Why are professors liberal?

Neil Gross & Ethan Fosse

Published online: 29 January 2012# Springer Science+Business Media B.V. 2012

Abstract The political liberalism of professorsan important occupational groupand anomaly according to traditional theories of class politicshas long puzzledsociologists. This article sheds new light on the subject by employing a two-stepanalytic procedure. In the first step, we assess the explanatory power of the mainhypotheses proposed over the last half century to account for professors liberalviews. To do so, we examine hypothesized predictors of the political gap betweenprofessors and other Americans using General Social Survey data pooled from 19742008. Results indicate that professors are more liberal than other Americans becausea higher proportion possess advanced educational credentials, exhibit a disparitybetween their levels of education and income, identify as Jewish, non-religious, ornon-theologically conservative Protestant, and express greater tolerance for contro-versial ideas. In the second step of our article, we develop a new theory of professorspolitics on the basis of these findings (though not directly testable with our data) thatwe think holds more explanatory promise than existing approaches and that sets anagenda for future research.

Keywords Professors . Intellectuals . Politics . Liberalism . New class

Sociological research consistently finds that American professors generally havesocial and political attitudes to the left of the US population. Not long after WilliamF. Buckley famously railed against liberal academe in God and Man at Yale (1951), alandmark survey showed nearly half of academic social scientists scoring high on anindex of permissive attitudes toward communism (Lazarsfeld and Thielens 1958).

Theor Soc (2012) 41:127168DOI 10.1007/s11186-012-9163-y

N. Gross (*)Department of Sociology, University of British Columbia, 6303 NW Marine Drive, Vancouver,BC V6T1Z1, Canadae-mail: ngross@interchange.ubc.ca

E. FosseDepartment of Sociology, Harvard University, 33 Kirkland Street, Cambridge, MA 02138, USAe-mail: efosse@wjh.harvard.edu

The next large-scale survey of professors political views, in the late 1960s (Ladd andLipset 1976), found that just under half identified as left of center, compared to abouta fifth of the US population, and that they voted 2025% more Democratic than theAmerican electorate. Recent studies echo these conclusions, confirming that profes-sors are decidedly liberal in political self-identification, party affiliation, voting, and arange of social and political attitudes (Gross and Simmons 2007; Rothman et al.2005; Schuster and Finkelstein 2006, p. 506; Zipp and Fenwick 2006).

Understanding why American professors tend to be liberal is important for threereasons. First, doing so sheds light on an occupation of great sociological signifi-cance. The approximately 1.4 million college or university professors and instructorsworking in the United States today play pivotal social roles producing new knowl-edge and technology; teaching and credentialing growing numbers of students;advising government, industry, and nongovernmental organizations; and shapingsocial narratives in what Alexander (2006) calls the civil sphere. Politics do notbear directly on all work professors do, but higher education institutions as loci ofknowledge production and dissemination may be influenced in important ways bytheir political views.1 Second, a major public controversy surrounds professorialpolitics. Conservative critics charge that many liberal and radical academicians abusetheir positions, indoctrinating students politically and silencing conservative voiceson campus (e.g., DSouza 1991; Horowitz 2006, 2007; Kimball 1990; see Brub2007; Messer-Davidow 1993; Smith et al. 2008). Public opinion polls find that abouta third of Americans believe liberal bias in academia to be a serious problem (Grossand Simmons 2006), and the issue of professors and politics has received extensivecoverage in the press. Critics have proposed various remedies, some with far-reachingimplications. Discovering the sources of professorial liberalism could inform thedebate that has ensued. Third, given the large number of social, cultural, andinstitutional processes with which the phenomenon of American professorial liberal-ism appears connected, it represents a fruitful case for study, providing an opportunityto reexamine core issues of politics and society.

In part because of the last of these points, over the years sociologists haveadvanced numerous hypotheses to explain the liberalism of professors.2 Ladd andLipset (1976), taking up Lazarsfeld and Thielens (1958) notion of the academicmind, argued that intellectualisma rational, critical, creative mindset linked to theWestern intellectual traditionis naturally at odds with most forms of conservatism.

1 There is growing recognition among sociologists of knowledge that professors politics matter. Forexample, sociologists of ideas attend to the ways in which commitment to paradigms and approaches,especially for humanists and social scientists (Gross 2008; Rojas 2007) but for natural scientists as well(Frickel 2004), may be bound up with political identity. Scholarship on the sociology of intellectualsconsiders how the economic conservatism of a small number of academics, namely economists, powerfullyshaped late twentieth-century social and economic policy (Babb 2004; Fourcade-Gourinchas and Babb2002). Likewise, a large body of research in science studies examines how scientists explicit and tacitpolitical commitments may influence their investigations (e.g., Barnes 1977; Shapin and Schaffer 1985;Haraway 1989).2 In this article liberalism refers to views that would be classified as left of center in the contemporaryAmerican context, including favorable attitudes toward certain aspects of the welfare state and governmentprograms to reduce inequality, support for a politics of inclusion vis--vis minority groups, and a preferencefor multilateralism and diplomacy in international affairs. We address below the question of historicalvariation in liberalisms meaning.

128 Theor Soc (2012) 41:127168

Ladd and Lipset mobilized this argument to explain the especially high levels ofliberalism of American professors at elite institutions, who, they theorized, exhibit themost intellectualism (also see Lipset 1982; Lipset and Dobson 1972). Politicalsociologists in the 1970s and 1980s engaging the new class problem developedother hypotheses (for discussion see Bruce-Briggs 1979; King and Szelnyi 2004). Tothese scholars, professors belong to the broader occupational category designated byBrint (1985, pp. 401402)following Kristolsocial and cultural specialists,which also includes artists, writers, journalists and social scientists [outsideacademe]. For some, like Brint, the groups liberalism reflected high levels ofeducation and the experience of having come of age in the 1960s (Bell 1976, 1979;Brint 1984). Others saw its political dispositions and sympathies to be a function ofclass location and interests, variously understood (Gouldner 1979; Lamont 1987,1992). For still others the liberalism of professors and of the professional classesgenerally evidenced the rise of a new political culture in the advanced industrialdemocracies characterized by new cross-class alignments and concerns (Clark andHoffmann-Martinot 1998; Clark and Lipset 1991; see Manza et al. 1995). Finally,some social scientists today suggest the facultys liberal tilt may reflect discriminatoryhiring and promotion practices (Klein and Stern 20042005, 2005; Klein and Western20042005; Rothman et al. 2005).

The aim of this article is to move discussion of professors and politics forwardempirically and theoretically. In the first half of the article we do what we believe noother scholars working on the subject have done: use multivariate data to assess therelative explanatory power of leading theoretical accounts. After discussing theliberalism of professors in more detail, we identify, from a review of research onprofessorial politics published over the past half-century, six hypotheses that speak tothe question of why professors tend to have views that are distinctive in the Americancontext. We test these hypotheses using a variant of the Oaxaca-Blinder regressiondecomposition procedure, an econometric technique typically used to examine wagedisparities, on data from the General Social Survey (GSS) pooled over the years19742008. During this period 326 GSS respondents with non-missing values onour outcome variable were employed as professors or instructors in higher educa-tion, and we examine how much of the gap between their politics and those of otherAmericans can be accounted for with variables related to each of the hypotheses.These variables together account for about 43% of the gap, with the most importantfactors advanced education, the disparity between professors educational levels andtheir incomes, the fact that a higher proportion of professors than non-professorshave no religious affiliation or are Jews or non-theologically conservative Protes-tants, and intellectualism (operationalized as tolerance for controversial ideas not of aliberal nature).

Although our models account for a large proportion of the gap between professorsand other Americans, in conducting our statistical analyses we came to believe thatnone of theories we had tested was fully adequate. In the second half of our article weoutline the new theory of professorial politicsnot directly testable with our datathat came into view as we reflected on our empirical findings. Bringing togetherinstitutionalist theory (Clemens and Cook 1999; Powell and DiMaggio 1991; Scott1995) and social psychology-influenced approaches to status attainment and occupa-tional choice (Sewell and Hauser 1975), we argue that, given the way the American

Theor Soc (2012) 41:127168 129129

academic profession was institutionalized and its twentieth-century history, it hasacquired such a strong reputation for liberalism and secularism that over the last35 years few politically- or religiously-conservative students, but many liberal andsecular ones, have formed the aspiration to become professors. We theorize thatthese commonplace aspirational differences, combined with more occasional down-stream assessments of organizational fit, are the main proximate causes of thepreponderance of liberals in academe today. To buttress our argument we cite anumber of studies that offer indirect support, concluding with a call more researchon how institutionalized occupational reputations may interact with political identityto shape educational and occupational aspirations and choices in the professoriate andbeyond.

