Political Representation and the Geography of Legislative Districts Jaclyn Kaslovsky Harvard University Jon C. Rogowski Harvard University May 30, 2018 Abstract The process of assigning voters to districts is inherently political and has motivated a voluminous literature on phenomena such as malapportionment and racial and partisan gerrymandering. Yet while the districting process also induces unavoidable geographic variation in legislative districts, scholars have devoted less attention to studying their implications for legislative behavior. In this paper, we argue that district geography affects lawmakers legislative capacities and document the representational consequences of the geographic size of U.S. House districts from 1985 to 2008. Using a within-member design, we show that legislators from larger districts accrued substantially fewer policy accomplishments and secured significantly less federal funds. Additional evidence suggests that these representational consequences result from how district size structures legislators capacities for investing resources in policy-focused activities. Our findings have important implications for theories of representation in single-member districts and suggest new avenues for studying how geography affects political outcomes. We thank Chris Gibson, Mike Olson, Dan Smith, and Jim Snyder for helpful suggestions. Ph.D. Candidate, Department of Government, 1737 Cambridge St, Cambridge, MA 02138; jkaslovsky@g. harvard.edu. Assistant Professor, Department of Government, 1737 Cambridge St, Cambridge, MA 02138; rogowski@ fas.harvard.edu.
Drawing district lines is one of the central tasks of republican governments. Districts define political constituencies who elect representatives of their choice and to whom they delegate authority for making political decisions on their behalf. The placement of district boundaries further affects the electoral fortunes of incumbent and would-be officeholders and the quality of political representation afforded to constituents. And at the aggregate level, how district lines are drawn has implications for which political party controls government. The considerable normative implications of districting have generated a voluminous literature from political scientists, legal scholars, historians, and others which studies the consequences of malapportionment (e.g., Ansolabehere, Gerber, and Snyder 2002; Horiuchi and Saito 2003), the politics of racial and partisan districting (e.g., Cain 1985; Chen and Rodden 2013; Engstrom 2013; King and Browning 1987; Niemi et al. 1990), and the effects of redistricting on voter turnout, election outcomes, responsiveness, and substantive representation (e.g., Ansolabehere, Snyder, and Stewart 2000; Cameron, Epstein, and O Halloran 1996; Fraga 2016; Gelman and King 1994a,b; Hayes and McKee 2009, 2012). In this paper, we argue that the physical geography of legislative districts induced by the districting process affects how legislators represent their constituents. Existing scholarship details how principles such as equal apportionment, compactness, contiguity, partisan and racial composition, and communities of interest are (or should be) considered when drawing district maps (Altman 1998; Gilligan and Matsusaka 2006; McDonald 1996; Niemi et al. 1990). Yet mapmakers must also contend with the physical nature of the territory and the dispersion of the population across space. These characteristics lead to variation in the geographic size, shape, and accessibility of legislative districts as a mechanical feature of the districting process and have been almost entirely overlooked as a contributor to distinct patterns of legislative behavior. While existing scholarship documents a wide range of political factors that contribute to variation in a legislator s political influence or effectiveness (Alexander, Berry, and Howell 2016; Anzia and Berry 2011; Berry, Burden, and Howell 2010; Berry and Fowler 2016, 2018; Hitt, Volden, and Wiseman 1
2017; Lee 1998; Lee and Oppenheimer 1999; Lee 2000; Rogowski 2016; Volden, Wiseman, and Wittmen 2013; Volden and Wiseman 2014, Forthcoming), to our knowledge none has focused on the geographic features of political jurisdictions. We argue that larger districts provide political incentives for legislators to dedicate more of their resources to district-focused activities which reduces their capacity to engage in the legislative process. We study the importance of political geography for representation by focusing on legislative behavior in the United States House of Representatives from 1985 to 2008. Using panel data on legislative outcomes and a within-legislator design, we document new relationships between the geographic size of congressional districts and legislative behavior. Across a variety of measures of policy accomplishment, we find that increases in district size were associated with systematically lower levels of legislative effectiveness and significantly smaller shares of federal program spending. These results are robust across a range of model specifications and additional analyses. Consistent with our proposed mechanism, we use original data on the allocation of legislators staff to Washington, D.C. and district-based offices and show that legislators representing larger districts allocate greater shares of their staff members to district offices. Our findings demonstrate how the geography of legislative districts, including territorial size, affects legislative behavior and suggest that districting procedures and principles can affect political representation in possibly unintended ways. The results also suggest that the capacity to engage in policymaking may vary across legislators and is an important source of variation in legislative productivity. Electoral Districts and Political Representation Electoral districts are central for political representation because the placement of district lines defines representatives constituencies. In turn, the behavior of reelection-seeking legislators depends on the composition of their constituency (e.g., Mayhew 1974). For instance, legislators have electoral incentives to craft roll call voting records that are consistent with the political and 2
