Category Archives: Environment and Devlopment



S. Galiani, P. Gertler, & E. Schargrodsky

Journal of Political Economy, Vol. 113 No. 1 (2005)

A Short Summary

Principal Research Question and Key Result Does the privatization of water services have an effect on health outcomes? Using data from the privatization of the Argentinian water companies the authors show that child mortality decreased differentially in those that privatized as opposed to those that did not.


Theory There may be efficiency gains from privatization yet these do not necessarily translate into improved health outcomes, or to alleviate poverty. In particular private water companies may not take into account the significant positive/negative health externalities from providing good/poor quality and so may underprovide decent quality water. Also the poor may be hurt if water companies target only affluent areas for clean water, or if they overcharge which reduces the availability of good quality water to poor households.


Motivation Privatization is often touted as a means of delivering services that have traditionally been the domain of the state (due to natural monopoly, inelastic demand, and externalities).


Data Data are from Argentina, which in the 90s as part of the structural reform decentralized many services to local governments are allowed them to privatize them. Some local governments did privatize and others did not which allows for exploitation of data in a difference in difference strategy.


Strategy Difference in difference. The plot the child mortality rates in privatized, non-privatized localities and the visual representation lends some support to the parallel trends assumption, although it is far from perfect parallelism.
Results Privatization of the water services is associated with a 5.3% reduction in child mortality in the average privatizing region. This is robust to the inclusion of GDP, inequality measures, public spending measures (to capture the effect of the other programmes occurring at that time) and the identity of the political party.


Robustness Evidence for their conclusion is added to when they disaggregate results by type of disease, and find that effects are significant for infectious and parasitic diseases (i.e. from water) but no other types of disease/accident.

They a poverty indicator variable and find no relation between death and water in richer areas, but a significant relationship in poorer areas


Problems So much was going on in the 90s Argentina, that even despite some nice looking parallel trends evidence, it is hard to conclude that the water system alone was responsible for the observed outcomes on child mortality. However, that the reduction was primarily happening through lessened water borne health issues, there is some extra support that it was the privatization that caused the improved outcomes.

There is still a problem of omitted variables in my opinion. Whilst they do control for the identity of the political party, they cannot control for the ability of the local government. If the privatizers were governments that had a certain level of administrative capacity that was needed to successfully privatize, and that ability also allowed them to benefit differentially from the other decentralizations (such as health care) that were occurring at the time, then the parallel trends assumption is violated, as the outcome under governments with more ability would have developed differently from the non-privatizing irrespective of the fact of privatization of the water company.

There are clear external validity problems here, there is no telling whether privatization could work in other settings as incentives facing companies, the extent of the market, existing infrastructure, identity of private investors, means of privatization will all play a role in determining outcomes in an individual setting.


Implications Privatization may increase access, efficiency and service provision when it comes to services such as water. They cannot find evidence that such privatizations hurt the poor disproportionately; in fact they may stand the most to gain. However, this was only one privatization of many, so it is not clear how the privatizations interact, or if all such industries would benefit from privatization.





R. Burgess, M. Hansen, B. Olken, P. Popatov & S. Sieber (2012)

A Short Summary

Principal Research Question and Key Result In the management of natural resources do incentives facing local politicians/bureaucrats play a role in determining patterns of resource use? If districts engage in Cournolt type competition, then increasing the number of districts should reduce the price of the resource, and also increase its depletion. The context is Indonesia. The key result is that rainforest depletion increases with the number of districts per region, and prices are reduced which is consistent with the idea of Cournolt competition between districts for rents. If alternative rents are available, such as oil royalties then the effect on the forest is diminished, although this effect tapers out after local elections which indicates that rent seeking politicians may be more likely to stand in regions with oil royalties and potential rents from forest depletion.


Theory Competition between jurisdictions for rents from the forest mean that increasing the number of districts in a region should decrease the price of logs, and increase the amount of logging. This is because an official’s market power diminishes as the number of competitive districts increases, so he has to lower the price of extraction and increase the amount of it in order to sustain rent extraction levels. As alternate sources of rent are made available (such as oil) the cost of being found cheating (expulsion) increases, and so he decreases rent extraction.


Motivation Deforestation is very rapid in the tropics and contributes more to greenhouse gas emission than the whole world’s transport sector. There are also biodiversity loss issues. Yet the vast majority of these forests are managed by national governments which rely on official to monitor and implement the enforcement strategies for illegal loggers. If incentives play a part in the unsanctioned deforestation, then how deforestation reacts to those incentives faced by politicians are important for designing power structures that oversee the forest, and also so that alternate mechanisms of protection can be identified.
Data Indonesia is the setting. This is ideal as it has a huge amount of tropical forest, and also it has recently experienced a variation in the number of districts due to t a huge decentralization of power since the 90s. Data on deforestation are created from repeated satellite images that track how much forest is being lost on a 250m by 250m lattice that produces 34.6 million pixels.

