American Political Science Review, Page 1 of 18doi:10.1017/S0003055418000606 © American Political Science Association 2018

Legislative Staff and Representation in CongressALEXANDER HERTEL-FERNANDEZ Columbia UniversityMATTO MILDENBERGER University of California Santa BarbaraLEAH C. STOKES University of California Santa Barbara

Legislative staff linkMembers ofCongress and their constituents, theoretically facilitating democraticrepresentation. Yet, little research has examined whether Congressional staff actually recognize thepreferences of their Members’ constituents. Using an original survey of senior U.S. Congressional

staffers, we show that staff systematicallymis-estimate constituent opinions.We then evaluate the sources ofthese misperceptions, using observational analyses and two survey experiments. Staffers who rely moreheavily on conservative and business interest groups for policy information havemore skewed perceptionsof constituent opinion.Egocentric biases also shape staff perceptions.Ourfindings complicate assumptionsthat Congress represents constituent opinion, and help to explain why Congress often appears so unre-sponsive to ordinary citizens. We conclude that scholars should focus more closely on legislative aides askey actors in the policymaking process, both in the United States and across other advanced democracies.

INTRODUCTION

Representative democracy rests on the idea thatelected officials understand and act on theircitizens’ opinions. Yet research has struggled to

demonstrate the link between representatives and theirconstituents, especially in the United States. Instead,there is growing evidence that elected officials aremoreresponsive to interest groups and wealthier individualsthan to average citizens (Bartels 2008; Druckman andJacobs 2015; Gilens 2012; Gilens and Page 2014) andthat state politicians from both parties systematicallyoverestimate the conservatism of their constituents(Broockman and Skovron 2018).1 Still, political scien-tists do not have a clear picture of why legislatorsmisperceive constituent preferences and respond sounequally to the mass public. In this paper, we examineone important and previously underappreciated mech-anism propagating inequalities in legislative responsive-ness: senior legislative staffers’perceptions, attitudes, andbehaviors. Legislative staff act as a bridge between

electedofficials, thepublic, and interestgroups.Aswewilldocument, despite their best efforts to estimate thepreferences of their Members’ constituents, senior leg-islative staff have very skewed perceptions of publicattitudes. We also find that egocentric bias and interestgroup contact, especially with conservative groups andbusinesses, may drive some of the mismatch betweenstaffer perceptions and actual public opinion.

We reach these conclusions using an original surveyof senior legislative staffers in Congress merged withmass public opinion data on five policies: gun control,carbon pollution restrictions, repeal of the AffordableCare Act, infrastructure spending, and raising theminimum wage. This approach allows us to examinehow well senior Congressional aides can characterizethe public’s policy preferences.Across all five issues, wefind that staffers do not accurately identify their districtor state’s preferences and often overestimate theirconstituents’ conservatism. We examine four explan-ations for themismatch in staffers’ perceptions: electoralcompetitiveness; staffers’ personal policy preferences;staffers’ experience in Congress; and staffers’ inter-actions with interest groups. We find that staffers’ per-sonal policy preferences and their interest group contactcorrelate most strongly with the opinion-representationgap. Staffers whose personal opinions deviated fromtheir constituents’ opinions were less accurate in theirestimates of district and state preferences. In addition,stafferswho reported greater contactwith corporate andideologically conservative interest groups over liberaland mass-based citizen groups—whether measuredthrough staffers’ own reports or campaign contributionsto that staffer’s Member—were less likely to get theirconstituents’ preferences right.

We present results from two survey experimentsembedded in the legislative staffer survey that providecredible causal evidence for interest groups’ role inexplaining the opinion-representation gap. Using a listexperiment, we find that about 45%of senior legislativestaffers report having changed their opinion aboutlegislation after a group gave theirMember a campaigncontribution. In a second experiment, we show thatstaffers are more likely to interpret correspondence

Alexander Hertel-Fernandez, Assistant Professor, School of Inter-national and Public Affairs, Columbia Universty, alexander.hertel@gmail.com.Matto Mildenberger, Assistant Professor, Department of PoliticalScience, University of California Santa Barbara.AU1Leah Stokes, Assistant Professor, Department of Political Science,University of California Santa Barbara, lstokes@ucsb.edu.Authors listed alphabetically and contributed equally to the projectdesign and manuscript.AU2

The authors thankGeoffreyHenderson for terrific research supportand are grateful to Raymond O’Mara III, Lee Drutman, and KevinKosar for aid in drafting and disseminating the survey. Thanks to TimLaPira, Robert Shapiro, Kent Jennings, and participants at the UCSanta Barbara Psychology, Environment and Public Policy (PEPP)Seminar and the Columbia Sustainable Development Workshop andthe MPSA Political Institutions and Elite Behavior 4 mini-conferencefor feedback on earlier drafts. The Dirksen Congressional Centerprovided funding for the project through its Congressional ResearchGrant program.Replicationfiles are available at theAmerican PoliticalScience Review Dataverse: https://doi.org/10.7910/DVN/OWQNVF.

Received: February 4, 2017; revised: March 27, 2018; accepted: 15August 2018. First published online: nnn

1 But see Enns 2015; Erikson 2015 for opposing perspectives.

from businesses as being more representative of theirconstituents’ preferences than correspondence fromordinary constituents. These findings provide evidencethat donor and interest group contact can shape howstaffers perceive policies, leading to constituent opinionmis-estimates.

These results contribute to our understanding ofCongress, legislative responsiveness, and politicalinequality. Despite the fact that 80%of senior legislativeaides reported that constituent opinion was “extremely”or “very” important in their recommendations to theirbosses, most had only a limited ability to characterizewhat their constituents actuallypreferred.Moreover, theprospect ofmore competitive elections—long thought tobe an incentive for democratic representation—does notappear to spur greater congruence between seniorstaffers and their constituents. Instead,ourobservationaland experimental evidence points to interest groupcontact and egocentric bias as sources of the staffer-constituent representation gap. One reason why Con-gress may seem so unresponsive to ordinary citizens isthat interest group lobbying and campaign contributionsskew staffers’ perceptions of their constituents. SinceCongress, including staffers, hearsmuchmore frequentlyfrom businesses, donors, and organized interest groupsthan individual citizens (Schlozmanetal. 2012), staffmaysubstitute interest groups’ and donors’ positions forconstituent preferences. Our results highlight the needfor scholars of legislatures, lobbying, and public policy tofocusgreaterattentionon legislativeaidesaskeyactors inthe policymaking process, both in the United States andother advanced democracies.

LEGISLATIVE STAFF ANDPOLITICAL REPRESENTATION

Why focus on Congressional staff? Even as observers ofCongress note Members’ growing dependence on theiraides, few studies have explicitly examined staffers’ rolein the policymaking process. Yet, scholarly accounts ofCongress give us good reason to think that staffers are anessential part of the legislative process (Kingdon 1984;Hall 1996; DeGregorio 1988; Hammond 1996; Malbin1980; Montgomery and Nyhan 2016; Romzek and Utter1997; Fenno Jr. 1978).AsHall (1996) described it, “facedwith the press of excessive obligations and the frequentprospect described of needing to be two places at once,members have responded by relying increasingly onstaff.”Thispowerhas led someCongressional scholars toworry about staff serving as “unelected representatives,”acting on their own to shape policy unmoored fromelectoral accountability (Malbin 1980).

The presence of staff in Congress has changed dra-matically over time.Congressional staffing grew rapidlyfromthe1950s through the late1970s, reachingapeak inthe 1980s and followed later by amodest decline.2 Evenwith these reductions, however, Members of Congress

still act as the heads of large “legislative enterprises,”with up to 18 full-time staff in US House offices andpotentially dozens more in US Senate offices andCongressional committees (cf. Salisbury and Shepsle1981).3

While some authors have argued that Congressionalstaff ought tomerely reflectMembers’ preferences givenstaffers’ career incentives for loyalty (DeGregorio1988;Kingdon 1984, 1989), other research suggests thatstaff exert a strong, independent effect on Memberbehavior.After conducting extensive fieldwork in the1970s, Malbin (1980) concluded that while someCongressional staffers acted as effective surrogatesfor their Members and constituents, in many othercases they were entrepreneurial, shaping legislationon their own in meaningful ways. More recentanalysis backs up this conclusion. Using a longi-tudinal dataset of staffers, Montgomery and Nyhan(2016) show that Members who exchange moresenior staff behave more similarly than would oth-erwise be expected, even after taking into accountMember characteristics.

Given this evidence on staffers’ importance, thispaper poses three questions: To what extent do staffersrely on their constituents’ preferences when craftingrecommendations about policy for theirMembers?Arestaffers able to accurately assess those constituents’preferences? And, if not, why? These are similarquestions to those long asked about Members ofCongress (Erikson et al. 1975; Hedlund and Friesema1972;Miller andStokes1963);we insteadplace the focuson staffers (see alsoBroockman and Skovron 2018 for asimilar approach to ours).

Of course, staff may not possess the same incentivesas politicians, and so it is worth reflecting on theirrelationships to politicians and constituents.4 Our the-oretical framework assumes that, as with lawmakers ingeneral, staffers consider both career incentives andpersonal policy preferences in their job. To advance onCapitol Hill, we assume that staff face incentives todevelop a reputation as competent and loyal aides totheir Members. This entails helping Members to passlegislation, service constituents, and engage in publicactivities thatpermitMembers toposition themselves aseffective representatives. We assume that developingsuch a reputation is essential, even for staff who planto leave government. For instance, connections withMembers are very valuable to aspiring lobbyists(Bertrand et al. 2014; McCrain, forthcoming).

