of,

21:38 of the virus spread dynamics.

21:41 Then,

21:41 uh,

21:41 in March,

21:42 uh,

21:43 uh,

21:44 an important paper by Van Dormael and painted a slightly darker picture saying,

21:48 well,

21:48 actually,

21:48 we have some evidence

21:50 that suggests that SARS-COV-2 can survive up to 3 days on some surfaces

21:55 and,

21:56 uh,

21:56 concluded that aerosol transmission might actually be feasible.

22:01 So,

22:02 uh,

22:02 from that perspective,

22:03 uh,

22:04 if this virus will be,

22:06 uh,

22:06 airborne,

22:07 then you might expect similar dynamics as we know exists with other airborne,

22:11 uh,

22:11 viruses.

22:12 So there's a quite a huge body of literature on this,

22:16 and,

22:16 uh,

22:17 the general takeaways is that for various classes of,

22:20 uh,

22:20 viruses

22:21 that are airborne,

22:23 uh,

22:23 we know that the presence of fine particulate matter,

22:26 um,

22:26 uh,

22:27 actually increases the infection risk.

22:30 So,

22:30 um,

22:31 if that holds true,

22:31 then we might expect some similar dynamics,

22:34 uh,

22:34 uh,

22:35 in,

22:35 in COVID-19.

22:37 So when I started looking at the data early on,

22:39 I quickly concluded,

22:41 um,

22:41 that,

22:42 uh,

22:43 uh,

22:43 in at least in the Netherlands,

22:45 uh,

22:45 the case,

22:46 cases seem to double in areas where,

22:48 uh,

22:48 pollution concentrations increased by just 20% above guidelines.

22:53 We're gonna talk a little bit more about,

22:55 uh,

22:55 how we got there,

22:57 um.

22:58 But I want,

22:58 wanted to show a quick picture,

23:00 uh,

23:01 uh,

23:02 to everybody

23:03 to show that we can,

23:04 uh,

23:05 quite well see,

23:06 uh,

23:06 uh,

23:06 pollution from space.

23:08 So it's quite easy to track,

23:11 uh,

23:11 uh,

23:11 pollution distributions,

23:13 uh,

23:13 across countries and within countries.

23:16 So if we just have a,

23:17 a general look at the situation in Italy,

23:20 which was,

23:21 uh,

23:21 really,

23:21 uh,

23:22 escalating,

23:22 uh,

23:23 around the time when I started,

23:24 uh,

23:25 thinking about this problem.

23:26 Uh,

23:27 and we see that in the north,

23:28 uh,

23:29 the nitrogen dioxide,

23:30 which is just one pollutant,

23:32 but,

23:32 uh,

23:32 one for which,

23:34 uh,

23:34 data is quite easily accessible,

23:36 we see that,

23:37 uh,

23:38 that the highest concentrations are really centered in the north,

23:41 and then if you just pull up,

23:42 uh,

23:42 a map,

23:43 and this is from,

23:43 from,

23:44 from early April,

23:45 then you can see that the case densities,

23:47 uh,

23:47 seem to,

23:48 uh,

23:48 have some similarity in how they are distributed across the country.

23:53 Um,

23:54 the situation has seemed to be quite stable.

23:56 This is data from,

23:57 uh,

23:57 from last week,

23:59 uh,

23:59 in the middle again,

24:00 uh,

24:00 the cases on the left,

24:01 the same pollution map,

24:03 and then on the right,

24:04 I just quickly pulled up a population density map,

24:07 uh,

24:07 just to show that actually

24:09 by,

24:09 uh,

24:10 briefly looking at the data,

24:11 the similarities between pollution and case densities

24:14 are much more striking than,

24:16 uh,

24:17 than,

24:17 uh,

24:17 with population density.

24:19 So that begs the question,

24:20 uh,

24:21 what,

24:21 what is driving

24:22 this,

24:23 you can also look at,

24:24 at some other countries in France,

24:26 you see that,

24:26 uh,

24:27 the highest case densities are in the northern part across the border with Belgium,

24:31 Germany,

24:32 and that's generally where you see a lot of cases.

24:34 And then if you start looking across Europe,

24:36 um,

24:37 And this seems to be a recurring pattern,

24:40 but actually not in all places.

24:41 Um,

24:42 first,

24:43 I highlighted the Netherlands,

24:44 which I personally analyzed and we'll talk about the details there.

24:47 In the bottom you see France,

24:49 uh,

24:49 and Italy.

24:50 Um,

24:51 I'm putting,

24:51 uh,

24:52 on the right,

24:52 uh,

24:53 uh,

24:53 the case distribution in Poland.

24:55 Uh,

24:55 just to point out that in many places,

24:58 the regions are actually quite,

24:59 quite big,

25:00 so even though there seems to be some similarity,

25:02 uh,

25:03 with the pollution distribution,

25:04 it's very,

25:05 very hard to,

25:06 to draw any solid conclusions here just by the fact that

25:09 these regions are also large and we only have a few of them.

25:13 Uh,

25:13 then in Austria,

25:14 we actually see a different pattern.

25:16 Uh,

25:17 actually the highest case densities are,

25:19 uh,

25:19 in places where pollution is quite low,

25:21 but we know that,

25:22 uh,

25:23 uh,

25:23 this country has,

25:24 has mountains,

25:25 uh,

25:25 so the terrain is,

25:26 is completely different from one place to another,

25:29 so there could just be,

25:30 uh,

25:30 other factors than,

25:31 than pollution that are important.

25:33 And then,

25:34 uh,

25:34 in,

25:35 uh,

25:35 the United Kingdom,

25:36 England,

25:36 uh,

25:37 you,

25:38 you see that it's not at all that clear.

25:40 You see in the,

25:41 in,

25:41 in the south part,

25:42 a lot of pollution cases seem to be a bit higher,

25:45 but then in the north,

25:45 you also see a lot of cases

25:47 per population,

25:49 and that just shows that these areas are all quite large,

25:52 uh,

25:52 and they don't have a lot of people,

25:54 um,

25:55 so.

25:56 Uh,

25:56 it's not actually all that clear,

25:58 uh,

25:58 how the relationship with population comes into play here.

26:01 So apart from just an obvious similarity,

26:04 there's a lot of other factors that you have to think of.

26:07 Here,

26:07 I'm just quickly showing the same pollution map that David,

26:10 uh,

26:11 pulled up in the previous presentation

26:13 and then overlaying it with,

26:14 uh,

26:15 with,

26:15 with cases

26:16 just to again see the,

26:17 the striking similarity here.

26:20 So

26:20 when it comes to thinking through what's behind,

26:23 what's what's behind this,

26:24 uh,

26:24 this correlation,

26:25 there,

26:25 there could be many explanations.

26:28 Uh,

26:28 there could be actually direct relationships and,

26:31 uh,

26:31 uh,

26:32 Ana will talk a bit more about that.

26:34 But this would have to do with that pollution,

26:36 uh,

26:37 increases susceptibility to the virus,

26:39 uh,

26:39 through health impacts

26:41 or it might even be the case that,

26:43 uh,

26:43 the virus,

26:44 uh,

26:44 attaches on or suspends into aerosols

26:47 and then uh that particulate matter,

26:49 uh,

26:49 helps the spread of the virus.

26:52 There could be other forms of associated relationships,

26:55 uh,

26:55 meaning that air pollution could just proxy,

26:57 uh,

26:58 various risk factors,

26:59 and this could be quite broad.

27:01 It could mean that

27:02 pollution tends to be produced in,

27:04 uh,

27:05 in working environments where a lot of people work in cramped conditions,

27:09 and it's really just the fact that there's a lot of people working in cramped,

27:12 uh,

27:13 locations that increases the transmission,

27:15 um.

27:16 And then finally,

27:17 there are of course completely different uh explanations

27:20 to these correlations which could be just incorrect,

27:23 uh,

27:24 spurious trends,

27:25 it just happens to be that pollution uh trends in the same way

27:29 and uh there could be other confounding relationships,

27:32 meaning that,

27:33 uh,

27:34 it's really population density that proxies for activity and

27:37 then you expect both more cases and more pollution.

27:40 So some of these factors,

27:41 we can treat them in a statistical method.

27:44 Um,

27:45 but in the end,

27:46 uh,

27:46 we can exclude some of these explanations,

27:49 but will always remain,

27:50 uh,

27:51 uh,

27:51 that,

27:52 uh,

27:52 even if we can show that there's a robust explanation,

27:54 these type of analysis that at least I'm gonna show,

27:57 uh,

27:57 are not intended to really explain,

27:59 uh,

28:00 why you see this,

28:01 uh,

28:01 uh,

28:02 this correlation.

28:04 So just uh the analytical strategy,

28:06 uh,

28:07 to summarize,

28:07 we're gonna need granular data on uh COVID-19

28:11 cases and very precise air pollution measurements.

28:14 This already puts a lot of constraints on,

28:16 on what,

28:16 what,

28:17 what areas we can study rapidly.

28:19 Ideally,

28:20 we would have patient level data,

28:21 but this is not available everywhere or not always open.

28:24 Uh,

28:25 we're going to need reliable data on a number of

28:27 control variables that have to do with health pre conditions,

28:30 which are certainly not available everywhere,

28:32 and then we're also going to need a reasonable

28:34 sample size or at least very small area district,

28:36 uh,

28:37 this

28:38 units.

28:39 Uh,

28:39 so one of the,

28:40 the places that kind of have this,

28:42 uh,

28:43 at least early on,

28:44 uh,

28:44 uh,

28:45 making that data available is the Netherlands.

28:47 So I analyzed 355 municipalities in the Netherlands,

28:50 which are all quite small.

28:52 Um,

28:52 I looked at different,

28:53 uh,

28:54 dependent variables,

28:56 confirmed cases,

28:57 uh,

28:57 per per capita

28:58 with confirmed hospital admissions per capita.

29:01 I also did,

29:02 uh,

29:02 case counts,

29:03 um,

29:03 and then I used a variety of air pollution,

29:06 uh,

29:06 data sets.

29:07 Uh,

29:08 2 based on ground measurements,

29:09 PM 2.5,

29:10 PM 10,

29:11 and I also looked at remote sense data.

29:14 Um,

29:14 generally,

29:15 these considerations don't really change the conclusion,

29:18 which is good.

29:19 Uh,

29:19 then I looked at a number of control variables including population density,

29:23 pre-existing,

29:24 uh,

29:24 health conditions,

29:26 uh,

29:26 and,

29:26 uh,

29:27 uh,

29:27 proxies for case severity,

29:29 use a number of demographic,

29:31 uh,

29:31 data sets including age,

29:33 household composition,

29:34 uh,

29:35 And then uh in terms of methodological considerations,

29:38 uh,

29:38 I looked at non-linearities,

29:40 uh,

29:40 I looked at alternative distributional assumptions,

29:42 the influence of outliers,

29:44 um,

29:45 uh,

29:45 using a variety of tools to control for spatial dynamics and so on.

29:50 You can read the details in the paper.

29:52 I'm just gonna present you the main simple,

29:54 uh,

29:54 estimation results as

29:56 most of the more sophisticated considerations

29:59 do not really seem to impact the main conclusions.

30:02 Uh,

30:02 here in this table,

30:03 I'm presenting very simple linear regression results.

30:06 The first model includes,

30:08 uh,

30:08 all the control variables that I could get my hands on,

30:11 which are,

30:11 uh,

30:12 22,

30:13 uh,

30:13 ranging in,

30:14 in the categories that I just mentioned.

30:16 Uh,

30:17 in this model,

30:17 we see immediately that PM 2.5 is an extremely significant and a strong predictor.

30:23 Uh,

30:23 every point in PM 2.5 seems to raise the number of cases,

30:28 uh,

30:29 per 100,000 by 10.

30:30 And at this point in the data,

30:32 the average was about 20 cases for every 100,000 people in these areas.

30:38 So,

30:39 uh,

30:39 meaning that a 2 point increase

30:41 seems to correlate with about a doubling on average.

30:45 Um,

30:46 then the other models,

30:47 uh,

30:47 they take into account spatial dynamics,

30:50 uh,

30:50 spatial dependence,

30:51 you see that the,

30:52 the parameters,

30:52 uh,

30:53 value changes a bit in magnitude.

30:56 Uh,

30:56 but those familiar with these types of models,

30:58 uh,

30:59 when,

30:59 when you're modeling spatial feedback,

31:01 there there's a multiplier effect.

31:03 Uh,

31:04 so generally if you take into account,

31:05 it actually seems to suggest the same,

31:07 uh,

31:08 relationship linearly,

31:10 doubling in cases when,

31:11 uh,

31:11 pollution increased by 2 points,

31:14 um.

31:16 What I,

31:16 uh,

31:16 then did is,

31:17 uh,

31:18 I looked at the nonlinearities,

31:19 uh,

31:19 uh,

31:20 in this relationship.

31:21 There's a lot of factors that might actually influence or might be important

31:25 when it comes to the relationship between pollution

31:28 and,

31:28 and COVID cases.

31:30 Uh,

31:30 there could be,

31:31 uh,

31:31 weather variables,

31:32 uh,

31:33 that influence how impactful pollution actually is.

31:36 Uh,

31:37 there might also be economic,

31:39 uh,

31:40 uh,

31:40 conditions that,

31:41 that,

31:41 uh,

31:42 increase the vulnerability.

31:43 Uh,

31:44 uh,

31:44 of people.

31:45 So,

31:46 um,

31:46 for example,

31:47 in poor regions,

31:48 people might not be able to,

31:50 uh,

31:50 protect themselves all too well against pollution,

31:53 hence the impact from pollution onto COVID,

31:56 uh,

31:57 uh,

31:57 might be higher.

31:58 So there's a lot of reasons to even in this,

32:00 uh,

32:01 in,

32:01 we're thinking about a simple correlation,

32:03 there might be a lot of influencing factors that,

32:05 that Uh,

32:07 that,

32:07 uh,

32:08 that determine how this,

32:09 this correlation actually looks from one place to another.

32:12 So I used the nonlinear regression,

32:14 uh,

32:14 method,

32:15 use a number of,

32:16 uh,

32:16 important control variables,

32:18 and I also used X and Y coordinates.

32:20 So effectively,

32:21 uh,

32:22 this type of regression,

32:23 uh,

32:23 uh,

32:23 actually allows you to model

32:25 the relationship between COVID and air pollution.

32:29 In such a way that this relationship varies from,

32:31 uh,

32:32 across levels in the data and in this

32:33 case also explicitly from one location to another.

32:37 Then,

32:37 uh,

32:38 I did a simple prediction exercise,

32:40 um,

32:41 predicting,

32:42 uh,

32:43 what,

32:43 what the expected or at least

32:45 from,

32:45 from the model's perspective,

32:47 how the case distribution looks like,

32:49 uh,

32:49 when,

32:49 when areas at about 10 p.m.

32:53 Levels and then increasing it to 12,

32:56 uh,

32:56 calculate the,

32:57 the change,

32:57 uh,

32:58 and then you see that,

32:59 that,

32:59 that this,

33:00 this,

33:00 uh,

33:00 this effect differs,

33:02 uh,

33:02 from one area to another,

33:04 so you can draw a bit of a distribution

33:06 and then you see that there's a lot of heterogeneity,

33:09 uh,

33:09 in,

33:09 uh,

33:10 in,

33:10 in the increase,

33:11 uh,

33:12 in cases.

33:13 So

33:14 again,

33:14 it seems to center around,

33:16 uh,

33:16 a doubling in in cases when pollution increases from 10 to 12,

33:21 uh,

33:21 points.

33:22 But,

33:23 uh,

33:23 this relationship really seems to vary from one area to another.

33:27 This is also important because,

33:29 uh,

33:29 at least from the World Bank's perspective,

33:31 we're,

33:31 we're pretty much concerned about developments in,

33:34 in poorer countries,

33:35 and these tend to have higher pollution concentrations,

33:38 um,

33:39 and,

33:40 uh,

33:40 just this heterogeneity in,

33:42 uh,

33:42 in a very small country like the Netherlands

33:44 already

33:45 shows that it,

33:46 it's not,

33:47 not that trivial,

33:48 uh,

33:48 to extrapolate.

33:50 This relationship to to countries where the conditions might be,

33:53 uh,

33:54 might be different.

33:55 So that's kind of like brings me to a bit,

33:57 a bit of a summary of the main takeaways for,

33:59 from,

34:00 uh,

34:00 from the analysis,

34:01 and that is that PM 2.5 is a,

34:03 is a well-known health risk factor

34:06 and we already know that short-term exposure to pollution has

34:10 shown to decrease infection risk for various viral infections.

34:14 Uh,

34:14 this is my last slide.

34:16 Uh,

34:16 and then emerging analysis,

34:18 uh,

34:19 across the board seem to point toward PM 2.5 as a strong predictor,

34:23 uh,

34:23 of COVID-19 incidents,

34:25 and that seems to be robust to,

34:27 uh,

34:27 uh,

34:27 a number of control variables and different methodological approaches.

34:31 There's some cautionary notes in order,

34:33 uh,

34:33 I would say,

34:34 and,

34:34 uh,

34:35 first,

34:35 these type of analysis,

34:37 uh,

34:37 they're not intended to explain why,

34:40 uh,

34:40 or how this correlation works.

34:42 Uh,

34:42 we also know that the available data on COVID-19 is at best sub-optimal,

34:47 um,

34:48 and,

34:48 uh,

34:49 we don't yet know from just analysis in a country like this

34:53 how this relationship extrapolates to higher PM 2.5 concentrations.

34:58 Uh,

34:58 a systematic review on the applicability of these findings across countries,

35:02 uh,

35:02 would be highly beneficial at this point.

35:05 Uh,

35:05 and as most of this literature is,

35:07 is still in early,

35:08 uh,

35:08 stages,

35:09 uh,

35:09 I'd really caution everybody on,

35:11 on how,

35:11 how,

35:11 how to interpret these results.

35:13 Uh,

35:14 so,

35:14 uh,

35:14 I'm happy to take a lot of questions in the end.

35:17 Uh,

35:17 I'll keep it at,

35:18 uh,

35:18 at this for,

35:18 for now.

35:19 Thank you all so much.

35:21 Thank you,

35:21 thank you very much,

35:22 Bo.

35:22 That was fascinating as well,

35:24 in particular the quantitative magnitude that you've

35:27 come up with for the Netherlands.

35:28 Um,

35:29 and,

35:29 it's over to you now to give us some

35:31 insight as economists as to what's really behind it.

35:35 Uh,

35:35 teach us some epidemiology

35:37 of what's going on here.

35:38 Thank you.

35:38 Over to you,

35:39 Anna.

35:40 Thank you,

35:40 I'll just,

35:41 uh,

35:42 set up sharing my screen.

35:44 So,

35:44 uh,

35:45 can you hear me?

35:46 Yes,

35:47 perfect clarity.

35:48 Yes,

35:49 yeah,

35:49 fantastic.

35:50 So

35:51 I started my career as a respiratory.

35:55 Into public health and for the last 20

35:58 years or so,

35:59 I've been working in environmental epidemiology,

36:02 primarily in air pollution but uh other

36:04 uh types of uh environmental exposure.

