Abstract
This article explores the extent, and possible causes, of income-based biases in representation of citizens by members of the 110th Congress. The author finds that the preferences of wealthier citizens are modestly but significantly better reflected in the choices of their congressional representatives than are the preferences of poorer citizens. More importantly, the author shows that education, political sophistication, political engagement, ethnicity, and other sociodemographic factors can explain only a small part of this representation gap. Biases in representation across income lines appear to be driven by income alone, or at least not by politically relevant factors correlated with income.
A substantial body of research shows that the electorate’s policy preferences are well (if certainly not perfectly) represented in legislative behavior and policy outcomes. This research has typically dealt with aggregated preferences of the entire electorate (e.g., Erikson, MacKuen, and Stimson 2002; Wlezien 2004), explicitly precluding the possibility that certain types of citizens might be better represented than others. But spurred in part by the American Political Science Association’s (2004) Task Force on Inequality, scholars have paid increasing attention to the topic of equality of representation, particularly across income lines. There remains substantial debate over the nature of income-based biases in representation, with some arguing that preferences of richer citizens are substantially better represented than those of the poor and others suggesting that representational biases are minimal (see, e.g., Bartels 2006, 2008; Gilens 2005, 2011; Soroka and Wlezien 2010; Rigby and Wright 2011; Erikson and Bhatti 2011).
More importantly, despite the growing line of work exploring the extent of income-based biases in policy representation, there has been little theoretical or empirical consideration of the factors that might lead the preferences of wealthier citizens to be better reflected in policy making. The development of a fuller explanation for why income-based biases in representation might persist is important to building a theoretical understanding of the role of income in policy representation and is relevant to the evaluation of possible substantive reforms that might help to equalize political influence across income lines.
This article takes a first step toward addressing this issue, working to shed light on the roles that political resources, political engagement, and sociodemographic factors play in driving income-based biases in representation. I develop measures of dyadic representation of citizens by their congressional representatives in the 110th House and then use these measures to assess the degree to which wealthier citizens’ preferences are more congruent than poorer citizens’ preferences with the voting behavior of their members of Congress (MCs) and the degree to which such biases in representation are driven by politically consequential factors correlated with income, as opposed to income itself.
I find evidence of differences in representation across income lines: MC voting behavior in general, and votes on important bills in particular, tends to correspond more closely to the preferences of wealthier than poorer constituents. Representation gaps are small, given that differences in preferences across income lines are themselves small but systematic. In addition, I find wealthier citizens’ greater levels of education, political knowledge, political activity, voting, and other relevant factors can explain only a small part of these gaps. Politically engaged and knowledgeable wealthy citizens are better represented than the comparably engaged and knowledgeable poor, and MCs tend to represent wealthy citizens who share their party affiliation better than poorer citizens who also share that affiliation.
This article makes three contributions. First, it contributes to the ongoing debate regarding the extent of economic inequalities in representation, providing one of the first studies of income and policy representation within the U.S. House of Representatives. Second, it pushes the theoretical foundations of this debate forward by examining more closely the factors that might lead to these inequalities. While this analysis cannot definitively explain why income biases in representation exist, it does show that biases are not driven by education, political engagement, or a number of other theoretically and politically relevant correlates of income. As such, it takes a step toward developing a more general understanding of why wealthier citizens are often more likely to get what they want from policy-making institutions. Third, these analyses provide a method through which scholars might more deeply explore the individual and contextual factors that exacerbate or diminish income-based biases in representation.
Income and Economic Biases in Representation
Scholars have taken an increasing interest in the role of income in shaping political outcomes and patterns of political influence in the United States (e.g., Jacobs and Skocpol 2005; Bartels 2008; Kelly 2009). Central to these discussions are the nature and role of inequalities in policy representation. The meaning of political “inequality” is a complex and far-ranging concept. But the consensus definition of inequality in the representation of preferences in policy is more straightforward. “Inequality” in representation exists if the preferences of some citizens are weighted more heavily than others in policy-making decisions or if the preferences of some citizens are more congruent with policy outcomes than the preferences of other citizens (Gilens 2005; Griffin and Newman 2005; Ura and Ellis 2008; Soroka and Wlezien 2010). At the level of representation of citizens by MCs (my interest in this article), economic biases in representation mean that the voting decisions of MCs are systematically more likely to reflect the views of certain (typically, richer) citizens than others (Bartels 2006).
