Every Vote Counts, But What About Theirs?
Jails, Voter ID Laws, and Voter Turnout
Yifei Luo
May 2021
“Count every vote!” protesters shouted after the 2020 presidential election. But what about those deprived of voting rights in the first place?
A final project blog post for ECON 0210.
Introduction
Soon after Georgian voters granted Democrats 51-50 control of the Senate with not one, but two runoff elections, (Cohn and Smart 2021) Georgia’s Republican legislature moved to pass a new voting law. Among other provisions, absentee voting now requires some form of identification. (Brewster and Huey-Burns 2021)
Two obstacles are widely believed to suppress turnout: voter ID laws to stop the poor and ID-less from voting, and a world-leading mass incarceration rate, with 1 in 239 U.S. residents imprisoned in 2019. (Carson 2020) Although underrepresentation itself is enough to undermine democracy, its effect can only worsen should this underrepresentation be biased on political preference, race, or indeed any other variable. We therefore arrive at this important question:
Do voter ID laws and mass incarceration suppress turnout, and if so, to what extent?
Data for Analysis
As jail and prison figures are often delayed, this analysis looks at data for the 2012 U.S. presidential election, compiled by the Vera Institute. (Subramanian, Henrichson, and Kang-Brown 2015) We can add jail and prison rates together, to form an incarceration indicator of how many people are incarcerated per 100,000. Voter ID laws, on the other hand, are categorical: either there is no ID requirement, there is one, or there is a strict one. (State Legislatures 2020) We estimate turnout as a percentage of the voting-age population to include disenfranchised felons; income inequality and partisanship are two potential variables used for control.
Interactive Map
This interactive map gives a full picture of how turnout varies across the U.S. As noted in the statistical appendix below, the map does not include data from Alaska, which reports turnout differently.
Darker shades indicate higher turnout.
Scroll to zoom; hover for independent variables in each county.
Artificial Divides
Voter ID laws vary by state, so we can observe divides in similar places.
Take Arizona and New Mexico for example: back in 2012, Arizona had a strict voter ID requirement, while New Mexico did not require voter IDs at all. The result? Despite being neighbors, Arizona only had an average turnout of 46.49%, 7.78% smaller than New Mexico.
The same applies to mass incarceration.
Collin County and Denton County share numerous similarities: they are both in Texas, near a large city (Dallas), returned similar votes for Democrats (33%), and top 5% income for the two counties are both near 30%. Going from Collin County to Denton County, however, as incarceration climbs from 459.21 to 629.88, voter turnout decreases from 60.68% to 55.35%, a decrease of more than 5%.
Analysis
Voter ID laws
Voter ID laws are prevalent in the US, even for in-person voting - only 16 states do not require them, while every state requires IDs for at least some of its mail-in ballots. (State Legislatures 2020) But America’s lack of a universal ID had sparked a fierce debate, between the need to verify voter identity and the need for everyone eligible to vote to actually be able to.
A common case for voter ID laws is prevention of voter fraud. But voter ID laws neither boost voter confidence in elections or lower rates of voter impersonation fraud, a study published in the Stanford Law review found; rather, voter ID laws have become “more salient and partisan in tone.” (III, Ansolabehere, and Persily 2016) Another study, published February 2021, refuted each and every voter fraud claim “statistics” deployed by former president Trump’s supporters. (Eggers, Garro, and Grimmer 2021)
Hover for turnout averages and 95% confidence intervals.
What voter ID legislation can do, however, is to limit access to the polls and suppress turnout. As shown in the above plot, preliminary analysis shows a clear downward trend in turnout, suggesting that the presence of voter ID laws limits turnout, and stricter voter ID laws reduces it further.
There is one potential confounder, namely whether the county leans Democrat or not. An analysis of restrictive voter ID legislation from 2001 to 2012 has found that Republicans have used these voter ID laws to “maintain support while curtailing Democratic electoral gains,” especially in states where elections are more competitive. (Hicks et al. 2015) Republicans are also “markedly more united in their support of voter ID laws,” another analysis found, whereas Democrat supporters’ opinion on such laws is more divided. (Gronke et al. 2019) Nevertheless, similar to how Georgia’s Republican legislature went to action after a Democrat victory, county-level support does not always align with state-level legislature support.
