Literature Review and Theory

Since 1970, the number of people convicted and punished for crimes in the United States has skyrocketed. By the end of 2006, more than 7.2 million people were being supervised in jail, in prison, on probation, or on parole at all levels of government (Reference Glaze and BonczarGlaze & Bonczar 2009; Reference West and SabolWest & Sabol 2008). About 2.3 million of these individuals are incarcerated (Reference West and SabolWest & Sabol 2008). People with criminal backgrounds make up an ever-increasing proportion of the population, particularly among those of low socioeconomic status: “Nine out of ten prison inmates are male, most are under the age of 40, African Americans are seven times more likely than whites to be in prison, and nearly all prisoners lack any education beyond high school” (Western et al. 2004: 1). For high school dropouts, incarceration is fast becoming “a pervasive event” in the life cycle: 32.4 percent of young black male high school dropouts aged 22–30 were in prison or jail; for comparable whites, the figure is 6.7 percent (Western et al. 2004: 7).

Much of the research on the political effects of increasing conviction and supervision rates has focused on explaining the origins and impact of felon disfranchisement laws. Currently, 48 states retain some restriction on the voting rights of felons and/or misdemeanants who are serving sentences; 12 states bar some or all offenders who have finished serving their time from the ballot box, at times for life. Reference Manza and UggenManza and Uggen (2004) estimate five million offenders and ex-offenders were disfranchised during the 2004 general election. Fellner and Mauer estimated that, as of 1998, 13 percent of all adult black men were disfranchised nationwide; in several states, including Florida and Alabama, more than a quarter of black men were disfranchised (Reference Fellner and MauerFellner & Mauer 1998). In many states, a growing percentage of Latinos are disfranchised. As of 2003, 6 percent of the Latino voting age population was disfranchised in Washington and Nebraska and 5 percent of the Latino voting age population was disfranchised in Arizona and Florida (Reference Demeo and OchoaDemeo & Ochoa 2003).

Clearly, felony disfranchisement laws affect a considerable share of the population, especially the black male population. However, despite the growth in disfranchisement rates and conviction rates more generally, very few studies attempt to estimate the participation rates and candidate preferences of ex-felons directly, perhaps because of the methodological and theoretical difficulties of such a task. Most cross-sectional surveys do not ask questions about criminal history. Panel studies that can track incarceration and other incidents often suffer from attrition or selection bias. However, even if survey organizations did ask about experiences with criminal justice, most would find current and former felons incredibly difficult to reach. A final problem is conceptual: determining the impact of convictions and disfranchisement on voting depends on making counterfactual estimates, which requires finding an appropriate control or comparison group against which to compare felons.

The existing research on ex-felon voter participation and the effect of disfranchisement laws is limited by these problems. In an extensive consideration of this question, Reference Manza and UggenManza and Uggen (2004) find that disfranchisement laws prevented hundreds of thousands of felons from voting in 2000 and 2004 and may have changed the outcomes of one presidential and seven senate elections. They estimate that, in the absence of disfranchisement laws, people convicted of felonies would vote at rates of 35 and 24 percent for presidential and mid-term elections, rates that are comparable to those of people with characteristics similar to ex-felons who have not been convicted of crimes (Reference Uggen and ManzaUggen & Manza 2002).Footnote ³ Moreover, they estimate 70 to 80 percent of ex-felons nationally would support Democratic candidates. These results are based on estimates of turnout and vote choice of respondents to the Current Population Survey and American National Election Studies rather than actual offenders. Reference Uggen and ManzaUggen and Manza (2002) assume explicitly that in the absence of disfranchisement laws, “the political behavior of disfranchised felons would approximate that of nonfelons matched to them in terms of age, race, gender, education, income, and marital status.” Subsequent tests of their model on a sample of youth in Minnesota shows no difference in participation between people with and without convictions once “sociodemographic factors” are taken into account (Reference Manza and UggenManza & Uggen 2004: 499). The implication of this argument is that being convicted of a crime affects behavior only to the extent that a person is disfranchised legally.

Other studies conclude that voter registration and turnout among ex-felons is virtually nonexistent. In a cross-state analysis, Reference Grose and YoshinakaGrose and Yoshinaka (2002) find consistent evidence that disfranchisement affects turnout in the South. When extended to the entire nation, however, Reference MilesMiles (2004) finds no correlation between disfranchisement regime and voter turnout across states. He ascribes the lack of a treatment effect to the fact that most offenders do not vote even when they have the right to do so because “the same demographic and socioeconomic factors that correlate with participation in criminal activity, and by implication with disfranchisement, also correlate with the decision to forgo voting” (Reference MilesMiles 2004: 115). By this logic, the relationship between disfranchisement and participation is spurious and disappears when demographic factors such as poverty and race are taken into account. In line with these findings, Burch estimates that about 15 percent of ex-felons in Georgia and Michigan and 10 percent of ex-felons in North Carolina and Missouri would have voted in the 2000 general election (Burch Reference Burch2007a). Similarly, Reference HaselswerdtHaselswerdt (2009) finds single-digit turnout in 2004 among 660 recently released ex-prisoners in Erie County, PA. However, Burch shows that turnout among prisoners, especially those still serving sentences, is low relative to that of probationers (Reference Burch2007b). Thus, Haselswerdt's findings may underestimate ex-felon turnout overall.