For any given policy issue, staffers must gather andsynthesize information about the implications of the

2 See: https://www.brookings.edu/multi-chapter-report/vital-statistics-on-congress/.

3 While Representatives are allotted a maximum of 18 full-time,permanent staff members, US Senate office staffing depends on abudget assigned to each office that takes into account the size of theSenator’s state and their distance from Washington, D.C.4 Members of Congress have a strong incentive to select and retainstaff who will be faithful “agents” to them; for instance, because theyshare a common set of priorities and preferences. While this selectionand retention process complicates assessments of staffers’ effects onMembers, it should not bias our efforts to understand how staffersperceive public opinion.

Alexander Hertel-Fernandez, Matto Mildenberger, and Leah C. Stokes

2

various alternatives, as well as the stances of differentinterests, including their Member’s constituents. Typi-cally, the senior-most policy staff in a Congressionaloffice—Chiefs of Staff or Legislative Directors—willtake these various considerations and incorporate theminto a recommendation for their Member: for instance,whether to vote for a bill, oppose a proposal, or release astatement. Senior staffers have incentives to considerdistrict or state opinion when making recommendationsto theirMembertohelp theirbossesappearresponsive totheir constituents (Hall 1996; Kingdon 1989). Gatheringconstituent opinion, however, is not a costless endeavorfor staffers. As a result, not all staffers may be equallyable to judge theopinionof the citizens in their states anddistricts across all policy issues. Staffers (especially in theHouse) cannot regularly field representative polls oftheir constituents to figure out what those constituentsthink. Instead, we assume that staffers primarily rely onalternative methods to gauge public opinion in theirdistrict: managing and aggregating constituent corre-spondence; holding constituent meetings; hostingtownhalls; keeping in touch with community leaders;following local media; and using interest group publicopinion polls.

Constituent preferences are not the only intereststhat staffers rely upon when making recommendationsto their Members, however. Staffers may also turn toCongressional support agencies, think-tanks, mass-based membership groups, or trade associations.Especially if staffers do not have a good sense of whattheir constituentswant, staffmay rely on interest groupsthat claim to speak for other constituents, like busi-nesses that claimto represent theeconomic interestsofaMember’s district or state. And staffers may also sub-stitute their ownpreferences andopinionswhenmakingrecommendations to theirMembers, or at least use theirown preferences as a filter for understanding what theirconstituents want.

Gaps between constituents’ opinions and staffers’perceptions could thus emerge in a variety of ways.First, senior legislative staff might have idiosyncraticbiases in their preferences for public policy that runagainst constituent attitudes. These biases mightemerge from staffers’ ideologies, racial or ethnicidentities, personal finances, occupational experi-ences, or gender (e.g., Carnes 2013; Swers 2002). Ifstaffers communicate their own preferences to theirbosses, rather than the public’s preferences, this couldintroduce a mismatch in representation. Second,staffers could have a different picture of what theirconstituents want based on the contact staffers havewith a subset of their constituents, like organizedinterest groups, donors, or activists, perhaps becauseof their Member’s electoral pressures or their expe-rience on the job. If a staffer only has contact with anespecially vocal subset of theirMember’s constituency,they might well develop a distorted perception of theirdistrict or state (Stokes 2016). Indeed, this is often anexplicit strategy organized interest groups use whenlobbying Members (Arnold 1990; Kingdon 1984;Kollman 1998). Consistent with this strategy, Miler(2010) has shown that Members of Congress and their

staffs are more likely to mention resource-rich con-stituencies, like businesses or interest groups, whenrecounting the actors who were important to them inpast policy debates. A final possibility is that staffaccurately represent their constituents’ preferences,but that any representation gaps are created elsewherein the legislative process.

To gauge the extent towhichCongressional staffersare, or are not, representing their constituents’preferences, we follow a long tradition in politicalscience and examine the difference between theperceptions that legislative staffers hold of theirconstituents’ opinions and their constituents’ actualpreferences (Broockman and Skovron 2018; Eriksonet al. 1975; Hedlund and Friesema 1972; Miller andStokes 1963). As in earlier studies, we operationalizepolitical representation as the degree of congruencebetween what constituents say they want on surveysand what staffers think their constituents want onthose same policy issues. Our empirical strategy thusrelies on merging an original survey of senior Con-gressional staffers with survey data on Congressionalconstituencies’ opinions.

It is worth acknowledging here that we are notexamining legislative staffers’ effect on specific policychoices or outcomes. Our evidence speaks to staffers’abilities to correctly perceive their constituents’ pref-erences, not their ability to shape legislation. And,identifying a congruence between constituent opinionand legislative staffers’ perceptions of that opinion doesnot necessarily imply that constituent preferences arecausally informing legislative staffers’ actions (indeed,seeJacobsandShapiro2000 forevidence to thecontraryin a related context). While recognizing these limi-tations, we note that staffers’ correct estimation ofpublic opinion is a necessary though insufficient con-dition for citizens’ representation inCongress. If staffersare not getting their constituents’ preferences right,then it is unlikely that staffers’ inputs into the legislativeprocess will reflect the majority preferences of thecitizens in their districts and states.5

THE 2016 LEGISLATIVE STAFFER SURVEY

In August 2016, we fielded a survey instrument tar-geting senior legislative staff in each Congressionaloffice.Typically, the target staffers’ job titleswereeither‘Chief of Staff’ or ‘Legislative Director’. We chose tosurvey these aides because these are the individualswithin each Congressional office responsible fordeveloping the legislative and political agenda for their

5 We also acknowledge that we are focused on public preferencesrelated to specific policy issues, rather than the mass public’s moregeneral ideological orientation. It may well be that individuals whoespouse a “liberal” position on these specific issues would still expressa conservative political outlook (see canonically Free and Cantril1967). Such an extension of our studywould be a fruitful line of futureresearch.

Legislative Staff and Representation in Congress

3

Member of Congress and for reporting directly to thatMember (CMF 2011).6 Chiefs of Staff and LegislativeDirectors are thus the individuals ultimately in chargeofreviewing the information an office receives fromordinary constituents, interest groups, and other sour-ces, and then incorporating that information into rec-ommendations for their Member. Thus, our surveytargeted the population that plays a crucial role in thepolicymaking process, connecting the preferences ofconstituents with Members of Congress. It may well bethe case that other staffers within a Congressionaloffice—like district directors—could more accuratelyjudge their constituents’ opinions. However, we viewChiefs of Staff and Legislative Directors as the criticalsample for our study’s research question because theyare central in shaping Members’ policy decisions. Dis-trict staff’s perceptions of public opinion mean little ifthey cannot be accurately conveyed through moresenior staff to shape Member decision-making.

The survey asked staffers a range of questions aboutpolicy preferences, including staffers’ own preferencesand their beliefs about their constituents’ attitudes. Italso asked staffers aboutwhere their policy informationcomes from, including their reliance on organizedinterest groups. In addition, the survey included severalexperiments. One list experiment measured staffers’willingness to change their beliefs on public policy as aresult of campaign contributions from an organizedinterest group. Another experiment gauged howstaffers would respond to communications from dif-ferent constituents, including business and advocacygroups.

To construct the senior Congressional staff pop-ulation,we used theLeadershipDirectories database toidentify Chiefs of Staff and Legislative Directors ineveryUSHouse andUSSenate office as of July 2016. Incases where staffers did not have those exact titles, weidentified other individuals who would serve as the toppolicy staffers in an office. We sent an initial emailinvitation to these staffers toparticipate inour surveyonAugust 18, 2016. Two follow-up emails were sent tostaffers on August 24 and August 31.7 Our final surveysample includes 101 respondents from 91 offices, for atotal response rate of 9.6%, in line with other studies ofCongressional staff.8

A number of Congressional offices have policiesagainst participating in surveys, and we received emailsindicating that many staff would not be able to

participate in our study as a result of those policies. Inother cases, however, staffers reported making anexception for our survey because of its confidential andacademicnature—two facts thatweheavily emphasizedin our correspondence. These policies would be con-cerning if certain types of Congressional offices weremore likely to implement them, or certain types ofstaffers were more likely to follow these policies morestrictly, causing our survey results to represent a biasedsample of staffers and offices. Reassuringly, however,our sample of respondents is quite close to the overallpopulation of senior legislative staffers and Congres-sional offices on a range of observable characteristics.This provides evidence that certain types of offices orstaffers did not systematically answer our survey.

Figure 1 compares our survey respondents to theoverall population of senior legislative staffers inCongress, contrasting differences inmeans for a varietyof characteristics with 95% confidence intervals.9 Fullbalance tables are provided in Appendix C. In general,our sample closely resembles the overall population.Still, there are some differences. The largest gapbetween our survey sample and the overall populationof staffers was in partisanship: slightly over half of oursample (54%) came from Democratic offices, while inCongress as a whole only 43% of senior staffers work inDemocratic offices. Nevertheless, we still have a suffi-cient numberofRepublican respondents todisaggregateour analyses by party. Within both Democratic andRepublican respondents, moreover, the respondingoffices’ ideological orientations are quite similar to theoverall distribution of Congressional ideology, asmeasured by standard first-dimension DW-NOMI-NATE ideal points (Carroll et al. 2015).