The 2021 Georgia runoffs also demonstrated how Democrat-leaning increases voter participation: turnout was 92% in precincts that voted for Biden, and 88% in precincts that voted for Trump, yielding a 4% difference. (Cohn and Smart 2021) Further evidence comes from a summary by the University of Notre Dame, indicating that the Democratic candidates have benefited more from higher turnout, indicating the effect of partisanship on turnout. (Radcliff 1994)
Mass Incarceration
You most likely have heard of the word “incarceration” before. As Pettit and Gutierrez note, the U.S. has continued to “incarcerate a historically and comparatively large segment of the population,” despite crime rates declining for almost two decades. Prisons may well hold convicted felons, but jails are for those awaiting trial, or poorer people who cannot afford bail for a misdemeanor. (Pettit and Gutierrez 2018)
People in jails can still vote: as the BBC reports, there are as much as 750,000 of them in 2020. (Seales 2020) Nevertheless, without access to the Internet, or even the local Board of Elections, few actually do vote. Apart from turnout, mass incarceration also destroys trust in the legal system and the health of families; the burden is borne collectively and racially, “most notably by African Americans,” the research stresses. (Pettit and Gutierrez 2018)
Click and drag to zoom in, double-click to zoom back out.
The right-hand side of this scatter plot shows 5 outlier counties throwing 1 in every 10 people in jails and prisons; removing them from the linear regression yields a more downward-pulling impact of mass incarceration on turnout.
Even more unfortunate than voter suppression is racially biased voter suppression. One study found that receiving a short jail sentence for a misdemeanor decreased chances of voting in 2012 by 8.0% overall, but 14.6% for Black voters; (White 2019) the estimated effect of incarceration can range from -11% to -26%, another meta-study concludes. (Gerber et al. 2017)
Notably, one of the mechanisms of mass incarceration affecting turnout is through jailed voters. They are unable to show up at polling stations, and are therefore subject to stricter absentee voter ID laws. As such, incarceration itself is a moderating variable for voter ID laws; so when investigating the latter, there is no need to control for incarceration.
Linear regression
Using a linear regression, we can summarize how voter ID laws, incarceration and control variables affect turnout, thus addressing bias due to any confounders.
| Dependent variable: | |||||
| turnout | |||||
| (1) | (2) | (3) | (4) | (5) | |
| Voter ID laws | -0.018*** | -0.015*** | -0.015*** | ||
| (0.002) | (0.002) | (0.002) | |||
| High Incarc. | -0.049*** | -0.048*** | |||
| (0.004) | (0.004) | ||||
| Dem. Voting | 0.026*** | 0.027*** | 0.034*** | 0.034*** | |
| (0.004) | (0.004) | (0.004) | (0.004) | ||
| High Inc. Ineq. | -0.007* | -0.003 | |||
| (0.003) | (0.004) | ||||
| Constant | 0.592*** | 0.576*** | 0.579*** | 0.578*** | 0.579*** |
| (0.003) | (0.003) | (0.004) | (0.004) | (0.004) | |
| Observations | 3,104 | 3,104 | 3,087 | 2,520 | 2,509 |
| R2 | 0.020 | 0.037 | 0.038 | 0.113 | 0.113 |
| Adjusted R2 | 0.020 | 0.036 | 0.037 | 0.113 | 0.112 |
| Residual Std. Error | 0.096 (df = 3102) | 0.096 (df = 3101) | 0.095 (df = 3083) | 0.090 (df = 2517) | 0.090 (df = 2505) |
| F Statistic | 63.009*** (df = 1; 3102) | 58.832*** (df = 2; 3101) | 40.876*** (df = 3; 3083) | 160.785*** (df = 2; 2517) | 106.863*** (df = 3; 2505) |
| Note: | p<0.1; p<0.05; p<0.01 | ||||
In the initial linear regression, stricter voter ID laws (No to Yes, Yes to Strict) decrease turnout by an average of 1.8%; after factoring in Democratic support, the number drops to 1.5%. Support for Democrats is an omitted variable, increasing turnout by 2.7%. The results indicate that Democratic support only confounds voter ID laws’ influence by 0.3%, although providing further empirical evidence on Democratic support boosting turnout.
To test for whether prevalent issues will affect turnout, high income inequality is added as a control variable; the table shows no decrease in either variable’s effect on turnout due to income inequality, so it is not a confounder.
For two otherwise identical counties, stricter voter ID laws decrease turnout by 1.5%, and high incarceration decreases turnout by 4.8% after controlling for partisanship and income inequality. A Democrat-voting county will have 2.7% to 3.4% more turnout than otherwise; all three figures are significant at the 0.01 level.
