7/30/2019 Althaus_1998

1/15

Information Effects in Collective Preferences

Author(s): Scott L. AlthausSource: The American Political Science Review, Vol. 92, No. 3 (Sep., 1998), pp. 545-558Published by: American Political Science AssociationStable URL: http://www.jstor.org/stable/2585480 .

Accessed: 03/01/2011 09:46

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at .http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless

you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you

may use content in the JSTOR archive only for your personal, non-commercial use.

Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at .http://www.jstor.org/action/showPublisher?publisherCode=apsa. .

Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed

page of such transmission.

JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of

content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms

of scholarship. For more information about JSTOR, please contact support@jstor.org.

American Political Science Association is collaborating with JSTOR to digitize, preserve and extend access to

The American Political Science Review.

http://www.jstor.org

http://www.jstor.org/action/showPublisher?publisherCode=apsahttp://www.jstor.org/stable/2585480?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/action/showPublisher?publisherCode=apsahttp://www.jstor.org/action/showPublisher?publisherCode=apsahttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/2585480?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=apsa

7/30/2019 Althaus_1998

2/15

AmericanPoliticalScience Review Vol. 92, No. 3 September1998

InformationffectsnCollectivereferencesSCOTT L. ALTHAUS Universityof Illinois at Urbana-Champaignontrary o much of the literatureon collective opinion, I find that the low levels and uneven socialdistributionof political knowledge n the mass public oftencause opinion surveys o misrepresent hemix of voices in a society. To assess the bias introduced by information effects,I compare '{fullyinformed" collectivepreferencessimulated from actual surveydata to collectivepreferencesrevealed in theoriginal data. Analysis of policy questions from the 1988 and 1992 American National Election Studiesshows that group differences in knowledge, along with the public's modest average level of politicalknowledge,can cause significantdistortions n measures of collective opinion. The masspublic may appearmoreprogressiveon some issues and more conservativeon othersthan would be the case if all citizenswereequally well informed. To the extent that opinion polls influence democraticpolitics, this suggests thatinformationeffectscan impairthe responsivenessof governmentsto their citizens.

A number of path-breaking studies in recent yearshave suggested that the mass public's inattentionto politics may have less bearing on the qualityof its collective opinions than previously thought (e.g.,Converse 1990; Page and Shapiro 1992; Popkin 1991;Sniderman, Brody, and Tetlock 1991; Wittman 1995).These studies emphasize that while most individualstend to be ill informed about the political world, theavailability of heuristic shortcuts and the filtering pro-cess of statistical aggregation may help compensate forthis lack of knowledge in measures of collective opin-ion, such as election results or opinion surveys. If thisline of thinking is correct, then we can conclude withPage and Shapiro (1992, 385) that opinion surveysprovide a "good deal of coherent guidance about

policy." If the mass public is unable to compensateeffectively for its lack of political knowledge, then theuse of surveys and other measures of collective opinionas inputs to the political process may be rightly ques-tioned.In this article I extend recent work by Delli Carpiniand Keeter (1996) and Bartels (1996) on the measure-ment of information effects in collective preferences.Contrary to much of the literature on collective opin-ion, this study finds that the low levels and unevensocial distribution of political knowledge in the masspublic often cause opinion surveys to misrepresent themix of voices in a society. Correcting for informationasymmetries reveals that many collective policy prefer-ences would look quite different if all citizens wereequally well informed about politics.Because knowledgeable respondents are better ableto form opinions consistent with their political predis-Scott L. Althaus is Assistant Professor of Speech Communicationand Assistant Professor of Political Science, University of Illinois atUrbana-Champaign, Urbana, IL 61801-3694 (salthaus@uiuc.edu).Earlier versions of this article were presented at the 1996 annualmeeting of the American Political Science Association and the 1997annual meeting of the International Communication Association. Iam grateful to the ICPSR and the University of Michigan Center forPolitical Studies for making available the NES data. I am indebted toBen Page, Susan Herbst, Jane Mansbridge, Patricia Conley, ScottKeeter, Michael Delli Carpini, Dennis Chong, Paul Gronke, andSteven Klepper for their helpful comments and suggestions over thecourse of this project. All analyses were conducted using SPSS forWindows 7.0.

positions (Bennett 1995; Converse 1964; Delli Carpiniand Keeter 1996; Lockerbie 1991; Stimson 1975; Zaller1992), and because they tend to give opinions morefrequently than other people (Althaus 1996a; DelliCarpini and Keeter 1996, 230-1; Krosnick and Milburn1990), the demographic characteristics of well-in-formed people-who tend to be more affluent, older,white, and male compared to the ill informed-cancause collective preferences to reflect disproportion-ately the opinions of some groups more than others.These two dynamics can create information effects inmeasures of public opinion. By information effect, Imean a bias in the shape of collective opinion causedby the low levels and uneven social distribution ofpolitical knowledge in a population. While others haveexamined the individual-level effects of political knowl-edge on response stability (e.g., Delli Carpini andKeeter 1996, 231-4; Feldman 1989, following Converse1964) and the role of political knowledge as a linkbetween political predispositions and policy or votingpreferences (e.g., Bennett 1995; Delli Carpini andKeeter 1996, chapter 6; Zaller 1992), the focus here ismore narrowly on the macrolevel impact of theseeffects on measures of collective opinion.Building on the simulation approach developed in-dependently by Delli Carpini and Keeter (1996) andBartels (1996), this article examines the substantiveimpact of information effects in policy preference datafrom the American National Election Studies (NES).Simulating "fully informed" collective preferencesfrom actual survey data shows that group differences inknowledge, along with the public's rather modest aver-age level of political information, can cause significantdistortions in measures of collective opinion. Moreimportant, the direction and magnitude of these dis-tortions fall into predictable patterns. The unevensocial distribution of political knowledge causes themass public consistently to appear more progressive onsome issues and more conservative on others thanmight be the case if all citizens were equally wellinformed about politics. To the extent that opinionpolls influence democratic politics, this findingsuggeststhat information effects may impair the responsivenessof governments to their citizens.

545

7/30/2019 Althaus_1998

3/15

Information Effects in Collective Preferences September 1998HOW POLITICALKNOWLEDGE AFFECTSCOLLECTIVEOPINIONSurveyaftersurveyhas shown hatcitizensareoftenata loss to relate basicfacts aboutthe players, ssues, andrules of the game that structureAmericanpolitical ife(Delli Carpini and Keeter 1996). The extent of thisignorance ed several early and influentialstudies tosuggestthat the public's political opinions are oftenfickle and not to be trusted (Almond 1950; Berelson,Lazarsfeld,and McPhee 1954; Converse1964; 1970).Yet, in recent years this pessimisticview of publicopinionhasbeenchallengedon several ronts.Thefirstchallenge stems from work in cognitive psychologyshowing hatpeoplewho are ill informedaboutpublicaffairscan nonetheless form opinionsconsistentwiththeir political predispositions.They can do this bytaking cues from trusted political elites aboutwhichpolicies they should preferandby harnessing varietyof heuristicstrategies o deduce theirpoliticalprefer-ences, thus avoiding he need to inferpreferences romfactualbits of knowledge tored in long-termmemory(e.g., Ferejohn and Kuklinski 1990; Lupia 1994;Mondak 1994; Popkin 1991; Smithand Squire 1990;Sniderman,Brody, and Tetlock 1991; Zaller 1992).Fromthis perspective, he public's ow levels of infor-mationmaynot be a significant roblembecausemanypeople apparentlycan compensate for their lack ofknowledgewith information hortcuts.The second challenge comes from a line of worksuggestingthat aggregate opinion may be able toreflectthe public's nterestseven whenthe opinionsofmost individuals re ill informed,ambivalent,ndiffer-ent,or inconsistente.g.,Converse1990;FerejohnandKuklinski1990; Page and Shapiro 1992). This viewsuggests that the process of aggregating opinionsshouldtend to cancel out the more or less randomopinions given by ill-informedrespondents.To theextent that this occurs,measuresof collectiveopinionshould reflect the nonrandomopinionsof knowledge-ablerespondents.A relatedargument temming romCondorcet's ury theorem suggestsa similarconclu-sion. Based entirely on statisticalprobabilities,thetheoremshowsthat, undercertainconditions,groupstend to providemoreinformeddecisions han individ-uals(Condorcet 1785]1972;GrofmanandOwen1986;Ladha 1992;Miller 1986;but see Austen-SmithandBanks1996).From this perspective, t is the aggrega-tion processitself whichgenerates meaningfulcollec-tive opinions.Furthersupportfor "revisionist"hallengescomesfromexperimentaltudiessuggestinghat the commonmethodsused to measure informationabout politicsmay actually est recallabilityrather hanknowledge-in-use. Knownas the "on-line" r "impression-driven"model of information rocessing Lodge,McGraw, ndStroh1989;Lodge, Steenbergen,and Brau1995),thisviewsuggests hatmanypeople process nformation tthe time they are exposedto it, updatetheir opinionaccordingly, and then quickly forget the informationitself while retaining he updated summary udgment.Thus, people may express nformedpreferenceseven

