WEATHER AND DEATH IN INDIA - ROBIN BURGESS1 OLIVIER DESCHENES2 DAVE DONALDSON3 MICHAEL GREENSTONE4

Page created by Mathew Mcdonald
 
CONTINUE READING
WEATHER AND DEATH IN INDIA - ROBIN BURGESS1 OLIVIER DESCHENES2 DAVE DONALDSON3 MICHAEL GREENSTONE4
Weather and Death in India

Robin Burgess1 Olivier Deschenes2
Dave Donaldson3 Michael Greenstone4
          1 LSE,   CEPR and NBER
             2 UCSB   and NBER
          3 MIT,   CIFAR and NBER
             4 MIT   and NBER
Motivation
   • Large rural populations still mostly rely on
     weather-dependent economic activities.
        • Limited access to ‘adaptation’ technologies
          (irrigation, air conditioning, etc).
        • Famines are now largely “history” (O’Grada). But
          do weather shocks may still affect mortality?

   • Extensive literature on consumption smoothing
     via risk-sharing arrangements within villages.
        •   But how severe are aggregate shocks?

   • Concern over climate change brings these
     issues to the fore.
        •   Size of health risks poorly understood, especially in
            LDCs were they are likely to be largest.
Weather and Death: Key Questions
   1. Does the weather affect mortality in developing
      countries?
      (a) Why do these effects exist?
      (b) Are these effects different across rural and urban
          regions?

   2. What do these effects imply for policy?
      (a) Could weather-based income transfers help?
      (b) Can these effects be mitigated? (In progress. Not
          for today.)
      (c) The health costs of climate change?
This Paper
   1. Estimate ‘semi-parametric’ relationships
      between daily weather and annual mortality
      rate in India (1957-2000).
         • Utilize new dataset on finest geographic level of
           data (district level) for both weather and mortality.
         • Emphasis on temperature effects (more novel in
           literature).
         • Contrast rural/urban effects given that rural
           population is more weather-dependent.

   2. Estimate analogous effects of weather on
      rural/urban real incomes (output, wages,
      prices):
   3. Implications of results for predicted costs of
      climate change.
Daily Temperatures in India: 1957-2000
Using our data we construct one of these histograms per district and year

  80
                                                                               73

                                                                                       63
  60                                                                   56

  40
                                                               34
                                                                                               31
                                                      24
                                                                                                       20
  20                                          18
                                      13
                              9                                                                                10
       4              5                                                                                               3
              3
   0
       
                                        Distribution of Daily M ean Temperature (C)
                                              (Days Pe r Year in Each Interval)

                                                           Days Per Year
Weather and Death in the US and India
Estimates for US from Deschenes and Greenstone (2009)
Weather and Death in the US and India
US and All India
Weather and Death in the US and India
US and Urban India
Weather and Death in the US and India
US and Rural India
Our Main Result
US and Rural/Urban India
Summary of Results I

    • Large reduced-form effects of extreme
     temperatures on mortality.
        •   Approximately 20× larger than in USA.
        •   Large and persistent lagged effects.
        •   Dire upper-bound (no-adaptation) consequences of
            climate change.
Summary of Results II
    • Cluster of findings that suggests this
      could plausibly work through an
      agricultural mechanism:
        •   Mortality effects occur in rural areas only (even
            among infants).
        •   Within rural, no effects during non-growing season
            (hottest time of the year).
        •   Rural incomes (productivity, nominal wages and
            prices); large effects in rural areas, no effects in
            urban areas.
        •   Bank deposits: drop in rural areas, no change in
            urban areas
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
Weather and Death 1: Direct Effects
    • ‘Heat stress’ (cardiovascular failure while
      attempting to stabilize body’s core
      temperature):
         •   Large public health literature on this (eg survey in
             Basu and Samet (2003)).
         •   Very little work on developing countries. But Hajat
             et al (2005) find very small effects in Delhi (around
             one heat wave).
         •   More stress while working hard, outdoors. (US
             Army guideline is to do ‘moderate work’ for only
             15 mins/hour when outside temp is 32 C.)
         •   Near term displacement (‘harvesting’) important.
Weather and Death 1: Direct Effects