The literature on ex-felon voter participation provides a murky picture of the turnout rates of ex-felons who can vote and the effect of disfranchisement laws on ex-felons who cannot. On the one hand, both Miles and Haselswerdt suggest that turnout rates would be closer to zero. On the other, Manza and Uggen estimate a national turnout rate of one-third of ex-felons based on 2004 Current Population Study (CPS) data (Reference Manza and UggenManza & Uggen 2006), By extension, Manza and Uggen might argue that the turnout rates among ex-felons in the most recent election would approximate those of disadvantaged respondents to the 2008 CPS as well. As shown in Figure 1, these rates are extremely high even among respondents who did not obtain high school diplomas. Nearly 46 percent of black men and 53 percent of black women who completed some high school reported voting in 2008, both representing an increase of 6 percentage points over 2004. Thirty-five percent of white men and 41 percent of white females with less than a high school diploma reported voting in 2008; both figures represent a decrease from 2004.

Figure 1. This graph shows reported voter turnout in the 2000, 2004, and 2008 general election among select racial and gender groups who completed some high school without earning a diploma, as reported by the U.S. Census Bureau.

A more nuanced analysis of felon voting patterns would take into account the unique characteristics of people who get convicted of crimes as well as the special burdens criminal convictions further place on their ability to participate in politics. To be sure, persons convicted of crimes often face many disadvantages that would have lowered their turnout relative to the rest of the population even if they had not been convicted. For instance, it is no secret that people convicted of crimes are worse off relative to the rest of the population in ways that are difficult to measure in a survey. For instance, in a national sample of state prisoners, about 70 percent of state inmates and 40 percent of state probationers did not have a high school diploma—in comparison, only 18 percent of the general population lacked high school diplomas (Reference HarlowHarlow 2003). Age is another factor that affects participation; because they tend to be young, people convicted of crimes already would be less likely to participate in politics than the average citizen. The Bureau of Justice Statistics estimates that almost 60 percent of U.S. inmates are under age 35 (Reference West and SabolWest & Sabol 2008). People convicted of crimes also face other problems that would affect their likelihood of participating in politics. Because of their lower socioeconomic status and involvement with crime, it could be the case that mortality rates are higher among offenders as well. Reference Langan and LevinLangan and Levin 2002) find that among prisoners, the mortality rate was 1.46 times higher than that of the general population. Psychological and emotional disturbances are more common among offenders (Reference DittonDitton 1999). Sexual abuse is high among these populations; 16 percent of male and 57 percent of female prisoners report having been victimized sexually prior to their entry into prison (Reference DittonDitton 1999). Drug and alcohol dependence is high among people who are convicted of crimes; one-third of federal and more than half of state prisoners reported committing their crimes while under the influence of alcohol or drugs (Reference MumolaMumola 1999). Twenty-one percent of state and 16 percent of federal prisoners showed signs of past alcohol abuse, while 57 percent of state and 40 percent of federal inmates had used drugs in the month prior to committing their offense (Reference MumolaMumola 1999). These types of physical and mental disabilities often render political activity difficult, even impossible.

In addition to these factors, persons convicted of crimes may differ from the general population in other ways. They may be engaged in drug abuse or other criminal activities that make them unwilling or unable to provide their personal information to government agencies, including boards of elections (Reference MumolaMumola 1999). Likewise, they may be avoiding further contact with law enforcement officials for fear of arrest. Finally, people who commit crimes may be less connected to their communities, families, peers, and government (Reference Fleisher and DeckerFleisher & Decker 2001; Reference SampsonSampson 1988; Reference Sampson and GrovesSampson & Groves 1989; Reference SampsonSampson et al. 2002). These networks reinforce the norm in favor of political participation and the norm against criminal behavior.