546

though they may be unable to recall the factual infor-mation used to shape those preferences. From thisperspective, the public's apparently low levels of polit-ical knowledge may be a red herring. Its judgments maybe more informed than they seem in light of theapparently poor performance of those citizens onknowledge tests.Despite the note of optimism sounded by theserevisionist arguments, empirical evidence in support oftheir claims is quite modest. A number of studies havedetailed how people can use on-line processing andvarious information shortcuts to make up for a lack offactual knowledge, but there is surprisingly little evi-dence that large numbers of people do use thesestrategies effectively, regularly,and across a wide rangeof situations and issues. More glaring still is the lack ofsupport for the collective rationality hypothesis. I amaware of only two attempts to test this idea empirically.One study (Bartels 1996) found that aggregationhelped voters act as if they were somewhat betterinformed than they actually were, but not as if theywere fully informed. The other (Althaus 1995) foundthat statistical aggregation had quite limited informa-tion-pooling qualities.Revisionist perspectives also tend to overlook animportant fact: Low information levels are only half theproblem. Just as important is the observation that somekinds of people tend to be better informed than others.Knowledge of politics is concentrated among thosewho are politically and socially advantaged. Collegegraduates and relatively affluent people tend to beconsistently well informed, while high school dropoutsand relatively poor people tend to be consistently illinformed (Delli Carpini and Keeter 1996; Neuman1986; Sigelman and Yanarella 1986). Political knowl-edge also is distributed unevenly among groups withdistinctive and potentially competing political interests.For instance, whites tend to be more informed thanblacks, men more than women, and older people morethan younger people (Bennett 1988; Delli Carpini andKeeter 1996; Neuman 1986; Sigelman and Yanarella1986).There are two ways that information asymmetriesamong groups can undermine representation in opin-ion surveys. The first is by affecting the demographiccorrespondence between a survey sample and thegroup of people who give substantiveresponses. Thosewho are poorly informed about politics tend to give"don't know" and "no opinion" responses at muchhigher rates than more knowledgeable people (Althaus1996a; Delli Carpiniand Keeter 1996, 230-1; Krosnickand Milburn 1990). This tendency leaves the group ofopinion givers disproportionately well educated, afflu-ent, male, white, and middle aged relative to thepopulation they are supposed to represent (Althaus1996b). Because these voices tend to be overrepre-sented in the ranks of opinion givers, the particularneeds, wants, and values expressed by some groupsmay come to be represented disproportionately incollective preferences.The second way that information symmetries ffectrepresentation s by influencing he qualityof opinions

7/30/2019 Althaus_1998

4/15

American Political Science Review Vol. 92, No. 3that respondents provide. Our ability to form prefer-ences consistent with our political predispositions isoften mediated by the quality and quantity of politicalinformation we can bring to bear on an issue. Theimportance of knowledge to the formation of policypreferences comes from the way that values are con-nected to attitudes through beliefs: beliefs about thestate of the world, cause-and-effect processes, whatgovernment is currently doing, and the likely outcomesof government actions (Delli Carpini and Keeter 1996,chapter 6; Downs 1957, 79-80). Some ill-informedpeople may believe a policy is the "correct" solutionwithout knowing whether it is consistent with theirpredispositions or whether it is the best way to achievea given end. Others may believe a policy serves theirinterests, while someone with more perfect or com-plete information can see that the policy is diametri-cally opposed to them. Because the well informed arelikely to have more accurate beliefs than the ill in-formed, they are more likely to express policy prefer-ences consistent with their political predispositions(Converse 1964; Lockerbie 1991; Stimson 1975; Zaller1992). As a result, the interests of respondents who arerelatively well informed may come to be more accu-rately reflected in measures of collective opinion. Inother words, such measures may reflect the needs,wants, and values of whites better than those of blacks,men better than women, and the rich better than thepoor.It would seem that the low levels and uneven socialdistribution of political knowledge may indeed have animportant bearing on the quality of surveyed publicopinion. The problem is how best to measure anypotential distortion in collective preferences broughtabout by information asymmetries. One must find away to estimate how the opinions people express insurveys might change if respondents were more com-pletely informed about the issues.MEASURING INFORMATIONEFFECTS INCOLLECTIVEPREFERENCESEver since Marx suggested that "false consciousness"distracts workers from their material interests, studentsof politics have grappled unsuccessfully with how todetermine whether people's interests are at odds withtheir opinions. Some, like Marx and Edmund Burkebefore him, argue that political interests are objectiveand can be identified for any group of people withoutregard to their stated preferences. Some, like JeremyBentham and John Stuart Mill, claim that interests aresubjective and thus inseparable from the expressedwishes of individuals speaking for themselves. But inlight of inherent problems with each of these defini-tions, more recent work has focused instead on inter-ests as "fully informed" or "enlightened" preferences(Bartels 1990; Connolly 1972; Dahl 1989, 180-1; DelliCarpini and Keeter 1996; Lippmann 1955, 42; Mans-bridge 1983, 24-6). In this perspective, as Jane Mans-bridge (1983, 25) puts it, interests are revealed in "thepreferences hat people would have f their nformationwereperfect, ncluding he knowledge heywouldhave

in retrospect if they had a chance to live out theconsequences of each choice before actually making adecision." By equating interests with hypothetical fullyinformed preferences, this perspective provides a use-ful approach for determining when the preferencesexpressed by individuals may be at odds with their"fully informed" needs, wants, and values.The difficulty lies in sorting out the best amongseveral possible ways of operationalizing this definitionof interests. Two traditional approaches have beenused to explore what collective opinion might look likeif opinion givers were relatively better informed thanthe general public. The first is to purge ill-informedrespondents from among the ranks of opinion giversthrough the use of filtering questions. While this is aparticularly blunt method-it ignores completely howindividual opinions might change with more politicalknowledge-it is nevertheless widely used by surveyorganizations to isolate "informed" public opinion.

The second approach manipulates the amount andquality of information available to people who expresspreferences of various kinds. Experimental methodsfor assessing the role of information in preferenceformation have been used extensively in social andcognitive psychology. These methods have also beenused to explore what fully informed collective policypreferences might look like. To date, the most ambi-tious use of such methods has been by James Fishkin(1991, 1995) in experiments with deliberative opinionpolls, which bring a random sample of ordinarycitizensto a central location where they are provided withdetailed policy information and an environment inwhich to discuss issues. In this way, argues Fishkin(1995, 171), "the deliberative poll can be thought of asan actual sample from a hypothetical society-thedeliberative and engaged society we do not have."Yet, even experimentation provides an unsatisfac-tory measure of fully informed collective preferences.It is possible in an experimental setting to providepeople who are normally oblivious of the politicalworld with information that they can use to formulatepreferences, but ill-informed subjects cannot beequipped with other important traits normally associ-ated with being well informed: the cognitive styles andinformation processing strategies characteristic of po-litically knowledgeable people (Fiske, Lau, and Smith1990;Krosnick and Milburn 1990;McGraw and Pinney1990; Sniderman, Brody, and Tetlock 1991), the knowl-edge stored in long-term memory that affects how newinformation is perceived and used to update attitudes(Delli Carpini and Keeter 1996; Krosnick and Milburn1990; Zaller 1992), and the confidence, developedthrough experience, that one is able to understandcomplicated political issues and events (Krosnick andMilburn 1990). Moreover, typical experimental settingsalso fail to duplicate the social contexts in whichpolitical information is acquired and used to formpreferences (Huckfeldt and Sprague 1995). It wouldseem that while experimental methods are especiallyuseful in differentiating informed from ignorant peo-ple, they are less suited to predicting the sorts ofpolicies a fully informedpublic might prefer.

547

7/30/2019 Althaus_1998

5/15

Information Effects in Collective Preferences September 1998Where filter questions and experimental methodsfall short, a third way of simulating fully informedpreferences offers promise. Independently pioneeredin work by Delli Carpini and Keeter (1996, chapter 6)and Bartels (1996), this approach uses multivariateregression to simulate how individual opinions might

change if opinion givers were better informed aboutpolitics. Unlike filter questions and experimental meth-ods, this approach is explicitly premised on the conceptof "enlightened preferences" and on the social con-struction of political interests. Estimates of fully in-formed opinions are generated by assigning the pref-erences of the most highly informed members of agiven demographic group to all members of that group,simultaneously taking into account the influence of awide range of demographic variables. For instance, ifpolicy preferences of well-informed respondents fromunion families differ from those of ill-informed respon-dents from union families, then this approach assignsthe mix of fully informed preferences to all respon-dents from union families. But instead of consideringonly the bivariate relationship between union member-ship and policy preferences, this method looks at unionrespondents who are women, from a certain incomelevel, who live in eastern states, are married, ownhomes, of a certain age, and so on. If the mostinformed people sharing all these characteristics havedifferent preferences from the least informed people,then their mix of fully informed preferences is assignedto everyone who shares their demographic character-istics.The present study extends this basic approach inseveralways. First, I use logistic regression to avoid therestrictive assumption that political information musthave a linear relationship with preferences.1 Second, Iestimate fully informed preferences for people whogive "don't know" and "no opinion" responses in theactual data, following the assumption that as informa-tion levels rise, the proportion of people who giveopinions or turn out to vote should also rise. Third,while Bartels examined vote choices in presidentialelections and Delli Carpini and Keeter analyzed fivescales representing differentpolicy domains, I conductsimulations on individualpolicy questions representing

a broad range of political issues. Although the individ-ual questions analyzed here may be less reliable mea-sures of attitudes than scales composed of multiplequestions, analyzingthe marginalsof specific questionshas two distinct advantages. It can reveal the influenceof question wording on information effects and, sincemarginal percentages seem to be the lingua franca ofopinion surveys in the political sphere, how the sur-veyed opinion that "counts" politically may be skewedby the social distribution of political knowledge.The logit model I use for simulating fully informedopinions is structured as follows:1 Delli Carpini and Keeter (1996) use ordinary least squares toestimate parameters in their simulation; Bartels (1996) uses a probitmodel to estimate parameters, which are then transformed in a waythat assumes linearity of information effects.