    • Change in disease environment:
        •   Malaria thrives in hot and wet conditions (though
            malaria rarely directly fatal in India). ⇒ We check
            for this using malarial incidence data below.
        •   Intestinal infections and deaths peak in rainy
            season (Dyson (1991) on India; Matlab studies on
            Bangladesh; Chambers et al (1981) on Lesotho).
            Some suggestive evidence from Matlab studies that
            these infections are worse in wetter years. ⇒ We
            can’t measure infections directly, but we explore
            interaction of rain and temperature below.
Weather and Death 2: Income Effects

   • ‘Heat stress’ also harms plants and hence rural
     incomes (wages for workers and yields for
     owner-cultivators/sharecroppers).
       •   Focus of huge ‘climate change and agriculture’
           literature in agronomy (eg IPCC reports) and
           economics (eg Deschenes and Greenstone (2007)
           on US, Guiteras (2009) on India).
Weather and Death 2: Income Effects
   • Could a transitory income shock like this pass
     through to death (in year t or t − 1...)?
        •   Income ⇒ Consumption: depends crucially on
            household’s ability to smooth consumption across
            periods — ie ability to borrow and/or engage in
            insurance.
        •   Consumption ⇒ Death: depends on background
            state of health (eg ‘Synergies hypothesis,’ or
            strong interaction between malnutrition and
            exposure to disease). But note that consumption
            here could mean both health goods and food.

   • Clearly this ‘income effect’ should only be at
     work in rural areas, and from weather shocks
     that occur during the growing season.
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
A Simple Model of Survival Choice
Objective function

        • Consider a household with the following
           lifetime expected utility function:
                          "∞                              #
                            X
                   V =E        (β)t (Πtt 0 =0 st 0 )u(ct )
                               t=0

        • Where:
           • u(ct ) is the utility enjoyed in t if alive.
           • ct is consumption in t of goods that the consumer
             values directly.
           • st is the probability of being alive in t conditional
             on having been alive in t − 1.
           • β is a discount factor.
A Simple Model of Survival Choice
Objective function

        • Now suppose that st = s(ht , Tt ), where:
               •   ht is consumption of ‘health goods’ (air
                   conditioning, component of nutrition that is purely
                   for survival) by the consumer. They are only valued
                                                             ∂st
                   for their survival-enhancing properties ( ∂h t
                                                                  > 0,
                     2
                   ∂ st
                    ∂ht2
                         < 0). These goods sell for p (assumed
                   constant for simplicity) relative to ct goods.
               •   Tt is the temperature (or other weather variable)
                   in period t. Tt enters s(.) because of the ‘direct’
                                                            ∂st
                   effect of weather on death. We expect ∂T    t
                                                                 < 0.
A Simple Model of Survival Choice
Timing and Constraints

       • Finally suppose that income is given
          exogenously, but is a function of Tt :
          yt = y (Tt ). This is the root of the
          ‘income-based’ effect of weather on death.
       • Timing is as follows:
             1. At the start of period 0, T0 is drawn.
             2. The consumer chooses c0 (T0 ) and h0 (T0 ).
             3. Th consumer’s death shock arrives.
             4. Consumption occurs, if the consumer is alive.
             5. Period 1 begins.
A Simple Model of Survival Choice
Timing and Constraints

       • For simplicity, imagine the consumer has access
          to a full and fair annuity market, and can
          borrow (at R = 1 + r ) so the intertemporal
          budget constraint is:

          c0 + ph0 − y (T0 )
                 "∞                                            #
                  X
             = E      (R)t (Πtt 0 =0 st 0 )(ct + pht − y (Tt )) .
                         t=1
A Simple Model of Survival Choice
First-order conditions

        • Consider choice at period 0, after T0 is known.
           FOCs:

           Euler : u 0 (c0 ) = βRE [u 0 (c1 )]
                         "             "∞                               #
                   ∂s0                  X
               h :          u(c0 ) + E       (β)t (Πtt 0 =1 st 0 )u(ct )
                   ∂h                    t=1
                        ∂s0
                         ≡   V0 = λp
                        ∂h
                 c : s(h0 , T0 )u 0 (c0 ) = λ
A Simple Model of Survival Choice
Result on WTP/WTA for T

      • Suppose a policy were to pay out money as a
         function of the current temperature:
         M = M(T0 ). What would this policy look like
         if it were to hold utility constant?
      • Can show that it will satisfy:

                          dM                 dy0   ds0 V0    ∂h0
         WTA(dT0 ) ≡            dV =0   =−       −        +p
                          dT0                dT0 dT0 λ       ∂T0

      • Vλ0 is the value of a statistical life, conditional
         on surviving in period 0.
A Simple Model of Survival Choice
Result on WTP/WTA for T
      • Can therefore potentially measure welfare
         impact of weather variation from data on:
              dy0
            • dT
                 0
                  ,which we have.
               ds0
            •  dT0
                   ,
                   which we have.
               ∂h0
            •  dT0
                   ,
                   which we don’t have.
            • Value of a statistical life, which other studies
              estimate.

      • NB: Above is all subject to caveats:
          • Lack of borrowing constraints hard to believe.
          • Other instruments for consumption smoothing (eg
            insurance) left out. (Is T0 aggregate or
            idiosyncratic?)
          • Constancy of p betrays evidence below that food
            prices are correlated with Tt .
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
Data Sources
   • Historical Weather:
        •   Precipitation: daily precipitation at each 1 × 1
            degree lat/long gridpoint, from Indian
            Meteorological Department. Spatial interpolations
            based on over 7,000 underlying stations.
        •   Temperature: modeled daily precipitation at each 1
            × 1 gridpoint, from National Center for
            Atmospheric Research.
        •   Gridpoints mapped to districts by inverse-distance
            weighting (within 100 km radius).
   • Mortality Rates:
       • Vital Statistics of India (VSI), 1956-2000.
       • Universe of registered deaths. But
         under-registration problematic (estimated at
         approx 35 percent).
Data Sources
   • Economic outcomes:
       •   Agricultural outcomes (yields, prices, wages),
           1956-86, from World Bank Agriculture and Climate
           Database.
       •   Urban wages and output, 1965-1990, relate to
           state-level registered manufacturing at state-level,
           from ASI via Ozler, Datt and Ravallion (1996).
       •   Urban price index 1965-1990, CPI for Industrial
           Workers, at state-level, from NSS via Ozler, Datt
           and Ravallion (1996)
       •   Bank deposits in registered banks, 1970-1998,
           from Reserve Bank of India.
Daily Temperatures in India: 1957-2000
 80
                                                                              73

                                                                                      63
 60                                                                   56

 40
                                                              34
                                                                                              31
                                                     24
                                                                                                      20
 20                                          18
                                     13
                             9                                                                                10
      4              5                                                                                               3
             3
 0
      
                                       Distribution of Daily M ean Temperature (C)
                                             (Days Pe r Year in Each Interval)

                                                          Days Per Year
Empirical approach: Graphical

    • To construct ‘flexible’ temperature impact
      graphs, estimate regressions of following form
      and plot the 15 θ’s
                      b (best seen graphically):