The potential for such biases has long concerned scholars of public policy and American political processes. Equality of representation is a central part of a well-functioning democratic system: in the extreme case, severe biases may even beg the question of whether the United States can be called a democracy (Bartels 2008). Even more subtle inequities in representation, though, can have substantial consequences for the protection of citizen rights and the nature of the policy-making process (see Verba and Orren 1985; Griffin and Newman 2008).
There remains substantial debate regarding the extent of income-related biases in representation of this sort. Evidence of unequal representation can be difficult to find, simply because on many issues, preference differences across income lines are not large enough for rich and poor citizens to send substantively different messages to policy makers (Ura and Ellis 2008). If rich and poor are asking for the same thing from government, then the representation of one group’s preferences would necessarily ensure the representation of the other’s. Where differences in opinion across income lines do exist, the dominant view is that the preferences of the wealthy generally matter more to policy making (e.g., Gilens 2005, 2011; Jacobs and Page 2005; Bartels 2006, 2008). Others argue, however, that income biases are small or inconsistent (e.g., Soroka and Wlezien 2010; Rigby and Wright 2011; Erikson and Bhatti 2011).
Behind the empirical debate regarding the extent of income bias is the comparably more neglected question of why lower income citizens might be less well represented. On one hand, it is perhaps intuitive to think that wealthy citizens will have representational advantages over poor ones. On the other, though, reelection-seeking policy makers, in theory, have little incentive to systematically ignore the preferences of any of their constituents, lest they face electoral sanction (Soroka and Wlezien 2010). It is thus not straightforward to expect that they will discount the preferences of certain types of constituents, at least without compelling reason to do so.
Possible Reasons for Economic Biases in Representation
In general, extant work has largely left open the question of whether it is income alone, as opposed to politically relevant factors correlated with income, that are behind representational inequities. Low-income citizens are disadvantaged in the political process in many ways that go beyond income: they vote less, participate less, tend to know and care less about policy, and are less likely to have the political “resources” necessary to voice their views (e.g., Althaus 2003; Verba, Schlozman, and Brady 1995). They are also more likely to be members of demographic groups historically disadvantaged in the political process (e.g., Griffin and Newman 2008). These factors, to the extent that they shape whose preferences legislators hear and whose they deem important to heed, might explain why poorer citizens would be less well represented.
In fact, much prior discussion on this topic—in particular, the American Political Science Association’s (2004) Task Force report—implicitly or explicitly point to these sorts of resource- or engagement-based factors as the driving forces behind biases in representation. Proposals to institute compulsory voting (or to make voting easier), to limit the amount that private citizens can donate to campaigns, and to educate and stimulate political engagement among the poor all have their roots, at least to some extent, in a desire to remedy perceived inequalities in political influence across income lines (e.g., Piven and Cloward 1988; Lijphart 1997).
In addition to income alone, then, there are many different plausible, but indirect, factors that might help to drive observed economic biases in representation. These diverse sets of factors have implications for understanding the significance of representational biases and, if such biases are deemed to be damaging to representative democracy, for the types of solutions that might be employed to remedy them. Exploring in more detail the role of some of these factors will help to shed light on the possible reasons for, and consequences of, economic biases in representation.
Voting and Political Participation
For many reasons, poorer citizens are typically less politically active than wealthier ones (Rosenstone and Hansen 1993). Income disparities in participation are modest when it comes to voting but larger with respect to more active forms of participation (such as working for political causes and donating money to political campaigns). There are two central reasons to expect participants to be better represented than nonparticipants. First, participants are more likely to be represented by legislators who share their political worldviews. No individual citizen can ensure that he or she is represented by a like-minded legislator. But at least at the margins, those who are active in politics are more likely to be represented by legislators who share their views than those who are not. Voters are by definition more influential in shaping election outcomes than nonvoters, and political activists can also be disproportionately influential in helping to elect candidates that reflect their views (e.g., Layman et al. 2010).
Second, the opinions of participants are more likely to affect policy even apart from elections. Classic models of representation of public opinion are driven in large part by the threat of electoral sanction. 1 Legislators must balance many of competing motivations when deciding how to act, and it makes little sense to take into account the preferences of particular subsets of constituents when there is minimal threat that ignoring their views will lead to electoral punishment. Those who participate are perceived as proving a more credible threat of electoral sanction and are better represented as a result (Griffin and Newman 2005). To the extent that income is correlated with levels of political activity, we might see that biases in representation are largely a function of differential levels of political engagement across income lines.