548

prob (Yi= 1)= a + l1i + Z kkDik+ Z k(Ii*Dik) + ei,

where Yi is respondent i's dichotomous policy prefer-ence (e.g., 1 = "favor,"0 = "oppose"); Ii is respondenti's score on a scale of political information; Dik isrespondent i's score on the kth demographic charac-teristic; Ii e Dik is the product of respondent i'sinformation score multiplied by respondent i's score onthe kth demographic characteristic; and ei is the errorterm for the ith observation. In this equation, Pt is thecoefficient for the information variable, Pk is thecoefficient for the kth demographic characteristic, and8k is the coefficient for the kth interaction term.The measures of political information are scalesdeveloped for NES data by Delli Carpini and Keeter(1993, 1996). These scales, which are detailed in Ap-pendix A, are built primarily from direct measures offactual knowledge, in contrast to the exclusive relianceby Bartels on subjective ratings of interviewers.2Thepolitical information scale for the 1992 NES has amaximum value of 23, with a mean of 12.7 andstandard deviation of 5.8; the scale for the 1988 NEShas a maximum value of 20, mean of 10.3, and standarddeviation of 5.0. These scales have respective alphareliabilities of .893 and .876.The Dik terms account for the effects of Education,Income, Age, Partisanship,Race, Gender,MaritalStatus,Occupation, Religious Affiliation, Union Membership,Homeowner Status, Parental Status, Financial Status,Region, and Type of Community (see Appendix A forcoding details). For the 1992 data I also includedReceiving WelfareBenefits andReceivingOtherBenefits.3These characteristics represent all the available demo-graphic variables that tend to be relatively stablefeatures of a respondent's makeup and that may beexpected to have some bearing on policy preferences.Excluded from the analysis were attitudinal variablesthat may be determined by or confounded with levelsof political information. The resulting mix of demo-graphicvariables is quite similarto that used by Bartels(1996) and by Delli Carpini and Keeter (1996), withone significant exception. I include party identificationbecause it is a relatively stable trait, which puts it onparwith the other demographic variables;more impor-tant, partisanship is a widely used cueing mechanism2 Interviewer ratings of respondent knowledge levels discriminatewell relative to factual knowledge scales (Zaller 1985), but correla-tions between these measures run between .57 and .68, whichdemonstrates that the two are not synonymous (Delli Carpini andKeeter 1992, 1993; see also Luskin 1987). While interviewer ratingsare frequently used as components of political knowledge scales(Delli Carpini and Keeter 1996; Zaller 1985), these findings suggestthat they may be poor stand-alone substitutes for direct measures ofpolitical knowledge.3 These two variables were not available in the 1988 data. Includingthe constants and interaction terms, there were 52 parametersestimated in the 1988 data and 56 parameters in the 1992 data. Whilethe number of parameters is substantial, the large sample size of theNES studies (1,775 completed interviews in 1988; 2,255 in 1992)should ensure the asymptotic properties of efficiency, lack of bias,and normality (Aldrich and Nelson 1984, 53; also see King 1989,74-80).

7/30/2019 Althaus_1998

6/15

American Political Science Review Vol. 92, No. 3and information shortcut for issue positions (Campbellet al. 1960; Page 1978; Rahn 1993; Rahn and Cramer1996). To the extent that party identification serves asa heuristic shortcut, excluding it from analysis couldexaggerate the apparent importance of factual infor-mation to policy preferences (cf. Dimock and Popkin1995).This model was applied to a set of policy questionsasked in the 1988 and 1992 NES (Miller et al. 1991,1993). Since a logistic regression (logit) model wasused to estimate coefficients, the data set consists of allpolicy questions from these studies with binary re-sponse options and any other policy questions that canbe collapsed straightforwardly nto dichotomous distri-butions. This made for a total of 45 usable questionsout of the approximately 100 available.The simulation proceeds in four steps. First, policypreferences are regressed on the full set of informa-tion, demographic, and interaction variables. This stepestimates the relationships among these variables andprovides a set of coefficients for simulating fully in-formed opinions. Second, each respondent's score onthe political information scale is changed to the highestpossible value. In the 1992 NES, for instance, thatvalue was 23 points, so all respondents in the 1992study were assigned a score of 23. Third, each respon-dent's fully informed opinion is calculated by pluggingthe coefficient values obtained from step one into eachrespondent's actual demographic characteristics, sub-stituting only the new values of the altered informationvariable and interaction terms. Fully informed opinionsare also estimated in this step for respondents whogave "don't know" or "no opinion" responses in theactual survey. Thus, respondents counted as "missing"in the actual data were assigned the same mix of fullyinformed preferences as the people who share theirdemographic characteristics and who actually gaveopinions. In essence, this is a sophisticated way ofweighting the preferences of opinion givers who have acertain combination of demographic characteristicsbythe number of respondents who share those character-istics. Fourth, all the individual fully informed opin-ions, including those of people who originally re-sponded "don't know" and "no opinion," areaggregated into a fully informed collective preference.I use the mean of the Yjprobabilities to construct fullyinformed marginal percentages that can be compareddirectly to actual marginal percentages. Using margin-als to gauge the substantive influence of informationeffects provides what is probably the most easily inter-pretable test statistic that can be generated with logisticregression methods.It is important to note that the marginals resultingfrom this simulation process can be compared directlyto the actual marginals only because they are opera-tionalized as the mean of the individual probabilitiesthat Yj = 1 instead of the sum of the predicted valuesof Yi. For instance, if 60% of all respondents sharing acertain demographic profile favor a certain policy, eachrespondent has a probability of .6 for favoring thepolicy. The mean of these probabilitiesrecoverstheactual percentage n favor (.6 = 60%), but since the

predicted value of Yjfor each respondent is 1 (because.6 > .5), estimates from predicted values would mis-takenly show a group that is 100% in favor of thepolicy. The upshot is that most ordinary least squaresapplications (including the simulation method used inDelli Carpini and Keeter 1996) are less suited forestimating information effects than maximum likeli-hood applications because they introduce largeamounts of error into such estimates.The end results of this four-step transformation areuniformly high information levels across demographicgroups and substantive opinions for all respondents.Using only the observed differences between well- andill-informed respondents, this method imputes to allrespondents the information processing strategies andcognitive styles employed by well-informed people. Italso allows political information to interact with demo-graphic characteristics in ways that may move prefer-ences in one direction for some groups and in theopposite direction for other groups. This flexibilityallows the model to reflect accurately the social diver-sity of needs, wants, and values. Of course, increasinglevels of information may lead instead to greaterconsensus of opinion across groups, and the simulationmethod leaves that possibility open as well. Mostimportant, this method is not predisposed to findingany information effects at all: If well- and ill-informedrespondents give essentially the same mix of prefer-ences, then the shape of the resulting fully informedpreference should be about the same as the actual one(Bartels 1996, 208-9).

INFORMATION EFFECTS IN COLLECTIVEPREFERENCESThe appropriatetest for the significanceof informationeffects in the model used here is the likelihood ratiotest (Bartels 1996,' 209).4 This test was found to besignificant at the p < .01 level in 84.4% of the 45questions and at the p < .05 level in 88.9%. In otherwords, the unrestricted model, which takes informationeffects into account, tends to provide a substantiallybetter fit to the data than does the restricted model,which assumes no information effects. Informationeffects in policy questions appear to be the norm ratherthan the exception.

4 Similar to the F-test used to compare improvement of fit in OLSequations, the likelihood ratio test compares the log likelihoods of anunrestricted logit model, which includes all the variables, and arestricted logit model, which lacks the information and interactionterms, to determine the significance of the differences between them.For example, the -2 log likelihood of the unrestricted equation inTable 1 is 2063.7, while that of the restricted equation withoutinteraction terms (not shown) is 2159.0. The result is a x2 value of95.3, with m = 28 degrees of freedom, where m is equal to thenumberof parametersn the restrictedmodel. This value is signifi-cant at thep < .00001 level,so thenull hypothesis f no informationeffectscan be safelyrejected.