               15
               X
    ln Ydt =         θj Tdtj + δRAIN Controlsdt
               j=1
                           + αd + βt + γr1 t + γr2 t 2 + εdt
Empirical approach: Graphical
               15
               X
    ln Ydt =         θj Tdtj + δRAIN Controlsdt
               j=1
                             + αd + βt + γr1 t + γr2 t 2 + εdt
        •   dt: unit of observation is a district×rural/urban
            area, observed annually
        •   Ydt : crude death rate (deaths in year per 1,000
            pop at mid-year), or other outcome.
        •   Tdtj : Number of days in dt in which daily mean
            temperature was in ‘bin’ j
        •   Use various RAINdt controls; results extremely
            robust to this.
        •   γr1 t, γr2 t 2 are IMD climatological region-specific
            quadratic trends (Nr = 5).
Empirical approach: Graphical
           P15       j
ln Ydt =     j=1 θj Tdt   + δRAIN Controlsdt + αd + βt + γr1 t + γr2 t 2 + εdt

           • Intuition for these functional forms:
                • Temperature is not storable, so total annual
                  impact is approximated by sum of each day’s
                  impact (with unknown lags).
                 •   Water is much more storable (in soil, plant, or
                     irrigation systems), so appropriate measures will
                     aggregate over many days.

           • Other adjustments:
               • Weight by population/area as appropriate.
               • Cluster at district level (main results robust to
                 state block bootstrap).
Empirical approach: Tables
    • For tables, also estimate more parsimonious,
      parametric specifications:
                              3
                              X
      Ydt = θCDD32dt +             δk 1 {RAINdt in tercile k}
                              k=1
            + αd + βt +    γr t + γr2 t 2
                            1
                                            + εdt
    • CDD32dt : Cumulative number of degrees
      (above 32◦ C)-times-days in year t
        • Collapses ‘flexible approach’ (15 θ’s)
                                            b into a spline.
        • Choice of 32◦ C is somewhat arbitrary (but in line
          with agronomy antecedents); robust to sensitivity
          checks below.
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
Recall: Our Main Result
US and Rural/Urban India
Weather and Death: Rural vs Urban
All-ages death rate

                                                                                  Rural                 Urban
      Dependent variable: log (total mortality rate)
                                                                                   (1)                   (2)
      Temperature (degree‐days over 32C) ÷ 10                                    0.0134                 0.0045
                                                                             (0.0032)***             (0.0020)**
      Rainfall in lower tercile                                                  0.0318                ‐0.0030
                                                                              (0.0145)**              (0.0105)
      Rainfall in upper tercile                                                 ‐0.0003                ‐0.0148
                                                                               (0.0184)               (0.0110)
      Observations                                                               11,121                 11,525
Notes: Regressions include district fixed effects, year fixed effects and climatological region‐specific quadratic time
trends. Regressions weighted by population. Standard errors clustered by district.
Temperature and Infant Mortality: Rural
Rural infant mortality: mortality rate under age of 1
       0.04

       0.03

       0.02

       0.01

       0.00
               
       -0.01

       -0.02
                   Estimated Impact of a Day in 15 Temperature (C) Bins on Log Rural Annual Mortality Rate,
                                             Relative to a Day in the 22° - 24°C Bin
                                             -2 std err      coefficient      +2 std err
Temperature and Infant Mortality: Urban
Urban infant mortality: mortality rate under age of 1
       0.04

       0.03

       0.02

       0.01

       0.00
               
       -0.01

       -0.02
                   Estimated Impact of a Day in 15 Temperature (C) Bins on Log Urban Annual Mortality Rate,
                                              Relative to a Day in the 22° - 24°C Bin
                                             -2 std err      coefficient      +2 std err
Temperature and Mortality: Adaptation?
Rural total mortality, districts split into ‘hot’ and ‘cold’ by median cutoff
        0.04