The liberalism of American professors

Although the liberal tendencies of American professors have grown over time, suchtendencies have been evident for much of the twentieth century. Before then, in theeighteenth century and first half of the nineteenth, the primary mission of Americancolleges and universities was to inculcate civic and moral virtues and transmitclassical culture. Professors were often clergymen expected to teach accepted veritiesand live uncontroversial lives. With the birth of the American research university inthe second half of the nineteenth century, the capacity to produce new knowledgecame gradually to be valued above political or religious conformism (Hofstadter andMetzger 1955; Veysey 1965). This trend continued in the early twentieth century,coinciding with the involvement of many professors, especially in the emergingsocial sciences, in Progressive Era politics (Cohen 2002; Ross 1991). Simultaneously,private money pouring into research universities raised new demands for control, asbusiness interests clashed with an emerging public intellectualism that saw someleading academics criticize laissez-faire capitalism and imperialism. Professors lever-aged their newfound social power into institutionalized protections of academicfreedom (Haskell 1996)important enabling conditions for a new wave of facultyradicalism, which struck during the Depression with uneven force across disciplinesand institutions. Two decades later, radicalism in the university was under attack.McCarthy and his followers were convinced that national security was threatened notjust by the alleged Communist Party membership and sympathies of some Americanacademics but also by their ostensible tendency to support redistributionist policies,broad civil liberties protections, and dtente with the Soviet Union (Schrecker 1986).

It was to assess how well social scientists were faring in the context ofMcCarthyism that the Lazarsfeld and Thielens (1958) survey was undertaken.However, their most striking findings pertained not to academic freedom but topolitics. Three years after the 1952 presidential elections, in which 55% of theelectorate voted for Dwight Eisenhowera politician whose moderate views werea sign the Republican Party had not yet veered right (Rae 1989)only about 16% ofthe social scientists surveyed reported usually voting for Republicans; 47% wereDemocrats and the rest Independents.

Other studies in the late 1950s and early 1960s, most restricted to a few fields orinstitutions, confirmed these statistics but also found social scientists to be the most

130 Theor Soc (2012) 41:127168

liberal of all professors. Reviewing this research, Faia (1974, p. 173) concluded thatwhen professors are included [in surveys] who are either in the natural sciences orprofessional schools, the percentage claiming a Democratic affiliation will dropaccordingly and will be well below the proportion indicating a Democratic affiliationin the electorate generally.

Political mobilization in the late 1960s and early 1970s, much of it centered oncollege campuses, was the context for the major survey of American professors,reported by Ladd and Lipset (1976) in The Divided Academy. Liberalism andDemocratic support, they found, were no longer limited to the social sciences.Examining all fields, Ladd and Lipset found 46% of professors had political identitiesof left or liberal, 27% middle of the road, and 28% conservative, with younger facultymore liberal than older. As was increasingly the case outside academe, whereSouthern conservatives were abandoning the Democratic Party in the wake of theCivil Rights movement (Black and Black 2002; Lassiter 2005) and where excitementamong conservative populists about Barry Goldwaters 1964 presidential cam-paign was heightening commitment to the GOP (Critchlow 2007), Democraticsupport went hand in hand with left or liberal self-identification. Some differences byfield remained, but theywere not as great as they had once been. In 1968 and 1972 socialscientists and humanists supported Democratic candidates Hubert Humphrey andGeorge McGovern, respectively, in nearly equal measure; professors in the physicaland biological sciences were next most likely to vote Democratic. Republicans com-manded a majority only among professors of business, engineering, and agriculture.

New trends were observed in the 1980s. Comparing faculty survey data from 1975and 1984 to the data Ladd and Lipset had analyzed, Hamilton and Hargens found aconservative trend in faculty politics from the 1970s to the early 1980s, parallelingattitudinal developments outside academe that helped precipitate the Reagan revo-lution. Professors identifying as leftist increased by about 1%, but liberal self-identifications declined by roughly seven points. Conservative identifications, bothmoderate and strong, showed increases, these together being approximately equal tothe liberal losses (Hamilton and Hargens 1993, p. 607). In 1984, 6% of professorsidentified themselves as on the left, 34% as liberal, 27% as middle of the road, 30% asmoderate conservative, and 4% as strong conservative (Here and throughout thepaper we round, so percent tabulations may not add to 100). Aggregate conservativegrowth was found to result from increasing conservative self-identification in themost conservative fields.

The trend was short-lived. Research shows that in recent decades the Americanprofessoriate has become an even greater liberal and Democratic stronghold. Forexample, voter registration records of faculty at the University of California, Berkeleyand Stanford reveal that of the 67% for whom information could be obtained,almost 50% were registered Democrats and about 5% Republicans (Klein andWestern 20045; but see Cohen-Cole and Durlauf 2005). A national survey offaculty in six social sciences and humanities showed ratios of Democratic to Repub-lican voters somewhere between 7:1 and 9:1 (Klein and Stern 20042005). A 1999study of faculty in all fields reported that 72% identify as left of center and 15% rightof center; that 50% are registered Democrats, compared to 33% Independents and11% Republicans; and that faculty attitudes display an across the board commitmentto positions that are typically identified with contemporary liberal ideals (Rothman

Theor Soc (2012) 41:127168 131131

et al. 2005, p. 8).3 Another faculty survey using a more detailed scale found thatpercentages of liberals, moderates, and conservatives fall around 48, 31, and 17,respectively; that in the 2004 elections 72% reported voting for John Kerry and 25%for George Bush; and that professors hold liberal views on business and the freemarket, support for international institutions, and separation of church and state(Tobin and Weinberg 2006).

Descriptively these studies tell a consistent story. Yet much research on the topicrecently has been ideologically motivated and beset by problems of sampling and ques-tionnaire design (Gross and Simmons 2007). Drawing most of their questions verbatimfrom long-standing surveys of the US population such as the General Social Survey(GSS) and the American National Election Study (ANES), Gross and Simmons(2007) were able to conduct a more reliable and representative survey of Americanprofessors and to make meaningful comparisons to the general population. Theirresearch shows that in 2006, 62% of professors described themselves as any shade ofliberal, 18% as middle of the road, and 20% as any shade of conservative. Bycontrast, data from the 2004 ANES showed 29% of Americans describing themselvesas liberal and 39% as conservative. Gross and Simmons argue that those self-identified as slightly liberal or slightly conservative should be considered mod-erates, changing the overall distribution to 44% liberal, 47% moderate, and 9%conservative and revealing a trend toward less extreme views in the youngest facultycohorts. Even so there is no question that liberal self-identification and attitudes in anumber of domains are much more common among professors than in the general USpopulation. (Among other points, Gross and Simmons argue that the social liberalismof professors is more pronounced than their economic liberalism.)4

Professors are also more likely to be registered Democrats and vote Democratic.Gross and Simmons (2007) reported that 51% of professors had Democratic affili-ations, 35% Independent (with Democratic-leaning Independents greatly outnumber-ing Republican-leaning ones), and 14% Republican. Gallup polls in December 2007showed 31% of voting-age Americans to be Democrats, 41% Independents, and 28%Republicans (Newport 2008). As for voting, in the 2004 presidential elections, Bushwon 51% of the popular vote and Kerry 48%. Of professors who reported voting,78% said they voted for Kerry and 20% for Bush.

The origins of professors politics

But why are American professors liberal? Six testable hypotheses can be derived fromthe sociological literature on the subject. Although, as indicated above, we ultimately

3 However, the authors miscalculated the numbers for political self-identification. Later analysis revealedthat the correct figures were 62% liberal, 26% moderate, and 12% conservative (see Rothman and Lichter2008).4 Given that this is so, some might ask whether professors should truly be seen as liberal at all. Our view,developed elsewhere (Gross et al. 2011), is that liberalism and conservatism are not objective categories ofideology and political practice, with the analyst entitled to specify a priori what makes someone a liberal orconservative. Rather, these are historically-variable collective identities, the meanings of which areconstantly in dispute among actors in the political field, with significant consequences. We thus hewclosely to our statement above that we mean by liberalism those views and self-understandings seen asliberal in the contemporary American (neoliberal) context.

132 Theor Soc (2012) 41:127168

found none of these fully satisfactory, we unprejudiciously lay out the rationale for eachas a prelude to reporting our empirical test of them.

Hypothesis 1: The sociodemographic characteristics of the professoriate explain whyprofessors tend to be liberal Political allegiances change over time, and analysts errwhen they treat as natural social groups defined by such categories as ethnicity, race,and gender (Brubaker 2004). Nevertheless, data show a strong, consistent associationbetween membership in core U.S. sociodemographic groups and social and politicalattitudes. For example, class held constant, African-Americans tend to have moreliberal views than whites on economic and racial matters (Schuman et al. 1998).Likewise, social psychological research suggests that men tend to be morecomfortable than women with social hierarchies and that their worldviews inthis regard shape political engagement in important ways (Sidanius and Pratto2001).

Occupations obviously vary in sociodemographic makeup and, accordingly, ourfirst hypothesis explores how the composition of the professoriate might relate to itspolitics. As they are linked to broader literatures, we evaluate education, religion, andclass separately, below, and consider under Hypothesis 1 the effect of cohort, familial,gender, and geographic composition.