Significance
How significant are these turnout differences? We look at percent differences between counties to find out. Suppose that those prevented from voting only voted for one party. Then, a hypothetical change in these barriers may alter the county’s votes enough to change the winning party.
In fact, eliminating voter ID laws nationwide will flip 47 counties, while lowering incarceration will flip 80 counties. This amounts to a total of 11,135,295 voters in 102 counties, or 3.55% of the whole U.S. population, being potentially affected! Of course, this is an unrealistic overexaggeration; however, these voters could indeed benefit from removing voter ID laws and high incarceration as barriers to voting.
Conclusion
Access to voting is central to any government’s claim that it was chosen by the people. Despite this, the analysis above indicates that voter ID laws and high incarceration reduced overall turnout by 1.5% and 4.8% on average respectively.
Voter ID laws and effects have been examined closely in previous literature, with mixed results. The observed effect was negative and racial, (Hajnal, Lajevardi, and Nielson 2017) negative but not racial, (Hood III 2012) or there was no effect at all. (Mycoff, Wagner, and Wilson 2009) A team of researchers in 2017 identified problems limiting the effectiveness of such research: voting surveys themselves are hard to reach for the hard-to-reach populations (such as those without a photo ID), and discrepancies in the data prevented the recovery of plausible estimates. (Grimmer et al. 2018)
A few assumptions exist for these causal claims, most notably that partisan support in a county is independent from high incarceration in that county. This is difficult to verify or disprove: local sheriffs typically have larger influence over policing practices, and the fear-mongering that led to historical jail & prison populations has been bipartisan since its birth. (Haney 2010) The issue also lies in the broad scope and variance in the political parties themselves, their policies varying across each region, state, or precinct.
The data collected were for the whole population, either from departmental reports or the Census Bureau’s estimates. While offering a whole picture of voting in the U.S., the nature of this data dictates that we cannot draw observations for the affected population only, for example how felons vote less after being convicted.
Also absent from the overall analysis is racial data, since there is no turnout data for each race and ethnicity; existing analyses have relied on surveys as an estimate of the whole voting population. Further mechanisms exist to racially exclude minorities from voting. Racial and ethnic minorities are less likely to possess a valid ID to vote; (Barreto et al. 2019) they are also asked for IDs at higher rates than White voters. (Cobb, James, and Quinn 2010)
As for mass incarceration, the story has been about race from before its birth, when Southern states codified a range of voting barriers to counter Black suffrage, including criminal disenfranchisement. (Shapiro 1993) Literacy tests and property requirements were gradually eliminated, but disenfranchisement through incarceration remains; in as late as 2000, 15% of all NYPD arrests were for marijuana, and most arrestees were Black or Hispanic. (Golub, Johnson, and Dunlap 2007) In connection with recent policing events, this analysis adds yet another piece of evidence on the impact of current policing and sentencing practices.
Statistical Appendix
A complete replication package with additional details and source files can be found in this shared folder.
Data sources
Mass Incarceration
The Vera Institute of Justice, a nonprofit research organization, compiled county-level data for prisons and jails 1970-2018. The data came from a variety of sources, including the National Corrections Reporting Program (NCRP), the Annual Survey of Jails (ASJ), and the FBI Uniform Crime Reporting Program, among others; Vera researchers then spent a great amount of time verifying and manually correcting the data, such as contacting local jurisdictions themselves.
The unit of observation for the cleaned dataset is one county, containing jail and prison populations in that county during 2012.
Voter ID laws
The National Conference of State Legislatures (NCSL) keeps track of voter ID laws in effect at each time period, so we can compile a list of voter ID laws by state. (State Legislatures 2020)
The unit of observation is one state, with numbers 2, 1, and 0 indicating the strictness of voter ID laws in that state, in 2012.
Turnout
Traditional turnout figures report the ratio between votes cast and the number registered voters. For our purposes, this is problematic because convicted felons, and sometimes those without a valid ID to register too, are left out of these calculations.
Instead, we turn to the Census Bureau’s Citizen Voting Age Population (CVAP) estimates. These figures are tabulated from the American Community Survey, drawing from 5 years of ACS data to estimate the number of voting age citizens in each county. (Bureau 2021)
Election returns were gathered from The Guardian, townhall.com, Fox News, Politico, and the New York Times, compiled by these contributors on GitHub.com. (McGovern 2020)
The unit of observation for the cleaned dataset is one county, containing voting-age populations and election returns in that county, back in 2012.