        0.03

        0.02

        0.01

        0.00
               
       -0.01

       -0.02
                   Estimated Impact of a Day in 15 Temperature (C) Bins on Log Rural Annual Mortality Rate,
                                             Relative to a Day in the 22° - 24°C Bin
                                              'Colder' Districts    'Hotter' Districts
Temperature and Mortality: Improvement
over Time?
Estimate separate rural CDD32 coefficient, by 11-year period
Temperature and Mortality: Improvement
over Time?
Estimate separate urban CDD32 coefficient, by 11-year period
Temperature and Mortality: Malaria?
Annual Malarial Parasite Incidence (in randomly collected blood slides, by NMEP,
1965-2006)
Temperature and Mortality: Malaria?
Annual Falciparum Parasite Incidence (in randomly collected blood slides, by
NMEP, 1965-2006)
Temperature and Mortality: Malaria?
‘Malarial deaths’ according to NMEP, 1965-2006
Temperature and Mortality: Malaria?
‘Radical Anti-Malarial Treatments’ given out by NMEP, 1965-2006
Growing Seasons
   • Now explore differential timing of heat shocks
     within a calendar year.
   • Exploit stark beginning of the agricultural
     growing/planting season in India, which is
     oriented around the sudden arrival of
     (Southwest) monsoon rains in June.
        •   Studied in detail in AP region by Gine, Townsend
            and Vickery (2007).

   • Therefore define:
        • ‘Growing season’: period between ‘normal’
          monsoon arrival (in a district) and December 31st.
        • ‘Non-growing season’: three-month period before
          ‘normal’ monsoon arrival (in a district).
‘Normal’ Monsoon Onset
From website of Indian Meteorological Department; we map our districts to this.
GS vs NGS
Rural only
GS, with Lagged Effects
Rural only
GS vs NGS, with Lagged Effects
Rural only
GS vs NGS, with Lagged Effects
Urban only
GS vs NGS, with Lagged Effects
Rural and Urban
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
Temperature and Productivity: Rural
Rural Productivity: Real aggregate agricultural output per acre
 0.010

 0.005

 0.000
          
 -0.005

 -0.010
                 Estimated Impact of a Day in 15 Temperature (C) Bins on Log Agricultural Total Product
                      (Sum Weighted by Average Crop Price), Relative to a Day in the 22° - 24°C Bin

                                           -2 std err      coefficient      +2 std err
Weather and Productivity: R vs U
Rural: real agricultural output per acre;
Urban: (state-level) registered manufacturing output

                                                                                  Rural                 Urban
           Dependent variable: log (productivity)
                                                                                   (1)                   (2)
      Temperature (degree‐days over 32C) ÷ 10                                  ‐0.0100                 ‐0.0000
                                                                             (0.0035)***               (0.0055)
      Rainfall in lower tercile                                                ‐0.0915                 ‐0.0435
                                                                             (0.0097)***               (0.0327)
      Rainfall in upper tercile                                                 0.0036                 ‐0.0595
                                                                               (0.0063)                (0.0414)
      Number of observations                                                      8,604                  512
      R‐squared                                                                    0.87                  0.99
Notes: Regressions include district fixed effects, year fixed effects and climatological region‐specific quadratic time
trends. Column (1) regression weighted by average cultivated area; column (2) regression weighted by state urban
population. Standard errors clustered by district.
Weather and Productivity: GS vs NGS
Rural only. Real agricultural output per acre

             Dependent variable: log (productivity)                                     (1)                    (2)
  GROWING SEASON:
     Temperature (degree‐days over 32C) ÷ 10                                       ‐0.0090                 ‐0.0089
                                                                                 (0.0031)***             (0.0031)***
  NON‐GROWING SEASON:
     Temperature (degree‐days over 32C) ÷ 10                                                                0.0015
                                                                                                           (0.0022)
  Number of observations                                                              8,604                  8,604
  R‐squared                                                                            0.87                   0.87
  Notes: Regressions control for rainfall terciles and include district fixed effects, year fixed effects and climatological
  region‐specific quadratic time trends. Regressions weighted by average cultivated area. Standard errors clustered
  by district.
Temperature and Nominal Wages: Rural
Rural Wage: District-level agricultural wage
 0.010

 0.005

 0.000
          
 -0.005

 -0.010
                  Estimated Impact of a Day in 15 Temperature (C) Bins on Log Real Agricultural Wage,
                                         Relative to a Day in the 22° - 24°C Bin