Concerning cohort, numerous sociologists have followed Ladd and Lipset (1976,p. 23) in viewing the 1960s as a watershed moment in faculty politics, a decade whena new generation of academics, who knew neither Hitler nor Stalin and found noreason to hold back their criticisms of American society, entered the university. Laddand Lipset reported that already by 1969 the political leanings of this generation hadpushed overall faculty politics further to the left, and predicted that effects wouldcontinue to be felt for decades since the 1960s were a time of major and expansion ofthe American higher education sector. We therefore examine whether a large numberof baby boomers in the academic ranks might help account for professors liberalattitudes. In terms of family structure, some observers have noted that Americanprofessors tend to have fewer children on average than others (see Bassett 2005;Hochschild 2003). This is so in part because, for most academics, normative child-bearing years coincide with the struggle for tenure, a fact affecting academic womenespecially. Higher rates of childbearing in a group may be associated with politicalconservatism not simply because religious conservatives, in particular, place morevalue on childbearing (Greeley and Hout 2006), but also because larger families aremore expensive, which could militate toward support for politicians and partiesfocused on tax reductions and maximization of take home pay, as the RepublicanParty has been. Lower rates of childbearing among professors might thereforecontribute to their liberalism. As for gender, while women have historically beenunderrepresented among professors, in recent decades they have registered majorgains, and now make up just under half of entering faculty cohorts (Schuster andFinkelstein 2006:449). We consider the possible connection between professorialliberalism and a feminized academic labor force.

Finally, other observers have linked professorial liberalism to the fact that mostcolleges and universities are located in urban areas, whose residents typically havemore liberal views; and to the fact that, relative to other regions of the country, thepolitically-conservative South has traditionally been underserved by higher

Theor Soc (2012) 41:127168 133133

education, such that a disproportionately small number of professors reside there(Wilson 2008). On the first of these points, it does not seem plausible to us thatprofessors political views would be much affected by the population density of thecommunities in which they eventually settle. We know too many examples ofacademics who are lonely voices of radicalism in their small, rural towns. But wedo examine whether a higher proportion of professors than others might have grownup in big cities, reflecting differential educational opportunities and normative occu-pational aspirations, and whether this could help explain their liberalism. With respectto residence outside the South, we examine the effects of this directly on theassumption that region may powerfully shape the political culture of institutions,including institutions of higher education, and that institutional culture in turn mayaffect the politics of individual professors. We additionally consider whether profes-sors are more likely to be foreign born, given the internationalism of the academiclabor market and that the American professoriate has served as an important vehicleof social mobility for some immigrant groups; and, separately, whether professors aremore likely to have lived in a foreign country as teenagers, given that, as we discussbelow, many professors come from relatively privileged families in which travel andexposure to other cultures are common. Both factors might be associated withliberalism. To be sure, the political attitudes of immigrants to the United States varydramatically by country of origin, class background, religion, and other factors. ButAmerican conservatives have often favored restrictions on immigration and on thegranting of political and social rights to immigrants, and all else being equal thisshould lead to less conservative political self-identification within immigrant com-munities (though professors who are immigrants tend to be buffered against suchrestrictions.) For non-immigrants, living abroad while growing up might foster thedevelopment of a more cosmopolitan, international orientation (for discussion seeSaito 2011) that would be unswayed by conservative appeals to American nationalidentity.

Hypothesis 2: High levels of educational attainment among professors account fortheir liberalism Debate continues over the relationship between education and socialand political attitudes. On the one hand, a large body of research from the 1950s,1960s, and 1970s showed that the more years of education people have, the morecomfortable they tend to be with social diversity, the more tolerant of political dissent,the less traditional their action-orientations and thinking, the more they tend to bereligious skeptics, and the more coherent and constrained their worldviews (fordiscussion see Feldman and Newcomb 1969; Pascarella and Terenzini 1991). Fourmechanisms were posited to account for these effects: socialization into an enlight-enment culture institutionalized in schools; cognitive upgrading; exposure to peerculture and to students from diverse backgrounds; and class differentiation processesin which middle- and upper-middle-class students symbolically embrace tolerance todistance themselves from putatively authoritarian working-class culture. On the otherhand, critics of this perspective viewed findings of greater tolerance among theeducated as either artifacts, since the educated understand survey items better andknow the socially acceptable answers, or irrelevant, inasmuch as education in capi-talist societies encourages individualism and impedes meaningful collective action(Jackman and Muha 1984).

134 Theor Soc (2012) 41:127168

Since these positions were first staked out, overstated claims on both sides havebeen tempered. Work in the tradition of Newcomb (1943), who followed studentsduring their years at Bennington College (and beyond), continues to show that atcertain kinds of schools and for certain kinds of students, higher education, at least,can be a socially and politically transformative experience. Pascarella and Terenzininote that studies drawing on national probability samples invariably indicatechanges during the college years in students political attitudes and values towardmore liberal political stances (1991, p. 278). More specifically, studies dealing withchanges in attitudes and values related to civil rights, civil liberties, racism, anti-Semitism, or general tolerance for nonconformity uniformly report shifts towardsocial, racial, ethnic, and political tolerance and greater support for the rights ofindividuals in a wide variety of areas (p. 279). However, other research suggestssuch effects may vary cross-nationally and generationally; reflect dispositional differ-ences leading students to opt in or out of college; depend on specific educationalexperiences; be restricted to social attitudes and not touch economic attitudes or viewsof foreign policy; and have ambiguous consequences for political behavior (e.g., Dey1997; Highton 2009; Jennings and Stoker 2008; Pascarella and Terenzini 2005;Phelan et al. 1995; Weakliem 2002; Weil 1985).

These lines of investigation suggest a straightforward hypothesis about the liber-alism of professors. In most colleges and universities professors are required to holddoctoral or other advanced degrees. Insofar as rates of advanced degree holding aresubstantially higher among professors than among non-professors, and insofar asattainment of an advanced degree requires many years of schooling, this may explainprofessors greater liberalism.

To an extent, the hypothesis has already been explored. Analysts examining theinfluence of class position on political views and behavior have long included in theirmodels how peoples politics vary with education. For example, Brint (1984),examining empirically the new class thesis of Gouldner and others, argued thatthe liberalism of knowledge workers had been overstated, and that what liberalismcould be found was a function in part of the liberalizing effects of a much expandedhigher-education system (60). Gerteis (1998) analyzed American middle-class pol-itics in two periods of political change, the 1970s and the 1990s, and found that inboth a signal characteristic of occupations tilting most heavily left was educationalattainment. Likewise, Manza and Brooks (1997) found that growing support forDemocrats among professionals between the 1950s and 1980sa key politicalrealignmentwas driven in part by educational experiences. Important as thesestudies are, none focuses on the professoriate specifically.

Hypothesis 3: Professors tend to be liberal because they exhibit high levels ofintellectualism In a 1972 article with Dobson, Lipset identified a structural tensionin contemporary societies. In such societies intellectualswhom Lipset, quotingHofstadter, defined as those persons who have creative, critical, and contemplativemindsare given considerable power to educate the young and inform publicpolicy. Yet intellectuals are often social critics, especially humanists whose workinvolves a tendency to appraise [everything] in terms of general conceptions of thedesirable (p. 138). If the conceptions thus favored by intellectuals are at odds withprevailing political and institutional practices, society may come under intense

Theor Soc (2012) 41:127168 135135

scrutiny from a core subsystem, and social conflict may arise. Why do intellectualsespecially contemporary academic intellectualsso often embrace the role of socialcritic? Lipset and Dobson never fully answered this question, but, drawing from thesociology of science, they pointed to several relevant features of academic life. Statushierarchies in academe create large groups of perpetually frustrated professors whomay channel resentment and alienation into social criticism.5 Likewise, the orga-nized skepticismMerton posited as part of the normative structure of science may bea general tendency, found among nearly all intellectuals, to doubt conventionalverities and reigning myths (p. 160). Finally, the growth of academe in the lattertwentieth century led to an increasing concentration of intellectuals within certainsocial and geographic settings (p. 161)such as college townswhich couldintensify political sentiment. Lipset and Dobsons general thesis was that occupationsexhibiting the most intellectualism tend naturally to be the most liberal and thatwithin occupations such as college teaching liberalism and intellectualism arecorrelated.