Education and Political Knowledge
Income is also strongly correlated with the possession of political “resources,” particularly education and political knowledge (Delli Carpini and Keeter 1997; Verba, Schlozman, and Brady 1995). Such resources are critical in shaping citizens’ abilities to form informed opinions on policy issues and their abilities to hold policy makers accountable for their actions. Again, there are reasons to believe that the more educated and knowledgeable will be better represented. First, educated and knowledgeable citizens are better equipped to form and voice coherent, stable, and well-developed preferences on matters of public policy. The inability to develop coherent preferences necessarily precludes policy representation, as it is unclear what “representation” of public preferences even means if those preferences are not themselves systematically meaningful (Soroka and Wlezien 2010). If educated or knowledgeable citizens send signals that are less “noisy,” and more systematic, than the less educated or knowledgeable, it is logical to expect that the former will have their preferences better represented than the latter, simply because those preferences are more likely to be cleanly reflected in measures of public opinion and will be clearer, more stable, and easier for policy makers to perceive. 2
Second, educated and knowledgeable citizens are generally more attentive to politics in general and to their MC’s behavior in particular. Given the election-seeking motivations of legislators, an MC will be more likely to disregard a segment of their district’s preferences if that segment of the constituency is unlikely to notice that their preferences have been disregarded (Erikson and Bhatti 2011). The association between income and these types of resources means that income biases could be driven largely by poorer citizens’ lower levels of education and knowledge.
Race, Ethnicity, and Gender
For a variety of reasons, racial and ethnic minorities are typically less well represented in the behavior of Congress members than white citizens (Hero and Tolbert 1995; Griffin and Newman 2008). To a lesser extent, the same may be true of the representation of women compared to men (Griffin and Newman 2007). Since race, ethnicity, and gender are correlated with income, gaps in representation across income lines may be explained in large part by these factors, inexorably intertwined with long-standing issues of racial and gender inequality.
Income
Finally, it is possible that biases across income lines will persist independent of these factors, going beyond income-based differences in engagement, education, race, or other politically relevant correlates. Instead, the biases might stem from factors related to the possession of income itself that gives wealthier citizens a greater formal and informal voice in designing and directing social institutions. There are many reasons to think that income-based gaps in representation might exist above and beyond what can be explained by politically relevant correlates. In many contexts, wealthier people possess advantages in the political arena that go beyond the ability to participate meaningfully in politics or to donate to political causes. Wealthier citizens, regardless of their levels of education or political interest, have better access to formal and informal political and social networks that help to set policy agendas and shape public debate on political issues (e.g., Lindblom 1997; Schlozman 1984). They are also more likely to internalize the behavioral norms that make them apt to have their preferences taken seriously by public bureaucracies and other forms of public and private authority (e.g., Galanter 1999; Beeghley 2007). Furthermore, most MCs are themselves well-off and thus might be more apt to interact with constituents who are well-off (Erikson and Bhatti 2011).
All of these ideas suggest, consistent with a number of lines of work in political science, sociology, and other fields, that wealthier citizens’ preferences are “visible” (Fallows 2000) in ways that go beyond their levels of education or political engagement. These ideas also raise the possibility that income-related biases would persist even after taking participation- or resource-based factors into consideration. In this account, the issue of income bias is more subtle than if it were largely a function of these factors: policies designed to boost lower income voter turnout, restrict the role of private money in politics, or engage low-income citizens in the political process would be incomplete, at best.
Measuring Representation
Following the discussion above, the goals in this article are twofold. First, I wish to examine the extent to which lower and upper income citizens have their preferences represented in the choices of MCs. Second, I wish to take steps toward understanding why such biases exist—in particular, in examining the extent to which biases can be explained by resource- or engagement-based factors, and the extent to which they persist after taking such factors into account. The focus here is on dyadic representation, the degree to which citizens’ preferences are reflected in the policy decisions of their own representatives. 3 I develop two measures of representation, both based on roll-call voting data for members of the 110th House and mass opinion data from the 2008 Cooperative Congressional Election Study (CCES).
The first, building from the methodology of Griffin and Newman (2005, 2007, 2008) and others, is a basic measure of the ideological proximity between citizens and representatives. This measure simply relates the ideological leanings of citizens to the ideological positions of the MCs who represent them and provides a measure of the absolute distance between a citizen’s self-expressed political ideology and the ideological location of his or her MC.