partisan views of their constituents (Canes-Wrone, Brady, and Cogan 2002; Carson et al. 2010), pursue legislative action on the issues prioritized by their constituents (Sulkin 2011), and secure federal resources for their districts and advertise their success in doing so (Grimmer, Messing, and Westwood 2012). A legislator s constituency affects not only her choice about what roll call vote to cast, issues to prioritize, or federal projects to pursue, but it also affects the relative mix of attention and emphasis a legislator places on the various components of her job (e.g., Ashworth and Bueno de Mesquita 2006; Grimmer 2013). Yet not all legislative districts are created equal. Some districts are heavily Democratic or Republican while others are more evenly balanced across partisan lines. Some districts have dense concentrations of specific interest groups, such as public sector employees unions or health care providers, while other districts have more diffuse sets of interests. Some districts are heavily populated by constituents from a single racial or ethnic group, while other districts are composed of constituents from a more varied set of racial and ethnic backgrounds. Each of these sources of variation has important consequences for which candidates emerge, win election, and ultimately behave while in office. Legislative districts also differ in geographic size. Some legislators represent small, compact, densely populated urban centers; other legislators represent larger districts knitted together across suburban sprawl; and still others represent large swath s of a state s territory composed mostly of rural communities which consist mostly of farmland and/or rugged terrain. We argue that these differences in geographic size contribute to differences in political representation. Legislators who represent larger districts must expend greater resources to provide constituency service and make their presence felt at home. This allocation of resources, however, constrains legislators capacity to engage in the legislative process while in Washington, D.C. These factors lead us to expect that legislators who represent large geographic districts experience decreased Washington-focused productivity, as we will elaborate below. Our focus on geographic size as a contributor to legislative representation contrasts with ex- 3
isting research that studies the predictors of legislative behavior. These perspectives generally emphasize how representation is affected by constituency demand, generally operationalized as demographic and political characteristics, in addition to a legislator s background, personal ideology, and interest in satisfying party leaders or capitulating to presidential pressure. These accounts overlook, however, the ways in which representation can be affected by the process of dividing a state into districts whose populations are approximately equal. The incentives for these patterns of representation are therefore mechanically induced rather than politically motivated. Though scholarly accounts of legislative behavior generally exclude district geography as a contributing factor, lively debates during the American Founding centered on the geographic nature of political representation. According to Zagarri (1987), delegates to the Constitutional Convention from small states viewed territorial units as the basis for representation while large-state delegates perceived population and demography as the basic unit of representation. Delegates from small states worried that their states small geographic territories would therefore limit their potential political influence, thus leading William Paterson from New Jersey to argue that if we are to be considered as a nation, all state distinctions must be abolished. 1 A similar perspective appeared a decade earlier in debates over apportionment in the Virginia state legislature, when a writer to The Virginia Gazette argued for redrawing each county s boundaries so that each contained the same amount of territory. 2 Later scholarly accounts point out that the only difference between the upper and lower chambers of state legislatures is that the upper chamber consists of fewer members representing larger districts (Wilson 1911, 487) and identify House members representation of smaller and sometimes shifting parts of states as one of the most important differences between the U.S. House and Senate (Froman 1967, 6-7). Variation in the geographic size of legislative districts, however, has been less appreciated by subsequent research. Our research has some parallels with, yet is conceptually distinct from, research which stud- 1 Madison debates, June 9, 1787; available at http://avalon.law.yale.edu/18th century/ debates 609.asp. 2 June 7, 1776; quoted in Zagarri (1987, 69). 4
ies how population size and district magnitude affect representatives behavior and legislative outcomes. When some districts have greater numbers of constituents than other districts under malapportionment, constituents in districts with smaller populations wield relatively greater influence over aggregate political outcomes (e.g., Lee 2000). Consistent with this logic, Ansolabehere, Gerber, and Snyder (2002) demonstrate how court-mandated redistricting in the mid- 1960s helped equalize transfers to counties on the basis of population. The Senate may be the most infamous example of malapportionment in the United States, which leads Senate legislative outcomes to better reflect the preferences of Senators from less populous states and whose votes are easier to secure for the purposes of forming legislative coalitions (Lee 1998; Lee and Oppenheimer 1999; Lee 2000). This literature focuses on how mechanisms such as vote dilution and variation in the price of coalition-building link district size and legislative behavior. Though somewhat more distantly related, other research focuses on how district magnitude influences political representation, and argues that voters difficulty in attributing outcomes to individual legislators from multi-members districts produces free-riding incentives. The free-riding problem reduces levels of constituency service and weakens relationships between roll call voting behavior and constituency opinion (Ashworth and Bueno de Mesquita 2006; Portmann, Stadelmann, and Eichenberger 2012; Rogowski 2017). By contrast, we posit that district geography induces officeholders to adopt representational strategies that reflect the mechanistic and political incentives generated by the dispersion or concentration of constituents across space. How Geography Affects Representation We develop the argument outlined above in the context of the United States House of Representatives, though its core elements apply to legislatures where representatives are chosen by electoral districts. Overall, we argue that legislators who represent larger districts confront resource constraints which limit legislators focus on Washington-based accomplishments. Primarily, we posit 5
