Education and Crime across America: Inequity’s Cost

Round 1

Reviewer 1 Report

Manuscript Title: Education and Crime Across America: Inequality’s Cost

Manuscript ID: socsci-1226666

Reviewer Comments:

This study combines datasets from 2003 to 2018 to examine the relationship between school district spending and crime. Consistent with past research that shows how education can reduce crime, the author(s) find a negative relationship between school district spending and crime. More specifically, results suggest that for every $1,000 increase in school district spending on education, crime drops (on average) by 2.8%. The negative relationship between school district spending and crime was more pronounced among property crime compared with violent crime. Below are my comments about this manuscript, which I hope are of some assistance to the authors.

With regard to the review of literature, I recommend that the author(s) include some discussion pertaining to the theoretical reasoning for why education is strongly associated with decreases in crime. In its current form, the focus is on past studies that show how beneficial education is for reducing crime; incorporating some discussion of why these relationships exist would strengthen this section. The authors could also mention how past studies find a strong, negative relationship between education and crime at both the micro and macro level.

Turning to the data and methods, the authors mention on page #4 that they used multiple imputation to address missing data, and they combined their results from 5 imputed datasets. I would recommend that the authors incorporate more imputations into their final analysis, as research is increasingly recommending a higher number of imputations when using multiple imputation (see, for instance, Graham, Olchowski, and Gilreath 2007).

On page #7, I recommend that the authors use alternative language when discussing their findings regarding the relationship between race and crime. Perhaps something along the lines of, “every ten-percentage point increase in the proportion of individuals in a city or town who are non-Hispanic White…”

Overall, I believe that the discussion section is strong. I would, however, encourage the authors to spend more time talking about the implications of their findings rather that re-reviewing the specifics of their findings. At some places, the discussion feels a bit repetitive and too similar to what was already mentioned in the results section. Indeed, some recap is expected, but perhaps this could be streamlined a bit more in this section.

Minor Comments:

In the abstract, I recommend removing the word “positive” from the last sentence. I suspect that the authors are alluding to the fact that education is beneficial for reducing crime, but in this context, it seems to suggest a contradicting direction in the relationship between education and crime.

On page #12, line #292, it looks like the word “home” should be “hone.”

Recommended References:

Graham, John W., Allison E. Olchowski, and Tamika D. Gilreath. 2007. “How Many Imputations are Really Needed? Some Practical Clarifications of Multiple Imputation Theory.” Prevention Science 8:206-213.

Author Response

Please see the attachment.

Reviewer 2 Report

Please see my comments below. I am also attaching a PDF with all my comments.

1. Mixed-effects model: The paper uses a linear mixed effect model to explore the interactions between education expenditure and crime. The authors should explain why this model is appropriate to answer their question, since this is a different methodology to what has been used in the literature (i.e., fixed effects model, instrumental variables, among others).

2. Literature: There are at least two articles that are more closely related to the paper given the crime measure being employed, which I consider should be cited. \cite{Lochner_Moretti_2004_AER}, \cite{MerloWolpin_2015}, \cite{mancino2016separating} also employ self-reported crime data to study the interaction between educational attainment and crime, as opposed to exploring arrest or conviction. This can explain the differences in the reported effects.\\    

3. Objective and discussion of results: Although the main objective of the paper is to understand the relationship between education expenditure and crime, the paper briefly discuss the corresponding findings in the results section, without pointing to possible mechanisms. Only in the conclusion, the paper discuss possible mechanisms driving the results. I would move this discussion to the main text, instead of just discussing them in the conclusion.\\  

4. Model selection: The paper starts with a baseline model and then add explanatory variables following their impact on the AIC. First, given that the variable of interest is education expenditure, I consider that the baseline model should be one that only includes such variable (and maybe population and population density). The paper should then use that estimate as benchmark and analyse how the coefficient on the variable of interest changes as new explanatory variables are added to the regression model. Second, I consider that the order in which the control variables are added in the model is not as relevant. What is more important is what the final model we choose is and how the effect of education expenditure on crime changes across specifications. Thus, showing the coefficient on education expenditure across specifications (together with the AIC) is key (maybe this can be better summarized in a table). Furthermore, it would be interesting to see which model is preferred if another information criterion is used instead, for example, the Bayesian Information Criterion, since the results can be quite different when different criteria are used.\\    