Congress members’ ideologies are operationalized using first dimension W-NOMINATE scores for the 110th Congress (Poole and Rosenthal 2007). First-dimension NOMINATE scores reflect members’ positions on the liberal–conservative continuum that drives most elite political conflict in the United States and thus provide a useful summary measure of the general ideological leanings of MCs. I measure the ideological location of citizens through responses to a CCES question asking respondents to place themselves on a 0 (extremely liberal) to 100 (extremely conservative) scale. I then can use these data to calculate measures of the absolute ideological distance between individual citizens and their congressional representatives.
This measure has several strengths. The first of these is breadth: NOMINATE scores reflect the sum total of MC voting behavior on contested issues, and ideological self-placement can serve as a simple summary measure of citizens’ political beliefs. As such, the measure provides a useful way to examine representation using the widest possible conception of congressional voting. The second strength is a strong grounding in the literature: measures of representation that relate legislators’ roll-call votes to citizens’ self-placement have been used extensively in work addressing both representation and representational inequity (e.g., Bartels 2006, 2008; Griffin and Newman 2005, 2007, 2008; Clinton 2006; Griffin and Flavin 2007).
The measure also has some disadvantages. Foremost, of course, is the long-noted difficulty of placing legislator behavior and citizen preferences on the same scale in a way that makes the computation of ideological proximity possible. Scholars have taken three general approaches to dealing with this problem. Some suggest that ideological identification and roll-call voting measures can simply be scaled to the same range (Achen 1978). Others have argued that both measures should be standardized in some way (Wright 1978). Finally, some suggest that MCs’ ideologies need to be rescaled to a narrower range than that of citizens, given the paucity of truly “extreme” legislators (Powell 1982). There remains debate over the appropriate way to deal with this problem (see Burden 2004). Fortunately, however, the empirical consequences of the choice are relatively minor: all results to come remain statistically and substantively similar regardless of the approach used (results available on request).
More importantly, it is clear that none of these approaches provides an ideal measure of ideological distance. But these problems are mitigated because of the focus on relative, not absolute, representation. I make no strong claims about the substantive meaning of any particular level of ideological distance—whether being twenty ideological “points” away from an MC, for example, signifies that one’s preferences are represented “well” or “poorly.” Rather, I argue that a citizen whose preferences are closer to those of his or her MC than another’s according to this measure is likely better represented. 4 I derive my ideological distance measure of representation using the Wright (1978) approach. To create this measure, I first standardize both citizen and legislator ideological placements. 5 I then compute the measure of ideological representation by calculating the absolute distance between a citizen’s standardized ideological identification and his or her MC’s standardized NOMINATE score. To aid in interpretation, I rescale this measure to a 0 to 100 metric. All else equal, I assume that a citizen whose ideological leanings fall very close to those reflected in the voting behavior of his or her MC is better represented than one whose ideology is distant from his or her MC. We would see evidence of bias across economic lines, then, if ideological distance decreases as income increases.
The second measure of representation addresses the degree to which citizens get their preferences represented on a small number of salient issues. The CCES asked respondents for whether they supported or opposed key pieces of legislation voted on by the 110th House: bills that would (1) provide a timeline for the withdrawal of troops from Iraq, (2) increase the minimum wage, (3) permit stem cell research, (4) expand eavesdropping activities under of the PATRIOT Act, (5) expand access to subsidized health insurance for children, (6) ban gay marriage, (7) provide federal assistance to homeowners, and (8) expand key NAFTA provisions. Respondents were simply asked whether they, in principle, “favored” or “opposed” the legislation. Because respondents were asked about the same specific pieces of legislation on which MCs actually voted, I can compute a measure of representation that calculates the number of times that a citizens’ preference corresponded to his or her legislator’s actual vote.
I again use survey responses and roll-call data to compute measures of representation for each respondent. First, I simply code whether a citizen’s preference was reflected in the vote of his or her MC on a particular issue, creating dummy variables coded 1 if a citizen supported a piece of legislation and his or her MC voted for it or if a citizen opposed the legislation and his or her representative voted against it, and 0 if the citizen supported the legislation and his or her MC voted against it (or vice versa). I then sum across all issues to compute a measure, which I label key vote representation, that divides the number of issues on which a citizen had his or her preference represented by the number of issues on which there is usable data (i.e., the issues on which a citizen expressed a preference and his or her MC cast a vote). I again scale this measure to a 0 to 100 metric, where the highest score indicates a citizen whose preferences were reflected in the votes of his or her MC for all of the issues on which there are valid data and the lowest score indicates a citizen whose preferences were the opposite of his or her MC’s votes on all issues with valid data. We would see evidence of economic bias according to this measure if key vote representation increases as income increases. 6
This measure has advantages over the ideological distance scale. The first of these is the comparably stronger connection between concept and measure. Because respondents are asked about actual, specific roll-call votes, the scaling of citizens’ and legislators’ preferences together into a measure of “representation” is less arbitrary: if a citizen supported a piece of legislation, and his or her representative voted for it (or vice versa), we can say with a reasonable degree of confidence that the citizen’s preference was represented. Second, the eight votes are prominent ones, chosen in part because of their high levels of salience and policy importance. The measure thus may provide a better, or at least a different, sense of representation than a measure that aggregates over all votes, even minor and obscure ones.