that legislators from large districts allocate greater shares of their staff resources to district offices. Larger geographic districts require more district offices to minimize the time required for constituents to interact with a representative s district staff (Griffin and Flavin 2011). Every House member receives a fixed representational allowance for personnel salaries and have been limited to no more than 18 permanent personal staff in recent decades (Brudnick 2016). This budget constraint requires legislators from large districts to forego some Washington, D.C.-based staffers in favor of maintaining greater staff presence in district offices. Because, as Salisbury and Shepsle (1981, 560) wrote, the core of any congressional enterprise is the personal staff of the member, the allocation of staff resources by legislators in larger districts deprives the D.C. offices of valuable human capital, expertise, and staff time (Fenno 1978; Hall 1996). In turn, this constraint reduces legislators ability to pursue legislative accomplishments because Washington-based staff are generally dedicated to legislative matters and help legislators write, recruit support for, and navigate legislation through Congress (Montgomery and Nyhan 2017; Schiff and Smith 1983). Thus, legislators who represent large districts likely have less capacity to pursue legislative accomplishments. Consistent with this expectation, Schiller (1995) shows that the size of a Senator s personal legislative staff has significant effects on the number of bills Senators introduce. Increases in district size may also present political incentives for legislators to strategically allocate greater staff resources to district offices. For instance, larger districts may also be more complex political environments in which policy-focused accomplishments may have more limited appeal. In contrast with small urban districts, which tend to be comprised disproportionately by Democratic constituencies (Chen and Rodden 2013), larger and more rural districts are likely to be composed of many small groups of constituents with distinct political interests. These districts may be more politically fraught because there are more groups within the electorate who can be picked off by an entrepreneurial challenger (Fenno 1977, 877). Consistent with this expectation, Wichowsky (2012) shows that district complexity increases with the percentage of a district s rural population. This political diversity may also contribute to greater competitiveness 6
in larger districts, particularly if their constituents are drawn from collections of rural, suburban, and urban communities. District complexity and competitiveness are both likely to decrease legislators emphasis on Washington-focused accomplishments (Wichowsky 2012) and instead lead legislators to dedicate their resources to crafting local reputations for constituency service. This explanation is consistent with the model posited by Ashworth and Bueno de Mesquita (2006), in which more competitive electoral environments increase legislators provision of constituency service. To recapitulate, our argument posits that district size is an important source of variation in legislative behavior. More specifically, we expect that legislators in larger districts place less emphasis on and experience less success in the legislative process because of decreased legislative capacity. Our focus on the relationship between district size and legislative behavior contributes to scholarly interest in identifying the determinants of legislative effectiveness (Anzia and Berry 2011; Hitt, Volden, and Wiseman 2017; Matthews 1960; Volden and Wiseman 2014, Forthcoming). This research generally focuses on individual-level predictors of effectiveness, such as gender, seniority, partisanship, and ideology. We offer two contrasts. First, we study variation in legislative behavior across various indicators of effectiveness; second, we focus on how these patterns are shaped by district-level factors specifically, the geography of political jurisdictions rather the personal attributes of legislators or their institutional power within the legislature. The remainder of our paper examines our expectations empirically and explores our hypothesized mechanism. Before proceeding, we point out several normative implications suggested by our argument. First, to the extent representatives in larger districts produce fewer legislative achievements, district size may be an important and understudied source of political inequality. Second, and related to the first, constituents from larger districts may also exert less influence in policy debates over the substance of legislative outcomes. Instead, policy debates may be shaped disproportionately by legislators who represent smaller and more densely populated districts. Third, other research suggests that less-populous states have more influence relative to their size on legisla- 7