5. Dependent variable: The authors choose to add a constant of one to the crime rate. If the intention is to smooth the crime values, I consider that a better strategy is to take the natural logarithm of the crime rate. If I understand correctly, there are no zero crime rates in the dataset, thus this should not be a problem. Furthermore, I would suggest checking the robustness of the results to simply using the crime rate in levels.\\  

6. Missing crime data: The paper opts to impute missing crime data, instead of deleting missing values. Given the imputation procedure, it would be nice to check how sensitive the results are to deleting the missing values instead.\\

7. Discussion of results - poverty: While the main variable of interest is education expenditure (according to the introduction of the paper), the beginning of section 5.1 makes it seems like the proportion of children in poverty is a key variable as well. I suggest that either the authors discuss the relevance of this variable before, or omit figure 1c at all. Furthermore, the paper does not discuss the the raw correlations between the key variables of the model illustrated in figures 1a, 1b, and 1c. This important and should be used as motivating facts/ benchmark for the main results of the model.\\  

8. It is not clear from the text how a ``place'' is defined; does it refer to towns, cities, zip codes? It would help if the paper gave an example of what a place is.\\    

9. Model specification - Second crime model: I do not understand how the second crime model is specified. Is the dependent variable the crime rate in place \textit{p}, year \textit{t}, for crime type \textit{c}? In that case, that specification should contain and indicator for violent crime, as well as its interaction with the education expenditure variable. Is this what the paper does? Lastly, is the difference with the third crime model that the regressions are run separately for violent and property crime? The discussion of the different specifications needs to be clarified in the main text.\\  

9. Model specification - Covariates: There are a few surprising results that need to be discussed. First, the paper finds no significant effect of unemployment on crime. Is unemployment rate included in the selected specification? or is it dropped because it is not significant? In that case, by how much does the coefficient on education expenditure change when we include unemployment rate? \\   Second, the paper finds no significant effect of law enforcement on crime. This result is very surprising and should be further explored. For example, is law enforcement significant when the model does not account for per capita income? Can the results be driven by measurement error in the law enforcement data?\\  

10. Education spending interacted with crime type: The paper finds that, in most cases, education expenditure is not related to violent crime. One potential explanation for this result is that schooling can increase the concentration of young people, leading to more violent confrontations, and consequently causing increased criminal activity (Jacob and Lefgren, 2003).\\  

11. Range in education spending on crime: It is not clear to me how the range in education spending is defined. This should be clarified in the model, specially since it the results for violent crime change dramatically under that specification. Furthermore, the authors should provide some intuition behind the huge difference this specification has on violent crime. Having a better understanding of how the range variable is defined will certainly help in this matter.\\  

12. Comparison of results with the literature: The authors fail to compare how do the results in the paper differ from those in the literature. The paper should specifically compare the results with the papers that are more closely related (See listed papers above.)\\  

13. Conclusion: In line 377, the paper suggests that the crime data is not valid at all. Although I consider that this is a bit extreme, the paper makes it sound like we should not believe the results at all given the quality of the crime data.\\  

14. Conclusion: In line 307, the paper makes reference to a model that accounts for a three-way interaction between crime, race, and per capita income. However, I cannot find these results in the main text. It may be worth including them in the main text, together with a table of results.  

15. Minor - Effect of Education Expenditure: In future research, the authors could also try to break down the effect of education expenditure on crime. For instance, is the effect coming through expenditure on teachers, infrastructure, or other?\\  

16. Minor - Abstract: The results by crime type reported in the abstract do not match those in the text. \\  

Author Response

Please see attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

[My comments are attached in a PDF as well]

Summary

This paper combines a multitude of datasets to study the relationship between crime and educational spending in the U.S. from 2003 to 2018. The data combination allows for analysis of the correlations between criminal activity and school district spending, controlling for a rich set of covariates, fixed effects, and random effects. The results indicate a negative correlation between educational spending and crime. Within-city analysis demonstrates a positive correlation between educational spending and property crime, but a negative correlation between educational spending and negative crime.

Overall, this paper provides a sturdy empirical base to an interesting problem. An empirical framework to consider a causal question would greatly strengthen this work. Even if the authors intend to only consider correlations, more work is needed to connect this analysis to the literature and clarify the relevance of the results. More work to empirically consider different pathways would also bolster the work.