The weaknesses of the measure are twofold. First, the votes included in the CCES, while important, are not selected at random. It is thus possible that results are driven by the particular mix of issues chosen. Second, the votes used to create the measure were cast prior to the CCES being in the field. This leaves open the possibility that constituent preferences might be affected by MC votes (Hill and Hurley 1999). I cannot say the degree to which such effects exist and the degree to which this might affect the results. Each measure has advantages and limitations, so employing both and finding consistent results can help to validate the findings for each.
Income and Mass Preferences
Representation gaps across income lines can exist only to the extent that there are differences in preferences across income lines. Table 1 presents a basic illustration of such differences, showing the opinions of citizens in three income terciles on the 100-point self-placement scale, each of the eight roll-call issues, and a summary measure of preferences aggregating across all issues. 7 Preference differences, though generally small, are systematic. Citizens in the richest tercile place themselves as modestly (less than 2 points on the 100-point self-placement scale) but significantly (p < .05, two-tailed) more conservative than poorer ones. While these differences are small, even very minor shifts in the preferences of the electorate—on the order of one or two points—can have sizable policy consequences (Erikson, MacKuen, and Stimson 2002), so even these relatively small differences can have substantive implications for policy making if the preferences of one group are better represented than another’s.
Issue and Ideological Positions by Income Tercile
The differences are larger when it comes to specific policy issues. On five of the eight issues, preferences of rich and poor differ by more than eight percentage points. In addition, on three of the eight issues, the majority preferences of rich and poor differ. When aggregating the ideological direction of preferences across all issues (coding positions that were supported by a majority of House Democrats as liberal and those supported by House Republicans as conservative), we see that wealthier citizens are significantly more conservative—taking, on average, slightly less than one additional conservative position than poorer citizens.
Income and Representation: Bivariate Results
With these measures in hand, I first take a basic look at whether these differences in preferences translate into differences in representation. As a first cut, I simply examine representation across the three income terciles. With respect to both ideological and win ratio representation, we see that the poorest citizens are significantly (p < .05) less well represented than the middle tercile, who are in turn less well represented than the richest. With respect to key vote representation, MCs vote in a way congruent with their poorest constituents an average of 56.8 percent of the time (σmean = 0.28), which is less often than they vote with the middle tercile (M = 57.7, σmean = 0.29) or the wealthiest tercile (M = 58.5, σmean = 0.27). When it comes to ideological representation, the poorest citizens are on average 26.1 points away from their MC (σmean = 0.19), which is again further away from their representative than are the middle (M = 25.0, σmean = 0.16) and upper (M = 24.7, σmean = 0.16) terciles.
The sizes of the representation gaps are not overwhelming, which is to be expected given the small size of the preference differences across groups. But the differences are important, particularly given that small preference differences make it difficult to uncover evidence of unequal representation even if it did exist (Ura and Ellis 2008).
To get a sense of the relative magnitude of these gaps, Figure 1 compares the size of the gaps between the richest and poorest terciles to gaps across other lines of political difference. These other bivariate results generally conform to intuition: educated and politically knowledgeable citizens are better represented than their less educated and knowledgeable counterparts, whites are better represented than blacks, and voters and activists are generally better represented than nonvoters and nonactivists. 8 The gaps between rich and poor appear large when compared to these other factors: gaps between the richest and poorest terciles, for example, are larger than the gaps between voters and nonvoters, or between those with high and low levels of knowledge. Only the gap between blacks and whites is consistently larger than that between the richest and poorest terciles.
Representation gaps among politically relevant subgroups
Income, Resources, and Representation Gaps
I now examine the possible reasons for of these biases, exploring the extent to which economic biases are driven by politically relevant correlates of income. To do so, I estimate a series of ordinary least squares (OLS) regression models in Table 2, modeling citizens’ ideological distance (columns 1–4) and key vote (columns 5–8) representation as a function of income and the engagement, resource, and demographic factors discussed above. Columns 1 and 5 simply model individual representation as a function of a more fine-grained measure that divides income into fourteen distinct categories. 9 Each one-category increase in income is expected to decrease the distance between a citizen and his or her MC by roughly 0.23 points and increase the proportion of the time that his or her preferences on the eight-issue scale are represented by 0.21. The expected differences in representation between the richest and poorest citizens are roughly 3.0 points on the ideological preference scale and 2.6 points on the key vote scale.