tive outcomes and presidential elections due to the design of the U.S. Senate. Identifying how district geographic size affects legislative behavior has important implications for characterizing the quality of political representation and understanding how these factors may offset the institutional advantages enjoyed by less-populous states in the design of the federal system. Data and Empirical Strategy We begin our analysis by studying the relationship between district geography and representation using data from the 99th through 110th House of Representatives, which cover the period 1985 to 2008. We focus on two key dependent variables which capture two dimensions of legislative achievements that have been studied in recent scholarship (Anzia and Berry 2011; Berry and Fowler 2016; Volden, Wiseman, and Wittmen 2013; Volden and Wiseman 2014; Hitt, Volden, and Wiseman 2017; Volden and Wiseman Forthcoming). First, we use a measure of legislator effectiveness with scores developed by Volden and Wiseman (2014). These scores describe a legislator s Volden and Wiseman (2014, 18) proven ability to advance a member s agenda items through the legislative process and into law while also accounting for variation in the substantive importance of each bill. The scores are calculated for each congress, where legislators with higher scores sponsored more legislation, were more successful in moving those bills through the legislative process (i.e., through the committee stage, when receiving floor votes, etc.), and did so on bills that were substantive and/or significant. The scores are normalized to have a mean of 1 in each congress. Second, we use data on the provision of federal outlays to study legislators success in procuring federal program spending for their districts. Credit-claiming and advertising have long been recognized as fundamental motivations for reelection-seeking incumbents (Mayhew 1974) and federal resources provide opportunities for both. Theoretical work further supports the role of federal spending as a signal to constituents of a legislator s underlying quality (Ashworth and 8
Bueno de Mesquita 2006). Consistent with the expectations offered by these accounts, a variety of evidence links the provision of federal resources to increased electoral support for members of Congress (Alvarez 1997; Grimmer, Messing, and Westwood 2012; Lazarus and Reilly 2010; Levitt and Snyder 1997) and the president (Kriner and Reeves 2012). Following prior research (Berry, Burden, and Howell 2010; Dynes and Huber 2015), we use information from the Federal Assistance Award Data System (FAADS) to measure high-variance program spending in individual House districts for the congresses under study. These data measure most federal transfers to domestic recipients with the exclusion of defense spending and procurement contracts and were obtained from Dynes and Huber (2015). Dollar amounts are converted to 2010 real values. The median level of district outlays was approximately $458 million and ranged from $13 million to $26 billion. 3 We follow previous research (Anzia and Berry 2011; Berry, Burden, and Howell 2010; Berry and Fowler 2016; Dynes and Huber 2015) in using the logged values of this measure. We make several assumptions by using these measures as our dependent variables. We assume, first, that both measures reflect legislator quality. To the extent we find similar results across both measures increases our confidence that larger values of them indicate more effective legislators. Second, we assume that legislators have common incentives to secure legislative achievements; in the case of the legislator effectiveness scores, we also assume legislators have sincere preferences in favor of the bills they introduce. Third, we assume that writing original legislator and securing federal program spending come with opportunity costs by requiring the costly commitment of resources in both time and expertise. Fourth, and finally, we assume that constituents prefer to elect high-quality officials and provide more positive assessments of effective legislators that less effective legislators. The key explanatory variable in our analysis is the geographic size of U.S. House districts, 3 Receipt of these funds likely reflect some combination of legislator effort in addition to federal formulas and agency decisions. While our focus is on the former, we cannot fully rule out the latter as potential influences on the allocation of program spending. So long as formulas, agency decisions, and related factors outside the hands of congressional representatives are uncorrelated with the size of congressional districts, the results support our interpretation that these data can be used to measure legislators representational priorities. 9
which we measure with the square mileage of district area. This variable is measured for each redistricting cycle. The median-sized district in our data is 2,146 square miles in area, which is slightly larger than the size of Delaware. The smallest district was NY-15 following the 1980 redistricting at seven square miles, and the largest was the at-large district for Alaska (572,000 square miles). The distribution of district size in square miles is highly skewed and thus we use the natural log of its values. Studying our hypotheses in systematic fashion presents several identification challenges. If legislators were randomly assigned to districts of varying size, we could simply compare legislators accomplishments on the basis of whether they were assigned to represent small or large districts. Without random assignment, however, a simple cross-sectional comparison of the relationship between district size and legislative accomplishments is subject to two key sources of confounding. First, legislators could have different time-invariant levels of quality or ability. Without accounting for these (potentially unobservable) characteristics, cross-sectional comparisons could generate inappropriate empirical findings about the relationship between district size and legislative accomplishments if legislators from larger districts tend to be lower quality on average. 4 Second, legislators from districts of varying size could have fundamentally different preferences. For instance, smaller districts are (by definition) more densely populated, more urban, and more African American, and each of these factors is associated with the election of more Democratic and liberal representatives. In the context of federal outlays, representatives from smaller districts could secure more federal funds due to their preferences for greater federal spending. In both cases, any differences in legislative accomplishments would be inappropriately attributed to district size when they instead result from unmeasured or unobserved factors that are correlated with it. We address these issues and test our hypotheses using panel data on legislative outcomes and 4 A negative correlation between district size and legislator quality could arise from, for instance, a relatively small pool of potential candidates with political experience in large districts, or the decreased ability of voters to distinguish candidates on the basis of quality in large districts, perhaps due to sparser information environments. 10
