Write a paper on Causal Mechanisms of Protected Areas on Poverty
This study focuses on answering the question of how protected forest areas reduce poverty.
Studies done in the past have already claimed that protected areas have played a role in reducing
poverty in the surrounding areas, but they do not provide the mechanisms that lead to these
poverty reductions. Using Andam et al. (2010) data set, the study identifies three mechanisms
that might lead to poverty reductions in Costa Rica due to protection of forested land: tourism,
infrastructure development and ecosystem services mechanisms. To estimate the impacts of these
three mechanisms, the manuscript uses proxies: creation of park entrances, changes in road
networks and changes in forest cover. Since, mechanisms are affected by the treatment
(protection) treating mechanisms as an independent variable and controlling for them in a
regression will generally produce biased results. The main aim of the study is to estimate the
Mechanism Average Treatment Effect on the Treated (MATT). The manuscript provides a
detailed theoretical econometric foundation for the unbiased causal estimates. It uses a two stage
estimation strategy, adjusting for bias and uses robust standard errors to deal with
heteroskedasticity. The results show that tourism plays a vital role in reducing poverty relative to
the other two mechanisms, which are found to be insignificant and close to zero. Several
robustness checks produce similar results.
I view this manuscript as an interesting study, with necessary attention given to the
methodology section. The theoretical background provided to develop the causal relationship
between the mechanisms and poverty is rigorously explained. Particular attention is given to the
assumptions being made and the estimate being evaluated. The paper is well structured with no
unnecessary descriptions and very few repetitions. The paper makes significant contributions to
the literature; however, some key issues have not been addressed. Besides not addressing these
ECN 331 – Empirical Methods
Referee report example
2
issues, the author/s should also consider further discussion/explanation of the main ideas
(discussed under minor comments). The manuscript also needs to be edited in certain places
(discussed under minor comments).
Major comments:
1) Clarity and style: The introduction is precise and clearly states the motivation. However, an
explanation of how the paper progresses at the end of the introduction should be provided.
Giving more information on the data set being used would be helpful. Not enough credibility is
provided for the proxies that are used, particularly for infrastructure development. The discussion
on standard errors could be placed in an appendix as the reader does not get to the results until
page 21.
2) Explicit assumptions: Reductions in poverty near protected areas might reduce poverty in
other areas nearby due to a boost in economic activity in the region. Poverty might also be
affected if protection encourages migration to the towns closer to the protected lands. This might
create further financial differences between towns near protected and unprotected parcels.
Hence, poverty is being affected not due to the mechanisms but through other sources. I believe
the authors are implicitly making the assumption that no other sources are at play here. The main
problem with this assumption is that its violation, in the context of this study, is not really clear
in which direction would it bias MATT. Another assumption implicitly assumed is that of
common support: ( | ( ) ( ) ) . These assumptions should be
explicitly stated.
ECN 331 – Empirical Methods
Referee report example
3
3) Reversal of sub-group rankings: Lands where protection had the greatest effect on
deforestation are different from parcels of land where protection had the greatest effect on
poverty. This was one of the findings of Ferraro and Hanauer (2011). Protection reduced
deforestation mostly on those parcels which had high land use capacity, whereas poverty
reduction was found on lands with low land use capacity. Lands of different productivity are
used in the matrix of covariates and each is given equal weightage. This might bias the results.
Since, low land use capacity sees the most poverty reduction; more weightage should be given to
this covariate. On the other hand, less or no weightage should be given to high land use capacity.
There seems to be a trade-off between deforestation and poverty alleviation.
4) Give details regarding follow-up surveys. The authors used follow-up surveys for the main
analysis. What was the attrition rate? Were there incentives associated with survey responses?
5) Was the sample truly random? Did the researchers in any way influence who was given the
treatment and/or who was placed in the control group? The authors need to elaborate on this?
6) Results. The authors stated that the income effect was 5%. However, Figure 3 shows that the
incomes were decreasing throughout the program. Was this taken into account in the
computation of the income effect?
7) Tables 3 and 4 provide similar information and may be combined into one table.
ECN 331 – Empirical Methods
Referee report example
4
8) Discussion. There was no mention of the limitations of the study, one of which is the
apparently high dropout rate. Also, mention how your results compare to (reference given to
author) another study which was published very recently.
Minor comments:
1) No abstract is provided with the manuscript. It would be nice if the audience is made aware of
how the paper is structured at the end of the Introduction.
2) On page 8 it is stated that if mechanism effect of ecosystem services is negative, one of the
explanation for this outcome would be that protection has “no effect on poverty or their positive
effect is smaller than the negative effects”. Isn’t this misleading, since a negative coefficient for
this mechanism would imply that ecosystem services actually increase poverty?
3) The mechanisms help to explain only 60% of the effects on poverty. There is no discussion of
other mechanisms. Moreover, only those mechanisms are studied in this paper that affect poverty
positively. What about factors that might have negative effects on poverty? The MATT
coefficient for infrastructure development was not significant. This might be due to the proxy
that is being used. Construction of roads might not be capturing all the effects. Maybe roads are
not being constructed because of strict protection laws. There might be other infrastructure
development that might capture the effects of this mechanism on poverty. Secondly, after
matching the improvement in mean difference for roadless volume is only 58.75%. Shouldn’t
this be a concern?
ECN 331 – Empirical Methods
Referee report example
5
4) Footnote 7 on page 13 does not provide the correct definition of ATT. The definition gives an
unconditional mean. ATT should be conditional on the treatment and covariates. Hence
ATT should equal
[ ( ( )) ( ( ))| )] [ ( ( )) ( ( ))| )]
5) Matching technique: Mahalanobis covariate matching with replacement technique is used for
covariate balancing. One reason for using this technique is it was used in Andam et al. (2010).
Providing some minor details like
(i) The Mahalanobis metric is ( ) ( ), where is the variance-
covariance matrix.
(ii) It generates good matches and is recommended for achieving balance on higher-order
moments.
about this technique might be beneficial.
6) On page 22 results of Table 3 are being discussed. In the last paragraph for the forest cover
proxy, numbers of Table 4 are used. For example the MATT estimate from Table 3 is 0.103 and
not 0.099 (which appears in Table 4). Similarly, had protection not affected deforestation, 23%
of the protected census tracts, on average, would have been deforested between 1960 and 1986.
ECN 331 – Empirical Methods
Referee report example
6
The expression “over 22%” in this paragraph I think is referring to the counterfactual mechanism
0.223 that appears in Table 4.
7) There are a few editorial issues: no space is given between certain words. On page 18 line 1,
there should be space between “obtained from”, in the line after equation (13) there should be
space between “of times”, in footnote 10, there should be space between “is available” and
finally on page 23 after equation (23), there should be space between “ ̂ represent”. In equation
(5), an is missing and below equation (10) it should be not . It is not clear in Table 1 that
the summary statistics for land productivity are for which year.