You are on page 1of 27

Southern Political Science Association

The Systematic Beliefs of the Mass Public: Estimating Policy Preferences with Survey Data Author(s): John E. Jackson Reviewed work(s): Source: The Journal of Politics, Vol. 45, No. 4 (Nov., 1983), pp. 840-865 Published by: Cambridge University Press on behalf of the Southern Political Science Association Stable URL: http://www.jstor.org/stable/2130415 . Accessed: 01/03/2013 00:03
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 forms of scholarship. For more information about JSTOR, please contact support@jstor.org.

Cambridge University Press and Southern Political Science Association are collaborating with JSTOR to digitize, preserve and extend access to The Journal of Politics.

http://www.jstor.org

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

The Systematic Beliefs of the Mass Public: Estimating Policy Preferences with Survey Data

John E. Jackson
University of Michigan
This paper develops and estimates a model that relates individuals' policy preferences to three basic terms: (1) a common orientation that affects positions on all items in an issue domain and that explicitly represents the constraint concept; (2) a question-specific term that represents the influence of impacts and interests associated with the explicit policy in question; and (3) party identifications. Each of these components is hypothesized to vary among individuals, based on their social, economic, and demographic characteristics. This model is estimated with the domestic policy items included in the first American National Panel Study (1956-60) using the LISREL statistical method. The results indicate the existence of a large and stable common orientation that strongly influences positions on the jobs, education aid, and health care items, but that is only weakly related to the power and housing item. The model also finds small yet significant policy-specific and partisan effects. The composition and influence of these components are quite stable over the four years of the study. Thus we conclude that the structure underlying individual preferences exhibits considerable constraint and stability, in contrast to some previous analyses with these data.

he continual study of individual political attitudes is justified by and attests to the critical role of these preferences in democratic theory. (For excellent summaries of previous studies, see Converse, 1975; and Kinder and Sears, 1982.) Each additional study hopes to exploit either new data, time periods, and populations, or more refined statistical procedures in its effort to measure attitudinal structure and stability. This study is the latter type. It is a reexamination of the domestic issue questions asked during the first American National Panel Study (1956-60),
T

* This research was supported by grants from Resources for the Future, The National Science Foundation, SES-8023332 and the Guggenheim Foundation, none of whom is responsible for the contents. This research was assisted by the availability of the SRC/CPS 1956-60 American Panel Study, available through the ICPSR, University of Michigan.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

841

and it takes advantage of recently perfected statistical methods to estimate a model of preferences incorporating specific hypotheses developed from current theories about attitude structures. Reexamination of these data is justified on two grounds. Some authors argue there were substantial changes in the American electorate in the ensuing decade (Pomper, 1972; Nie, with Anderson, 1974; and Nie, Verba, and Petrocik, 1979). Analysis of the 1956-60 data forms a baseline for assessing any change. The more accurate are our estimates of the nature and structure of these preferences as well as of the preferences of specific representative individuals in 1956, the better we can observe and describe changes in preferences. A more important reason for continued study of these data with newer statistical tools is their historical importance as the basis for pioneering work on individual preferences. This work led to the generally pessimistic conclusion that the mass electorate is unable to hold and to maintain coherent preferences on a range of widely discussed national issues (Converse, 1964; 1970). Our reanalysis of these data develops new results that expand the earlier findings and conclusions. This is far from the first reanalysis of these data (see Pierce and Rose, 1974; Achen, 1975; Dean and Moran, 1977; Erikson, 1978; 1979), but it is the first to specify and estimate a model with explicit terms representing the hypothesized constructs underlying preferences. Individuals' opinions of specific government policies are systematically related to three separate measurable influences -a common orientation, a policy-specific component, and party identification. The common orientation for any year is hypothesized to affect positions on all policies in that year and explicitly represents the constraint concept. The policyspecific components represent the aspects of preferences that are associated with the particular characteristics and impacts of each policy. Party identifications capture the cue-taking and leadership effects associated with partisanship. Each of these influences is then assumed to vary systematically among respondents based on individual social, demographic, and economic characteristics. Combining these two sets of relations permits us to measure individual policy preferences and to assess their underlying structure. This paper presents the statistical estimation of this structure for the domestic policy questions in the 1956-60 American Panel. We evaluate the various propositions about preferences, including the hypothesis that they do not exist in any coherent fashion, from the size and stability of the parameters in this model. This procedure provides both a definitive test of the models of attitudes and estimates of the preferences held by different individuals.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

842
MODELING

THE JOURNAL OF POLITICS, VOL. PREFERENCES AND THE SURVEY RESPONSE

45, 1983

The complete model developed and estimated in this paper consists of three separately identifiable parts. The first describes people's positions on specific policy items and is derived from current hypotheses about preference formation.' These hypotheses are the core of our model and contain the systematic aspects of preferences referred to in the previous section. The second part relates the different systematic terms to a set of social and demographic variables describing each respondent. These variables assess different individuals' preferences and identify whatever preference changes may have occurred during the panel study. The final part of the model specifies the relationship between these systematic parts and the survey response. (A formal presentation of the model is given in the Appendix.) The roles of the three systematic terms and their relationship to the social and demographic variables (parts one and two) are depicted in figure 1. The boxes in the center of the figure represent the preferences of a given person, on a specific policy item, in each year of the study. The Systematic Preference Components Individual's positions on the specific policies in each year of this study are functions of the person's party identification and common orientation that year (shown at the top of the figure), and of his/her policy-specific determinations (indicated by the boxes along the bottom of the figure). Party Identification. Numerous studies relate a person's party identification to his or her position on specific issues (see Sears et al., 1980, for theoretical arguments; and see Jackson, 1975a; Page and Jones, 1979; Converse and Markus, 1979, for empirical support). This partisanship influence could arise from the individual's need for cues about what positions to adopt, from the influence derived from being part of a political coalition, or from some other psychological force developed from the nature of party identification. It is less important for our model to discuss the causal reasons for any partisan influence than to estimate such influence on preferences. An important implication of a large partisanship influence is that parties are responsible for the development of preferences of the American public and thus should be evaluated in light of this influence. The presence of this influence also means that parties are not just responsive to
I The questions are the four domestic policy items from the 1956-60 American Panel Study: jobs, education aid, health care, and power and housing. The variables are coded so that 0.0 indicates strong opposition to federal government provision of services, 1.0 indicates support for the particular actions, and 0.5 indicates no opinion or don't know.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC FIGURE 1