a differences-in-differences design to estimate the within-legislator effects of district size. Our unit of analysis is a legislator i serving in congress c. This design represents the strongest possible approach short of random assignment. In particular, we observe the outcomes of legislative behavior over multiple congresses, and the same legislator often represents districts of varying size. Most commonly, this variation is induced by the decennial Census, after which virtually every state redraws its district lines to account for population shifts, particularly in states which gained or lost seats in the House. During the period under study, several states (including North Carolina in 1998 and Texas in 2003) also redrew U.S. House districts in intercensal years following court-ordered redistricting. We leverage these sources of variation to identify how they relate to changes in legislative behavior. This within-legislator design is particularly appropriate for testing our hypotheses because it allows us to hold constant each legislator s ability and preferences. Failing to do so, our results would be vulnerable to omitted variable bias due to the failure to account for each legislator s underlying characteristics which also contribute to different observable outcomes. Overall, about 78 percent of the legislators in our data served more than one term between 1984 and 2008. Of these, about 27 percent of the legislator-congress observations experienced at least some change in the size of their district from the previous congressional term. Approximately 15 percent of legislators districts grew in size while the other 12 percent reduced in size. The average congress-to-congress absolute change in district size was about 12 percent, or 321 square miles. As noted above, virtually all continuing legislators (> 98%) experienced at least some change in district size following the 1990 and 2000 Census and a number of other legislators experienced changes in district size due to intercensal redistricting. We use this variation to identify our key quantities of interest, which we estimate with the following linear regression model: Y ic = β 0 + β 1 District size ic + X ic Ω + δ i + T c + ε ic, (1) 11
where Y is the relevant dependent variable, District size is the district s area in square miles, and Ω is a vector of coefficients for a matrix of controls X ic described below. The subscripts i and c index legislators and congresses, respectively. We include legislator fixed effects, denoted with δ i, to characterize legislator-specific factors that contribute to differences in legislative behavior, and indicators for each year (T c ) to account for differences in legislative behavior that are correlated with time. With this specification, estimates of β 1 reflect within-legislator differences in behavior as they represent districts of varying size. Finally, ε ic is a random error term, which we cluster on legislator. Though we report results below from a baseline model which excludes control variables (see, e.g., Lenz and Sahn 2017), our model specification follows related literature on legislator effectiveness (Berry and Fowler 2018; Volden, Wiseman, and Wittmen 2013; Volden and Wiseman 2014; Hitt, Volden, and Wiseman 2017; Volden and Wiseman Forthcoming) and the distribution of federal outlays (Alexander, Berry, and Howell 2016; Anzia and Berry 2011; Berry, Burden, and Howell 2010; Berry and Fowler 2016). Accordingly, we estimate models which include a range of control variables that may affect a legislator s incentives or advantages to secure particular legislative outcomes. First, following Alexander, Berry, and Howell (2016), we include the population of each member s district, district median income (in 2010 dollars, logged), and the ideological distance (using DW-NOMINATE scores) between legislator i and the chamber median in congress c. 5 Second, we estimate models which include the district s support for the sitting president in the most recent presidential election, the percent of district votes cast for the Democratic presidential nominee in the most recent election, and linear and squared measures of the percentage of the vote received by the legislator in the most recent election. Finally, in our most fully specified models, we also account for whether the legislator is a member of the president s party, the majority party, serves as a committee or subcommittee chair, is a majority or minority party 5 District square mileage, population, and demographics are taken from a combination of the decennial census and Scott Adler s data on congressional districts. See https://sites.google.com/a/colorado.edu/ adler-scott/data/congressional-district-data. 12
leader, or serves on an important committee (which includes Appropriations, Ways and Means, and Rules); the legislator s seniority in years, plus its squared value; and the size of the House delegation from the legislator s state. Results We begin by studying how district size affects legislative effectiveness. These results are shown in Table 1 below. In our baseline model, shown in column (1), we examine the bivariate relationship between effectiveness and district size. In column (2), we include the population of each member s district, district median income (in 2010 dollars, logged), and the ideological distance between a legislator and the House median. The results shown in the third column account for the electoral environment and constituency preferences by including the president s vote share in the district (as an indicator of the constituency s alignment with the sitting presidential administration), the district s support for the Democratic presidential candidate in the most recent election (as an indicator of district preferences), and linear and quadratic expressions of the vote share received by the sitting legislator (as indicators of electoral competitiveness). Our fully specified models, shown in column (4), also account for the legislators political alignment with the president and House majority party, time in office, chamber and committee leadership positions, and service on key committees in addition to the size of legislators state delegations to the U.S. House. In each model, we include legislator and congress fixed effects. Across all four models, we find strong, consistent, and robust evidence that district size negatively affects legislator effectiveness. The coefficients for District size are negative and statistically significant in each of the four models. The magnitudes of the estimates are relatively stable across each and range from -0.112 to -0.129. Based on the most conservative estimate from this set of models, provided in column (4), our results imply that a one standard deviation increase in district size (31,040 square miles) from the mean district size (8,088 square miles) would reduce a legislator s effectiveness score from 0.83 to 0.66, or by approximately 20%. 13
