THE RELATIONSHIP BETWEEN LENGTH OF RESIDENCE AND VOTING BEHAVIOR IN THE UNITED STATES

Cole Gessner

Abstract: Prior research has identified a number of factors that are related to voter turnout, including socio-economic status, race, age, and sex. A substantial literature also suggests the existence of a positive relationship between length of residence and likelihood of voting.  However, these studies use data from before 2016. The present study fills that gap in the literature by estimating the relationship between length of residence and the probability of voting using the most recent data from the United States Census Bureau Current Population Survey Voting Supplement. More specifically, I estimate the influence of length of residence on the probability of voting in the 2020 U.S. general election. Consistent with earlier studies, I find evidence of a small, positive, and statistically significant association between length of residence in the same home and the likelihood of voting in the November 2020 elections. Given that political participation is widely regarded as an indicator of a healthy and fully functioning democracy, policymakers interested in preserving American democracy should explore the potential for a causal relationship between length of residence and voting.

Introduction

More than two-thirds of Americans recently surveyed by American National Election Studies (ANES 2021) believe that public officials in the United States do not care what their constituents think. Almost as many respondents believe that American citizens do not have a say in public policy decisions or in what their government does (ANES 2021). Perhaps because of these concerns, the United States is an outlier among its democratic peers in terms of voter turnout (Hill 2006). Some 67% of eligible Americans voted in the 2020 presidential election, and only half of eligible voters participated in the 2018 midterm elections (United States Election Project 2020). While these turnout figures represented historic highs for the United States (Desilver 2021), the country still ranks relatively low in voter turnout when compared to other democratic countries (Desilver 2020). A total of 56% of the United States voting age population (VAP) participated in the 2016 presidential election (Desilver 2021). In comparison, during recent national elections, 80% of VAP participated in Australia, 77% participated in South Korea, 77% participated in The Netherlands, and 69% in Germany. More than 80% of VAP participated in Australia’s 2019 election (Desilver 2020).[1]

Low voter turnout has potentially important implications. For example, multiple studies have found that American voters tend to be of higher socioeconomic status than non-voters (Leighley and Nagler 1992; Lijphart1997; Nielson 2015). This finding suggests that, in the American political system, the poor and other groups could be underrepresented (Hill 2006). Low turnout also has implications for electoral outcomes. Wattenberg and Brians (1998) estimated that, in the 1994 midterms, the Republican Party may have lost nearly three percent of its vote because registered Republican voters decided to not participate.

In addition, low turnout levels are related to a variety of policy outcomes and to the overall health of a democracy. In his study of the impacts of compulsory voting on public policy outcomes in Australia, Fowler (2013) found that the increase in voting after the adoption of the country’s compulsory voting laws was linked to increased public spending on pensions. Relatedly, Anzia (2012) found that changes in voter turnout in Texas school board elections were linked to changes in teacher salaries.[2] Other studies have shown that, in the United States, voting is a clear indicator of democratic collective action (Chapman 2016), and that high voter turnout supports democratic values such as popular sovereignty, legitimate elections, and majority rule (Hill 2006).[3]

Given the implications of voter turnout for the healthy functioning of government, it is important to understand its determinants. One potential determinant is voters’ sense of connectedness to their communities, which may be related to how long they have resided in their homes. A considerable literature on this subject provides evidence of a positive relationship between length of residence and likelihood of voting. However, the data used in this research dates from before 2016.  However, these studies are limited to data collected before 2016, leaving a gap in the research from 2016 to the present. In this study, I use the most recently available (2020) release of the United States Census Bureau’s Current Population Survey (CPS) to study this relationship.

Background

Prior research has come to differing conclusions regarding the significance and direction of the relationship between length of residence and voting. Since the 1980s, interstate migration within the United States has been on the decline (Molloy et al. 2014). This trend is particularly true for those who rent their homes. One study finds that 21.7% of renters in the United States moved to a different state in 2017, a lower rate than in any other year since 1988 (U.S. Census Bureau 2017). This decreasing trend has been found to be associated with lower rates of job turnover and decreases in the wage gains associated with such job changes, among labor market outcomes (Molloy et. al.  2014). 