There are at least three problems with these intriguing claims. First, they are to anextent tautological: being a social critic from the left is virtually built into Lipset andDobsons definition of intellectualism. Second, Lipset and Dobson never adequatelyoperationalized intellectualism. Ladd and Lipset tried to do so by assuming that themost successful academicsthose teaching at prestigious schools or with the mostpublicationsare the most intellectual, but many other factors are associated withacademic success (see, e.g., Cole and Cole 1973; Leahey 2007). Third, a growingconsensus among social scientists holds that there is little scholarly warrant and muchanalytic danger in trying to distinguish conservatism from liberalism on the basis ofthe putative dogmatism of the former and the greater creativity of the latter. Forexample, had the American conservative movement not exhibited considerable ideo-logical and organizational creativity in the 1970s and 1980s, it would never haveflourisheda point easily missed if conservatism is equated with cognitive rigidity(e.g., Brennan 1995; Coontz 2000; Diamond 1995; Himmelstein 1990; Pierson andSkocpol 2007; Shulman and Zelizer 2008).6

Nevertheless, we make a preliminary effort to evaluate the hypothesis. First,lacking more direct measures, we examine whether professors may be more intellec-tual than other Americans in the sense of being willing to give a hearing to ideas thatmost people would regard as abhorrent, and whether this helps to explain theirliberalism. Critics of political correctness in the academy (e.g., Downs 2005;Hamilton 1995) may scoff at the notion that professors believe in an unrestrictedmarketplace of ideas, and there is ample anecdotal evidence of intolerance on the partof the faculty toward a variety of conservative groups and opinions. Nevertheless,

5 In apparent contradiction, Lipset and Dobson also insisted, in line with Ladd and Lipsets findings, thatthe highest-status academics tended to be the most liberal. Also see Lipset (1982).6 Of course, a political movement could exhibit creativity at the ideological and organizational level andstill be characterized by rigidity among the rank and file. And, indeed, running in the opposite directionfrom work by historians on the contemporary conservative movement is scholarship by political psychol-ogists, reprising themes of the authoritarian personality, who claim that support for conservatism isundergirded by a psychological need for uncertainty avoidance; intolerance of ambiguity; needs for order,structure, and closure; perception of a dangerous world; and fear of death (Jost and Hunyady 2005, p.261). In agreement with Martin (2001), we find such claims theoretically thin and empirically problematic.

136 Theor Soc (2012) 41:127168

commitment to rational deliberation in the sense outlined by Habermas (19841986),in which no ideas are excluded a priori, is part of what Lipset had in mind byintellectualism (though there are other important components, such as a tendencytoward contemplation, that we cannot capture using political tolerance as a proxy).Second, we examine how far professors superior average performance on a vocab-ulary test may go toward accounting for the political gap between them and others.Intellectualism and intelligence, which vocabulary tests are often seen as proxies for,are distinct, as Hofstadter (1966) famously noted. But in addition to the fact that anumber of psychologists have recently drawn attention to the connection betweenintelligence and liberalism, theorizing intellectualism as a mediating factor (Deary etal. 2008; Kanazawa 2010), we think a legitimate rationale for measuring intellectu-alism partly on the basis of a vocabulary test can be found in the fact that, onceeducation is controlled for, verbal ability is associated with the kind of culturalliteracy often seen as a hallmark of intellectuals.

Hypothesis 4: The religious characteristics of professors may explain theirliberalism Mid-twentieth-century social scientists studying the relationship betweenreligion and politics in the United States focused on the political consequences ofaffiliation with the Jewish or Catholic faiths and mainline Protestant denominations.Jews were seen to have liberal views and to support the Democratic Party. Catholicswere seen to hold relatively conservative social views but typically to vote Demo-cratic. Mainline Protestants were seen as a key constituency for the Republican Party(for review, see Wuthnow 2007). Social scientists sometimes invoked these tenden-cies, in part reflecting ethnic and class compositions of different faiths, to explainoccupational politics, including the politics of professors. For example, Ladd andLipset noted that after decades of discrimination, Jews entered American academe inlarge numbers following World War II. By the 1960s, the college enrollment rate ofJews was twice that of non-Jews, and by 1969 Jews made up 10% of professors andwere heavily represented on the faculties of Ivy League schools and other eliteprivate universities (Ladd and Lipset 1976, p. 150). According to Ladd and Lipset,the liberalism of American professors could therefore be traced in part to religiouscomposition. Such claims rested on assumptions about how faith and religiouscommunity shape political views and behavior, and also about how members ofparticular faiths tend to pursue certain occupations, not least because of ethnic andimmigrant niching in the labor market.

With the revival of American conservatism in the 1970s, a group not often consideredin these analysestheologically conservative Protestants, many of whom lived in theSouth and had long voted Democraticbecame a significant political force. Mobilizedinto a diverse conservative coalition through their churches, their worldviews at oddswith major cultural shifts of the 1960s, conservative Protestants soon proved key to thesuccess of conservative Republican candidates. By the mid 1990s, Protestant religiousorthodoxy had become a consistent and important predictor of party affiliation andvoting; mainline and conservative Protestants were realigned, with Republican supportdwindling among the former and gaining dramatically among the latter (Manza andBrooks 1997, 2002; Layman 1997; Woodberry and Smith 1998). Nor could thesetendencies be explained away by other variables (Manza and Wright 2003; Miller andHoffmann 1999). Sociologists of religion observed that conservative Protestant

Theor Soc (2012) 41:127168 137137

politicization had important implications for occupational politics. On the one hand,the strength of theologically-conservative Protestantism in lower-middle- andworking-class communities may have pushed many occupations centered in thosecommunities in a politically conservative direction, creating social values voterswhose politics were at odds with their own economic interests, conventionallyunderstood. On the other hand, religious belief may have constrained educationalaspirations and occupational choices (Darnell and Sherkat 1997), so that someoccupations became normatively out of bounds for conservative Protestants.

The professoriate has three distinctive religious characteristics that may incline ittoward greater liberalism. First, what was true in Ladd and Lipsets day remains truetoday: Jews are overrepresented among professors (Schuster and Finkelstein 2006, p.460). The intervening period has witnessed the growth of a dynamic Jewish wing ofthe Republican Party, but Jews are still much more likely than non-Jews to beDemocrats and liberals (Mazur 2007). Second, although evidence suggests thatconservative Protestants are beginning to make their way into knowledge work fieldslike academia and journalism (Lindsay 2007; Schmalzbauer 2002), they remainsignificantly underrepresented among professors, especially at elite research univer-sities (Gross and Simmons 2009); academia is an occupation in which, among thefaithful, mainline Protestants predominate. Third, although a majority of Americancollege and university professors are religious believers, many surveys have docu-mented that religious skepticism is far more common among professors than withinthe US population (e.g., Ecklund and Schietle 2007). To the extent that the Americanconservative movement and Republican Party have defined themselves partly aroundissues of religious faith, professors who claim no affiliation with any religion may bepredisposed toward liberalism, explaining some of the political gap between profes-sors and other Americans.

Hypothesis 5: The liberalism of professors is explained by their class positions Althoughwe treat the politics of the professoriate as important in its own right, most work onthe topic has been under the banner of class analysis. On the one hand, insofar associologists have looked at occupational politics in general, it has been chiefly tounderstand how occupations aggregate into politically aligned classes or class frac-tions; how occupations with rising or falling memberships and economic fortunesexert pressure on the political system; or what political alignment by occupationreveals about class as social cleavage.

On the other hand, a high proportion of the work sociologists have produced on theliberalism of intellectuals specifically proceeds from class-analytic assumptions. Acore issue in this research is whether intellectuals in modern societies form a cohesiveclass or stratum with its own interests, as argued by Weber (see Sadri 1992) andtheorists of the New Class such as Gouldner (1979) and Konrd and Szelnyi(1979); or whether, located close to the poles of social and cultural power but furtherfrom the pole of economic power and internally differentiated in terms of classbackground and area of expertise, intellectuals are best thought of as a free-floatinggroup (Mannheim 1949) or as occupying one of a number of contradictory classlocations (Wright 1978, 1985). Based on one or more of these assumptions andvarying conceptions of class, scholars have developed theories of the likely effects onintellectuals political views and behavior.

138 Theor Soc (2012) 41:127168

Of these theories, the most influential has been that of Bourdieu. For Bourdieu,intellectuals are defined structurally by their possession of high levels of culturalcapital and moderate levels of economic capital. This structural position, Bourdieuasserts, shapes their politics, though he also insists that, for intellectuals, theconnection between position in the class structure and political activity is mediatedby participation within cultural fields (Swartz 1997, p. 234). Deprived of economicsuccess relative to those in the world of commerce, intellectuals are less likely to beinvested in preserving the socioeconomic order, may turn toward redistributionistpolitics in the hopes of reducing perceived status inconsistency (see Goffman 1957;Vaisey 2006), and may embrace unconventional social or political views in order todistinguish themselves culturally from the business classes (Lamont 1987). At thesame time, many intellectuals owe whatever material success they have found to thesocial functions they play for the economic elite in terms of maintaining the realm ofhigh culture on which elites rely to legitimate their dominance, and, in the case ofteachers and professors, granting educational credentials, a phenomenon alsowrapped up with the reproduction of social inequality (Bourdieu and Passeron1979; Collins 1979). Bourdieu theorizes that this indebtedness imposes a constrainton intellectuals radicalism.

In Homo Academicus, Bourdieu ([1984] 1988) extends these insights into a theoryof how political views may vary within the intellectual stratum. He argues that undernormal democratic political conditions,7 artists and independent journalists as well asaspiring academicians who find their mobility blockedbeing subject to the fewestof these constraintsdo tend toward radicalism. Among professors, Bourdieu sug-gests that progressive views should be most common in social sciences, humanities,and natural science fields because scholars in such fields tend to come from lesselevated class positions than do professors in law, medicine, or business. Where thelatter, because of their background, current class position, and even direct socialfunction, will tend to exhibit a taste for order, identifying unreservedly with thestatus quo, the former reject everything which profoundly ties the maintenanceof social order to the most insignificant practices of social custom the hierarchieswhich they evoke, the vision of social divisions which they imply (p. 51). Evenwithin each of these fields, Bourdieu theorizes, class positionunderstood from thestandpoint of his multidimensional account of social stratificationshould be corre-lated with political outlook. Bourdieu finds empirical support for these claims in asurvey of tenured faculty in Paris administered in the wake of May 1968.