549

7/30/2019 Althaus_1998

7/15

InformationEffects n CollectivePreferences September1998

TABLE 1. Logit Coefficients for Oppositionto Spousal Notification Laws (1 = Oppose)InteractionIndependent Effects (XVariables MainEffects Information)

Information (1-23) -.027 (.089) ... ...Education years) -.043 (.072) .013** (.005)Income (percentile) .011 (.008) -.000 (.001)Age (years) .009 (.014) -.001 (.001)Republican .568 (.443) -.075* (.033)Democrat -.149 (.408) .015 (.031)Black .663 (.447) -.057 (.034)Female .062 (.358) .053* (.024)Married .783* (.367) -.072** (.026)Unionfamily -.353 (.460) .028 (.031)Homeowner -.948** (.358) .048 (.025)Parentw/child athome .069 (.352) -.003 (.024)Receive welfarebenefits .045 (.472) .020 (.034)Receive otherbenefits .397 (.428) -.046 (.034)Financiallyworseoff .178 (.321) -.015 (.022)Protestant .186 (.473) -.041 (.032)Catholic .188 (.518) -.063 (.035)Other religion -.353 (.587) -.006 (.040)East -.328 (.524) .004 (.034)Midwest - .683 (.479) .024 (.031)South -.645 (.456) .030 (.030)Urban -.194 (.352) .021 (.023)Rural -.575 (.398) .026 (.028)Retired -.021 (.670) -.018 (.047)Homemaker -.058 (.499) -.004 (.038)Executive/Professional -.244 (.573) .020 (.036)Clerical .606 (.531) - .036 (.039)Technical/Sales .873 (.585) -.050 (.039)Constant -1.808 (1.262)Note: Standard errors of parameter estimates are in parentheses.Beginning log likelihood = -1221.9; ending log likelihood = -1031.8.Beginning correct classifications = 65.4%; ending correct classifica-tions = 73.1%. N = 1,894. *p < .05, **p< .01.

Information Effects in Attitudes towardSpousal Notification LawsBefore discussing the general findings from these sim-ulations, it will be helpful to take a close look at anexample of how information asymmetries can biascollective preferences. A question from the 1992 NESasked: "Would you favor or oppose a law in your statethat would require a married woman to notify herhusband before she can have an abortion?" To illus-trate how the fully informed collective preferencesdiscussed later were obtained, the logit coefficients forthis question are displayed in Table 1. Although thisequation correctly predicts responses for nearly three-quarters of opinion givers and represents a statisticallysignificant improvement over the restricted model,5 therelative paucity of significant coefficients is notable. Asimilar pattern was found in the Bartels simulation ofinformation effects in vote choices: Only between 14%5See footnote 4.

550

and 24% of coefficients achieved conventional levelsof significance (Bartels 1996, calculated from tables 1and 4-8). The reason for this pattern appears to be amulticolinearity problem stemming from the abun-dance of interaction terms correlated with politicalinformation scores as well as with the demographicvariables.6 Examining the restricted form of this equa-tion confirms that, absent the information and interac-tion terms, almost half the coefficients attain conven-tional levels of significance (data not shown).Rerunning the regression using only the significantvariable clusters identified in Table 1 produces nounexpected changes in the size or direction of coeffi-cients. These findings suggest that the relatively largestandard errors of many of the coefficients in theunrestricted equation should have little or no effect onthe unbiasedness of the coefficients themselves.A more substantial concern with the data shown inTable 1 is the potential for specification error thatcomes from excluding attitudinal variables. Given theneed to exclude such obviously relevant measures asattitudes toward abortion and women's rights, in theanalyses that follow I avoid any suggestion that themodel used here can explain individual-level opinions.Instead, I focus on what the model is intended tocapture: differences in opinion between groups as wellas the relationship between information and policypreferences within groups.Figure 1 provides a detailed comparison of theactual marginal percentages for this question and thosesimulated for a public with uniformly high levels ofpolitical knowledge. In the actual marginals, nearlytwo-thirds of respondents said they favored a spousalnotification law. Yet, in contrast to this apparentlystrong majority opinion, the fully informed collectivepreference is almost evenly divided on the issue. Theonly difference between these two measures is that thelatter controls for the uneven social distribution ofpolitical knowledge.As a precaution, I also tested whether the presenceof insignificant coefficients in Table 1 affected thesimulated marginal percentages. A second measure offully informed collective opinion was estimated using aregression containing only the significantvariable clus-ters shown in Table 1 (information, education, parti-sanship, gender, marital status, and home ownership).The result of this alternative simulation was a fully6 Most of this multicolinearity comes from the nature of the demo-graphicvariable clusters. For instance, in the 1992 NES, Republicanpartisanship is correlated with Democratic partisanship at -.78.Likewise, living in a suburbanarea is correlated at -.57 with livinginan urban area and -.42 with living in a rural area. The rest is due toassociations between different variables. For example, the scale ofpolitical information is correlated at .56 with education and .42 withincome. Being married is correlated at .42 with income and .33 withbeing a homeowner. All of this is complicated by the set ofinteraction terms, which are linearly but imperfectly related to boththe information scale and the set of demographic terms. Despitethese associations, the theoretical approach to information effectsadopted here dictates that all these variables be included in the logitmodel. The result s a largenumberof insignificant oefficients, utthe coefficients hemselves should nonetheless be valid for thepurposes o whichthey are put.

7/30/2019 Althaus_1998

8/15

American Political Science Review Vol. 92, No. 3

FIGURE 1. Actual and Fully InformedOpinion on Spousal Notification Laws[ ActualOpinion 2 SimulatedOpinion]

75- 70.7 62.9 61.0

IL~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

U_50- 40.9

25 Men WomenGender

TotalSample Favor Oppose NActualmarginals 65.5% 34.5 100% 1894)Simulatedmarginals 51.2% 48.8 100% 1969)

informed collective preference that was 52% opposedand 48% in favor of spousal notification laws. Deviat-ing only about three percentage points from the simu-lated marginalsof the full equation, this result confirmsthe accuracyof the model shown in Table 1, despite thefrequency of insignificant coefficients.7Even more interesting than the substantive shift inthe collective preference are the underlying dynamicsof change among male and female respondents. Figure1 shows that the direction of change from actual tosimulated opinion is identical for both genders: Con-trolling for information asymmetries led to decreasedsupport among both men and women for spousalnotification laws. But the magnitude of this change wasmuch smaller for men than for women. Opinion amongmale respondents shifted only eight percentage points,compared to twenty points among females. Thissmaller change among males was primarily due to aceiling effect from their higher average informationscores and to the weaker relationship for males be-tween political information levels and opposition tonotification laws (data not shown).After assigning all respondents equally high informa-tion levels, the group preference of fully informed men(i.e., support for notification laws) remained un-changed from that of actual opinion. Thus, the fifteen-point swing in collective opinion came mostly from7 In this particular case, a shift on the magnitude of three pointsnevertheless signals a reversal in the collective preference simulatedfor a fully informed public. Although a shift of this magnitude doesnot count as a change in preference ordering for this study (it wouldstillbe countedas a tie: see footnote9 below),it demonstratesheimportanceof examining he size of point changes n conjunctionwithany changes n preferenceordering.

changes in the opinions of female respondents. Asshown in Figure 1, group opinion among womenshifted twenty percentage points in a pro-choice direc-tion once information levels were raised and standard-ized. This resulted in an almost perfect reversal ofmajorityopinion among female respondents: Whereas61% of women said they favored a spousal notificationlaw in the actual data, 59% said they opposed it in thesimulated data.These findings suggest that the information imbal-ances between men and women suppress the magni-tude of pro-choice opinion revealed in opinion surveys.The fully informed majority preference of women onthe issue of spousal notification laws is diametricallyopposed to that of men, while their actual majoritypreference is the same as that of men.General Patterns of Information Effects inCollective PreferencesWith this insight into the effects of information dispar-ities on collective preferences, we now turn to thegeneral findings from the simulation data. Table 2displays these overall results for each of four issuecategories and for all questions together (Appendix Bcontains results for individual questions). Foreign pol-icy covers the use of military force abroad, economicsanctions, arms agreements, and whether the UnitedStates should become more involved in solving prob-lems around the world. The fiscal category addresseslevels of government spending and taxation. Operativeissues deal with the size and scope of the federalgovernment, the election of legislators, and govern-ment regulation of the economy. Social policy coverssuch topics as abortion, gay rights, the death penalty,and affirmativeaction.The first column shows the mean percentage pointchange between actual and fully informed opinion forall questions in each category. A score of five, forexample, indicates a five-point shift in the marginalsofone of the response options, as when collective opinionchanges from 50% to 55% in favor. Table 2 shows thatthe average difference between actual and fully in-formed opinion was about seven percentage points.Questions dealing with fiscal issues had the largestaverage point change, while social policy questions hadthe smallest average difference between actual andsimulated opinion measures.These averages give a somewhat misleading picture,however, as the distribution of point differences isskewed toward the low end of the scale. Twenty of the45 questions (44%) had a difference of between zeroand five points, 13 (29%) differed between six and tenpoints, nine questions (20%) had differences of be-tween eleven and fifteen points, and three (7%) dif-fered by more than fifteen points. In the more than halfof observations in which nontrivial changes were ob-served-that is, changes on a magnitude greater thanfive percentage points-the differences between actualand fully informed opinion averaged 10.8 points. Whencollectiveopinionchangedbymore thana few points,it tendedto shift quitedramatically.