                                          -2 std err      coefficient       +2 std err
Temperature and Nominal Wages: Urban
Urban Wage: State-level earnings per worker in manufacturing sector
 0.030

 0.020

 0.010

 0.000
          
 -0.010

 -0.020

 -0.030
                   Estimated Impact of a Day in 15 Temperature (C) Bins on Log Manufacturing Wage,
                                          Relative to a Day in the 22° - 24°C Bin

                                          -2 std err     coefficient      +2 std err
Weather and Nominal Wages: R vs U
Rural: Agricultural wages;
Urban: State-level manufacturing earnings per worker

                                                                                  Rural                 Urban
          Dependent variable: log (nominal wage)
                                                                                   (1)                   (2)
      Temperature (degree‐days over 32C) ÷ 10                                   ‐0.0045                 0.0065
                                                                             (0.0015)***               (0.0057)
      Rainfall in lower tercile                                                 ‐0.0167                ‐0.0223
                                                                              (0.0066)**               (0.0647)
      Rainfall in upper tercile                                                  0.0050                ‐0.0105
                                                                               (0 0069)
                                                                               (0.0069)                (0 0746)
                                                                                                       (0.0746)
      Number of observations                                                      7,994                  482
      R‐squared                                                                    0.95                  0.96
Notes: Regressions include district fixed effects, year fixed effects and climatological region‐specific quadratic time
trends. Column (1) regression weighted by rural district population; column (2) regression weighted by state urban
population. Standard errors clustered by district.
Weather and Nominal Wages: GS vs NGS
Rural only. Nominal agricultural wage

            Dependent variable: log (nominal wage)                                      (1)                    (2)
  GROWING SEASON:
     Temperature (degree‐days over 32C) ÷ 10                                        ‐0.0041                 ‐0.0043
                                                                                  (0.0015)**              (0.0015)**
  NON‐GROWING SEASON:
     Temperature (degree‐days over 32C) ÷ 10                                                                0.0014
                                                                                                           (0.0014)
  Number of observations                                                              7,994                  7,994
  R‐squared                                                                            0.95                   0.95
  Notes: Regressions control for rainfall terciles and include district fixed effects, year fixed effects and climatological
  region‐specific quadratic time trends. Regressions weighted by rural population. Standard errors clustered by
  district.
Temperature and Prices: Rural
Rural Prices: Price index of agricultural prices
 0.010

 0.005

 0.000
          
 -0.005

 -0.010
                    Estimated Impact of a Day in 15 T emperature (C) Bins on Log Agricultural Prices
                  (Sum Weighted by Average Crop Production), R elative to a D ay in the 22° - 24°C Bin

                                          -2 std err       coefficient       +2 std err
Weather and Prices: R vs U
Rural: Agricultural prices (mostly ‘farm gate’);
Urban: state-level industrial workers’ CPI

                                                                                  Rural                 Urban
               Dependent variable: log (prices)
                                                                                   (1)                   (2)
      Temperature (degree‐days over 32C) ÷ 10                                   0.0019                  0.0014
                                                                              (0.0007)**               (0.0094)
      Rainfall in lower tercile                                                 0.0107                  0.0108
                                                                             (0.0029)***               (0.0066)
      Rainfall in upper tercile                                                 0.0014                  0.0035
                                                                               (0 0029)
                                                                               (0.0029)                (0 0051)
                                                                                                       (0.0051)
      Number of observations                                                      7,994                  592
      R‐squared                                                                    0.95                  0.99
Notes: Regressions include district fixed effects, year fixed effects and climatological region‐specific quadratic time
trends. Column (1) regression weighted by average cultivated area; column (2) regression weighted by state urban
population. Standard errors clustered by district.
Weather and Prices: GS vs NGS
Rural only. Agricultural prices (mostly ‘farm gate’)