36:07 So I'm going to give a talk very much

36:10 from an epidemiological and public health

36:14 perspective on this.

36:15 Um,

36:16 so

36:17 it's a,

36:17 a,

36:18 a different starting point from,

36:20 um,

36:21 our previous speaker on this.

36:23 Um,

36:23 but I will start with a similar,

36:25 uh,

36:25 uh,

36:26 slide here

36:27 showing,

36:27 um,

36:28 the air pollution levels in China

36:31 and Italy.

36:32 And these were

36:33 two of the epicenters of COVID-19,

36:37 um,

36:37 or two of the initial epicenters,

36:39 and it wasn't very long before people said the air pollution levels are high

36:43 here and in Northern Italy,

36:45 we've seen high amounts of COVID-19.

36:47 Could these potentially be

36:49 related?

36:50 And there have been um a number of papers um

36:53 that have been published and

36:55 more er er opinions that have also been put out.

36:58 I've picked out this particular paper

37:00 um from

37:01 um

37:02 which was one of the first um which was er um for Europe,

37:06 which was er er suggesting that the highest levels of COVID-19 deaths

37:11 were in the most polluted regions.

37:14 So uh that's

37:15 throws up a hypothesis,

37:16 but it's a correlation.

37:18 And um you need to look beyond that to see what else is going on because

37:23 um these areas with a high air pollution level tend

37:25 to be the areas with the high population density.

37:28 They tend to be very well connected areas,

37:30 so they're where the infection came in first,

37:33 got a hold spread

37:34 um through the the density of the population.

37:37 Um,

37:38 they also may have uh um areas of deprivation.

37:41 Um,

37:42 and that,

37:43 uh,

37:43 in itself is a,

37:44 is a risk factor.

37:45 So I think to be able to understand,

37:47 um,

37:48 the

37:48 relationship,

37:49 uh,

37:50 or potential relationship between air pollution and COVID-19.

37:53 Need to,

37:54 uh,

37:55 first start off thinking about the epidemiology.

37:58 So I've got some figures here that uh come from the UK

38:01 um there are similar types of figures that

38:03 are coming out in terms of COVID-19 mortality

38:07 in relation to age.

38:08 So age

38:09 is a really big uh risk factor for COVID-19 as is,

38:13 uh,

38:13 being male.

38:14 Um,

38:15 but that,

38:15 those risk factors only seem to come into play,

38:18 uh,

38:18 for severe disease when you're over about the age of 40 unless you have

38:22 other underlying conditions.

38:24 So that means if you're a developing country with,

38:26 um,

38:27 a,

38:27 a large percentage of your population

38:30 in younger age groups,

38:32 um,

38:32 it probably means that the risk of,

38:35 uh,

38:35 death and severe complications,

38:37 um,

38:37 is,

38:37 is much lower in your population if you have a,

38:40 a younger population.

38:42 So,

38:43 um,

38:44 the second thing to comment is just about this

38:46 top right hand side about the usual all cause mortality

38:49 rates.

38:49 So COVID-19 is a new disease,

38:51 it's not behaving like,

38:53 uh,

38:53 some other diseases.

38:54 So pneumonias and,

38:56 uh,

38:56 uh,

38:57 various other respiratory diseases usually have this J shape

39:00 relationship with,

39:01 uh,

39:01 mortality with a,

39:03 uh,

39:03 a high,

39:04 uh,

39:04 higher impact in those less than 1 year old in babies.

39:08 Um,

39:08 we don't see that or we haven't seen that with COVID.

39:12 And then thinking about the risk factors,

39:14 um,

39:14 people think of it as a respiratory disease,

39:16 but actually,

39:17 uh,

39:18 the comorbidities that are showing up really clearly are cardiovascular ones,

39:22 hypertension,

39:23 um,

39:23 heart disease,

39:24 also,

39:25 uh,

39:25 metabolic disease such as diabetes,

39:28 and,

39:28 um,

39:28 in fact,

39:29 certainly in the UK our asthmatics have been

39:31 underrepresented in those with the more severe disease,

39:34 which is interesting.

39:36 It's a systemic disease.

39:37 I think we're finding that the severe.

39:39 Uh,

39:40 uh,

39:40 COVID-19 is

39:42 not just impacting on the lungs,

39:43 it's,

39:44 uh,

39:44 affecting the blood vessels as well,

39:46 and that might be important for long term sequelae,

39:49 um,

39:49 of the disease and,

39:51 um,

39:51 implications for follow up.

39:54 So I've put other risk factors here,

39:56 obesity and non-white ethnicity in Western countries.

40:00 Um,

40:01 so whether,

40:02 uh,

40:02 people who are,

40:03 uh,

40:03 from Asia,

40:04 South Asia and Africa,

40:06 um,

40:07 experience,

40:07 uh,

40:08 higher rates

40:09 in Asia and Africa,

40:11 I think we still have to,

40:13 uh,

40:13 you know,

40:13 we still need more information on

40:16 in

40:16 the Western countries,

40:17 these non-white ethnicities tend to uh be more deprived.

40:21 Um,

40:21 they live in the,

40:22 um,

40:22 uh,

40:23 very densely populated,

40:24 uh,

40:25 areas,

40:26 and,

40:26 uh,

40:26 they have higher rates of comorbidity which

40:29 might well be related to the deprivation,

40:31 and it's very difficult to untangle,

40:34 um,

40:34 that,

40:35 uh,

40:35 those,

40:35 uh,

40:36 all those factors.

40:38 So,

40:39 um,

40:39 how might air quality

40:41 impact on COVID-19,

40:42 and I think it's worth thinking through what mechanisms

40:46 might underlie any,

40:47 uh,

40:48 uh,

40:48 observed relationships.

40:50 So I think an obvious one is that the,

40:53 uh,

40:53 air pollution impact is indirect.

40:55 There's lots of evidence that air pollution increases the risk of chronic disease,

40:59 and we know that chronic disease gives you a risk of more severe COVID-19

41:03 and death from COVID-19.

41:06 And if that's the case,

41:07 and the only mechanism,

41:08 you might expect the risks to be of a similar order of magnitude

41:12 to previous studies looking at air pollution and mortality.

41:16 There are some other potential mechanisms which I,

41:19 I,

41:19 I think are more short-term and direct,

41:22 and they would be around increasing infectivity.

41:25 So,

41:26 um,

41:26 that might occur if the virus is er being carried

41:30 on particulates.

41:32 It might occur through inflammation of the lung,

41:34 er,

41:35 from air pollution which we know is a a a feature of air pollution exposure,

41:39 um,

41:39 or it might occur through some specific mechanisms.

41:42 I've got a slide on that.

41:44 Um,

41:44 so that would give you,

41:45 uh,

41:46 an interaction,

41:47 so you would see higher infection risk in polluted areas.

41:50 So,

41:51 uh,

41:51 they,

41:51 you wouldn't,

41:52 you'd have higher risk than you see with

41:53 the previous studies of air pollution and mortality.

41:56 Um,

41:57 you

41:58 might also have,

41:59 uh,

41:59 an impact of air pollution

42:00 that if you have severe,

42:02 uh,

42:02 if you have,

42:03 uh,

42:03 an infection,

42:04 it might make the infection more severe,

42:06 which might be through these sort of general inflammation,

42:09 uh,

42:09 mechanisms,

42:10 and we know that,

42:11 uh,

42:11 air pollution can cause inflammation through body systems.

42:16 So I'll just tackle uh a couple of those uh uh direct mechanisms.

42:22 So um as Bo uh

42:23 uh uh talked about,

42:26 there are some studies that suggest that you can detect coronavirus

42:30 on particles of air pollution.

42:32 Uh,

42:32 there's not a lot of studies around.

42:34 Those studies don't tell us whether

42:36 uh the virus uh material that's detected is viable

42:41 nor whether uh you can pick up enough from the air

42:44 um to get an infected dose.

42:47 I'm not sure we actually know at the

42:48 moment how much virus constitutes an infected dose.

42:52 And I think the,

42:52 the role of aerosols,

42:54 um,

42:55 in terms of infection is

42:57 not very well established.

42:58 We know that it's a droplet.

43:00 Uh,

43:00 transmission,

43:01 um,

43:02 and if those droplets contaminate surfaces,

43:04 you can pick them up from surfaces.

43:06 Um,

43:06 aerosol transmission is probably a small

43:09 risk factor,

43:10 but we don't know how small it is.

43:12 And,

43:12 um,

43:13 it's,

43:13 it's quite important I think to get a handle on

43:16 that because it would affect our social distancing policies,

43:19 uh,

43:19 right through to is it safe to use a public toilet,

43:22 for example.

43:23 And that's,

43:23 uh,

43:24 um,

43:24 some of the things we're grappling with in the UK

43:26 at the moment uh as we're coming out of lockdown.

43:30 So there's another uh mechanism which is

43:33 uh really interesting from a medical perspective.

43:36 This is a busy slide,

43:37 um,

43:37 just concentrate on the middle bit here,

43:39 we've got a red uh SARS um

43:42 uh COV-2 virus here

43:45 and this is the ACE2 receptor in green,

43:47 so the uh virus,

43:49 coronavirus attaches to this receptor and that's how it enters the cell.

43:53 And uh there is evidence that the uh that

43:56 both NO2 and particulate pollution

43:59 can up regulate,

44:01 increase the numbers of these ACE2 receptors.

44:03 Um,

44:04 what we don't have is direct evidence of,

44:07 uh,

44:08 um,

44:09 air pollution levels

44:10 plus virus,

44:11 um,

44:11 and some cells,

44:13 um,

44:13 in a,

44:14 uh,

44:14 a mechanistic study to say

44:16 does,

44:17 uh,

44:17 the virus,

44:18 uh,

44:18 plus air pollution give you a more sort of potent,

44:21 uh,

44:21 risk of infection.

44:23 Um,

44:24 than,

44:24 uh,

44:24 not having the air pollution,

44:25 and I think we urgently need that,

44:27 uh,

44:28 to be able to understand some of the

44:30 epidemiological studies.

44:33 So there are a few epidemiological studies and as Bo said it's quite early days.

44:38 Um,

44:38 I think one of the problems we've got is where do I look?

44:41 So,

44:41 uh,

44:42 my look didn't actually turn up in your paper,

44:44 I'll apologize for that.

44:45 So,

44:46 so,

44:46 um,

44:46 I tend to look in medical studies on.

44:48 PubMed.

44:49 There is a WHO database that,

44:51 uh,

44:52 is,

44:52 is sort of a,

44:53 I think PubMed Plus.

44:54 Um,

44:55 you can also look on Med Archive,

44:57 which is,

44:58 uh,

44:58 for pre-prints,

45:00 but it's,

45:00 I put a question mark there because they're very variable quality

45:04 and,

45:04 um,

45:05 and there's a huge amount.

45:06 So people are just putting things up there.

45:08 They get a press release,

45:09 they come out in the,

45:10 in the,

45:11 um,

45:12 um,

45:12 in the media and you're having to react to,

45:14 to them,

45:14 um.

45:15 And um there's now concern in the,

45:17 this is from the British Medical Journal,

45:19 um this article here saying that er

45:21 we're actually getting perverse incentives to publish early

45:25 and possibly too early

45:26 so that er um there's too much data and not enough information.

45:31 We have another problem about outcome measures

45:34 and um so um

45:36 uh I think deaths

45:38 are

45:39 well there's a there's a whole issue about uh

45:41 how you interpret deaths,

45:42 excess deaths,

45:43 deaths from COVID-19.

45:44 They're all defined differently and in different countries.

45:47 Um,

45:48 the thing that is really problematic is a case,

45:51 um,

45:51 because a case,

45:52 um,

45:53 uh,

45:54 is dependent on testing

45:55 and the testing,

45:56 who gets selected to be tested has varied,

45:59 um,

46:00 within the areas in within countries,

46:02 um,

46:03 and between countries and also has changed over time,

46:06 so.

46:06 Um,

46:07 you know,

46:07 it's been changing in some cases every week as to who's,

46:10 uh,

46:11 eligible to be tested,

46:13 um,

46:13 and there are very few countries that have tested,

46:15 uh,

46:15 large numbers of the population.

46:17 I think South Korea and Germany stand out there.

46:20 Um,

46:21 I think we've got another challenge which is around the methodology,

46:24 um,

46:25 about disease propagation and the previous,

46:28 uh,

46:28 talks,

46:29 I think of trying to get a handle on this,

46:31 um,

46:31 but we're seeing this sort of pulse of

46:34 infections go out,

46:35 um,

46:36 and that's not the type of methods that we've developed to look at these short-term,

46:40 um,

46:41 interactions,

46:42 um,

46:42 and then some of our control measures are impacting

46:45 on the air pollution as well as the transmission.

46:47 And it then gets very tricky to start unpicking

46:51 um what the

46:53 true associations might be.

46:56 I'm gonna focus on two,

46:59 I think quite,

47:00 uh,

47:00 authoritative

47:01 studies or certainly ones that have

47:04 provoked a lot of discussion.

47:06 Um,

47:06 and they're both from the States,

47:09 um,

47:09 in part because I think you have,

47:11 uh,

47:12 good access to data.

47:14 Uh,

47:14 one of them came out in April,

47:16 the second came out in May.

47:18 Um,

47:18 they're similar but slightly different designs

47:21 and the results are slightly different.

47:23 And um

47:24 they're both from very well respected um air pollution epidemiology

47:29 groups.

47:30 Uh,

47:30 the XX uh Francesca Dominici group is at Harvard,

47:33 um,

47:34 and the uh

47:35 Dong Haiang and Ha Chang are um

47:38 at Emory,

47:39 but there's also some Harvard.

47:43 So,

47:43 um,

47:44 this slide

47:45 picks out some of the,

47:47 uh,

47:47 key,

47:48 um,

47:48 aspects of these studies.

47:52 So the colors are not meaningful.

47:53 They're just so that you can pick out both.

47:55 Um,

47:56 the,

47:56 one's using death counts,

47:58 uh,

47:59 and the Liang study is using deaths and cases.

48:02 I'm just going to concentrate on the mortality here.

48:05 Um,

48:05 they're both looking up to,

48:07 uh,

48:07 nearly the end of April.

48:09 They are area level studies.

48:10 They're comparing US counties.

48:13 Um,

48:13 for some reason,

48:14 there are

48:14 differences in the number of deaths picked up between the different databases.

48:19 They both use very well established and good air pollution models um to

48:25 uh give you the,

48:25 the air pollution levels,

48:27 um,

48:28 but the

48:29 wall paper,

48:29 the first paper

48:30 just looked at PM 2.5.

48:32 The second paper looks at NO2 and ozone as well.

48:35 Slightly different years,

48:37 but they're looking at long term averages,

48:39 um,

48:39 in relation to,

48:40 uh,

48:41 COVID-19.

48:42 So there are some differences in the models that they've used,

48:46 um,

48:46 these statistical models.

48:48 Um,

48:48 so

48:49 the second paper has used a zero inflated model because um a

48:53 lot of the counties had zero deaths during this time period.

48:57 And I think the other major difference,

48:58 and we have just touched on that is that the second paper

49:01 also adjusted for spatial autocorrelation

49:04 and probably explains why

49:06 there are uh uh lower uh results found.

49:10 These are really impressive

49:11 amounts of analysis,

49:13 um,

49:14 with,

49:15 The

49:17 wallpaper,

49:17 they found an 8% increase in mortality rate

49:21 per 1 mcg

49:23 per meter cubed increase of PM 2.5,

49:25 and this is adjusted for about 20 different confounders,

49:28 including

49:29 attempts to look at the state of the

49:32 progression

49:33 of the pandemic,

49:35 uh,

49:35 infection rates,

49:37 and population density and various population vulnerability factors.

49:41 Um,

49:42 the Leanne paper finds,

49:43 uh,

49:44 an association that's not statistically significant with PM 2.5,

49:48 and it's about a third the size.

49:51 Um,

49:52 it does find a significant association with

49:55 NO2

49:56 and it expresses that per interquartile range

50:01 um of the NO2 exposure

50:02 and there's no association with ozone.

50:05 And then both papers give us some.

50:07 Uh,

50:08 uh,

50:08 public health type interpretation

50:10 and the thing that made me sit up and notice on

50:12 the Wu Dominici paper was that the coefficients were actually higher

50:16 than those for all cause mortality that they'd found in the previous,

50:20 um,

50:20 analysis.

50:21 And uh I think the Lianne paper commented it's around about

50:25 7% of this might have been avoided if you had lower NHs.

50:30 I think the other thing to note is the P.M.2 levels were

50:33 quite low here.

50:35 They're not comparable to what you'd see in a developing

50:37 country.

50:39 Um,

50:41 In terms of the inference of this,

50:43 um,

50:44 we can come back to this in questions probably,

50:46 but,

50:47 uh,

50:47 epidemiological studies,

50:48 you want to look at the totality of the evidence

50:51 and interpret in the light of what you already know,

50:53 um.

50:57 Hi,

50:58 you're giving me a minute to go.

51:00 That's great.

51:01 Um,

51:01 I will just say that the ecological studies are

51:05 often used in the initial

51:07 assessments and they're very,

51:09 uh,

51:09 good if you want to get a quick,

51:11 uh,

51:11 answer,

51:13 um,

51:13 for,

51:14 uh,

51:15 for,

51:15 uh,

51:16 these types of questions,

51:17 but,

51:17 uh,

51:18 you probably want a long term individual level could study

51:21 uh to help you with that.

51:23 Um.

51:25 I think it would be surprising if we didn't

51:27 see a link between air pollution and COVID-19 given

51:30 what else we know about air pollution and COVID-19.

51:33 Um,

51:34 I have got some couple of slides here just on

51:36 public health perspectives about what we can learn from SARS.

51:39 I can come back to those,

51:40 um,

51:41 in,

51:42 uh,

51:42 sorry,

51:43 I can come back to those in questions if you'd like to,

51:45 uh,

51:45 know a bit more.

51:46 Um,

51:47 but,

51:48 uh,

51:48 in conclusion,

51:49 um,

51:50 the air pollution,

51:52 we already know is associated with risk of chronic disease.

51:55 And um.

51:57 Also with mortality,

51:59 it's unlikely

52:00 that we don't also see a link with COVID-19.

52:03 We're early days in interpreting the evidence.

52:06 I don't think,

52:06 and also the Committee of the Medical Effects of Air Pollution,

52:09 which I sit on in the UK

52:11 doesn't feel that the evidence is strong enough to

52:13 get specific measures against COVID-19 in high air pollution

52:17 areas that are different from low air pollution areas.

52:20 Um,

52:21 I think there's various gaps that we need

52:22 to fill in to understand this relationship better.

52:25 And I think one of the positive things

52:27 uh just from a public health perspective is that

52:30 if you have a well functioning public health system,

52:33 you can

52:34 uh achieve low

52:35 disease transmission.

52:36 I think that's really important to just hold in mind when we're thinking about the uh

52:40 um air pollution levels that actually

52:43 public health measures really important.

52:45 Don't lose our focus on that uh

52:47 even when we're thinking about other risk factors.

52:50 I'll stop there.

52:51 Thank you.

52:55 Thank you,

52:56 thank you so much.