Controls
Income inequality
The Economic Policy Institute published a list of income percentiles 2010-2015 in their report “The new gilded age: Income inequality in the U.S. by state, metropolitan area, and county.” (Sommeiller and Price 2018) Their data source was tax data from the Internal Revenue Service.
The unit of observation for the cleaned dataset is one county, containing average top 5% income percentiles for that county in 2010 and 2011.
Partisan voting
To determine whether voting for a different party affects turnout, we can use the election returns dataset which includes voting percentages for both parties.
Data cleaning
The data were on county or state levels, so one-to-one and one-to-many merges sufficed.
As noted above, Alaska does not report election returns by county, so this analysis does not include data for Alaska.
Turnout is generated by dividing the number of votes by the citizen voting age population (CVAP).
An index for incarceration is generated by adding prison and jail rates for every county. The rates are the number of people in jail/prison for every 100,000 people aged 15 to 64 in a county. (Subramanian, Henrichson, and Kang-Brown 2015)
Summary statistics
| Variable | N | Mean | Std. Dev. | Min | Pctl. 25 | Pctl. 75 | Max |
|---|---|---|---|---|---|---|---|
| Turnout | 3104 | 0.576 | 0.097 | 0.174 | 0.509 | 0.643 | 1 |
| % Democrat | 3105 | 0.385 | 0.148 | 0.046 | 0.278 | 0.476 | 0.934 |
| Incarceration | 2520 | 1212.323 | 1008.748 | 45.45 | 682.392 | 1487.823 | 26300.81 |
| Top 5% Income % | 3088 | 28.333 | 5.486 | 16.25 | 24.848 | 30.717 | 73.907 |
Robustness checks
Alaska
Alaska does not report election returns by county, so we cannot produce turnout data for Alaska. More people in Alaska vote by mail, and voter ID laws are stricter for mail-in ballots, so not accounting for Alaska may reduce the observed effect of voter ID laws on turnout.
Turnout
Estimated turnout exceeded 100% in 5 counties; in one of these counties, Harding County NM, there are reports of turnout hitting 100% in 2012. (Fouriezos 2019) Although this explains estimated turnout exceeding 100%, we still discard these estimates, as they are unreliably unreasonable.
Incarceration index
The incarceration index exceeded 100,000 in King County, TX and Daggett County, UT.
Both counties have a CVAP of less than 1,000. Small counties like these may not have prisons of their own, so prisoners are sent to nearby counties, creating errors during approximation. As with the turnout data above, we discard these unreliably unreasonable estimates. Additional outliers are treated in a separate regression, as seen above.
Partisan policies
Both the Democratic Party and the Republican Party have a wide range of policies regarding voter ID laws, income inequality, incarceration and turnout, and these policies often differ from one state to another. This analysis uses voting figures to estimate the effect of partisan policy, thereby assuming that a Democratic-voting county will have Democrat-leaning policies, and mitigates this effect using linear regression.
Heteroskedasticity
In a common linear regression, heteroskedasticity is unaccounted for and will produce invalid standard errors.
We can test whether heteroskedasticity exists using a White test, the null hypothesis being that heteroskedasticity is not present in the data.
| # | statistic | p.value | parameter | method | alternative |
|---|---|---|---|---|---|
| 1 | 6.021054 | 4.926570e-02 | 2 | White's Test | greater |
| 2 | 10.312828 | 3.547527e-02 | 4 | White's Test | greater |
| 3 | 26.093041 | 2.139297e-04 | 6 | White's Test | greater |
| 4 | 27.905920 | 1.303268e-05 | 4 | White's Test | greater |
| 5 | 53.508566 | 9.267359e-10 | 6 | White's Test | greater |
With p-values smaller than 0.05, we reject the null hypothesis and conclude heteroskedasticity is present in the data. We therefore need to replace standard errors with heteroscedasticity-consistent ones using the starprep function from the estimatr package. We have done this above when using stargazer (Hlavac 2018) to display the linear regression table, ensuring robustness under heteroskedasticity.
