1956018

8/12/2019 1956018

1/17

Effects of Public Opinion on Policy

Author(s): Benjamin I. Page and Robert Y. ShapiroSource: The American Political Science Review, Vol. 77, No. 1 (Mar., 1983), pp. 175-190Published by: American Political Science AssociationStable URL: http://www.jstor.org/stable/1956018.

Accessed: 06/03/2014 13:41

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at.http://www.jstor.org/page/info/about/policies/terms.jsp

.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 formsof scholarship. For more information about JSTOR, please contact [email protected].

.

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

The American Political Science Review.

http://www.jstor.org

This content downloaded from 128.151.244.46 on Thu, 6 Mar 2014 13:41:52 PMAll use subject to JSTOR Terms and Conditions
http://www.jstor.org/action/showPublisher?publisherCode=apsahttp://www.jstor.org/stable/1956018?origin=JSTOR-pdfhttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/stable/1956018?origin=JSTOR-pdfhttp://www.jstor.org/action/showPublisher?publisherCode=apsa

8/12/2019 1956018

2/17

Effects of Public Opinion on PolicyBENJAMINI. PAGEUniversity of Chicago

ROBERTY. SHAPIROColumbia UniversityandThe National Opinion Research Center, University of ChicagoThe responsiveness of government policies to citizens' preferences is a central concern of variousnormative and empirical theories of democracy. Examining public opinion and policy data for theUnited Statesfrom 1935 to 1979, we find considerable congruence between changes inpreferences andin policies, especially for large, stable opinion changes on salient issues. Wepresent evidence thatpubic opinion is often a proximate cause of policy, affecting policy more than policy influencesopinion. One should be cautious, however, about concluding that democratic responsiveness pervadesAmerican politics.

The responsiveness of government policy tocitizens' preferences is a central concern in nor-mative democratic theory (Dahl 1956; Arrow1963; Sen 1970), and there is no shortage of em-pirical theorizing about the extent to which policydoes or does not respond to public opinion.Economists' perfect information theories of elec-toral competition (the first section of Downs1957;Davis, Hinich, and Ordeshook 1970) predicta high degree of responsiveness, under variousassumptions about the nature of preferences andthe behavior of politicians and voters.Certain interest group theorists, on the other

We aregrateful o the HooverInstitution or a Na-tional Fellowship,and to theNationalScienceFounda-tion for providingsupportunder Grant SES-7912969A01. John M. Gillroy capably helped collect andorganizemuch of our data. NormanBradburn,RobertMichael, Alexander Hicks, Tom W. Smith, LutzErbring, Calvin Jones, Eric Uslaner, and RaymondWolfingerprovidedusefulsuggestions ndcriticisms.We also wish to thank our many research ssistantsand students,who helpedwith variousphasesof datacollection and analysis: MichaelGalati, Joe Altonji,John Dister, Eric Schmaler,Glenn Dempsey, JohnSchultz,Ted Zang, Jack Knight, John Treantafelles,LeahKnowlton,KathleenBawn,MyronMykyta,SusanBrooks,Peter Skarzynski,Paul Gronke,Susan Sher-manKarplus,CharlesM. Silver,MaryJordanLangen-henning,Lenny Kusnitz,ElizabethMcNichol, RonaldE. Toles, BethDorris,EhsanFeroz, SusanEsser,Fran-cis Kwakwa,WallyMihailovich,BrucePeterson,SteveRoeder,Susan Kuczynski,MitchellFink, David Mc-Cluskey,RichardSarkissian,KennethWong, RichardHeinle,RobertM. Wynne,CarolynOckels,and MyrtleStephens.An earlierversionof this paperwas presented t the1981annualmeetingof theAmericanPoliticalScienceAssociation,New York.

hand (e.g., Schattschneider 1960; McConnell1966), suggest that the will of the general publicmay be obstructed because the system of pressuregroup politics is biased in favor of well-organizedbusiness and professional groups. Similar resultsare predicted by political economists who build in-formation costs and transaction costs into theirmodels (thus accounting for political inequality),and who stress that the free-rider problem im-pedes the organization of diffuse interests (thesecond part of Downs 1957; Olson 1965; Hardin1982).Those who see government as a problem solver,administered by statesmen who reason about thepublic interest and employ objective techniques(e.g., cost-benefit analysis) to reach policy deci-sions, likewise do not necessarily expect any closecorrespondence between specific policies and thepublic's preferences of the moment as expressedin opinion polls.Although certain other scholars do anticipateagreement between opinion and policy, they seethe causal relationship as partly or wholly re-versed: politicians and policies themselves affectpublic opinion. Under conditions of limited infor-mation this could occur if statesmen lead andeducate ordinary citizens by helping them tounderstand which policies will further their in-terests (Mill 1962; Key 1961); or if politiciansmanipulate the public with lies and deceptive sym-bols (Edelman 1964; Wise 1973; Miliband 1976).Finally, of course, an observed relationship be-tween opinion and policy could be spurious, theproduct of a concurrent influence upon both bysome outside factor.By no means are all of these processes mutuallyexclusive; they could occur in various combina-tions. Furthermore, there is reason to expect thatthey might vary across issue areas. Foreign policy

175

This content downloaded from 128.151.244.46 on Thu, 6 Mar 2014 13:41:52 PMAll use subject toJSTOR Terms and Conditions
http://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsphttp://www.jstor.org/page/info/about/policies/terms.jsp

8/12/2019 1956018

3/17

176 The American Political Science Review Vol. 77decisionsmight be relativelyautonomousfromthe public,or theymight nvolvemore leadershipand manipulationof opinion than do domesticpolicies.Politicalresponsivenessmightbe greatestconcerninghighly salient issues for which thescope of conflict is broad. There mightalso besignificantvariationsacrosspoliticalsystemsandhistoricalperiods,or amongdifferentpolitical n-stitutions. ncontemporary mericanpolitics,forexample,onemightexpect hepresident ndCon-gress o bemoreresponsiveo publicopinionthanare state governmentsor the courts. One mightalso predictthat governmentwill respondmorefrequentlyto cases of large sustainedopinionchangethan to changesthat are slight or tem-porary.Empirical videnceon thesematters endsto beinconclusive.Someof the best research oncernsrepresentationn theU.S. Congress.Thesestudiesindicate hat thereare moderate elationships e-tweencongressmen'soll call votes and theircon-stituents' survey-measuredpolicy preferences(Millerand Stokes 1963, 1966).But the principaldata(fromthe 1958studyby the SRC,Universityof Michigan)arebased on small andunrepresen-tativedistrict amplesand are susceptibleo vary-ing interpretationsAchen 1978; Erikson 1978).(The recentuse of congressionaldistrictsas pri-marysamplingunits promises mproved amples;see Erikson1981;Page et al. 1981.) The causaldirectionof the underlyingprocesses s left am-biguous, since we cannot be sure whethercon-stituents'opinionsaffectcongressmen's otes, orwhether-perhaps less plausibly-congressmeninfluenceopinionin their districts.The use of demographicensusdata to simulatedistrictopinions,whichmaysuggest tronger ep-resentationalinks(Erikson1978,but see Page etal. 1981; eealsoWeber t al. 1972-73;WeberandShaffer1972)eases the samplingproblemandin-creasesour abilityto drawcausal nferences.Butsimulationalso introducesnew problemsof itsown (see Seidman1975;Kuklinski1977).Somebut not all of thesedifficultiesare reducedwhenreferendum results rather than demographiccharacteristics re used as indicatorsof prefer-ences(e.g., Kuklinski1978).The most important imitationof representa-tion studies is that they concern the microlevelbehaviorof individualcongressmenand do notnecessarily evealmuch about the responsivenessof the politicalsystemas a whole.Theydealwithlegislators'votes, not policy results (Weissberg1978).Some empiricalstudies have pursued a dif-ferent, more macrolevelstrategyof comparinggovernmentpolicies with aggregate (surveyed)publicopinion. Erikson(1976)found substantialcovariation f opinionandpolicy nthreeareasof

state policymaking. Weissberg's thorough casestudies of eleven policy topics provided severalex-amples of "majoritarian" congruence, in whichpolicy corresponded to what the majority ofAmericans said they favored, but also severalcases of noncongruence (Weissberg 1976; see alsoDevine 1970). The small number of cases limitedgeneralizability, however, and there were pro-digious difficulties in policy measurement. In suchstudies alternative policy measures often suggestdifferent results, and some opinion items are soambiguous that they are not easily matched withspecific policies. Causal inference is quite un-certain, with little hope of distinguishing policyresponsiveness from leadership or manipulationof public opinion.Based on a much larger number of cases (248),Monroe's (1978) work has suggested that there isconsiderable-though far from complete-consis-tency between opinion and policy, especially forforeign policy and highly salient issues. Again,however, policy measurement is difficult, andcausal inference is problematic; it is hard to tellwhether correspondence between opinion andpolicy arises from democratic responsiveness,from leadership or manipulation of opinion, orfrom some combination of these.