The coefficient estimates for the other covariates are also of substantive interest and provide several new findings. Legislators who represent more populous constituencies exhibit greater effectiveness, possibly because of more substantial electoral pressures or constituencies to please. Interestingly, legislative effectiveness is negatively associated with district income; as legislators represent more well-to-do constituencies, they are less effective. One potential explanation, though there certainly may be others, is that legislators in wealthier districts spend greater amounts of effort fundraising for themselves and their party at the expense of legislating. In models (2) and (3), moreover, we find that legislators located further from the chamber median are less effective. These legislators may find it more difficult to assemble a successful legislation coalition behind a bill that reflects their sincerely held preferences. However, we do not want to overinterpret this result because the coefficient switches sign and loses statistical significance in model (4). We do not find much evidence that electoral considerations or constituency preferences are associated with legislator effectiveness. However, we do find that legislators gain effectiveness when they gain positions of power within the institution including majority party status and serving as committee and subcommittee chairs. However, members who ascend to particularly powerful committees lose effectiveness, consistent with evidence from related research in this area. Finally, we find no evidence that experience alone, as measured by seniority, is systematically related to legislator effectiveness, which suggests that learning may not be a significant contributor to effectiveness. Table 2 shows results from our regressions of federal outlays on district size. Consistent with findings reported in Table 1, increases in district size are associated with significantly fewer federal program dollars for a legislator s constituents. Across each column, the coefficient estimate for District size is relatively constant and around -0.06. The coefficient estimates imply that federal funds reduce by about 0.7% for every ten percent increase in district size. Following the 1990 and 2000 rounds of districting, the size of legislators districts changed (in absolute terms) by an
Table 1: District Size and Legislative Effectiveness in the U.S. House, 1985 2008 Legislative Effectiveness Score (1) (2) (3) (4) District size -0.120* -0.120* -0.129* -0.112* (0.059) (0.053) (0.053) (0.048) Population (logged) 2.296* 2.152* 1.463* (0.391) (0.398) (0.354) Median income (2010 dollars, logged) -1.096* -1.139* -1.019* (0.303) (0.305) (0.272) Absolute distance from floor median (DW-NOMINATE) -3.420* -3.376* 0.084 (0.118) (0.121) (0.330) President s vote share in the district -0.003-0.000 (0.002) (0.002) Presidential Democratic vote share in the district -0.010-0.005 (0.006) (0.005) Percent vote received to enter this Congress 0.015 0.033* (0.016) (0.014) Percent vote received to enter this Congress, squared -0.000-0.000* (0.000) (0.000) Same party as the President 0.005 (0.052) Majority party member 0.691* (0.132) Seniority -0.022 (0.045) Seniority, squared 0.002 (0.001) Committee chair 2.996* (0.100) Subcommittee chair 0.735* (0.061) Size of House delegation -0.005 (0.014) Key committee member -0.190* (0.081) Majority party leadership 0.218 (0.153) Minority party leadership -0.279 (0.147) Constant 1.145* -16.616* -14.232* -8.949 (0.435) (5.522) (5.734) (5.123) Member Fixed Effects Yes Yes Yes Yes Congress Fixed Effects Yes Yes Yes Yes Observations 4983 4983 4983 4983 The dependent variable is the legislator s effectiveness score from Volden and Wiseman (2014). Entries are linear regression coefficients with standard errors clustered on legislator in parentheses. * p < 0.05 15
average of 67 percent; our estimates suggest that an increase of this size reduced federal outlays for the average district by nearly five percent. For the median congressional district which received $458 million annually during the period under study, these estimates imply that a district which experienced an average increase in size would receive about $30 per capita less in federal funds. 6 Altogether, the results in Tables 1 and 2 provide strong support for our argument. Increases in a district s geographic size reduce a legislator s effectiveness in advancing legislation and securing federal program spending for their constituents. Moreover, we find no evidence that the results are driven by any particular congressional term or wave of redistricting. Our substantive findings for both dependent variables are unchanged when estimating models in which we sequentially omit one congress at a time. Our theoretical argument posited that legislators from larger districts must devote more time and effort to constituency service and other means of representation beyond proposing legislation and securing program spending. Consistent with this proposed mechanism, in additional analyses we continue to find that members from larger districts appear to disengage from the legislative process relative to their colleagues. We estimated similar models to those above while predicting how frequently legislators cosponsored legislation proposed by their colleagues. These results are displayed in Table A.1. Compared to authoring legislation and shepherding it through the legislative process, cosponsorship is a relatively costless way for legislators to signal their policy preferences to their colleagues and constituents and can provide inexpensive opportunities for position-taking. Nevertheless, we continue to find that legislators from larger districts cosponsor significantly fewer pieces of legislation. Even when presented with relatively costless opportunities to engage in the legislative process, legislators from larger districts appear to devote less attention to doing so. 6 This figure is obtained by dividing a 4.7 percent decrease in outlays ($21,600,000) by the average population per district in 2010 (308.7 million divided by 435 districts = 710,000 residents per district). 16