                  Dependent variable: log (prices)                                      (1)                    (2)
  GROWING SEASON:
     Temperature (degree‐days over 32C) ÷ 10                                        0.0019                  0.0020
                                                                                  (0.0007)**             (0.0008)***
  NON‐GROWING SEASON:
     Temperature (degree‐days over 32C) ÷ 10                                                               ‐0.0010
                                                                                                           (0.0006)
  Number of observations                                                              7,994                  7,994
  R‐squared                                                                            0.98                   0.98
  Notes: Regressions control for rainfall terciles and include district fixed effects, year fixed effects and climatological
  region‐specific quadratic time trends. Regressions weighted by average cultivated area. Standard errors clustered
  by district.
Temperature          and Bank Deposits: Rural
Estimated Response Function Between Temperature Exposure and Log Ban
Deposits
Bank     Per per
     deposits Capita,
                 capitaRural Areas
                        in Rural areas
     0.020

     0.010

     0.000
             
    -0.010

    -0.020
              Estimated Impact of a Day in 15 Temperature (C) Bins on Log Bank Deposits Per Capita,
                                       Relative to a Day in the 22° - 24°C Bin

                                            -2 std      impact       +2 std
Temperature          and Bank Deposits: Urban
Estimated Response Function Between Temperature Exposure and Log Ban
Deposits
Bank     Per per
     deposits Capita,
                 capitaUrban  Areas
                        in Urban areas
     0.020

     0.010

     0.000
             
    -0.010

    -0.020
              Estimated Impact of a Day in 15 Temperature (C) Bins on Log Bank Deposits Per Capita,
                                       Relative to a Day in the 22° - 24°C Bin

                                            -2 std      impact       +2 std
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
Implications of Climate Change I
    • We have documented a large reduced-form
      impact of both temperature and rainfall
      extremes on mortality in India from 1956-2000
    • Looking into the future: As India’s climate
      changes throughout the 21st Century, what are
      the implications for mortality?
        • Clearly one has to be very skeptical of the use of
          reduced-form estimates based on the past to
          predict the future.
        • But it is unclear what the alternative is.
        • Under most scenarios, our estimates (based on
          short-lived shocks) will place an upper-bound on
          the effects of a long-run change.
Implications of Climate Change II
    • Climatological models predict ∆Td (and
      ∆RAINd , though these are more controversial)

    • We use our earlier estimates of the mortality
      consequences of weather variation to estimate
      the mortality consequences of predicted ∆Td :

            X                    3
                                 X
    ∆Y
    dd =          θbj ∆Tdj   +         δbk ∆1 {RAINd in tercile k}
              j                  k=1

    • Report pop-weighted average of these
      district-level impacts.
Implications of Climate Change III
    • Feed in 2 standard C.C. models:
        1. Hadley Centre’s 3 A1F1 (corrected) model and
           NCAR’s CCSM 3 A2 model
         • Both are ‘business as usual’ scenarios (no CO2
           mitigation)
         • Both do not include ‘catastrophic scenarios’
           (Himalayan glaciers melt, monsoon terminates, sea
           level rises, more cyclones)

    • Details:
        • Models simulate full daily time path of temp. and
          rain from 1990-2099
        • Different time paths for each district in India
        • Define ∆Td ≡ Td2070−2099 − Td1957−2001 etc
        • Compute ∆Y dd for each district d and take
          population-weighted average
Predicted Change in Temp. Distribution
 60

 40

 20

  0
       
 -20

 -40
                             Predicted Change in Distribution of Daily Mean Temperatures (C),
                                        Change in Days Per Year in Each Interval