Data and MethodsOur own approach also employs a macrolevel,aggregate design but uses what Weissberg (1976)calls the "covariation" model, examining rela-tionships between changes in preferences andchanges in policy in the United States. The in-stances of opinion change are our units ofanalysis.Our change-oriented design (discussed morefully in Page and Shapiro (1980) and Shapiro(1982)) permits simple, ordinal measurement ofpolicy. We need only ascertain whether policy

moves in the same direction as opinion-that is,congruently-or in the opposite direction, or notat all. We can gather multiple measures of policyto make sure that they all point in the same direc-tion or, if they do not, to show how the findingsvary with different kinds of measures.Most important, we aim to illuminate the causalstructure of the processes by which policy andpublic opinion are related. We can use temporalasymmetriesto get at causal priority, noting whenopinion changes before policy and when the op-posite occurs.We began by sorting through marginal frequen-cies from several hundred surveys of nationalsamples of Americans, which were conducted be-tween 1935 and 1979 by three survey organiza-tions: Gallup, the National Opinion ResearchCenter, and the Survey Center/Center for Politi-


8/12/2019 1956018

4/17

1983 Effects of PublicOpinionon Policy 177cal Studies (Michigan). We archived several thou-sand (3,319) questions about policy preferences,of which some 609 items were repeated in identicalform at two or more points in time. We then iden-tified every instance in which there was a signifi-cant change (6 percentage points or more, assum-ing samples of 1500 and fairly even divisions ofopinion) in opinion from one survey to another.We chose the end points of instances of change soas to maximize the extent of movement in a givendirection without significant internal reversal.In all we found 357 instances of significantchange in Americans' policy preferences between1935 and 1979. These instances of change are de-scribed elsewhere (Page and Shapiro 1982;Shapiro 1982);a list is available from the authors.These changes in policy preferences encompassmany different kinds of policies at the federal,state, and local levels: foreign and domestic(about one-half each); spending, taxation, regula-tion, military action, trade, diplomacy. Theycover the period from 1935 to 1979 (but with rela-tively few cases before 1940). They vary inmagnitude (from as large as 38 percentage pointsto as small as a barely significant 6 percentagepoints); in duration (some span many years,others as little as one month); and in gradualnessor abruptness of change. Some represent long-term trends; a few are parts of fluctuating timeseries. Each of these characteristics is explicitlydescribed by variables in our data set, so that wecan investigate whether the frequency of policycongruence varies with characteristics of opinionchanges.For each of these instances of opinion changewe measured policy outputs during the periodbeginning two years before the date of the initialopinion survey and ending four years after thefinal survey. We used multiple indicators of policyand examined various lags between the end of theopinion change and the measurement of policy.Our techniques of policy measurement are de-scribed in an Appendix found in Page andShapiro (1981a, b) and also available from theauthors.The analysis in this article is based on the full357 instances of opinion change, for all of whichwe have been able to code covariational con-gruence (or noncongruence) using at least onesuitable measure of government policy. In manyBases (57 percent), the best available measure isidentical to the theoretically ideal measure agreedapon by both senior investigators on the basis ofthe precise wording of each survey item. For pur-posesof comparision with other research, we alsoattemptedto code the extent of majoritariancon-Pruencebetween policy and opinion at the begin-ling of each instance of change, at the end of thechange, and one year and four years after the end.

These data allow us to determine how muchagreement there is between changes in opinionsand policies; how the results of our change designcompare with those for majoritarian congruence;whether the extent of congruence varies betweenforeign and domestic issues, for large versus smallopinion changes, or in other ways; and, in cases ofcongruence, whether opinion or policy usuallymoves first.

FindingsHow Much Congruence

There has been a great deal of congruence be-tween changes in policy and changes in opinionduring the last half century; more, in fact, thaninitially meets the eye. Table 1 displays our find-ings for all 357 cases of opinion change, using thebest available policy measure.' These figures referto congruence after a one-year lag; that is, policychange is measured from the moment (Tl) of thefirst opinion survey to a point one year after thefinal (T2) opinion measurement for the instanceof change. As one would expect, congruence ap-pears more frequent when the policymaking pro-cess is allowed time to react to change in opinion,and a one-year lag is a reasonable time interval."

Congruent changes in policy were clearly muchmore frequent than noncongruent changes. TableI also shows, however, that in 33 percent of thecases policy did not change at all. What are we tomake of these cases of no change in policy?In one sense, cases of no change would seem tosignify a lack of congruence: the public's prefer-ences changed significantly, but policy did not. Ifall the no-change cases were treated as non-congruent with opinion, the extent of congruencewould be a rather unimpressive 43 percent.But a closer look demonstrates that many, infact most, of the no-change cases are inappropri-ate for handling by our change-oriented researchdesign, and they conceal congruence of a differentkind. In the first place, 65 (54 percent) of these

'Results or the caseswiththeoretically"ideal"pol-icy measures regenerally imilarbut sharper; omewillbe mentioned later. Results for second- and third-ranked measuresare slightly weaker than those dis-cussed here. Dichotomous udgmentsof congruence/noncongruenceusing the second and third measurescorrelated 82 and .50 (Yule's Qs), respectively,withthosefor the bestpolicy measure.2The66 percent congruencewith a one-year ag isslightlyhigher han the 61 percentwith no lag. It is ap-proximatelythe same as with a two-year lag (66percent),a bit more than with threeyears(63 percent),and- forunknown easons-slightlylessthanthe67.5percentwitha four-year ag.


8/12/2019 1956018

5/17

178 TheAmericanPoliticalScienceReview Vol. 77Table 1. Congruence Between Opinion and Policy, 1935-1979

Cases withTotal cases policy changeto N t N

Congruent change in opinion and policy 43 (153) 66 (153)Noncongruent change in policy 22 ( 78) 34 ( 78)No change in policy 33 (120)Uncertain 2 ( 6)100 (357) 100 (231)

Note: Each case is an instance in which public policy preferences changed significantly, according to repeatedadministration of identical survey items.

cases concern policies that have reached floors orceilings, making it impossible for them to respondany further to opinion. For example, when moreAmericans came to oppose "requiring all youngmen to give one year of service to the nation-either in the military forces or in nonmilitarywork," and there was no compulsory military (ornonmilitary) service before this opinion change,the political system could not respond to thechange by drafting fewer than zero people. Simi-larly, when there was an increase in the size of themajority opposing fines for people who fail towear seatbelts in automobiles, it was impossible torespond by lowering fines, which were alreadyzero. These policies had reached a point wherethey could not move in the same direction aspublic opinion. We consider such cases inappro-priate for covariational analysis and set themaside.We also observe that 9 (16 percent) of the other55 no-change cases eventually show policy changein a congruent direction (and then stay con-gruent), but only after time lags greater than theone year we are presently discussing. We do notattempt to take them into account, because of off-setting changes in the opposite direction.But it is more important that many of the re-maining no-change cases involve a dichotomouspolicy or a relatively discontinuous one, such asspecific legislation or executive action, so thatsmall opinion movements would not necessarilybe expected to yield policy changes unless opinionhappened to surpass some threshold-perhaps asizable majority.3When, for example, approval of

3Inperhaps more than half these cases policy could beconsidered maximally noncongruent, and as a reviewerhas pointed out, this floor or ceiling effect preventsmovement in a noncongruent direction. The cases of nopolicy change which are the most likely examples ofnonresponsiveness often involve political reform issues(e.g., the electoral college, voting age, terms for federalofficials, a national primary). All these cases show con-

CommunistChina'sadmission o the UnitedNa-tions rose from 7 percentto 20 percentin the1950s,we would not expectthe UnitedStatestobeginvotingfor admission.Whensupport oroneof Gallup'spet projects ike nationalprohibitionor six-year erms orpresidents osefromapprox-imately20 to 30 percent,we would not expectthose plans to be enacted.Policy responsivenesswouldmoreplausibly akethe formof increasedpre-policyactivity:more and strongerproposalsbeing made, more bills introduced,and furtheradvancementof proposals through the policy-makingprocess.For a small subset (28) of these no-changecases, we have data on suchpre-policymeasuresas the numberandcontentof proposalsandtheirprogresstoward enactment.4In 57 percent ofthosecases,the pre-policymeasuresmoved n thesamedirectionas publicopinion. In only 14per-centof thecasesdidit movein the oppositedirec-tion; in the resttherewas still no change.Sincethese cases with pre-policymeasuresrepresentamoreor less representativeampleof caseswith-out policy change,5 t is reasonable o estimatethat approximatelyhe same57 percentof all no-

changecases without floor or ceilingeffects in-volvedcongruentpre-policyactivity.

gruent pre-policy activity (discussed below). See Shapiro(1982) and Monroe (1978).