Table 2: District Size and Federal Outlays in the U.S. House, 1985 2008 Federal outlays, 2010 dollars (logged) (1) (2) (3) (4) District size -0.059* -0.063* -0.054* -0.057* (0.018) (0.018) (0.018) (0.018) Population (logged) 0.414* 0.447* 0.431* (0.130) (0.132) (0.131) Median income (2010 dollars, logged) -0.323* -0.278* -0.260* (0.101) (0.101) (0.101) Absolute distance from floor median (DW-NOMINATE) -0.038-0.069-0.290* (0.039) (0.040) (0.122) President s vote share in the district 0.000 0.001 (0.001) (0.001) Presidential Democratic vote share in the district 0.008* 0.008* (0.002) (0.002) Percent vote received to enter this Congress -0.008-0.008 (0.005) (0.005) Percent vote received to enter this Congress, squared 0.000 0.000 (0.000) (0.000) Same party as the President -0.031 (0.019) Majority party member -0.075 (0.049) Seniority -0.152* (0.017) Seniority, squared -0.000 (0.000) Committee chair 0.001 (0.037) Subcommittee chair -0.018 (0.023) Size of House delegation -0.032* (0.005) Key committee member -0.002 (0.030) Majority party leadership 0.012 (0.056) Minority party leadership 0.017 (0.054) Constant 20.254* 18.175* 17.214* 17.957* (0.131) (1.837) (1.902) (1.893) Member Fixed Effects Yes Yes Yes Yes Congress Fixed Effects Yes Yes Yes Yes Observations 4983 4983 4983 4983 The dependent variable is the logged value of federal outlays in 2010 dollars. Entries are linear regression coefficients with standard errors clustered on legislator in parentheses. * p < 0.05 17
Institutional Power and Potential Moderators Theoretical and empirical scholarship identifies a number of institutional and personal factors which advantage some representatives over others in the legislative process. For instance, committee chairs, members of the majority party, copartisans of the president, and members of powerful committees may be able to translate their institutional power into greater legislative effectiveness and shares of federal spending (e.g., Berry, Burden, and Howell 2010; Berry and Fowler 2016, 2018; Fourinaies 2018; Rogowski 2016; Volden and Wiseman 2014). Legislators may also be able to leverage their personal characteristics to achieve more favorable legislative outcomes; more instance, more senior representatives have greater knowledge about the legislative process while more moderate legislators may be common targets of vote-buying (e.g., Alexander, Berry, and Howell 2016). In additional analyses, we explored whether these factors moderated the relationship between legislative accomplishments and district size. In doing so, we study whether institutionally well-positioned legislators are insulated from the negative effects of increased district size. We find little evidence, however, that institutional power insulates legislators from the negative consequences of district size for legislative accomplishments. The statistically significant negative effects of district size on legislative effectiveness and federal outlays persist across legislators on the basis of their service as committee chairs, members of the majority party, copartisans of the president, members of powerful committees, seniority, and distance from the chamber median. The coefficients for the interaction terms between these factors and district size are inconsistently signed and small in magnitude, and none reach conventional levels of statistical significance. On the basis of these findings, legislators with institutional power do not appear to be invulernable to the negative effects of increased district size. 18
District Size, Legislator Capacity, and Political Representation The results above document the relationship between district geography and congressional representation. Following our theoretical discussion, we explore a potential mechanism that may contribute to the relationships shown above. Specifically, we suggested that limits on capacity may constrain legislators from larger districts from pursuing legislative accomplishments at the same rates as their colleagues from smaller geographic districts. Crafting legislation and jockeying for federal spending is costly, and the costs of doing so may simply be too high for legislators from large districts. Geographically large districts require legislators to invest greater effort and resources to district-based activities. For instance, to effectively serve their constituents, they likely have incentives to open more district offices and dedicate greater percentages of their allocated staff to serve district needs rather than work in their Washington office. Because staff are an important contributor to a representative s legislative capacity, the reduction of Washington staff for legislators serving in large districts may reduce those legislators capacities to craft legislation of their own. We provide a preliminary analysis of this proposed mechanism using original data on the allocation of House members staff. Overall, the number of personal staff for House members has remained relatively constant over the last 35 years, averaging around 16 staffers per legislator between 1979 and 2015 (Reynolds 2017). 7 During this period, members of Congress have made increased use of staff resources in district offices; while about one-third of House personal staff worked in district offices in 1978, nearly half of House staffers worked in district offices by 2015. 8 We collected data on the locations of each House members staffers from the United States 7 However, we note that the number of personal staff declined by 14% between 2010 and 2015; it is unclear whether this is a consequence of Republican House leadership or indicative of a more permanent trend. 8 Both these trends also characterize the Senate. The number of personal staff per Senator varied between 37 and 43 during this period while the percentage of personal Senate staff in state offices increased from 25 to 43 percent between 1978 and 2015. 19