551

7/30/2019 Althaus_1998

9/15

Information Effects in Collective Preferences September 1998

TABLE 2. Differences between Actual and Fully Informed Collective PreferencesAverage Point N withDifferent FullyInformedOpinionRelativeCategory Change PreferenceOrder to ActualOpinionForeignpolicy issues 6.13 1 of 8 Moredovish and interventionist(N = 8) (3.80)Fiscal issues 11.61 3 of 6 Morewilling o pay for more services(N = 6) (7.28) and deficitreduction, ess willingto maintainhigh defense spendingOperative ssues 9.22 2 of 8 Moreopposed to government(N = 8) (6.41) control of economy, more in favorof free marketapproachesSocial policy issues 5.49 3 of 23 Moreprogressivein general,but(N = 23) (4.16) slightly ess progressiveon someracial ssuesAllissues 7.08 9 of 45(5.35)

Source: Pooled data from 1988 and 1992 NES.Note: Standard deviations are in parentheses.

Earlier it was suggested that the point differencesbetween actual and simulated opinion might be exag-gerated if party identification variables were omittedfrom the simulation equations. This expectation turnedout to be incorrect. Rerunning the simulations withoutparty identification variables (data not shown) pro-duced estimates of information effects that were essen-tially identical to the simulation results reported here.8While Table 2 shows that correcting for informationasymmetries can lead to substantial shifts in collectiveopinion, changes of equal size may not be equallymeaningful. A change from 80% to 90% in favor of apolicy is in some ways less important than a changefrom 45% to 55% support; the former merely rein-forces the majority opinion, while the latter indicates asubstantive shift in majority opinion. The relative im-portance of the differences between actual and fullyinformed opinion can thus be clarified by noting whenthey cause collective policy preferences to change.The middle column of Table 2 shows the number ofquestions in each category for which the preferenceorder of fully informed opinion differed from that ofactual opinion. Three possible collective preferenceswere considered: a majority in favor, a majority op-posed, or a tie between options.9Any case in which thecollective preference in actual opinion differed fromthat in the simulated marginals was coded as a changein preference order.The frequency with which preference shifts occurredis a significant finding. After controlling for informa-tion effects, collective preferences changed in fully8 Of the simulations controlling for party identification, 57.8% (n =26) produced smaller estimates of information effects than thesimulations that omitted party identification variables, hardly animprovement over chance. Moreover, the mean difference betweensimulations that omitted party identification and the measures offully informed opinion listed in Appendix B was only two-thirds of apercentage point. Contrary to expectations, controlling for partyidentification seems to have little influence on the size or direction ofinformation effects in these data.9 A tie was defined as any marginals falling within plus or minus 3percentage points of the 50% mark. These boundaries approximatethe 95% confidence interval for point estimates from these data. Itfollows that a majority is defined as at least 53.5% of responses.

552

one-fifth of policy questions. Change was most fre-quent in the fiscal category and least frequent in thearea of foreign policy, although the small number ofquestions makes it difficult to draw any firm conclu-sions about differences between issue categories. Theimportant point is that controlling for the effects ofinformation on individual responses often results innew collective preferences. This contrasts somewhatwith the findings of Bartels (1996), who studied infor-mation effects in presidential election outcomes. Al-though Bartels found information effects in votechoices that are comparable to the patterns examinedhere, in every election he studied, correcting the pop-ular vote totals for the presence of information effectswould have resulted in the same outcome as washistorically the case.10In contrast, the collective order-ing of policy preferences seems to be much moresensitive than aggregate vote choices to informationeffects.My interpretation of what fully informed opinionlooks like relative to actual opinion is given in the thirdcolumn of Table 2. Fully informed opinion on foreignpolicy issues is relatively more dovish and intervention-ist than actual opinion. For example, while 29% ofrespondents in 1988 agreed that the United Statesshould stay out of problems in other parts of the world,the level of agreement dropped to just 18% of fullyinformed responses. An example of dovishness in fullyinformed foreign policy preferences comes from an-other question from the 1988 NES: "Do you favor oroppose using American military forces in the MiddleEast to protect oil shipments?" While 64% of actualresponses favored military deployment, supportdropped to 58% of fully informed responses.10 Calculated from Table 2 of Bartels 1996 and popular vote totalsfrom the Statistical Abstract of the United States. Correcting actualpopular vote totals with the estimated aggregate deviations from fullyinformed voting reported in Table 2 of Bartels 1996 has the followingresults: Nixon wins by 59.0% instead of 60.7% in 1972, Carter by51.5% instead of 50.1% in 1976, Reagan by 56.3% instead of 50.7%in 1980 and by 53.9% instead of 58.8% in 1984, Bush by 56.4%instead of 53.4% in 1988, and Clinton by 45.2% instead of 43.0% in1992.

7/30/2019 Althaus_1998

10/15

American Political Science Review Vol. 92, No. 3Simulated opinion on fiscal issues is more favorabletoward paying for deficit reduction and for a largernumber of governmental services than is actual opin-ion. The proportion of respondents willing to payhigher taxes in order to reduce the deficit rose from31% in the actual data to an impressive 53% in the

simulated data. And while 35% of actual respondentsagreed that "the government ought to cut taxes even ifit means putting off some important things that need tobe done," agreement dropped to 22% when controllingfor information effects. Likewise, only 30% of actualrespondents in the 1992 NES said that they would bewilling "to pay more in taxes so that the governmentcould spend more on the services you favor," butsupport rose to 47% among fully informed respon-dents. There were only two exceptions to the tendencyfor fully informed opinion to favor increased federalspending. While 68% of actual respondents felt thatthe United States should maintain high levels of de-fense spending, support dropped to 64% for fullyinformed respondents, and in terms of satisfaction withcurrent levels of Social Security, 40% of actual respon-dents said benefits were about right or too high,compared to 48% of fully informed respondents.With regard to operative issues, fully informed opin-ion is relatively more opposed than actual opinion tothe idea of "big government." While 63% of actualopinion givers in 1988 agreed that "the government isgetting too powerful," the level of agreement rose to71% among fully informed respondents. Likewise, in1992, 65% of actual respondents felt "there are morethings that government should be doing," compared toonly 50% of fully informed respondents. In simulatedopinion, this option tied with "the less government thebetter."In the area of social policy, fully informed opinion ismore progressive than actual opinion on most issues.Simulated opinion is consistently more pro-choice thanactual opinion (39% versus 25% opposed to parentalnotification laws) and more supportive of gay rights(40% versus 28% favored allowing gay couples toadopt children). A similar direction of change wasobserved for most social policy issues, the one excep-tion being specific programs aimed at giving preferen-tial treatment to African Americans. On matters suchas affirmative action and school integration, fully in-formed opinion tends to be slightlyless supportive thanactual opinion. At the same time, simulated opinion isslightly more progressive than actual opinion on thegeneral need for government to guarantee equal op-portunity and fair treatment to blacks (see AppendixB).The general pattern emerging from the simulationsis for fully informed opinion to be more dovish andinterventionist on foreign policy, more progressive onsocial and fiscal issues, and more conservative onoperative issues. It is interesting that the same generaltrends were found when the simulations were reesti-mated using years of education in place of the factualknowledge scales and using only stastaticallysignificantvariableclusters data not shown).Furthermore, hesegeneralpatternsof differences etweenactualandfully