SCHEMATIC REPRESENTATION

843

SOCIO- DEMOGRAPHIC

SOCIO- DEMOGRAPHIC

SOCIO- DEMOGRAPHIC

COMMON 1956

PARTY 956

COMMON 958

PARTY 1958

COMMON 960

PARTY 960

rJOBS

ED. AID IIPOWER

IIHEALTH

IJOBS

ED. AID IIPOWER

IJOBS

ED. AID IIPOWER

IIHEALTHI

JOBS

EDUCATION

~~AID

POWS HOUSING

HEALTH CR

SOCIO - DEMOGRAPHIC

and aggregations of the distribution of policy preferences of the electorate in the hierarchical manner of the formal spatial models. Parties help determine this distribution by their actions, and so make the electoral process nonrecursive. Common Orientations. Recent work (Sears et al., 1979; Kinder and Kiewiet, 1979; and Sears et al., 1980) depicts policy preferences as derived from an individual's broad symbolic or ideological orientation. There are several explanations offered as the basis of this component, derived from references to ideologies about the federal government's role in providing services, to statements about people's fundamental values such as a concern for greater equality, and to respondent's identification with the interests of specific social groups. (For a very clear description of these different derivations, see Kinder and Sears, 1982.) We are not in a position to discriminate among these alternative sources for a common orientation, but we are able to assess its influence on each particular policy position.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

844

THE JOURNAL OF POLITICS, VOL.

45, 1983

Individual's orientations, such as the desire for the federal government to provide services that redistribute income, should motivate complimentary positions on a range of specific policy items. The common orientation term explicitly models the constraint concept. Originally, constraint meant the presence of an underlying orientation from which positions on a range of specific policies were derived. This is a much more specific, and theoretically more interesting, concept than its operationalization as the simple bivariate correlation among preferences on pairs of issues. Such correlations may also arise from other sources, such as a similarity of self-interested motives or the effects of party identification on preferences, instead of from the important presence of a common basis for preferences. Policy-Specific Considerations. These components assess people's propensity independently to develop their positions on specific policies. The policy-specific terms are shown at the bottom of figure 1. There are a number of arguments for these terms. Sears et al. (1980) develop these influences from the personal impact of a given policy, such as the individual's being unemployed or eligible for a government health care program. Other reasons for such a derivation of policy-specific preferences exist as well. Musgrave and Musgrave (1976), for example, argue that individuals may want policies that subsidize other people's consumption of certain commodities, such as health care, housing, or education, but not ones that redistribute income per se. In this case there is no common structure to preferences; specific policies are supported or opposed on the basis of one's feeling about other people's consumption of certain services and how those services should be provided. Again, our model cannot discriminate among these explanations, but it can assess the presence and importance of policy-specific considerations. Modeling Common and Policy-Specific Components The second part of the model relates positions on the common orientation and policy-specific terms to the individual characteristics shown in table 1 so that we can evaluate the preferences of different respondents. This part of the model appears in figure 1 as the links from the sociodemographic boxes to the other components. The coefficients in these equations permit us to predict the positions of particular individuals and in what manner party identifications, common orientations, and specific considerations, and thus preferences, vary with income, race, religion, etc. There is, unfortunately, little hard theory to guide the selection of the individual variables in the underlying models. The objective is to pick

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

OF MASS PUBLIC SYSTEMATIC BELIEFS THE


TABLE 1 EXOGENOUS DEMOGRAPHIC VARIABLES Age (in tens of years) Race Catholic Jewish Easta Southb Central City Rural
a
b

845

Education (in tens of years of schooling)c College Degree Union Income (in tens of thousands of 1960 dollars) Father's Party Identification-Democratd Father's Party Identification - Independentd Born prior to 1910 (entered the electorate prior to 1932)d

New England, N.Y., N.J., Pa., Del., Md., W.Va., and Washington, D.C. Va., N.C., S.C., Ga., Fla., Ala., Miss., La., Ark., Tenn., Tex. c Included in exponential form. d Included only in the party identification equation.

variables from the survey that are strongly related to people's ideological orientations toward the federal government's role in our society, to their evaluations of specific domestic policy's impacts, and to their group associations. In most cases, these considerations lead to overlapping sets of variables. For example, education, age, and income differences are related to different ideological views, and they are simultaneously associated with the expected impact of jobs, education, and health programs. Similarly, variables such as race and religion may be associated with group associations and identifications, or they may indicate differences in basic orientations and ideologies. Fortunately, these considerations are not crucial for the purposes both of estimating the size and influence of the three underlying components and of comparing the preferences of different individuals across issues and over time. Consequently we resolve the dilemma by including all the variables in table 1 for each underlying component, except as noted in the party ID equations. The most important consideration is to make sure we have included variables that systematically vary with preferences. The manner in which they vary will help us to identify and estimate the characteristics and influence of the underlying components. The relations between preferences and the socio-demographic variables have several important functions in the model. Besides permitting us to ascertain the preferences of different respondents, correlations between these variables and the responses to each of the specific policy questions contain statistical information with which to identify the common orientation and policy-specific components and to estimate their influence on preferences. For example, similar covariance patterns between the sociodemographic variables and responses to different policy questions indicate an important common orientation.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

846

THE JOURNAL OF POLITICS,

VOL.

45, 1983

The relationship between the underlying components and the sociodemographic variables includes a stochastic term representing the part of people's preferences not explained by the socio-demographic variables. Although stochastic, such terms represent important parts of our preference structure. The stochastic element for the common orientations and policy-specific terms contains any aspect of an individual's underlying components arising from personal experiences, such as health problems, or from contextual circumstances, such as the wealth and stability of the local economy, not measured by the individual variables. Model Structure and Interpretation The structure just described contains most of the commonly proposed preference models, and the relative prominence of the different models will be determined by the size and stability of the various parameters in the structure. There are four major comparisons: 1. Constraint is measured by the magnitude both of the coefficients relating common orientations to preferences and of those relating orientations to the individual characteristics. Furthermore, if the composition and influence of the common orientations are stable, the result will be coherent, stable preferences. 2. If differences on the policy-specific components are large, stable, and influential, relative to the common orientations, we will have stable preferences but with little constraint. 3. If variations in party identifications are large, and the sole determinants of preferences, the proposition that preferences are strictly derived from the cues and structure provided by individual party identifications will be supported. 4. If the coefficients in the models for common orientations and policyspecific terms and the coefficients relating these terms and party identifications to specific policy preferences are small and insignificant, we will reject the hypothesis that preferences have any systematic structure to them. This last result will be the strongest evidence for the argument that members of the electorate do not have structured preferences on public policy issues. Our statistical objective is to estimate the various parameters as accurately as possible in order to examine these different propositions. Temporal Preference Changes The model has two explanations for temporal preference changes. The first is change in an individual's position on any of the underlying components. For whatever reasons, people's views of the federal govern-