However, a number of recent studies have shown that while the population as a whole is less likely to migrate, younger, college-educated Americans are more likely to move in order to find new economic opportunities (Kodrzycki 2001; Kelly 2010; Sapra 2014). These demographic correlates on interstate migration may also have implications for voter turnout, as studies have shown that those with more education and higher incomes are more likely to vote (Pew Research Center 2006; Nielson 2015). In other words, those in the United States who are most likely to vote may also be the most likely to migrate. 

The United States is also experiencing a decline in homeownership (GAO 2020) and a rise in rent prices (U.S. Census Bureau 2021), both of which may be related to migration patterns. A 2020 report from the Government Accountability Office (GAO) found that, between 2004 and 2016, the rate of U.S. homeownership decreased from approximately 69% to 62% and that the ratio of rented to vacant household units had increased compared to the ratio of rented to owned housing units. Additionally, the GAO report found that, between 2010 and 2018, homeownership in nine major cities had either declined or remained relatively constant.[4] This factor, combined with the Census Bureau’s (2021) finding that rent levels have increased steadily since 1997, could have important implications for intra-state migration. Given that residential leases tend to be set for a particular length of time and that renters in the United States are typically more mobile than homeowners (Joint Center for Housing Studies of Harvard University 2020), Americans who rent may be incentivized to relocate from neighborhood to neighborhood within their communities in search of more affordable rent levels. Such highly mobile Americans who move from lease to lease may be less likely to register to vote in their new places of residence.

In addition, the Joint Center for Housing Studies (2020) reported that an increasing number of renters in the United States are younger, college-educated, and white. This demographic group is also highly likely to vote (Wattenberg 1998; Pew Research Center 2006), which suggests that a demographic group important to voter turnout is also more likely to rent rather than own and, therefore, to have higher residential mobility levels than other Americans. If the demographic groups that are more likely to vote are also more likely to relocate, registration and voting rates could be affected.

Literature Review

A substantial literature suggests that there is a positive relationship between the length of time that an individual resides in his or her home and that individual’s likelihood of voting (Wolfinger and Rosenstone 1980; Squire et al. 1987; Rosenstone and Hansen 1993; Highton 2000). Most importantly, Squire et al. (1987) found that living in a community for two or fewer years is associated with a significantly lower probability of voting. Relatedly, Ansolabehere and Lovett (2008) found that moving across state lines has a significant, negative relationship with a mover’s opinions of the destination state’s senators, which the authors consider to be a proxy for political engagement. In other research, Burke et al. (2019), in their analysis of data on North Carolinians who moved to a new voting precinct between the 2012 and 2016 elections, found that a change in polling station was associated with a decrease of nearly one percentage point in the probability of voting. 

Researchers have advanced two theories to explain the relationship between length of residence and voting behavior. The “social cost” school of thought posits that moving disrupts one’s social network, and that these disruptions in turn have implications for an individual’s political participation. The “administrative cost” school suggests that the administrative burdens associated with moving, such as finding a new polling place and re-registering to vote, reduce an individual’s likelihood of voting. I describe both theories in more detail below.  

The Social Cost School

Several studies examine the relationship between social networks and political participation. For example, Fieldhouse and Cutts (2018) find that sharing those who identify with the same political party as others in their social networks are more motivated to vote. Field experiments have also provided evidence of a positive relationship between social pressure and the motivation to vote (Gerber, Green, and Larimer 2008; Panagopoulos et al. 2014).  Magre et al. (2016) found that the longer individuals live at their residences, the more likely they are to participate in local community activities. More specifically, the authors found that community engagement increases significantly after five years of residence in one’s community.

Researchers who study the impact of residential mobility on political behavior often argue that moving disrupts social networks that serve as important sources of political information and motivation to engage in politics (Wolfinger and Rosenstone 1980; Rosenstone and Hansen 1993). For instance, in his analysis of panel data from Danish elections between 2009 and 2013, Hansen (2016) finds evidence of a negative relationship between moving and political participation, even among groups of voters who maintain the same polling station after a move—and who therefore do not need to re-register. Hansen interprets this as evidence that the social value of voting decreases when an individual moves to a new residence. This research directly supports the social cost school’s argument that disruptions to social networks, such as those caused by a move from one community to another, could reduce one's likelihood of voting.