4Weare especially grateful to John Dister and JosephAltonji for their skillful and diligent assistance in col-lecting much of this historical information on legislativeand executive activity.'The cases were not chosen by any systematic pro-cedure. They are, however, quite diverse, covering

issues such as civil liberties, political reform (the votingage, terms for federal officials, and others), atomicenergy, arms control, relations with particullar foreigncountries or regions, education, Hawaiian statehood,welfare, inflation, gun control, the draft, wars, labor,Puerto Rico, taxes, and a few others. They do not mani-fest any obvious bias.


8/12/2019 1956018

6/17

1983 Effects of Public Opinion on Policy 179In a largemajority approximately0 percent)of no-changecases, then, the lack of changeresultedroma discontinuityn policy,whichcon-cealedeithera floor or ceiling effect or a pre-policymovementn the samedirectionas opinion

(see Table 2). We are reluctant, therefore, tocountno-changecasesas noncongruent.But weare also hesitantto considerany of these casesfullycongruent.Rejectedbillsarenot laws;pro-posals are not official policies; and no-changecaseswhichmayhavealreadyresponded ully topublicopinionstill shouldnot be calledcongruentchanges.For most purposes, herefore,wewillsetaside all cases of nonchange n policy, pointingout when treatmentof some of them (thosewithoutfloor or ceilingeffects)as noncongruentwouldalterour findings.Examininghe 231casesfromTable 1in whichwe arecertain hatpolicychangedn onedirectionor the other,we find thatpolicychangewascon-gruentwith opinion change n 66 percentof thecases.If our collectionof cases s treatedas a sam-ple, we can rejectthe null hypothesisof random(probability5) congruenceat betterthanthe .01level,basedon a cumulativebinomial est. More-over, the 66 percent figure may be depressedsomewhatby measurement rror. When we re-strictour attention o the 203 casesfor whichwehave theoretically"ideal" policy measures,the

extentof congruences an evenmore striking76percent.Our66 percentresult s quitesimilar o the 64percentthat Monroe (1979) found for majori-tariancongruence,hatis, for agreement etweenmajorityopinion and policy. In our data, ma-joritarian ongruence ppears o be less frequent:only 54 percentat T2 (based on the cases thatcouldbe codedwithconfidence),andonly49 per-cent with opinion measuredat T1. Neither issignificantlydifferentfrom the amountof con-gruencethat would be found in a sample bychance f therewereno realrelationship t all be-tweenopinionandpolicy.Butthisis probablyanartifactof the researchdesignbasedon casesof

opinion change, since majoritarian congruenceduring the process of change might be expected tobe temporarilyout of kilter. And in fact, when weexamine policy a year after T2, giving opinion andpolicy time to reach a new equilibrium, the pro-portion of congruent cases goes up to 61 percent.With a more generous four-year lag, it rises to 65percent; and for any time within a four-year span(the criterion Monroe used), it is 68 percent.These latter results are very similar to Monroe'sfindings and to our 66 percent for covariationalcongruence.As expected, majoritarian congruence some-times proved difficult to assess. In 22 percent ofthe 357 cases (for majoritarian congruence with aone-year lag), we had to resort to codes of "prob-able" congruence or noncongruence, or completeuncertainty. In contrast, we had very little dif-ficulty coding covariational congruence, as longas there was discernible change in policy; only 2percent (6) of the cases were coded probable oruncertain using the one-year lag.Thus we begin with a finding of rathersubstan-tial congruence between changes in opinion andpolicy. This result is reinforced when we lookmore closely at the cases of noncongruence.RelativelyFew Cases of Noncongruence

When we examine the cases of noncongruence,it becomes apparent that the 34 percent in Table 1overstates their frequency and importance. Someare artifacts of imperfect policy measures. (As wenoted, there is less noncongruence-only 24percent-among the cases for which we have idealpolicy measures.) Approximately a quarter of thecases of noncongruence vanish when lags longerthan one year are allowed for. Some involve verysmall changes in opinion, to which it is hardlystartling that policy does not respond. Finally, afew instances of noncongruence involve fluc-tuating or temporary opinion changes in whichthe changes in opinion are closely preceded orfollowed by movements in the opposite direction.

Table 2. Kinds of Nonchange in PolicyProportion of cases withno policy change

SO NPolicy already maximally congruent with opinion change 54 (65)Pre-policy congruent with opinion change 26 (31, estimated*)Nonresponsive to opinion change 20 (24, estimated*)

100 (120)*Estimates are based on the subsample of cases with pre-policy measures.


8/12/2019 1956018

7/17

180 The AmericanPoliticalScienceReview Vol. 77Table 3. Frequency of Congruence for Opinion Changes of Different Sizes

Direction of policy changeSize of Congruent Noncongruent Totalopinion change % N S N t N6-7 percentage points 53 (25) 47 (22) 21 (47)8-9 64 (32) 36 (18) 22 (50)10-14 62 (32) 38 (25) 29 (65)15-19 69 (22) 31 (10) 14 (32)20-29 86 (18) 14 ( 3) 9 (21)30+ 100 (10) 0 (0) 4 (10)

65 (147) 35 (78) 100 (225)Gamma = -.29.

It would be asking a great deal to expect policy tofollow each small twist and turn of opinion.If all the weak cases are disregarded, there re-main relatively few solid, convincing examples ofnoncongruence between opinion and policy. Butthere are some which must be taken seriously.The extent of congruence for opinion changesof different sizes is shown in Table 3. Noncongru-ence is much more frequent when opinion changeis small; it reaches 47 percent for cases with abarely significant (6 or 7 percentage point) changein public preferences. That is, for very small opin-ion changes, policy moves in a congruent direc-tion no more than would be expected by chance(assuming a .5 probability). Among cases of verylarge change in public opinion, by contrast, policyalmost always goes in the same direction as opin-ion. When there is opinion change of 20 per-centage points or more, policy change is con-gruent an overwhelming 90 percent of the time.Similarly, noncongruence is relatively commonwhen there is evidence that opinion may be revers-ing direction or fluctuating.6 Based on the 136cases for which we have adjacent opinion mea-surements to detect fluctuations, noncongruenceagain nears the 50 percent level (it is 44 percent)among cases of fluctuation or reversal. It is only21 percent among cases of steady or one-timechange. When an opinion change is sustained,policy usually moves in a congruent direction.When we put all these factors together and dis-regard all cases where there is small (less than 10

6As we have argued elsewhere (Page and Shapiro1982), public opinion has seldom fluctuated in the senseof suddenly moving significantly in opposite directions.In the analysis reported here, we more loosely call anopinion change part of a fluctuation when there existseither an opinion point within two years before Tiwhich is not significantly (6 percentage points) differentfrom T2; or one within two years after T2 which is notsignificantly different from T1.

percent) opinion change, or our data show opin-ion fluctuation, or the policy measures are im-perfect, we find that noncongruence is quite un-common indeed. Fully 87 percent of the remain-ing cases are congruent (see Table 4). We take thisfigure to be a better estimate of the extent of con-gruence than the 66 percent reported earlier.When Americans' policy preferences change by asubstantial amount, without reversal, publicpolicy (if it changes at all) overwhelmingly tendsto move in the same direction.Still, some cases of noncongruence remain andare worth illustrating. Although public dis-approval of economic assistance to EasternEurope rose 10 percentage points (from 42 to 52percent) between November 1956 and April 1957,for example, U.S. aid (in current and constantdollars) actually increased. Similarly, fromAugust 1952to June 1956public support for send-ing our allies economic rather than military aidrose by 26 percentage points (from 55 to 81percent ), but economic assistance as a propor-tion of total aid (military and economic) declined.From November 1971 to April 1973support for"more strict" wage and price controls increasedfrom 42 to 56 percent, but responsiveness to thiswas transient; after President Nixon's 60-dayprice freeze of June 1973, the strictness of con-trols decreased through "Phase IV" and the ter-mination of the Economic Stabilization Programin May 1974. During the mid-to-late 1970s, thefederal government moved toward the adoptionof the metric system (with the Metric ConversionAct of 1975), while public opposition to such ac-

70ur data permit us to identify the presenceorabsenceof fluctuationsn only59 percent N= 136)ofour cases. A numberof cases whereopinionfluctuated-but therewerenosurveyso reveal he fact-no doubtremain n our data set and depressthe proportionofcongruenceamong the "stable" instancesof opinionchange n Table4.