Comments

1. It is difficult to understand the connection between the papers cited in the literature review and this work. The authors are studying educational spending and criminal experience (the number of crimes in an area). The papers listed study educational attainment and criminal involvement. Of course these are related, but I wonder if papers in public, rather than in labor, related to educational spending and public safety would be better suited here. The authors need to clarify what papers, if any, exist that study the relationship between these two entities. Additionally, it would be helpful to discuss papers studying the effects of increasing educational funding on other outcomes besides crime and papers studying policies that reduce crime rates.
2.  There are many details about the data construction, but not as much information about the data itself. I think it would be more helpful to describe the variables, perhaps adding a table of descriptive statistics. Perhaps a table demonstrating which variables come from which datasets would be helpful.
3. What is the exact empirical approach? It would be helpful to the reader to have an equation of the models to be precise about the fixed and random effects and geographic units.
4. I think it would be useful to clarify what the experience of crime is. This is especially important when considering differences within municipalities. Does someone who lives in a high-income area of a municipality face the same crime rate as someone who lives in a low-income area? Redoing the data section to be clear about geographies and adding the equation of the models would help here as well.
5. While some potential mechanisms are touched on in the conclusion, I believe it should be more formal and detailed. While reading the paper, I was thinking about what could be some mechanisms at play. Even the discussion in the conclusion left me with questions. Enriching the discussion of mechanisms and adding empirical tests where possible would help the reader understand the theoretical underpinnings of the work.
   1. How does increasing educational spending affect student outcomes? Does it actually increase attendance or improve skills? It could be that a neighborhood is changing demographically. As the authors discuss in the conclusion, higher income families moving into a neighborhood would change the composition of students in a district and increase the spending. I do not believe that the fixed and random effects capture endogenous sorting. Citing papers that study the effects of increasing educational spending would further help here.
   2. Is there a complementarity relationship between education and public safety spending? Perhaps any relationship between educational spending and crime is partially related to complementary increases in public safety spending? This is an empirical question that the authors are well-suited to answer. This kind of exercise would greatly fortify the discussed mechanisms.
6. The authors have an excellent dataset constructed and may want to explore some causal analysis, if they have not already done so for a separate project. Are there some districts or states that exogenously changed educational expenditure? Passage of millages or other local public finance measures may be an avenue to explore. It is understandable if the authors prefer to present an analysis that focuses on correlations, but there may also be opportunity here.
7. Stylistic and Minor Comments
   1. In Figure 1c, why are the proportions less than 0? Do the estimates in Figure 1 account for random effects? Is it that the covariates are "centered" (centered at 0? This is not consistent across figures). It is not clear from the writing.
   2. The description of the results is not clear. Often, it is not clear which figures the authors are referencing (example, line 188 discussing the relationship between population density and crime) or the direction of the estimate (example, line 207 presenting the relationship between Asian population and crime rate).
   3. The instances of informal and unscientific writing detract from the work. Examples include: contractions (didn't, we've, would've), vs, we decided to, beta values, "white-ness,'" ed spending.
   4. Some of the plots are are hard to read, especially Figure 6. I suggest the authors just plot the lines and remove the clouds of individual data points.
   5. The last paragraph is light-hearted, yet imprecise. The conclusion is already long and I wonder if this last paragraph can be cut. It does not seem to add much substance.
   6. The title seems not entirely appropriate for the analysis. The authors cannot make a causal statement on the inequity of educational funding and crime.
   7. Line 329. I do not think it is counter-intuitive. Those who are wealthier and more educated live in richer neighborhoods where there is a higher expected benefit for property crimes.
      
      

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Please see the attached referee report.

Comments for author File: Comments.pdf

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 3 Report

I thank the authors for seriously considering the comments in my report. The description of the related literature and the empirical approach are much clearer. I also found the updated conclusion more compelling. I suggest a few minor changes to be made that are mostly superficial.

• Is Table 1 missing? I did not see it in the updated main paper.
• In considering the complementarity between public safety spending and education spending, did you look within place? In your response, you seem to be describing the overall trend. It might be enlightening to investigate the correlation within place over time.
• Minor formatting changes and spellcheck will be necessary to remove all typos.

Author Response

We thank this reviewer for his/he comments; they have much improved the manuscript and the scope and utility of the work.

We've included Table 1 with sample demographics as well as the latest figures.

So regarding complementarity of total law enforcement (unless this reviewer is talking about adding a new variable of public safety spending) would be capturing the mean within place change over time. Given the number of places, it might be work breaking this down by larger cities, and we'd be happy to collaborate in such a future project with this reviewer. We have also added this point to the discussion.

Thank you for your time and insight!