It is unclear whether and how the experience of conviction and punishment would affect political behavior. For many individuals, a criminal conviction serves as the wake-up call that provides an opportunity for rehabilitation and training (Sourcebook of Criminal Justice Statistics [“Table 6.73”] 2003; Reference Mumola and KarburgMumola & Karburg 2006). However, the economic and social burdens associated with criminal convictions severely restrict the ability of offenders to garner resources such as time, money, and civic skills that would help them participate in politics after they serve their time (Reference VerbaVerba et al. 1995). The unemployment rate among ex-offenders is much higher than that of the general population, often as the result of employment discrimination (Reference HolzerHolzer et al. 2004; Reference Pager and QuillianPager & Quillian 2005). Federal educational and housing assistance, as well as cash assistance such as TANF and SSI, can be denied to people convicted of drug offenses (Reference Travis, Mauer and Chesney-LindTravis 2002; Reference UggenUggen et al. 2006; Western et al. Reference Western and Pattillo2004a). Socially, it is well documented that people are more likely to get involved in politics when they are mobilized by friends or political parties, and people are more likely to be mobilized when they take part in community and social activities (Reference Rosenstone and HansenRosenstone & Hansen 1993; Reference Verba and NieVerba & Nie 1972; Reference VerbaVerba et al. 1995). Criminal offenders tend to be less residentially stable; moreover, for incarcerated individuals, already-fragile social networks “are made tenuous by the distance between home and prison” (Reference Abu-JamalAbu-Jamal 1995: 12; Reference Fleisher and DeckerFleisher & Decker 2001).

Still, even in light of these findings, the predictions of both Miles and of Haselswerdt seem to underestimate the rate at which felons would vote. Miles's analyses at the state level accurately gauge the absence of effects big enough to change electoral outcomes; however, cross-state analyses may be too blunt to pick up smaller differences in the turnout of felons across states. Likewise, Haselswerdt's analysis is based on a small sample of parolees and thus reflects the turnout patterns of those ex-felons who are least likely to vote. Even though the turnout is low, it is still likely that tens of thousands of persons with felony convictions voted in this and in previous elections.

Even if one accepts the claim that some people with felony convictions vote, it is not altogether clear for whom convicted offenders vote when given the chance. Most criminal offenders are males of low socioeconomic status (Reference HarlowHarlow 2003). Is this group more likely to vote Democratic? Most research today says yes; poor men vote their class interests, so lower-class voters are more likely to support Democratic candidates (Reference BartelsBartels 2005; Reference Brewer and StonecashBrewer & Stonecash 2001; Reference CampbellCampbell et al. 1960; Reference EriksonErikson 1995; Reference Piven and ClowardPiven & Cloward 2000; Reference StonecashStonecash 2000; Reference TuckerTucker et al. 1986). However, as Reference KristofKristof (2004) has pointed out, “One of the Republican Party's major successes over the last few decades has been to persuade many of the working poor to vote for tax breaks for billionaires.” Many other observers often lament the propensity of white lower-class voters to vote seemingly against their own class interests in favor of Republicans (Reference DeNardoDeNardo 1980; Reference FrankFrank 2004; Reference Hillygus and ShieldsHillygus & Shields 2005; Reference Nagel and McNultyNagel & McNulty 1996).

Black men across classes consistently support the Democratic Party and therefore it is expected that black ex-felons would overwhelmingly support Democratic candidates (Reference DawsonDawson 1994). For whites, however, the answer to this question is still unsettled in the literature and remains quite controversial, as noted above. Bartels finds consistent support for the Democratic Party among lower-class whites throughout the last three decades (Reference BartelsBartels 2005). Other empirical evidence suggests lower rates of support for the Democratic Party among white male offenders. Manza and Uggen find that people who have been incarcerated or arrested are more likely to describe themselves as Independents and less likely to describe themselves as Democrats (Reference Manza and UggenManza & Uggen 2006: 124). People with criminal backgrounds were also more likely to support Jesse Ventura rather than major party candidates in Minnesota's 1998 gubernatorial election (Reference Manza and UggenManza & Uggen 2006: 124). Anecdotal evidence such as that found in Reference FrankFrank (2004) also calls into question the idea that lower-class whites vote Democratic.

Research Design

Measuring the participation of people convicted of crimes is difficult, as noted above, because data on the criminal history of individuals have not been included regularly in studies that measure participation. Moreover, determining the effects of disfranchisement is even more problematic because it involves estimating a particularly difficult counterfactual scenario.