informed opinion are consistent with results from othersimulation methods (Althaus 1996a; Delli Carpini andKeeter 1996; chapter 6), with the broad patterns ofopinion change among participants in the NationalIssues Convention held in January1996 (Public Broad-casting Corporation 1996), and with the results ofquestion filter experiments (Bishop, Oldendick, andTuchfarber 1983). They are also consistent with previ-ous studies comparing differences between well- andill-informed survey respondents on a variety of policyissues (e.g., Althaus 1996a; Dimock and Popkin 1995;Key 1961, 331-41; Neuman 1986, chapter 3; Popkinand Dimock 1995).CONCLUSIONCorrecting for the low levels and uneven social distri-bution of political knowledge can change our under-standing of collective preferences in significant ways.This study found that the effects of information asym-metries on collective opinion are both larger and morecommon than suggested by previous work. Controllingfor information effects produces an average change ofseven percentage points in question marginals andreveals that one in five policy questions might have adifferent collective preference if everyone were equallywell informed about politics.Obviously,these findings only suggest what collectiveopinion might look like in a hypothetical world ofpolitically attentive citizens. While it is tempting tosuppose that simulations of fully informed opinionsomehow reflect the underlying political interests ofthe mass public, I draw no such conclusion from thesedata. The key contribution of this study lies not withspeculating about the public's true interests but withdemonstrating that survey results sometimes look theway they do because so many people are ill-informedabout politics and because certain people tend to bebetter informed than others.Interpreting the findings presented here in light ofthe existing literature on public opinion is a surpass-ingly difficult task, for the results of this study seem toboth challenge and support the idea that the masspublic's command of political knowledge is related tothe quality of its collective preferences. On the onehand, the results can be seen as validating the hypoth-esis that on-line processing, heuristic shortcuts, andstatistical aggregation help the mass public compensatefor its lack of political knowledge. After all, the opti-mist might say, nearly half the question marginalsshifted less than six points after correcting for informa-tion effects, and the ordering of collective preferencesremained unchanged for eight in ten questions. On theother hand, the findings can be seen as challenging theview that heuristics, on-line processing, and statisticalaggregation help an ill-informed public express policypreferences similar to those it might give if it weremore knowledgeable about politics. After all, the pes-simist might say, more than half the question marginalsshifted at least five percentage points, and one-quartershifted more than ten points after correcting or infor-mation effects.Moreover, he pessimistmight add, the

553

7/30/2019 Althaus_1998

11/15

Information Effects in Collective Preferences September 1998ordering of collective preferences changed for one outof five questions when information effects were con-trolled.Choosing between these competing interpretationsrequires a standard for determining when informationeffects are a problem. Earlier I made use of three suchstandards for evaluating individual questions: whenlikelihood ratio tests for the presence of informationeffects attain statistical significance, when differencesbetween actual and fully informed collective opinionsexceed five percentage points (which correspondsroughly to exceeding the 95% confidence interval forpoint estimates made from 1988 and 1992 NES data),and when correcting for information effects results indifferent preference orderings for collective opinions.Yet, such standards are of little help when movingbeyond individual questions to evaluate more generalpatterns in groups of questions. Here the issues ofinterest are how often preference orderings mustchange for information effects to be called frequent,and how sizable mean point changes must be forinformation effects to be called large. As the answers tothese questions varywith the vantage point and expec-tations of the individual observer, I submit my ownassessment of the relative importance of informationeffects, recognizing that it is one among several validinterpretations of these results.Given the frequency of shifts in collective prefer-ences and the sizable point changes that occurredwith some regularity when controlling for informa-tion asymmetries, I am inclined to see these resultsas challenging more than supporting the revisionistarguments that collective preferences are "rational"and reliable guides for public policy. Yet, the famil-iar warnings about the public's civic incompetencethat have been made since the early 1950s -aresimilarly challenged by this analysis. It would seemthat a middle position fits best with the resultspresented here: Sometimes collective policy prefer-ences are significantly influenced by the public'smodest, level of knowledge about politics, and some-times they are not. Most of the time, at least in NESdata, the aggregate effect of information asymme-tries seems likely to be palpable but not decisive.

APPENDIX A: VARIABLES AND CODINGDemographicsIncome is measured as income percentiles,Education ismeasuredasyearsof formalschooling,andAgeis measuredin years. Republicanand Democrat are coded 1 for theappropriatepartisanidentification including ndependent"leaners") nd0 for all others.Black is coded 1 for AfricanAmericansand0 for allothers,Female s coded1for womenand0 formen, UnionFamily s coded1 if anymemberof therespondent'shouseholdbelongsto a union. Homeowner scoded 1 for people who own theirown home and 0 for allothers, and Parent with a Childat Home is coded 1 forrespondentswhose minorchildren esidewiththemand0 forall others. Receiving WelfareBenefits is coded 1 for peopleliving n a householdwhere they or familymembersreceivefood stamps,Medicaid,unemployment, nd/orAFDC and is

554

coded 0 for all others. Receiving Other Benefits is coded 1 forpeople living in a household where they or family membersreceive payments from Medicare and/or Social Security.Financially Worse Off is coded 1 for respondents who re-ported being worse off financially compared to one yearbefore and 0 for all others. Protestant, Catholic, and OtherReligion are each coded 1 for the appropriate religiousaffiliation and 0 for all others, with atheists, agnostics, andthose with no religious affiliation comprising the referencecategory for this set. East, Midwest, and South are coded 1 forrespondents living in the respective census region and 0 forall others, with those from western states as the referencecategory for this set. Urban and Rural are coded 1 for thosefrom the appropriate census-defined place of residence and 0for all others, with suburban as the reference category for thisset.Retired,Homemaker, Executive/Professional,Clerical,andTechnical/Sales are coded 1 for the respective occupationalstatus and 0 for all others.Political InformationThe information measures used in this study were originallyconstructed and tested by Delli Carpini and Keeter (1993,1996). These scales are primarily additive measures of correctanswers to factual knowledge questions (correct = 1; incor-rect or no answer = 0). They also incorporate a subjectiveassessment of respondent knowledge level made by theinterviewer at the conclusion of each interview. Three kindsof factual knowledge items were used to construct thesescales: relative location tests, in which correct answers areconstructed by comparing responses to two different ques-tions; open-ended questions asking respondents to identifythe job or political office held by a public figure; andclosed-ended questions testing knowledge of constitutionalpowers, which party held majority status in both houses ofCongress, and which party was more conservative than theother. An example of a correct answer to a relative locationtest is placing the Republican Party as relatively moreconservative than the Democratic Party on a seven-pointideology scale, regardless of where on the ideology scale arespondent actually places the two parties.Besides the interviewer rating score (v555, reverse coded),the questions for the 1988 NES information scale includedidentifying the offices held by Ted Kennedy (v871), GeorgeShultz (v872), Margaret Thatcher (v875), Yasser Arafat(v876), William Rehnquist (v873), Michail Gorbachev (v874),and Jim Wright (v877); naming the majority party in theHouse (v878) and Senate (v879); identifying the relativeideological locations of Bush and Dukakis (v231, v232),identifying the relative ideological locations of the Republi-can and Democratic parties (v234, v235), and locating therelative positions of the Republican and Democratic partieson national health insurance (v321, v322), government ser-vices (v307, v308), defense spending (v315, v316), and jobassurances (v328, v329).Aside from the interviewer rating score (v4205, reversecoded), the questions for the 1992 NES information scaleincluded identifying the offices held by Dan Quayle (v5916),William Rehnquist (v5917), Boris Yeltsin (v5918), andThomas Foley (v5919); identifying which branch of thefederal government was responsible for deciding the consti-tutionality of laws (v5920) and for nominating federal judges(v5921); naming the majority party in the House (v5951) andSenate (v5952); identifying which was the more conservativeparty (v5915); identifying the relative ideological locations ofRepublicans and Democrats (v3517, v3518), Bush and Clin-ton (v3514, 35 5); and dentifying he relativepositionof theparties on governmentservices (v3704, v3705), Bush and

7/30/2019 Althaus_1998

12/15

American Political Science Review Vol. 92, No. 3Clinton on government services (v3702, v3703), the partieson defense spending (v3710, v3711), Bush and Clinton ondefense spending (v3708, v3709), the parties on job assur-

ance (v3721, v3722), Bush and Clinton on job assurance(v3719, v3720), and Bush and Clinton on abortion (v3733,v3734).APPENDIX B: SIMULATIONRESULTS FOR INDIVIDUALQUESTIONSTable B-1. Levels of Actual Opinion and Simulated Estimates of Fully Informed Opinion,by Question

Year Actual SimulatedQuestion Type (Var#) Opinion Opinion

"I am going to read you a statement about U.S. foreign policy Foreign 1988 71.1 81.6and I would like you to tell me whether you agree or disagree: (v254)'This country would be better off if we just stayed home and 1992 73.7 85.8did not concern ourselves with problems in other parts of the (v3604)world.' " (percent disagreeing)"The United States and the Soviet Union have recently reached Foreign 1988 12.6 8.7agreements to reduce the number of nuclear arms. Do you (v364)approve or disapprove of these agreements?" (percentdisapproving)"Do you favor or oppose using American military forces in the Foreign 1988 35.7 42.6Middle East to protect oil shipments?" (percent opposed) (v366)

"Some people think that the U.S. should increase the pressure on Foreign 1988 41.3 34.5the South African government to change its racial laws. Others (v861)think the U.S. should not do this. What do you think-shouldthe U.S. apply more pressure or not?" (percent opposed tomore pressure)"The U.S. should maintain its position as the world's most Foreign 1988 42.9 48.2powerful nation even it if means going to the brink of war." (v972)(percent disagreeing)"Do you think we did the right thing in sending U.S. military Foreign 1992 21.0 18.4forces to the Persian Gulf or should we have stayed out?" (v3608)(percent saying should have stayed out)