52:57 Um,

52:57 without further ado,

52:59 let's go to our discussions.

53:01 And our first discussion is,

53:02 uh,

53:02 Doctor Sherm Lowe

53:04 from the Urban Unit in the World Bank.

53:06 Shermick,

53:06 over to you.

53:08 Thank you very much,

53:09 Richard.

53:09 And uh

53:11 I really appreciate

53:13 the analytic work and empirical analysis that was presented this morning

53:19 that tells us a little bit about the pathways

53:22 for the spread of coronavirus and the COVID-19 disease.

53:26 I think,

53:27 to me,

53:27 at least from what we do at the World Bank,

53:29 this work will be very useful in the near term,

53:32 as cities in many developing countries.

53:35 are really trying to prioritize

53:38 how and where to allocate medical and civil

53:42 resources to save lives and to save livelihoods.

53:45 In particular,

53:46 I really like the way both David and Bo have chosen to be very spatially aware

53:53 in their approaches to examine the spread of the virus.

53:56 And I think,

53:57 David,

53:57 your modeling draws on a large body of work in economic geography

54:02 that relies on market access and

54:04 gravity-based approaches to measure connectivity.

54:07 And this is really apt for measuring contagion.

54:10 And Bowe's work does a really nice

54:13 job in drawing on the spatial econometrics literature

54:16 pioneered by the likes of Luke Ansellin

54:19 to take into account of spatial trends and spatial clustering.

54:24 And

54:25 such a spatial

54:26 approach is really important because much of the work I see is

54:30 spatially sort of blind.

54:32 And this makes the work very robust

54:34 and uh

54:36 uh and it improves our understanding of the

54:38 correlates linked to the spread of the virus.

54:41 My one suggestion here

54:43 is on both the papers,

54:45 we may want to address the so-called reflection problem

54:49 that's often seen in spatial models of social interaction.

54:54 And here,

54:54 infection rates of individual zip codes and neighborhoods

54:58 are endogenous to the broader neighborhood in general.

55:02 And,

55:02 and we'd probably want to think about some way of instrumenting for that.

55:07 Now,

55:07 I'd just like to tell you a little bit about my feelings on the spatial correlates.

55:12 And first is the density in coronavirus,

55:14 where I think,

55:15 I'm just gonna share one slide of mine.

55:18 Uh,

55:19 and hope you can,

55:20 uh,

55:21 see it,

55:22 uh,

55:22 right here,

55:23 uh,

55:24 where

55:25 there's been a lot of attention.

55:28 Mons

55:31 The downsides of density

55:33 and the fact that in many dense urban hubs,

55:37 uh,

55:37 instead of focusing on the benefits in terms of productivity and livability,

55:43 this density in this time has been a real

55:47 transmission belt for the virus.

55:49 And I think,

55:49 David,

55:50 your analysis shows the significance of density and transmission,

55:54 while Bo's work shows density is not

55:57 really relevant.

55:58 But I think what we really need to do is make clear

56:01 the distinction between density and crowding,

56:04 and I'd like to just put together,

56:06 show you some data we put together for New York,

56:09 where if you were to look at Jackson Heights in Queens,

56:12 that's the most

56:13 worst affected neighborhood.

56:15 Cases are like 4000 per 100,000 residents.

56:18 In Chelsea on the other side,

56:20 it's 925 per 100,000 residents.

56:24 But what's relevant to note here.

56:27 Is that Queens is not the densest neighborhood in New York City.

56:30 Queens has a density of 12,000 people per square kilometer,

56:35 where in Chelsea,

56:36 it's about 31,000.

56:38 So what makes a difference here

56:40 is neighborhood incomes and associated characteristics,

56:44 which really temper the extent to which complementary investments in housing

56:48 and infrastructure,

56:50 and amenities

56:51 can really transform places from being crowded to becoming dense and livable.

56:56 And

56:56 Good planning

56:58 and land development regulations,

57:00 these sort of valuable places,

57:02 developers have the incentive to build up tall structures,

57:05 create,

57:05 create a lot of floor space.

57:07 So even when you have a pandemic and you want to shut down and keep social distance,

57:12 dense places do pretty well.

57:14 And

57:15 in fact,

57:15 to support the World Bank's work on the pandemic

57:18 in developing country cities.

57:21 We have developed a methodology that can help city leaders prioritize

57:26 places and resources towards these potential hotspots,

57:29 and these are the places with high contagion risk,

57:32 uh,

57:32 even under a lockdown.

57:34 And what we do here is we distinguish

57:37 density from crowding by accounting for differences in floor space

57:41 and availability of amenities.

57:43 So I think we need to take this density argument a little carefully.

57:48 Second,

57:48 on the issue of air quality,

57:50 as

57:51 Anna kind of pointed out,

57:53 the evidence isn't all that clear.

57:56 And if you were to look at the numbers,

57:59 air pollution is a really silent killer.

58:01 It takes about 4 million lives annually,

58:04 and,

58:05 and,

58:05 and most

58:06 parts of the world,

58:07 especially in developing countries,

58:09 cities,

58:10 are way over the WHO guidelines.

58:12 So I'd be surprised not to see a correlation.

58:15 But I don't really know the extent to which we're untangling

58:19 the fact

58:20 of

58:21 the air pollution on life,

58:23 quality or expectancy

58:25 from its

58:27 from getting COVID-19,

58:29 and this is particularly true because over the last two months we've seen

58:32 a lot of cities shutting down and there's a lot of lockdown,

58:36 a lot of NPIs,

58:37 so air quality measures may have really changed as well.

58:40 But here,

58:41 when we think about it,

58:41 and I'll just take one more minute.

58:44 we want to think

58:45 if you want to do robust work,

58:46 we need to think structurally.

58:48 And if PM 2.5 is a combination of emissions from industry

58:53 and transportation

58:54 and there are places with dirty industries,

58:57 they're likely to have low land prices.

58:59 And the places with a lot of pass through are not going to be great places

59:04 to live,

59:05 right?

59:06 So

59:06 in response,

59:07 people are going to sort,

59:09 and there's going to be sorting in the land market in cities

59:12 where

59:13 a lot of these low value places may get lower income communities

59:17 who may not be able to afford the homes

59:20 and the neighborhoods where,

59:21 you know,

59:22 high quality of living or social

59:25 amenities are feasible.

59:26 And the pollutants may also maybe exacerbate pre-existing health conditions.

59:32 So I think it's really important to think about the sorting issues

59:36 as we try to build a robust case on air quality.

59:40 Uh,

59:40 thank you again for these excellent presentations.

59:43 Really appreciated it.

59:46 Thank you so much,

59:47 Shermik.

59:48 Uh,

59:48 I won't pass any comment over to you,

59:51 Rishi.

59:52 Uh,

59:52 we're getting some fascinating insights.

59:54 Thank you.

59:56 Uh,

59:56 thank you,

59:57 Richard.

59:57 Let me just start by saying,

59:59 um,

1:00:00 that really thanking all three presenters for excellent presentations.

1:00:04 I think,

1:00:05 um,

1:00:06 this is an area of,

1:00:08 um,

1:00:08 where knowledge is just emerging.

1:00:10 There's a lot that we don't know,

1:00:13 and therefore I think they have done a fantastic job of getting us,

1:00:17 you know,

1:00:17 up sort of filling us in quickly

1:00:21 on what is it that,

1:00:22 um,

1:00:24 We know and where are some of the unknowns,

1:00:27 uh,

1:00:28 so,

1:00:28 uh,

1:00:28 let me actually begin

1:00:30 by uh recapitulate uh

1:00:32 uh

1:00:34 some of what we've heard.

1:00:35 So,

1:00:36 what do we know about COVID-19 and air quality?

1:00:41 What is known,

1:00:42 uh,

1:00:42 we've heard

1:00:43 a lot of stories.

1:00:45 There's been a lot of newspaper reports that air quality has improved,

1:00:48 you know,

1:00:49 a lot of reports coming out that the Himalayas,

1:00:51 for example,

1:00:51 in South Asia,

1:00:53 are visible for the first time.

1:00:55 So we know that the lockdown has actually improved

1:00:59 air quality.

1:01:00 Right,

1:01:00 so that's an important thing for us to remember

1:01:03 as we're thinking about how a poor air quality

1:01:07 can lead to more infections.

1:01:09 So if air quality is improving,

1:01:11 should we not worry,

1:01:12 right?

1:01:12 Are we OK?

1:01:14 So the other thing that we know is,

1:01:16 as a number of our speakers

1:01:19 have pointed out.

1:01:21 That places with higher air pollution,

1:01:23 a number of studies have come out

1:01:25 to show that there's higher uh COVID infection rates in these areas,

1:01:30 as um,

1:01:31 uh,

1:01:32 you know,

1:01:32 the speakers have told us.

1:01:33 And now,

1:01:34 uh,

1:01:34 the other thing that I'd like to throw into the mix um

1:01:38 uh in today's uh talk is

1:01:40 what um what.

1:01:41 What does this all mean for when countries um grow back,

1:01:46 you know,

1:01:46 should they,

1:01:47 does all of this mean that they should,

1:01:49 uh,

1:01:50 does this give them a,

1:01:51 um,

1:01:52 uh,

1:01:52 you know,

1:01:53 should they pursue green fiscal stimulus,

1:01:55 for example,

1:01:56 to deal with air pollution as they grow back?

1:01:58 What does all of this mean in terms of policy?

1:02:03 Now,

1:02:03 let me also point you to what is not known.

1:02:07 We've heard a lot about,

1:02:10 um,

1:02:10 because we have fantastic satellite data,

1:02:13 we've got,

1:02:14 um,

1:02:15 we've seen these images before and after lockdowns,

1:02:19 where NOX levels,

1:02:21 nitrogen dioxide levels have actually come down dramatically.

1:02:25 But we don't really know um

1:02:28 uh what has happened to the most harmful of air pollutants,

1:02:32 which is PM 2.5,

1:02:35 and um we also don't know at the moment what is driving some of these improvements,

1:02:40 right?

1:02:41 Then as uh both Anna and uh both uh

1:02:45 uh

1:02:46 discussed,

1:02:47 we really don't know what are the mechanisms uh between

1:02:50 that of transmission between air uh air pollution and um.

1:02:56 Uh,

1:02:56 COVID-19 virus,

1:02:57 does it,

1:02:58 does the,

1:02:59 uh,

1:02:59 some people have said that the virus hitchhikes

1:03:03 on,

1:03:03 uh,

1:03:03 particulate matters.

1:03:04 Is that the case?

1:03:06 Or is it that,

1:03:07 uh,

1:03:08 the virus,

1:03:09 um,

1:03:10 you know,

1:03:10 air pollution makes people more susceptible,

1:03:13 uh,

1:03:13 to,

1:03:14 uh,

1:03:14 any infection and therefore it makes it susceptible to,

1:03:18 uh,

1:03:18 the COVID-19 infection as well.

1:03:22 And then of course,

1:03:23 if countries were to think about growing back greener,

1:03:27 uh what does this mean um in terms of,

1:03:30 how would they design uh this kind of fiscal stimulus package?

1:03:36 Now,

1:03:36 as we heard from a number of speakers,

1:03:39 and this talk today is really about air pollution and the COVID infections,

1:03:44 I'd like to um try and summarize,

1:03:46 and I'm not going to do full justice to the wealth of information that was shared,

1:03:51 but there are 3 different mechanisms through which

1:03:54 air pollution and COVID-19 infections can be correlated.

1:03:58 One is transmission,

1:03:59 and I,

1:03:59 as I mentioned.

1:04:01 Um,

1:04:01 uh,

1:04:02 here,

1:04:03 what my understanding,

1:04:05 um,

1:04:05 and talking to epidemiologists and as Anna shared with us,

1:04:09 it is not likely that the,

1:04:11 um,

1:04:12 uh,

1:04:12 the COVID-19 infection is actually airborne or that it is,

1:04:16 uh,

1:04:16 that the

1:04:18 virus itself is hitchhiking

1:04:20 on air pollution partic particles,

1:04:23 and this comes because of evidence coming out of China,

1:04:27 out of Wuhan.

1:04:28 Is showing that most of the transmission happened indoors within,

1:04:33 sort of uh within families and it's not happening

1:04:37 outside and which would mean that it's not,

1:04:39 the virus is not hitchhiking,

1:04:42 but this still um there is still a way

1:04:45 in which air pollution can transmit the disease because,

1:04:49 And,

1:04:49 and especially spikes in air quality,

1:04:51 uh,

1:04:52 sorry,

1:04:52 spikes in air pollution,

1:04:54 uh,

1:04:54 can irritate the throat,

1:04:56 and Bo pointed this out as well,

1:04:58 and coughing

1:05:00 is a way in which you can transmit,

1:05:03 so when you,

1:05:03 you know,

1:05:04 if your your throat is irritated and you cough a lot,

1:05:07 that is a way in which you can transmit,

1:05:09 uh,

1:05:10 so that there is a way in which air pollution can help,

1:05:14 can increase the transmission

1:05:16 of uh the infection.

1:05:18 Then,

1:05:18 um,

1:05:19 it also air pollution,

1:05:21 uh,

1:05:22 uh,

1:05:23 will degrade,

1:05:24 um.

1:05:26 The uh upper airways um

1:05:29 in um uh

1:05:31 uh

1:05:32 in um

1:05:33 in your,

1:05:34 in your nostrils,

1:05:36 and that degradation makes you more susceptible uh to um

1:05:41 getting uh the infection,

1:05:44 and then finally,

1:05:45 as Anna pointed out and um

1:05:47 Bo also pointed our attention,

1:05:49 air pollution increases the risk of a number of uh uh diseases.

1:05:56 And these diseases have,

1:05:58 people with these diseases have also been um have

1:06:01 been found to have higher rates of hospitalization,

1:06:05 um,

1:06:06 so what,

1:06:07 to summarize,

1:06:08 what we do is we do expect,

1:06:10 we do expect that air pollution

1:06:12 is going to make

1:06:14 the COVID-19 infection um worse.

1:06:17 however,

1:06:19 as uh our speakers have told us,

1:06:21 we do not yet have the data.

1:06:24 Uh,

1:06:25 to be able to actually estimate this relationship carefully,

1:06:28 and I'd like to give a shout out to both,

1:06:31 because of all the papers that I have seen,

1:06:34 um,

1:06:35 um,

1:06:36 and as Anna pointed out,

1:06:37 I think some of them are coming out way too quickly,

1:06:41 um,

1:06:41 and some of them are once put out,

1:06:44 are being criticized heavily.

1:06:47 But uh

1:06:48 uh Bo's paper really stands out in doing a careful um

1:06:53 job of both first understanding,

1:06:56 um first controlling for um you know,

1:06:59 whether you have um,

1:07:01 the infection and then given that you have the infection,

1:07:04 are you more likely,

1:07:05 uh,

1:07:06 are you going to see more deaths,

1:07:07 so you know,

1:07:08 of all the papers that we see out there,

1:07:11 Bo has done.

1:07:12 Really a commendable job

1:07:14 of trying to separate,

1:07:15 but I think all in all,

1:07:17 my

1:07:18 uh takeaway from all of this is,

1:07:19 listen,

1:07:20 we expect there to be an interaction,

1:07:24 but it's way too early

1:07:26 for us at this time,

1:07:28 sorry,

1:07:28 I'm getting a call,

1:07:30 I don't know if

1:07:31 um

1:07:32 it's related to us anyway um.

1:07:34 And so,

1:07:35 um,

1:07:37 Uh,

1:07:37 now,

1:07:38 so we know that air pollution levels matter,

1:07:41 but uh.

1:07:44 The question is,

1:07:45 um,

1:07:47 what has happened to air quality?

1:07:48 If air quality has actually improved due to the COVID-19 um uh pandemic.

1:07:55 Do we then care?

1:07:56 Should we be worrying about air quality during this time?

1:07:59 And I'd like to point out to you

1:08:02 that air pollution level,

1:08:03 air quality,

1:08:05 uh,

1:08:05 matters now,

1:08:07 because even despite the lockdowns,

1:08:09 we have not seen improvements

1:08:12 in PM 2.5 levels.

1:08:14 So what I've shown you here is some work that we are doing in the World Bank.

1:08:18 Where you know we're trying to complement

1:08:21 the literature

1:08:22 on um nitrogen uh where we are seeing a lot of um you know,

1:08:28 images

1:08:29 of improvements in uh

1:08:31 nitrogen dioxide.

1:08:35 But

1:08:36 are we seeing improvements in p.m. 2.5,

1:08:38 which are more correlated with health impact?

1:08:42 And here,

1:08:43 um,

1:08:44 what I'm showing you are three different graphs,

1:08:47 one for,

1:08:48 uh,

1:08:48 the Hubei,

1:08:48 where the city of Wuhan is located in China,

1:08:52 then,

1:08:53 uh,

1:08:53 for PM 2.5 levels in,

1:08:55 uh,

1:08:55 France,

1:08:56 and finally PM 2.5 levels in the Indo-genetic plain,

1:09:00 and the green,

1:09:01 um,

1:09:02 is for 2019,

1:09:03 the blue lines are for 2018.

1:09:06 And the yellow lines are for 2020.

1:09:08 Now if you look um in um

1:09:12 for France,

1:09:12 for example,

1:09:13 we see almost no decline

1:09:16 in p.m. 2.5 levels,

1:09:18 even though as Bo had shown up there was decline inOx levels.

1:09:22 So

1:09:23 the pollutant that really matters is not declining

1:09:26 now um in the endogenetic plane,

1:09:29 on the other hand,

1:09:30 we.

1:09:30 We do see a decline in p.m. 2.5 levels after the lockdown,

1:09:34 but what has happened is in India,

1:09:37 um,

1:09:37 in this region,

1:09:38 PM 2.5 levels have declined even,

1:09:41 um,

1:09:42 ahead

1:09:42 of the lockdown.

1:09:43 So we don't,

1:09:44 uh,

1:09:45 and the reason I'm pointing this out

1:09:47 is because the lockdown in some ways is a

1:09:49 way for us to better understand pollution levels.

1:09:52 can you begin to wind up,

1:09:53 please.

1:09:55 Yes,

1:09:56 and so this is my last slide,

1:09:57 um,

1:09:58 and so I think what I'd like to,

1:10:00 you know,

1:10:01 like to sort of leave you with

1:10:03 is that

1:10:04 even though

1:10:05 we've

1:10:06 heard

1:10:06 that air pollution has improved,

1:10:09 um,

1:10:09 with the lockdown,

1:10:11 the pollution,

1:10:11 pollutant that matters,

1:10:13 PM 2.5,

1:10:14 um,

1:10:15 has not necessarily declined uniformly,

1:10:18 so it is still an,

1:10:19 um,

1:10:20 a problem.

1:10:22 Then the uh discussion that we heard from Bo,

1:10:25 as well as from Anna,

1:10:26 and the uh as David pointed out,

1:10:29 there,

1:10:29 you expect to be there to be a relationship

1:10:32 between uh air pollution and uh COVID-19 infections,

1:10:37 although the data at the moment is not there for us to be able

1:10:41 to,

1:10:42 um,

1:10:43 you know,

1:10:44 draw that relationship out definitely.