In light of the influence of Bourdieus theory, we test whether the disjuncturebetween professors educational levels and their income helps account for theirliberalism. We also consider the significance of differential levels of confidencein business between professors and other Americans, a measure of symbolicdistinction from the business classes. In addition, we evaluate the Bourdieu-inspired claim that professors may be liberal because they are more likely tocome from highly educated families that encourage their children to value cultural overeconomic capital.

7 Bourdieu (1991) gives a different account of these dynamics when considering the case of pre-NaziGermany.

Theor Soc (2012) 41:127168 139139

We also consider three other class-related explanations. First, as Lamont (1987)and others have argued, new class workers in the government and nonprofit sectormay have direct economic stakes in liberalism inasmuch as it entails greater statesupport for education, social services, the arts, and so on. Accordingly, we evaluatehow much of the political difference between professors and others is a functionof the fact that many professors are public employees.8 Second, a variety ofapproaches to class analysis point to the uniqueness of selfemployed workers,theorizing that they hold key positions in the class structure and have distinctiveeconomic interests. In Goldthorpes view (2007), for example, the self-employed areunique because the maintenance of their class positions may well appear bestguaranteed not through educational attainment. but rather through their directinheritance of family businesses as going concerns or through the intergenerationaltransmission of capital sufficient to enable them to start up enterprises of their own(pp. 173174). These interests should lead the self-employed to favor low inheritanceand other taxes and on these grounds to favor conservative economic ideology. Sincethere are obviously few self-employed professors, we examine whether their lowlevels of self-employment help to explain their liberalism. Finally, we test whetherprofessors may be more liberal because they tend to have relatively high rates ofunionization. While progressive views among workers should foster union growth,the causal direction is uncertain since large, active unions, long courted by theDemocratic Party, may nurture liberalism among workers.

Hypothesis 6: Intellectualism aside, professors have distinctive job values thatinfluence their politics In a recent paper, Summers (2007) argues that during and aftercollege young people bound for middle- and upper-middle-class lives face manychoices about what careers to pursue. Echoing themes advanced in another context byInglehart (1990), Summers suggests that a key distinction is between those who feelcomfortable with the world of commerce and see the main purpose of a job as beingto provide as much income as possible, and those who find business distasteful andput higher value on the meaningfulness of work, its social utility, and so on. Theformer will gravitate to careers in the for-profit sector; the latter may becometeachers, social workers, and professors. All else being equal, according to Summers,people with a preference for business and making money tend to hold more conser-vative views and support the Republican Party, which might explain the liberalism ofprofessors and other workers in the nonprofit sector. We test this hypothesis byexamining how much importance professors place on the meaningfulness of workas opposed to other aspects like remuneration, and whether this accounts in part fortheir liberal views.9

8 No measure is available in the GSS of whether respondents are located in the broader nonprofit sector.While a substantial number of American professors teach at private institutions and might benefit moreindirectly from the largess of the state, if the new class direct economic interest hypothesis were robust thehigh percentage of professors who are public employees should substantially reduce the political gapbetween professors and other Americans.9 Because of space and data constraints, we do not examine any hypotheses linking professorial liberalismto meso- or macro-level features of the American political-economic context, as would be suggested bycomparative class approaches to intellectuals and politics of the kind developed by Brym (1980), Karabel(1996), and Wright (1978).

140 Theor Soc (2012) 41:127168

Data and methods

Although most of the hypotheses enumerated above have been under considerationby sociologists for some time, there have been, once again, no attempts to test themagainst one another using multivariate data. In our initial thinking about this article,our plan was simply to provide such a test. Data limitations imposed a majorconstraint: we could think of no way to examine the hypotheses empirically exceptby using nationally-representative survey data containing a sufficiently large numberof professors and non-professors that the factors associated with the politics of eachgroup could be reliably identified. The dataset that best met this requirement was theGSS when pooled over the period 19742008. As mentioned previously, 326respondents in the GSS have non-missing values on the outcome variable we choseto use and occupational codes indicating that at the time of the interview they wereworking as professors or instructors in higher education. By contrast, datasets like theANES contain less than half the number of professors than the GSS, decreasingstatistical power and increasing the probability of spurious findings. For reasons wedescribe below, the GSS data are imperfect for our purposes. Yet they permit at least apreliminary empirical examination of claims long locked in the realm of theory.

To estimate the size of the political gap between professors and non-professors, wefocus on one of the most widely used measures of political orientation: self-identification along a continuum from liberal to conservative. Political sociologistshave sometimes been skeptical of measuring liberalism or conservatism in terms ofself-identification, in part because doing so might obscure the differences betweendifferent types of liberals and conservatives (e.g., between social and economicliberals). However, while the correlation between political self-identity and socialand political attitudes varies over time and across issue domains and populationsubgroups (Baldassarri and Gelman 2008), political self-identity represents, for mostAmericans, a unique assessment of their overall position in political space. Politicalpsychologists have demonstrated that, at least in the contemporary United States, self-identification in terms of liberalism or conservatism functions as a powerful cognitiveand affective structure (Jost 2006; Jost et al. 2009), and it has become an increasinglyimportant predictor of voting and other political behaviors. For reasons of parsimonywe therefore rely on self-identity as our main outcome measure. In the GSS, politicalself-identity is measured with a seven-point Likert item asking respondents to locatethemselves on a scale ranging from extremely liberal to extremely conservative. (Inthe analyses below we reverse the direction of the scale.) At the same time, we aresensitive to the concern that our models should have direct purchase on attitudes aswell. In our discussion of robustness we therefore briefly report the results ofalternative models using a variety of attitudinal outcomes.

Our hypothesis testing involves six sets of variables. For our first hypothesis onsociodemographic traits, we consider membership in the boomer cohort (defined bythe US Census as having a birth year between 1946 and 1964), the number ofchildren respondents have, being female, having lived in a big city at the age of 16,living at the time of the interview outside the South (defined as the East South Centraland South Atlantic U.S. Census regions), having lived in another country at the age of16, and being foreign born. For our second hypothesis on education, we considerwhether respondents have any degree beyond a bachelors degree. We operationalize

Theor Soc (2012) 41:127168 141141

intellectualism, our third hypothesis, with a scale measuring the number of wordsrespondents score correctly on a 10-word vocabulary test. We also consider thewillingness of respondents to tolerate the expression of controversial ideas. To doso, we created a summated, standardized scale of six items: answering yes or noto the questions of whether racists and militarists should be allowed to speak, teach,or have a book in a library. We chose these items over others from the Stouffertolerance scale, such as those pertaining to communists or homosexuals, sincethe latter may capture aspects of liberal ideology rather than tolerance for controver-sial ideas per se. Although consisting of only a few items, the scale exhibits a highlevel of internal consistency with a Cronbachs alpha of 0.822, above the standardbenchmark of 0.700 (Nunnally 1978). In addition, principal components factoranalysis indicates that the scale displays unidimensionality, loading primarily onone (unrotated) factor with an Eigenvalue of 2.685.

Our fourth hypothesis on religion examines whether respondents identify asJewish, not religiously affiliated, or members of non-theologically conservativeProtestant denominations according to the coding scheme developed by Smith(1990).10 Some sociologists of religion have questioned whether Smiths scheme isa reliable measure of Christian conservatism generally as opposed to more strictlyfundamentalist identity (Steensland et al. 2000). But alternative coding schemeshave come in for criticism as well (see Hackett and Lindsay 2008), and when weexperimented with one of thesethe scheme developed by Steensland and col-leagueswe observed patterns in the representation of conservative Protestants inacademe at odds with expectations based on prior research (e.g., Schmalzbauer2002). Accordingly, we rely on Smiths long-established measure (which, in onlyslightly modified form, forms the basis for Greeley and Houts [2006] definitive studyof Christian conservatives).

A core component of our fifth hypothesis, dealing with class position and mobi-lization, examines levels of cultural and economic capital. Here he follow the lead ofscholars of status inconsistency like Hope (1975), first standardizing years of educationand constant household income and creating a variable that is the sum of the twoameasure of respondents total level of cultural and economic capital; and secondcreating a variable that reflects the difference between standardized education andstandardized income. Both variables are included in our model so that we examine theeffect of status inconsistency holding constant overall position in the class structure.11

To avoid multicollinearity with the advanced degree variable, however, we do notinclude it and the Bourdieu status inconsistency variable in the same model.

Hypothesis 5 also considers whether professorial liberalism may be a function ofhigher rates of parental advanced degree holding. To measure differentiation from thebusiness classes, we use a dummy variable indicating whether or not respondentshave hardly any confidence in major American businesses. Also related to classposition and mobilization, we consider whether respondents are public employees,self-employed, and union members. Our sixth hypothesis focuses on job values,

10 Our dummy variable here codes for whether respondents are Protestants and members of liberal ormoderate denominations.11 We are grateful to Omar Lizardo for suggesting this approach.