"Some people think that the U.S. and its allies should have Foreign 1992 36.7 35.4continued to fight Iraq until Saddam Hussein was driven from (v3630)power. Others think that the U.S. was right to stop fightingafter Kuwait was liberated. What do you think? Should the warhave continued or should it have stopped?" (percent sayingshould have stopped)"Inorder to reduce the size of the federal budget deficit are you Fiscal 1988 68.9 46.3willing or not willing to pay more in federal taxes?" (percent (v249)unwilling)"The government ought to cut taxes even if it means putting off Fiscal 1988 64.5 78.0some important things that need to be done." (percent (v944)disagreeing)"Some people say the U.S. should maintain its position as the Fiscal 1992 31.9 35.6world's strongest military power even if it means continuing (v3603)high defense spending." (percent disagreeing)"Would you personally be willing to pay more in taxes so that the Fiscal 1992 70.0 52.9government could spend more on the services you favor or (v5922)would you rather keep your taxes the same even if this meantthe government couldn't increase its spending as you wouldlike?" (percent saying keep taxes the same)"Inyour opinion, are Social Security retirement benefits too low, Fiscal 1992 40.1 48.7about the right amount, or too high?" (percent saying about (v6132)right/too high)

"Do you favor or oppose taxes on Social Security benefits?" Fiscal 1992 87.0 81.0(percent opposed) (v6134)

555

7/30/2019 Althaus_1998

13/15

InformationEffects n CollectivePreferences September1998Table B-1. (continued)

Year Actual SimulatedQuestion Type (Var#) Opinion Opinion"Some people have suggested placing new limitson foreign Operative 1988 23.7 39.7

imports n order to protect American obs. Otherssay that such (v376)limitswould raise consumer prices and hurtAmericanexports. 1992 33.1 38.5Do you favor or oppose placing new limitson imports,or (v3802)haven't you thought much about this?" (percentopposed)"Whats your feeling,do you think he government s gettingtoo Operative 1988 36.7 29.2powerfulor do you think he government s not gettingtoo (v848)strong?" percent saying government not gettingtoo strong) 1992 29.9 31.1(v6016)"Over he past yearwouldyou say that the economic policies of Operative 1992 43.2 60.0the federal governmenthave made the nation's economy (v3541)better, worse, or haven't they made much differenceeitherway?" (percentsaying worse)"A aw has been proposed that would limitmembers of Congress Operative 1992 18.5 19.8to no more than 12 years of service in that office.Do you favor (v3747)or oppose such a law?" percent opposed)"Choosewhichof two statements Iread comes closer to your Operative 1992 64.9 50.3own opinion .. One, the less the government he better;or (v5729)two, there are more things that governmentshould be doing?"(percent saying there are morethings governmentshould do)"Choosewhichof two statements I readcomes closer to your Operative 1992 26.5 41.5own opinion .. One, we need a strong government o handle (v5730)today's complex economic problems;or two, the free marketcan handle these problemswithoutgovernmentbeinginvolved?" percent favoring ree market)"Doyou favor or oppose laws to protecthomosexuals against job Social 1988 45.8 41.8discrimination?"percent opposed) (v852)1992 39.9 34.5(v5923)"Doyou favor or oppose the death penaltyfor persons convicted Social 1988 19.9 24.8of murder?"percent opposed) (v854)1992 20.6 24.0(v5933)"Some people say that because of past discrimination, lacks Social 1988 79.9 83.1should be given preferencein hiringand promotion.Otherssay (v856)that such preferencein hiringand promotionof blacks is wrong 1992 79.9 84.0because it gives blacks advantages they haven'tearned. What (v5935)about your opinion-are you foror against preferentialhiringand promotionof blacks?"(percent against)

"Should he government nWashington ee to it that black Social 1988 42.7 39.7people get fairtreatment njobs or is this not the federal (v865)government'sbusiness?"(percent saying this is not the 1992 42.7 42.9government's business) (v5938)"Somepeople say that because of past discriminationt is Social 1988 63.7 65.8sometimes necessary forcolleges and universities o reserve (v869)openings for blackstudents. Othersoppose quotas because 1992 67.4 68.4they say quotas give blacksadvantages they haven't earned. (v5947)Whatabout your opinion-are you foror against quotas toadmit black students?"(percent against)"Equalopportunity or blacks and whites is very importantbut it's Social 1988 43.2 49.0not really he government's ob to guarantee it."(percent (v965)disagreeing)"Do you thinkthe governmentshould requirecompanies to allow Social 1992 68.8 63.0up to six months unpaid eave for parents to spend time with (v371 )their newborn or newly adopted children,or is this somethingthat should be left up to the individual mployer?" percentfavoringdecision left to employers)556

7/30/2019 Althaus_1998

14/15

American Political Science Review Vol. 92, No. 3

Table B-1. (continued)Year Actual SimulatedQuestion Type (Var#) Opinion Opinion

"Wouldyou favoror oppose a law inyour state that would Social 1992 24.5 39.2requireparentalconsent before a teenager under18 can have (v3735)an abortion?" percent opposed)"Would ourfavoror oppose a lawinyourstate that would allow Social 1992 49.1 46.9the use of government unds to help pay for the costs of abortion (v3737)forwomen who cannot afford hem?" percentopposed)"Would ou favororoppose a lawinyourstate that wouldrequire Social 1992 34.5 48.8marriedwomanto notifyher husbandbeforeshe can have an (v3739)abortion?"percentopposed)"Doyou thinkgovernment hould provide hildcare assistanceto Social 1992 38.8 40.3low and middle ncome workingparents,or isn't it the (v3745)government's esponsibility?"percent aying government s notresponsible)"Doyouthinkhomosexualsshould be allowed o serve in the United Social 1992 41.4 37.2States ArmedForcesor don't you think o?" (percent ayinggays (v5925)should not be allowed o serve)"Doyouthinkgay or lesbiancouples, inotherwords,homosexual Social 1992 72.3 59.0couples,should be legallypermittedo adoptchildren?"percent (v5927)opposed)"Doyouthink he government nWashington houldsee to itthat Social 1992 50.8 59.5white and blackchildren o to the same schools, [or hatit (v5932)should] tay outof this area as it is notthe government'sbusiness?" percent aying government hould stay out)"Somepeoplefeel thatwe should use government undsonlyto Social 1992 42.9 33.5supportchildrenwho go to public chools, others feel thatwe (v6023)should use government unds to supportchildren's choolingregardlessof whether heirparentschoose to send them to apublic,private, r parochialchool. How do you feel, or haven'tyou thoughtmuch aboutit?" percent nfavorof fundingallschooling)"Doyou favoror oppose expandingMedicare o payfornursing Social 1992 12.0 21.5homecare and long hospital tays forthe elderly?"percent (v6136)opposed)"Doyoufavora lawmakingEnglishhe official anguageof the Social 1992 30.6 28.7UnitedStates, meaninggovernmentbusiness would be (v6233)conductedinEnglishonly,or do you oppose such a law?"(percentopposed)"Doyou think hatimmigrantswhocome to the U.S. shouldbe Social 1992 78.9 70.8eligibleas soon as they come hereforgovernment ervices such (v6242)as Medicaid,FoodStamps, Welfare, rshould they have to behere a yearor more?" percent aying immigrantshould wait ayearor more)

REFERENCESAldrich, John, and Forrest Nelson. 1984. Linear Probability,Logit,and Probit Models. Beverly Hills, CA: Sage.Almond, Gabriel A. 1950. The American People and Foreign Policy.New York: Harcourt Brace.Althaus, Scott L. 1995. "Can Collective Opinion Redeem an Ill-Informed Public?"Presented at the annual meeting of the Amer-ican Association of Public Opinion Research, Ft. Lauderdale, FL.Althaus, Scott L. 1996a. "Opinion Polls, Information Effects andPolitical Equality:ExploringIdeological Biases in Collective Opin-ion." Political Communication 13(January-March):3-21.Althaus, Scott L. 1996b. "Who Speaks for the People? PoliticalKnowledge, Representation, and the Use of Opinion Surveys inDemocratic Politics." Ph.D. diss., Northwestern University.Austen-Smith, David, and Jeffrey S. Banks. 1996. "InformationAggregation, Rationality, and the Condorcet Jury Theorem."American Political Science Review 90(March):34-45.

Bartels, Larry. 1990. "Public Opinion and Political Interests." Pre-sented at the annual meeting of the Midwest Political ScienceAssociation, Chicago, IL.Bartels, Larry. 1996. "Uninformed Votes: Information Effects inPresidential Elections." American Journal of Political Science40(February):194 230.Bennett, Stephen. 1988. "'Know-Nothings' Revisited: The Meaningof Political Ignorance Today." Social Science Quarterly69(June):476-90.Bennett, Stephen Earl. 1995. "Americans' Knowledge of Ideology,1980-1992." American Politics Quarterly23(July):259-78.Berelson, Bernard, Paul Lazarsfeld, and William McPhee. 1954.Voting:A Study of Opinion Formation in a PresidentialCampaign.Chicago: University of Chicago Press.Bishop, George, Robert Oldendick, and Alfred Tuchfarber. 1983."Effects of Filter Questions in Public Opinion Surveys." PublicOpinion Quarterly47(Winter):528-46.