To estimate the turnout rates of ex-felons and thus the effect of ex-felon disfranchisement laws, the ideal analysis would measure the extent to which each offender undertakes different political activities now and then compare that figure with what that same person would have done if the conviction or its various consequences had never taken place. Although it may be possible to measure the participation of people once they have been convicted, it is impossible to observe that same individual's participation in the counterfactual condition. Thus, it is not possible to test the effects of conviction on any particular individual directly (Reference HollandHolland 1986). Using a control subject who is similar to what the unobserved person would be if he had not been convicted is the best substitute for approximating the level of participation in the counterfactual condition (Reference HollandHolland 1986). However, people who have been convicted of crimes often are very different from people who have never been convicted in ways that cannot be measured, complicating the task of finding a control or comparison group against which to measure disfranchised ex-felons.Footnote ⁴

Although no surveys or data sources measure the effects of convictions on individuals directly, state governments collect very detailed data on convicted offenders and on one form of political activity, voting. As a result, it is possible to combine these records and use them to directly measure at least one form of participation for offenders with different criminal justice experiences.Footnote ⁵ Because these data also contain some demographic information, one can also separate the effects of an offender's background and disadvantages from the effects of the conviction.

Having data on the personal characteristics and voter participation of offenders makes identifying and measuring their voting behavior possible. Even more importantly, these data enable potentially confounding factors such as age, race, gender, crime severity, and education to be taken into account. These data allow the voter participation of ex-offenders, many of whom are allowed to vote after completing their sentences, to be measured. Constructing appropriate counterfactual groups against which to compare offenders barred from voting is more difficult. For instance, the CPS turnout rates discussed previously provide one estimate of the likely participation among offenders. Manza and Uggen use this comparison group to estimate felon participation in the counterfactual case at the national level (Reference Uggen and ManzaUggen & Manza 2002; Reference Manza and UggenManza & Uggen 2004, Reference Manza and Uggen2006). However, analyses that use this control group to estimate the effect of convictions, like those of Manza and Uggen, are biased because they cannot account for the differences between people with and without convictions that may affect participation.

Instead, looking at turnout rates among persons convicted of their first felony after the 2008 general election best accounts for baseline differences between the general population and offenders. At the time of the election, these offenders had not yet been convicted or taken into custody; thus, their turnout patterns cannot be attributed to the experience or consequences of justice supervision. Rather, the gap between these offenders and the general population, if any, represents the influence of those unmeasured factors that eventually led this group to be convicted of crimes. Conceptually, this group represents people who would have been punished and disfranchised during this election had they been captured and convicted just a few months earlier.

Estimating candidate preferences among offenders is difficult without survey evidence. To get a sense of vote choice, party registration among offenders in states with closed primaries can be used. Although party registration differs from partisan identification and vote choice, it is the best information available on the preferences of offenders (Reference Finkel and ScarrowFinkel & Scarrow 1985).

Hypotheses

To summarize the argument, being criminally convicted can affect an individual's desire and ability to participate in politics. Given the fact that persons convicted of crimes already lack the economic and educational resources of other citizens and suffer from substance abuse, mental health issues, and social isolation at higher rates than nonoffenders of similar social background even before their convictions, these individuals should demonstrate lower levels of political activity than similar citizens who have not been convicted of crimes. Likewise, the experience of being convicted and punished for crimes should prevent participation of offenders because convictions impose barriers that tend to decrease the likelihood of participating in politics even further. To restate:

With respect to vote choice, although there is no way of knowing the true candidate preferences of offenders for this analysis, it is possible to make some plausible assumptions about vote choice based on the available evidence. Criminal offenders do not vote at the same rates as people with similar backgrounds who have not been convicted of crimes. However, it is possible that criminal offenders would have the same preferences as nonoffenders who share their demographic characteristics. There is a consistent finding that in general, nonvoters tend to have the same preferences as voters, such that registration laws have little effect on the policy preferences of the electorate (Reference Rosenstone and WolfingerRosenstone & Wolfinger 1978). Applying this logic to disfranchised offenders, then, one should expect the candidate preferences of offenders to mirror those of other people with the same race, gender, and educational level.

Data

One of the most exciting facets of this study is that it explores the political participation of real offenders using records maintained by the departments of corrections in the states selected for the analysis: Florida, Georgia, Michigan, Missouri, and North Carolina. In each state, the department of corrections maintains highly detailed data on all offenders convicted of felonies and misdemeanors who are sentenced to state supervision in prison or in the community through probation or parole. People convicted of felonies in federal court who served sentences under federal authorities and people convicted of misdemeanors or infractions in municipal court who served time in county jail are not included in the analysis.Footnote ⁶ In all states in this analysis, all offenders accused of state felonies are tried, convicted, and punished by state authorities, such that the files represent a complete list of people who have or are being supervised for felony convictions by that state.Footnote ⁷ People with misdemeanor convictions who were supervised by local authorities are absent from the data, as are people convicted under federal law. As a result, these findings estimate turnout only among citizens convicted by state authorities, and cannot be generalized to those citizens who are convicted under federal or local authorities.