Modeling Socioeconomic Biases in Representation
Table entries are ordinary least squares coefficients (standard errors, clustered by congressional district, are in parentheses). Estimates for congressional district fixed effects, included in each model, are not shown.
p < .05.
The successive columns build on this model to explore the roles that engagement, resources, and other politically relevant factors play in shaping income biases in representation. Columns 2 and 6 add in measures of political knowledge and education (measured as a respondent’s highest degree earned) as well as a battery of demographic variables. 10 These variables generally behave as expected (though when I include education and knowledge in the same model, only knowledge remains as a significant predictor of representation). But we see that while the effects of knowledge and demographics on representation are statistically significant and fairly powerful, they are very nearly orthogonal to those of income: the inclusion of these variables decreases the income coefficients from 0.22 to 0.18 for the distance measure and from 0.21 to 0.20 on the win ratio measure.
Columns 3 and 7 add in indicators of political engagement: respondents’ reports of voting in the 2008 election, participating in the 2008 campaign in ways that went beyond voting (a scale measuring whether respondents, in the past year, attended political meetings or rallies, worked for a political cause, put a political sign on their home or car, blogged about politics, attempted to persuade others how to vote, or contacted an elected official to voice an opinion or concern), and donating to a political cause (with a separate variable for those who donated more than $1,000 in the past year). Again, although all behave as expected when considered separately, these participation variables behave somewhat erratically in the multivariate context—contacting an elected official and donating large amounts of money tend to decrease ideological distance, for example, while being an activist increases it and voting has no effect.
The inclusion of the participation measures also tends to mute the independent effects of other relevant variables—knowledge and race/ethnicity are no longer significant predictors of ideological representation, for example, once these factors are taken into account. The reason for this, at least in large part, is the relatively high degree of intercorrelation between predictors. Including this many correlated variables—education, knowledge, political engagement, race—makes it very difficult to read too much into the effects of any particular one. But one exception to this rule appears to be income: the effects of income remain significant and substantively similar, decreasing to only 0.17 in the distance model and to 0.18 in the key vote model.
Finally, columns 4 and 8 include one additional, and more expansive, factor: a simple dummy variable that indicates whether a citizen is represented by a MC that shares his or her party affiliation. Since party drives a substantial portion of both citizen and legislator behavior, one would expect that those whose MC is a “copartisan” to be substantially better represented than those whose MC is not. Many of the factors that lead to copartisan representation are the result of simple geographical happenstance: few congressional districts are so gerrymandered that there would not be a substantial number of wealthy, educated, and engaged citizens who simply live in an area where they are the political minority. But at least a portion of copartisanship is systematic: voting and otherwise working to influence elections can play a substantial role in determining whether one is represented by someone who shares one’s partisan views. Clearly, party representation is critical: the amount of variance in representation explained by copartisanship is substantially more than that of all of the model’s other variables combined. And the inclusion of the copartisan variable does dampen the independent effects of income a bit—the income coefficient falls to 0.15 for the distance measure and 0.16 for the key vote measure. But again, the effects of income remain strong even with this measure included.
Taken together, the results show that whether considering broad ideological leanings or preferences on salient issues, the preferences of wealthier citizens are more closely associated with their MC’s behavior than the preferences of poorer citizens. More importantly, they show that such gaps in representation cannot be explained in whole, or even in large part, by resources, political engagement, or demographics. The inclusion of the full list of resource and engagement variables, including copartisanship, diminishes the independent effects of income by only about 35 percent for the ideological distance measure and 25 percent for the key vote measure.