1:10:46 And so

1:10:48 this,

1:10:48 all of this says that this is a chance.

1:10:51 Air pollution is going to be,

1:10:53 is an issue now.

1:10:55 It's going to become more of an issue as countries lift their lockdowns,

1:10:59 and so,

1:11:00 you know,

1:11:00 this is a time for countries to start thinking about

1:11:04 ways to.

1:11:05 Um,

1:11:06 green

1:11:06 their fiscal stimulus,

1:11:08 and here it is a matter of both,

1:11:11 you know,

1:11:12 putting in measures to reduce their pollution,

1:11:14 but also measures to stimulate

1:11:17 demand,

1:11:17 and I'll stop here,

1:11:18 Richard.

1:11:18 Thank you very much.

1:11:20 Thanks so much,

1:11:20 Ruvishi.

1:11:21 Thank you to our speakers and our discussions.

1:11:32 I'm going to ask these you whatever else comes in throwing it at,

1:11:36 at the panelists and the discussions.

1:11:39 So our first two questions I think are more for

1:11:42 Anna than anyone else.

1:11:44 So the first question from a viewer in Bangladesh asks,

1:11:48 can the virus survive dehydration?

1:11:52 And there's a second question

1:11:54 from someone who read The Economist,

1:11:56 and there was a piece in The Economist that says,

1:11:59 smokers don't seem to be heavily impacted by coronavirus.

1:12:03 Why?

1:12:04 And then there is a third question,

1:12:06 um,

1:12:06 I guess it could be Anna Berubashi as well and anyone else can take it

1:12:10 from one of our colleagues in the bank,

1:12:12 Iwande,

1:12:13 who asks or suggests,

1:12:14 so should we really be targeting PM 2.5.

1:12:18 So,

1:12:18 Anna,

1:12:18 if you don't mind taking at least the first two questions.

1:12:21 And uh and the 3rd 1 too if you want to,

1:12:24 but otherwise we could then take the 3rd.

1:12:28 OK.

1:12:29 So yeah,

1:12:30 no,

1:12:30 the,

1:12:30 the first question's about,

1:12:31 uh,

1:12:32 there's a wider context about how long the,

1:12:34 the virus can survive in different environments.

1:12:36 Um,

1:12:36 I can't give you the exact figures of that,

1:12:38 but I think it's sensitive to dehydration.

1:12:40 It's certainly sensitive to

1:12:41 ultraviolet light,

1:12:43 um,

1:12:44 and then how long it lasts on,

1:12:46 uh,

1:12:46 various surfaces.

1:12:47 There are some studies,

1:12:48 um,

1:12:49 on this,

1:12:49 so it lasts longer on plastic,

1:12:50 for example,

1:12:51 than on paper.

1:12:52 So paper might be a few hours,

1:12:53 whereas plastic might be.

1:12:55 Um,

1:12:55 up to a week or more,

1:12:56 but obviously the viability of the virus decreases,

1:12:59 um,

1:13:00 um,

1:13:00 over time.

1:13:02 So,

1:13:02 uh,

1:13:02 what was the second question about the smokers?

1:13:04 Um,

1:13:05 I groan every time I get that question about the smoking,

1:13:07 uh,

1:13:08 um,

1:13:09 because it's,

1:13:09 uh,

1:13:10 I think if you smoke long enough to give yourself a chronic illness,

1:13:13 you're at increased risk from COVID-19.

1:13:16 Um,

1:13:16 there were some early studies that suggested

1:13:18 smokers were underrepresented in people that,

1:13:21 uh,

1:13:21 were coming into hospital and then some other studies that

1:13:24 suggested there was selection bias,

1:13:26 um,

1:13:26 involved in that and that,

1:13:28 uh,

1:13:28 it wasn't a,

1:13:29 a,

1:13:29 a true,

1:13:30 um,

1:13:31 association.

1:13:32 Um,

1:13:32 but there is some interesting,

1:13:34 um,

1:13:35 uh,

1:13:36 research about the impact of nicotine on the receptor.

1:13:39 There's the,

1:13:39 uh,

1:13:40 ACE2 receptor that the virus enters and suggesting that,

1:13:43 uh,

1:13:43 um,

1:13:44 it might modify

1:13:46 the infectivity,

1:13:47 but I think the,

1:13:48 uh,

1:13:48 I think the message is don't smoke.

1:13:50 Um,

1:13:50 it's not gonna stop you getting COVID-19,

1:13:52 but,

1:13:53 um,

1:13:53 there,

1:13:54 there may be some interesting mechanistic,

1:13:56 uh,

1:13:56 um,

1:13:57 issues to,

1:13:58 to follow up on that around the nicotine impact on the receptor.

1:14:03 Did you also,

1:14:04 I'm,

1:14:04 I'm going to throw a couple more because there's one more which is directed at you.

1:14:09 So it turns out that they wanted you to,

1:14:10 uh,

1:14:10 the person uh who asked the question wanted you to answer

1:14:14 whether 2.5 should be targeted.

1:14:16 And there's also a question about diesel particles.

1:14:20 And the question reads,

1:14:21 we know that diesel particles serve as carriers for carcinogenic compounds.

1:14:27 Even though we know this,

1:14:28 could

1:14:29 COVID virus particles stick to these?

1:14:33 Eg through 2.5.

1:14:35 Yeah.

1:14:36 OK,

1:14:36 so which,

1:14:36 which should we target?

1:14:38 Um,

1:14:38 I would,

1:14:39 I would go for both PM 2.5 and NO2,

1:14:42 and the NO2 because there is,

1:14:44 uh,

1:14:44 mechanistic evidence about

1:14:46 NO2 exposure,

1:14:48 um,

1:14:48 increasing susceptibility to respiratory viruses.

1:14:52 And there is a direct inflammatory um effects,

1:14:55 um,

1:14:56 in,

1:14:56 in the lungs from NO2.

1:14:58 PM10 we have the most uh information about uh impacts generally on the,

1:15:02 on the body.

1:15:03 I mean PM 2.5,

1:15:04 we,

1:15:04 we know a lot about how,

1:15:06 uh,

1:15:06 um,

1:15:07 how,

1:15:07 how bad it is for you,

1:15:08 so I would target that as well.

1:15:10 In a lot of countries,

1:15:11 PM 2.5 and NO2 are very highly correlated because

1:15:14 they have the same source which is road transport.

1:15:16 So actually by tackling the road transport,

1:15:18 then you,

1:15:18 you bring both of them down.

1:15:20 Um,

1:15:21 diesel,

1:15:21 that's a very good point.

1:15:22 Um,

1:15:23 that

1:15:23 nobody's done the study yet on the,

1:15:25 on the diesel as to whether,

1:15:26 uh,

1:15:27 they're more dangerous.

1:15:28 Certainly they look like they

1:15:30 are potentially the more dangerous

1:15:32 components of the particulate fraction,

1:15:34 um,

1:15:35 is diesel,

1:15:36 and that's been part of the reason why,

1:15:38 uh,

1:15:38 um,

1:15:39 there's been a lot of efforts to try and reduce

1:15:41 emissions from diesel vehicles and to reduce the diesel fleet,

1:15:44 on the,

1:15:44 on the roads,

1:15:45 um,

1:15:46 in,

1:15:46 in quite a number of countries.

1:15:48 Can I just while I just talk about PM 2.5 as well in,

1:15:52 in Europe.

1:15:53 So,

1:15:53 um,

1:15:54 in the UK we have had PM 2.5 episodes

1:15:57 during lockdown which have been related to the weather.

1:16:00 So,

1:16:00 um,

1:16:01 we've,

1:16:02 uh,

1:16:02 because the precursors are around,

1:16:04 um,

1:16:04 and the

1:16:05 hot sunny weather that we've had.

1:16:07 Um,

1:16:08 over the initial period of our lockdown,

1:16:10 um,

1:16:10 encouraged,

1:16:11 uh,

1:16:12 the formation of secondary formation of particulates,

1:16:14 and that's why we've seen quite high levels of probably

1:16:17 impacting on France,

1:16:18 um,

1:16:19 as well.

1:16:20 Um,

1:16:21 and I think,

1:16:21 you know,

1:16:22 there are lots of different sources of particulate matter,

1:16:25 whereas the NO2 tends to primarily come

1:16:27 from transport and from industrial sources.

1:16:30 So if you

1:16:30 lock down on those,

1:16:31 then you reduce the NO2 levels.

1:16:33 So I think that's,

1:16:33 that's some of the differences that uh.

1:16:36 Uh,

1:16:36 um,

1:16:37 that,

1:16:37 that have been commented on in terms of what you actually see of levels.

1:16:42 Thanks.

1:16:43 There's 2 questions and anyone from the panel can take them.

1:16:46 The first one

1:16:47 is,

1:16:47 do we have any information

1:16:49 on the impacts of indoor air pollution?

1:16:53 That's one question.

1:16:54 I'm going to throw another one.

1:16:56 What kinds of green stimulus policy should we use

1:17:00 for,

1:17:01 so that we get improvements in air quality.

1:17:04 So,

1:17:04 let's take these two and there's another 2 which are coming up.

1:17:09 Uh,

1:17:11 yeah,

1:17:12 one of the,

1:17:13 so,

1:17:14 um,

1:17:15 I would,

1:17:15 you know,

1:17:15 I'll,

1:17:16 uh,

1:17:17 um,

1:17:18 appreciate if Anna will come in on the indoor.

1:17:19 I haven't seen any studies actually.

1:17:22 Looking at the infections largely because,

1:17:25 you know,

1:17:26 the data

1:17:27 in developing countries where there is more use of,

1:17:31 um,

1:17:31 you know,

1:17:31 solid fuels for cooking and indoor,

1:17:33 so higher levels of indoor air pollution,

1:17:35 the data are just not there,

1:17:38 but one would expect,

1:17:39 right?

1:17:39 As I said,

1:17:40 I think here

1:17:42 we are in a situation where,

1:17:44 you know,

1:17:44 you can't wait for the perfect data

1:17:47 to be able to draw some of these conclusions.

1:17:50 I just don't think,

1:17:51 I mean,

1:17:51 we have enough evidence on the harmful effects

1:17:54 of air pollution and we have enough evidence.

1:17:58 Uh,

1:17:58 of the kinds of mechanisms that are likely to be,

1:18:01 uh,

1:18:02 making the virus,

1:18:03 infections of the virus worse.

1:18:04 So in

1:18:05 high levels of indoor air pollution,

1:18:07 you're going to have a lot of coughing,

1:18:09 you have people living together,

1:18:11 you're going to see more infections.

1:18:13 I mean,

1:18:13 it is,

1:18:14 it makes sense,

1:18:15 and here I think sense has to overtake lack of um data,

1:18:19 so that's,

1:18:20 um,

1:18:20 but you know,

1:18:21 Anna can come in.

1:18:22 In terms of what kind of green fiscal stimulus policies,

1:18:25 and I think this comes back also to what Anna said,

1:18:28 I think it's important to understand the source structure

1:18:31 of uh PM 2.5 pollution,

1:18:33 and what we're finding,

1:18:35 unlike in developed countries where,

1:18:38 Uh,

1:18:38 there's a huge correlation between NOx,

1:18:40 uh,

1:18:40 so nitrogen dioxide

1:18:42 and PM 2.5,

1:18:43 um,

1:18:44 and therefore traffic is the biggest contributor.

1:18:47 In developing countries,

1:18:48 it's actually a very complicated source structure for PM 2.5,

1:18:52 and it's a very multi-sector problem,

1:18:55 so household,

1:18:57 uh,

1:18:57 solid fuel use is a big contributor,

1:19:00 agriculture.

1:19:02 Is a big contributor and not just because of the crop burning,

1:19:05 but also because of the secondary particle formation,

1:19:08 high use of,

1:19:09 we're finding in India,

1:19:11 for example,

1:19:12 very high emissions of nitrogen uh uh of of sorry of ammonia,

1:19:16 from agriculture because of overuse of fertilizers

1:19:19 and um is leading to secondary particle formation

1:19:23 where it interacts with mocks and socks.

1:19:25 So the policies in fact that we need

1:19:28 um cut across these sectors.

1:19:30 So let me give you just one example.

1:19:32 For example,

1:19:33 a green policy,

1:19:34 a green fiscal policy

1:19:36 could be uh removing some of the subsidies

1:19:39 from uh of fertilizers,

1:19:42 because what has happened in India is there's very high subsidies

1:19:45 for uh for um urea,

1:19:48 which has led to overuse of urea in uh agriculture,

1:19:52 which in turn is leading to high ammonia emissions.

1:19:55 And so removing the subsidy will give the fiscal space for governments,

1:20:00 and those subsidies can be used as an economic stimulus.

1:20:04 So you have the stimulus part of it,

1:20:06 and then you're reducing the emissions by reducing ammonia emissions,

1:20:11 and so that is one example,

1:20:12 but happy to discuss,

1:20:13 you know,

1:20:14 almost every sector one can come up with a set of policies.

1:20:17 Thanks.

1:20:18 OK,

1:20:19 great,

1:20:19 Ruvishi,

1:20:20 because you asked another question there which I won't put in front of you,

1:20:23 which was sort of really questioning,

1:20:24 you know,

1:20:25 is the data right?

1:20:26 Is pollution really coming down?

1:20:28 There's a question for David,

1:20:29 but others might wish to chime in as well.

1:20:31 The question specifically is a classic one

1:20:34 that I was waiting to check to,

1:20:35 to come from Fhad,

1:20:37 which is really given

1:20:38 non-pharmaceutical interventions and social distancing are endogenous.

1:20:43 How have you taken care

1:20:45 of all of these endogenous impacts,

1:20:47 you know,

1:20:47 policies,

1:20:47 etc.

1:20:49 Uh,

1:20:49 like NPIs and social distancing.

1:20:51 David,

1:20:52 over to you,

1:20:52 and boy,

1:20:52 you might want to jump in as well.

1:20:54 Uh,

1:20:54 Richard,

1:20:55 can you hear me OK?

1:20:55 I see my images seems to be frozen,

1:20:57 but I'm hoping the microphone is not.

1:21:00 We can hear you loud,

1:21:01 loud and clear,

1:21:02 David,

1:21:02 please go ahead.

1:21:03 I'll let you know if you,

1:21:04 we can't hear you.

1:21:05 Well,

1:21:06 there's a whole suite of technical issues that one has

1:21:08 to address and thinking about estimation in this context and

1:21:11 You know,

1:21:11 Bowe has

1:21:13 touched on some of them on,

1:21:14 on the issue of

1:21:16 how one might control for differential policies

1:21:20 across spatial units of,

1:21:22 um,

1:21:23 for example,

1:21:24 social distancing.

1:21:25 Uh,

1:21:26 of course,

1:21:26 in principle,

1:21:27 the answer is,

1:21:28 um,

1:21:29 one would,

1:21:30 would like to be able to do that in the context of doing work like this

1:21:34 for 3000 US counties or 1500 municipalities in the Philippines,

1:21:38 uh,

1:21:38 the information base is sparse.

1:21:41 To put it mildly,

1:21:42 and so you're really thrown back then on the issue of whether

1:21:45 the errors and variables problem

1:21:47 here might haunt you,

1:21:49 uh,

1:21:49 perversely and looking at the,

1:21:51 the results we've gotten,

1:21:52 that is to say,

1:21:53 could there be

1:21:54 some covariates.

1:21:56 Of distancing,

1:21:58 which also would affect the variables that we're trying to explain.

1:22:01 So that we're getting a biased view of

1:22:03 the effect of the variables that we've observed.

1:22:05 In fact,

1:22:05 let me generalize that.

1:22:07 Uh,

1:22:07 I think,

1:22:08 uh,

1:22:08 a couple of colleagues have

1:22:11 Refer to the errors and variables problem in the measurement of cases.

1:22:15 For uh coronavirus.

1:22:17 In the US,

1:22:17 for example,

1:22:18 the,

1:22:19 there's the New York Times database,

1:22:20 the Johns Hopkins database,

1:22:22 there are other databases.

1:22:23 Everyone knows that there are

1:22:25 uh variations in reporting.

1:22:28 And the same problems apply there.

1:22:30 Now,

1:22:30 econometrically,

1:22:31 there are 3 cases.

1:22:31 The first case is,

1:22:33 uh,

1:22:34 there,

1:22:35 there is a sort of a random incidence

1:22:37 of reporting.

1:22:38 Well,

1:22:39 that's OK.

1:22:39 Econometrics was basically built to handle

1:22:42 random variation,

1:22:43 uh,

1:22:43 in,

1:22:43 in the variable being explained.

1:22:45 That's not a problem.

1:22:46 The Second case would be whether,

1:22:48 uh,

1:22:48 there might then be,

1:22:49 um,

1:22:50 some,

1:22:51 some variation in

1:22:53 Variables that is perverse in the sense that there may be

1:22:57 a covarying factor on the right-hand side of these equations,

1:23:01 which causes an effect to be over or underestimated.

1:23:04 Now,

1:23:04 I think that's,

1:23:05 that would be the question that we need to

1:23:07 look at in the context of whether or not distancing

1:23:10 is a problem.

1:23:11 We have only the roughest of notions,

1:23:13 by the way,

1:23:13 of whether or not

1:23:15 stated policies of states or counties in the US are having any effect on distancing.

1:23:20 Uh,

1:23:20 it seems quite evident that

1:23:22 many so-called red states have many metro

1:23:24 areas where distancing has been fairly effective.

1:23:27 There are many so-called blue states where it

1:23:28 is quite obvious that in some contexts,

1:23:31 distancing has been absolutely ineffective.

1:23:34 And so,

1:23:35 we then have to ask ourselves whether or not

1:23:36 there might be some underlying model of causality,

1:23:39 which would cause some of the characteristics

1:23:41 we're looking at to affect distancing.

1:23:43 The problem I have in trying to do these estimates right now,

1:23:46 and the problem that's kind of frustrating is,

1:23:48 we,

1:23:49 there has never been a test of That.

1:23:50 So,

1:23:50 we have no idea what those

1:23:52 covariates might be or what the signs in the equation might be.

1:23:56 So,

1:23:56 this is all a bit of throwing up of the hands here.

1:23:58 I fervently agree that

1:24:00 there might be an errors of variables problem here.

1:24:03 It might lead to biased estimates of the effects that we're looking at.

1:24:06 Uh,

1:24:07 but we simply at this point,

1:24:08 do not know enough to say anything systematic.

1:24:11 There's one further thing I,

1:24:12 I'd like to add,

1:24:12 and it interests me.

1:24:14 The problem with spatial covariation as Bo has said,

1:24:16 is very important in this context.

1:24:19 And it's not clear,

1:24:19 for example,

1:24:20 in the US that the county is the right unit of observation because

1:24:23 many phenomena are more regionally general and they vary over space.

1:24:27 I didn't have time,

1:24:28 for example,

1:24:28 to point this out,

1:24:29 but in the US determinants of

1:24:31 life expectancy equation,

1:24:33 temperature enters into the OLS version of that,

1:24:35 but not into the spatial econometric version.

1:24:38 The reason for that is temperature is much more widely diffuse

1:24:41 on average

1:24:42 than county level units would suggest.

1:24:44 So,

1:24:45 there really are not that many effective.