142 Theor Soc (2012) 41:127168

which we measure using an item asking respondents if meaningful work is the mostimportant thing to them in a job versus its income potential, degree of job security, orwhether it affords opportunities for promotion.

We pooled the GSS across years to ensure that the largest possible number ofprofessors would be included in the sample, not because we are examining changeover time. A common objection we have heard to pooling in this way is that, over thecourse of the 34 years covered by our data, the meanings of liberalism and to an evenlarger extent conservatism changed dramatically, rendering incomparable politicalself-identification at different time points. Undoubtedly such changes have occurred.(For a discussion of the conservative side of the ledger, see Gross et al. 2011.) Butonce again, our analysis is not intended to be dynamic, and, accordingly, westatistically control for shifts in the meaning of our outcome variable, along withother time-dependent exogenous factors that might otherwise bias results, by includ-ing year dummies in all models.

Not all of the items we use to construct our variables were asked during every GSSyear. To deal with the problem of missing data we multiply imputed values usingRubins (1987) rules, fitting our models on each of m020 imputations and runningthe appropriate post-imputation averaging procedures to obtain the final point esti-mates and variances.12 Following the suggestion of Von Hippel (2007), we includedthe outcome variable in our imputation models, and then dropped cases that wereoriginally missing on the outcome variable when conducting our analysis. Multipleimputation is one of the most effective ways of handling missing data, and is lesslikely to lead to bias than other approaches, including mean substitution, dummyvariable adjustment, and hot-deck imputation (Allison 2001). Nevertheless, since notall questions were asked each year, several variablesnamely, our measures ofverbal ability, openness to controversial ideas, union membership, work in the publicsector, confidence in business, and job valueshad levels of missingness above 25%and in a few instances above 60%. Notwithstanding the reliability of multipleimputation, and the fact that we retained a very large overall sample size of 44,029cases, we urge some caution in interpreting the coefficients for these variables.13

To assess our hypotheses we use a variant of the counterfactual decompositiontechnique developed by Oaxaca (1973) and Blinder (1973). Commonly used byeconomists to isolate the contributions of individual predictors to the gap on someoutcome (e.g., wages) between dichotomous groups (e.g., men and women)andused with increasing frequency in sociology as wellin this study we use thetechnique to assess how much of the politics gap between professors and non-professors is accounted for by each of our variables. We compute results from thedecomposition based on reference coefficients from a pooled model over both groupsand, following the recommendation by Neumark (1988), include the group indicator

12 For those interested in replication, the specifications of our imputation modelincluding the randomseedare available from the authors.13 As a check on the reliability of the imputation procedure, we reran all the models reported belowexcluding imputed cases for the six variables with high levels of missingness. Results were substantivelyunchanged.

Theor Soc (2012) 41:127168 143143

as an additional control variable. We adjusted decomposition results using thestandard GSS weight variable.

We noted above that while the GSS is the best available dataset for our purposes, itis still imperfect. One way in which this is so is that there are some hypotheses, suchas the one on intellectualism, that we would have preferred to test using more directlyrelevant variables. Of potentially greater concern to some, however, is the question ofwhat kind of professors are included in the GSS sample. Our professorial respondentsstated that they were currently working as professors or instructors in higher educa-tion, but the institutional landscape of American higher education is incrediblydiverse, ranging from community colleges and proprietary schools to liberal artscolleges and elite research universities. We have no information about the institu-tional location of those in our professors subsample, but all signs point to thelikelihood that they reflect the full diversity of the field, and that the subsample ismore or less representative of American academe as a whole.14 For scholars drawn tothe topic of professorial liberalism because of its implications for class politics, thiscould be seen as a problem. As we suggested above, much of the traditionalpuzzlement among sociologists about the liberalism of professors revolves aroundthe question of why members of a high status occupational group would favorredistributionist policies that could erode their social standing. Social scientistsmotivated by this question might prefer an analysis focused exclusively on professorsat elite institutions, reasoning that professors at lower tier schools are not in equiv-alent social positions and hence do not have as anomalous politics. But there aremany other valid reasons to be interested in professorial liberalism. For example, onecould think the topic important because of an interest in how professors liberal viewson social issues might influence their teaching, the broader political environment oncampus, the distribution of political sentiment among college students, and the workdone by professors at all levels in the civil sphere. Since 72% of undergraduates in theUnited States are enrolled in nondoctoral-granting institutions, a broad sample ofprofessors of the kind reflected in the GSS data would seem entirely appropriategiven such a focus. We do acknowledge, however, that while we have found noreason to doubt that our findings apply to scholars at elite schools specifically, it is alimitation of our data that they do not permit us to assess the fit of the models forscholars at different types of institutions.

Results

Exactly how large is the political gap between professors and non-professors? Overthe 34 years of our study period professors are on average 0.571 points more liberal

14 For example, the occupational coding is specific enough to tell us that of the 326 GSS respondents whoare professors, 11% are social scientists, 25% are humanists or in the fine arts, and 8% are physical orbiological scientists, with the remainder located in other fields. Extrapolating from nationally-representativesurveys of the American faculty conducted in 1975, 1984, 1988, 1992, and 1998, we would expect 12% ofrespondents to be social scientists, 24% to be humanists or in the fine arts, and 20% to be physical orbiological scientists (Schuster and Finkelstein 2006, p. 447). Similarly, as we report below, the distributionof political self-identity within the professors subsample in the most recent period is almost exactly thesame as that found in specialized surveys of the professoriate covering professors in all types of institutions.

144 Theor Soc (2012) 41:127168

than non-professors on the seven-point political self-identification scale.15 Although0.571 might seem small, in relative terms the difference is substantial. One way to putthe gap in perspective is to compare it to differences between other groups. Thedifference in political self-identity between professors and other Americans is over1.5 times that between blacks and whites (0.352), over twice as great as that betweenthe bottom and top deciles in constant household income (0.251), and more thanseven times larger than that between women and men (0.078). Examining the politicsof different occupations in the United States from 19742008 using the 126 occupa-tion classification scheme of Weeden and Grusky (2005), we find that the professo-riate is tied with the category authors and journalists for the highest proportion ofliberals. For the period 20002008, it is the single most liberal occupation.16

What might account for the size of the politics gap between professors andother Americans? As a first step toward assessing our hypotheses, we examinehow professors score on the predictor variables. Table 1 compares their values tothose of non-professors. As expected, professors tend to have fewer children, aremore likely to have resided in an urban area or foreign country while growing up, andare more likely to be foreign born. However, despite recent shifts women remainunderrepresented in academia for the period as a whole. Professors are very similar toother Americans in terms of membership in the boomer cohort and residence outsidethe South.

More striking differences are found with respect to education, intellectualism, andreligion. Compared to other Americans, professors are 12 times more likely to haveadvanced degrees. In addition, they have above-average verbal ability and exhibitconsiderably higher levels of tolerance for controversial ideas. As well, Jews andthose with no religious affiliation are overrepresented in academia, while conserva-tive Protestants are underrepresented.

Professors are also fairly distinctive when it comes to class. Compared to otherAmericans, they tend to come from better-educated families and exhibit a greaterdisparity between their education levels and their income, in line with Bourdieus

15 This is the size of the gap for which the regression decompositions attempt to account. The numbershown in Table 1 is slightly smaller because the data there are unweighted.16 For the period 20002008, 43% of professors in the GSS sample stated their political identity asextremely liberal or liberal, and 9% as extremely conservative or conservative. These numbersare extremely close to the 44% versus 9% figure numbers from Gross and Simmonss (2007) survey,enhancing our confidence in the reliability of GSS data for studying the politics of professors. Consistentwith the observation made by a reviewer that a mean score of 4.45 on the 7 point GSS political self-identification scale for the 19742008 period is far from extreme, Gross and Simmons argue that there is asizable and often ignored moderate bloc in academe. This is an important point often ignored byconservative critics, but it does not change the fact that in relative terms the American professoriate isclearly to the left. A further question arises about these descriptive statistics, however. Another reviewerobserved that the standard deviation on the political self-identification variable is considerably higher forprofessors than for non-professors, and expressed concern that this could signal that a small number ofprofessors in our sample from elite institutionswho have long been found to have more liberal viewscould be driving up the mean for all professors, making the professoriate appear more liberal than it actuallyis. However, Gross and Simmons found that while there is an institutional status gradient in professorialpolitics, professors in lower tier institutionsin community colleges and 4-year, BA-granting schoolsremain much more liberal than most Americans. Indeed, if the contemporary professoriate consistedexclusively of professors from such schools, it would still be one of the most liberal major occupations,at 38% liberal. We therefore think the alternative interpretation of the large standard deviation is correct:that it reflects the greater variance typically associated with small subsamples (see Gelman and Hill 2007).

Theor Soc (2012) 41:127168 145145

theory. They are much more likely than other Americans to be employed in the publicsector, and much less likely to be self-employed. Professors are somewhat morelikely to belong to unions than are other Americans, but, contrary to the expectationsof some new class theorists, they are no more likely to say they lack confidence in big

Table 1 Univariate Statistics, Professors vs. Non-Professors, 19742008

Professors Non-Professors

Variable Mean Min. Max. Std. Dev. Mean Min. Max. Std. Dev.