                                              CCSM 3 A2        Hadley 3 A1FI, Corrected
Climate Change and Mortality
                     dd = P θbj ∆T j + P3 δbk ∆1 {RAINd in tercile k}
Percentage impacts: ∆Y     j      d     k=1
in 2070-2099 (on average)
                                                                            Temperature
                            Impact of Change in Days with        Total
                                                                                and
                                    Temperature:              Temperature
                                                                            Precipitation
                                                                Impact
                          < 16 C     16 C‐32 C       > 32 C                   Impacts
                           (1a)        (1b)           (1c)        (2)            (3)
Based on Hadley 3, A1F1
    Pooled                 ‐0.010       ‐0.139       0.659       0.510         0.462
                          (0.030)      (0.045)      (0.126)     (0.125)       (0.142)
    Rural Areas            ‐0.030       ‐0.164       0.853       0.658         0.617
                          (0.039)      (0.055)      (0.153)     (0.126)       (0.173)
    Urban Areas            0.036        0.013        0.090       0.112         0.116
                          (0.033)      (0.058)      (0.105)     (0.101)       (0.116)
Based on CCSM3, A2
    Pooled                 ‐0.010       0.039        0.145       0.176         0.116
                          (0.013)      (0.042)      (0.028)     (0.061)       (0.084)
    Rural Areas            ‐0.015       0.071        0.189       0.245         0.207
                          (0.016)      (0.049)      (0.035)     (0.074)       (0.099)
    Urban Areas            0.009        0.016        0.028       0.052         ‐0.078
                          (0.013)      (0.042)      (0.022)     (0.054)       (0.092)
Climate Change and Mortality
                    dd = P θbj ∆T j + P3 δbk ∆1 {RAINd in tercile k}
Percentage impacts: ∆Y    j      d     k=1
All India
                                                                            Temperature
                            Impact of Change in Days with        Total
                                                                                and
                                    Temperature:              Temperature
                                                                            Precipitation
                                                                Impact
                          < 16 C     16 C‐32 C       > 32 C                   Impacts
                           (1a)        (1b)           (1c)        (2)            (3)
Based on Hadley 3, A1F1
    2010‐2039              ‐0.009       0.027        0.057       0.075         0.037
                          (0.014)      (0.026)      (0.012)     (0.039)       (0.054)
    2040‐2069              ‐0.011       ‐0.013       0.270       0.246         0.205
                          (0.025)      (0.028)      (0.050)     (0.069)       (0.086)
    2070‐2099              ‐0.010       ‐0.139       0.659       0.510         0.462
                          (0.030)      (0.045)      (0.126)     (0.125)       (0.142)
Based on CCSM3, A2
    2010‐2039              ‐0.006       0.082        ‐0.073      0.003         ‐0.054
                          (0.010)      (0.016)      (0.015)     (0.018)       (0.041)
    2040‐2069              ‐0.008       0.097        0.002       0.091         0.031
                          (0.006)      (0.023)      (0.005)     (0.023)       (0.050)
    2070‐2099              ‐0.008       0.039        0.145       0.176         0.116
                          (0.013)      (0.042)      (0.028)     (0.061)       (0.084)
Outline of Talk
  Background
     Weather and Human Health
     Simple Theoretical Framework
  Data and Empirical Strategy
  Results:
     Weather and Death
     Weather and Income
  Implications of Climate Change
  Conclusion
Summary: Aim to Make 3 Contributions
   1. Document new facts about the role that
      temperature extremes play in the mortality
      risks faced by LDC residents.
        •   One SD more degree-days (over 32 C) leads to 9
            % higher crude death rate. (20 × US effect.)
   2. Understand the plausibility of various
      mechanisms that could explain these effects.
        •   Cluster of findings (rural not urban, GS not NGS,
            mirror effects in rural incomes) consistent with
            these effects working through agricultural income.
   3. Use these estimates to inform estimates of
      health costs of climate change scenarios.
        •   No-adaptation (upper-bound) costs: annual
            mortality rate to rise by 10-50%.
Summary: Implications
   • Smoothing of marginal utility in rural India
     seems far from complete.
        •   Aggregate (between-village) income/health shocks
            are not being insured in the way that idiosyncratic
            (within-village) shocks are (eg Townsend (1994)).

   • Weather-dependent income transfers are likely
     to save lives cheaply.
   • Temperature extremes seem to matter a great
     deal already for rural incomes and health
     status.
        •   And this will only get worse (or costly adaptations
            will be incurred) if temperatures rise in the future.
You can also read