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

847

ment's role in providing social services, that is, their common orientation; their evaluations of individual services and personal impacts, that is, the policy-specific term; and/or their party identifications may shift. (The policy-specific components are assumed to be stable over time in the basic model, but this assumption is tested and altered for specific policies.) These shifts may occur for only some people, resulting in different relationships between positions and our exogenous variables. For example, differences in preferences related to income may change but those related to race, education, or region may remain the same. Thus, contrary to previous studies that assumed any changes are equally likely for all people (Achen, 1975; Erikson, 1978; 1979), or are nonexistent (Converse, 1964; 1970; Pierce and Rose, 1974), this model relates changes to social and demographic characteristics, and thus permits changes to differ among respondents. Changes in the omitted effects are assessed by the temporal correlations of the stochastic terms. Positions on specific policies may also shift because of changes in the relative influence of the common, policy-specific, and party identification components. Increased common orientation influence relative to that of the policy-specific term means both greater differences among people whose ideologies differ and reduced variation among people whose policy-specific considerations differ. An increased common influence will produce higher correlations among positions on all items, and so will lead to observations of more ideological politics and greater constraint. Increased party influences result in greater differences among Republicans and Democrats on specific policies. If such shifts occur on all topics, the result will be a greater partisan character to the distribution of preferences and the appearance of party-derived constraint among the electorate. The Survey Response The last part of the statistical model describes the relationship between preferences and the response to the survey questions. The response model includes a random element unique for each response and presumably unrelated to systematic preferences.2 This term measures the deviation of any measured preference from that described by the three components. These deviations derive from measurement errors associated with the survey instrument, from deviations induced by respondent randomness, from the effects of omitted variables, or from whatever error sources are
2 We use random to denote the unspecified part of the survey response model. This will distinguish it from the unspecified parts of the common orientation and issue-specific components, referred to as stochastic terms. Both terms are stochastic, however, in that they are randomly and independently drawn from some distribution for each respondent.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

848

THE JOURNAL OF POLITICS, VOL.

45, 1983

represented by the stochastic term in any linear model. Our point is that we cannot distinguish statistically among these sources of randomness; moreover, it is much more important to estimate the systematic parts of the model and to base our inferences about individual preferences on these estimates. The sole concern here is to recognize the presence of the random terms and to estimate their variance at the same time that we estimate the components in the systematic part of people's preferences in order to get accurate estimates of the entire structure. The model of the survey response estimated here also includes a systematic response bias, such as an agree or do not know, regardless of the person's actual expected response. This systematic measurement error exists in addition to the previously described preference structure and random terms. Its presence, if not removed, seriously biases any measure of stability or constraint. If the response biases to the same question at different waves or to different questions in the same waves are similar, these measures will be overestimated. Thus, it is important to remove the effects of the biasing terms. Previous analysis of the 1956-60 panel data (Jackson, 1979) finds a significant negative relationship between education level and the probability of giving a biased response. These probabilities vary by question and year, but generally range from 0.3 for people with an eighth grade education to nearly zero for those with a college education. This analysis uses the results of that study to purge responses of any expected systematic errors. Our resulting measures of constraint and stability should be lower than those found by Achen and Erikson, and the estimated structure should better describe true preferences. Empirical Estimation The statistical model is a combination of a factor analysis and a hierarchical set of regression equations (see Appendix). The common orientations and the policy-specific components are latent, unobserved constructs, as are the components in a factor analysis, and the responses to the issue and party identification questions are the observed variables. As shown in figure 1, this model differs from the more common factor analysis in two ways. First, each observed variable is related to only a small and prespecified set of the latent variables rather than to all latent factors.3 For example, preferences on a particular policy in a given year are unaffected by party identifications and common orientations in other years and by the other policy-specific components. Second, this model
3This use of the factor analytic model corresponds to Joreskog's confirmatory factor analysis model, rather than to the traditional use, which he labels the exploratory model Joreskog, 1969).

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

849

specifies that the latent variables are functions of the observed sociodemographic variables. Thus the common orientations, policy-specific components, and party identifications are cast as intervening variables in a hierarchical system, with the individual characteristic variables treated as exogenous and the observed survey responses considered to be the resulting endogenous variables. The individual equations relating positions on specific policy questions (purged of the systematic response bias) to the common orientation, policy-specific, and party identification components, and those relating these components to the socio-demographic variables are linear regression equations. The coefficients in these equations measure the expected difference in positions for different orientations, question-specific considerations, and party identifications, and the expected component differences for various individual characteristics. This structure, with the purged responses, conforms to the linear structural equation systems described by Joreskog (1973). The program developed by Joreskog and Sorbom (1976), referred to as LISREL, can be used to estimate the full structure, using all the questions in the panel study, once the response biases are removed. (For a brief discussion of LISREL, see Hanushek and Jackson, 1977, ch. 10.) Party identifications are measured by the seven-point CPS scale and vary from zero for strong Republicans to one for strong Democrats. They are treated as endogenous variables to eliminate biases that might result from the effects of preferences on identifications. (We do not estimate these reciprocal effects, however.) This treatment is accomplished by relating party identifications to the full set of exogenous variables, including the variables presumed to relate to identifications and not to preferences - father's identification and whether the person reached voting age prior to 1932. This procedure parallels that of traditional two-stage least squares. The equations relating party identification in a given year to the exogenous variables are comparable to the reduced form equation in the first stage of the two-stage procedure.
ESTIMATED PREFERENCE STRUCTURE AND SURVEY RESPONSE MODEL

The statistical results provide an excellent fit to the observed responses and indicate both the presence of sizable systematic influences on individual preferences and that these influences create considerable constraint and stability.4 The common orientations are large relative to the policy-specific terms, and exert a substantial and stable influence on jobs,
4The chi-squared statistic measuring the goodness of fit is 98.4 with 105 degrees of freedom, which has a significance level of 0.666. Thus we cannot reject our estimated model as being the true model, since we could get a worse fit by chance two-thirds of the time.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

850

THE JOURNAL OF POLITICS, VOL.