1:24:47 Degrees of freedom

1:24:48 as far as temperature is concerned,

1:24:50 uh,

1:24:51 uh,

1:24:51 that we're looking at.

1:24:52 So,

1:24:53 I,

1:24:54 I mean,

1:24:55 I,

1:24:55 I confess to some frustration with this.

1:24:58 Uh,

1:24:58 it's not evident to me that anything that we're looking at

1:25:01 is leading to any systematic bias,

1:25:04 uh,

1:25:04 one way or the other in the results,

1:25:05 but I've readily acknowledged

1:25:07 that we could undoubtedly explain the variation

1:25:10 that remains unexplained if we had better

1:25:13 microinformation about distance.

1:25:18 OK,

1:25:18 thank you.

1:25:19 Thank you very much,

1:25:19 David.

1:25:20 That was very comprehensive.

1:25:22 Bo,

1:25:22 did you want to give a brief answer because there's a couple of other things,

1:25:25 uh,

1:25:26 and if you wanted to have any concluding remarks,

1:25:28 you've got the floor

1:25:30 for a sentence or two to go of,

1:25:32 of your words of wisdom for us to conclude with,

1:25:34 and I'm going to do the same for the others.

1:25:37 I can,

1:25:37 I can add 22 things to what David already said,

1:25:40 and that is on the errors in variables.

1:25:43 So if there's a lot of measurement error,

1:25:46 which there likely is,

1:25:47 then uh you would actually basically uh expect the bias towards zero.

1:25:52 so that will mean that if there's a lot of extra noise,

1:25:55 uh,

1:25:56 in your,

1:25:57 in your data,

1:25:58 uh,

1:25:59 which includes,

1:25:59 uh,

1:26:00 noise that has No correlation at all.

1:26:02 You would actually expect your,

1:26:04 your biases to be towards zero.

1:26:05 So the fact

1:26:07 that you're still finding a very significant uh parameter,

1:26:11 uh,

1:26:11 actually in,

1:26:12 in,

1:26:12 in the situation where you do expect measurement

1:26:15 errors is actually even stronger because you would expect

1:26:17 the results to be more conservative,

1:26:19 uh,

1:26:20 when there's a lot of measurement error.

1:26:21 Second,

1:26:22 uh,

1:26:22 if this measurement error is actually not all too random.

1:26:25 And there's a a a systematic measurement error,

1:26:29 which uh we could also expect.

1:26:31 So if there's a convenience sampling.

1:26:33 So,

1:26:34 for example,

1:26:35 doctors are more aware of uh respiratory illnesses

1:26:39 in areas with high pollution because historically there

1:26:42 have been a lot of patients there,

1:26:44 hence they're more eager to test for corona,

1:26:47 hence you see more uh cases.

1:26:49 That'll be more uh problematic because then you would expect,

1:26:52 uh,

1:26:53 uh,

1:26:53 just purely because of the sampling strategy

1:26:56 to have more cases in polluted areas.

1:26:58 And then I would say,

1:26:59 well,

1:27:00 it really becomes the question whether that uh

1:27:02 as a theory is more far-fetched than simply

1:27:05 following the evidence that suggests that particulate matter

1:27:09 is bad for your health,

1:27:10 period,

1:27:11 and we see,

1:27:12 uh,

1:27:12 a correlation with COVID cases,

1:27:15 uh,

1:27:15 and then it's really saying,

1:27:17 well,

1:27:17 what is more far-fetched and uh I would

1:27:19 lean toward following the evidence right now.

1:27:21 Um,

1:27:22 so

1:27:23 that's just a

1:27:24 quick addition to what David already said.

1:27:27 OK,

1:27:28 great.

1:27:28 Thanks.

1:27:29 Look,

1:27:29 we're,

1:27:29 we,

1:27:29 we're coming to the fact that we've got 3 minutes left.

1:27:32 So,

1:27:33 Anna,

1:27:33 in a minute,

1:27:33 you wanted to come in on indoor air pollution,

1:27:36 anything else?

1:27:36 And

1:27:37 if there's time,

1:27:37 I'd like to give a minute to Ubuhi and to show me

1:27:41 if they want to take a minute to just

1:27:43 any,

1:27:43 any last words of wisdom.

1:27:44 Thanks.

1:27:45 Over to you,

1:27:45 Anna.

1:27:47 Uh,

1:27:48 just to comment on the indoor air,

1:27:50 it's something we don't know very much about in,

1:27:52 uh,

1:27:52 developed countries either.

1:27:54 Um,

1:27:54 I think we neglect it as our peril.

1:27:56 Um,

1:27:57 and one of the things we've been discussing is increased use of cleaning agents,

1:28:00 so we've got more volatile organic compounds and,

1:28:03 uh,

1:28:03 bleach

1:28:04 fumes,

1:28:04 um,

1:28:04 in the house,

1:28:05 uh,

1:28:06 potentially than previously.

1:28:07 Um,

1:28:08 I will

1:28:09 come back to the

1:28:11 issue of,

1:28:12 uh,

1:28:12 what we learned from SARS,

1:28:14 I think as my final comment.

1:28:15 Um,

1:28:17 so I think the positive thing about SARS was that public health measures

1:28:21 helped

1:28:22 conquer the disease before we had a vaccine

1:28:25 and,

1:28:25 um,

1:28:26 before we even genotyped it at that,

1:28:28 at that stage,

1:28:28 so.

1:28:29 Um,

1:28:29 I think,

1:28:30 you know,

1:28:31 we need to keep in mind public health,

1:28:33 just basic public health,

1:28:34 um,

1:28:34 uh,

1:28:35 uh,

1:28:35 measures can be very effective in controlling,

1:28:38 um,

1:28:38 this disease,

1:28:39 and,

1:28:40 um,

1:28:40 that's,

1:28:41 that's where I'll leave it at,

1:28:42 but thank you.

1:28:43 Thank you so much,

1:28:44 Anna,

1:28:44 those are,

1:28:45 those are indeed wise words.

1:28:47 Over,

1:28:47 over to you,

1:28:48 you've got a minute.

1:28:49 Yeah,

1:28:49 thanks Richard.

1:28:50 I won't take,

1:28:51 I just want to say again,

1:28:52 uh,

1:28:52 you know,

1:28:53 I've learned so much from these excellent presentations,

1:28:56 they're very interesting.

1:28:59 Oh,

1:29:00 can you hear me now?

1:29:01 Yes,

1:29:01 we can.

1:29:06 No,

1:29:06 we can't.

1:29:07 You're muted.

1:29:08 Now you're not muted.

1:29:09 Yes,

1:29:10 yes,

1:29:10 now we can hear you.

1:29:11 Please go ahead.

1:29:12 OK,

1:29:13 great,

1:29:13 uh,

1:29:13 just to say that,

1:29:14 uh,

1:29:15 you know,

1:29:15 excellent presentations.

1:29:16 I learned a lot,

1:29:18 but I just want to say that listen,

1:29:20 air pollution matters.

1:29:21 Uh,

1:29:22 we know it's a problem,

1:29:23 but we also need to

1:29:26 point out that it still matters,

1:29:28 and so we don't want policymakers to sort of take their eye off the ball,

1:29:33 and I'll leave it at that.

1:29:35 Thanks,

1:29:35 Urishi.

1:29:37 Show me over to you,

1:29:37 then I'll have to one sentence and we will then call it a day.

1:29:41 Thanks,

1:29:42 everyone.

1:29:42 Uh,

1:29:43 I really found the presentations to be

1:29:46 insightful and providing

1:29:47 at least,

1:29:48 uh,

1:29:48 policymakers in developing country cities,

1:29:51 a lot more information on the transmission parts of

1:29:55 the coronavirus.

1:29:56 As we go forward,

1:29:58 I'd like to make

1:29:59 an urge to do more work on three dimensions.

1:30:03 But

1:30:03 let's try to get a better understanding about land use and transportation.

1:30:07 Because

1:30:08 a lot of patterns we see today,

1:30:09 both in terms of connectivity and the spread of the virus,

1:30:13 but the emissions and the p.m. 2.5

1:30:15 are sort of driven by an underlying economic structural

1:30:18 issue of how land in the city is organized

1:30:21 and it's organized around industrial use,

1:30:23 residential use,

1:30:24 and commercial use,

1:30:25 and how transportation and the choice of transportation modes

1:30:29 link up these places.

1:30:31 So getting a better understanding of land

1:30:32 and transportation would be really helpful.

1:30:35 Second,

1:30:36 I think

1:30:37 we,

1:30:38 in,

1:30:38 in all the work that we're seeing on measures of density,

1:30:42 we should be more sort of proactive and thinking about amenities

1:30:47 and living space and not only physical measures of density.

1:30:51 So let's look at economic geography,

1:30:53 not only physical geography.

1:30:55 And the final part,

1:30:56 I think,

1:30:57 is that it'll be really useful

1:31:00 for the group

1:31:01 in kind of trying to in the short term distill.

1:31:05 So what is it that city leaders can really do?

1:31:08 If a lot of the work,

1:31:10 you know,

1:31:10 will be led by mayors

1:31:12 and the remit of mayors in developing countries is very limited,

1:31:16 what is it that we can really help city leaders do

1:31:19 both on the density issue but also the quality issue?

1:31:23 Versus where city leaders need to work with higher levels of government.

1:31:27 Thank you.

1:31:30 Thank you very,

1:31:31 very much,

1:31:31 Shermik.

1:31:32 Um,

1:31:33 I'd like to thank the presenters very much

1:31:35 and the discussers too,

1:31:37 equally.

1:31:39 And,

1:31:39 uh,

1:31:39 might I say my final comment,

1:31:41 I'm sure everyone that was listening in

1:31:43 has learned something new,

1:31:44 if not a lot new.

1:31:46 And we really have to thank you all.

1:31:47 It was rich in material,

1:31:48 rich in new research,

1:31:50 and we've all come away knowing an awful lot more.

1:31:53 So thank you.

1:31:54 My final point is that

1:31:56 apparently,

1:31:56 the recording of this event and all the materials are on a web page.

1:32:01 Uh,

1:32:01 perhaps there can be an email letting people know what the link to that web page is.

1:32:06 So with that,

1:32:06 we're just 2 minutes over time,

1:32:08 which is not so bad.

1:32:09 Let me really profusely thank our speakers,

1:32:12 particularly those joining from a different time zone in Europe.

1:32:16 And indeed,

1:32:16 uh,

1:32:17 David as well,

1:32:17 even if you're on the same time zone,

1:32:19 Shomi

1:32:20 Uvishi,

1:32:21 thank you so much and thank you to all our listeners.

1:32:23 This was a really,

1:32:25 really excellent session that we had.

1:32:27 Many thanks

1:32:28 and have a great day and a great evening,

1:32:30 everyone.

1:32:30 Bye-bye.

1:32:31 Thanks,

1:32:32 Richard.