POLICY CONSIDERATIONS FORLEGISLATIVE STAFFERS

Before assessing the relationship between staffer per-ceptions of their constituents and constituent opinion,we first consider the information sources that staffersreport relying on. This allows us to compare, in staffers’own minds, the importance of constituent opinionagainst other potential considerations. We askedstaffers: “Think about the policy proposals you haveworked on during your time on the Hill. What shapedyour thinking onwhether yourMember should supportor oppose these policies? Indicate how important eachof the following considerations was in shaping youradvice to your Member on various policy proposals.”

Figure 2 reports this item’s results by party. Onaverage, staffers reported constituent communicationand attitudes were most important, and concerns aboutprimary opponents were least important. Over 80% ofstaffers view either constituent opinion or communi-cation as extremely or very important in shaping theiradvice to Members. These results fit well with pastpolling of Congressional staffers by the Congressional

6 According to the guide developed by the Congressional Manage-ment Foundation, the job description for Chief of Staff is “Top staffperson responsible for overall office functions; oversees staff andbudget; advises Member on political matters; responsible for hiring,promoting, and terminating staff; establishes office policies andprocedures”; the description for Legislative Director is “Establisheslegislative agenda; directs legislative staff; serves as resource personfor LAs [Legislative Assistants]; briefs Member on all legislativematters; reviews constituent mail.”7 See Appendix A for copies of recruitment materials.8 For instance, our response rate is similar to the response rates theCongressional Management Foundation obtained (~15%) in itsstudies ofCongressional offices (See: http://www.congressfoundation.org/publications/1048-managing-changes-in-budgets-and-benefits). 9 We clustered standard errors by Member office.

Alexander Hertel-Fernandez, Matto Mildenberger, and Leah C. Stokes

4

Management Foundation, which has found that over90% of surveyed staffers say that constituent contactwould be important in deciding an office’s stance on anissue.10 In fact, Congressional offices appear to priori-tize collecting and responding to constituent opinions somuch that the CMF has found that nearly half of sur-veyed staffers said that their office had shifted resourcesaway fromother activities towardmanaging constituentcommunications.11 This is understandable, given thatthe Internet makes it easier for constituents to com-municate with their elected officials’ offices. Of course,these estimates are based on staffer self-reports. Wemight be concerned that, even in the context of aconfidential survey, staffers downplay their reliance oncertain information sources that are viewed as politi-cally unpopular. To evaluate this potential bias, we alsoasked stafferswhat considerations other offices used forpersuasion during past policy debates. This questionpartially taps into revealed behavior, rather than simplystaffers’ own judgments. If another office was tryinghard to persuade a staffer and their Member about anissue, they would likely pick considerations they knewwould be taken more seriously by the target office. Weprovide full details on this question and results from thissurvey item in Appendix D. In general, there was astrong relationship between the types of considerationsand information sources described as personallyimportant by staffers and as frequently used by otheroffices for the purposes of persuasion.

In follow-up, semistructured interviews with some ofour survey respondents, we also probed staffers abouttheir efforts to process constituent communications andopinions.12 Our interviews indicated that to trackconstituent opinion offices generally log every phonecall, e-mail, letter, and fax they receive into a database.These databases can be used to sort pieces of corre-spondence into batches relating to particular issueareas, such as health care or trade, and to report thenumber of contacts the office has received on each issueas well as the stances conveyed in the messages. Whenthe American Health Care Act (the GOP bill to repealthe Affordable Care Act) was introduced in the Housein 2017, for instance, staffers used this information totrack howmany contacts their office received in favor ofthe bill, and how many opposed it.

While this process is remarkably consistent acrossoffices, interviewees varied in the extent to which theyreported that their office engages in additionalmethodsof assessing constituents’views.Theothermethods theycited include face-to-face meetings with constituents intheir offices, polling, monitoring constituents’ com-ments on social media, and even door-to-door can-vassing by Members themselves. Several intervieweesalsomentioned townhalls andother public events in thedistrict or state, such as civic association meetings, as auseful means of keeping abreast of the issues that aremost salient to constituents.

Staffers in our interviews consistently indicatedcorrespondence tallies on a given bill (e.g., how manypeople asked a Member to vote for or against it) areoften incorporated into recommendations aboutwhether to vote for the bill. Similarly, intervieweesreported that the information about which issues theyare hearing about from their constituents plays a role intheir office’s decisions about which issues to focus on.For instance, at weekly meetings with the Memberwhere the office decides its priorities, staffers reviewwhich issues they have been hearing about most fromconstituents during the past week. We will evaluate theextent to which staffers actually perceive their con-stituents’ preferences and take correspondence fromconstituents as seriously as they indicate on this ques-tion in the sections that follow.

Returning toFigure 2, therewere large differences byparty in the considerations staffers reported were rel-evant for their policy advice, aside from constituentopinion. The largest divide was on information fromunions: nearly half (49%) of Democratic staff reportedthat union informationwas extremely or very importantto their deliberations,while only 7%ofRepublican staffsaid the same. Another striking difference was theimportance of information from party leaders (Curry2015). Over half (55%) of Democratic staffers reportedthat they found information from their party’s leader-ship to be extremely or very important to them, while

FIGURE 1. Balance Between SurveyRespondents and Overall Population of Staffers(with 95% Confidence Intervals)

10 http://www.congressfoundation.org/storage/documents/CMF_Pubs/cwc-perceptions-of-citizen-advocacy.pdf.11 http://www.congressfoundation.org/storage/documents/CMF_Pubs/cwc-mail-operations.pdf.

12 Our research team conducted 30-minute interviews withrespondents from the survey from July through September 2017.These interviews were conducted with 11 Democrats and sevenRepublicans. Thirteen interviewparticipants wereChiefs of State andfive were Legislative Directors.

Legislative Staff and Representation in Congress

5

only one in five Republican staffers reported the same(21%). This divide may reflect the ongoing struggleswithin the Congressional Republican caucus as GOPleaders like Speaker Paul Ryan and Senate MajorityLeader Mitch McConnell attempt to bring togethermore traditional, business-friendly Members withnewer and more ideological Freedom Caucus and TeaParty members (Mann and Ornstein 2016).

LEGISLATIVE STAFFERS’ PERCEPTIONS OFTHEIR CONSTITUENTS

Over three-quarters of both Democratic and Repub-lican staffers reported in our survey that their topconsideration when thinking about legislation was theirconstituents’ opinion. But do staffers know what theirconstituents want? To answer this question, we com-pare staffers’ estimates of public opinion in their districtor state to the actual opinion in that district or state.Specifically, we ask respondents “Now, consider onlythe people living in your Member of Congress’s con-stituency. To the best of your knowledge, what per-centage of the people living in your Member’sconstituency would agree with the following policystatements? The slider below goes from 0% (no one)to 100% (everyone). Just give your best guess.”Respondentsmoved a slider bar to answer the question.

As part of our follow-up interviews with surveyrespondents, we validated our interpretation of thisquestion as representing estimates of district-level

public opinion. In particular, these interviews fore-closed the unlikely interpretation of our question asquerying the narrow preferences of their Member’spolitical supporters. All our interviewees unanimouslyinterpreted our question wording as referring to theresidents of their Member’s district or state. No stafferinterpreted it as referring to political supporters.13

We selected the five policy domains included on thesurvey for several reasons: public opinion data avail-ability, issue salience, partisan polarization, and interestgroup involvement. With the exception of the climatechange item, the items came from the CooperativeCongressional Election Study (CCES), which offerssufficiently large sample sizes to estimate downscaledpublic opinion for states and Congressional districts.The CCES uses respondents sampled from YouGov, awell-respected online polling firm. The sample aims tobe representative of the national adult population and isvery large with around 50,000 respondents duringelection years. We use data from the 2016 CCES toestimate district- and state-specific attitudes usingmultilevel regression and post-stratification (MRP),detailed in Appendix F. Our climate question uses datapreviously down-scaled by Howe et al. (2015). In theanalyses that follow, we compare Senate staffer

FIGURE2. ConsiderationsStaffersReportedasExtremelyorVery Important inShapingAdvice toTheirMembers, by Staffer Party

13 Moreover, when asked whether constituents included peopleresiding outside the district or state, some staffersmentioned that theyreceived correspondence from people living outside the member’sjurisdiction, but that they were able to tell whether the contact camefrom a resident of the district or state.

Alexander Hertel-Fernandez, Matto Mildenberger, and Leah C. Stokes

6

perceptions with state-level opinion estimates, andHouse staffers with district-level opinion estimates. Forall five policy areas, we used essentially the samequestion wording in our survey as in these othersurveys.14

All five policy domains are highly salient, andtherefore are issues on which staffers have received atleast someconstituentand interest groupcontact.This isespecially true because Congress considered bills thatwould address all five proposals within the prior year tothe survey.15 The fact that Congress had recently votedon measures related to these policies means that toplegislative staffers ought to have considered, relativelyrecently, the positions of their constituents on theseissues. The five issues also exhibit varying degrees ofpartisan polarization. In two cases—ACA repeal andthe minimum wage—the issue is highly polarizedbetween the two parties. There is essentially no Con-gressional Democratic support in favor of repeal andvirtually all Congressional Republicans were suppor-tive of ending the health reform law, at least in principlebefore the Trump administration; while the reverse istrue for boosting the minimum wage. By contrast, bothparties have tended to be more divided internallytoward climate change and gun control: even thoughDemocrats have tended to be much more favorabletoward measures to address climate change and guncontrol than Republicans, there are more divisions onthese issues within the parties. Some conservativeDemocrats, for instance, have stymied Congressionalclimate proposals and new gun control measures. Somemore moderate Republicans have occasionally sup-ported legislation in these domains as well. Andinfrastructure spending is generally viewed as a bipar-tisan issue historically supported by both parties as anecessary investment in the economy. Indeed, aninfrastructure spending bill was one of the few bipar-tisan accomplishments of the otherwise gridlocked114th Congress.