They use data on the timing of the district splits to see if there are changes in forest outcomes. These district splits were not driven by availability of forest rent extraction.


Strategy They do a Mac Likelihood estimator whereby the count of the number of deforested cells is on the left, and number of districts in the region and fixed effects are on the left. They do a similar OLS for the log price of timber.
Results Annual deforestation increases 3.85% when another district is added, and this effect is similar in regions where some logging is legal and all logging is illegal (so it is not just an increase in legal permits being issued or permits being renegotiated as between the district and the central government).  The effects appear to be even larger when lagged right hand variables are added, so the effects in three years are even stronger than in the contemporaneous year. This indicates that it is not just a problem of new governments being less experienced at enforcement, as after three years the effect is increasing by which time they should be up to speed with enforcement methods.

Prices decrease by around 1.7% when a new district is added although this is not significant. In fact, the results only become significant in the medium term when variables are lagged. This evidence is not that strong.

When oil and gas revenues are included, an extra $1 per capita in oil reduces deforestation by 0/3%, and this occurs in both the illegal and legal logging zones. This effect tapers after a few years. Specifically they interact the oil variable with a post-election variable, and then the deforestation coefficient is no longer significant, which potentially indicates that rent seeking politicians deliberately target districts with oil revenues, and they are also more callous regarding deforestation.


Robustness They present results per island which are similar. They disaggregate by type of forest zone and find there are still strong effects in zones where all logging is illegal.

They look separately at the districts that split off as opposed to the parts they left behind. If enforcement was the issue then there would be an increase in the new part and little change in the old part whereas in fact the data show that there is little difference between the old and the new in terms of they both increase deforestation, and if anything the old part slightly more.


Problems There are some, but they are not of major importance for the purpose of this summary.
Implications Many people argue that putting resource management in the hands of local communities is the best way to protect the resources. This paper offers evidence that such decentralization does not always work in practice. Where jurisdictions are big enough to have market power for wood, and politicians can earn rents from illegal extractions, decentralization can actually increase deforestation. The bottom line is that local political structures seem to matter for environmental outcomes.

There could be the possibility of better top down monitoring and enforcement (although scope for this may be limited in LDCs) or to provide politicians with alternative sources of rent, such as cash to protect the forest.





R. BURGESS, O. Duschenes, D. Donaldson & M. Greenstone (2011)


Principal Research Question and Key Result Do weather conditions affect mortality, and if so through what mechanisms does weather so affect death rates? Using data from India, the results indicate that weather does affect mortality, and that this appears to occur through the dent that poor weather makes in agricultural incomes. Evidence for this mechanism are that there is no observed effect in urban areas and also that only poor weather in the growin season is capable of affecting mortality.


Theory The authors propose a theoretical framework. An individual gets utility from the value of consumed goods times the probability that he will be alive to consume them. That probability measure is related to two things, consumption of health commodities (broadly defined), and the weather. The ability of the individual to consume health commodities is constrained by his income which itself is a function of weather (inasmuch as he is dependent upon agriculture). Thus the weather may have a direct effect (heat stress) on the chance of survival, or an indirect effect (which is the effect it has on income and hence the ability to consume health commodities).

This model therefore predicts a minimal weather/death relationship in urban areas, a stronger relationship in rural areas, and a particularly strong effect when the weather shock occurs in the growing season as this will most impact the incomes of the individual farmers.


Motivation Structural changes in the developed world mean that people are able to use the resources at their disposal in order to protect themselves from weather shocks. Additionally most developed world individuals are not dependent upon agriculture as the prime means of incomer generation, unlike in the developing world. Whilst there may be informal institutions at the village level in rural economies that enable a certain amount of income smoothing, when the shock affects a wide range of villages all in the same way, there is little scope for smoothing incomes and this may result in increase mortality.  Understanding the link between weather and death is important for three reasons: firstly in order to effectively design policy. Secondly for understanding what may happen in the event that the world experiences significant warming in the coming years. Thirdly, if poverty is affected by weather then counteracting the negative effects of weather shocks could be one way to fight poverty.


Data The context is India. This is a good setting to examine the above hypotheses as there is a sharp rural/urban divide, with 76% agricultural income in rural areas, and only 7% in urban areas. Also food is a much smaller part of the overall budget in urban areas.