45, 1983

education aid, and health care preferences. The common orientation influence on preferences about the federal government's role in providing power and housing is small but stable. The policy-specific components, although smaller than the common orientation, are important nonetheless and exhibit some interesting differences among policy items. The most significant variations concern the power and housing question, where the composition of people's policy-specific evaluation changes each election year. This variation contrasts strongly with that in the other policy areas, where the composition of the specific term remains constant, as hypothesized in the basic model. Finally, party identifications exert an important but far from dominant influence on policy preferences. The most substantial determinants of preferences, however, are variations in the exogenous variables related to an individual's common orientations. Policy Preference Equations The coefficients assessing the influence of party identifications, common orientations, and policy-specific terms on policy preferences are shown in table 2. (All but the education aid-specific coefficients are statistically significant at the 0.001 level.) Each row in the table corresponds to the coefficients in a simple regression model relating preferences on the specific policy in a particular year to the underlying components.5 Interpretation of the size and composition of the common and specific components, and thus of how fully to interpret the coefficients in table 2, is given in the following section. The important point here is to compare the relative size of these coefficients for each policy and their stability for different years. Differences in people's common policy orientations produce corresponding and nearly equal variations in responses to the jobs, health care, and education aid questions, but not to the power and housing question, where the resulting differences are a third as large. These relationships indicate considerable constraint among the former issues, but not for pairs involving the power and housing question. These effects are quite stable in magnitude, except for education aid in 1958.

5 These coefficients are analogous to regression coefficients and are interpreted as the expected difference in policy preferences for a unit difference in people's latent components or party identifications. Thus for education aid in 1956, we expect two people whose difference in common orientations equals one to differ in their support of federal education programs by 0.84. Similarly, the party identification coefficient of 0.21 in the same equation indicates the expected difference in support between a strong Republican and a strong Democrat. Certain coefficients have been set equal to 1.0 in order to scale the unobserved components. This specification determines the absolute size of the component and its coefficients in the other equations, but leaves the relative size of all the coefficients unchanged.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

851
2

TABLE ESTIMATED

PREFERENCES EQUATIONS

COMMON POLICY - YEAR ORIENTATION

POLICY SPECIFIC

PARTY IDENTIFICATION

Jobs- 1956
1958 1960 Education - 1956 1958 1960 Health - 1956 1960 Power & Housing - 1956 1958 1960

1.OOa
1.00a 1.00a

1.0Oa 1.74 1.22


1.00a

0.19
0.15 0.22 0.15 0.22 0.21 0.24 0.30 0.22 0.24 0.22

0.84 1.18 0.84 0.99 0.95 0.39 0.34 0.32

0.45 -1.77
1.00a

1.74
1.00a

1.12 1.02

Coefficients set to one to scale structural components.

The policy-specific effects are statistically significant, indicating that preferences are influenced by considerations related only to that particular policy. These effects are not as stable as those of the common policy orientations. This instability is particularly evident with education aid preferences, where the directional effect is reversed in 1960, and with opinions about public power and housing, where the structure of the component changes every year. The instability of the education aid component effects prompted a test of whether the composition of this term changed in 1960, as did the power and housing component. The results of the test were negative, leading to the conclusion that only the effects of this component changed between 1958 and 1960.6 The effects of the policy-specific terms are statistically significant, indicating that preferences are influenced by considerations related only to that particular policy. The party identification effects are about 0.2, with the exception of 1960 health care preferences. Thus the likelihood of a strong Democrat agreeing with the policy proposition is about 0.2 greater than that of a strong Republican. The party influences increase slightly, by about 0.05,
6 Subsequent analysis shows that education aid preferences are related to a common civil rights orientation, which does change between 1958 and 1960. This omitted influence likely accounts for the dramatic coefficient change noted here, which disappears in the expanded model.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

852

THE JOURNAL OF POLITICS, VOL.

45, 1983

between 1956 and 1960 for all but the power and housingquestion. When we compare these influences with those of the individual exogenous variablesdiscussedin the next section, we see that party is one of the most influential single variables, although the combined effect of all the sociodemographicvariables and of common orientations is greater for most typical individuals. Common Orientationand Policy-SpecificComponents The relationsamong the common orientationand policy-specificterms and the socio-demographic variables provide important information about the composition of the components and the relative positions of each respondent on these underlying dimensions. Each component is related to the separate social and demographicvariables to estimate the expected differences in position on these underlying components of individuals whose characteristicsdiffer, e.g., people of different incomes, race, etc. The magnitudeof these individual differencesdeterminesthe relative size of each component, which is summarizedfor our sample of individualsby the variance of the different components. The larger the equation'scoefficients, the more people differ in their orientationsor in their specific evaluations, and consequently the greater the sample variance in these terms. When this informationis combined with the information in table 2, which concerns the effect of each component on preferences, we can then infer the relative importance of common and specific considerations. In table 3 we presentthe standarddeviationsfor both the total component and the stochastic term of the different components and the differences in the expected values of the basic componentsof two representative but quite different respondents.7 RespondentA is assumedto be fifty-five, college educated, white, Protestant, earning $20,000 per year (in 1960), and living in a midwesternsuburb. RespondentB is twentyfive, black, with an eighth gradeeducation, earning $5,000 per year, and living in an eastern central city. The components'standard deviation indicates there are much greater differences among people's common orientationsthan there are in their policy-specificevaluations-by a factor of two in the cases of jobs, education aid, and health care. Only the power and housing specific componentshave variationsthat approachthose of the common orientations.
7We express these variations by their standard deviations, rather than by their variances, because the ratio of the standard deviations of two components can be compared to the ratio of their coefficients in the preference equations shown in table 2. Comparisons of such ratios are perfectly analogous to comparisons of standardized regression coefficients. An appendix with the estimated coefficients is available on request from the author.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

853
3
AND DIFFERENCES

TABLE STANDARD DEVIATIONS

OF COMPONENTS

FOR Two

CHARACTERISTIC

RESPONDENTS

STANDARD DEVIATION COMPONENT Common Orientation TOTAL STOCHASTIC

DIFFERENCE" PERSONS BETWEEN A AND B

1956 1958 1960 Policy Specific Jobs Education-Systematic


Education-Stochastic Only

0.228 0.230 0.230 0.118 0.084


-

0.202 0.202 0.212 0.095 0.077


0.183

0.517 0.572 0.434 -0.050 -0.076 0.042 -0.091 -0.034 -0.077 0.416 0.483 0.520

Health Care Power & Housing 1956 1958 1960 PartyIdentification 1956 1958 1960

0.130 0.167 0.195 0.170 0.339 0.343 0.343

0.109 0.155 0.155 0.155 0.283 0.277 0.283

a Expected differencesin values of each component for representative individualsdescribedin the text. Entriesindicatehow much more personB supportsfederalgovernment social programs and identifieswith the Democratsthan does personA.