1:32:35 You

showAllTimestamps

no

transcript

Right on the dot at 11 o'clock. Here we are. All right. Um, good morning, everyone. Good morning, ladies and gentlemen. Good evening, ladies and gentlemen, wherever you are, and good afternoon, wherever you might be. So, welcome to everyone. Um, today, we're going to deal with a really important topic in this series on COVID, and it's on the relationship and the link between COVID-19 and air pollution. If there's one thing that we know about COVID, it's that we don't know very much about it. And one of the great unknowns about COVID too, are the links that COVID might have with air pollution. So, we hope for some clarity and insight from the 3 speakers that we have today in our 2 discussions, so that I don't interrupt the flow of what we're going to hear today. Let me begin by introducing our 3 speakers in the order in which they will appear. And our discussions. And after, after our speakers and discussions have gone, we will open the floor for questions. You can either, you can either ask your questions verbally or through the chat line and I will forward them to, uh, to, to our speakers and our discussions. So we have 3 extremely well qualified speakers to address this issue. Our first speaker of the day is going to be David Wheeler. David is very well known to the World Bank and he has a very long CV too. Currently, David is a senior fellow at WRI which is the World Resources Institute. Prior to that, he was a lead economist in the World Bank, worked in the CGD, and prior to that, he was a professor at Boston University. And one of the marks I've known David for a very long time, one of the marks of David's work. is that he's always ahead of the curve, especially when it comes to data. Long ago, it was on pollution, then it was on forests, and now David is on to the whole COVID issue. And it'll be wonderful to hear from, from, from David. Then we'll move to Bo Peter Andre, who's currently a consultant in the bank who's doing work on crisis analytics. Bo has had stints not only at the World Bank, but also at the OECD, at the EC, ADB. And at the Dutch government. And he has a book, The Theory and Applications of Dynamic Spatial Analysis. And I downloaded that book last night to read it and I can assure you, it is not bedtime reading. You really, really have to focus because you want to understand about the dynamics of spatial analysis. Um, finally, last but not least, we'll have Anna Hansel, who's professor of environmental epidemiology at the University of Leicester. Prior to that, she held positions at Imperial College and was associate director of the Health Statistics Unit in the UK. And then each of our speakers will speak for about 14 minutes at about a 13-minute mark. I will rudely interrupt you and ask you to wind up. And then we will give 7 minutes to each of our discussant. Our first discussant is going to be ho Mal, who's a lead columnist in the Urban Development unit and a global lead. For anyone that knows anything about urban economics, you will know and have read Schmick's work. He's extremely well published and has led a lot of our thinking in the bank on the economics of urbanization, agglomeration, and development. And, and then we will finally have comments from Urvashi Narain, who's who's also a lead economist in the Environmental and Natural Resources unit at the World Bank, and she's also extremely well known. Especially on air pollution where she's doing incredible things in countries with deep air pollution problems and is our go to person in the World Bank on problems of air pollution. Wherever, she too is widely published in journals like Jean and elsewhere. So, Without further ado, I would like to hand the virtual floor over to our first speaker, who's David Wheeler. So, David, it's all yours and you have 14 minutes and I will interrupt you at the 13-minute mark. Thank you so much. Thank you, Richard, and I'm now going to go to share mode here. And bring up my slides. Uh, thanks very much for your kind introduction and welcome to everyone. Uh, I'm here, uh, really as a reporter on joint work which has recently been undertaken by the Development Research Group and the Bank's Development Economics vice presidency and the Global Practice Group for Social, Urban, Rural and Resilience. And I'm here to provide some introductory context by talking about An enterprise that we began very recently, uh, to look at the question of whether one might be able to model the spread of COVID-19 in a tractable way and in a way that would be useful for developing countries. And the focus this morning, as you can see, will be on the US and Philippines, uh, with this brief, uh, visual timeline of the spread of the epidemic in the two places, but I'll return to that, uh, momentarily. We're trying to respond to a challenge here which is very unusual. As we all know, COVID-19 is spreading very rapidly. Uh, countries need to anticipate how the spread will occur as best they can. But at the same time, for many countries, uh, particularly low-income countries, time is scarce, resources are limited, and there, there really is a need for some methodologies that will help. Uh, to identify places that may be subject to the spread of COVID that are not currently afflicted. Our project then was, was a major challenge for us because it has been formed and implemented very, very rapidly. We only began in March. And our task has been to develop a model that is conceptually sound, but at the same time, simply designed, uh, easy to explain. Easy to transport across regions and income levels, easy to estimate and robust nevertheless, to estimation problems, and there are 3 case studies that we're undertaking for the US, the Philippines, and, uh, we are now moving into South Africa as well. Uh, I thought that, uh, to begin with, I would actually prevent you, present you a visual timeline of the evolution of COVID quickly, uh, in the US at continental scale, in the US state of Georgia. And in Luzon, Philippines. And the reason I'm going to start with this is to give you a sense of the, the central problem that we face in modeling, which is handling what I'll call multi-polar spread. You see on the screen then, a picture of the United States at the county level, the small outlines there are US counties in early March before this problem really emerged uh to any significant degree. And as these slides unfold, you're going to see color scaling going from blue to green, yellow, brown into red, and finally lavender to give you a sense of the severity of Uh, the infection rate in each county, so that what you are going to see now, uh, is standardized by population to make viewing easy. As you see these slides for each case, uh, it's good to look for three things. The first is the appearance of poles for the outbreak, places where things begin, then the extension outward from those poles, and finally, the joining together of geographic extensions into larger patterns. So here we are on March 3rd. We see the beginning of the emergence of some spot on March 17th. March 24th shows us clear emergence of some poles, which then extend outward by the 31st. April 7th, you can see some severely impacted areas forming. Here is April 21st. The process continues through May 12th, and finally, June 2nd, and here, after the process of emergence and extension, you see a joining up where huge regions of the United States have now been affected. Just to reinforce these ideas, let me now go to a uh regional case. This is the US state of Georgia. Uh, and let me begin, um, on March 15th, which is about two weeks after, uh, there was a funeral in Albany, Georgia, which is a circled city in the southwest of the diagram. Uh, lots of people attended, they came from various areas. On the map, you see three circled areas. In the northwest is Atlanta. Uh, in the southwest is Albany, and in the middle is Macon. Atlanta is, of course, a huge transportation hub and very densely populated. Now, watch the evolution and spread of COVID as we go forward into March. You see what happened after the funeral in the Southwest, you see the spread from Atlanta. And now you see uh the virus beginning to spread out. By April 19th becoming more severe. Here's May 3rd when join-ups begins between the Atlanta area and the Albany area. And by June 1st, most of the map has been overwhelmed, including the area around Macon. Now, let's move to Luzon, Philippines. Again, we're gonna start on March, in early March. Here, there was hardly any incidents then. You see 3 circles again, one near Manila, 2 in other regions which will turn out to be affected quickly. We see the beginning spread from Manila. We see it moving outwards from that area. Then we see a seeding occurring in other areas. Two poles in the circle areas then become much more severe and there's continued spread, uh, which then Uh, simply evolves and progresses. So, this is a pattern that we see everywhere, and I'll refer to it as fractal, meaning there's a sort of an eerie replication of this pattern, uh, at continental scale and at regional scale. In all these cases, we see similar patterns, an emergence in several places, a non-uniform pattern of spatial propagation outward. A link up ultimately the fastest propagating salience in those patterns, and finally, a spread into the areas that were initially surrounded by fast moving salient. This has been the challenge for us. How can we model this process in a way that allows us to predict what may happen next? And in this presentation, I'm going to avoid almost all technical detail. There's a huge amount of it, and I'll be happy to go back and answer any questions that you may have uh after my presentation in the Q&A session. But let me just say that there are 3 elements here in trying to model this. The first, as you can see the spread pattern involves interaction with neighboring areas. That's critical. This thing moves from one area to neighboring areas quite inexorably. And in order to model that, uh, interaction between two areas, we've used a variant of the well-known gravity modeling approach, familiar from trade theory, uh, from, uh, other branches of economics and other disciplines, including epidemiology. Uh, in, in many applications. And the basic principle is simple. Uh, in this context, interactions between two areas of people who may be infected are simply proportional to the product of the populations of those areas that are infected and, uh, in, inversely related to the distance between the two populations. But in this case, we're going to innovate a bit, not by using distance, but by using the travel time, which is a much better uh measure of proximity. So, we're going to then include an element which involves interactions with neighboring areas. Another element which involves interactions within each area, and for that, we're gonna use a local measure of population density as our best proxy. And we have to put all this together into a growth model, a model that has growth dynamics. We're going to use the Gompert's model for that. It's a figure model from technology diffusion, uh, from epidemiology, and for population studies. I'd like to touch very briefly on two aspects of this that are relevant. On the interactions with neighboring areas, uh, there was a challenge in trying to quantify travel time in this context. And without dwelling on this too long, I'd like to advertise a bit for a wonderful new resource that we use that's available for everyone in the world now for free, called the open-source routing machine, which allows you to use OpenStreetMaps to compute travel times between any two points for free on a massive scale. For this particular application, we had to compute hundreds of thousands of travel times. Uh, fortunately, for replication of this approach, you only have to do this once and we did it all for free. The basic idea is for each area. Uh, you take all areas with which it might feasibly interact, and we set a radius of 200 miles around the area. And then you calculate travel times to the centroids of, in this case, other areas, counties in the US, uh, and, and that gives you the domain within which you do measurement. Down below, you can see, uh, the travel time radii that are color coded from 5 US counties and 2 areas in the Philippines, and of course, depending on road conditions, travel times can vary a lot. Uh, once you have this, then you we the infections in each neighboring unit by inverse travel time and add all those up for each unit or each county in the US case, for example, to get a total measure of interaction. So, that's one of the terms in uh the modeling approach. On the growth modeling, this is my only brush past mathematics, and I'll be very quick. Just one point I wanted to make. The Gompert's function is quite useful in this context. It provides a pretty good fit to actual data on the spread of infections. You can see on the right there, uh, an exemplary picture of South Korea. And the reason I wanted to bring up the math is that To highlight that red alpha, which is there in the middle, which we will see again in the econometrics. The basic idea behind the Gompert's curve is quite simple. It says that the rate of change of something is proportional to the gap between its ultimate destination and the current state. So, the bigger that gap, the faster the growth will be. Having said all this by way of introduction, then, uh, I'm gonna pass immediately to some econometric results which we got, uh, which suggests that in fact, this very simple model, two modes of interaction and a, and a growth, a modeling context give you fairly powerful results. So, this data, It involves both the US and the Philippines. We're looking at changes from March 17th to April 14th. We have the three terms I've already discussed as variables in this model. The first is the log of the initial infection rate, and the, this is basically minus the value of alpha from the mass or Doppert's curve. We have the travel time weighted. Measure that I had, uh, discussed. And finally, we have local population density. In the case of the US and Philippines, you could see by the standard canons of statistics, these are really quite powerful results. They're very robust. Uh, in each case, the signs are right. Uh, in the case of maximum population density locally, uh, the two coefficients are even very close to one another. For the others, they differ, uh, and that's something one could discuss. But in any case, the model seems robust. I should point out one. Uh, side issue here, as part of fitting this model, we had to use a grid search algorithm to fit the parameter for travel time. Uh, in this relationship by which if you have an infection in an in, in, in a, an interacting area, you wait by inverse travel time. That isn't necessarily travel time, uh, to the first power, just dividing by T as you can see up above. It might be a different power. The only reason I'm bringing this up is that it makes a big difference. We got an exponent of 0.7 and the green line, uh. In this graph shows you that in fact, that gives you a much broader reach for this infection than you would get if you were dividing by travel time alone, that's the blue line or the square of travel time. Uh, that's the, the reddish brown line. So you can see that an implication of our result is that the influence of, of places that are fairly far away is still significant in determining what happens in a particular county. All that said, let's move to the central question here, which is, uh, what are the predictions look like from this model? In order to set this up, I predicted outside of the sample in the case of both the US and Philippines. By that, I mean that we start as if, uh, we're in April 14th and all we know is what we have learned up until April 14th, and we forecast 28 days ahead. Then having done that, we compare that with what actually happened to see how well we do. In order to illustrate this in a very simple way, uh, I've divided the predictions into three groups, the lowest 10th percentile, the middle 80, the middle 80th percentile points, and the highest 10th percentile, or the 90th percentile or higher. Uh, the graph here shows you on the vertical axis, the actual infection growth, and uh, on the horizontal axis, we're showing then the, the, uh, Predictions offered by the model, uh, at the lowest, the medium, and the highest levels. And what this says is simply that we do an awful lot better. Our prediction is that growth is, uh, much higher in counties that in fact do grow more quickly. And as you can see in some cases, this is fairly spectacular. I've isolated um the case on the right, because that's uh the most spectacular of all. The peas in here are, are different population groups, which are defined above P1 through P4. P1 is the largest cities, P4 is the smallest. Turn to the Philippines, we see the same thing again. Uh, as we move up through the prediction percentiles, we see actual growth going up. And again, what these results suggest, as they do in the case of the US is that, uh, in the highest 10th percentile, we actually go a long way toward identifying areas that in the next period of time, in the month following the modeling exercise, we will see rapid growth of the virus. So this model seems to be pretty robust. Having gone all this distance, let me then stop, uh, and extend a little bit. We want to extend the model and look at some other measures that might be relevant. Uh, and one of those is vulnerability, and here I join my colleagues briefly in, uh, providing a very quick introduction to topics that Bo and Anna will then cover in much more detail. One of the things you might want to look at is vulnerability. And in this particular case, life expectancy uh might be a measure that you wouldn't want to look at in that context. This is a regression result for all the covariants of life expectancy that enter into these discussions. Uh, there's no, I don't want to dwell on this. I simply want to point out that many of the covariants that have been identified with uh the spread of PM10 and vulnerability are here at present for life expectancy, and I get extremely strong results for PM 2.5 as one of those covariants. Now, that in itself, uh, this, this, uh, finding of covariation doesn't necessarily mean that we have causation here. However, I wanted to turn quickly before I close to this one graph. This is built up from the census tract level in the US, very detailed data that shows something which is quite remarkable. The red here, uh, denotes counties which are most afflicted by 2.5, and this is to an almost eerie degree, a picture of coal fired. power in the US, particularly in Georgia, there, the, the red zone areas there are some of the biggest coal-fired power plants in the US. So it's clear that pollution generally defined seems to have a, a measure of correlation that's fairly significant with life expectancy. That said, I've gone back to the original model and put in the life expectancy begin to wind up, David. Sure, I'm, I'm, this is effectively wrapping it up. Uh, and what we find is that they have, the results have the expected signs. They're not tremendously powerful statistically. I don't think they add too much to our predictive power, but they certainly suggest that there is a correlation here beyond the part that is explained, uh, by the initial model that I talked about. So just to summarize then very quickly. The challenge was very rapidly to put together a model of COVID spread. We put together elements including a gravity model of interactions, local interactions in the highest density areas, used the Gomert's formulation to uh characterize change. We got similar strong fits for both the US and Philippines, a similar pattern for the extent to which travel time affects the influence of infections. The, the forecasts look promising. It looks like they do differentiate subsequent experiences of cities, and we're now undertaking some extended experiments with the model. You've seen a quick illustration for the US. We're working on Philippines and subsequently we'll work on South Africa. Let me just close there. Thank you so much, David. That was really fascinating. Thank you as always. Um, Bo, over to you now, please. Thank you so much. I'm gonna share my screen as well. Here we go. All right, thanks everybody, and thanks, uh, uh, Richard for your introduction. Um, like you said, my, uh, my background, uh, is, uh, is, is quite technical. I did my PhD in econometric theory, mostly working with, uh, processes that evolve over space and over time, and that's really the angle, uh, from which, uh, I come at this. Um. I work in a, in a unit strategy, analytics, finance and solutions and knowledge in the World Bank, and what we're doing is, uh, we're working on crisis analytics. So whenever there's a crisis somewhere I try and put together, uh, uh, the relevant data set to, to come at some, some useful policy advice and guide through that, uh, and it's quite unnatural to have a crisis at large, but, uh, still, uh, it's, it's the angle from which I, I come at this. Um, so I started thinking about this actually early in February and, uh, there was, uh, the, the first signs of, of how this virus spreads were, were quite, quite a, a, a positive picture. Uh, uh, the first tense was that this virus was not airborne, uh, and that airborne transmission was not believed to, uh, to explain a substantial part of, of the virus spread dynamics. Then, uh, in March, uh, uh, an important paper by Van Dormael and painted a slightly darker picture saying, well, actually, we have some evidence that suggests that SARS-COV-2 can survive up to 3 days on some surfaces and, uh, concluded that aerosol transmission might actually be feasible. So, uh, from that perspective, uh, if this virus will be, uh, airborne, then you might expect similar dynamics as we know exists with other airborne, uh, viruses. So there's a quite a huge body of literature on this, and, uh, the general takeaways is that for various classes of, uh, viruses that are airborne, uh, we know that the presence of fine particulate matter, um, uh, actually increases the infection risk. So, um, if that holds true, then we might expect some similar dynamics, uh, uh, in, in COVID-19. So when I started looking at the data early on, I quickly concluded, um, that, uh, uh, in at least in the Netherlands, uh, the case, cases seem to double in areas where, uh, pollution concentrations increased by just 20% above guidelines. We're gonna talk a little bit more about, uh, how we got there, um. But I want, wanted to show a quick picture, uh, uh, to everybody to show that we can, uh, quite well see, uh, uh, pollution from space. So it's quite easy to track, uh, uh, pollution distributions, uh, across countries and within countries. So if we just have a, a general look at the situation in Italy, which was, uh, really, uh, escalating, uh, around the time when I started, uh, thinking about this problem. Uh, and we see that in the north, uh, the nitrogen dioxide, which is just one pollutant, but, uh, one for which, uh, data is quite easily accessible, we see that, uh, that the highest concentrations are really centered in the north, and then if you just pull up, uh, a map, and this is from, from, from early April, then you can see that the case densities, uh, seem to, uh, have some similarity in how they are distributed across the country. Um, the situation has seemed to be quite stable. This is data from, uh, from last week, uh, in the middle again, uh, the cases on the left, the same pollution map, and then on the right, I just quickly pulled up a population density map, uh, just to show that actually by, uh, briefly looking at the data, the similarities between pollution and case densities are much more striking than, uh, than, uh, with population density. So that begs the question, uh, what, what is driving this, you can also look at, at some other countries in France, you see that, uh, the highest case densities are in the northern part across the border with Belgium, Germany, and that's generally where you see a lot of cases. And then if you start looking across Europe, um, And this seems to be a recurring pattern, but actually not in all places. Um, first, I highlighted the Netherlands, which I personally analyzed and we'll talk about the details there. In the bottom you see France, uh, and Italy. Um, I'm putting, uh, on the right, uh, uh, the case distribution in Poland. Uh, just to point out that in many places, the regions are actually quite, quite big, so even though there seems to be some similarity, uh, with the pollution distribution, it's very, very hard to, to draw any solid conclusions here just by the fact that these regions are also large and we only have a few of them. Uh, then in Austria, we actually see a different pattern. Uh, actually the highest case densities are, uh, in places where pollution is quite low, but we know that, uh, uh, this country has, has mountains, uh, so the terrain is, is completely different from one place to another, so there could just be, uh, other factors than, than pollution that are important. And then, uh, in, uh, the United Kingdom, England, uh, you, you see that it's not at all that clear. You see in the, in, in the south part, a lot of pollution cases seem to be a bit higher, but then in the north, you also see a lot of cases per population, and that just shows that these areas are all quite large, uh, and they don't have a lot of people, um, so. Uh, it's not actually all that clear, uh, how the relationship with population comes into play here. So apart from just an obvious similarity, there's a lot of other factors that you have to think of. Here, I'm just quickly showing the same pollution map that David, uh, pulled up in the previous presentation and then overlaying it with, uh, with, with cases just to again see the, the striking similarity here. So when it comes to thinking through what's behind, what's what's behind this, uh, this correlation, there, there could be many explanations. Uh, there could be actually direct relationships and, uh, uh, Ana will talk a bit more about that. But this would have to do with that pollution, uh, increases susceptibility to the virus, uh, through health impacts or it might even be the case that, uh, the virus, uh, attaches on or suspends into aerosols and then uh that particulate matter, uh, helps the spread of the virus. There could be other forms of associated relationships, uh, meaning that air pollution could just proxy, uh, various risk factors, and this could be quite broad. It could mean that pollution tends to be produced in, uh, in working environments where a lot of people work in cramped conditions, and it's really just the fact that there's a lot of people working in cramped, uh, locations that increases the transmission, um. And then finally, there are of course completely different uh explanations to these correlations which could be just incorrect, uh, spurious trends, it just happens to be that pollution uh trends in the same way and uh there could be other confounding relationships, meaning that, uh, it's really population density that proxies for activity and then you expect both more cases and more pollution. So some of these factors, we can treat them in a statistical method. Um, but in the end, uh, we can exclude some of these explanations, but will always remain, uh, uh, that, uh, even if we can show that there's a robust explanation, these type of analysis that at least I'm gonna show, uh, are not intended to really explain, uh, why you see this, uh, uh, this correlation. So just uh the analytical strategy, uh, to summarize, we're gonna need granular data on uh COVID-19 cases and very precise air pollution measurements. This already puts a lot of constraints on, on what, what, what areas we can study rapidly. Ideally, we would have patient level data, but this is not available everywhere or not always open. Uh, we're going to need reliable data on a number of control variables that have to do with health pre conditions, which are certainly not available everywhere, and then we're also going to need a reasonable sample size or at least very small area district, uh, this units. Uh, so one of the, the places