8/12/2019 1956018

8/17

1983 Effectsof PublicOpinionon Policy 181Table4. Frequencyof Congruenceor Large,StableOpinionChangesUsingIdealPolicyMeasures

Directionof policy changeCongruent Noncongruent Total

Dpinionchange S N % N % NLarge 10%+)and stable 87 (53) 13 ( 8) 54 (61)Small nd/orfluctuating 63 (33) 37 (19) 46 (52)

76 (86) 24 (27) 100 (113)Yule'sQ= .58.

-ion rose 15 percentage points (from 50 to 65 per-.ent among people who said they "know what thenetric system is"). These few cases are sufficient:o indicate that responsiveness to public opinion islot perfect.Variations in Congruence

The Nature of Opinion. The extent of congru-ance appears to vary according to the political in-ititutions and the types of issues involved. Most:f these differences, however, reflect more basicDifferencesin the nature of public opinion or inthe way it changed.The most basic variations in congruence, inFact, are probably those we have already dis--ussed. Policy tends to move in the same directionas public opinion most often when the opinionChange s large and when it is stable-that is, notreversed by fluctuations.Similarly, policy congruence is higher on salientthan on nonsalient issues. There has been muchmore frequent congruence, for example, in caseswhere an opinion question was repeated often, anindicator, presumably, of substantial public in-terest as well as relatively large opinion changes.

UPollstershave incentives to ask questions thatwill attract readers' attention.) Policy moved in aDirection congruent with opinion change in 76

percent of the 74 cases involving items that wererepeated six or more times, but in only 60 percentof the 65 cases with just two repetitions.Salience can be measured more directly,although still not perfectly, by the proportion ofrespondents answering "don't know" or "noopinion" to survey questions. When the propor-tion of don't knows is relatively low-that is,when more people are willing to offer a preference-it is a sign of more public interest and attentionand perhaps also stronger, more intensely heldopinions. As Table 5 indicates, congruence wassubstantially more common when "don't know"responses were relatively few (73 percent congru-ence for 1-5 percent don't knows) than when theywere many (56 percent congruence for 15 percentor more don't knows).6

'Since the literature (e.g., Smith, 1978) indicates thatproportions of "don't knows" vary markedly, depend-ing on the question formats and interviewing techniquesused in different surveys, we conducted the sameanalysis separately for different survey organizationsand types of surveys. While the average frequencies of"don't knows" did vary, the same moderate tendencytoward more congruence with fewer "don't knows" ap-peared, although it was weaker in Gallup poll items.There were particularly large salience effects in theNORC (State Department) foreign policy surveys.

Table5. Frequencyof Congruence y IssueSalienceDirectionof PolicyChange

Proportion esponding Congruent Noncongruent Total"don'tknow"at T2 (%6) %N %N %N15+ 56 (14) 44 (11) 13 (25)10-14 61 (25) 39 (16) 22 (41)5-9 70 (48) 30 (21) 36 (69)1-5 73 (40) 27 (15) 29 (55)

67 (127) 33 (63) 100 (190)aamma = -.20.


8/12/2019 1956018

9/17

182 TheAmericanPoliticalScienceReview Vol. 77These findings suggest that Schattschneider's(1960) argument concerning the scope of conflicthas some validity. On issues about which thepublic has more well-defined opinions and showsmore concern, where the scope of conflict is

broad, policy tends to move in harmony withpublic opinion.The greatercongruence when opinion change islarge and stable and issues are salient is consistentwith the interpretation of congruence as showingan effect of opinion upon policy. The theoreticalgrounds for expecting greater congruence in thesecases have to do with democratic responsiveness.It is natural to expect that when the public hasdefinite opinions, when those opinions change bylarge amounts, and when the changes endure overtime, the political system will more often respondto the public's preferences. Moreover, as we willsee below, multivariate analysis shows that someother variations in congruence-variations ac-cording to issue type and political institutionresult from characteristics of public opinion.

Type of Polcy Issue. One expected kind of vari-ation in congruence, a difference between domes-tic and foreign issues, does not occur even at thesimple bivariate level. One might think congru-ence should be more frequent on domestic thanforeign policy issues, since the public presumablytends to care more about matters close to homeand is more insistent that politicians follow itswishes on domestic policy. On the other hand, wemight expect congruence of a different sort onforeign policy issues. Precisely because the publictends to be less involved and have less informa-tion, it might be easier for officials to changepolicy and get citizens to go along.We found, however, that there is little dif-ference between foreign and domestic issues. Theapparently slightly greatercongruence for domes-tic issues (70 percent) than foreign (62 percent)ac-tually reverses a finding from our small prelimi-nary data set (see Page and Shapiro 198 a, p. 16).But the difference is not statistically significant,and it disappears when no-change cases arecounted as noncongruent: then there is 53 percentcongruence for domestic policy and 54 percent forforeign.The reason foreign policy does not show sub-stantially less (or more) congruence than domesticis probably that opinion changes were not ap-

We also used Tom Smith's (1980) time series for the-Gallup poll's "most important problem" question toclassify issues by salience. Differences in congruencewere small and erratic, however, probably because thegross response categories failed to differentiate sharplyamong our specific issues, especially foreign policies.

preciably smaller or less stable on foreign thandomestic issues, although foreign affairs weresomewhat less salient, having significantly moredon't knows (see Shapiro 1982; Page and Shapiro1982). It is also possible that the gross similarityhides a difference in the processes by which con-gruence comes about, but we have not yet foundany compelling evidence for this.We did find more congruence (78 percent) forthe salient and large-changesocial issues, than oneconomic or welfare policies (66 percent) or mostof foreign policy. There are too few cases to besure about more refined categories of issues, but itis worth noting that congruence was particularlyhigh on cases concerning abortion (100 percent,n 10), civil liberties (89 percent, 9), civil rights(74 percent, 19), and World War 11(100 percent,7), for which our instances of change were mostlylarge and sustained. In contrast, congruence wasless frequent on matters of collective security (50percent, 14), national defense (53 percent, 19),relations with Russia (64 percent, 14), foreign aid(65 percent, 26), political reform (63 percent, 19,including big government and labor issues), or theeconomy (68 percent, 22) where opinion changestended to be smaller and sometimes fluctuating.The frequency of congruence has varied in dif-ferent historical periods: there was more in the1970s (75 percent of 92 cases) than in the Eisen-hower years (59 percent of 37) or the Kennedy-Johnson years (54 percent of 26), and the levelwas about average from 1935 through 1952. Butbefore concluding that government was intrin-sically more responsive during the Nixon, Ford,and Carter administrations, we should note thatthere has been a trend toward a better-educatedcitizenry and greater issue salience, concurrentwith the rise of important new social issues (seeScammon and Wattenberg 1970;Davis 1975;Nie,Verba, and Petrocik 1976; Taylor 1978; Shapiro1982). Moreover, opinion data have been better inrecent years, with identical questions repeatedmore regularly. More time has passed, so thatthere has been more opportunity to discover casesof large, sustained opinion change.

Political Institutions. McConnell (1966) andothers have argued that state governments, withtheir narrow constituencies and low visibility, areless responsive to the public than the federal gov-ernment is. Similarly, Schattschneider (1960)claimed that the scope of conflict is broader andresponsiveness greater when the president ratherthan Congress makes policy. By the same token,the judiciary might have the lowest visibility andbe the most insulated from the public.To test these hypotheses we noted which politi-cal institutions-state governments, Congress, thepresident, or the courts-were specified (or im-


8/12/2019 1956018

10/17

1983 Effects of PublicOpinionon Policy 183plied) in each opinion question as the relevantpolicymaking bodies. We then compared the fre-quency of congruence across political institutions.