These standard deviations are specific to this sample, but when taken with the coefficients in table 2 indicate that among these respondents likely differences in common orientations are more important in determining preference variations than are differences in policy-specific evaluations and in party identifications. (The easiest way to make these comparisons is to multiply the coefficients in table 2 by the standard deviation of the appropriate component.) A better way to compare the differences among people on these underlying components is by the expected values for representative individuals. For our two individuals described above, the differences in their positions are shown in the third column of table 3. The differences in expected party identifications and common orientations are quite large, while those for the policy-specific evaluations are much smaller. Person A, the older, wealthier white, is much less likely to favor any form of federal government programs and is more likely to be a Republican

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

854

THE JOURNAL OF POLITICS, VOL.

45, 1983

than is person B, the younger, poorer, central city black. These underlying differences translate into substantial differences in positions on specific policy issues. According to the coefficients from the equation for the 1956 positions on whether the government should provide jobs, for example (the first line table 2), the expected difference in these two persons' positions is 0.546, with person B being much more supportive of government efforts to provide jobs. Most of this difference is accounted for by the difference in the two individuals' positions on the common orientation (1.00*0.517). We see from comparing these two individuals an example of what was noted above, that individual differences in common orientations are very important in accounting for differences in people's positions on government policies. Only for power and housing preferences are issue-specific considerations equally important in determining individual positions. The individual variation in the specific components is large relative to that for the other policy-specific evaluations, and the policy-specific concerns play an important part in determining individual positions (as seen in the coefficients in table 2). Furthermore, the power and housing specific component exhibits little stability between elections in contrast to those for the other policies.8 We conclude that positions on the power and housing question are determined more by an unstable policy-specific component than by the common orientations that are influential in positions on the jobs, education aid, and health care policies. The second comparison concerns the relations among the underlying components, and particularly the year-to-year correlations of the common orientations. These correlations, shown in table 4, determine the stability and coherence of specific policy preferences. By far the largest systematic relations are among the three common orientations, which have correlations between 0.7 and 0.8 for the two-year intervals and almost 0.7 for the four-year interval. Thus we find that the common orientation is both large and stable. This result, combined with its large and relatively stable influence, produces a set of preferences with considerable constraint and stability for the jobs, education, and health questions. The policy-specific components are relatively independent of the common policy orientations. This results, in part, from the specification that the stochastic terms in the specific components are independent of the stochastic terms in the common policy orientations. However, even the
8 The correlations among the systematic parts of the power and housing-specific component illustrate this instability. The 1956 and 1958 parts have correlations of 0.53, while the 1960 part's correlations are -0.21 and 0.10 with 1956 and 1958 respectively. Statistical tests reject the null hypothesis of no difference in the systematic parts of the component at the 0.01 level.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

855
4

TABLE CORRELATIONS

AMONG STRUCTURE COMPONENTS

COMMON POuCY

ISSUE SPECIFC

POWER

1956 1.00

1958 0.80
1.00

1960 0.66
0.74 1.00

Jobs 0.11
0.03 -0.01

Health 0.08
0.01 -0.02

School -0.14
-0.10 -0.03

1956 -0.09
-

1958 -0.06 -

1960 -0.02

1.00

0.29 1.00

-0.13 -0.11 1.00

-0.18 -0.16 0.12

-0.13 -0.12 0.09

0.07 0.03 -0.05

correlations among systematic parts, exclusive of the stochastic terms, are still not very strong, which suggests that the policy-specific components measure an aspect of preferences that is relatively independent of the common policy orientations. The specific components for the jobs and health questions are moderately positively correlated (0.3), but the others, including the three power and housing terms, are generally unrelated to each other, further validating our view of these evaluations as policy specific.

Coherenceand Stability
We can easily summarize these findings about the structure of individual preferences by the across question and over time correlations of respondents' positions. These correlations, shown in table 5, are based on the systematic components and exclude the random term. They hold only for the individuals included in the study and would differ for different samples, but they do give an idea of the relative coherence and stability among these people's preferences. The relatively high over time correlations confirm our observations about the stability of the components and about their influence on preferences. These two-year correlations vary from 0.85 to 0.90 and are about 0.80 for the entire four-year interval, with the exception of education aid in 1958-60 and 1956-60, which is attributable to the changing influence of the issue-specific component. These correlations are about 0.1 lower than those estimated by Erikson and by Achen. The cross issue correlations confirm the earlier observations that the jobs, education aid, and health care issues compose a fairly consistent and cohesive group and that positions on the question about who should provide power and housing are only weakly related to these other preferences. The correlations among the first set range, with one exception, between 0.6 and 0.7, while the correlations of the power and hous-

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

856

THE JOURNAL OF POLITICS, VOL.

45, 1983

TABLE SAMPLE CORRELATIONS

5
AMONG PREFERENCES

OvEn TIME 1956-58 Jobs Education Health Power 0.87 0.86 0.91 1958-60 0.84 0.74 0.85 1956-60 0.79 0.63 0.81 0.87

AcRoss QUESTIONS 1956 Health-Jobs Education-Jobs Education-Health Power-Jobs Power-Health Power-Education 0.86 0.61 0.59 0.35 0.36 0.33 1958 0.63 0.22 0.32 1960 0.72 0.61 0.52 0.42 0.37 0.33

ing responses with the other preferences are between 0.3 and 0.4. These results illustrate in a different way the observations made earlier. There is a noticeable constraint among preferences on the first three domestic issues, accounted for by the large size and influence of the common orientation component, but this constraint does not extend to responses to the question about public provision of power and housing, which are only weakly related to the common orientation.