Methods

This analysis counts the raw registration, vote totals, and party registration for two groups of offenders: men who experienced their first conviction and sentence after the 2008 electionFootnote ¹¹ and men who finished serving their sentences before the 2008 election. The probability of voting among ex-offenders and pre-conviction offenders is estimated using multivariate analyses in order to determine whether variation across the two groups in racial composition, age, punishment type, and education (where available) can account for the differences between the pre- and post-conviction groups, if any. Knowing voter registration and turnout among the pre-conviction group gives a sense of what would have happened if these offenders had been convicted and disfranchised during the election; they provide estimates of the counterfactual turnout rates among current and former offenders barred from voting during this election cycle. Voter turnout in the 2008 general election for CPS respondents who report completing some high school without earning a diploma also will be referenced for comparison following the example of Reference Manza and UggenManza and Uggen (2004).Footnote ¹²

The analysis is decomposed into three steps. In the first, the participation rates of offenders before they experience a conviction are tabulated in order to provide a lower bound on what participation would have been without these interventions. This group best serves as a proxy for how people currently serving felony convictions might have voted had they not been convicted. Next, voter turnout among offenders who have completed their sentences is presented. Finally, multivariate estimates of voter turnout are presented for each state, along with the predicted probabilities of voting among pre- and post-conviction offenders with certain characteristics. The purpose of this regression is to account for demographic differences between pre- and post-conviction offenders that might explain differences in the turnout rates of the two groups. However, this regression cannot account for differences in life circumstances between pre- and post-conviction felons. The probability of voting is calculated for male felony offenders who were eligible to vote in the 2008 general election based on models that account for race, age, whether the sentence was served before or after the election, whether the offender had served or would serve time in prison, and where available, educational attainment. The dependent variable, Vote 2008, is a dichotomous variable where “1” indicates a vote in the 2008 presidential election. The analysis includes both ex-offenders and pre-conviction offenders; Pre-conviction is the dummy indicator for offenders who were captured and convicted after the election. Race is entered as dummy variables for White and Black, with Hispanics and other races represented as the baseline condition. Age is in years and includes offenders between the ages of 18 and 65. Prisoner is dichotomous and indicates whether an offender had served or will serve a prison sentence. Finally, offender education is available only for Misouri and Georgia; for these states, Education is a 0 1 variable that indicates whether an offender completed high school or its equivalent. Age, race, and educational attainment have been shown in previous research to be important determinants of voting (Reference CampbellCampbell et al. 1960, Reference Verba and NieVerba & Nie 1972; Reference VerbaVerba et al. 1995). Similarly, voter turnout among prisoners also is lower (Burch Reference Burch2007a). Because voting, the dependent variable, is dichotomous, these models are estimated using logistic regression. The predicted probabilities of voting for offenders with several characteristics are simulated using the means and standard deviations estimated from these models.

Results

The results indicate that many ex-felons voted in the 2008 general election. As Table 1 shows, 22 percent of ex-felons voted in Georgia, 19.4 percent voted in Missouri, and 24.2 percent voted in North Carolina. In Michigan, where felony probationers were never disfranchised, nearly 35 percent of ex-felons voted in the general election. However, in Florida, only 11.1 percent of eligible ex-felons voted in the general election. This low turnout rate is likely due to the fact that this is the first presidential election in which some of Florida's ex-felons could vote. Among offenders who served time for their first offense after the election, turnout surprisingly was lower than that of ex-felons. As Figure 2 shows, in Florida, 9.4 percent of people convicted of crimes after the election voted; in Georgia, 16.7 percent; in Missouri, 11.7 percent; and in North Carolina, 17.3 percent. Again, Michigan is exceptional; 38.9 percent of offenders in Michigan voted before they served time.

Figure 2. This graph shows the percent of persons registered to vote and turning out to vote in each state before they were taken into custody and convicted of crimes after the election. Data were calculated by matching department of corrections data to voter registration and history data from each state.

Table 1. Participation Rates among Male Ex-felons

Note: Data were calculated by matching department of corrections data to voter registration and history data from each state.

When compared with turnout in past presidential elections, it is clear that turnout in the 2008 election represents a substantial increase in political participation among ex-felons from previous elections, as depicted in Figure 3. Since the 2000 general election, when voter turnout was roughly similar in the four states where ex-felons could vote, ex-felon voter turnout has increased 68 percent in Georgia, 95 percent in Michigan, 88 percent in Missouri, and 75 percent in North Carolina. Despite this substantial growth, voting among offenders still remains low relative to that of similar people who do not have criminal backgrounds. As expected, the turnout in the 2008 general election of both ex-felons and those offenders who began serving time after the election is lower than that of CPS respondents who had attended high school without earning a diploma reported earlier in this article.

Figure 3. This graph shows the percent of ex-felons turning out to vote in each state. Data were calculated by matching department of corrections data to voter registration and history data from each state.