Overall we found little difference among theexecutive, Congress, or even the federal courts;for each, congruence was at the 63 to 65 percentlevel. New congressional legislation, however, ishard to get. About two-thirds of the cases requir-ing it showed no change at all, often (presumably)because the changes in opinion had not yetreached a threshold of magnitude and durationsufficient to push bills through the legislative pro-cess. But when legislation did change, it wentoverwhelmingly (in 92 percent of 25 cases) in thesame direction as public opinion. Other congres-sional action that was joint with the executive(which usually involved continuous measuressuchas expenditures) was congruent with aboutaverage frequency, in 59 percent of 125 cases.This lack of variation among national govern-ment institutions need not contradict the reason-ing that led us to expect differences. All our casesinvolve significant opinion changes on issues ofrelatively high public salience, quite possiblyhigher salience than is usual with court decisionsor legislative actions.To our distinct surprise, state policies turnedout to be congruent most often of all, in 81 per-cent of 26 cases.9 Again, however, the nature ofthe opinion changes, rather than institutionalcharacteristics, probably underlie this finding.Most of the state issues we examined were moralor social concerns like capital punishment,divorce laws, and abortion; issues of high saliencefor which there were often large changes in publicopinion (cf. Erikson 1976). As we will see, in amultivariate analysis controlling for the charac-teristics of opinion change, the apparent effect ofstate policymaking vanishes.We take this as supporting the core of McCon-nell's (1966) and Schattschneider's (1960) argu-ment: policy is more responsive to the public onissues of high salience and visibility. Most statepolicy may be less salient than our cases, moresubject to special economic interests and smallconstituencies, and less in harmony with thepreferencesof the general public.

Ideological Direction.One ratherstartlingvaria-tion in congruence, however, persists even when

'Our opinion data come from national samples, sothat the corresponding measuresof state policy are "na-tionalized": They are aggregate tendencies across allstates, with each state's policy weighted by its popula-tion (see Weissberg 1976). Thus one or two states candominate the results with responsive changes, if allotherstates keep policy constant. C.f., Weber and Shaf-fer (1972) and Erikson (1976).

other factors are taken into account. That is thestrong tendency for policy to move congruentlywith public opinion more often when opinionchanged in a liberal direction. Policy moved con-gruently with liberalopinion changes a substantial86 percent of the time, but with conservativechanges only 53 percentof the time (see Table 6).10In part, this simply reflects the fact that some ofthe most salient issues and biggest opinionchanges of the last two decades involved sweepingliberal trends in such areas as civil rights, civilliberties, and abortion (see Page and Shapiro1982). Not surprisingly, policy moved mostly inharmony with those trends. On the average ourinstances of liberalchange are larger than the con-servative shifts (means of 14.5 versus 10.5 percen-tage points); and all but one of the 14 changes of30 percentage points or more were in a liberaldirection. But in the multivariate analysis con-trolling for such opinion characteristics, theliberal effect is still substantial.We hesitate to conclude, however, that there isa liberal bias in the polity. Some of the conser-vative changes are very recent and relatively small,and the period of our data collection may haveended just before opinion shifted furtherin a con-servative direction and government policies, witha time lag, began to respond to them. The liberal-conservative difference is small for large andgradual opinion changes, particularly when thecases of no policy change are counted as non-congruent (Shapiro 1982). Moreover, differentfrequencies of congruence say little or nothingabout the degree of congruence for changes in dif-ferent ideological directions, nor about whetherthe average level of policy differs systematicallyfrom citizens' desires. Our finding does not ex-clude the possibility that policy could tend to beless liberal than the public wants. In fact, accord-ing to our data, policy a year after T2 was (ma-joritarian) congruent with a liberal majority onlya bit more often, and not significantly so (61 ver-sus 55 percent), than with a conservative majority.Still, the finding of more congruence for liberal

'The codingof policychangesas liberalor conserva-tive was necessarilyubjective,but the two authors g-noredconceptual omplexities nd simply ried o applythe labels as usedin journalistic nd populardiscoursein the 1960s and 1970s, resachingnearperfectagree-ment(99.5 percent)on the 215of 357 cases that eachwas willing to assign to one of the dichotomouscate-gories.The remaining142 cases were set aside as noteasily locatableon the liberal-conservativeimension,although they were includedas nonliberal hanges inour multivariate nalysisto maximize he numberofcases. Table 6 also excludes the cases of no policychange;the difference n congruence s approximatelythesame whenthesecases are included.


8/12/2019 1956018

11/17

184 TheAmericanPoliticalScienceReview Vol. 77Table 6. Congruence for Liberal vs. Conservative Opinion Changes

Direction of policy changeDirection of Congruent Noncongruent Totalopinion change t N t N t NLiberal 86 (62) 14 (10) 58 (72)Conservative 53 (28) 47 (25) 42 (53)

72 (90) 28 (35) 100 (125)Yule's Q = .69.

changesmustbe reckonedwith. Onepossibilitysthatpolicyhada liberaldynamic n the 1960sand1970s regardlessof public opinion; that whenpublic opinion happenedto move in the samedirection,policyappeared ongruentwith t. Thuspart of the relationshipbetween opinion andpolicy mightbe spurious.This seemsmoreplau-sible with regard to spending issues, however(where incrementalism nd bureaucratic xpan-sionismmightholdsway),thanwithrespect o thesweepingsocial liberalizationn civil rights,civilliberties,and lifestyle ssues.Inanycase,themultivariateesultspresentednTable 7 confirmthat in severalcasesthe magni-tudes and even the directionsof relationships

changewhen we move from bivariate orrelationsto multipleregression nd logisticanalysesof thevariablesaffecting congruence.Taken as a group, the highly intercorrelatedcharacteristics f opinion-the size of opinion

"The presence of a dichotomous dependent variablereduces the size of coefficients estimated by correlationand regression techniques and reduces the efficiency ofestimates. Logit (or probit) analysis is more appropri-ate. We have included OLS and logit results in Table 7,because standardized regression coefficients are morereadily interpretable and because here, as is often thecase, the main findings are very similar. See Hanushekand Jackson (1977).

Table7. Multivariate nalysisof CongruenceBivariate Regression LogisticVariables Correlation* Stand.Beta Estimate t***

Size of opinionchange %O) .23 .06 .04 1.31No fluctuation** .19 .11 .57 1.15Howlongchange ook (months) .18 .01 .00 .45Lowsalience %DKs) -.10 -.09 -.05 1.63Socialissues .14 .00 -.09 .16Economy .01 .03 .17 .26Defense -.09 -.03 -.29 .55WorldWar I .13 .15 9.75 .00Otherforeignpolicy -.14 -.11 -.58 1.36Liberalopinionchange .28 .25 1.38 2.65States' policy .11 -.05 -.08 .101969-1979period .14 .06 .11 .26

MultipleR .40R2 .16Note: The dependent variable is direction of policy change from T1 to one year after T2, scored 0(noncongruent)or 1 (congruent). Some independentvariableswith insignificantbivariatecorrelationsareomitted.Theexcludedcategoryof issuetypesconsistsof remaining omesticpolicies.See Shapiro 1982).*Correlationswith magnitudesof .11 or betterare significantat the .05 level;those .07 or greater,at the .1level.

**Threecategories; ases with insufficienttime points to detect fluctuationwere coded in a middlecategory.***Ratioof logisticestimateto its asymptoticstandard rror.Weconsider valuesof 1.96 or greater tatisticallyreliable.


8/12/2019 1956018

12/17

1983 Effectsof PublicOpinionon Policy 185change,whether t fluctuated,how long it took,and the degreeof policysalience-are nearlysig-nificantat the .05 level(F(4,177df)= 2.181,p=.072),buttheir ndividual ffectsdiminish unlikeother variables n the table, however, most oftheir ogisticestimatesaregreater hantheirstan-darderrors).The recencyof opinionchange(the1969-1979period)makes little or no difference;social issues lose their distinctiveness; nd statepolicymaking witchesto an insignificantnega-tive effect.The liberaldirectionof changeretainsa sub-stantial mpactand is significantlydifferent romzero (p < .05) according to a t-test. Again,though, we do not want to overstate he impor-tanceof this effect. Nor wouldwe conclude hatother variables are irrelevant,especially sincemulticollinearityndmeasurementrrormake heestimatesunstable,and the causalstatus of theliberal-conservativevariable is unclear. Wesuspect hat,evenafter"controlling" orallotherfactors n Table7, someof the apparent ffect ofliberalopinionchangeuponthedegreeof congru-ence results from the fact that liberal changestendedto be large,salient,and stable.