Outcome

Political Self-Identification 4.45 1.00 7.00 1.51 3.89 1.00 7.00 1.36

Hypothesis 1: Demographics

Baby Boomer 0.37 0.00 1.00 0.48 0.39 0.00 1.00 0.49

Number of Children 1.44 0.00 6.00 1.46 1.92 0.00 8.00 1.77

Lived in an Urban Area at Age 16 0.52 0.00 1.00 0.50 0.41 0.00 1.00 0.49

Female 0.41 0.00 1.00 0.49 0.55 0.00 1.00 0.50

Lives Outside the South 0.76 0.00 1.00 0.42 0.74 0.00 1.00 0.44

Lived in a Foreign Country at Age 16 0.13 0.00 1.00 0.34 0.05 0.00 1.00 0.23

Born in a Foreign Country 0.13 0.00 1.00 0.33 0.08 0.00 1.00 0.27

Hypothesis 2: Educational Attainment

Graduate Degree 0.72 0.00 1.00 0.45 0.06 0.00 1.00 0.24

Hypothesis 3: Intellectualism

Verbal Ability 8.04 0.00 10.00 1.93 6.01 0.00 1.00 2.14

Tolerance Scale 0.47 2.05 2.91 0.58 0.01 3.34 4.03 0.73Hypothesis 4: Religious Affiliation

Jewish 0.06 0.00 1.00 0.24 0.02 0.00 1.00 0.14

No Religious Affiliation 0.20 0.00 1.00 0.40 0.10 0.00 1.00 0.30

Protestant, Non-ConservativeChristian

0.50 0.00 1.00 0.50 0.40 0.00 1.00 0.49

Hypothesis 5: Class Position and Mobilization

Mother with a Graduate Degree 0.05 0.00 1.00 0.22 0.02 0.00 1.00 0.15

Father with a Graduate Degree 0.18 0.00 1.00 0.39 0.05 0.00 1.00 0.22

Total Composition of Capital(Bourdieu)

2.10 4.75 6.83 1.40 0.05 7.32 7.97 1.61

Relative Composition of Capital(Bourdieu)

1.20 5.58 5.89 1.12 0.02 5.52 6.68 1.09

Union Member 0.18 0.00 1.00 0.39 0.13 0.00 1.00 0.34

Public Sector Employee 0.50 0.00 1.00 0.50 0.17 0.00 1.00 0.38

Self-Employed 0.03 0.00 1.00 0.16 0.12 0.00 1.00 0.32

Low Confidence in Business 0.14 0.00 1.00 0.35 0.14 0.00 1.00 0.35

Hypothesis 6: Job Values

MeaningfulWork is Most Important 0.80 0.00 1.00 0.40 0.57 0.00 1.00 0.49

GSS, 19742008. N0326 professors; N043,703 non-professors. Univariate statistics are calculated overm020 multiple imputations. Sampling weight not shown.

146 Theor Soc (2012) 41:127168

business.17 Finally, although 80% of professors report that meaningful work is amongthe most important qualities they look for in a job, nearly 60% of other Americansfeel the same.

Although informative, the statistics in Table 1 are unadjusted and accordinglycannot indicate to what extent these variables predict the gap in political self-identification between professors and other Americans. The decompositions inTable 2 provide estimates of the relative contribution of each variable in explainingthe political identification gap. Because of high multicollinearity, Model 1 includesthe advanced degree variable but excludes variables related to status inconsistency,while Model 2 includes status inconsistency but excludes advanced degree. Overallthe variables in Model 1 account for about 43.5% of the politics gap betweenprofessors and non-professors, whereas the variables in Model 2 account for about27.8%.

Beginning with an assessment of Hypothesis 1, the decompositions provide onlymodest support for the idea that sociodemographic factors help account for profes-sors liberalism. Adjusting for other predictors, the only sociodemographic factorsthat account for more than one percent of the political gap are number of children andliving in an urban area at age 16. Together these factors explain about 5.6% of thegap. Gender increases the size of the gapas a main effectgiven the under-representation of women in academe. There is greater evidence for Hypothesis 2:after controlling for other predictors, graduate degree holding among professorsaccounts for about 19.8% of the gap. As for Hypothesis 3, dealing with intel-lectualism, the decompositions show that the greater average verbal ability ofprofessors does nothing to account for the political gap between professors andother Americans. The tolerance scale, our other measure of intellectualism,accounts for about 10.3% of the gap. Turning to Hypothesis 4, the data supportthe theory that religious differences between professors and non-professors helpto account for political differences. The coefficients for all three variablesexamined here are statistically significantwith lack of religious affiliationbeing the most powerfuland together account for about 14.7% of the gap. ForHypothesis 5, on class, there is mixed support. The coefficient for the Bourdieustatus inconsistency variable, denoted in the table as the relative composition ofcapital, is large and statistically significant, accounting for about 13.0% of the gap.While the coefficients for most of the other class variables are statistically signifi-cant, together they account for only a small proportion of the gap in political self-identification. All else being equal, preference for a meaningful jobthe variable weuse to test Hypothesis 6increases the political gap between professors andothers.18

17 This is consistent with the findings of Gross and Simmons (2007), who report that, on average,professors are more likely than other Americans to believe that business corporations make a fair andreasonable profit. Although Gross and Simmons find that professors do tend to favor government action toreduce income inequality, as we note below, they also find that the majority are not consistently progres-sive in their economic views. Again, it is the social liberalism of professors that stands out.18 Alternative versions of the model using different specifications of this set of GSS variablesforexample, whether the respondent thinks high income is the most important aspect of a jobfound nosubstantively meaningful associations.

Theor Soc (2012) 41:127168 147147

Table 2 Oaxa-Blinder Twofold Pooled Decomposition of Political Identity for Professors and Non-Professors, 19742008

Mean Political Identity of Professors 4.4521

Mean Political Identity of Non-Professors 3.8813

Professors vs. Non-Professors Gap 0.5708

Model 1 Model 2

Hypothesis 1: Demographics Coefficients % of GapExplained

Coefficients % of GapExplained

Baby Boomer 0.0015* 2.26% 0.0018* 0.32%(0.0006) (0.0007)

Number of Children 0.0204*** 3.57% 0.0189*** 3.31%

(0.0009) (0.0009)

Lived in an Urban Area at Age 16 0.0114*** 2.00% 0.0123*** 2.15%

(0.0008) (0.0008)

Female 0.0207*** 3.63% 0.0170*** 2.98%(0.0010) (0.0009)

Lives Outside the South 0.0018*** 0.32% 0.0017*** 0.30%

(0.0004) (0.0003)

Lived in a Foreign Country at Age 16 0.0015* 0.26% 0.0029** 0.51%

(0.0008) (0.0008)

Born in a Foreign Country 0.0030*** 0.53% 0.0034*** 0.60%

(0.0005) (0.0006)

Hypothesis 2: Educational Attainment

Graduate Degree 0.1129*** 19.78%

(0.0045)

Hypothesis 3: Intellectualism

Verbal Ability 0.0264*** 4.63% 0.0057** 1.00%(0.0016) (0.0017)

Tolerance Scale 0.0588*** 10.30% 0.0662*** 11.60%

(0.0014) (0.0015)

Hypothesis 4: Religious Affiliation

Jewish 0.0212*** 3.71% 0.0231*** 4.02%

(0.0016) (0.0017)

No Religious Affiliation 0.0463*** 8.11% 0.0439*** 7.69%

(0.0025) (0.0024)

Protestant, Non-Conservative Christian 0.0164*** 2.87% 0.0171*** 3.00%

(0.0011) (0.0012)

Hypothesis 5: Class Position and Mobilization

Mother with a Graduate Degree 0.0045*** 0.79% 0.0050*** 0.88%

(0.0006) (0.0006)

Father with a Graduate Degree 0.0036** 0.63% 0.0069*** 1.21%

(0.0010) (0.0010)

Total Composition of Capital (Bourdieu) 0.1125*** 19.71%(0.0025)

148 Theor Soc (2012) 41:127168

Sensitivity analysis

Recently sociologists have demonstrated that differences in model specification candramatically alter the results of quantitative research (Western 1996; Young 2009). Toassess the robustness of our findings we considered several alternative ways ofspecifying our models. In particular, we examined how results changed after reducingthe set of conditioning information, predicting a binary version of the outcomevariable, including various interaction terms, and using alternative outcomemeasures.

First, we sought to determine whether the coefficients for our main findings weresensitive to the particular set of conditioning information we included in our models.Thus, in two other models (not shown) we retained only those variables (besides thetime dummies) accounting for at least 2% of the political gap in Models 1 and 2,respectively. None of the coefficients changed direction or size to a substantialdegree. Moreover, for both models the total amount of the gap explained was largelyunchanged.