557

7/30/2019 Althaus_1998

15/15

Information Effects in Collective Preferences September 1998Campbell, Angus, Phillip Converse, Warren Miller, and DonaldStokes. 1960. The American Voter. New York: John Wiley.Condorcet, Nicolas Caritat de. [1785] 1972. Essai sur application del'analyse a la probabilitydes decisions renduesa la plurality des voix.New York: Chelsea.Connolly, William E. 1972. "On 'Interests' in Politics." Politics andSociety 2(Summer):459-77.Converse, Philip. 1964. "The Nature of Belief Systems in MassPublics." In Ideology and Discontent, ed. David Apter. New York:Free Press. Pp. 206-61.Converse, Philip. 1970. "Attitudes and Nonattitudes: Continuationof a Dialogue." In TheQuantitativeAnalysis of Social Problems,ed.Edward Tufte. Reading, MA: Addison-Wesley. Pp. 168-89.Converse, Philip. 1990. "Popular Representation and the Distribu-tion of Information." In Informationand DemocraticProcesses. ed.John Ferejohn and James Kuklinski. Urbana: University of IllinoisPress. Pp. 369-88.Dahl, Robert A. 1989. Democracy and Its Critics. New Haven, CT:Yale University Press.Delli Carpini, Michael X., and Scott Keeter. 1992. "An Analysis ofInformation Items on the 1990 and 1991 NES Surveys."Report tothe Board of Overseers for the National Election Studies.Delli Carpini, Michael X., and Scott Keeter. 1993. "Measuring

Political Knowledge: Putting First Things First."American Journalof Political Science 37(November):1179-206.Delli Carpini, Michael X., and Scott Keeter. 1996. WhatAmericansKnow about Politics and WhyIt Matters. New Haven, CT: YaleUniversity Press.Dimock, Michael A., and Samuel L. Popkin. 1995. "Knowledge,Trust, and InternationalAttitudes."Presented at the annualmeetingof the American PoliticalScience Association, Chicago, IL.Downs, Anthony. 1957. An Economic Theory of Democracy. NewYork: Harper Collins.Feldman, Stanley. 1989. "MeasuringIssue Preferences: The Problemof Response Instability." In Political Analysis, ed. John Stimson.Ann Arbor: University of Michigan Press. Pp. 25-60.Ferejohn, John, and James Kuklinski, eds. 1990. Information andDemocratic Processes. Urbana: University of Illinois Press.Fishkin, James S. 1991. Democracy and Deliberation: New Directionsfor Democratic Reform. New Haven, CT: Yale University Press.Fishkin, James S. 1995. The Voice of the People: Public Opinion andDemocracy. New Haven, CT: Yale University Press.Fiske, Susan F., Richard R. Lau, and Richard A. Smith. 1990. "Onthe Varieties and Utilities of Political Expertise." Social Cognition8(1):31-48.Grofman, Bernard, and Guillermo Owen, eds. 1986. InformationPooling and Group Decision Making. Greenwich, CT: JAI Press.Huckfeldt, Robert, and John Sprague. 1995. Citizens, Politics, andSocial Communication: Information and Influence in an ElectionCampaign.New York: Cambridge University Press.Key, V.0. 1961.Public Opinion andAmerican Democracy. New York:Knopf.King, Gary.1989. UnifyingPoliticalMethodology:The LikelihoodTheoryof Statistical nference.Cambridge:CambridgeUniversityPress.Krosnick, Jon A., and Michael A. Milburn. 1990. "PsychologicalDeterminants of Political Opinionation." Social Cognition 8(1):49-72.Ladha, Krishna K. 1992. "The Condorcet Jury Theorem, FreeSpeech, and Correlated Votes." American Journal of PoliticalScience 36(May):617-34.Lippmann, Walter. 1955. The Public Philosophy. Boston: Little,Brown.Lockerbie, Brad. 1991. "The Influence of Levels of Information onthe Use of Prospective Evaluations."PoliticalBehavior 13(Septem-ber):223-35.Lodge, Milton, Kathleen M. McGraw, and Patrick Stroh. 1989. "AnImpression-Driven Model of Candidate Evaluation." AmericanPolitical Science Review 83(June):399-419.Lodge, Milton, Marco R. Steenbergen, and Shawn Brau. 1995. "TheResponsive Voter: Campaign Information and the Dynamics ofCandidate Evaluation." American Political Science Review89(June):309-26.Lupia, Arthur. 1994. "Shortcuts versus Encyclopedias: Informationand Voting Behavior n California nsuranceReformElections."American Political Science Review 88(March):63-76.

558

Luskin, Robert. 1987. "Measuring Political Sophistication." Ameri-can Journal of Political Science 31(November):856-99.Mansbridge, Jane J. 1983. Beyond AdversaryDemocracy. Chicago:University of Chicago Press.McGraw, Kathleen M., and Neil Pinney. 1990. "The Effects ofGeneral and Domain-Specific Expertise on Political Memory andJudgment."Social Cognition 8(1):9-30.Miller, Nicholas. 1986. "Information, Electorates, and Democracy:Some Extensions and Interpretations of the Condorcet JuryTheorem." In InformationPooling and Group Decision Making, ed.Bernard Grofman and Guillermo Owen. Greenwich, CT: JAIPress. Pp. 173-92.Miller, Warren E., and the National Election Studies. 1991. Ameri-can National Election Study, 1988: Pre- and Post-Election Survey[computer file] (Study #9196). Conducted by Center for PoliticalStudies, Ann Arbor, MI: University of Michigan, Survey ResearchCenter [producer], 1991. Ann Arbor, MI: Inter-university Consor-tium for Political and Social Research [distributor], 1991.Miller, Warren E., Donald R. Kinder, Steven J. Rosenstone, and theNational Election Studies. 1993. American National Election Study,1992: Pre- and Post-Election Survey [enhanced with 1990 and 1991data] [computer file] (Study #6067). Conducted by University ofMichigan, Center for Political Studies. ICPSR ed. Ann Arbor, MI:University of Michigan, Center for Political Studies, and Inter-university Consortium for Political and Social Research [produc-ers], 1993. Ann Arbor, MI: Inter-universityConsortium for Polit-ical and Social Research [distributor], 1993.Mondak, JefferyJ. 1994. "Question Wording and Mass Policy Prefer-ences: The Comparative Impact of Substantive Information andPeripheralCues." Political Communication l(April-June):165-83.Neuman, W. Russell. 1986. TheParadoxof Mass Politics: Knowledgeand Opinionin the American Electorate. Cambridge, MA: HarvardUniversity Press.Page, Benjamin. 1978. Choices and Echoes in PresidentialElections.Chicago: University of Chicago Press.Page, Benjamin, and Robert Shapiro. 1992. The Rational Public:FiftyYearsof Trends n Americans' Policy Preferences. Chicago: Univer-sity of Chicago Press.Popkin, Samuel. 1991. The Reasoning Voter: Communication andPersuasion in PresidentialCampaigns. Chicago: University of Chi-cago Press.Popkin, Samuel L., and Michael A. Dimock. 1995. "CognitiveEngagement and Citizen World Views." Presented at the annualmeeting of the PEGS Conference on Citizen Competence and theDesign of Democratic Institutions, Washington, DC.Public Broadcasting Corporation. 1996. "National Issues ConventionDeliberative Poll Reveals Significant Change in Delegates' Viewson Key Issues." Press release by Public Broadcasting Corporation,January25. *Rahn, Wendy. 1993. "The Role of Partisan Stereotypes in Informa-tion Processing about Political Candidates."American Journal ofPolitical Science 37(May):472-96.Rahn, Wendy M., and Katherine J. Cramer. 1996. "Activation andApplication of Political Party Stereotypes: The Role of Televi-sion." Political Communication 13(April-June):195-212.Sigelman, Lee, and Ernest Yanarella. 1986. "Public Information onPublic Issues: A Multivariate Analysis." Social Science Quarterly67(June):402-10.Smith, Eric, and Peverill Squire. 1990. "The Effects of PrestigeNames in Question Wording." Public Opinion Quarterly54(Spring):97-116.Sniderman, Paul, Richard Brody, and Philip Tetlock. 1991. Reason-ing and Choice: Explorations in Political Psychology. New York:Cambridge University Press.Stimson, James. 1975. "Belief Systems: Constraint, Complexity andthe 1972 Election." American Journal of Political Science 19(Au-gust):393-417.Wittman, Donald. 1995. The Myth of Democratic Failure: WhyPolitical Institutions Are Efficient. Chicago: University of ChicagoPress.

Zaller, John. 1985. "Pre-Testing Information Items on the 1986 NESPilot Survey." Report to the National Election Study Board ofOverseers.Zaller, John. 1992.The Nature and Origins of Mass Opinion. NewYork:CambridgeUniversityPress.