TheEffectof PublicOpinionon PolicyThe mereobservationof congruencebetweenopinionandpolicytells us little, of course,aboutwhich causeswhich. Congruencecould indicatethat there is democratic responsiveness:thatchanges in public preferencescause changes inpolicy.Butit might nsteadresultfrompolicyaf-fecting opinion. Policy changes might leadcitizens o change heiropinionsas theysawgoodresults romthe newpolicies,or as theyrational-ized that whatever he governmentdoes mustbeacceptable.Or persuasive hetoricby politiciansandothersaccompanying ewpoliciesmightcon-

vince citizensof theirmerits. Such processesoflearning, persuasion, and rationalization areindistinguishablerom responsiveness,f all weknow s thatopinionandpolicymovein thesamedirection.

The use of temporal asymmetries, however,permits us to draw some inferences about causalpriority. In particular, when public opinionchanges before policy does, it is generallyreasonable to infer that policy has not affectedopinion.A high proportion of the changes in publicopinion since 1935 did in fact occur before con-gruent changes in policy. To begin with, of 148cases of congruence after a one-year lag, 26 per-cent showed no congruence during the time inter-val of the opinion change. That is, at the end ofthe instance of opinion change (T2), policy hadnot yet moved in a congruent direction. Opinionclearly moved first.Furthermore, for a small subset of congruentcases, additional information is available: thereexists an opinion survey at a point (Tla) betweenTi and T2 which shows that opinion had alreadychanged significantly from what it was at T1. Forthese cases we measuredpolicy change from TI toTla. In a high (48 percent) proportion of them,policy had not yet moved in a congruent directionby Tla; in fact it had usually moved in the oppo-site direction (see Table 8).12 Again, opinion hadclearly begun to change before policy changed.

If we are willing to consider these cases withTla's as a random sample of all cases of congru-ence (which, strictly speaking, they are not),"3we

'2Onecaveat s in order: n a few cases the brevityofthetimeintervalrom TI to Tla mayhaveprevented sfrom detectingchanges in aggregatepolicy measures(e.g., annualbudget igures),even thoughpolicydidinfact change."However,these cases are quite diverse,coveringsuchissuesas relationswithRussia,foreignaid, policytoward China, generalactiveness n internationalaf-fairs, armscontrol, tariffs, taxes, inflation,abortion,civil liberties,nationaldefense,crime,civilrights,and

Watergate.They appear o haveno glaring ubstantivebias, but theywereselectedby the availability f dataratherhanthroughanysamplingprocedure.Onepossi-ble problem s that the focuson the casesof significantopinionchange romTI to theinterveningmeasurement

Table8. Movement f OpinionBeforePolicyProportionof Cases

% ~~~~NOpinionmovementbeforepolicy 48 (22)Earlierpolicymovementcannot be ruledout 52 (24)

Note. Entriesare basedon caseswith interveningTia) opinion measurements eforeT2, in which opinionhad alreadychangedsignificantly.Opinionmovedbeforepolicy if policybetweenTi andTla did not changeormovedcontraryo the opinionchange.


8/12/2019 1956018

13/17

186 The AmericanPoliticalScienceReview Vol. 77can infer that in at least the same proportion of allcases (about half), opinion changed before policydid. The actual proportion may be much higher,because in cases where policy had moved con-gruently by Tla, opinion (unmeasured) may stillhave moved first.In half or more of our cases of congruence,then, we rule out the possibility that policy af-fected opinion. We think that it is reasonable inmost of these cases to infer that opinion changewas a cause of policy change, or at least a proxi-mate or intervening factor leading to governmentaction, if not the ultimate cause.The main alternative possibility is that opinionand policy were only spuriously related, both in-dependently affected by world events, elite leader-ship, interest group campaigns, technologicalchange, or some other exogenous factor. We can-not definitely rule out spuriousness (or gauge itsextent) without multivariate analysis incor-porating exogenous variables for which we do notpresently have measures, and/or intensive casestudies of congruent changes in opinion andpolicy. (We are conducting a number of casestudies, which we shall report upon at lengthelsewhere.) But spuriousness becomes less likelywhen the first movement comes unequivocallyfrom the normally sluggish public (e.g., Key 1961;Page and Shapiro 1982) rather than policy-makers; the time asymmetrysuggests a genuine ef-fect of opinion. Based on our careful study of in-dividual cases to date, our judgment is that opin-ion changes did often affect policy.To cast doubt on the likelihood of spuriousnessis by no means to deny that third factors affectboth opinion and policy, which they surely do.We only argue that public opinion is a real influ-ence-often an intervening one-upon policy, inmany (probably more than half) of our cases ofcongruent change. When some third factor affectsboth opinion and policy, it tends to affect policythrough opinion; policy changes only becauseopinion changes. A few brief illustrations lendsupport to this interpretation.In a number of our cases policy moved in a con-gruent direction only well after opinion changed:up until the time period one year after the opinionchange, policy had changed contrary to publicopinion or remained the same. The cases showingthis apparent responsiveness occur among a fairlywide range of policy issues, and most of theserelationshipsdo not appearto be spurious, at leastnot a consequence of major external events.

elminating asesin whichopiniondid not changedur-ingtheinterveningime,and where t is somewhatmorelikelythatpolicy changed irst.

In domestic affairs, examples include a majorcivil-rights issue and abortion policies. Publicopinion on these issues had probably changed asthe result of generational replacement and na-tional liberalizing trends since the 1950s(see Davis1975; Taylor 1978), and perhaps through eventsand media influences. Although it has been dif-ficult to determine with confidence the causal im-pact of opinion concerning school desegregationand some other civil rights issues (McNichol 1980;Schmaler 1981), we found that the law givingblacks "the right to be served in public places"(i.e., the Civil Rights Act of July 1964)was passedonly after public support for it had increased from54 to 66 percent during the period from June 1963to mid-August 1963 and then January 1964.(Opinion had begun to change before Kennedy'sassassination in November 1963.) Similarly, per-mitting abortions for pregnant women who didnot want to have a child (because they were"single" or "poor" or simply did not want to)came about (with Roe v. Wade, February 1973)only after public opinion had changed substan-tially, by 32 percentage points from December1965 to April 1972.14Other instances in which opinion change prob-ably contributed to policy change were thepassage of the Federal Election Campaign Act of1971(public opinion since late 1968 had shown in-creased support for campaign spending laws; seeDempsey 1982) and the October 1979 legislationgiving the president authority to ration gasoline(public approval increased from May to August1979).In foreign affairs there is comparable evidencefor government responsiveness. From April 1948to June 1949, public opposition to militaryaid forthe Nationalist Chinese rose 16 percentage points;during the period from March to September 1949,disapproval of the Marshall Plan rose 14 percen-tage points; and from March of that year to April1950, more people thought that we were spendingtoo much on European recovery (a change of 19percentage points). From November 1950 toAugust 1952 (continuing through March 1955),more of the public thought that aid to "back-ward" countries was a good idea (a 15percentage-point shift in opinion). In all of these casesgovernment-assistance expenditures, in constant

"Our analysis (and Karplus's 1979) of states' policiespermitting abortions under a variety of more stringentconditions (e.g., rape, the possibility of a birth defect,danger to the health of the mother) provided less evi-dence for a causal interpretation, since there were noopinion data between 1965and 1972 to indicate whetheropinion had changed before some states revised theirlaws.


8/12/2019 1956018

14/17

1983 Effectsof PublicOpinionon Policy 187dollars, moved contrary to opinion until the yearafter a significant opinion shift had occurred, andthen moved congruently.A more striking case of policy following thelead of opinion was the United States' support forthe admission of Communist China to the UnitedNations. Support among Americans for admis-sion rose 33 percentage points from January 1964to May 1971, with a 13 percentage-point shift by1966. Had public preferences not shifted, theNixon administration very well might not havealtered U.S. policy toward the People's Republicof China (but cf. Kusnitz 1980).

