This is an evaluation of Liang et al (2021).

Summary measures

Overall assessment

Answer: 88

Confidence (from 0 - 5): 5

Quality scale rating

“On a ‘scale of journals’, what ‘quality of journal’ should this be published in?: Note: 0= lowest/none, 5= highest/best”

Answer: 4

Confidence (from 0 - 5): 5

See here for a more detailed breakdown of the evaluators’ ratings and predictions.

Written report

[Evaluation manager/editor’s note: I made some very copy-editing corrections below.]

Thank you for the opportunity to read this paper. I truly enjoyed reading it.

This work investigates the relationship between economic growth and biodiversity in the United States. It leverages an extensive database that collects an array of time-consistent species surveys at different locations to build yearly biodiversity indices for eight taxa across 50 years. Using fixed-effects regressions, the authors relate the biodiversity outcomes with state-level GDP measures in an unbalanced data set. They find a robust negative correlation between economic growth and biodiversity outcomes, with an elasticity of 1.5 to 3.5%. Although varying in size, estimates are consistent across taxa, production industries, biodiversity indices used, or alternative observational units (e.g., counties, eco-regions). The paper further delves into a causal identification of the impacts of economic growth on biodiversity using exogenous variations in military spending in a shift-share IV approach. The second part of the paper further explores two potential mechanisms, pollution and land-use change, highlighting their role in the decline of biodiversity across the United States.

The paper makes a highly valuable contribution to understanding the trade-offs between economic growth and its impacts on biodiversity. Furthermore, the authors spearhead a new spatially granular database on biodiversity measures that has received little attention from environmental and ecological economists. I am convinced that readers will be eager to follow and use the BioTIME database to understand the economic and political impacts on biodiversity.

Although I am fully convinced about the timeliness and importance of the paper, I have some comments related to framing and the empirical approach of the mechanism analysis. Nonetheless, many of the raised points can be easily addressed. If the authors pursue these improvements and others they receive elsewhere, I believe they will make a highly valuable contribution to the environmental economics literature.

Major Comments

1. The paper combines a multitude of empirical approaches and units of analysis. Although each approach has its advantage, switching back and forth between different approaches decreases the manuscript's readability. Although it is undoubtedly a matter of writing style and a question of the target journal, I believe that reordering some sections could ease the reading experience and streamline some of the arguments.

a. I suggest to order the arguments that are also linked to empirical approaches as follows: A) Correlation between biodiversity and GDP using OLS, B) Establish causality with IV; C) Robustness on measurements biases using the IV approach (instead of OLS); D) analyzing heterogeneities (e.g., by taxa, EKC, distributional, etc.) with the IV approach (instead of OLS); D) Pollution channel analysis with Wind-direction-IV E) Land-use channel analysis using OLS correlation: Though I would put this to the appendix or skip it entirely (see comments below). F) Environmental regulations channel analysis using TWFE. Though I would leave out the fuzzy regulation-GDP-biodiversity link G) Protected area channel analysis. However, I would replace it with a panel on protected area creation and use a TWFE estimator (instead of a heterogeneity analysis with interaction terms.

b. Before delving into potential heterogeneous results in section 3.2, I would address the potential biases. Section 2.3 (Title should be: “Potential sources of measurement bias”) deals with the potential measurement biases, while only section 4.1 addresses the omitted variable bias and reverse causality bias. I think it is best to address all biases next to each other. I would first describe the empirical strategy and then, directly after, the potential a) measurement biases and b) the potential omitted variable and reverse causality biases together.

c. After addressing potential biases I would create two separate chapters a) on heterogeneities (now section 3.2) and on potential channels (now sections 4.2 and 4.3).

2. After the main results, you choose to contrast two samples: All taxa vs non-bird taxa. The comparison group is missing. One should present results for either both groups, i.e., Bird taxa + non-bird taxa, or the three samples, i.e., all, bird, and non-bird. Results are mostly better for non-bird taxa, while results for the full sample are often insignificant. This bears the question if results for birds only are always insignificant or even point in a different direction.