The five issues also implicate different constellationsof interest group involvement, which we anticipatemayhelp explain Congressional representation. On climatechange, fossil fuel and other extractive industriesintensely oppose tighter regulations. These businesseshave invested a considerable sumofmoney in campaigncontributions, lobbying, and other political activities toprevent government action to curb greenhouse gasemissions (Mildenberger 2015). We might characterize

this policy as engendering the opposition of a con-centrated set of corporate interests. Tighter gun regu-lations represent a similar issue where there is intensecorporate opposition from a concentrated set of inter-ests, in this case, the firearm and ammunition industryand its advocacy groups, especially the National RifleAssociation (Cook and Goss 2014). Raising the mini-mum wage is also an issue that business interests haveopposed; but in this case, opposition is less concentratedin particular industries andmore widespread across thebusiness community (Hacker andPierson 2010).Healthreform, on the other hand, features a mixed set ofcorporate interests. Although health insurers and somemedical providers strongly opposed health reformlegislation, including the ACA (Jacobs and Skocpol2015), these sectors face adifferent set of incentives nowthat the ACA is law. Repealing some or all of the ACAwould undermine these companies’ current businessmodel, which nowdepends on enrollingAmericans intonewregulatedmarketplaces andanexpandedMedicaidsystem. At the same time, these industries would sup-port rolling back many individual provisions within theACA. Lastly, infrastructure spending is an issue sup-ported bymany segments of the business community asa useful investment in theAmerican economy. The veryconservative US Chamber of Commerce, typicallyopposed to most government intervention, backed theinfrastructure spending bill considered by the 114thCongress.16

Figure 3 compares the relationship between Con-gressional staffers’ perceptions of their constituents’opinions and their constituents’ actual opinions in eachpolicy area.17 If stafferswere able to accurately perceivepublic opinion, the data should fall on the dashed 45degree line; when staffers are underestimating support,the dots fall to the left-hand side of the line, when theyare overestimating support, they fall to the right. Thelocally weighted (lowess) regression line shows thatstaffers estimate public opinion across the full range,from 0% to 100% support. By contrast, public supportin each domain falls within a narrower range acrossdistricts: from 50% to 82% for carbon regulation; 39%to 69% for repealing theACA; 73% to 95% for gun saleregulations; 72% to 86% for infrastructure spending;and 57%to90%for boosting theminimumwage.Whilethere are modestly positive slopes across all five issueswhen pooling across Democratic and Republican

14 For more information on the CCES, see: https://cces.gov.harvard.edu/.AppendicesB andE summarize thewording of our questions, aswell as the sources for the survey dataweused to estimate district- andstate-specific attitudes using multilevel regression and post-stratification (MRP). We detail our MRP models in Appendix F aswell.15 InDecember 2015, Congress voted on ameasure to curb theEPA’sability to regulate CO2 emissions. Congress voted on ACA repeal asrecently as February, 2016, when the House voted to override Pres-ident Obama’s veto of repeal legislation H.R. 3762. In June, 2016,Congress voted on background check measures. In May 2015,Democrats released a $12 an hour minimum wage proposal. And inDecember 2015, Congress voted on a major infrastructure spendingbill.

16 See, e.g.: https://www.uschamber.com/blog/its-time-raise-federal-gas-tax.17 Wemightbe concerned that these results are in somewayanartifactof sampling uncertainty in ourMRPestimates. Fortunately, we have astrategy to directly evaluate and reject this possibility. TheHoweet al.(2015) data we rely on for our climate-based estimates involvedexternal validation of the MRP estimates against independent local-level polls to estimate the sampling uncertainty associated with themodel. In that research, Howe et al. report that congressional districtestimates are within67 percentage points of their external validationdataset. Taking this estimate, we find that 66 staffers (72% ofrespondents that answered this question) have a mismatch that isgreater than this uncertainty estimate. This should give us high con-fidence that staffer misperceptions are not a simple function of MRPsampling uncertainty.

Legislative Staff and Representation in Congress

7

staffers, these correlations generally disappear once weconsider within-party staffer misperceptions.18 Overall,then, in none of the five areas are staffers estimatingtheir constituents’ preferences with any degree of rel-ative or absolute accuracy.19 Staffer perceptions are farmore extreme than the public’s actual policy prefer-ences. Still, the strength of the relationship betweenstaffer estimates and constituent preferences varied

across the five domains: the correlation between stafferestimates and constituent preferences was strongest forattitudes about raising the minimum wage, followed byCO2 limits, repealing the ACA, gun sale checks, andthen infrastructure spending.20

If staffers donot haveaccurateperceptions of publics intheir districts or states, do theseperceptions instead reflectthe distribution of public attitudes among co-partisans?Wecanevaluate thispossibility in theclimate change issuearea by leveraging data from a partisan MRP model ofclimateandenergyattitudes (Mildenbergeret al. 2017). InFigure 4, we compare staffers’ perceptions of their con-stituents’ preferences for carbon regulation with thefraction of the Democratic or Republican publics intheir elected official’s district supporting this position.We again find little relationship. Democratic and

FIGURE 3. Comparing Staffer Perceptions of Constituent Preferences to True District-LevelPreferences, by Issue Area

18 Considering the 10 issue-party relationships between publicopinionand staffers’ guesses, only two cases are statistically significantat conventional levels: Democratic staffers on theminimumwage andon repealing the ACA. The remaining eight slopes are all far fromstatistical significance.19 Whilewedonothave theability toestimate thepreferencesof likelyvoters—as opposed to all constituents—at the district level, we cancompare differences between the true national preferences amonglikely voters and all US citizens. Subsetting CCES data to individualswho report that they will definitely vote in a given election, we findnegligibledifferencesbetweentheaveragepreferencesof likelyvotersand all citizens. This suggests that the misperception gap is not afunction of staffers simply discounting the preferences of unlikelyvoters from their estimates. Further MRP analysis that distinguishedbetween validated voters and nonvoters would be fruitful to test thispossibility in even greater detail.

20 InAppendixQ,wealso performa robustness checkon these resultsby redoing our analysis after dropping staffers who estimate that thepublic in their district is over 90% or less than 10% in support of anyindividualpolicy.Wethus showthat the surveyresults are robustwhenexcluding staffers who offered potentially nonrealistic answers tothese questions.

Alexander Hertel-Fernandez, Matto Mildenberger, and Leah C. Stokes

8

Republican staffers’ estimates of the general publics’preferences do not seem driven by the distribution oftheir co-partisan publics’ beliefs. If staffers’ perceptionsof public opinion were structured by primary or generalelection partisan voters, we would have expected to seea stronger relationship in this figure.

Our data also allow us to evaluatewhether staffers aresystematically over- or underestimating public attitudes.Previous researchfinds thatboth liberalandconservativestate politicians systematically over-estimate their con-stituents’ conservatism (Broockman and Skovron 2018).We similarly find a conservative bias for four of our fiveissues. The vast majority of surveyed staffers (91%)underestimated their constituents’ support for back-ground checks on gun sales. There was a similar, thoughless pronounced, bias toward staffers underestimatingsupport for regulating carbon dioxide emissions—78%of staffers underestimated constituent support for CO2limits in their districts and states. Therewas less of a clearpattern for stafferswhenestimating support for repealingthe ACA: 65% underestimated and 35% overestimatedmass support. However, there were clear partisan divi-sions in staffermisperceptions.EveryDemocratic stafferunderestimated support for repeal and nearly everyRepublicanstafferover-estimatedsupport for repeal.Onboth the minimum wage and infrastructure spending,nearly all GOP staffers underestimated their con-stituents’ support. Democratic staffers tended tounderestimate support, on average, though the biaswas less pronounced than with their Republican coun-terparts. Thus, overall, we find a conservative bias instaffers’ estimations.

EXPLAINING LEGISLATIVE STAFFERS’GAPS IN REPRESENTATION

What could explain these strikingly large mismatchesbetween staffer perceptions and constituent opinion?We examine four potential correlates for the repre-sentation gap: elections; staffers’ own preferences;staffer experience; and interest group contact. Wesummarize these hypotheses in Table 1.21

Standard models of representation predict thatelections should drive responsiveness for politicianswho want to win and hold office (Arnold 1990; Gilens2012; Mayhew 2004). Therefore, we might expect thatstafferswhoseMemberswere in tighter races at the timeof the survey would be more likely to correctly assessconstituent opinions because they want their Membersto win reelection to retain their jobs. We assessed thishypothesis using a question on the survey that askedstaffers to predict their Members’ reelection margin, iftheir Member was up for reelection in 2016. Wereceived 75 valid responses, ranging from a predicted 2-to 100-percentage point victory, with amean of 35 and amedian of 23-percentage points. Notably, no stafferpredicted a loss.