Does Policy Affect Opinion?The finding that in many cases public opinionaffects policy does not rule out the possibility thatthere is a reciprocal influence as well. It is quitepossible that policy also affects opinion, in someor all of the ways we have mentioned: by citizenslearning about a policy's impact, rationalizing itsexistence, or heeding the persuasive efforts ofpoliticians, interest groups, or others.Time asymmetries are not as helpful for dis-covering the extent to which policy affects opin-ion as they are for making the opposite inference.First of all, with intermittent poll data it is some-times possible to be sure that opinion changedbefore policy, but it is seldom certain that policychanged before opinion. Public opinion may havechanged at some early point when there were nosurveys. Secondly, policymakers can (and haveincentives to) anticipate the reactions of thepublic, so that public opinion may affect policyeven when policy changes first. A mere showingthat policy changed before opinion in a numberofcases could not be conclusive about causation.Nonetheless, investigation of temporal asym-metries can at least give some idea about the ex-tent to which policy affects opinion, perhaps iden-tifying an upper bound for it. Our data are not in-consistent with some effect of policy on opinion,though it is likely much less than the influence ofpreferences on policy.A first point to be made is that during the yearbefore our instances of opinion change (from oneyear before Ti to Ti), policy moved in the samedirection as the subsequent opinion change only

25 percent of the time. Usually (46 percent of thetime) policy did not change at all over that period,and when it did change it was as often (27 percent)in a noncongruent direction as congruent. In apreliminary report we therefore inferred-per-haps too quickly-that policy could not be affec-

ting opinion to any considerable degree (Page andShapiro 1981a).'The interpretation changes somewhat whenonly the cases of congruence between opinion andpolicy (using a one-year lag) are considered. Ofthese cases, a higher 39 percent show congruentmovement by policy in the year before the begin-ning of the opinion change. (The figure is a bitlower for the cases with ideal policy measures.)But examining policy movements before thebeginning of opinion changes does not allow forthe impact of new policy that comes into beingbetween the first and last opinion measurementsin our instances of change. For the most pertinentinformation on how policy may affect opinion,we must turn to another subset of cases: those 81cases for which there are opinion data at an in-tervening time point (Tic) between Ti and T2,where opinion had not yet changed to a statis-tically significant extent." A finding of congruentchange from Ti to Tic-that is, policy changefrom the moment of our baseline opinion mea-surement up to a point at which opinion had notyet changed significantly-would seem to supportthe likelihood of policy affecting opinion.By this criterion, possible effects of policy onopinion appear more frequent. In 46 percent ofthe 41 congruent cases for which data areavailable, policy between Ti and Tic changed inthe same direction as the subsequent opinionchange (see Table 9).Thus our data suggest that policy may affectopinion in close to half the cases of congruencebetween opinion and policy. But we would em-phasize that this is only an upper limit; the extentof influence is likely much less. Policy change isonly barely (and not significantly) more often inthe direction of opinion change than it is in theopposite direction. The congruence could repre-sent nothing more than chance. Furthermore, aswe have noted, opinion can still be affectingpolicy (through anticipated reactions) ratherthanthe opposite, even when policy moves first.Most important, the inference of policy effects

"We are still confident that this policy-first congru-ence occurs through processes (presumably of educa-tion, leadership, manipulation, rationalization, or rati-fication) different from those through which opinion af-fects policy. This congruence is not related to thevariables examined earlier (under "Variations in Con-gruence"), which distinguish conditions under which intheory policies will be more or less responsive to publicpreferences. See Shapiro (1982)."If several such points existed we picked the last one,in order to maximize the chance of detecting policymovement before significant opinion change. For agiven case there can be both points Tic and Tia; in thatevent Tic always occurs first in time.


8/12/2019 1956018

15/17

188 The American Political Science Review Vol. 77Table 9. Movement of Policy Before Significant Opinion Change

Proportion of Cases% N

Influence of policy upon opinion possibleChange in same direction as subsequent policy change 46 (19)

Influence of policy upon opinion ruled outNo change in policy 20 (8)) 54 (22)Change in opposite direction 34 (14) 5

100 (41)Note. Table 9 is based on cases with intervening (Tic) opinion measurement showing no significant change inopinion from Tl to Tlc. Policy movement is measured from Tl to Tlc.

upon opinion depends on viewing the cases withTlc's as a sample of all our cases of congruence,but there are strong reasons to doubt the validityof the sample. The TIc's were selected as points atwhich opinion had not yet "significantly"changed from Ti: any change that had occurredwas less than 6 percentage points. Yet the defini-tion of our instances of opinion change itself en-sures that at point Tlc opinion had alreadymoved slightly in the direction of T2. Thereforeall the cases of policy congruence displayed inTable 9 actually occurred after a small shift inopinion had already begun. Consequently the 46percent figure almost certainly overstates the fre-quency with which policy moved before opinion.17To be sure, there are some cases in which ourbest judgment indicates that policy change or ac-companying rhetoric or both probably affectedpublic opinion. For example, as the AtlanticCharter and early wartime cooperation with the

"'The atado not, however,ndicatea higher evelof(Ti to Tlc) congruenceamong the cases with largeropinion shifts from Ti to Tlc, exceptfor the largest(andnearly ignificant) hangesof 5 percentage oints.A thoroughgoingtudyof the effectsof policyuponopinionwouldnotworkbackwardsromthe dependentvariableas we havedone here,butwouldbeginwithaset of policymeasurementsnd ascertainn whatdirec-tion, if any, opinionsubsequentlymoved.Wearenowstartingsuch a study. A focus solely on instancesofopinionchangebiases hedatatoward howinganeffectof policyon opinion; his particular iasdoesnotoccurfor ourassessment f opinion'seffectsuponpolicy.A furtherbias n this useof Tlc's results romthefactthat in some cases public opinion at Ti may itselfalreadyhavechanged roma prior unmeasured)oint.Pollstersmaysometimeshaveaskeda questionat Ti,preciselybecausecircumstances ad changed n a waythat led them to suspecta changein public opinion.Here,again,policythatmovedonlyafterthebeginningof a shift in opinionwould appear o changebefore t.

Allies formed the basis for the United States'greater institutionalized participation in inter-national affairs, the public quite likely reacted tothis; support for joining an international organi-zation after the war rose sharply (22 percentagepoints from May 1941 to June 1942).Similarly, the Watergate investigation, congres-sional hearings, and court actions may haveswayed the public (perhaps in part by merely un-covering the facts) against President Nixonthrough a process of political leadershipor educa-tion. From June 1973 to May 1974, opposition toimpeaching Nixon and compelling him to leaveoffice dropped 34 percentage points. Of course inboth these cases public opinion may have subse-quently affected government action as well.Deception and political manipulation are theless appealing counterparts of education or lead-ership. The escalation of the Vietnam War may bethe best example of this in our data set. FromAugust to November 1965 public support forsending more troops to Vietnam rose by 13 per-centage points (more respondents said they wouldvote for a congressman who advocated sendingmore men to Vietnam), while by July the numberof troops there had risen to approximately 75,000,although the public was not informed (and maystill not be) about the uncertaintysurrounding theTonkin Gulf incident (see Wise 1973).But we cannot be sure how often policy changecauses congruence between opinion and policy.Very likely the frequency of this process is lessthan the upper limit suggested by Table 9, and lessthan is implied by some fashionable images of ac-tive policy makersand a pliant public. Almost cer-tainly the opposite process of opinion affectingpolicy is the more prevalent one.