Furthermore, the effects of income on representation are unusual in their consistency across model specifications and measures. Factors such as education, knowledge, participation, and ethnicity seem, to some extent, to be competing to explain much of the same variance, and thus the independent role of each is contingent on model specification. But the role of income is unique, persisting above and beyond all of these factors. Although these models do not capture the universe of factors that might affect representation, they do represent a large list of possible explanations for income-based inequities. Together, they show that economic biases in representation go deeper than what would be expected from resource or sociodemographic-based explanations
Knowledge, Income, and Representation
We have seen that lower income citizens tend to be less well represented by their MCs. But we also know that lower income citizens tend to be less politically knowledgeable than their wealthier counterparts. Representational gaps across income lines clearly deviate from the ideal of perfectly equal representation. But given the strong associations between income and knowledge (and other indicators of cognitive engagement with politics), some have entertained the argument that we might want the preferences of more knowledgeable (and thus, at the margins, wealthier) citizens to be comparably better represented in policy making (e.g., Verba 2003). The trade-off for unequal representation, in other words, would be policy that conforms to the wishes of the more attentive portion of the electorate. But this argument carries weight only to the extent that it is knowledge itself, and not income, that is the proximate driver of unequal representation—if, in the context of this analysis, knowledgeable poorer citizens were as well represented as their knowledgeable wealthier counterparts and low-knowledge poorer citizens were as poorly represented as their low-knowledge wealthier counterparts.
To examine the role of knowledge in moderating income-based gaps in representation, I segment citizens into six categories, dividing each income tercile into “high” and “low” levels of political knowledge based on whether their scores on the six-question knowledge scale are at or below, or above, the mean. 11 Figure 2 graphs the expected distance between citizens and their MCs for each of these groups. The results are derived from an OLS model similar to that in Table 2, with dummy variables for five of the six income and knowledge categories in place of the income and knowledge scales (models using the key vote measure produce similar results).
Ideological distance gaps by knowledge and income levels
These results illustrate the largely distinct roles that income and knowledge play in representation. Knowledge matters: at each income level, the more knowledgeable are better represented than their less knowledgeable counterparts (though the groups are not statistically distinguishable within the lowest income tercile). But the effects of knowledge do not eliminate the representation gap, as high knowledge, but low-income, citizens are still significantly less well represented than low-knowledge citizens in the richest income tercile. The results point to a modest reinforcing effect of knowledge, as we can say that knowledgeable higher income and middle income citizens are better represented than less knowledgeable ones within the same income category. But stratifying by knowledge does little to remediate income-based gaps in representation. If anything, they suggest that higher levels of knowledge may matter least to representation for those in the lowest income tercile, as the only group that is not significantly better represented (p < .05, two-tailed) than the low-knowledge poor is the high-knowledge poor.
Copartisanship, Income, and Representation
Copartisanship clearly is relevant to representation, more so than any other predictor: the inclusion of this variable in a bivariate regression model explains more than 25 percent of the variance in both the distance and key vote measures of representation. Poor citizens are significantly less likely than wealthier ones to be represented by an MC who shares their party identification: 45 percent of respondents in the lowest income tercile are represented by a copartisan, compared to 50 percent of middle income and 53 percent of upper income citizens. And as we have seen in Table 2, the inclusion of copartisanship has a modest effect on the size of income biases in representation.
But copartisanship does not make the effects of income go away: instead, it highlights another aspect of the representation gap. Figure 3 explores this issue more fully, showing results of models predicting ideological distance as a function of income tercile but restricting the analysis only to those who share a party affiliation with their MCs (i.e., including only Democrats who are represented by Democrats and Republicans represented by Republicans).
Ideological distance gaps by income level, copartisans only
Here, we see that gaps between high- and low-income citizens are modestly larger when considering only citizens who share their MC’s party affiliation. Again, we can say with confidence (p < .05, two-tailed) that the preferences of the wealthiest group are better represented than those of the middle tier, which are in turn better represented than those of the poor. Citizens in the richest income tercile are on average more than two points closer to their MCs than citizens in the middle tercile and roughly three points closer than citizens in the poorest tercile (again, the findings of a parallel analysis using the key vote measure produce similar results). The effects of copartisanship are substantially greater than that of income—even poor copartisans are substantially better represented than richer citizens who do not share their MC’s party affiliation. But although in-party citizens are substantially better represented than out-party ones, the benefits of in-party status accrue disproportionately to upper income citizens.
This basic discussion helps to round out the analysis of economic biases in representation by showing that poor citizens are disadvantaged when it comes to representation in two critical ways. First, they are less likely to be represented by a copartisan MC—the factor that is more important than any other in how well one’s views are reflected in one’s MC’s voting decisions. Second, even if poor citizens are represented by a copartisan, MCs tend to better represent the preferences of upper income copartisans better than those of poor copartisans.