Figure 5 plots the relationship between the staffer-constituent opinion mismatch and staffers’ electionpredictions. Looking across all policy domains, we seelittle consistent relationship between estimated racecompetitiveness and the accuracy of staffers’ percep-tions. Staffers for Members in tighter races were gen-erally no more likely to get their constituents’preferences right compared to staffers working forMembers in relatively uncontested races. The strongestexception was for Republican staffers on gun policy.Here, we see the expected relationship: GOP staffersfrom safer districts and states were more likely tounderestimate their constituents’ support for back-ground checks. Yet, even for staffers in highly com-petitive races there was a large mismatch: staffers whothought that their bosses would only win by two per-centage points were estimated to be off of their con-stituents’ preferences by about 43-percentage points.Electoral competitiveness thus does not seem to offermuch leverage in explaining the staffer-constituentmismatch (see Appendix H for evidence that actualrace competitiveness does not predict representation,either).

Next, we examined the role of staffers’ personalopinions (Figure 6). For this hypothesis, we wereinterested in understanding whether egocentric biascould explain mismatches. Egocentric bias is a con-sistent finding in psychology that suggests individualsuse their own beliefs as a heuristic for estimating thebeliefs and opinions of others (Epley et al. 2004;Nickerson 1999). We found the clearest case for ego-centric bias in health policy. On average, staffers whosupported the ACA repeal overestimated constituent

FIGURE 4. Staffer Perceptions of ConstituentPreferences Against Distribution of Co-PartisanConstituent Preferences

21 Wediscuss these explanations separately below, reflecting our viewthat we cannotmake strong causal claimswith our observational data.Nevertheless, we also provide multivariate regressions adjudicatingthe importance of each factor in Appendix L.

Legislative Staff and Representation in Congress

9

support for repeal while staffers who opposed repealunderestimated constituent support for repeal. Bycontrast, all staffers underestimated their constituents’support for climate regulations, infrastructure spend-ing, boosting the minimum wage, and gun backgroundchecks, although this dynamic was significantly mod-erated by staffers’ own beliefs. This finding indicates apotential role for staffers’ownperceptionsandattitudesin accounting for the representation gap betweenCongressional aides and the public. Our evidencesuggests that, despite any political incentives that mayexist to reward unbiased estimation of constituentbeliefs, staffers do not transcend common egocentricbias. Divergence in staffer misperceptions among indi-viduals who agree and those who disagree with eachpolicy statement are statistically significant in all fivepolicydomains; substantively, thedifferences range from17-percentage points (infrastructure) to 36-percentagepoints (gun regulations).

The third factor we considered was staffer workexperience. There was broad variation in the length oftime that senior staffers in our survey had worked fortheir particular Member, and for Congress in general.Onaverage, staffers reportedworking for theirMemberfor aboutfiveyears, and inCongress for nine. It couldbethe case that staffers with more work experience aremore knowledgeable about their constituents’ prefer-ences, either because they have had more time to learnthe public’s attitudes, or because they gained skillsin estimating constituent opinion. To assess this

explanation, we compare staffer tenure against staffer-constituent preference mismatches. We find no rela-tionship. In none of the policy domains we examinedwas staffer experience, either in aMember’s office or inCongress in general, related to the accuracy with whichstaffers could discern their constituents’ attitudes.Staffers who have served in Congress for longer are notmore likely to get their constituents’ preferences rightcompared to newer staffers (see Appendix I for fullresults).

Finally, we examined whether interest group contactcould account for the representationgap. In the absenceof strong signals from ordinary citizens, staffers mightrely on interest groups that claim to represent con-stituentswithin a staffer’s state or district. Staffersmighteven simply substitute interest group preferences forthose of their constituents. Building on the interestgroup literature (Schlozman et al. 2012; Walker 1991),we distinguish betweenmass-based interest groups andcorporate-based interest groups. Business-orientedgroups represent the interests of for-profit companies.Mass-based groups represent individuals from themasspublic. We expect that greater relative staffer contactwith mass-based groups will help staffers assemble amore accurate perception of their constituents’ pref-erences given that mass-based groups have closerinteractions with the public. By contrast, we hypothe-size that greater relative staffer contact with business-based groups will result in less accurate perceptions ofpublic opinion because business-based groups are less

TABLE 1. Correlates of the Staffer Representation Gap

Factor Hypothesis Measure Result preview

Electoralcompetitiveness

Staffers in offices facingmore competitive raceswill more accurately perceive constituentpreferences

Staffers’ predictions of2016 race margin;actual margin of victory(Appendix H)

No evidence supportingthis hypothesis (Figure 5)

Staffer preferences Staffers will offer more accurate estimates of theirconstituents’ preferencesif they share theirconstituents’ preferences

Staffers’ personal supportfor policies

Staffers’ own beliefs arecorrelated withmisperceptions (Figure 6)

Staffer experience Staffers with greater workexperience in Congresswill be more likely toaccurately perceiveconstituent preferences

Staffers’ years of service inMember office andCongress

No evidence supportingthis hypothesis (Appendix I)

Interest groups Staffers with greater contactwith corporate groupswill less accurately perceiveconstituent preferences;staffers with greater contactwith mass-based groups willmore accurately perceiveconstituent preferences

Self-reported reliance oninterest groups (includingmass-based advocacygroups, businessassociations, and thinktanks); campaigncontributions from industrygroups to staffers’ Membersin the last electoral cycle;local labor union density

Interest group contact andcampaign contributionsare both correlated withmisperceptions (Figure 7and Appendices J and K)

Alexander Hertel-Fernandez, Matto Mildenberger, and Leah C. Stokes

10

likely to represent the preferences of the generalpublic—and may be opposed to them (Gilens and Page2014; Hacker and Pierson 2010). Consistent with themore general rightward movement of the organizedbusiness community in recent decades (Hacker andPierson 2016), most business interests in our survey arealso aligned with ideologically conservative positions,while our mass-based groups are supportive of liberalstands. Regardless of the cause of this division, thiscomplicates efforts to separate the ideological ori-entation of groups from the business versus mass groupdimension in our survey. In the discussion that follows,we therefore refer to groups by both characteristics.

We measure interest group contact in three ways:staffers’ self-reported reliance on interest groups rele-vant to the policy issues we study, data on campaigncontributions from themajor corporate sectors related tothese policy domains, and an objective measure of onemass group’s strength (labor unions). The self-reportedmeasurecomes fromasurvey iteminwhich staffers could

rate their reliance on various interest groups on a onethrough six scale, where six indicated the greatestimportance.22 We average and sum standardized ver-sions of these measures across multiple interest groupsinvolved in each of the five policy domains.23

Our second measure involves the share of a Mem-ber’s campaign contributions in their last election cyclethat came from businesses in specific industries thatrelated to our policy domains: health providers and

FIGURE 5. Staffer Misperceptions of Constituent Preferences and EstimatedMargin of Victory in NextElection, by Issue Area

22 The survey item was: “We are interested in knowing which groupsare most important to you when considering legislation. For eachgroup, please indicate how important the group’s positions, resources,and informationhavebeen to youwhendeliberatingover legislation.”23 We coded Americans for Responsible Solutions, Everytown, theSierra Club, League of Conservative Voters, and the AFL-CIO asliberal, mass-based interest groups. We coded the National RifleAssociation, the American Petroleum Institute, the Edison Institute,the US Chamber of Commerce, the National Association of Manu-facturers, the Club for Growth, and Americans for Prosperity asconservative or corporate-based interest groups.

Legislative Staff and Representation in Congress

11

insurers for the ACA; gun and ammunition manu-facturers for background checks; construction andcontractors for infrastructure; business associations forthe minimumwage; and extractive industries, includingconventional energy power plants, oil, gas, and coalextraction, for carbon dioxide emission limits.24 Fornow, we interpret the correlation between campaigncontributions and staffer perceptions not as signifyingtheweight of the contributions themselves, but rather asthe strength of the relationship between a Member’soffice and a particular set of corporate interests. Agreater reliance on contributions from an industrialsector signifies a stronger relationship between anofficeand businesses in that sector.

Thefinalmeasure involves thedensityof laborunionsin a staffer’s district or state, which we interpret as ameasureofunion strength.Werelyonestimatesof laborunionmembership byHouse district reported inBecheret al. (2018) and state union density from the Bureau ofLabor Statistics (seeAppendix J formore information).Unions represent an especially important mass-basedgroup and check on corporate influence. As KaySchlozman, Sidney Verba, and Henry Brady explain inTheUnheavenly Chorus, unions are essentially the onlyway that nonprofessional, nonmanagerial workersreceive representation in the organized interest com-munity (seeespecially chapter 14; Schlozmanetal. 2012).In a similar vein,MartinGilens finds in his analysis of the

congruence between public opinion and policy changethat“unionsemergedas the interestgroupswith themostconsistent and widespread tendency to share the pref-erences of low- and middle-income Americans (Gilens2012, 161).” And in their analysis of the contribution ofeconomic policy to rising top-end inequality, JacobHacker and Paul Pierson argue that unions are the“vigorous champion” of “pocketbook issues” for“middle- and working-class Americans”—not just theirimmediate members (Hacker and Pierson 2010, 143). Inlight of this research, we feel comfortable characterizingunions as an especially important mass-based grouprepresenting public preferences.