that kind of have this, uh, at least early on, uh, uh, making that data available is the Netherlands. So I analyzed 355 municipalities in the Netherlands, which are all quite small. Um, I looked at different, uh, dependent variables, confirmed cases, uh, per per capita with confirmed hospital admissions per capita. I also did, uh, case counts, um, and then I used a variety of air pollution, uh, data sets. Uh, 2 based on ground measurements, PM 2.5, PM 10, and I also looked at remote sense data. Um, generally, these considerations don't really change the conclusion, which is good. Uh, then I looked at a number of control variables including population density, pre-existing, uh, health conditions, uh, and, uh, uh, proxies for case severity, use a number of demographic, uh, data sets including age, household composition, uh, And then uh in terms of methodological considerations, uh, I looked at non-linearities, uh, I looked at alternative distributional assumptions, the influence of outliers, um, uh, using a variety of tools to control for spatial dynamics and so on. You can read the details in the paper. I'm just gonna present you the main simple, uh, estimation results as most of the more sophisticated considerations do not really seem to impact the main conclusions. Uh, here in this table, I'm presenting very simple linear regression results. The first model includes, uh, all the control variables that I could get my hands on, which are, uh, 22, uh, ranging in, in the categories that I just mentioned. Uh, in this model, we see immediately that PM 2.5 is an extremely significant and a strong predictor. Uh, every point in PM 2.5 seems to raise the number of cases, uh, per 100,000 by 10. And at this point in the data, the average was about 20 cases for every 100,000 people in these areas. So, uh, meaning that a 2 point increase seems to correlate with about a doubling on average. Um, then the other models, uh, they take into account spatial dynamics, uh, spatial dependence, you see that the, the parameters, uh, value changes a bit in magnitude. Uh, but those familiar with these types of models, uh, when, when you're modeling spatial feedback, there there's a multiplier effect. Uh, so generally if you take into account, it actually seems to suggest the same, uh, relationship linearly, doubling in cases when, uh, pollution increased by 2 points, um. What I, uh, then did is, uh, I looked at the nonlinearities, uh, uh, in this relationship. There's a lot of factors that might actually influence or might be important when it comes to the relationship between pollution and, and COVID cases. Uh, there could be, uh, weather variables, uh, that influence how impactful pollution actually is. Uh, there might also be economic, uh, uh, conditions that, that, uh, increase the vulnerability. Uh, uh, of people. So, um, for example, in poor regions, people might not be able to, uh, protect themselves all too well against pollution, hence the impact from pollution onto COVID, uh, uh, might be higher. So there's a lot of reasons to even in this, uh, in, we're thinking about a simple correlation, there might be a lot of influencing factors that, that Uh, that, uh, that determine how this, this correlation actually looks from one place to another. So I used the nonlinear regression, uh, method, use a number of, uh, important control variables, and I also used X and Y coordinates. So effectively, uh, this type of regression, uh, uh, actually allows you to model the relationship between COVID and air pollution. In such a way that this relationship varies from, uh, across levels in the data and in this case also explicitly from one location to another. Then, uh, I did a simple prediction exercise, um, predicting, uh, what, what the expected or at least from, from the model's perspective, how the case distribution looks like, uh, when, when areas at about 10 p.m. Levels and then increasing it to 12, uh, calculate the, the change, uh, and then you see that, that, that this, this, uh, this effect differs, uh, from one area to another, so you can draw a bit of a distribution and then you see that there's a lot of heterogeneity, uh, in, uh, in, in the increase, uh, in cases. So again, it seems to center around, uh, a doubling in in cases when pollution increases from 10 to 12, uh, points. But, uh, this relationship really seems to vary from one area to another. This is also important because, uh, at least from the World Bank's perspective, we're, we're pretty much concerned about developments in, in poorer countries, and these tend to have higher pollution concentrations, um, and, uh, just this heterogeneity in, uh, in a very small country like the Netherlands already shows that it, it's not, not that trivial, uh, to extrapolate. This relationship to to countries where the conditions might be, uh, might be different. So that's kind of like brings me to a bit, a bit of a summary of the main takeaways for, from, uh, from the analysis, and that is that PM 2.5 is a, is a well-known health risk factor and we already know that short-term exposure to pollution has shown to decrease infection risk for various viral infections. Uh, this is my last slide. Uh, and then emerging analysis, uh, across the board seem to point toward PM 2.5 as a strong predictor, uh, of COVID-19 incidents, and that seems to be robust to, uh, uh, a number of control variables and different methodological approaches. There's some cautionary notes in order, uh, I would say, and, uh, first, these type of analysis, uh, they're not intended to explain why, uh, or how this correlation works. Uh, we also know that the available data on COVID-19 is at best sub-optimal, um, and, uh, we don't yet know from just analysis in a country like this how this relationship extrapolates to higher PM 2.5 concentrations. Uh, a systematic review on the applicability of these findings across countries, uh, would be highly beneficial at this point. Uh, and as most of this literature is, is still in early, uh, stages, uh, I'd really caution everybody on, on how, how, how to interpret these results. Uh, so, uh, I'm happy to take a lot of questions in the end. Uh, I'll keep it at, uh, at this for, for now. Thank you all so much. Thank you, thank you very much, Bo. That was fascinating as well, in particular the quantitative magnitude that you've come up with for the Netherlands. Um, and, it's over to you now to give us some insight as economists as to what's really behind it. Uh, teach us some epidemiology of what's going on here. Thank you. Over to you, Anna. Thank you, I'll just, uh, set up sharing my screen. So, uh, can you hear me? Yes, perfect clarity. Yes, yeah, fantastic. So I started my career as a respiratory. Into public health and for the last 20 years or so, I've been working in environmental epidemiology, primarily in air pollution but uh other uh types of uh environmental exposure. So I'm going to give a talk very much from an epidemiological and public health perspective on this. Um, so it's a, a, a different starting point from, um, our previous speaker on this. Um, but I will start with a similar, uh, uh, slide here showing, um, the air pollution levels in China and Italy. And these were two of the epicenters of COVID-19, um, or two of the initial epicenters, and it wasn't very long before people said the air pollution levels are high here and in Northern Italy, we've seen high amounts of COVID-19. Could these potentially be related? And there have been um a number of papers um that have been published and more er er opinions that have also been put out. I've picked out this particular paper um from um which was one of the first um which was er um for Europe, which was er er suggesting that the highest levels of COVID-19 deaths were in the most polluted regions. So uh that's throws up a hypothesis, but it's a correlation. And um you need to look beyond that to see what else is going on because um these areas with a high air pollution level tend to be the areas with the high population density. They tend to be very well connected areas, so they're where the infection came in first, got a hold spread um through the the density of the population. Um, they also may have uh um areas of deprivation. Um, and that, uh, in itself is a, is a risk factor. So I think to be able to understand, um, the relationship, uh, or potential relationship between air pollution and COVID-19. Need to, uh, first start off thinking about the epidemiology. So I've got some figures here that uh come from the UK um there are similar types of figures that are coming out in terms of COVID-19 mortality in relation to age. So age is a really big uh risk factor for COVID-19 as is, uh, being male. Um, but that, those risk factors only seem to come into play, uh, for severe disease when you're over about the age of 40 unless you have other underlying conditions. So that means if you're a developing country with, um, a, a large percentage of your population in younger age groups, um, it probably means that the risk of, uh, death and severe complications, um, is, is much lower in your population if you have a, a younger population. So, um, the second thing to comment is just about this top right hand side about the usual all cause mortality rates. So COVID-19 is a new disease, it's not behaving like, uh, some other diseases. So pneumonias and, uh, uh, various other respiratory diseases usually have this J shape relationship with, uh, mortality with a, uh, a high, uh, higher impact in those less than 1 year old in babies. Um, we don't see that or we haven't seen that with COVID. And then thinking about the risk factors, um, people think of it as a respiratory disease, but actually, uh, the comorbidities that are showing up really clearly are cardiovascular ones, hypertension, um, heart disease, also, uh, metabolic disease such as diabetes, and, um, in fact, certainly in the UK our asthmatics have been underrepresented in those with the more severe disease, which is interesting. It's a systemic disease. I think we're finding that the severe. Uh, uh, COVID-19 is not just impacting on the lungs, it's, uh, affecting the blood vessels as well, and that might be important for long term sequelae, um, of the disease and, um, implications for follow up. So I've put other risk factors here, obesity and non-white ethnicity in Western countries. Um, so whether, uh, people who are, uh, from Asia, South Asia and Africa, um, experience, uh, higher rates in Asia and Africa, I think we still have to, uh, you know, we still need more information on in the Western countries, these non-white ethnicities tend to uh be more deprived. Um, they live in the, um, uh, very densely populated, uh, areas, and, uh, they have higher rates of comorbidity which might well be related to the deprivation, and it's very difficult to untangle, um, that, uh, those, uh, all those factors. So, um, how might air quality impact on COVID-19, and I think it's worth thinking through what mechanisms might underlie any, uh, uh, observed relationships. So I think an obvious one is that the, uh, air pollution impact is indirect. There's lots of evidence that air pollution increases the risk of chronic disease, and we know that chronic disease gives you a risk of more severe COVID-19 and death from COVID-19. And if that's the case, and the only mechanism, you might expect the risks to be of a similar order of magnitude to previous studies looking at air pollution and mortality. There are some other potential mechanisms which I, I, I think are more short-term and direct, and they would be around increasing infectivity. So, um, that might occur if the virus is er being carried on particulates. It might occur through inflammation of the lung, er, from air pollution which we know is a a a feature of air pollution exposure, um, or it might occur through some specific mechanisms. I've got a slide on that. Um, so that would give you, uh, an interaction, so you would see higher infection risk in polluted areas. So, uh, they, you wouldn't, you'd have higher risk than you see with the previous studies of air pollution and mortality. Um, you might also have, uh, an impact of air pollution that if you have severe, uh, if you have, uh, an infection, it might make the infection more severe, which might be through these sort of general inflammation, uh, mechanisms, and we know that, uh, air pollution can cause inflammation through body systems. So I'll just tackle uh a couple of those uh uh direct mechanisms. So um as Bo uh uh uh talked about, there are some studies that suggest that you can detect coronavirus on particles of air pollution. Uh, there's not a lot of studies around. Those studies don't tell us whether uh the virus uh material that's detected is viable nor whether uh you can pick up enough from the air um to get an infected dose. I'm not sure we actually know at the moment how much virus constitutes an infected dose. And I think the, the role of aerosols, um, in terms of infection is not very well established. We know that it's a droplet. Uh, transmission, um, and if those droplets contaminate surfaces, you can pick them up from surfaces. Um, aerosol transmission is probably a small risk factor, but we don't know how small it is. And, um, it's, it's quite important I think to get a handle on that because it would affect our social distancing policies, uh, right through to is it safe to use a public toilet, for example. And that's, uh, um, some of the things we're grappling with in the UK at the moment uh as we're coming out of lockdown. So there's another uh mechanism which is uh really interesting from a medical perspective. This is a busy slide, um, just concentrate on the middle bit here, we've got a red uh SARS um uh COV-2 virus here and this is the ACE2 receptor in green, so the uh virus, coronavirus attaches to this receptor and that's how it enters the cell. And uh there is evidence that the uh that both NO2 and particulate pollution can up regulate, increase the numbers of these ACE2 receptors. Um, what we don't have is direct evidence of, uh, um, air pollution levels plus virus, um, and some cells, um, in a, uh, a mechanistic study to say does, uh, the virus, uh, plus air pollution give you a more sort of potent, uh, risk of infection. Um, than, uh, not having the air pollution, and I think we urgently need that, uh, to be able to understand some of the epidemiological studies. So there are a few epidemiological studies and as Bo said it's quite early days. Um, I think one of the problems we've got is where do I look? So, uh, my look didn't actually turn up in your paper, I'll apologize for that. So, so, um, I tend to look in medical studies on. PubMed. There is a WHO database that, uh, is, is sort of a, I think PubMed Plus. Um, you can also look on Med Archive, which is, uh, for pre-prints, but it's, I put a question mark there because they're very variable quality and, um, and there's a huge amount. So people are just putting things up there. They get a press release, they come out in the, in the, um, um, in the media and you're having to react to, to them, um. And um there's now concern in the, this is from the British Medical Journal, um this article here saying that er we're actually getting perverse incentives to publish early and possibly too early so that er um there's too much data and not enough information. We have another problem about outcome measures and um so um uh I think deaths are well there's a there's a whole issue about uh how you interpret deaths, excess deaths, deaths from COVID-19. They're all defined differently and in different countries. Um, the thing that is really problematic is a case, um, because a case, um, uh, is dependent on testing and the testing, who gets selected to be tested has varied, um, within the areas in within countries, um, and between countries and also has changed over time, so. Um, you know, it's been changing in some cases every week as to who's, uh, eligible to be tested, um, and there are very few countries that have tested, uh, large numbers of the population. I think South Korea and Germany stand out there. Um, I think we've got another challenge which is around the methodology, um, about disease propagation and the previous, uh, talks, I think of trying to get a handle on this, um, but we're seeing this sort of pulse of infections go out, um, and that's not the type of methods that we've developed to look at these short-term, um, interactions, um, and then some of our control measures are impacting on the air pollution as well as the transmission. And it then gets very tricky to start unpicking um what the true associations might be. I'm gonna focus on two, I think quite, uh, authoritative studies or certainly ones that have provoked a lot of discussion. Um, and they're both from the States, um, in part because I think you have, uh, good access to data. Uh, one of them came out in April, the second came out in May. Um, they're similar but slightly different designs and the results are slightly different. And um they're both from very well respected um air pollution epidemiology groups. Uh, the XX uh Francesca Dominici group is at Harvard, um, and the uh Dong Haiang and Ha Chang are um at Emory, but there's also some Harvard. So, um, this slide picks out some of the, uh, key, um, aspects of these studies. So the colors are not meaningful. They're just so that you can pick out both. Um, the, one's using death counts, uh, and the Liang study is using deaths and cases. I'm just going to concentrate on the mortality here. Um, they're both looking up to, uh, nearly the end of April. They are area level studies. They're comparing US counties. Um, for some reason, there are differences in the number of deaths picked up between the different databases. They both use very well established and good air pollution models um to uh give you the, the air pollution levels, um, but the wall paper, the first paper just looked at PM 2.5. The second paper looks at NO2 and ozone as well. Slightly different years, but they're looking at long term averages, um, in relation to, uh, COVID-19. So there are some differences in the models that they've used, um, these statistical models. Um, so the second paper has used a zero inflated model because um a lot of the counties had zero deaths during this time period. And I think the other major difference, and we have just touched on that is that the second paper also adjusted for spatial autocorrelation and probably explains why there are uh uh lower uh results found. These are really impressive amounts of analysis, um, with, The wallpaper, they found an 8% increase in mortality rate per 1 mcg per meter cubed increase of PM 2.5, and this is adjusted for about 20 different confounders, including attempts to look at the state of the progression of the pandemic, uh, infection rates, and population density and various population vulnerability factors. Um, the Leanne paper finds, uh, an association that's not statistically significant with PM 2.5, and it's about a third the size. Um, it does find a significant association with NO2 and it expresses that per interquartile range um of the NO2 exposure and there's no association with ozone. And then both papers give us some. Uh, uh, public health type interpretation and the thing that made me sit up and notice on the Wu Dominici paper was that the coefficients were actually higher than those for all cause mortality that they'd found in the previous, um, analysis. And uh I think the Lianne paper commented it's around about 7% of this might have been avoided if you had lower NHs. I think the other thing to note is the P.M.2 levels were quite low here. They're not comparable to what you'd see in a developing country. Um, In terms of the inference of this, um, we can come back to this in questions probably, but, uh, epidemiological studies, you want to look at the totality of the evidence and interpret in the light of what you already know, um. Hi, you're giving me a minute to go. That's great. Um, I will just say that the ecological studies are often used in the initial assessments and they're very, uh, good if you want to get a quick, uh, answer, um, for, uh, for, uh, these types of questions, but, uh, you probably want a long term individual level could study uh to help you with that. Um. I think it would be surprising if we didn't see a link between air pollution and COVID-19 given what else we know about air pollution and COVID-19. Um, I have got some couple of slides here just on public health perspectives about what we can learn from SARS. I can come back to those, um, in, uh, sorry, I can come back to those in questions if you'd like to, uh, know a bit more. Um, but, uh, in conclusion, um, the air pollution, we already know is associated with risk of chronic disease. And um. Also with mortality, it's unlikely that we don't also see a link with COVID-19. We're early days in interpreting the evidence. I don't think, and also the Committee of the Medical Effects of Air Pollution, which I sit on in the UK doesn't feel that the evidence is strong enough to get specific measures against COVID-19 in high air pollution areas that are different from low air pollution areas. Um, I think there's various gaps that we need to fill in to understand this relationship better. And I think one of the positive things uh just from a public health perspective is that if you have a well functioning public health system, you can uh achieve low disease transmission. I think that's really important to just hold in mind when we're thinking about the uh um air pollution levels that actually public health measures really important. Don't lose our focus on that uh even when we're thinking about other risk factors. I'll stop there. Thank you. Thank you, thank you so much. Um, without further ado, let's go to our discussions. And our first discussion is, uh, Doctor Sherm Lowe from the Urban Unit in the World Bank. Shermick, over to you. Thank you very much, Richard. And uh I really appreciate the analytic work and empirical analysis that was presented this morning that tells us a little bit about the pathways for the spread of coronavirus and the COVID-19 disease. I think, to me, at least from what we do at the World Bank, this work will be very useful in the near term, as cities in many developing countries. are really trying to prioritize how and where to allocate medical and civil resources to save lives and to save livelihoods. In particular, I really like the way both David and Bo have chosen to be very spatially aware in their approaches to examine the spread of the virus. And I think, David, your modeling draws on a large body of work in economic geography that relies on market access and gravity-based approaches to measure connectivity. And this is really apt for measuring contagion. And Bowe's work does a really nice job in drawing on the spatial econometrics literature pioneered by the likes of Luke Ansellin to take into account of spatial trends and spatial clustering. And such a spatial approach is really important because much of the work I see is spatially sort of blind. And this makes the work very robust and uh uh and it improves our understanding of the correlates linked to the spread of the virus. My one suggestion here is on both the papers, we may want to address the so-called reflection problem that's often seen in spatial models of social interaction. And here, infection rates of individual zip codes and neighborhoods are endogenous to the broader neighborhood in general. And, and we'd probably want to think about some way of instrumenting for that. Now, I'd just like to tell you a little bit about my feelings on the spatial correlates. And first is the density in coronavirus, where I think, I'm just gonna share one slide of mine. Uh, and hope you can, uh, see it, uh, right here, uh, where there's been a lot of attention. Mons The downsides of density and the fact that in many dense urban hubs, uh, instead of focusing on the benefits in terms of productivity and livability, this density in this time has been a real transmission belt for the virus. And I think, David, your analysis shows the significance of density and transmission, while Bo's work shows density is not really relevant. But I think what we really need to do is make clear the distinction between density and crowding, and I'd like to just put together, show you some data we put together for New York, where if you were to look at Jackson Heights in Queens, that's the most worst affected neighborhood. Cases are like 4000 per 100,000 residents. In Chelsea on the other side, it's 925 per 100,000 residents. But what's relevant to note here. Is that Queens is not the densest neighborhood in New York City. Queens has a density of 12,000 people per square kilometer, where in Chelsea, it's about 31,000. So what makes a difference here is neighborhood incomes and associated characteristics, which really temper the extent to which complementary investments in housing and infrastructure, and amenities can really transform places from being crowded to becoming dense and livable. And Good planning and land development regulations, these sort of valuable places, developers have the incentive to build up tall structures, create, create a lot of floor space. So even when you have a pandemic and you want to shut down and keep social distance, dense places do pretty well. And in fact, to support the World Bank's work on the pandemic in developing country cities. We have developed a methodology that can help city leaders prioritize places and resources towards these potential hotspots, and these are the places with high contagion risk, uh, even under a lockdown. And what we do here is we distinguish density from crowding by accounting for differences in floor space and availability of amenities. So I think we need to take this density argument a little carefully. Second, on the issue of air quality, as Anna kind of pointed out, the evidence isn't all that clear. And if you were to look at the numbers, air pollution is a really silent killer. It takes about 4 million lives annually, and, and, and most parts of the world, especially in developing countries, cities, are way over the WHO guidelines. So I'd be surprised not to see a correlation. But I don't really know the extent to which we're untangling the fact of the air pollution on life, quality or expectancy from its from getting COVID-19, and this is particularly true because over the last two months we've seen a lot of cities shutting down and there's a lot of lockdown, a lot of NPIs, so air quality measures may have really changed as well. But here, when we think about it, and I'll just take one more minute. we want to think if you want to do robust work, we need to think structurally. And if PM 2.5 is a combination of emissions from industry and transportation and there are places with dirty industries, they're likely to have low land prices. And the places with a lot of pass through are not going to be great places to live, right? So in response, people are going to sort, and there's going to be sorting in the land market in cities where a lot of these low value places may get lower income communities who may not be able to afford the homes and the neighborhoods where, you know, high quality of living or social amenities are feasible. And the pollutants may also maybe exacerbate pre-existing health conditions. So I think it's really important to think about the sorting issues as we try to build a robust case on air quality. Uh, thank you again for these excellent presentations. Really appreciated it. Thank you so much, Shermik. Uh, I won't pass any comment over to you, Rishi. Uh, we're getting some fascinating insights. Thank you. Uh, thank you, Richard. Let me just start by saying, um, that really thanking all three presenters for excellent presentations. I think, um, this is an area of, um, where knowledge is just emerging. There's a lot that we don't know, and therefore I think they have done a fantastic job of getting us, you know, up sort of filling us in quickly on what is it that, um, We know and where are some of the unknowns, uh, so, uh, let me actually begin by uh recapitulate uh uh some of what we've heard. So, what do we know about COVID-19 and air quality? What is known, uh, we've heard a