The Administrative Cost School

Other studies of the relationship between length of residence and voting focus on the administrative burdens that are created when one moves to a new residence, including the costs associated with registering to vote (Ortiz 2009). In essesnce, the greater the administrative burden associated with moving to a new residence, the greater the potential impact on voter behavior. Using data from the 1980 National Election Study, Squire et. al. (1987) analyzes the impact of the ease of registration on voter turnout among American citizens who have recently moved. The authors find that, among those who had lived in their residence for six months or less, voter turnout was 18% higher in states with more permissive voter registration laws than in states with more restrictive laws; related research has arrived at similar conclusions (Highton 2000, 2004; McDonald 2008).[5]

In addition, Highton (2000) makes an important distinction between the impact of moving within one’s community (i.e., maintaining one’s social networks, while potentially facing administrative barriers to voting) versus moving to an entirely new community (i.e., disrupting social networks and facing administrative barriers to voting). The author finds that both forms of moving are negatively associated with the likelihood of political participation. Moreover, he finds that the administrative barriers to voting after a move have a larger negative association with political participation than disruptions to social networks. Using Current Population Survey (CPS) and American National Election Studies (ANES) data from people who move within the same county, Ortiz (2009) similarly finds that barriers to registration are more closely related to voting behavior than any other factor. 

The Present Study

A number of the studies that provide evidence of a positive relationship between length of residence and political participation use data from before 2016 (Wolfinger and Rosenstone 1980; Squire et. al. 1987; Rosenstone and Hansen 1993; Highton 2000; Ortiz 2009; Hansen 2016; Burke et. al. 2019). While scholars from both the administrative and social cost schools widely cite these studies, it is important to reexamine the relationship as society continues to change. As noted in my Background section, the United States is experiencing declining homeownership (GAO 2020) and rising rent rates (U.S. Census Bureau 2021). Many of the citizens most likely to vote are also the most likely to rent homes with set leases; these individuals must contend with the social and administrative costs described above as they move from apartment to apartment (Joint Center for Housing Studies 2020; Wattenberg 1998). By using the most current (2020) data, I provide updated research into this relationship.

Conceptual Framework

Based on the findings detailed in my Literature Review, I hypothesize that length of residence in one’s current home will have a positive correlation with one’s likelihood of voting in the 2020 election. My model will control for economic, demographic, and geographic factors that are plausibly related to my dependent and key independent variables. These factors are diagrammed in Figure 1 and discussed further below.

Demographic Factors

Previous studies have linked several demographic factors to voter turnout. Age, sex, and race are commonly included demographic controls in a number of these studies, which have typically found them to be associated with political participation (Wolfinger and Rosenstone 1980; Squire et al. 1987; Highton 2000, 2004; Ortiz 2009; Burke et al. 2019). In addition, previous research has shown that marital status is associated with voter turnout: adults who never marry are less likely to vote than married adults (Wolfinger 2008).

Economic Factors

A relationship has also been found between economic factors and voter turnout. For instance, research suggests that educational attainment is positively related to the likelihood of voting (Wolfinger and Rosenstone 1980; Squire et al. 1987; Pew Research Center 2006). Socioeconomic status, as measured by family income, has also been positively associated with political participation and voter turnout (Wolfinger and Rosenstone, 1980; Highton 2000, 2004; Ortiz 2009, Pew Research Center 2006). In prior research, occupation or industry of employment have been included as socioeconomic controls (Wolfinger and Rosenstone 1980). 

Geographic Factors

Finally, prior studies of voter participation have controlled for geographic factors. For example, when analyzing voter turnout, researchers frequently control for differences between states (Highton 2004; McDonald 2008; Ansolabehere and Lovett 2008; Ortiz 2009). Additionally, widely-cited prior research has controlled for the differences in metropolitan status – i.e., whether a residence is in an urban area, a suburban area, or a rural area (Wolfinger and Rosenstone 1980).