ConclusionThe finding of substantial congruence betweenopinion and policy (especially when opinion


8/12/2019 1956018

16/17

1983 Effects of Public Opinion on Policy 189changes are large and sustained, and issues aresalient), together with the evidence that opiniontends to move before policy more than vice versa,indicates that opinion changes are importantcauses of policy change. When Americans' policypreferences shift, it is likely that congruentchanges in policy will follow.It is tempting to conclude, then, that demo-cratic responsiveness pervades American politics;that the people get what they want from govern-ment. But there are several reasons to be cautiousabout such a conclusion:1. Our findings concern policies on which pub-lic opinion has been judged by survey researchersto exist and to be worth measuringat two or morepoints in time. Although these are concrete issues,not merely setting the outer bounds for govern-ment action (cf., Key 1961; Weissberg 1976), theyare still cases of relatively broad generality andhigh public salience. On many less visible mattersof more specific, detailed policy, the public mayhave less opinion and less voice.2. There are some cases among those in whichpolicy does not change at all after opinionchanges, which must be considered nonresponsiveto the public (see Shapiro 1982).3. There are a few clear cases of noncongru-ence in which policy unequivocally moves in theopposite direction from opinion. Although theseare not nearly as frequent as cases of congruenceand are particularly rare in the face of large,stable opinion changes, some of them concernserious matters of policy and should not be dis-missed out of hand. They do demonstrate thatresponsiveness is not perfect.4. Our findings are consistent with policy af-fecting opinion in a substantialnumber of cases ofcongruence.5. The covariational research design poses arather easy test for political responsiveness: onlythat there be some movement of policy, howeverbig or little, in the same direction as an opinionshift. Even if responsiveness of this sort occurredevery time, there could still be major biases in thesystem. Policy might, for example, always moveonly half as far as opinion dictated; a strong tilt ofpolicy toward the positions of special interestgroups could coexist with such widespread (butpartial) responsiveness to the public.The above points must be considered in ap-praisingthe extent to which public opinion deter-mines policy outcomes. Moreover, even to thedegree that policy does react to public opinion,one should be cautious about bestowing the nor-mative imprimatur of "democracy" without tak-ing account of the quality of that opinion: whatkind of information it is based on, what has in-fluenced it, and perhaps how closely it corre-sponds with objective standards of citizens'

interests.Even if public opinion is truly a proximatecause of policy,it may tself beaffectedby factorsnot whollycompatiblewithnormative onceptsofdemocracy. If, for example, interest groups orpoliticiansmanage o manipulate pinion hroughlies or deception, and policy subsequentlyre-sponds to the manipulatedopinion, we wouldhesitate to celebratethe result as a democraticone. And if leaders influence public opinionthrough a more benign process of educationorpersuasion,using reasonedargumentsand goodevidence,we might stillwantto call the outcomesomethingdifferentfrompure democracy.To studyprocessesof leadershipor manipula-tion of opinionrequires xaminingmediacontent,politicians'rhetoric, and other matters beyondthe opinion andpolicy variables onsideredhere.Beforesuch researchs doneit wouldbe unwise odrawnormativeconclusionsabout the extent ofdemocraticresponsiveness n policymaking.Wecan be confidentonly that publicopinion,what-everits sourcesand quality,is a factorthat gen-uinelyaffects governmentpolicies in the UnitedStates.

ReferencesAchen, C. H. 1978. Measuring representation. Ameri-

can Journal of Political Science 22:475-510.Arrow, K. J. 1963, orig. pub. 1951. Social choice andindividual values. New York: John Wiley and Sons.Dahl, R. A. 1956. A preface to democratic theory.Chicago: University of Chicago Press.Davis, J. A. 1975. Communism, conformity, cohorts,and categories: American tolerance in 1954 and1972-73. American Journal of Sociology 81:491-513.Davis, 0. A., Hinich, M. J., and Ordeshook, P. C.1970. An expository development of a mathematicalmodel of the electoral process. American PoliticalScience Review 64:426-48.Dempsey, G. R. 1982. Public opinion and public policy:a case study of campaign finance in the UnitedStates. Unpublished, University of Chicago.Devine, D. J. 1970. The attentive public: polyarchialdemocracy. Chicago: Rand McNally and Co.Downs, A. 1957. An economic theory of democracy.New York: Harper.Edelman, M. 1964. The symbolic uses of politics.Urbana, Ill.: University of Illinois Press.Erikson, R. S. 1976. The relationship between publicopinion and state policy: a new look based on someforgotten data. American Journal of Political Sci-ence 22:25-36.1978. Constituency opinion and congressionalbehavior: a reexamination of the Miller-Stokes rep-resentation data. American Journal of Political Sci-ence 22:511-35.1981. Measuring constituency opinion: the 1978congressional election study. Legislative StudiesQuarterly 6:235-45.


8/12/2019 1956018

17/17

190 The AmenrcanPolitical Science Review Vol. 77Hanushek, E. A., and Jackson, J. E. 1977. Statisticalmethods for social scientists. New York: AcademicPress.Hardin, R. 1982. Collective action. Baltimore: JohnsHopkins University Press.Karplus, S. S. 1979. Public opinion and policy change:

the legalization of abortion in the United States. Un-published, University of Chicago.Key, V. O., Jr. 1961. Public opinion and Americandemocracy. New York: Alfred A. Knopf.Kuklinski, J. H. 1977. Public opinion: a test of thesurrogate model. Public Opinion Quarterly 41:34-40.1978. Representation and elections: a policyanalysis. American Political Science Review 72:165-77.Kusnitz, L. 1980. Public opinion and foreign policy:the recognition of the People's Republic of China.Unpublished, University of Chicago.McConnell, G. 1966. Private power and Americandemocracy. New York: Random House.McNichol, E. 1980. School busing for integration: acase study in opinion change and policy change.Master's thesis, University of Chicago.Miliband, R. 1976, orig. pub. 1969. The state in capital-ist society. London: Basic Books.Mill, J. S. 1962, orig. pub. 1861. Considerations onrepresentative government. Chicago: HenryRegnery.Miller, W. E., and Stokes, D. E. Constituency influencein Congress. American Political Science Review 57:45-56.1966. Constituency influence in Congress. InElections and the Political Order, eds. A. Campbell,P. E. Converse, W. E. Miller. New York: JohnWiley and Sons, pp. 351-73.Monroe, A. D. 1978. Public opinion and public policy,1960-1974. Presented at the annual meeting of theAmerican Political Science Association, New York.1979. Consistency between public preferencesand national policy decisions. American PoliticsQuarterly 7:3-19.Nie, N. H., Verba, S., and Petrocik, J. R. 1976. Thechanging American voter. Cambridge, Mass.: Har-vard University Press.Olson, M. 1965. The logic of collective action. Cam-

bridge, Mass.: Harvard University Press.Page, B. I., and Shapiro, R. Y. 1980. Policy change andpublic opinion change. Unpublished, University ofChicago.1981a. Congruence between preferences andpolicy: a preliminary report. Presented at the annual

meeting of the Midwest Political Science Associa-tion, Cincinnati.1981b. Effects of public opinion on policy.Presented at the annual meeting of the AmericanPolitical Science Association, New York.1982. Changes in Americans' policy preferences,1935-1979. Public Opinion Quarterly 46:24-42.Page, B. I., Shapiro, R. Y., Gronke, P. W., and Rosen-berg, R. M. 1981. Constituency, party, and repre-sentation in Congress. Unpublished, National Opin-ion Research Center, University of Chicago.Scammon, R. M., and Wattenberg, B. J. 1970. Thereal majority. New York: Berkley Publishing Cor-poration.Schattschneider, E. E. 1960. Thesemi-sovereignpeople.Hinsdale, Ill.: The Dryden Press.Schmaler, E. 1981. Race and housing: a case study ofdemocratic responsiveness comparing opinionchange to policy response. Unpublished, Universityof Chicago.

Seidman, D. 1975. Simulation of public opinion: acaveat. Public Opinion Quarterly 39:331-41.Sen, A. K. 1970. Collective choice and social welfare.San Francisco: Holden-Day.Shapiro, R. Y. 1982. The dynamics of public opinionand public policy. Ph.D. dissertation, University ofChicago.Smith, T. W. 1978. In search of house effects: acomparison of responses to various questions by dif-ferent survey organizations. Public OpinionQuarterly 42:443-63.1980. America's most important problem-atrend analysis, 1946-1976. Public Opinion Quarterly44:164-80.Taylor, D. G. 1978. The diffusion and change ofpublicattitudes toward some social issues in recent Ameri-can history. Ph.D. dissertation, University ofChicago.Weber, R. E., and Shaffer, W. R. 1972. Public opinionand American state policy-making. Midwest Journalof Political Science 16:683-99.Weber, R. E., Hopkins, A. H., Mezey, M. L., andMunger, F. J. 1972-73. Computer simulation ofstate electorates. Public Opinion Quarterly 36:549-65.Weissberg, R. 1976. Public opinion and popular gov-ernment. Englewood Cliffs, N.J.: Prentice-Hall.1978. Collective vs. dyadic representation inCongress. American Political Science Review 72:535-47.Wise, D. 1973. The politics of lying. New York:Vintage Books.

Documents

1956018