lot of stories. There's been a lot of newspaper reports that air quality has improved, you know, a lot of reports coming out that the Himalayas, for example, in South Asia, are visible for the first time. So we know that the lockdown has actually improved air quality. Right, so that's an important thing for us to remember as we're thinking about how a poor air quality can lead to more infections. So if air quality is improving, should we not worry, right? Are we OK? So the other thing that we know is, as a number of our speakers have pointed out. That places with higher air pollution, a number of studies have come out to show that there's higher uh COVID infection rates in these areas, as um, uh, you know, the speakers have told us. And now, uh, the other thing that I'd like to throw into the mix um uh in today's uh talk is what um what. What does this all mean for when countries um grow back, you know, should they, does all of this mean that they should, uh, does this give them a, um, uh, you know, should they pursue green fiscal stimulus, for example, to deal with air pollution as they grow back? What does all of this mean in terms of policy? Now, let me also point you to what is not known. We've heard a lot about, um, because we have fantastic satellite data, we've got, um, we've seen these images before and after lockdowns, where NOX levels, nitrogen dioxide levels have actually come down dramatically. But we don't really know um uh what has happened to the most harmful of air pollutants, which is PM 2.5, and um we also don't know at the moment what is driving some of these improvements, right? Then as uh both Anna and uh both uh uh discussed, we really don't know what are the mechanisms uh between that of transmission between air uh air pollution and um. Uh, COVID-19 virus, does it, does the, uh, some people have said that the virus hitchhikes on, uh, particulate matters. Is that the case? Or is it that, uh, the virus, um, you know, air pollution makes people more susceptible, uh, to, uh, any infection and therefore it makes it susceptible to, uh, the COVID-19 infection as well. And then of course, if countries were to think about growing back greener, uh what does this mean um in terms of, how would they design uh this kind of fiscal stimulus package? Now, as we heard from a number of speakers, and this talk today is really about air pollution and the COVID infections, I'd like to um try and summarize, and I'm not going to do full justice to the wealth of information that was shared, but there are 3 different mechanisms through which air pollution and COVID-19 infections can be correlated. One is transmission, and I, as I mentioned. Um, uh, here, what my understanding, um, and talking to epidemiologists and as Anna shared with us, it is not likely that the, um, uh, the COVID-19 infection is actually airborne or that it is, uh, that the virus itself is hitchhiking on air pollution partic particles, and this comes because of evidence coming out of China, out of Wuhan. Is showing that most of the transmission happened indoors within, sort of uh within families and it's not happening outside and which would mean that it's not, the virus is not hitchhiking, but this still um there is still a way in which air pollution can transmit the disease because, And, and especially spikes in air quality, uh, sorry, spikes in air pollution, uh, can irritate the throat, and Bo pointed this out as well, and coughing is a way in which you can transmit, so when you, you know, if your your throat is irritated and you cough a lot, that is a way in which you can transmit, uh, so that there is a way in which air pollution can help, can increase the transmission of uh the infection. Then, um, it also air pollution, uh, uh, will degrade, um. The uh upper airways um in um uh uh in um in your, in your nostrils, and that degradation makes you more susceptible uh to um getting uh the infection, and then finally, as Anna pointed out and um Bo also pointed our attention, air pollution increases the risk of a number of uh uh diseases. And these diseases have, people with these diseases have also been um have been found to have higher rates of hospitalization, um, so what, to summarize, what we do is we do expect, we do expect that air pollution is going to make the COVID-19 infection um worse. however, as uh our speakers have told us, we do not yet have the data. Uh, to be able to actually estimate this relationship carefully, and I'd like to give a shout out to both, because of all the papers that I have seen, um, um, and as Anna pointed out, I think some of them are coming out way too quickly, um, and some of them are once put out, are being criticized heavily. But uh uh Bo's paper really stands out in doing a careful um job of both first understanding, um first controlling for um you know, whether you have um, the infection and then given that you have the infection, are you more likely, uh, are you going to see more deaths, so you know, of all the papers that we see out there, Bo has done. Really a commendable job of trying to separate, but I think all in all, my uh takeaway from all of this is, listen, we expect there to be an interaction, but it's way too early for us at this time, sorry, I'm getting a call, I don't know if um it's related to us anyway um. And so, um, Uh, now, so we know that air pollution levels matter, but uh. The question is, um, what has happened to air quality? If air quality has actually improved due to the COVID-19 um uh pandemic. Do we then care? Should we be worrying about air quality during this time? And I'd like to point out to you that air pollution level, air quality, uh, matters now, because even despite the lockdowns, we have not seen improvements in PM 2.5 levels. So what I've shown you here is some work that we are doing in the World Bank. Where you know we're trying to complement the literature on um nitrogen uh where we are seeing a lot of um you know, images of improvements in uh nitrogen dioxide. But are we seeing improvements in p.m. 2.5, which are more correlated with health impact? And here, um, what I'm showing you are three different graphs, one for, uh, the Hubei, where the city of Wuhan is located in China, then, uh, for PM 2.5 levels in, uh, France, and finally PM 2.5 levels in the Indo-genetic plain, and the green, um, is for 2019, the blue lines are for 2018. And the yellow lines are for 2020. Now if you look um in um for France, for example, we see almost no decline in p.m. 2.5 levels, even though as Bo had shown up there was decline inOx levels. So the pollutant that really matters is not declining now um in the endogenetic plane, on the other hand, we. We do see a decline in p.m. 2.5 levels after the lockdown, but what has happened is in India, um, in this region, PM 2.5 levels have declined even, um, ahead of the lockdown. So we don't, uh, and the reason I'm pointing this out is because the lockdown in some ways is a way for us to better understand pollution levels. can you begin to wind up, please. Yes, and so this is my last slide, um, and so I think what I'd like to, you know, like to sort of leave you with is that even though we've heard that air pollution has improved, um, with the lockdown, the pollution, pollutant that matters, PM 2.5, um, has not necessarily declined uniformly, so it is still an, um, a problem. Then the uh discussion that we heard from Bo, as well as from Anna, and the uh as David pointed out, there, you expect to be there to be a relationship between uh air pollution and uh COVID-19 infections, although the data at the moment is not there for us to be able to, um, you know, draw that relationship out definitely. And so this, all of this says that this is a chance. Air pollution is going to be, is an issue now. It's going to become more of an issue as countries lift their lockdowns, and so, you know, this is a time for countries to start thinking about ways to. Um, green their fiscal stimulus, and here it is a matter of both, you know, putting in measures to reduce their pollution, but also measures to stimulate demand, and I'll stop here, Richard. Thank you very much. Thanks so much, Ruvishi. Thank you to our speakers and our discussions. I'm going to ask these you whatever else comes in throwing it at, at the panelists and the discussions. So our first two questions I think are more for Anna than anyone else. So the first question from a viewer in Bangladesh asks, can the virus survive dehydration? And there's a second question from someone who read The Economist, and there was a piece in The Economist that says, smokers don't seem to be heavily impacted by coronavirus. Why? And then there is a third question, um, I guess it could be Anna Berubashi as well and anyone else can take it from one of our colleagues in the bank, Iwande, who asks or suggests, so should we really be targeting PM 2.5. So, Anna, if you don't mind taking at least the first two questions. And uh and the 3rd 1 too if you want to, but otherwise we could then take the 3rd. OK. So yeah, no, the, the first question's about, uh, there's a wider context about how long the, the virus can survive in different environments. Um, I can't give you the exact figures of that, but I think it's sensitive to dehydration. It's certainly sensitive to ultraviolet light, um, and then how long it lasts on, uh, various surfaces. There are some studies, um, on this, so it lasts longer on plastic, for example, than on paper. So paper might be a few hours, whereas plastic might be. Um, up to a week or more, but obviously the viability of the virus decreases, um, um, over time. So, uh, what was the second question about the smokers? Um, I groan every time I get that question about the smoking, uh, um, because it's, uh, I think if you smoke long enough to give yourself a chronic illness, you're at increased risk from COVID-19. Um, there were some early studies that suggested smokers were underrepresented in people that, uh, were coming into hospital and then some other studies that suggested there was selection bias, um, involved in that and that, uh, it wasn't a, a, a true, um, association. Um, but there is some interesting, um, uh, research about the impact of nicotine on the receptor. There's the, uh, ACE2 receptor that the virus enters and suggesting that, uh, um, it might modify the infectivity, but I think the, uh, I think the message is don't smoke. Um, it's not gonna stop you getting COVID-19, but, um, there, there may be some interesting mechanistic, uh, um, issues to, to follow up on that around the nicotine impact on the receptor. Did you also, I'm, I'm going to throw a couple more because there's one more which is directed at you. So it turns out that they wanted you to, uh, the person uh who asked the question wanted you to answer whether 2.5 should be targeted. And there's also a question about diesel particles. And the question reads, we know that diesel particles serve as carriers for carcinogenic compounds. Even though we know this, could COVID virus particles stick to these? Eg through 2.5. Yeah. OK, so which, which should we target? Um, I would, I would go for both PM 2.5 and NO2, and the NO2 because there is, uh, mechanistic evidence about NO2 exposure, um, increasing susceptibility to respiratory viruses. And there is a direct inflammatory um effects, um, in, in the lungs from NO2. PM10 we have the most uh information about uh impacts generally on the, on the body. I mean PM 2.5, we, we know a lot about how, uh, um, how, how bad it is for you, so I would target that as well. In a lot of countries, PM 2.5 and NO2 are very highly correlated because they have the same source which is road transport. So actually by tackling the road transport, then you, you bring both of them down. Um, diesel, that's a very good point. Um, that nobody's done the study yet on the, on the diesel as to whether, uh, they're more dangerous. Certainly they look like they are potentially the more dangerous components of the particulate fraction, um, is diesel, and that's been part of the reason why, uh, um, there's been a lot of efforts to try and reduce emissions from diesel vehicles and to reduce the diesel fleet, on the, on the roads, um, in, in quite a number of countries. Can I just while I just talk about PM 2.5 as well in, in Europe. So, um, in the UK we have had PM 2.5 episodes during lockdown which have been related to the weather. So, um, we've, uh, because the precursors are around, um, and the hot sunny weather that we've had. Um, over the initial period of our lockdown, um, encouraged, uh, the formation of secondary formation of particulates, and that's why we've seen quite high levels of probably impacting on France, um, as well. Um, and I think, you know, there are lots of different sources of particulate matter, whereas the NO2 tends to primarily come from transport and from industrial sources. So if you lock down on those, then you reduce the NO2 levels. So I think that's, that's some of the differences that uh. Uh, um, that, that have been commented on in terms of what you actually see of levels. Thanks. There's 2 questions and anyone from the panel can take them. The first one is, do we have any information on the impacts of indoor air pollution? That's one question. I'm going to throw another one. What kinds of green stimulus policy should we use for, so that we get improvements in air quality. So, let's take these two and there's another 2 which are coming up. Uh, yeah, one of the, so, um, I would, you know, I'll, uh, um, appreciate if Anna will come in on the indoor. I haven't seen any studies actually. Looking at the infections largely because, you know, the data in developing countries where there is more use of, um, you know, solid fuels for cooking and indoor, so higher levels of indoor air pollution, the data are just not there, but one would expect, right? As I said, I think here we are in a situation where, you know, you can't wait for the perfect data to be able to draw some of these conclusions. I just don't think, I mean, we have enough evidence on the harmful effects of air pollution and we have enough evidence. Uh, of the kinds of mechanisms that are likely to be, uh, making the virus, infections of the virus worse. So in high levels of indoor air pollution, you're going to have a lot of coughing, you have people living together, you're going to see more infections. I mean, it is, it makes sense, and here I think sense has to overtake lack of um data, so that's, um, but you know, Anna can come in. In terms of what kind of green fiscal stimulus policies, and I think this comes back also to what Anna said, I think it's important to understand the source structure of uh PM 2.5 pollution, and what we're finding, unlike in developed countries where, Uh, there's a huge correlation between NOx, uh, so nitrogen dioxide and PM 2.5, um, and therefore traffic is the biggest contributor. In developing countries, it's actually a very complicated source structure for PM 2.5, and it's a very multi-sector problem, so household, uh, solid fuel use is a big contributor, agriculture. Is a big contributor and not just because of the crop burning, but also because of the secondary particle formation, high use of, we're finding in India, for example, very high emissions of nitrogen uh uh of of sorry of ammonia, from agriculture because of overuse of fertilizers and um is leading to secondary particle formation where it interacts with mocks and socks. So the policies in fact that we need um cut across these sectors. So let me give you just one example. For example, a green policy, a green fiscal policy could be uh removing some of the subsidies from uh of fertilizers, because what has happened in India is there's very high subsidies for uh for um urea, which has led to overuse of urea in uh agriculture, which in turn is leading to high ammonia emissions. And so removing the subsidy will give the fiscal space for governments, and those subsidies can be used as an economic stimulus. So you have the stimulus part of it, and then you're reducing the emissions by reducing ammonia emissions, and so that is one example, but happy to discuss, you know, almost every sector one can come up with a set of policies. Thanks. OK, great, Ruvishi, because you asked another question there which I won't put in front of you, which was sort of really questioning, you know, is the data right? Is pollution really coming down? There's a question for David, but others might wish to chime in as well. The question specifically is a classic one that I was waiting to check to, to come from Fhad, which is really given non-pharmaceutical interventions and social distancing are endogenous. How have you taken care of all of these endogenous impacts, you know, policies, etc. Uh, like NPIs and social distancing. David, over to you, and boy, you might want to jump in as well. Uh, Richard, can you hear me OK? I see my images seems to be frozen, but I'm hoping the microphone is not. We can hear you loud, loud and clear, David, please go ahead. I'll let you know if you, we can't hear you. Well, there's a whole suite of technical issues that one has to address and thinking about estimation in this context and You know, Bowe has touched on some of them on, on the issue of how one might control for differential policies across spatial units of, um, for example, social distancing. Uh, of course, in principle, the answer is, um, one would, would like to be able to do that in the context of doing work like this for 3000 US counties or 1500 municipalities in the Philippines, uh, the information base is sparse. To put it mildly, and so you're really thrown back then on the issue of whether the errors and variables problem here might haunt you, uh, perversely and looking at the, the results we've gotten, that is to say, could there be some covariates. Of distancing, which also would affect the variables that we're trying to explain. So that we're getting a biased view of the effect of the variables that we've observed. In fact, let me generalize that. Uh, I think, uh, a couple of colleagues have Refer to the errors and variables problem in the measurement of cases. For uh coronavirus. In the US, for example, the, there's the New York Times database, the Johns Hopkins database, there are other databases. Everyone knows that there are uh variations in reporting. And the same problems apply there. Now, econometrically, there are 3 cases. The first case is, uh, there, there is a sort of a random incidence of reporting. Well, that's OK. Econometrics was basically built to handle random variation, uh, in, in the variable being explained. That's not a problem. The Second case would be whether, uh, there might then be, um, some, some variation in Variables that is perverse in the sense that there may be a covarying factor on the right-hand side of these equations, which causes an effect to be over or underestimated. Now, I think that's, that would be the question that we need to look at in the context of whether or not distancing is a problem. We have only the roughest of notions, by the way, of whether or not stated policies of states or counties in the US are having any effect on distancing. Uh, it seems quite evident that many so-called red states have many metro areas where distancing has been fairly effective. There are many so-called blue states where it is quite obvious that in some contexts, distancing has been absolutely ineffective. And so, we then have to ask ourselves whether or not there might be some underlying model of causality, which would cause some of the characteristics we're looking at to affect distancing. The problem I have in trying to do these estimates right now, and the problem that's kind of frustrating is, we, there has never been a test of That. So, we have no idea what those covariates might be or what the signs in the equation might be. So, this is all a bit of throwing up of the hands here. I fervently agree that there might be an errors of variables problem here. It might lead to biased estimates of the effects that we're looking at. Uh, but we simply at this point, do not know enough to say anything systematic. There's one further thing I, I'd like to add, and it interests me. The problem with spatial covariation as Bo has said, is very important in this context. And it's not clear, for example, in the US that the county is the right unit of observation because many phenomena are more regionally general and they vary over space. I didn't have time, for example, to point this out, but in the US determinants of life expectancy equation, temperature enters into the OLS version of that, but not into the spatial econometric version. The reason for that is temperature is much more widely diffuse on average than county level units would suggest. So, there really are not that many effective. Degrees of freedom as far as temperature is concerned, uh, uh, that we're looking at. So, I, I mean, I, I confess to some frustration with this. Uh, it's not evident to me that anything that we're looking at is leading to any systematic bias, uh, one way or the other in the results, but I've readily acknowledged that we could undoubtedly explain the variation that remains unexplained if we had better microinformation about distance. OK, thank you. Thank you very much, David. That was very comprehensive. Bo, did you want to give a brief answer because there's a couple of other things, uh, and if you wanted to have any concluding remarks, you've got the floor for a sentence or two to go of, of your words of wisdom for us to conclude with, and I'm going to do the same for the others. I can, I can add 22 things to what David already said, and that is on the errors in variables. So if there's a lot of measurement error, which there likely is, then uh you would actually basically uh expect the bias towards zero. so that will mean that if there's a lot of extra noise, uh, in your, in your data, uh, which includes, uh, noise that has No correlation at all. You would actually expect your, your biases to be towards zero. So the fact that you're still finding a very significant uh parameter, uh, actually in, in, in the situation where you do expect measurement errors is actually even stronger because you would expect the results to be more conservative, uh, when there's a lot of measurement error. Second, uh, if this measurement error is actually not all too random. And there's a a a systematic measurement error, which uh we could also expect. So if there's a convenience sampling. So, for example, doctors are more aware of uh respiratory illnesses in areas with high pollution because historically there have been a lot of patients there, hence they're more eager to test for corona, hence you see more uh cases. That'll be more uh problematic because then you would expect, uh, uh, just purely because of the sampling strategy to have more cases in polluted areas. And then I would say, well, it really becomes the question whether that uh as a theory is more far-fetched than simply following the evidence that suggests that particulate matter is bad for your health, period, and we see, uh, a correlation with COVID cases, uh, and then it's really saying, well, what is more far-fetched and uh I would lean toward following the evidence right now. Um, so that's just a quick addition to what David already said. OK, great. Thanks. Look, we're, we, we're coming to the fact that we've got 3 minutes left. So, Anna, in a minute, you wanted to come in on indoor air pollution, anything else? And if there's time, I'd like to give a minute to Ubuhi and to show me if they want to take a minute to just any, any last words of wisdom. Thanks. Over to you, Anna. Uh, just to comment on the indoor air, it's something we don't know very much about in, uh, developed countries either. Um, I think we neglect it as our peril. Um, and one of the things we've been discussing is increased use of cleaning agents, so we've got more volatile organic compounds and, uh, bleach fumes, um, in the house, uh, potentially than previously. Um, I will come back to the issue of, uh, what we learned from SARS, I think as my final comment. Um, so I think the positive thing about SARS was that public health measures helped conquer the disease before we had a vaccine and, um, before we even genotyped it at that, at that stage, so. Um, I think, you know, we need to keep in mind public health, just basic public health, um, uh, uh, measures can be very effective in controlling, um, this disease, and, um, that's, that's where I'll leave it at, but thank you. Thank you so much, Anna, those are, those are indeed wise words. Over, over to you, you've got a minute. Yeah, thanks Richard. I won't take, I just want to say again, uh, you know, I've learned so much from these excellent presentations, they're very interesting. Oh, can you hear me now? Yes, we can. No, we can't. You're muted. Now you're not muted. Yes, yes, now we can hear you. Please go ahead. OK, great, uh, just to say that, uh, you know, excellent presentations. I learned a lot, but I just want to say that listen, air pollution matters. Uh, we know it's a problem, but we also need to point out that it still matters, and so we don't want policymakers to sort of take their eye off the ball, and I'll leave it at that. Thanks, Urishi. Show me over to you, then I'll have to one sentence and we will then call it a day. Thanks, everyone. Uh, I really found the presentations to be insightful and providing at least, uh, policymakers in developing country cities, a lot more information on the transmission parts of the coronavirus. As we go forward, I'd like to make an urge to do more work on three dimensions. But let's try to get a better understanding about land use and transportation. Because a lot of patterns we see today, both in terms of connectivity and the spread of the virus, but the emissions and the p.m. 2.5 are sort of driven by an underlying economic structural issue of how land in the city is organized and it's organized around industrial use, residential use, and commercial use, and how transportation and the choice of transportation modes link up these places. So getting a better understanding of land and transportation would be really helpful. Second, I think we, in, in all the work that we're seeing on measures of density, we should be more sort of proactive and thinking about amenities and living space and not only physical measures of density. So let's look at economic geography, not only physical geography. And the final part, I think, is that it'll be really useful for the group in kind of trying to in the short term distill. So what is it that city leaders can really do? If a lot of the work, you know, will be led by mayors and the remit of mayors in developing countries is very limited, what is it that we can really help city leaders do both on the density issue but also the quality issue? Versus where city leaders need to work with higher levels of government. Thank you. Thank you very, very much, Shermik. Um, I'd like to thank the presenters very much and the discussers too, equally. And, uh, might I say my final comment, I'm sure everyone that was listening in has learned something new, if not a lot new. And we really have to thank you all. It was rich in material, rich in new research, and we've all come away knowing an awful lot more. So thank you. My final point is that apparently, the recording of this event and all the materials are on a web page. Uh, perhaps there can be an email letting people know what the link to that web page is. So with that, we're just 2 minutes over time, which is not so bad. Let me really profusely thank our speakers, particularly those joining from a different time zone in Europe. And indeed, uh, David as well, even if you're on the same time zone, Shomi Uvishi, thank you so much and thank you to all our listeners. This was a really, really excellent session that we had. Many thanks and have a great day and a great evening, everyone. Bye-bye. Thanks, Richard. You

showAllTranscripts

no

duration

PT1H32M36S

scene7File

worldbank/Air Pollution Exposure and COVID

scene7Domain

scene7FileAvs

worldbank/Air Pollution Exposure and COVID-AVS

title

Air Pollution Exposure and COVID

description

Air Pollution Exposure and COVID

showTimestampAndTranscript

yes