The Effect of Pollution on Labor Supply: Evidence from a Natural Experiment in Mexico City Paulina Oliva University of California Santa Barbara joint with Rema Hanna, Harvard, NBER, BREAD
Question and Motivation Does high pollution reduce work hours? Pollution may cause temporary illness Effect on hours worked is theoretically ambiguous Understanding this question is Particularly important for developing countries when assessing the effects of more stringent environmental regulation Important in designing optimal taxation policies (Schwartz and Repetto 2000, Williams 2003)
What we do Estimate the elasticity of hours worked with respect to air pollution Challenge: confounding factors decline in business activity may affect both pollution and labor markets Exploit natural experiment: closure of a large refinery in Mexico City in March of 1991 Compare the changes in hours worked for individuals who lived in neighborhoods located near the refinery (that experienced a fall in pollution) with those who lived in neighborhoods far from the refinery (that did not experience a fall).
What we find Neighborhoods in the vicinity of the refinery experienced 8% fall in SO 2 relative to other neighborhoods One percent increase in SO 2 results in 0.43-0.67% decrease in hours worked in the following week Translates to about a 756 Peso (USD 198) gain per worker over the course of a year
Health or alternative mechanisms? Direct effect on labor demand Labor demand contraction may increase unemployment underestimate elasticity Exploit variation in refinery-related pollution that is orthogonal to the distance to the refinery (wind, altitude) Results remain similar in magnitude and significance Selective migration People with high valuation for air quality may move to refinery neighborhood after closure Migration rates very small No differential changes in observable demographic composition
Outline of talk I. Conceptual Framework and Background II. Data III. Primary Methodology and Results IV. Labor Demand Mechanisms V. Selective Migration VI. Discussion
Conceptual Framework Partial equilibrium household model Pollution increases the disutility from work through own health health of dependents (e.g. Currie et al, 2009; Ransom and Pope, 1992)
Conceptual Framework Partial equilibrium: effect is ambiguous Positive component: better air quality less disutility of work more hours worked Negative component: more work more income lower marginal utility of income reduce the need for air-quality substitutes General equilibrium: productivity effects (Croker 1981, Graff Zivin and Neidell 2011), increased aggregate labor supply
Conceptual Framework Partial equilibrium: effect is ambiguous Positive component: better air quality less disutility of work more hours worked Negative component: more work more income lower marginal utility of income reduce the need for air-quality substitutes General equilibrium: productivity effects (Croker 1981, Graff Zivin and Neidell 2011), increased aggregate labor supply
Refinery Background Closed in March of 1991 At its peak: 174 ha, 14 refining facilities, 3 petrochemical facilities, 35% of the refining capacity in Mexico Ministry of Environment estimated that the refinery emitted 85 thousand tons of air pollutants per year (2% of total air pollution)
Data Intro Theory Data Empirics Discussion National Employment Panel Survey of Mexico (ENEU) Household rotative panel (1989-1993) Hours of work and employment status in last week Automatic Network of Atmospheric Monitoring (RAMA) Hourly measures of SO 2 for 15 stations (1989-1993) Other pollutant measures were not available Wind direction and wind speed patterns Compute SO 2 at census block level using station data weighted by distance (Currie and Neidell, 2005)
Data Intro Theory Data Empirics Discussion
OLS Intro Theory Data Empirics Discussion To consistently estimate β 1 : ln(y int )=β 0 + β 1 ln(p in(t 1) )+u int E(u int ln(p in(t 1) )=0 Likely violated. For example, wealthier households live in low-pollution areas
IV: Distance-Closure Interaction ln(y int )=β 0 + β 1 ln P in(t 1) + αt 1 + δ n + ν int (1) Instrument ln P in(t 1) with (Post(t 1) D in ) conditional on week and census block fixed effects E (Post (t 1) D in ) ν int α t 1, δ n = 0 (2) * Specifications with demographic controls and census block specific time trends
First Stage 8% relative reduction in pollution in 5 km perimeter of refinery relatively higher pollution within a census block after the closure for those located farther away M7*AE7.+N(D+(I+HA.IAE+56(O6L/!"#!$#!%#!&# '()*+,+-.()/ 01213$3,,,!121$$4# '()*+,+56)*789/ 1211:4,,, 1211:4,,, 1211;:,,,!1211":#!1211":#!1211$"# <//=+>? @ @ @ @ -/8)A)+B.(9=+>? @ @ @ @ 5/C(DE7FG69+-(8*E(.) @ @ -/8)A)+B.(9=+HF/96I69+J/7E+KE/8L) @