Data and Methods

My empirical analysis uses individual-level data collected monthly by the U.S. Census Bureau’s Current Population Survey (CPS).[6]  Using both an online questionnaire and person-to-person interviews, the survey gathers data on all members of the sampled household over the age of 15. In most cases, the owner of the household or primary renter responds on behalf of each of the household’s eligible members.[7]  The CPS collects data on a wide array of subjects, including general demographic information, education outcomes, employment, and income and wealth. These data points are used by federal agencies, such as the Bureau of Labor Statistics to create annual reports.

For my analyses, I use the November 2020 Voting Supplement to the CPS, which is the  most recent version of this survey.[8] In addition to the base set of survey questions, the Voting  Supplement includes questions on voter behavior, such as whether respondents voted in the  election earlier in the month, whether and how they registered to vote, and how long they have  lived in their place of residence (IPUMS 2021). Respondents below the age of 18 are not included in my analysis as they are not of legal voting age in the United States. Non-citizens are also removed from my analysis since they cannot participate in American elections. As discussed in my Conceptual Framework section, I control for several demographic, economic, and geographic factors that have been found in prior studies to be related to voter turnout. Data for these controls are all drawn from the 2020 CPS Voting Supplement. 

Due to the binary nature of my dependent variable, I estimate a linear probability model (LPM) to analyze the relationship between length of residence and the likelihood of voting. My primary specification is as follows, with the individual as the unit of analysis:

Pr⁡(Voted) = β0+ β1 Residence + β2Age + β3Sex + β4Race + β5MaritalStatus + β6Education + β7FamilyIncome + β8Industry + β9Metro + State + ε

 where State represents a vector of state fixed effects terms, ε is an error term, and β1 is my coefficient of interest. My dependent variable is a dichotomous measure of whether the respondent voted in the most recent (November) election. My key independent variable is a continuous measure of how long CPS respondents have lived at their current address (measured in months). The inclusion of state-fixed effects in the regression reduces the extent of bias in my estimates by controlling for all unobserved factors that are common to residents in each respective state. My data set contains information on 35,109 individuals who were surveyed for the November 2020 CPS Voting supplement. Table 1 provides definitions for all of the variables included in my model.

Descriptive Statistics

Table 2 reports descriptive statistics for my dependent, key independent, and control variables. All my results have been weighted using a variable created by the U.S Census Bureau to facilitate analysis of their monthly surveys. My analytic sample includes 35,109 individuals from all 50 states and the District of Columbia.[9]  Approximately 81% of the sample group reported that they voted in the November 2020 elections. As discussed in my Introduction, the national voting turnout in the 2020 election was 66.8% (United States Election Project 2020). Thus, the members of my sample have higher levels of political engagement than the U.S. population at large. Respondents’ average length of residence at their current address is 64 months. Sample members are 44 years old on average, and their average household income is $98,778 per year. The average family income in my sample is $31,257 higher than the U.S. family median income reported by the US Census Bureau ($65,521) in 2020 (Shrider et al. 2021). 

 Table 2 also shows that a plurality of participants is categorized as working in the “Professional Services” industry, which is a broad category defined by the Census Bureau to include occupations such as healthcare, education, management, and accounting service professionals. Additionally, a substantial majority of respondents have at least a high school diploma and live in either an urban or a suburban setting. 

Results

Table 3 reports my weighted regression results.[10] Given the dichotomous nature of my dependent variable; I estimate linear probability models (LPM).[11] I report robust standard errors beneath each coefficient. Table 3 presents results for five different specifications. Model (1) is a simple bivariate specification that estimates the relationship between the length of residence in one’s home (measured in years) and the likelihood of voting in the November 2020 election.  Model (2) includes several control variables to reduce the extent of bias in my estimates.  These controls include demographic factors (i.e., age, sex, marital status, and race), economic factors (i.e., income, educational attainment, and industry of employment), and geographic factors (i.e., metropolitan status and state of residence). Model (3) adds state-fixed effects to account for unobserved differences between residents of different states, such as civic engagement or the ease of voting registration.