Among ex-felons, personal characteristics influence voter turnout much as they do among the broader electorate. As shown in Table 1, older ex-felons were more likely to vote than were their younger counterparts. Also, as expected, ex-felons in all states who had served time in prison at some point before the election were less likely to vote than were ex-probationers who had never served time in prison. Ex-felons with a high school diploma were 33 percent more likely to vote in Georgia and 72 percent more likely to vote in Missouri, the two states where data on the educational attainment of ex-felons are available. The most interesting pattern, however, develops with respect to race. In four of five states, black male ex-felons were more likely to vote than whites, as shown in Figure 4. This pattern results from a surge in black turnout between 2004 and 2008 as Figure 5 displays; in previous years, racial differences in turnout were mixed; in North Carolina and Missouri, white ex-felons were more likely to vote than black ex-felons, while in Georgia and Michigan, the opposite was true.

Figure 4. This graph shows the percent of ex-felons turning out to vote in each state, by race. Data were calculated by matching department of corrections data to voter registration and history data from each state.

Figure 5. This graph shows the change in the turnout rates of ex-felons between 2004 and 2008 in each state. Data were calculated by matching department of corrections data to voter registration and history data from each state.

Because of correlations among several of the factors discussed previously, simple bivariate data might not provide an accurate comparison between pre- and post-conviction felons. For instance, blacks were more likely to vote in this election than whites, probationers more than prisoners, and the higher educated more than the less educated; thus the effect of the timing of the conviction could reflect differences in racial composition, sentence type, or educational attainment across the groups. Similarly, people convicted of their first offense after the election may be younger than people who served their sentences before the election. However, multivariate analyses confirm the patterns found in the simple tabulations; demographic differences in the sample do not account for the difference in turnout between pre- and post-conviction felons. The coefficient on pre-conviction is significant and in the expected direction. Age, race, imprisonment, and education are significant in the models, indicating that each still has a direct effect on turnout even after controlling for the other factors. These estimates can be found in the appendix. Logit coefficients are not easily interpreted so it is easier to discuss the probability of voting among offenders using the simulated scenarios in Table 2 (Reference KingKing et al. 2000: 348). According to the analysis, the probability of voting among black men convicted after the general election declines by 6 percentage points in North Carolina, 4 percentage points in Georgia, and 10 percentage points in Missouri, even after accounting for differences in age, race, and incarceration history. However, there is no significant difference between people who served time before and after the election in Michigan and Florida. Among whites, the probability of voting declines by 6 percentage points in North Carolina, 3 percentage points in Georgia, and 8 percentage points in Missouri. Again, there is no significant difference between pre- and post-conviction whites in Florida and Michigan.

Table 2. Predicted Probability of Voting in 2008 and First Differences for Felony Offenders

Once race is considered, it becomes clear that black ex-felons account for most of the turnout increase between the 2004 and 2008 presidential election in Georgia, Misssouri, and North Carolina. To reiterate the evidence from Figure 4, turnout among black and white male ex-felons increased in all states from 2004 to 2008. However, as shown in Figure 5, in Georgia, Missouri, and North Carolina, black ex-felon participation increased much more than white ex-felon participation in those years. Again, Michigan is exceptional; white ex-felon participation increased and surpassed that of blacks between 2004 and 2008. By comparison, turnout showed no consistent pattern by race prior to 2008. As shown in Figure 6, blacks were more likely to turn out in 2000 in Michigan and Georgia, while whites were more likely to vote in North Carolina and Misssouri.

Figure 6. This graph shows the percent difference in turnout rates for the 2000 general election between black and white ex-felons in each state. Data were calculated by matching department of corrections data to voter registration and history data from each state.

Party registration among ex-felons also shows surprising patterns. Table 1 shows that 51.4 percent of registered ex-felons in Florida registered as Democrats, compared with 27.9 percent as Republicans and 20.7 percent unaffiliated with the two major parties. One should note, however, that party registration is available only for the small subset of Florida ex-felons whose voting rights were automatically restored. In North Carolina, 56.3 percent of registered ex-felons signed up as Democrats, compared with 22.8 percent as Republicans and 20.9 percent unaffiliated or Independent. This pattern, as hypothesized, varies by race and thus reflects the racial composition of the offender population as shown in Figure 7. Black ex-felons overwhelmingly support the Democratic Party; 72 percent of black ex-felons in North Carolina and 84 percent of black ex-felons in Florida are registered Democrats. By contrast, white ex-felons in both states are divided almost evenly among the Democrats, Republicans, and unaffiliated categories, although slightly more white ex-felons support the Republican Party than the Democratic Party. This distribution closely mirrors the general breakdown of registration by party and race in those particular states. In Florida, 83 percent of blacks were registered with the Democratic Party in 2008, while only 35 percent of whites were (Florida Secretary of State 2008). Similarly, in North Carolina, ex-felon party registration by race closely mirrors that of the general population (author's own analysis; see Reference BurchBurch 2010).