The Random Term


The third part of our statistical model relates the systematic aspect of individual preferences just described to the survey response for the specific questions used in the study. This equation contains both a systematic error response bias term and the conventional random term. (For results estimating the likelihood and direction of any response bias see Jackson, 1979.) The random term, which is unique to each question in each interview, is completely independent of the common policy orientation, the policy-specific component, and the party effects. This random term includes any randomness in respondents' preferences, all measurement errors, and other contributions to deviations from the systematic relations shown in table 2. Our interest is in the variance of these terms and in what that variance indicates about the reliability of

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

857

survey responses as measures of the systematic aspects of individual preferences. The striking result is the high proportion of the variance that is totally unique to each question each time it is asked. Proportions are near 0.50 indicating a large nonsystematic term. The exceptions are health care and education aid in 1960, with proportions close to 0.35, and all responses to the power and housing question, where the proportions are consistently about 0.70. These results are close to those of Achen and Erikson in order of magnitude. As suggested previously, it is virtually impossible to interpret substantively these variances because of the many factors they represent.
CONCLUSIONS

Our findings about the systematic parts of individual preferences have several implications for discussions of individual and collective political behavior on both substantive and methodological topics. The most significant result is the observed presence of a large, stable, and influential common orientation that accounts for a substantial part of people's preferences on the domestic policy items of government-provided jobs, education aid, and health care. Thus, contrary to some earlier interpretations of these data, but consistent with the hypothesis of Sears et al. (1980), we find the existence of a common component that people use to develop their positions on a number of specific issues. For the policies included here, this dimension seems to be the question of government provision of services. However, opinions about who is best at providing power and housing, thought to be the most representative measure of this ideology, are only weakly related to the common orientation. Therefore, earlier analyses using this question to explore the hypothesis of whether the American people had any ideological content to their positions were unfortunate in selecting what now turns out to be a poor measure of this component and of the public's ideological structure.9 People's issue preferences are not totally a function of the common orientation, however. Preferences on all four items are related to a component specific to each policy. The policy-specific terms' size and influence are smaller and more varied than those of the common orientation, but they are important nonetheless. This result implies that posi9 This result is particularly important for Nie (1974) and Nie et al. (1979) who claim that preference constraint among the American public increased significantly between 1960 and 1964. Their pre-1964 measures of constraint on issue pairs involving domestic policy are based on average correlations that include the power and housing responses, which are only weakly related to the issue dimension being estimated. This attenuation of estimnatedconstraint is in addition to that caused by the large random components of these responses.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

858

THE JOURNAL OF POLITICS, VOL.

45, 1983

tions on any issue and responses to any policy question are influenced by concerns specific to that policy and that no single question can be used to assess the existence of and positions on the common orientation. A substantively important question is whether circumstances related to a campaign or to events surrounding an election affect the relative influence of these policy-specific components. For example, will a recession, as we had in 1958, increase the influence of the jobs-specific component, as we observe for 1958? Or, will extensive emphasis during the campaign on one particular policy and the way that policy is characterized, e.g., in personal impact terms rather than ideological ways, increase the influence of the issue-specific component? We have too few issues, circumstances, and campaigns to address systematically these hypotheses, but they are substantively important and warrant investigation with comparable structural models for other issue areas and during other campaigns. The evidence of a systematic structure to individual preferences has implications for the many efforts to estimate the relationship of policy preferences to other aspects of political behavior, such as party identification and voting decisions. Research based on correlation and regression analyses using the observed responses to these policy questions shows little influence of preferences on behavior (Nie et al., 1979, p. 34; Schulman and Pomper, 1975). These analyses treat the observed responses, which include the response bias and random terms as well as the systematic structure, as the true measure of preferences. More recent work using alternative statistical techniques (Jackson, 1975a; 1975b; Page and Jones, 1979; Markus and Converse, 1979; and Franklin and Jackson, 1983) bases estimates of these influences on the relation between measures of behavior and the systematic structure described here, and find much larger relationships.10 Hanushek and Jackson, for example, show a doubling of the coefficients assessing the influence of preferences on party identification and voting decisions in the 1964 election (0.44 to 0.77 and 0.39 to 0.90 respectively) using two-stage least squares rather than ordinary regression (see Hanushek and Jackson, 1977, pp. 246 and 271). The results showing larger influences of preferences on political behavior suggest that the presence of the random term in the observed responses attenuates the statistical relationship and obscures the true effect of preferences. Thus earlier studies giving policy preferences a minor role in determining various political behaviors may have seriously underestimated these effects because of the choice of statistical procedures and the implicit
10 The alternative structural equation techniques, such as two-stage and three-stage least squares, are appropriate if one wants to purge observed explanatory variables of any random elements that obscure the true values of those variables. See Hanushek and Jackson, 1977, pp. 288-89.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

859

assumptions that the random term present in the preference measure was an integral part of individual preferences. Finally, the observed coherence and stability of individual preferences have implications for discussions of the stability of aggregate electoral behavior. The preference stability observed in table 5 approaches the stability estimated for party identifications during this period. (Estimates of individual party identification stability range from 0.85 [Converse, 1964] to 0.92 [Asher, 1974], depending upon the measure used and the assumptions made about the size of the random component in measures of party identification.) These two studies, which relate systematic preferences and relative party positions to individuals' party identifications, combined with our observations about the stability of domestic policy preferences between 1956-60, imply that the relative stability in party identifications observed with both survey and aggregate data during this period is the result of stable preferences and relatively unchanging party positions. With preferences and identifications being simultaneously related and exhibiting high and nearly equal levels of stability we have the picture of an electoral system in relative equilibrium during the late fifties. Were individual preferences or the relative party positions to change, this equilibrium would be upset, and we would observe changes in party identification and voting choices. Findings of a relatively coherent and stable systematic component to policy preferences, strongly related to positions on a common orientation, are different from earlier interpretations of these data. These results indicate that we were correct not to reject completely the null hypotheses of constraint and stability, in spite of the earlier studies. They also show the power of more sophisticated statistical methods to provide greater discrimination among the various influences on preferences. However, we have only examined one set of policies, in one rather short time period. In addition, given the prominent role of these policies in our political culture and the relatively stable positions advocated by each party since the New Deal, we can argue that if constraint and stability of preferences were to exist, they would effect these policies during this period. We thus have the easiest test for those arguing for structural and coherent preferences. More difficult tests await.

