1

Air pollution and children’s

respiratory health:

Do English parents respond to

air quality information?

Katharina Janke

20th June 2012

Introduction

• Epidemiological research of short-term

effects of air pollution: time-series studies

Single city

Populations serve as their own controls

Weather-driven variations in pollutant levels

• Limitations

Central monitoring site

Single-pollutant models

2

Fixed effects approach

• Panel data (longitudinal data)

• 89 English local authorities in 2003 to 2007

Reduces city selection bias

Areas smaller => reduces measurement error

• Control for unobserved factors by adding

local authority dummies

outcomea,t = 1exposurea,t + 2controlsa,t

+ a + ea,t

3

Specification

admissionsa,t,w,q,y =

4i=0 [i

NONOa,t-i + iOZOZa,t-i]

+ 4i=0 M’a,t-ii

+ Xt’

+ a,q,y + w,y + ea,t,w,q,y

4

Air Pollution Forecast

5

6

7

Control for avoidance

admissionsa,t,w,q,y =

4i=0 [i

NONOa,t-i + iOZOZa,t-i]

+ 4i=0 iforecasta,t-i

+ 4i=0 M’a,t-ii

+ Xt’

+ a,q,y + w,y + ea,t,w,q,y

8

Air Pollution Forecast

9

Data • Hospital Episode Statistics (HES)

Hospital emergency admissions for respiratory

diseases and symptoms, age 5 to 19

(ICD-10 codes J00-J99, R05, R06)

Patient’s local authority: daily admission counts

Rates (per 100,000)

• UK Air Quality Archive

Distance-weighted mean of pollutant

concentrations at monitors in 5/10/15/20 km

radius around local authority’s population-

weighted centroid

10

11

12

Results

Without forecast

NO2 / 10 0.033*** (0.007)

O3 / 10 0.024*** (0.006)

Forecast

Robust standard errors in (round brackets), clustered at county

level. Coefficients are sum of coefficients on contemporaneous

value and four lags of pollutants. 148,210 observations in 89

local authorities with 23 county clusters.

* significant at 5%, ** significant at 1%, *** significant at 0.1%.

• Elasticity at mean:

• Example for NO2:

0.033 x (3.46/1.36) = 0.08

13

admissions

NOi

NOi

24

0

2

14

Results

Without forecast

NO2 / 10 0.033*** (0.007)

[0.083]

O3 / 10 0.024*** (0.006)

[0.099]

Forecast

Robust standard errors in (round brackets), clustered at county

level. Numbers in [square brackets] are elasticities at the mean.

Coefficients are sum of coefficients on contemporaneous value

and four lags of pollutants. 148,210 observations in 89 local

authorities with 23 county clusters.

* significant at 5%, ** significant at 1%, *** significant at 0.1%.

15

Results

Without forecast With forecast

NO2 / 10 0.033*** (0.007)

[0.083]

0.035*** (0.007)

[0.088]

O3 / 10 0.024*** (0.006)

[0.099]

0.025*** (0.006)

[0.103]

Forecast -0.033 (0.041)

[-2.398] Robust standard errors in (round brackets), clustered at county level.

Numbers in [square brackets] are elasticities at the mean for NO2 and O3

and percent change in admission rate (evaluated at the mean) for discrete

change of Forecast from 0 to 1. Coefficients are sum of coefficients on

contemporaneous value and four lags of pollutants. 148,210 observations in

89 local authorities with 23 county clusters.

* significant at 5%, ** significant at 1%, *** significant at 0.1%.

Results

• Subset of respiratory diseases:

Asthma

16

17

Asthma

Without forecast With forecast

NO2 / 10 0.011* (0.006)

[0.086]

0.013** (0.006)

[0.103]

O3 / 10 0.005 (0.003)

[0.067]

0.006* (0.003)

[0.080]

Forecast -0.035*** (0.010)

[-8.243] Robust standard errors in (round brackets), clustered at county level.

Numbers in [square brackets] are elasticities at the mean for NO2 and O3

and percent change in admission rate (evaluated at the mean) for discrete

change of Forecast from 0 to 1. Coefficients are sum of coefficients on

contemporaneous value and four lags of pollutants. 148,210 observations in

89 local authorities with 23 county clusters.

* significant at 5%, ** significant at 1%, *** significant at 0.1%.

Conclusions

• 10% increase in NO2 or O3 increases

children’s hospital emergency admissions

for respiratory diseases by 1%

• For asthma admissions:

Moderate or high air pollution forecast reduces

admissions by 8%

15% bias in pollutant coefficient estimates

18

Visitor data from Bristol Zoo

19

20

Daily visitor counts (Bristol Zoo)

Day visitors Members

Forecast 0.023 (0.030) -0.070** (0.028)

Rain -0.015*** (0.002) -0.013*** (0.002)

Max. temperature 0.027*** (0.005) 0.022*** (0.005)

Min. temperature -0.016*** (0.005) -0.017*** (0.005)

Wind speed -0.018*** (0.003) -0.021*** (0.003

Newey-West standard errors allowing for autocorrelation up to lag 10 in

(round brackets). Regressions include year-month dummies, dummies for

day of week, public holidays in winter, public holidays in summer, bank

holiday weekends, school holidays and school holiday weekends. 2,382

observations. * significant at 5%, ** significant at 1%, *** significant at 0.1%.