The top left panel of Figure 7 summarizes, acrosspolicy domains, the contact that staffers reported withboth liberal/mass groups and conservative/corporategroups. The horizontal axis is a standardized index ofthe frequencywithwhich staffers reliedonmass groupsminus corporate groups. Greater values indicate agreater relative reliance on liberal/mass groups overconservative/corporate groups. The strong negativerelationship indicates that staffers who reportedrelying on liberal/mass over conservative/corporateinterest groups—at the right end of the x-axis—tendedto more accurately perceive their constituents’ pref-erences. By contrast, staffers more reliant on con-servative/corporate groups over liberal/mass-basedgroups—at the left end of the x-axis—weremore likelyto get their constituents’ opinions wrong. The top rightpanel reports the campaign contribution measure. Weobserve a similar relationship as in the left panel,indicating that a greater reliance on corporate

FIGURE 6. Staffer Misperceptions of Constituent Preferences and Staffer Personal Opinions, by IssueArea. Lines Indicate 95% Confidence Intervals

24 Data from theNational InstituteonMoney inStatePolitics.Wefindsimilar results using logged campaign contributions in dollars as well.

Alexander Hertel-Fernandez, Matto Mildenberger, and Leah C. Stokes

12

contributions is correlated with larger constituentmismatches.25

The bottom left panel of Figure 7 shows consistentresults with the other two interest group panels: greatermass group-basedpressure onaMember’s office (in thiscase measured with local union strength) is related tomore accurate perceptions of constituent preferences.Staffers representing districts and states with strongerunions were more accurate in their predictions ofconstituent preferences along all five dimensions.

In summary, we find little support for a relationshipbetween electoral pressures or staff tenure and staffers’accuracy in estimating public opinion. We do find arelationship between accurate predictions and interestgroup interactions with and donations to offices, as wellas staffers’ own opinions. Greater contact with liberal,mass-based groups is related to more accurate per-ceptions of constituent opinion while greater contactwith conservative, corporate-based groups is related toless accurate perceptions. Staffer contact with theseorganizations may skew their perceptions of the publicbecause the groups aremore ideologically conservativethan the public along the range of issues we study. Thatmechanismwouldbe very consistentwith the analysis inBroockman and Skovron (2018), which shows that statepoliticians’ over-estimation of their constituents’ con-servatism can be attributed in part to conservatives’greater likelihood of contacting elected officials.

Still, these correlations are at best suggestive. Andimportantly, these results cannot explain why it is thatcontact with conservative and corporate groups skewsstafferperceptionsof thepublic.Tomore rigorously testwhether interest group interactions can shape stafferperceptions andpositions,we turn to evidence from twoexperiments embedded in the survey. These experi-ments provide more credible causal evidence about therole of interest group and donor contact in shapingstaffer perceptions and preferences.

EXPERIMENTAL EVIDENCE SUPPORTINGINTEREST GROUPS’ INFLUENCE ON THEREPRESENTATION GAP

We use two experiments embedded within the staffersurvey to test whether contact with organized interestgroupschanges senior legislative staffers’perceptionsoftheir constituents’ preferences. In a list experiment, weuncover the proportion of top Congressional aides

revealing they changed their opinion on a policy issueafter receiving a campaign contribution from anorganized interest group. We used a separate experi-ment to examine how communications from ordinaryconstituents, citizens’ associations, or employees of alarge business influence staffers’ opinions aboutpending policy debates and their perceptions of con-stituent attitudes.

Contribution List Experiment

Recent experimental researchbyKalla andBroockman(2015) gives reason to think that interest group electoralcontributions increase the likelihood that high-levelstaffers meet with interest groups. That study andrelated literature, however, is limited in its ability togauge whether contributions also shape staffers’ policyopinions. While contributions facilitate access, thosemeetings may not necessarily result in staffers changingtheir minds about legislation. For instance, if interestgroups generally make campaign contributions to allieswho already agreewith them, then contributions shouldnot change staffers’ minds (Hall and Deardorff 2006;Hall and Wayman 1990). Examining whether stafferschange their minds on legislation after their Repre-sentative or Senator receives campaign contributionsfrom an interest group is no easy task, however. It is amechanism that is difficult to observe without askingstaffers directly. Yet staffers likely face strong incen-tives against responding truthfully to such a question ona survey. Accordingly, we use a list experiment to elicitmore accurate responses from senior legislative staffersabout whether electoral contributions from interestgroups shaped their policy preferences. If interestgroups do shape staffers’ policy preferences, this couldhelpexplainwhy corporate contributions are correlatedwith a greater staffer-constituent opinion mismatch.

In a list experiment, respondents are shown a list ofseveral statements, and then indicate how many itemsthey agree with. Not all respondents see the same list,however. Half of respondents are shown a short listwithout the sensitive item the researcher is interested instudying,while theotherhalf sees a listwith the sensitiveitem included. The proportion of respondents agreeingwith the sensitive statement can be calculated by sub-tracting the average number of items selected amongrespondents who saw the full list from the averagenumber of items selected by respondents who saw theshort list. List experiments have been successfully usedin many other contexts to study similar taboo or sen-sitive behaviors (Glynn 2013).

In our list experiment, staffers read the followingprompt: “Below, you will find a list of [4/5] statements.Please tell us HOWMANYof them you agree with.Wedon’twant toknowwhichonesyouagreewith, justHOWMANY. Over the last year, I’ve developed a new per-spective about a policy under consideration in Congressafter speaking with…”Half the staffers were assigned tothe control group, where they saw four items: “a rep-resentative from a business”; “a legislative staffer fromthe opposing party”; “a legislative staffer from my ownparty”; and “a representative from a union.” The other

25 We standardized this variable, as with the mass versus corporategroup contact variable, to be comparable across policy domains.Unfortunately we cannot exactly match the “mass group minus cor-porate group” measure we used for the self-reported staffer contactpanel becausemany of the organizations we included in our survey donot make direct campaign contributions to Congressional candidates.We can replicate the exact measure for the CO2 limits question,however,andwefindaverystrongrelationshipbetweenthebalanceofcontributions from environmental, mass-based groups and extractiveindustry, corporate interests and staffer misperceptions. We alsocalculate business and labor contributions for each staffer’s office andagain find similar results looking across all issue domains.We includeboth results in Appendix K.

Legislative Staff and Representation in Congress

13

half of staffers were assigned to the treated group,wherethey saw those same four items plus an additional fifthitem: “a group that provided campaign contributions tomyMember.”This is the sensitive itemwhoseprevalencewe were attempting to estimate.26

Staffers in the control condition agreed with anaverage of 2.84 items (SE 5 0.18), while staffers in thetreatment condition agreed with an average of 3.30items (SE5 0.19). The difference inmeans between theconditions is 0.454 (SE5 0.27). This indicates that 45%of staffers agreed they haddeveloped a newperspectiveabout a policy after speakingwith a group that providedcampaign contributions to their Member.27 This resultsuggests that a potentially large share of senior Con-gressional staffers have had their thinking about policyshaped by interest groups that provided their Member

campaign contributions. Some of these staffers mightdevelop new policy perspectives from campaign con-tributors because they share broad policy objectives.Others may engage these contributors to secure con-tinued campaign financing. Critically, however, thisexperiment offers evidence that campaign contributorsnot only shape access to Congressional offices in thefirst place (Hall and Deardorff 2006; Hall andWayman1990;Kalla andBroockman 2015), but that contact withthese contributors shapes staffer policy perspectives. Inthis sense, the list experiment suggests a potentialmechanism for the correlations between corporateinteractions and staffer-constituent preference mis-matches we uncovered in the observational surveyevidence above.

Communications Survey Experiment

Our second survey experiment focused on anotherstrategy that interest groups use to shapeCongressionalpolicymaking: having their members write to Congresson pending policy proposals. These tactics are well-documented in past work on interest group behavior(e.g., Kollman 1998; Schlozman et al. 2012; Walker2014). In particular, large companies increasingly use

FIGURE 7. Staffer Misperceptions of Constituent Preferences and Interest Group Contact (Self-Reported Reliance on Mass Versus Corporate Groups in Top Left Plot; Corporate CampaignContributions to Staffer’s Member in Top Right Plot; Labor Union Membership in Bottom Left Plot)

26 SeeAppendixO for a discussion of our design andwording choices.27 Two-tailed test: p 5 0.098, one-tailed test: p 5 0.049; clustered byoffice. That we find a significant effect at all given such a small samplesize is striking; while our confidence intervals are necessarily widegiven this small sample, we also note that, ultimately, it is the directionof the effect that matters most for our argument that interactions withcampaign contributors may skew staffer perceptions of publicpreferences.

Alexander Hertel-Fernandez, Matto Mildenberger, and Leah C. Stokes

14

employee correspondence to shape legislative behavioron bills relevant to their firms and industries (Hertel-Fernandez 2018; Walker 2014).28 Legislative staffers’role in mediating communications to Members is lesswell understood, however. This oversight is importantbecause we know staffers play a central role in theinformation chain connecting outside communicationsto Members’ actions.

Existing literature on Congressional constituentcorrespondence is similarly unclear on whether thesecommunications provide anopportunity for all intereststo be represented in the policymaking process, orwhether they advantage certain voices over others. In aseries of correspondence experiments, Butler (2014)shows that municipal politicians discount constituentcommunications from individuals who appear to be oflow socioeconomic status because elected officialsperceive those individuals to be less knowledgeableabout the issues described in those communications.This research suggests important ways that politiciansmight discount public attitudes of certain groups. Weaim to extend this research, examining how politicalofficials evaluate messages from individual citizensalongside interest groups, and especially business. Weare thus interested in knowing whether interest group-initiated communications carry especially importantweight in the minds of Congressional staff comparedwith individuals’ communications. Can interest groups,and particularly businesses, use Congressional com-munications as a way of disproportionately represent-ing their interests in the policymaking process, thusskewing staffer perceptions of constituents?