House Telephone Directory for 2007 to 2015. 9 Figure 1 below shows the allocation of House staff to district offices from the 110th through the 113th congresses. As the figure shows, many legislators distributed their staff about evenly between the district and Washington. However, the figure also shows substantial variation, with some representatives choosing to keep as few as ten percent of their staff in district offices. Figure 1: Congressional Staff Allocated to District Offices, 110th to 113th Houses Source: U.S. House of Representatives Telephone Directory, 2008 2015. Collected from HeinOnline. We study the link between district size and legislative capacity by estimating models where the percentage of staffers allocated to district offices is the dependent variable. 10 Because there was only a single round of redistricting during the time period for which data are available, we 9 Unfortunately, we have been unable to locate electronic versions of earlier directories. However, we are in process of collecting these data for a wider time period for use in future research. 10 While we would like to evaluate our hypothesized mechanisms more formally in the context of the models we estimated above, unfortunately there is insufficient temporal overlap in the available staffing data and our primary models estimated above. 20
study the relationship between district size and the allocation of staff using a cross-sectional framework. This limits our ability to reach strong causal conclusions about the effect of district size on legislator capacity; however, our analysis allows to evaluate whether the data support a necessary condition if our proposed mechanism is operative in explaining the main results shown above. We estimate a model that regresses the percentage of staff in district offices on district size and the other covariates included in the models above. We also include state fixed effects to account for state-specific factors (such as, for instance, distance from Washington, D.C.) which may also influence legislators staffing decisions. Therefore, our main coefficients of interest are identified using within-state differences in how legislators allocate staff. We estimate these models separately for each Congress as well as pooled across the four. 11 Standard errors are clustered on states. The results are shown below in Table 3. Consistent with our theoretical argument, the results in Table 3 provide evidence that district size shapes how members of the House of Representatives allocate staff resources. In each column, the coefficient for District size is positive, indicating that representatives from larger districts allocated more of their staff to district offices rather than to their Washington, D.C. office. The coefficients are statistically significant in the pooled model (column 5) and for two of the four individual congresses, and imply that a ten percent increase in district size is associated with a legislator allocating approximately an additional four to nine percent of their staff to district offices, which equates to roughly one or two fewer staff members in Washington, D.C. The results in Table 3 thus provide evidence for a potential mechanism that links geographic size with legislative behavior. We emphasize that this is not the only possible mechanisms that could explain why legislators from larger districts are more successful at securing legislative outcomes and federal program spending. Nevertheless, the findings suggest that the outcomes of legislative districting have consequences not only for legislators electoral incentives, but that 11 Several of our covariates are not included in the congress-by-congress models due to perfect collinearity. 21
Table 3: District Size and the Allocation of Congressional Staff Percent of Staff in District 110th Congress 111th Congress 112th Congress 113th Congress All Congresses District size 0.807 0.796 0.511 0.362 0.609 (0.359) (0.321) (0.352) (0.303) (0.156) Population (logged) -211.324 292.800 81.731-21.148-0.654 (729.937) (658.466) (690.252) (29.239) (8.342) Median income (2010 dollars, logged) -10.646-8.361-7.611-8.932-9.341 (2.171) (2.057) (2.089) (2.442) (1.005) Absolute distance from floor median (DW-NOMINATE) 2.447-1.869-5.846-0.115-1.523 (2.839) (2.904) (3.010) (3.325) (1.404) President s vote share in the district -0.012 0.060 0.013 0.146 0.012 (0.060) (0.049) (0.055) (0.057) (0.019) Same party as the President -5.044 0.467 6.121-2.931-0.273 (2.188) (2.372) (2.232) (2.219) (0.553) Percent vote received to enter this Congress 0.185-0.599-0.054 0.262-0.026 (0.311) (0.276) (0.305) (0.345) (0.139) Percent vote received to enter this Congress squared -0.001 0.004 0.001-0.002 0.000 (0.002) (0.002) (0.002) (0.002) (0.001) Seniority 0.703 0.722 0.362 0.556 0.563 (0.289) (0.251) (0.265) (0.255) (0.122) Seniority squared -0.044-0.043-0.025-0.024-0.033 (0.015) (0.012) (0.013) (0.012) (0.006) Committee chair -1.541-0.420 0.865-0.066-0.563 (2.150) (1.904) (1.917) (2.018) (0.910) Subcommittee chair -1.509-0.540-1.594-1.860-1.547 (1.193) (1.046) (1.128) (1.151) (0.521) Key committee member -2.469-1.624-1.719-1.410-1.828 (0.913) (0.837) (0.937) (0.999) (0.433) Majority party leadership -2.684-2.846-2.180-2.247-2.634 (2.210) (1.888) (2.758) (2.774) (1.088) Minority party leadership -0.794 0.645-2.662-0.509-0.571 (2.932) (2.517) (2.263) (2.106) (1.111) Presidential Democratic vote share in the district 0.081 (0.022) Majority party member -0.259 (1.021) Size of House delegation -0.187 (0.254) Constant 2979.934-3766.314-967.963 410.012 154.366 (9765.148) (8809.084) (9234.645) (381.465) (111.591) State Fixed Effects Yes Yes Yes Yes Yes Congress Fixed Effects No No No No Yes Observations 410 423 424 412 1669 The dependent variable is the percentage of a legislator s staff allocated to district offices rather than the Washington office. Entries are linear regression coefficients with standard errors clustered on state in parentheses. p < 0.05 22