Figure 7. This graph depicts the party registration rates of black and white male ex-felons in North Carolina and Florida. Data were calculated by matching department of corrections data to voter registration and history data from each state.

Discussion

These findings help adjudicate the controversy in the literature over offender voter turnout and vote choice. As expected, many people with criminal backgrounds participate in politics both before and after they serve time. Thus, claims such as that of Miles and Haselswerdt that disfranchisement laws only affect people who would not have voted anyway are too harsh. Had Florida, Missouri, North Carolina, Michigan, or Georgia prevented ex-felon voting, thousands of people would not have cast ballots in this past election. However, the data also show that expecting people with criminal backgrounds to vote at the same rates as everyone else after controlling for socioeconomic status, race, and other factors is overly optimistic. Voter turnout among offenders before and after their convictions falls far short of the turnout demonstrated by similarly disadvantaged citizens who have not had a criminal conviction.

More interestingly, these data show that turnout and party registration diverged sharply from previous expectations about the political behavior of offenders. The most important claim here is that race mattered in the 2008 election in ways that it had not previously. As with the broader electorate, turnout among eligible offenders increased since 2004. However, also like the broader electorate, that turnout increase occurred primarily among racial minorities. Black offenders turned out at much higher rates than white offenders even after controlling for age, education, and incarceration history. Prior to 2008's election, turnout showed no consistent pattern with respect to race; in North Carolina and Missouri, white ex-felons were more likely to vote than black ex-felons. This pattern reversed in 2008. Thus, the expectation that ex-felons would respond uniformly to the increased excitement and engagement of the 2008 election was misplaced.

Race also matters with respect to party registration. White ex-felons and black ex-felons diverged sharply in their support of the two major political parties. While, as expected, overwhelming numbers of black ex-felons were registered Democrats, two-thirds of white ex-felons registered as something other than Democratic. These patterns are similar to those found among low-income male voters on national surveys (Burch Reference Burch2007a). Primarily, this finding points out that the potential effects of disfranchisement laws are not limited to Democratic supporters. Because low socioeconomic status (SES) whites were more likely to support Republican candidates in most recent elections, the answer to the question of which party is most affected by disfranchisement laws is, “it depends.” A complete answer to that question would take into account the racial makeup of the disfranchised population and the politics of the time under consideration. For instance, in light of these findings, it makes little sense to argue that Wyoming's disfranchisement of ex-felons particularly hurt the Democratic Party in that state. Likewise, though Democrats held an 11-point advantage in party registration in Florida in 2008, the evidence suggests that Florida's electorate leaned more Republican in earlier years. For instance, the partisan affiliation of Florida's Latinos shifted over the past decade; while a majority of registered Hispanics were Republicans in 2006, by 2008 Hispanic Democrats outnumbered Hispanic Republicans by nearly 70,000 people (Pew Hispanic Center 2008). These findings suggest the need to reexamine the widespread beliefFootnote ¹³ that Vice President Gore would have defeated President Bush in 2000 had ex-felons been allowed to vote in that state.Footnote ¹⁴

One curious result is that people who served time after the election vote less than ex-felons. One might think that the experience of conviction and punishment would suppress political participation by depleting the resources, efficacy, and social ties that foster participation, so individuals who had not yet experienced those effects would tend to vote more than those who had. That still may be the case. The pre-conviction group most closely approximates current felons, as noted at several points throughout this discussion. Pre-conviction felons and current felons may face a different set of life circumstances than ex-felons. For many people who are about to serve time for crimes, the time before they are caught and convicted is particularly difficult: they may be hiding from authorities or engaged in severe substance abuse, for instance, as suggested by Reference MumolaMumola (1999). In contrast, some ex-felons may have been able to turn their lives around and may be in better mental and emotional health than people about to enter the criminal justice system. Forty percent of state prisoners received treatment for substance abuse while in prison (Reference Mumola and KarburgMumola & Karburg 2006). Ten percent received psychotropic drugs and 12 percent received therapy or counseling for mental health conditions while in state correctional facilities Sourcebook of Criminal Statistics 2003).

Another possibility is that the negative consequences offenders experience because of convictions could increase their political activity. Hansen, for instance, argues that groups will increase their participation in politics under threat, such as when important resources are at stake (Reference HansenHansen 1985). Punishment that is perceived to be harsh or unfair, for instance, could hasten the “transformation of convicts into political militants” (Reference Davis and JamesDavis 2003: 69). Moreover, through contact with other offenders, a person may develop new habits of participation or even a revolutionary consciousness. Malcolm X, after learning of the teachings of the Nation of Islam, wrote that “It was right there in prison that I made up my mind to devote the rest of my life to telling the white man about himself—or die” (Reference XX 1965: 186).