To evaluate these questions, we designed an ex-periment wherein senior Congressional staffers wereinstructed to imagine that they had received constituentcommunicationsonapendingpolicy issue.Stafferswererandomly assigned to receive different descriptions ofthe constituent communications, with the variationsindicated in brackets as follows: “Imagine your office isconsidering a bill that is under debate inCongress.Youroffice receives [2; 20; 200] letters from constituents[supporting; opposing] this bill. The letters have very[similar; different] wording to one another. The letterwriters identify themselves as [employees of a largecompany based in your constituency; constituents;members of a nonprofit citizens group].”

We then asked staffers three questions about theeffect of those letters on their deliberations over thelegislation in question: “How likely are you to mentionthese letters to yourMember?”;“Howsignificantwouldthese letters be in your advice to your Member abouttheir position on the bill?”; and “How representative doyou think these letters are of your constituents’ opin-ions?” Responses to the three questions were all

provided on a one through four scale (“very”, “some-what”, “not very”, and “not at all”), with higher valuesindicating greater agreementwith the question. Each ofthese three questions taps into a potential mechanismfor how interest group-initiated communications withCongress could change how staffers—and ultimately,Members of Congress—perceive their constituents’opinions and use those perceptions to inform policydecisions. By randomly assigning staffers to differenthypothetical letters, we can estimate the effect of astaffer receiving messages that were coordinated inletter-writing drives from large businesses or citizenadvocacy groups, or simply written by individual citi-zens.29 As with any survey experiment, the treatmentand related questions are stripped of much of the richpolitical context that surrounds the actual legislativeprocess. Yet the virtue of this stripped-down experi-ment is that any differences we observe in howstaffers report evaluating these constituent communi-cations can be attributed to variation in the treatmentconditions—and not other characteristics of the staffersthemselves. Given our interest in understanding theeffect of organized interest groups on staffers’ per-ceptions of constituent opinion, we focus our attentiononevaluating theeffectsof these treatments, rather thantheother conditions in the experiment. It is important tonote that given three conditions, there are small samplesizes (N 5 17–36) in each treatment group (moreover,we did not pre-register our hypotheses).

With these caveats inmind, staffers reported that theywould be more likely to mention employee-authoredletters to their bosses, but the differences with theconstituent or citizens’ group conditions were not sig-nificantat this sample size. Stafferswerealsomore likelyto indicate that employee letters would shape theirthinking and recommendations about legislation, butthe difference between employee letters and lettersfrom ordinary constituents was not significant. Whencomparing employee letters against both citizens groupletters and ordinary constituents taken together, whichimproves the sample size, there is a larger effect (0.32units on the one to four scale, p5 0.089; 16-percentagepoints on the “very” or “somewhat” important scale,p 5 0.17).

The picture is more striking when looking at the finalquestion we posed to staffers: how representative theybelieved the letters were of their constituents’ opinionsas a whole. Employee messages are substantially morerelevant as staffers form perceptions of their con-stituents’opinions than eithermessages from individualcitizens or members of nonprofit advocacy groups. Afull 62%of staffers who sawmessages purportedly fromemployees of the same large business reported that themessages were “very” or “somewhat” representative of

28 See for instance the account provided in Hertel-Fernandez (2018):“A lobbyist for a telecommunications company reported … thatrecruitingworkers towrite to amember ‘creates a heightened sense ofimportance of an issue’ and permits their lobbying team to bring thosecontacts up in one-on-one meetings with the member and his or herstaff. The lobbyists might say, ‘We have 3,500 workers in your districtand this is an important issue for them’ (pages 50–1).”

29 We recognize the challenges with ensuring that staffers wereresponding under conditions of “information equivalence” (Dafoeet al. 2017). In response, we note that we have varied a number ofimportant background conditions about the bill and correspondence.In addition, we do not think any of these conditions would present thesort of extreme “edge cases” that Dafoe et al. are most concernedabout in their paper.

Legislative Staff and Representation in Congress

15

their constituents’ opinions, compared to just 34% ofstaffers who saw letters from ordinary constituents and29%of stafferswho saw letters fromnonprofit advocacygroup members. More dramatically, over a third ofstafferswith letters fromnonprofit advocacy groups andaround 10% of staffers with letters from individualconstituents reported that those letters were “not at allrepresentative” of their constituents’ real opinions. Thedifference between the employee letter condition andthe ordinary constituent condition was quite large,representing 0.39 units on the one to four scale (p 50.019), as was the difference between the employeeletter condition and the citizens group condition (0.68units, p , 0.01).30

Taken together, these results provide further evi-dence for a mechanism that businesses can use toshapeCongressional policymaking (Hertel-Fernandez2018; Walker 2014). They also illuminate a potentialmechanism by which staffers’ perceptions of theirconstituents’preferences can be skewed away from theattitudes held by most citizens. If a slew of employeeletters arrive supporting a particular bill, staffers maybe more likely to think that their district or state as awhole supports that bill, even if those employee lettersare not representative of public opinion. The resultsfrom the staffer communications survey, then, com-plement our observational analysis of staffer percep-tions of district and state opinion, showing onepotential mechanism by which organized interestgroups—and especially businesses—can distort therepresentation of constituents’ opinions in the eyes ofCongressional staff.

LEGISLATIVE STAFFERS, ORGANIZEDINTEREST GROUPS, AND AMERICANDEMOCRACY

As economic inequalities expand in the United States,many scholars and citizens have expressed concern thatthose at the top of the income distribution are advan-taged in the political process. A growing body ofresearch suggests that these fears may be well-founded.Wealthy individuals and business interests are moreinvolved in the political process than groups repre-senting the economically disadvantaged or workers(Drutman 2015; Schlozman et al. 2012). Moreover,when researchers compare policy changes to publicpreferences, results showbusiness’ and economic elites’preferences are much more predictive of policy changethan are the preferences of lower- and middle-income

citizens (Gilens 2012; Gilens and Page 2014). Overall,these results suggest a significant representation gap foraverage American citizens.

We contribute to this research, examining a poten-tially new mechanism for the breakdown in respon-siveness: the staffers serving Members of Congress andtheir contact with organized interest groups, partic-ularly businesses and ideologically conservative groups.Through our observational analysis and surveyexperiment on constituent communications, we havepresented evidence consistent with the argument thatwhen senior Congressional staff come into contact withcorporate or conservative interests, they are less likelyto have an accurate picture of what their constituentswant.

Our results cast a rather pessimistic light on dem-ocratic representation in Congress. Although seniorstaffers responsible for advising Representatives andSenators overwhelmingly report they would like tobase their decisions and recommendations on con-stituent opinion, in practice these staffers have only alimited understanding of constituent preferencesacross important policy issues. Our results also suggestthat far from being a tool for democratizing access tolawmakers, in practice Congressional correspondencecan be amechanism through which organized businessdisproportionately influences the policymaking proc-ess at the expense of other interests (cf. Yackee andYackee 2006).

The staffer survey we used to reach these conclusionsafforded us a number of advantages. It allowed us tostudy the top aides responsible for advisingMembers ofCongress on legislative and policy decisions. We werealso able to combine both observational and exper-imental evidence, finding support for interest groupsand egocentric bias in driving the mismatch. Yet ourfindings also have several important limitations.We canonly speak to staffer representationon the issues thatweincludedon the survey. Itmaybe that there are issuesonwhich staffers are much better positioned to evaluatetheir constituents’preferences. Still, given thepaucityofrepresentation on these five issues, which span verydifferent types of highly salient policies during anelection year, we think that this is unlikely to be true.

Our findings about representation are also subject tobroader concerns about the measurement of massopinion. Does it matter if legislative staffers do notcorrectly perceive their constituents’ opinions if con-stituent opinions simply represent ill-informed “non-attitudes” (Converse 1977)? We are agnostic in thisanalysis about the quality of constituent attitudesthemselves, simply noting that we have chosen salientpolicy issues that Congress has recently considered.These are the kinds of issues where we should expectcitizens to have developed relatively clearer attitudes.Moreover, even giving staffers a generous benefit of thedoubt in their precision, themagnitude of themismatchbetween staffer perceptions and constituent prefer-ences is quite striking.

A third limitation is that our results are necessarilybounded to the contemporary Congress. Recentincreases in Congressional contact under the Trump

30 Although we are hesitant to conduct subgroup analyses with thisexperiment given our small sample, in Appendix P we show thatstaffers from districts and states with higher levels of unemploymentweremuchmore likely to interpret employeecorrespondenceasbeingmore representative than were staffers from areas of lower unem-ployment. The fact that staffers who were potentially much moreattuned to unemployment in their constituency were much moreresponsive to the employee letters is consistent with theories ofstructural business power.At the same time, we reiterate that we viewthese results as requiring follow-up research.

Alexander Hertel-Fernandez, Matto Mildenberger, and Leah C. Stokes

16

Administration may change some of these findings. It isalso an open question whether these results would holdin other times or across different legislative bodies.Future work assessing legislative staffers’ role inmediating mass responsiveness in the states, or in othercountries, would be very useful.

Additional research could also examine the mecha-nisms underpinning our findings in greater detail. Inparticular, future work ought to disentangle con-servative and business-group contact with staffers. Doboth types of groups skew staffer perceptions of thepublic?Ourfindings in the communications experimentoffer some suggestive evidence that business groupsmay independently shape staffer perceptions of theirconstituents. But additional work might examine thereasons behind this bias in more detail. For example,staffers may be more deferential to business repre-sentatives because they view employees as moreknowledgeable about policy issues than ordinary citi-zens, especially on issues that affect their firms.