As I originally hypothesized, the results of Models (1), (2), and (3) suggest that there is a small, positive association between length of residence and voting. These correlations are all statistically significant at the 1% level. More specifically, the results of Model (1) indicate that a one-year increase in the length of residence is associated with a 1.6 percentage point increase in the likelihood of voting in November 2020. As indicated by Table 2 above, the average length of residence in my sample is about 5 years, and about 81% of sample members voted in the 2020 election. Thus, in practical terms, a one-year increase in length of residence is an approximate 20% increase relative to the mean, while a 1.6 percentage point increase in the likelihood of voting reflects an approximate 2% increase relative to the base proportion in my sample. In other words, a relatively large increase in the length of residence corresponds with only a small increase in the likelihood of voting. The addition of control variables in Model (2) results in a decrease in the estimated magnitude of this relationship. The results of this specification suggest that a one-year increase in length of residence is associated with an increase of just under one percentage point in the likelihood of voting; this estimate remains statistically significant at the 1% level. The addition of state-fixed effects in Model (3) has little impact on my estimate of interest.

Models (4) and (5) report the results of my subgroup analyses. Model (4) examines the role that age plays in the relationship of interest. This consideration is potentially important, given the homeownership and renting trends among young Americans discussed in previous sections. For this specification, I created a dichotomous age measure that indicates whether respondents are 30 years old or younger. I then interacted this dummied version of the age variable with the length of residence variable. Similarly, Model (5) examines the role that income plays in the relationship between length of residence and likelihood of voting. Income may moderate my relationship of interest because wealthier individuals could, as discussed earlier, be better able to overcome barriers to voting such as re-registration after migration. I dichotomized the income variable based on the within-sample median, then interacted this dummied version of the income variable with the length of residence variable.

Model (4) produces unexpected results, suggesting that there is a very small and positive, but statistically insignificant, difference in the association of length of residence and voting when examining the interaction of age. More specifically, a one-year increase in the length of residence is associated with a statistically significant 0.93 percentage point increase in the likelihood of voting (0.009 + 0.0003) among those who are 30 or younger, while comparable increase in length of residence is associated with a statically significant 0.9 percentage point increase in the likelihood of voting among those older than 30.[12] However, the difference between these two groups in terms of the association between length of residence and voting is very small and statistically insignificant. These results are not aligned with my expectation that increased length of residence would have a meaningfully greater impact on young people’s voting behavior.  

Model (5) is more aligned with my expectations. I find that there is a small, negative, and statistically significant difference between high and low-income individuals when examining the association between length of residence and likelihood of voting. More specifically, Model (5) suggests that a one-year increase in length of residence is associated with a statistically significant 0.51 percentage point increase (0.0115 - 0.0064) in the likelihood of voting for those with incomes greater than $87,000 per year. In comparison, a one-year increase in the length of residence is associated with a statistically significant 1.2 percentage point increase in the likelihood of voting for the lower income category.[13] The magnitude of this relationship is very small for both groups and there is little practical difference in the association between length of residence and voting when examining the difference across income groups.

In summary, my regressions provide evidence of a relatively weak, but positive and statistically significant, relationship between the length of residence in one’s home and the likelihood of participating in the November 2020 election. The small magnitudes of my estimates are not unexpected, considering the complex motivations for voting in American elections. I also find little evidence of meaningful variation in the relationship between length of residence and likelihood of voting according to age, and only limited evidence of variation according to income. The implications of my results are discussed in the conclusion, along with limitations of my analysis and recommendations for policy and future research.

Conclusion

My results suggest that there is a small, positive, and statistically significant association between length of residence in one’s home and the likelihood of voting in the November 2020 elections. Specifically, after controlling for a number of demographic, economic, and geographic factors, I find that an additional year of residence is associated with an increase of slightly less than a percentage point in the likelihood of voting. My estimates of this relationship are consistently significant at the one percent level. As discussed earlier, there is substantial literature supporting the positive relationship between the length of residence at a person’s address and her or his likelihood of political participation (Wolfinger and Rosenstone 1980; Squire et al. 1987; Rosenstone and Hansen 1993; Highton 2000). My results are consistent with these earlier findings when examining political participation in the November 2020 elections.