A third possible explanation for lower turnout among pre-conviction offenders is that this sample is contaminated with people who were actually ex-felons convicted by other state or federal authorities. Each offender classified as a “pre-conviction” felon is one for whom the first supervision by the state department of corrections for a felony took place after the 2008 election. If the turnout difference were in fact due to the inadvertent inclusion of ex-felons in the pre-conviction sample, then 16 percent of pre-conviction felons in Florida, 19 percent of pre-conviction felons in Georgia, 40 percent of pre-conviction felons in Missouri, and 29 percent of pre-conviction felons in North Carolina would actually be nonvoters with felony convictions in other jurisdictions, assuming a turnout rate among those ex-felons of zero. No data are available to evaluate the possibility that these individuals were convicted by other authorities directly, although the evidence suggests that the likelihood that people have felony convictions under other authorities is small. Sixty percent of pre-conviction ex-felons are between the ages of 18 and 24, suggesting that they are early in their adult criminal life. The percentage of felons convicted in federal courts accounts for only 6 percent of all felony convictions nationally. Furthermore, in the states for which data are available, most people convicted after the election were born in their state of conviction. More than three quarters of Georgia pre-conviction felons, and two-thirds of North Carolina pre-conviction felons were born in their respective states, suggesting that they had not moved to North Carolina after being convicted of a felony somewhere else.

Persons convicted after the election are interesting not just because they provide a close approximation to the turnout rates of felons if they had not been serving sentences at the time of the election, but also because their behavior provides a response to an important counterargument. Some might argue that this article underestimates the intent or desire of eligible offenders to vote. Many researchers can recall stories of offenders who do not vote because they believe they cannot. In fact, eligible offenders may be misinformed by government authorities (Reference EwaldEwald 2005). The explanatory force of such “de facto disfranchisement” claims diminishes, however, in the face of evidence that most felons did not vote before their convictions even after accounting for race, age, and other factors. Intimidation, fraud, or misinformation based on convictions cannot explain nonvoting among people who had not yet been captured, convicted, or punished at the time of the election. Instead, this evidence suggests that most people convicted of felonies are, at best, nonvoters or peripheral voters (Reference CampbellCampbell et al. 1960).

Even with full information, turnout might still be low. When Florida restored the civil rights of hundreds of thousands of ex-offenders, the Department of Corrections, Clemency Board, and Board of Elections teamed up to contact those eligible offenders to tell them that their rights had been restored automatically. These bureaus also held community events and ran advertisements to alert eligible ex-felons of the change in laws. Moreover, since 2006, the Department of Corrections in Florida has advised eligible ex-felons about the restoration of their rights as they exit supervision. Undoubtedly, the state found it difficult to reach ex-felons released before the change in the law (Reference MooreMoore 2008). However, turnout remains low even among those ex-offenders released after 2006, suggesting that lack of information is not the primary cause of nonvoting among ex-felons because this group was told explicitly about the automatic restoration of their rights.

Conclusion

This article argues that, on average, 22 percent of eligible ex-felons voted in states without ex-felon disfranchisement laws. This figure shows that a sizeable number of people would have been kept from voting had these states prevented ex-felon voting. However, this analysis suggests that this very disadvantaged group is even less likely to participate than previously indicated using the CPS or other data sources. Participation rates among eligible offenders, even during a time in which the broader society is highly mobilized, lag far behind those of even disadvantaged people who have not been convicted of crimes. That turnout is so low, even during this favorable climate, suggests that full democratic participation remains a remote prospect for the United States. However, these results raise many questions.

The results indicate the need for more research into the political behavior of people on the margins of society. An increasing number of people suffer from housing instability, food insecurity, joblessness, and isolation; these factors seem to lead to criminality as well as to nonvoting as this research suggests. These people are difficult to reach by traditional survey methods because they often do not have permanent residences or telephones.

Reasonable people disagree as to whether offenders should be encouraged to participate in politics on equal footing with law-abiding citizens. Liberal democratic theory brands an individual who breaks the law “a rebel and a traitor to the homeland” who deserves to be expelled from the community (Rousseau Reference Rousseau and Cress1762/1987). These normative judgments, however, are separate from the empirical understanding of how laws can potentially affect political participation. Achieving a deeper understanding of how being convicted, punished, and disfranchised for committing a crime affects political behavior is essential to comprehending voting patterns among disadvantaged citizens generally because so many disadvantaged citizens experience these phenomena. Likewise, future research should also explore how variation in laws and other conditions at the state level can influence those voting patterns.