Ultimately, our analysis should not be seen as anindictment of staffers alone. If the public does not con-sistently communicate its preferences to Congress—andinterest groups do—then it is hard to expect staffers toaccurately perceive what the public wants. In addition,some citizens may be more likely to contact their rep-resentatives than others, whether these disparities occurby issue, socioeconomicbackground,orgeography. Suchasymmetric mobilization can lead to distorted signaling,making it harder for staff to understand the preferencesof the whole public they seek to serve (Stokes 2016).Though efforts to increase mass participation in thepolicymaking process are likely to be challenging, theycould help to close the staffer-constituent opinion gapand ensure better representation in Congress. Regard-less of how future scholars and activists choose to pursuethe researchquestions andnormative issues raised in thispaper, the bottom line is that legislative staffers—andtheir resources, perceptions, and interactions—merit farmore attention in political science.

SUPPLEMENTARY MATERIAL

To view supplementary material for this article, pleasevisit https://doi.org/10.1017/S0003055418000606.

Replication materials can be found on Dataverse at:https://doi.org/10.7910/DVN/OWQNVF.

REFERENCES

Arnold, R. D. 1990. The Logic of Congressional Action. New Haven:Yale University Press.

Bartels,L.2008.UnequalDemocracy. Princeton:PrincetonUniversityPress.

Becher, M., K. Kappner, and D. Stegmueller. 2018. “Local UnionOrganization and Lawmaking in the U.S. Congress.” Journal ofPolitics 80 (2).AU3

Bertrand, M., M. Bombardini, and F. Trebbi. 2014. “Is It WhomYouKnow or What You Know? An Empirical Assessment of theLobbying Process.” American Economic Review 104 (12):3885–920.

Broockman, D., and C. Skovron. 2018. “Bias in Perceptions of PublicOpinion Among Political elites.” American Political ScienceReview. AU4

Butler, D. 2014.Representing the Advantaged. NewYork: CambridgeUniversity Press.

Carnes, N. 2013. White-Collar Government. Chicago: University ofChicago Press.

Carroll, R., J. Lewis, J. Lo, N. McCarty, K. Poole, and H. Rosenthal.2015. “‘Common Space’ DW-Nominate Scores with BootstrappedStandard Errors.” www.voteview.com/dwnomin_joint_house_and_senate.htm.

CMF. 2011. “Communicating with Congress.”Report, CongressionalManagement Foundation.

Converse, P. 1977. “TheNature of Belief Systems inMass Publics.” InIdeology and Discontent, ed. D. Apter. New York, NY: Free Press.

Cook, P., and K. Goss. 2014. The Gun Debate. Oxford UniversityPress. AU5

Curry, J. 2015.Legislating in theDark. Chicago:University ofChicagoPress.

Dafoe,A., B. Zhang, andD.Caughey. 2017. InformationEquivalencein Survey Experiments. Unpublished manuscript. AU6

DeGregorio, C. 1988. “Professionals in the U.S. Congress: AnAnalysis of Working Styles.” Legislative Studies Quarterly 13 (4):459–76.

Druckman, J., and L. Jacobs. 2015. Who Governs? Chicago: Uni-versity of Chicago Press.

Drutman, L. 2015. The Business of America is Lobbying. New York,NY: Oxford University Press.

Enns, P. 2015. “Relative Policy Support and Coincidental Repre-sentation.” Perspectives on Politics 13 (4): 1053–64.

Epley, N., B. Keysar, L. Van Boven, and T. Gilovich. 2004. “Per-spective Taking as Egocentric Anchoring and Adjustment.”Journal of Personality and Social Psychology 87 (3): 327.

Erikson, R. 2015. “Income Inequality and Policy Responsiveness.”Annual Review of Political Science 18: 11–29.

Erikson, R., N. Luttbeg, and W. Holloway. 1975. “Knowing One’sDistrict: How Legislators Predict Referendum Voting.” AmericanJournal of Political Science 19 (2): 231–46.

Fenno, Jr. R. 1978. Home Style. New York: HarperCollins.Free, L., andH. Cantril. 1967.The Political Beliefs of Americans. New

Brunswick, NJ: Rutgers Press.Gilens, M. 2012. Affluence and Influence. Princeton: Princeton Uni-

versity Press.Gilens,M., andB. Page. 2014. “TestingTheories ofAmericanPolitics:

Elites, Interest Groups, and Average Citizens.” Perspectives onPolitics 12 (3): 564–81.

Glynn, A. 2013. “What can We Learn with Statistical Truth Serum?Design and Analysis of the List Experiment.” Public OpinionQuarterly 77 (S1): 159–72.

Hacker, J., and P. Pierson. 2010.Winner-Take-All Politics. NewYork,NY: Simon and Schuster.

Hacker, J., and P. Pierson. 2016.American Amnesia. New York, NY:Simon and Schuster.

Hall, R. 1996.Participation in Congress. NewHaven: Yale UniversityPress.

Hall, R., and A. Deardorff. 2006. “Lobbying as Legislative Subsidy.”American Political Science Review 100 (1): 69–84.

Hall,R., andF.Wayman. 1990. “BuyingTime:Moneyed Interests andthe Mobilization of Bias in Congressional Committees.” AmericanPolitical Science Review 84 (3): 797–820.

Hammond, S. W. 1996. “Recent Research on Legislative Staffs.”Legislative Studies Quarterly 21 (4): 543–76.

Hedlund, R., and P. Friesema. 1972. “Representatives’Perceptions ofConstituency Opinion.” Journal of Politics 34 (3): 730–52.

Hertel-Fernandez, A. 2018. Politics at Work: How Companies TurnTheir Workers into Lobbyists. New York, NY: Oxford UniversityPress.

Howe, P., M. Mildenberger, J. Marlon, and A. Leiserowitz. 2015.“Geographic Variation in Opinions On Climate Change at Stateand Local Scales in the USA.” Nature Climate Change 5 (6):596–603.

Jacobs, L., and R. Shapiro. 2000. Politicians Don’t Pander. Chicago:The University of Chicago Press.

Jacobs, L., and T. Skocpol. 2015. Health Care Reform and AmericanPolitics. New York, NY: Oxford University Press.

Legislative Staff and Representation in Congress

17

Kalla, J., and D. Broockman. 2015. “Campaign Contributions Facil-itate Access to Congressional Officials: A Randomized FieldExperiment.”American Journal of Political Science 60 (3): 545–58.

Kingdon, J. 1984. Agendas, Alternatives, and Public Policies. Boston:Little, Brown.

Kingdon, J. 1989. Congressmen’s Voting Decisions. Ann Arbor: TheUniversity of Michigan Press.

Kollman,K. 1998.OutsideLobbying. Princeton: PrincetonUniversityPress.

Malbin, M. 1980.Unelected Representatives. New York: Basic Books.Mann, T., and N. Ornstein. 2016. It’s Even Worse than it Looks. New

York, NY: Basic Books.Mayhew, D. 2004. Congress: The Electoral Connection, 2nd edition.

New Haven: Yale University Press.McCrain, J. (forthcoming). “RevolvingDoorLobbyists and theValue

of Congressional Staff Connections.” Journal of Politics.Mildenberger,M. 2015. “Fiddlingwhile theWorldBurns: TheLogic of

Double Representation in Comparative Climate Policymaking.”PhD thesis. New Haven: Yale University.

Mildenberger,M., J.R.Marlon,P.D.Howe, andA.Leiserowitz. 2017.“The Spatial Distribution of Republican and Democratic ClimateOpinions at State and Local Scales.” Climatic Change 145 (3–4):539–48.

Miler, K. 2010. Constituency Representation in Congress. New York,NY: Cambridge University Press.

Miller, W., and D. Stokes. 1963. “Constituency Influence in Con-gress.” American Political Science Review 57 (1): 45–56.

Montgomery, J., and B. Nyhan. 2016. “The Effects of CongressionalStaff Networks in the U.S. House of Representatives.” AU7

Nickerson, R. 1999. “How We Know—and Sometimes Misjudge—What Others Know: Imputing One’s Own Knowledge to Others.”Psychological Bulletin 125 (6): 737.

Romzek, B., and J. Utter. 1997. “Congressional Legislative Staff:Political Professionals or Clerks?” American Journal of PoliticalScience 41 (4): 1251–79.

Salisbury, R., and K. Shepsle. 1981. “U.S. Congressman as Enter-prise.” Legislative Studies Quarterly 6 (4): 559–76.

Schlozman, K. L., S. Verba, and H. Brady. 2012. The UnheavenlyChorus. Princeton: Princeton University Press.

Stokes, L.. 2016. “Electoral Backlash against Climate Policy: ANaturalExperiment onRetrospectiveVoting andLocalResistanceto Public Policy.” American Journal of Political Science 60 (4):958–74.

Swers, M. 2002.The DifferenceWomenMake. Chicago: University ofChicago Press.

Walker, E. 2014. Grassroots for Hire. New York, NY: CambridgeUniversity Press.

Walker, J. 1991.Mobilizing Interest Groups in America. Ann Arbor,MI: University of Michigan Press.

Yackee, J. W. and S. W. Yackee. 2006. “A Bias towards Business?Assessing Interest Group Influence on the U.S. Bureaucracy.”Journal of Politics 68 (1): 128–39.

Alexander Hertel-Fernandez, Matto Mildenberger, and Leah C. Stokes

18