Conclusion
Using two measures of dyadic representation in the 110th House, I have shown that the preferences of lower income citizens are significantly less well represented than those of wealthier citizens in the voting behavior of their MCs. This analysis is one of the first to look at economic biases in the House of Representatives and, as such, helps to validate and extend prior research, typically based on older U.S. Senate data, on ideological representation (e.g., Bartels 2006, 2008; Erikson and Bhatti 2011). In addition, the use of CCES data allows us to develop a novel measure of representation that provides for the direct comparison of citizens’ preferences and MC voting behavior on a small number of highly salient issues. I find that income biases of relatively similar magnitude exist using this measure of representation.
More importantly, the results show that only a small part of this representation gap can be explained by patterns of participation, knowledge, education, or other factors correlated with income. The results suggest that a substantial portion of the income bias in representation is a function of income itself, or at the very least is not driven by factors related directly to unequal levels of participation or engagement or the fact that poor citizens come from historically underrepresented demographic groups. Differences in representation are a matter of degree, not magnitude, largely because differences in preferences across income lines are relatively small. But the fact that biases exist and are robust to a variety of different model specifications speaks to the pervasiveness of subtle but meaningful income-based gaps in representation.
While the analysis does not definitively rule out all possible resource- or engagement-based factors in explaining economic biases, it does show that explanations of why biases exist need to go well beyond the fact that wealthy people know, donate, and participate more than poorer citizens. This is not to say that greater equity in resources and engagement across income would not help to remediate biases in representation. We cannot explore, for example, the counterfactual situation where wealthy and poor did participate in politics at relatively equal rates. It is possible, for example, that an electorate in which participation was more equal across income lines might cause elected officials to pay more attention to the views of poor citizens. We can say only that for those interested in remediating income-based biases in representation, a focus on increasing the political engagement of the poor or placing restrictions on the political engagement of the wealthy (e.g., by limiting private financing of political campaigns) will be far from sufficient in eliminating income-based gaps in representation without more of a wide-ranging consideration of the role of wealth in shaping political and social systems.
The analysis comes with some limitations. First, the results are based on the 110th House alone. The magnitude of income gaps that I find corresponds well with those from other research using different time periods (e.g., Griffin and Newman 2005), and I see little reason why these results should be fundamentally affected by my choice of time period. But I clearly cannot say for certain whether the results are limited to this Congress alone. 12 In addition, while the results have demonstrated that MC votes are more representative of the preferences of wealthier constituents, we cannot directly observe evidence that wealthier citizens are more influential in influencing MC behavior. The greater congruence between wealthier citizens’ preferences and MC behavior could result from greater influence of the wealthy over MC action.
It might also result in part, however, from two other factors: the possibility that wealthier citizens’ preferences are more likely to be influenced by the behavior of their MCs and the possibility that wealthier citizens, because of their higher levels of political understanding and participation in public debate, are more likely to react to real-world political conditions in a way that mirrors their MCs. There is reason to think that meaningful differences in congruence also imply meaningful differences in influence (see, e.g., Gilens 2005 and Griffin and Newman 2008 for broader discussions of this point). But given my data and approach, we cannot rule out either possibility explicitly, and it is probably the case that both are true to at least some extent.
In addition, this analysis suggests possible directions for future research dealing with representational inequity across income lines. In particular, it would be useful to build on these results in the development of a broader theory of why income biases in representation exist. We have demonstrated that income biases in representation persist after controlling for a variety of political and sociodemographic factors known to affect citizen representation. This is a meaningful finding, from both substantive and theoretical perspectives. But it alone says little about why income matters above and beyond these correlates. Given this limitation, it would be useful to further unpack the role of the “income” variable, attempting to operationalize and test more directly some of the more subtle sorts of ways that wealth may translate into political influence. More generally, future research might address some of the more complex linkages between income and representation suggested by other lines of work: the possibility, for example, that wealthier citizens’ preferences are better represented because they are more likely to be voiced by organized interest groups (e.g., Baumgartner et al. 2009; Schlozman 1984).
Finally, this article has focused solely on individual-level factors that affect biases in representation, taking into account but not modeling contextual factors that might affect how well poor citizens are represented vis-à-vis the rich. But it is clear that context should matter as well. It is likely the case, for example, that poorer citizens will be better represented in certain types of House districts than others. Similarly, we might expect income gaps in representation to be greater at certain times or in certain political contexts than others. This approach to this topic has provided a method through which future work can explore how individual and contextual factors interact to shape the size and persistence of income-based gaps in representation.
Footnotes
Acknowledgements
A previous version of this article was presented at the 2010 annual meeting of the American Political Science Association. The author wishes to thank Scott Meinke, Kent Tedin, and Joseph Ura for helpful comments and criticisms.
Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.