APPENDIX

This appendix describes in more formal detail the model of policy preferences estimated in the paper's text. Although complex in its notation, the structure of the model is a quite simple set of hierarchical, nonrecursive linear equations. Certain of the intervening variables in the

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

860

THE JOURNAL OF POLITICS, VOL.

45, 1983

hierarchy are latent or unobserved, as is common in factor analytic models -a fact that complicates statistical estimation. This appendix is organized to follow the hierarchical structure of the model rather than the presentation in the text. Underlying Components The first parts of the model are the three sets of equations describing individuals' common orientations, policy-specific components, and party identifications. We denote these components as Y1t, Yv1, and Y3t respectively, with t = [1 = 1956, 2 = 1958, 3 = 1960] and j = [1 = Jobs, 2 = Education, 3 = Health, 4 = Power and HIousing]. The equations relate the values of these components to the individual socio-demographic variables shown in table 1, denoted as X, and to a stochastic term, and they follow the conventional linear form, Ylt = XB1t +
Y2j U1t, U21,
U3t.

(Al) (A2)
(A3)

= XB2j +

Y3t

= XB3t

The coefficients symbolized by B1, B2, and B3 assess how we expect values of these components to differ for differences in each of the individual variables in X. The stochastic terms U, to U3 represent the aspects of these components not accounted for by the individual variables and the linear model. (We have omitted the individual subscript, i, for Y, X, and U for simplicity of exposition.) Given that the panel study covers three election years and four policy questions, there are three separate equations represented by equations Al and A3 (as denoted by t) and four by A2 (as indicated by j), and for a total of ten equations in the underlying model shown in figure 1. The four policy-specific components differ for each policy, but are presumed not to vary over time. (This latter specification is tested and rejected for the power and housing question.) We include separate structural equations explaining party identifications, thus treating them as an endogenous influence. (We estimate the coefficients in these equations in the statistical procedures but do not report them, as they are not relevant for the discussion of preferences.) This method eliminates bias in the estimates of party identifications' influences on preferences attributable to the reciprocal effects of preference-based party evaluations on identifications (see Franklin and Jackson, 1983, for evidence of such effects). Included among the individual characteristic variables, X, are three variables related only to party identifications - father's identification and reaching voting age

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

861

before 1932. These three variables do not directly influence preferences, so they identify the effect of party on preferences.11 (The entries in B1A and Bv corresponding to these variables equal zero.) We summarize this first part of the model in matrix form as = Y= (Y1t,Y2j,Y3t) X(B1t,Bz1,B3t)+ (UltU21,U3y) = XB + U (A4); where Y, X, B, and U are the appropriately dimensional matrices for the ten underlying equations represented by equations A1-A3. Policy Preferences The influence of the above three components on individual policy preferences is also represented as a linear equation. We denote the preferences on issue j at period t as Wj, and write this expression as

+ Y1tA1j, Y2jA21,+ Y3tA3j, + ej,

(A5)

The coefficients A11i, A2j, and A3jt indicate how much we expect preferences to differ for differences in common orientations, policyspecific considerations, and party identifications, while ej, is the random term that includes for variations in responses not accounted for by the previous three explicit terms. Four questions and three time periods would mean that we have twelve separate equations represented by A5. However the health care question was omitted in 1958, giving eleven. We have double subscripted the A1, A2, A3 coefficients to indicate that they vary by policy item j, and by year t. (The estimated values are shown in table 2.) Thus we permit each component to have a different effect across policies and over time. Lack of such variation is the basis for our conclusion of constraint and stability. We write the policy preference equations in A5 in matrix form as

W = YA + e.

(A6)

W is the matrix of responses to the eleven policy questions; Y is the matrix of implicit values for the ten underlying components; A is the matrix of coefficients relating W to Y; and e is the matrix of eleven implicit random terms.
11 This procedure is directly comparable to two-stage least squares, with the equation for party identifications corresponding to the first stage, and our estimates of the effect of party identifications on preferences being the second stage. This is the conventional means for overcoming possible simultaneity bias (see Hanushek and Jackson, 1977, chs. 8 and 9).

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

862 Model Estimation

THE JOURNAL OF POLITICS, VOL.

45, 1983

The above set of equations constitutes a hierarchical, nonrecursive, linear system. Unfortunately, these equations cannot be estimated by the simple linear least squares procedures commonly used to estimate such structures, e.g., two-stage least squares. The intervening endogenous variables for the common orientations and policy specific terms, Y1,and Yv,, are not observed, so we cannot directly estimate equations Al and A2 and thus cannot estimate A5 in the second stage. The statistical problem then is how best to use the observed socio-demographic variables, X, and the observed responses to the policy and party identification questions in the survey, W, and P, respectively, to estimate the parameters in equations Al1, A2, and A5. The description of the procedure used relies on the matrix form of these expressions shown in equations A4 and A6. The statistical procedure is described in detail in Hanushek and Jackson (1977, ch. 10) and Joreskog and Sorbom (1976) and is only summarized here. It uses equations A4 and A6 to derive expressions for the expected variances and covariances among the observed variables, denoted in matrix form as Ex, Exw,and E,, as functions of the unknown parameters. We then use the variances and covariances among the observed variables, denoted as Sx, Sxw,and Sw, as estimates of Ex, Exw,and Ew, and calculate the parameter values that best fit the observed data. (Note that Sx directly estimates Ex.) To obtain expressions for .xw and Ew, note that substituting the expression in equation A6 for Y in equation A4 gives W= YA + e = (XB + U)A + e = XBA + UA + e. (A7)

We now assume that: 1. all stochastic terms are independent of the exogenous socio-demographic variables, i.e., X is independent of U and e (this is the assumption on which all regression analysis is based); and 2. the stochastic terms in the underlying components are independent of those in the policy preference equations (e is independent of U1, U2, and U3). With these assumptions, Fl. = E(X'W) = ExBA, and Ew = E(W'W) = A'B'X'XBA + A'E(U'U)A + E(e'e)
= A1B'2xBA + A1E2A +
Se

(A8)

