Harri Roblin

While at university, my tutor asked me to write a referee report on an unreleased economics paper for feedback. This paper has since been released at this address: https://www.cesifo.org/en/publications/2022/working-paper/no-surprises-please-voting-costs-and-electoral-turnout. Below is the report I produced which I thought would be of interest to my readers.

1. Summary

Understanding the determinants of voter turnout is essential in a functioning democracy to ensure maximum representation by citizens in the process. In this specific example of Munich, the election administrators aim to ‘facilitate voting as much as possible’. This paper can provide useful insights into the local Munich situation with the opportunity for wider generalisation. It studies the effects of voting costs on electoral turnout. The case was focused on eight elections spanning from 2013-2020. More specifically, polling station relocations and consequent voter reallocation were considered, looking at their effects on voting, especially looking at turnout and preferred voting method. An event-study design was used, where the event is the relocation of the polling place. This event was studied during these election periods, with the effect of the relocation being studied regarding the voter turnout and method of voting stated earlier.

The authors found that the changes in polling station location increased the rate of mail-in votes and decreased in-person voting. This paper shows turnouts are broadly maintained by voters changing from in-person to mail. The authors suggest this is due to search costs. This is an important finding and has implications broadly for democracies where mail posting is not offered (only 5% of countries worldwide offer mail-in voting). The drop in turnout was greater in cases where the polling station moved further away but there was no change to turnout if stations were moved closer to voters. This shows the search costs eliminates the increased chance of voting due to less travel. Elderly voters were most negatively affected if they were reallocated, which will be due to being less likely to try and find the new polling station.

This paper is the first report of its kind conducted outside the USA in a multi-party equal representation political setting. Secondly, it uniquely attributes the decreased voting turnout to the cost of searching for the new polling station. It is also the first paper that has shown evidence in favour of habit formation. This is where if a person votes once, they are more likely to continue to vote afterwards. These are all valuable contributions to the field.

I believe this is a good paper which I would recommend for revision and resubmission. Additional analysis might have been possible to compare re-allocated and re-assigned voters versus those who remained at the same polling station. This would be done on an election-by-election basis. However, this is a well-written study with an accurate model and statistics. The paper should also be revised by an experienced economic statistician to ensure the equations are appropriate, but it does have contributions worth publishing.   

2. Comments

• Strengths:

• The event study design provides a strong methodology which is easy to understand. Its strength in identifying causal relationships makes it perfect in understanding the effects of polling relocations. The static nature of the study allows precise estimation of timings of events and the elimination of confounding variables within the timeframe. Partnering this with strong empirical analysis gives this paper the tools to make strong conclusions.
• The paper also talks about having a lack of pretrends. This ensures results are not skewed through election officials changing precinct boundaries based off previous turnouts, as they do not have previous information available to them. This supports the identifying assumption that “trends in outcomes across comparison groups evolve smoothly except through the treatment”.
• The paper has a unique finding that reduction in turnout is not due to increased distance from polling place, but the search cost of finding the new place. This was found through testing the differences in event-time coefficient after absorbing the impact of distance. This discovery can be used to increase voter turnout by providing in-depth instructions for voters to find polling places to try and minimise the search costs. Studies like these might contribute to these voting systems allowing mail-in voting to increase turnout.
• All the statistical results used to make conclusions have a significant level difference (p<0.01). This makes the findings stronger in influencing voter turnouts as these results are significant at conventional levels.
• The balancing tests used also make the conclusions stronger. An example is the bivariate OLS regression results shown in panel A, which support the identifying assumption; ‘treated precincts (which had their polling location moved) would have experienced the same changes in outcomes as other precincts, absent treatment’. As no direct tests could be used to test the assumption, these tests are useful for confirming the validity of the results.
• The paper constructed a panel dataset of 618 precincts, complemented by time-varying indicators which were provided by the Munich Statistical Office. Therefore, robust and heteroskedasticity-consistent standard errors are required. The paper uses clustered standard errors at precinct level ‘to account for the correlation of model errors over time’. This standard error technique factors in individuals, within their groups, being independent. Thus, allowing for more robust inference. This is especially true in this study due to a plethora of clusters. Clustering the standard errors also leads to wider confidence intervals, which means findings are much less likely to be found vaguely significant.
• As mentioned in the summary, the paper is the first to be conducted outside the USA. The 7-year duration of the study allows the data to exhibit these transitory trends. This uniquely long research period is required for this paper as elections are not often, meaning longer is required to make findings.

• Weaknesses:
  
• One of the main weaknesses of the paper is that the results are demonstrating ideas that have been widely shown in previous papers. For example, the idea of increased voting cost causing shift to increased mail-in voting is a concept already proven.
• Despite the wide timeline of data the study uses, all the information is based on one location. This means it does not have universal applications as other cities might have factors that change the results (i.e. better transport links to reach a relocated polling station easier, different voting systems).
• The nature of the paper only recording data from polling periods opens the results to external shocks that occur outside of election time. These include possible changes in legislation, such as new incentives to vote or turning voting mandatory. This causes false correlations to be identified.
• Inference not being based on quasi-random sources of variation means the heterogeneity analysis can only be taken as suggestive. This is due to the possibility that other variables not included in the study produce differential causes, making it hard to isolate the variable of interest. This can also lead to selection bias, with different treatment effects getting misconstrued due to a confounding variable affecting the outcome.
• There are a few assumptions the paper takes that are unrealistic. One is that ‘polling place reassignments themselves are not driven by the expectation of changes in turnout’. In real life, this does not hold true. Officials use polling place reassignments to maximise voter turnout, which is stated in the conclusion of their paper.
• Another problem this paper has is the difference in views on voting in different districts. If projected votes show a one-sided contest, citizens have less incentive to vote at the reassigned polling locations. This is a similar problem to the external shocks stated earlier, as it will cause this reassignment to have a larger effect than it truly has.
• A variable not considered to the correct degree is precinct size. Reassignment of the polling place in a larger precinct is much more likely to increase the search cost, which is the main variable affecting voter turnout. In a smaller precinct, there is less possibilities for places it could reside, meaning less distance to travel and less unknown places for locals.
• Terms within the paper could have been defined better. ‘Search costs’, ‘bivariate OLS regression’ and ‘quasi-random’ are all terms known within the economics world. However, if this paper were to have wider applications, these terms should be described so a wider audience can understand the conclusions.
• Not all the elections within the study were the same type. This means this comparison should not be made, as it is natural to have different levels of turnout for the different elections as they hold varying importance.
• Better comparisons are capable on a vertical level as described in the summary. For example, a more comprehensive contrast could be made between the turnout of groups that have been reallocated and those that have not. This could act as a control to ensure that the reallocation is truly the reason for changes in turnout.
• This paper struggles to isolate and eliminate the effect of habit formation. If a voter does not temporarily vote due to reassignment, habit forming would increase the utility of not voting again. This would alter the results and is impossible to account for.