Reduced Form Relative rise of hours worked near refinery (Column 1) After the closure, we find a relative fall of 0.30-0.60 percent per km of distance from refinery (Columns 2-4)!"#"$%"$&'()*+),-".'-$/012*3'41*5"%6 /76 /86 /96 /:6 ;13&'<'=-13" >?>@@8<<< />?>7AB6 ;13&'<'!+3&)$C" D>?>>98<< D>?>>8E< D>?>>F><< />?>>7@6 />?>>7@6 />?>>8:6 4""5')$%'="$323'G-1C53'HI J J J J!"K1L*)#M+C'=1$&*1-3 J J ="$323'G-1C5'N#"C+O+C'P")*'Q*"$%3 J
IV Results 1% increase in SO 2 leads to a 0.43% fall in hrs worked (Col 3) Elasticity is 0.61 with census block specific time trends (Col 4)!"#"$%"$&'()*+),-".'-$/012*3'41*5"%6 /76 /86 /96 /:6 -$/;2-<2*'!+1=+%"6 >?@AAB8CCC >?@:D?ACC >?@:8BECC >?@A?FECC /?@8:AA6 /?@88976 /?@87BF6 /?@8A986 G$3&*2H"$&.'I13&'C'J-13" K G$3&*2H"$&.'I13&'C'!+3&)$L" K K K 4""5')$%'J"$323'M-1L53'NO K K K K!"H1P*)#Q+L'J1$&*1-3 K K J"$323'M-1L5';#"L+<+L'R")*'S*"$%3 K
Labor Demand Mechanisms: Theory IV assumption violated if labor markets are local If local, the labor demand contraction would lower eqm wages Upward sloping labor supply underestimate elasticity Downward sloping labor supply overestimate Labor supply elasticity is positive (Arceo and Campos, 2010) Unemployment threat if real wage rigidity overestimate No rigid wages when inflation is high (Fehr and Goette, 2005); inflation was 22%
Labor Demand Mechanisms: Evidence Directly test for localized wage effects Exploit variation in refinery-related pollution that is orthogonal from distance to the refinery (wind, altitude)
Wages No effect of the closure on wages Only suggestive: theoretical effect of wages is ambiguous!"#"$%"$&'()*+),-".'-$/0""1-2'0)3"4 /54 /64 789&':'!+9&)$;" <=>===? <=>==@A /=>==@B4 /=>==@?4 0""1'CD')$%'E"$9F9'G-8;1'CD H H!"I83*)#J+;'E8$&*8-9 H H E"$9F9'G-8;1'K#";+L+;'M")*'N*"$%9 H
Exploiting downwind/upwind patterns
Exploiting downwind/upwind patterns Instrument ln P in(t 1) with (Post(t 1) D in W in(t 1) ) : ln(y int )= β 0 + β 1 ln P in(t 1) +β 2 (Post (t 1) D in )+β 3 (Post (t 1) W in(t 1) ) +β 4 (D in W in(t 1) )+β 5 W in(t 1) + α t 1 + δ n + η int
Exploiting downwind/upwind patterns Estimated elasticities in FE models (Cols. 1 and 2) are similar in magnitude and significance to primary model Weak instruments with census block time trends (Col. 3) The estimated elasticities do not appear to be driven by local labor demand shocks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
Selective Migration The neighborhood of the refinery might become more appealing for individuals who value air quality after closure Estimated effect might be capturing differences in hours worked across different groups of people
Selective Migration No differential change in migration by distance to refinery No differential change in observable demographics by distance to refinery Fail to reject the null of the joint hypothesis test (p-value of 0.79) B'*.#$%P6"3@!"#$%&'( )'*%+',#'!%$$"'( -./01(23%&".4 5"#601(23%&".4 789 7:9 7;9 7<9 7=9 7>9!"#$%&'(&)$$*&+,-&.$#/0/&'%12*&+,-&"#3&.$#/0/&'%12*&45$26762&8$"9&:9$#3/?.@&0A0B"@&%43' CDCCC: CDCCCE CDC8C8 FCDCC8G FCDCCC; CDCCC: 7CDCCC=9 7CDCC889 7CDC:HC9 7CDCC8:9 7CDCC8C9 7CDCCCH9 I."4&0J'@&0?FK%+2' CDGE>E LM@'$N%&".4@ =8O=8E 8<;O;88 8<;O;88 8<;O;C: 8<;O:H: 8<;O:H:
Heterogeneous Treatment Effects The elasticity is smaller for women than for men, but the difference is not significant Suggestive evidence that workers with children under-five have higher elasticity Higher elasticity for salaried workers No significant difference between blue and white collar workers
Discussion Benefits of environmental regulation There are other welfare-inducing effects of lower pollution, so our estimate is not a comprehensive measure of the benefits However, effect on labor supply is a non-trivial component Implications for optimal pollution tax Pollution taxes are appealing because of double dividend Revenue from double dividend is not one for one if people substitute away from taxed goods, towards leisure: interaction effect (Parry 1995, Goulder & Bovenberg 1997) Interaction effect is mitigated if air quality directly increases labor supply (Schwartz & Repetto 2000, Williams 2003) Alargeandpositiveelasticityimpliesthattheoptimaltaxmay be closer to the marginal damage of air pollution
Conclusion This study suggests that gains in worker effort may help offset some of the economic losses that may result from more stringent environmental regulation 1% increase in SO 2 results in a 0.43 to 0.67% fall in hours worked This translates to about Peso 756 (USD 198) per-worker gain per year Pathways for future research: Distribution of gains Explore cumulative effects of pollution on labor supply