Additional plots
T-tests
Issue-based votes
Before looking at turnout: do counties with high incarceration and income inequality vote differently?
| Test statistic | df | P value | Alternative hypothesis |
|---|---|---|---|
| 8.334 | 2502 | 1.272e-16 * * * | two.sided |
| mean in group FALSE | mean in group TRUE |
|---|---|
| 0.4164 | 0.3697 |
| Test statistic | df | P value | Alternative hypothesis |
|---|---|---|---|
| -4.511 | 3055 | 6.692e-06 * * * | two.sided |
| mean in group FALSE | mean in group TRUE |
|---|---|
| 0.3734 | 0.3973 |
With p-values far smaller than 0.05, we can conclude that counties with higher incarceration returned on average 4.7% less votes for Democrats. High income inequality, however, increased Democratic votes by 2.4%.
Voter ID laws
Do voter ID laws affect turnout?
| Test statistic | df | P value | Alternative hypothesis |
|---|---|---|---|
| 7.552 | 2091 | 6.372e-14 * * * | two.sided |
| mean in group FALSE | mean in group TRUE |
|---|---|
| 0.5942 | 0.5662 |
| Test statistic | df | P value | Alternative hypothesis |
|---|---|---|---|
| 5.617 | 1182 | 2.421e-08 * * * | two.sided |
| mean in group FALSE | mean in group TRUE |
|---|---|
| 0.5809 | 0.5584 |
The difference was statistically significant, although only a difference in turnout less than 3% was observed.
High incarceration
Do high incarceration rates affect turnout?
| Test statistic | df | P value | Alternative hypothesis |
|---|---|---|---|
| 15.16 | 2488 | 9.919e-50 * * * | two.sided |
| mean in group FALSE | mean in group TRUE |
|---|---|
| 0.5998 | 0.5443 |
High incarceration in general meant 5.6% less Democratic votes; a scatter plot is needed to identify and address any possible outliers here.
Partisanship
Does being Democrat-voting affect turnout?
| Test statistic | df | P value | Alternative hypothesis |
|---|---|---|---|
| -8.794 | 3086 | 2.343e-18 * * * | two.sided |
| mean in group FALSE | mean in group TRUE |
|---|---|
| 0.5607 | 0.5911 |
Democrat-voting counties do have 3.4% higher turnout on average than other counties.
Income inequality
And, does income inequality affect turnout? As a potential control variable, people who care strongly about a social issue may be less affected by voting restrictions.
| Test statistic | df | P value | Alternative hypothesis | mean in group FALSE |
|---|---|---|---|---|
| 1.112 | 3048 | 0.2664 | two.sided | 0.5779 |
| mean in group TRUE |
|---|
| 0.574 |
The p-value of 0.3312 shows no significant difference in turnout based on income inequality.
Bar graphs
Voter ID laws
Do voter ID laws affect turnout? 
We see that stricter voter ID laws are associated with lower voter turnout.
Incarceration
Do high incarceration rates affect turnout? 
The error bars do not overlap, so there is a group difference in turnout between counties with and without mass incarceration.
Partisanship
Does voting for one party affect turnout? 
Yes! Counties supporting different parties have different turnouts. If partisan support affects mass incarceration, then we may have identified a potential confounder.
Inequality
Does high income inequality affect turnout? 
Overlapping error bars indicate that we cannot conclude a difference in turnout between counties with and without high income inequality.
Scatter plots
Voter ID laws
Do voter ID laws affect turnout? 
From this box plot, we can observe a weak negative correlation between voter ID laws and turnout.
Incarceration
Do high incarceration rates affect turnout? 
After omitting 5 outliers, there is a stronger downward trend of incarceration relating to decreased turnout. The outlier portion does not show a clear trend.
Partisanship
Does voting for one party affect turnout? 
There are few influencing outliers in the cloud. We see a weak positive correlation between Democrat vote share and turnout.
Inequality
Does high income inequality affect turnout? 
There does not seem to be a correlation between income inequality in either the regression part or in the outlier.
Causal diagram
Causal diagram for turnout
A less Democratic-leaning county will have stricter voter ID laws, and less turnout due to Democrat efforts in promoting turnout.
Strict voter ID laws directly lead to lower turnout, and may also limit turnout through racial bias in ID ownership. This racial bias is an unobserved variable.
With stricter voting ID laws, prisoners have reduced access to voting, leading to lower turnout. This effect is moderated by high incarceration: the higher the incarceration, the more prisoners have limited access to voting and thus there is lower turnout.
Even without stricter voter ID laws, prisoners still have reduced access to voting, so higher incarceration itself also leads to lower turnout.
Evidence from the data and from previous research have been discussed in the Analysis and Conclusion sections.