Second, we checked for robustness by using a different version of our outcomevariable. Specifically, we wondered whether some of the observed political gapbetween professors and non-professors might be a function of the fact that, giventheir high levels of education, professors could be less likely than others to locate

Table 2 (continued)

Mean Political Identity of Professors 4.4521

Mean Political Identity of Non-Professors 3.8813

Professors vs. Non-Professors Gap 0.5708

Model 1 Model 2

Relative Composition of Capital (Bourdieu) 0.0741*** 12.98%

(0.0020)

Union Member 0.0059*** 1.03% 0.0077*** 1.35%

(0.0006) (0.0008)

Public Sector Employee 0.0076*** 1.33% 0.0141*** 2.47%

(0.0013) (0.0013)

Self-Employed 0.0103*** 1.80% 0.0078*** 1.37%

(0.0005) (0.0005)

Low Confidence in Business 0.0013 0.54% 0.0011 0.19%(0.0010) (0.0008)

Hypothesis 7: Job Values

Meaningful Work is Most Important 0.0218*** 3.82% 0.0181*** 3.17%(0.0009) (0.0008)

Time Dummies 0.0121 2.12% 0.0069 1.21%All Covariates 0.2467 43.22% 0.1535 26.89%

GSS, 19742008. N0326 professors; N043,703 non-professors. Point estimates are calculated over m020separate imputations; standard errors are unadjusted. P

themselves near the midpoint on a politics scale. To address this possibility, wedichotomized the outcome variable so that it distinguished liberals from everyoneelse and reran Models 1 and 2 using Fairlie decomposition, which allows fornonlinear outcomes (see Fairlie 2005). The percentage of the politics gap explainedincreased somewhat in both models. Had systematic tendencies toward the extremesin survey response among professors biased our models, we would have expected thepercentage of the explained gap to decrease.

Third, we examined how results changed when we included various interactionterms. Of the interaction terms we examined, the only ones that were consistentlylarge and statistically significant involved advanced degree holding. Due to problemsof multicollinearity and interpretability of the coefficients, we could not includemultiple interaction terms with advanced degree holding in the same model. Table 3therefore summarizes the results of a number of models with interaction terms runserially. The table shows that several of the hypotheses that received only minorsupport from Model 1 receive much stronger support when considered in conjunctionwith advanced degree holding. In itself, the number of women in academe doesnothing to explain the politics gap, but the number of women with advanced degreesexplains 10.5% of the gap, which could only be because women with advanceddegrees, who compose nearly all women in academe, tend to be more liberal thanmen with equivalent educational credentials. Likewise, verbal ability does not reducethe gap, but verbal ability combined with advanced degree holding accounts for aremarkable 59.4% of the gap. Tolerance combined with advanced degree holdingaccounts for 19% of the gap. There are also interesting interactions with religion.Americans who are non-religiously affiliated and have advanced degrees, and non-conservative Protestants with advanced degrees, are more liberal than advanceddegree holders with other kinds of religious profiles, and accordingly both interactionterms decrease the politics gap between professors and non-professors. Although wethus find interaction terms with several variables that are large and statisticallysignificantand that demand theoretical interpretationin no instance did the inclu-sion of an interaction term with education meaningfully alter any of the non-interacting coefficients.

Finally, we considered how sensitive our results were to alternative measures ofpolitics. We reran Model 1 using as our outcome measures several social attitudesscales: one measuring attitudes toward sexuality, another abortion, a third on genderroles, and a fourth on race.19 We also considered an economic attitudes scale.Consistent with the findings of Baldassarri and Gelman (2008), in the generalpopulation these scales were correlated only moderately with political self-identity(with the strongest correlations for economic attitudes and attitudes on moral issueslike sexuality and abortion), indicating relatively low levels of attitudinal constraint inConverses (1964) classic sense of the term. Among professors the correlations wereabout twice as high, but in no case exceeded 0.43. Accordingly, we were unsure how

19 Details on these scales are available from the authors. All indices are standardized, summated scalesexhibiting desirable properties of parsimony (i.e., few items in the scale), reliability (i.e., high values ofCronbachs alpha), and unidimensionality (i.e., loading primarily on one factor).

150 Theor Soc (2012) 41:127168

well our model would explain the gap in attitudes between professors and non-professors. It turns out that, for most of the scales, the model accounted for a largeproportion of sizable gaps in attitudes: specifically, 83% of the gap on abortionattitudes, 76% of the gap on womens roles, 60% of the gap on sexuality, and 56%of the gap on views of race. On abortion, womens roles, and sexuality, advanceddegree holding and tolerance for controversial views were the main explanatoryvariables. On race, the key variables were advanced degree holding and ourvocabulary measure. In all these models advanced degree holding accounted forbetween 2025% of the attitudes gap between professors and non-professors,reflecting a coefficient roughly similar in relative magnitude to that in ourpolitical self-identity model. Religion was a lesser, though still important, com-ponent in several of these models, with higher rates of religious non-affiliationamong professors accounting for about 14% of the gap on abortion and 8% ofthe gap on sexuality. In light of previous research suggesting that professors aremore liberal on social than economic matters, and that their economic liberalismis selective, we did not expect to find a major gap registered by our economicattitudes scale. In fact, according to the scalewhich should be interpretedcautiously given missing datathere is no such gap, which means there isnothing to explain.20 For the others, however, the upshot of our analysis, largelyconsistent with our results using political self-identification as the outcome, is thateducation, intellectualism, and to a lesser extent religion are the key variablesaccounting for professors distinctive social and political views.

Toward a new theory of professorial liberalism

Although the results from the regression decompositions are large, statisticallysignificant, and robust to a number of variations in model specification, in reviewingthe hypotheses that had received empirical support we could not help but feel a littledissatisfied. First, for many of the hypotheses a number of different mechanismscould be driving the associations we observed. For example, concerning education, itcould be that advanced degree holders tend to be more liberal than others becausecognitive upgrading occurs in graduate or professional school that leads individualsto question conservative ideology or the values that underlie it. Alternatively, better-educated Americans might embrace liberalism to distance themselves culturally fromother status groups (although not presumably from the business classes, since ourmodels include a measure of such differentiation). While our results mark an advanceby distinguishing between hypotheses for which there is and is not systematic

20 We suspect that missingness, which we could obviously not correct for on outcome measures usingmultiple imputation, may account for the fact that we found no economic attitudes differences betweenprofessors and non-professors in the GSS, where other research would lead us to expect at least some. Forexample, for one of the items in our economic attitudes scale, which asks about the value of governmentefforts to reduce income differences, 44% of our professorial cases have missing data. Gross and Simmons(2007), asking the same question verbatim, found a 17.8 percentage point gap in liberal responses ascompared to general population data from the 2006 GSS.

Theor Soc (2012) 41:127168 151151

Table 3 Interactions with Graduate Degree using Oaxa-Blinder Twofold Pooled Decompositions ofPolitical Identity for Professors and Non-Professors, 19742008

Mean Political Identity of Professors 4.4521

Mean Political Identity of Non-Professors 3.8813

Professors vs. Non-Professors Gap 0.5708

Interaction with Female Coefficients % of Gap Explained

Graduate Degree 0.0344*** 5.33%

(0.0058)

Female 0.0182*** 3.26%(0.0009)

(Graduate Degree) (Female) 0.0604*** 10.51%

(0.0032)

Interaction with Verbal Ability

Graduate Degree 0.2069*** 36.51%(0.0182)

Verbal Ability 0.0324*** 6.55%(0.0016)

(Graduate Degree) (Verbal Ability) 0.3428*** 59.37%

(0.0192)Interaction with Tolerance Scale

Graduate Degree 0.0603*** 4.85%

(0.0051)

Tolerance Scale 0.0540*** 13.02%

(0.0013)

(Graduate Degree) (Tolerance Scale) 0.0681*** 18.99%

(0.0037)

Interaction with Public Sector Employee

Graduate Degree 0.1248*** 20.97%

(0.0055)

Public Sector Employee 0.0094*** 2.05%

(0.0014)

(Graduate Degree) (Public Sector Employee) 0.0181*** 3.05%(0.0048)

Interaction with Baby Boomer

Graduate Degree 0.1031*** 16.75%

(0.0059)

Baby Boomer 0.0014* 0.23%(0.0006)

(Graduate Degree) (Baby Boomer) 0.0074** 1.70%

(0.0029)

Interaction with Jewish

Graduate Degree 0.1076*** 18.08%

(0.0046)

Jewish 0.0199*** 3.42%

152 Theor Soc (2012) 41:127168

empirical evidence, we found it problematic that they did not more clearlyspecify the underlying causal processes. Second, at least one of the hypothesesthat received empirical supportthe intellectualism hypothesisappeared to haveconceptual problems, and it was unclear what it meant that the hypothesis hadbeen supported. Third, since a number of the hypotheses we examined receivedpartial support, and since we also uncovered a number of intriguing interactionswith advanced degree holding, the question aroseunanswerable by existingtheoretical approachesof how all the hypotheses, and the mechanisms under-lying them, should be thought of together. Fourth and finally, on political self-identity (less so social attitudes) our models left a sizable portion of the politicsgap between professors and other Americans unexplained.

Recognition of both the strengths and limitations of our study