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

863

We pick the values for A,B, E, and E. that best fit the observed variances and covariances, Sx. and S,. Best fit in this case is defined as maximizing the likelihood of obtaining the observed data under the assumption that U and e are distributed multivariate normal. We further assume that each ej, is independent of the others, so that Se is diagonal. If these assumptions hold, the maximum likelihood procedure provides a statistical test of the model's fit. This test, based on the chisquared distribution, indicates the probability of getting a worse fit by chance if the estimated model is correct. The higher that probability, the better the fit and the less likely we are to reject the estimated model as the correct one. The statistical procedures followed here also include an adjustment to purge the responses of the expected systematic response bias estimated in Jackson (1979). The survey response model, including a possible biased response Rjt for question j at time t for person i is Zi= V WijPrij,
+

Rjt(1-Prij,),,

where Zij, is the response of person i to question j at time t, Wij, is the person's "true" response given by equation A5, and Prij,is the probability the person responds with a true response rather than the biased response Rjt. The estimates for Rj, and Prij,reported in Jackson (1979) are used to purge the observed responses of any systematic bias. By denoting these estimates by Rjt and Prij, we have W t = I
Prijj -

R ]-

where qijtis any random error introduced by the estimates for Rjt and Prij,. This expression gives

w*t=

Rjj

_ Rjt

Wijt

qijt

Yi1tAot

Yi21A2ft +

Yi3tA3ft + eij,

qij,

The analysis discussed in the text applies to the adjusted, or purged, responses, denoted here as Wi*. This qij, term is treated as another type of measurement error which is independent of X and U, but not necessarily of eij,. The structure shown in equation A8 still applies, and we estimate the relevant parameters with the LISREL program.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

864

THE JOURNAL OF POLITICS, VOL.

45, 1983

REFERENCES Achen, Christopher (1975). "Mass Political Attitudes and the Survey Response." American Political Science Review 69: 1218-31. Asher, Herbert B. (1974). "Some Consequences of Measurement Error in Survey Data." American Journal of Political Science 18: 468-85. Converse, Philip E. (1964). "The Nature of Belief Systems in Mass Politics." In D. Apter (ed.), Ideology and Discontent. New York: The Free Press. (1970). "Attitudes and Non-Attitudes: Continuation of a Dialogue." In E. Tufte (ed.), The Quantitative Analysis of Social Problems. Reading, MA: AddisonWesley. (1975). "Public Opinion and Voting Behavior." In F. I. Greenstein and N. W Polsby (eds.), Handbook of Political Science. Vol. 4. Reading, MA: Addison-Wesley. Converse, Philip E., and Gregory B. Markus (1979). "Plus ca change ... : The New CPS Election Study Panel." American Political Science Review 73: 32-49. Dean, Gillian, and Thomas W. Moran (1977). "Measuring Mass Political Attitudes: Change and Uncertainty." Political Methodology 4: 383-424. Erikson, Robert S. (1978). "Analyzing One Variable-Three Wave Panel Data: A Comparison of Two Models." Political Methodology 5: 151-61. (1979). "The SRC Panel Data and Mass Political Attitudes: Change and Uncertainty." British Journal of Political Science 9: 89-114. Franklin, Charles H., and John E. Jackson (1983). "The Dynamics of Party Identification." American Political Science Review 77. Hanushek, Eric A., and John E. Jackson (1977). Statistical Methods for Social Scientists. New York: Academic Press. Jackson, John E. (1975a). "Issues, Party Choices and Presidential Votes." American Journal of Political Science 19: 161-85. (1975b). "Issues and Party Alignment." In L. Maisel and P. M. Sacks (eds.), The Future of Political Parties. Beverly Hills, CA: Sage Publications. (1979). "Statistical Estimation of Possible Response Bias in Close-Ended Issue Questions." Political Methodology 6: 393-423. Joreskog, Karl (1969). "A General Approach to Confirmatory Maximum Likelihood Factor Analysis." Psychometrika 34: 183-202. (1973). "A General Method for Estimating a Linear Structural Equation System." In A. S. Goldberger and 0. D. Duncan (eds.), Structural Equation Models in the Social Sciences. New York: Seminar Press. Joreskog, Karl, and D. Sorbom (1976). LISREL III: Estimation of Linear Structural Equation Systems by Maximum Likelihood Methods. Chicago: National Educational Resources. Kinder, Donald R., and D. R. Kiewiet (1979). "Economic Grievances and Political Behavior." American Journal of Political Science 23: 495-527. Kinder, Donald R., and David 0. Sears (1982). "Political Behavior." In G. Lindzey and E. Aronson (eds.), Handbook of Social Psychology. 3d. edition. Reading, MA: Addison-Wesley. Markus, Gregory B., and Philip E. Converse (1979). "A Simultaneous Equation Model of Electoral Choice." American Political Science Review 73: 1055-70. Musgrave, Richard A., and Peggy B. Musgrave (1976). Public Finance in Theory and Practice. New York: McGraw-Hill. Nie, Norman, with K. Anderson (1974). "Mass Belief Systems Revisited: Political Change and Attitude Structure." Journal of Politics 36: 540-91.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

SYSTEMATIC BELIEFS OF THE MASS PUBLIC

865

Nie, Norman, Sidney Verba, and Jon Petrocik (1979). The Changing American Voter. Cambridge, MA: Harvard University Press. Page, Benjamin I., and Calvin C. Jones (1979). "Reciprocal Effects of Policy Preferences, Party Loyalties and the Vote." American Political Science Review 73: 1071-89. Pierce, John C., and Douglas D. Rose (1974). "Non-Attitudes and American Public Opinion." American Political Science Review 68: 626-49. Pomper, Gerald M. (1972). "From Confusion to Clarity: Issues and American Voters, 1956-1968." American Political Science Review 66: 415-28. Schulman, Mark A., and Gerald M. Pomper (1975). "Variability in Electoral Behavior: Longitudinal Perspectives from Causal Modeling." American Journal of Political Science 19: 1-18. Sears, David O., et al. (1979). "Whites' Opposition to 'Busing': Self-Interest or Symbolic Politics." American Political Science Review 73: 369-84. (1980). "Self-Interest vs. Symbolic Politics in Policy Attitudes and Presidential Voting." American Political Science Review 74: 670-84.

This content downloaded on Fri, 1 Mar 2013 00:03:50 AM All use subject to JSTOR Terms and Conditions

You might also like