I also find that there is little variation in my relationship of interest according to age. This result is at odds with Wolfinger and Rosenstone’s (1980) finding that changes in political participation after a move tend to be more limited among young adults, compared to older adults. The authors suggested that moving in young adulthood is less disruptive than moving later in life because older adults usually have greater familial, financial, and material interests that complicate the moving process, all of which may have implications for political participation. I also find that my relationship of interest is modestly larger among those with lower incomes than among those with higher incomes.

My study has at least two limitations. The first is bias created by missing variables.  Previous studies that examined my relationship of interest (Squire et al. 1987; Highton 2000) controlled for the individual owning or renting their address of residence. These studies found that homeownership is positively associated with the likelihood of voting in elections. Owning a home is associated with longer lengths of residence, and therefore my coefficient is impacted by upward omitted variable bias (Haurin and Rosenthal 2004). In other words, because this source of bias is increasing the magnitude of the coefficients in my results, my small coefficients may be even smaller. Unfortunately, the 2020 edition of the Voter Supplement data from the CPS did not include data on homeownership, so I was unable to control for this potentially important variable; future research should attempt to include this variable. 

Similarly, in their study on the impact of residential mobility on political participation, Squire et al. (1987) controlled for individuals’ levels of political interest. The authors argue that being more politically interested is associated with higher levels of political participation.[14] However, there is limited research on the association between political interest and how long one resides at their address, making the direction of this omitted variable bias difficult to determine. The November 2020 CPS Voting Supplement did not include data on levels of political interest or engagement, so I was unable to include and control for this potentially impactful variable. 

The second limitation of my study is the likelihood of measurement error in my length of residence variable. As discussed in Table 1, my length of residence variable was created by taking the midpoint of a series of categories contained in the underlying variable available in the CPS, and I took the fairly conservative step of assigning a value of eight years for the “Longer than 5 years” category. This likely caused measurement error in the length of residence variable, which subsequently biased the resulting coefficients.

Future Research and Policy Implications

My results warrant additional exploration. First, future researchers should consider examining the relationship between length of residence and political participation during off-year election cycles. American society’s level of interest and engagement in the 2020 election was higher than any other point in the 21st century (Galston 2020), as evidenced by the approximately seven percentage point increase in voter turnout between the 2016 and 2020 elections (DeSilver 2021). Pew Research Polling indicated that 83% of Americans thought that the winner of the 2020 election would have important consequences for American policy and progress, compared to just 50% in 2000 (Galston 2020). Political engagement and participation are typically lower during mid-term and off-cycle elections (DeSilver 2014). Therefore, political participation may be more sensitive to individual-level factors in election years when society-level factors are less pronounced. In other words, individual-level factors, such as a person’s length of residence or a person’s level of education, may have greater influence on political participation in years when society as a whole is less interested in the election. Individual-level factors like my variable of interest may have had less influence in 2020 given the high levels of societal interest in the election. 

Second, future research should examine and control the impact of the COVID-19 pandemic on the association between length of residence and political participation. COVID-19 had considerable influence on the way in which the 2020 election was carried out. Mail-in ballots became much more prevalent, which may have contributed to the surge of participation across the country (DeSilver 2021). Future researchers should attempt to control for this impact. 

As previously discussed, political participation is a hallmark of a functioning and healthy democracy, and it supports democratic values such as popular sovereignty, majority rule, and legitimate elections (Hill 2006; Anzia 2012; Fowler 2013; Chapman 2016). My results suggest that length of residence is associated with increased political participation in the most recent U.S.  general election. This association is worth further exploration. If future research can establish a causal relationship between length of residence and voting, policymakers would benefit from promoting policies that facilitate length of residence, incentivize home ownership or long-term leases as a tool in their arsenal to increase political participation. Such policies could provide stability for American citizens, to engage with the needs of their community on a deeper level, and ultimately to increase their political participation and engagement in the American political system.

Appendix