3. Results on sectoral GDP (Table 2) are highly biased and, therefore not very informative. Regressing biodiversity on GDP is already subject to selection bias (though I think the paper is making a good point and presents a very valuable first analysis). Differentiating between sectoral GDP growth and jointly using all indicators in one regression must increase the problems of attribution, multicollinearity, reverse causality, and omitted variable bias. You do recognize these issues when discussing the positive estimate for agricultural GDP… I suggest excluding this argument (analysis) from the paper or trying a different empirical approach. Maybe it is possible to combine the IV strategy in combination with pre-period sectoral shares?

4. In the pollution channel analysis (section 4.2), I would like to see a more detailed reasoning for the selection of contributing counties. First, why are counties in a 300 km radius excluded? It would seem that close pollution sources are the most important. Second, the LASSO regression selects counties (see Figures 5a and 5b) that are thousands of miles away. Could it be that those relations are a statistical artifact and do not represent a true physical impact? I know that pollution from massive forest fires can travel large distances (e.g., Indonesia, Canada), but maybe you can back up your argument with natural science literature. On the other hand, far-away pollution sources might not pose a problem in your analysis as the IV is weighted by the inverse distance. Nonetheless, Figure 5 and your description is misleading at first sight.

5. The analysis of land-use policies (section 5.2) could be improved by using a panel of protected areas. I am unfamiliar with the expansion of protected areas in the United States. Still, if there is a significant expansion of protected areas in the vicinity of the sampling locations, that could be exploited in a quasi-experimental setting. You could use the panel of new protected areas in a similar empirical framework as the analysis on pollution attainment areas using TWFE.

Minor comments

1. I understand that many interdisciplinary journals prefer a graphical depiction of the main results, though if you choose an economic journal, I suggest showing results in Table format. I find it easier to read and understand the empirical strategy and the number of fixed effects and observations in a Table format.

a. The main results table could present the results of Figure 3a potentially using 3, 6 or 9 columns differentiating between all, bird, and non-bird taxa. It is also the set-up you choose to carry on in the text. Therefore, it might be more transparent to make the distinction right away. Figure 3b could pose as additional information for the appendix rather than a pre-step to show before summing everything up into bird vs. non-bird species.

b. OLS and IV estimates are easier to compare in Table format - Figure 4b does not easily convey how good the IV strategy is (first stage) and how small or large the difference between the IV and OLS estimates is. Table 3 is much more transparent but also mixes reduced form with IV estimates. In general, I don’t think it is not necessary to present a complete graph with a single line-plot just to show one point estimate.

2. I think the analysis of pollution regulations (section 5.1) can be cut to the TWFE estimation only. The exercises with GDP and the repeated comparison of overall vs. mechanism effects seems to massage the data a bit too much.

3. It might help to contrast the urbanization (section 4.3) and the construction-sector GDP estimates (section 4.3) next to each other.

4. Table 2 does not describe which FE are used.

5. You are missing a dotted zero line in Figure 3d

6. Figure A6 left has a wrong negative sign on the coefficient.

7. On page 15, you write, “ηt denotes year fixed effects to capture common shocks such as national recessions”. Year fixed effecs also capture common changes in federal environmental policies, regulations, laws, financing of protected areas, overall enforcement budget, etc.

8. Please provide a more straightforward argument why a state-level analysis is the preferred spatial unit. Later on, you sometimes shift the unit of analysis, which adds unnecessary to the complexity of the manuscript. I would recommend to choose one unit of analysis for the main text, as there are already many variations in the empirical strategy.

9. How is Table A.2 a panel regression if its outcome is constant over time (columns 2-6)?

10. Sometimes the paper mixes the data description with the estimation strategy: E.g., on page 9, “As previously noted, in all regressions we include … “. I would try to streamline the text for easier reading.

Evaluator details

1. How long have you been in this field?

I started my Ph.D. in 2012. Thereby I am already 11 years in the field.

2. How many proposals, papers, and projects have you evaluated/reviewed (for journals, grants, or other peer-review)?

I have reviewed 13 papers for 8 journals.