Mortality data are from the Indian Vital Statistics, and they construct an infant mortality rate, and an all ages rate by urban and rural areas separately (they are probably underestimates due to reporting problems). Data for weather are daily reports of a 1 degree by 1 degree lattice which are then matched to districts. There are 15 temperature bins with temps less that 10 being the lowest bin, then 13 2 degree intervals up to 36 and over. They have an alternate measure which is the cumulative number of days that temperatures are over 32 degrees.

They also have rainfall data, agricultural yield data, agricultural prices data, agricultural wage data, manufacturing productivity data, CPI data and manufacturing wages data.

They have panel of district level observations for the period 1957-2000


Strategy  It’s a pretty simple regression of log mortality by district and year and a measure of the number of days in a year that the temperature was in one of the 15 different temperature bins. Omitting the central bin. Rainfall is modeled such that the variable is high, medium or low depending on how it relates to a normal year. District fixed effects strip out idiosyncratic but time invariant district effects (such as supply of medical facilities), and time fixed effects strip out district consistent time varying effects (such as health reforms). There are also separate time trend dummies for each of India’s climactic regions in order to control for time varying factors that also vary across regions.

They use the temperature bins in order that they might estimate different coefficients for each 2 degree increment. This means that the specification is more flexible than if all the temperatures were pooled, and this is done in recognition that the relationship between weather and death is most likely not linear.

In a spate strategy they use the other cumulative measure of days in which the temperature was above 32 degrees. Errors are heteroskedaticity robust and clustered at the district level.


Results More hot days is associated with significantly more death. One extra day with mean temperature is associated with a 0.75% increase in mortality.

When the regressions are estimate separately for rural/urban areas, they find no significant relationship between weather and urban mortality, and a strong response for the rural areas. Coefficients in the temperature bins below the reference category are small and insignificant, but as the temperature climbs so do the coefficients and the significance. One extra day in the +36 degree bin leads to annual mortality rate increase of 1% in the rural sector. This is true both for all ages and for the infant mortality measures. This seems to indicate that urban populations are better equipped to adapt to temperature shocks, or more plausibly that there is a weaker connection as there is a lower dependence on weather contingent forms of production.

Results are similar for the cumulative measure.


  • They interact the rainfall and temperature variable and see no significant movement in the cumulative coefficient which indicates that the hypothesis that hot years kill people because they create perfect hot and wet conditions in which disease can blossom, is not supported.
  • Using data on when the monsoon usually arrives by district (and hence when the growing season arrives) they can assign weather days by whether they were in the growing season or not. They run separate regressions of mortality on the growing/non growing cumulative measure weather variable. Non-growing season coefficients are always close to zero and never significant, whereas they are large and significant in the growing season. Using lagged weather variables they find that the effects are still present even three years after the fact. There are no effects in urban areas.
  • They use the cumulative measure to estimate the model for 4 11 year periods to check that results are not being driven by earlier, pre agricultural revolution data. They find that although the relationship is lessened in more recent years, there has been little movement since the late 60s.
  • Agricultural indicators such as yield, wages and prices respond to inclement weather in exactly the same way as mortality which is indicating that the indirect effect of weather does indeed operate through agricultural incomes.  Yet when they use similar measures of urban wages and productivity they find no significant results.
  • They feed their results into three different climate models that predict how temperatures will change in the coming years, and find that if there is no adaptation, mortality rates could increase by 10-50% which is very high.
  • If weather shocks affect incomes and incomes affect health commodity purchasing possibilities it could have been nice to see some evidence that purchases of essential foodstuff, medicines, malnutrition etc. is affected by weather shocks. Additionally, if data on morbidity etc. could be used to show similar trends, this would have leant extra credibility to the findings.
  • The data are inter annual weather changes, so the results are necessarily short term, which does not say much about the long term effects of weather.
  • There are significant issues with feeding in past results to model future responses as they do when thinking about climate change. People will adapt to change, perhaps through urbanization, migration, technology etc. The authors do recognize this however, but at this stage it is not hugely clear what could be done.


Implications Weather shocks appear to affect rural populations especially hard. Whilst there is evidence of informal institutions allowing for the smoothing of idiosyncratic within village shock (Townsend), it appears that aggregate across village shock is not as easy to smooth away, and hence there is an effect of weather on mortality.  Given that these informal local institutions are failing to support farmers in crisis, there could be a role for formal state level institutions in preventing weather related death, such as health policy to help people cope with weather shocks, income/nutrient transfer programmes based on weather outcomes etc.