for A plus writer only

Cognition, Vol. 8, No. 1, 1940, pp 125-153

POLITICAL AWARENESS, ELITE OPINION

LEADERSHIP, AND THE MASS

SURVEY RESPONSE

IOHN ZALLER

University of California, Los Angeles

Political awareness affects virtually every aspect of citizens' political attitudes and

voting behavior. Among its effects are greater attitude stability, greater ideological

consistency, and greater support for a nation's "mainstream" values. Yet there

exists no comprehensive explanation of why political awareness has the pervasive effects that it has. Nor is there agreement on how the concept of political awareness

should be conceptualized and measured. This article addresses both concerns.

First, it draws on ideas from voting, belief-system, and other studies to develop a general theory of the effects of awareness. This account centers on how citizens

use cues and other information from political elites to translate their general value

orientations into support for particular polices. Second, the article argues that,

on both theoretical and empirical grounds, political awareness is best measured

by simple tests of factual information about politics.

Political awareness deserves to rank alongside party identification and

ideology as one of the central constructs in the public opinion field.

Citizens who are more politically aware are more likely to develop stable attitudes on major political issues (Feldman, 1989), to align their attitudes on the conventional liberal-conservative continuum

(Chong, McClosky, & Zaller, 1985; Converse, 1964; Stimson, 1975), and to coordinate their issue preferences with their party attachments

and voting decisions (Belknap & Campbell, 1951-1952). According to

many studies, better-informed persons are more likely to espouse racial and political tolerance and to support the "mainstream" or "of

ficial" U.S. government line on foreign policy (Chong et al., 1985;

Gamson & Modigliani, 1966; Key, 1961). Finally, political awareness

is a critical determinant of mass attitude change (Converse, 1962;

MacKuen, 1984; Zaller, 1987, 1989).

Notwithstanding all of this, no one has yet explained why political awareness has such pervasive effects. Several theories have been crafted

to explain effects on a particular outcome or finding, such as attitude

125

126 ZALLER

constraint, but no one has provided a comprehensive account of how

one gets from political awareness to each of its numerous dependent effects. The first purpose of this article is to provide such an account.

The second purpose is to determine, from both theoretical and

empirical reasons, how political awareness should be conceptualized and measured. Variables purporting to measure "political awareness,"

"political expertise," "political sophistication," "cognitive sophisti cation," "political information," "political involvement," "media ex

posure," and "political interest" appear regularly in the public opinion literature and are used (along with education) more or less inter

changeably to explain the same general family of dependent variables.

Questions thus arise: Are these alternative measures different in any

important ways? If so, what are the differences? If not, what is the

basic concept of which they are all indicators, and how is this concept best measured? This article will propose answers to these questions.

A MODEL OF ATTITUDE FORMATION

This section develops a model of how citizens respond to political news and argumentation and convert their responses into attitude

statements on opinion surveys. The rest of the article w7ill use this

model to explain why political awareness has such wide-ranging effects.

It is necessary to begin with some definitions. First, political awareness

denotes the extent to which an individual pavs attention to current

political events and understands what he or she has encountered. In

a phrase, political awareness denotes intellectual (as against merely emotional) engagement with public affairs. A person wrho is highlv aware knows the major political figures and their policy positions, understands the norms that govern the political system, and keeps current on new developments.

Because only a minuscule fraction of the public possesses regular, firsthand experience of politics, political awareness must normally

develop from exposure to the political media. So in a sense, political awareness denotes the extent to which individuals have encountered

and comprehended media reports of political events, issues, and per sonages.

A second primitive term in the model is consideration, which is

any reason that might induce an individual to decide a political issue one way or the other. Considerations are a compound of cognition and affect that is, a belief concerning an object and an evaluation of the belief. "The Pentagon wastes a huge amount of money" is an

example of a consideration that might impel an individual to say, in

response to a survey question, that he or she favors reductions in

POLITICAL AWARENESS 127

defense spending. The cognitive element in this consideration is the belief that the Pentagon wastes money; the implied evaluation of the belief is negative. Although more could be said about considerations, particularly their possible role in guiding perception, the present spare concept suffices for the tasks of this article.

Finally, there are two types of political communications per suasive

messages and cueing messages. Persuasive messages are rea sons tor taking a position or point of view; if internalized by a person, they become, by the definition just given, considerations.

Cueing messages consist of claims about the ideological or partisan implications of a persuasive message. Sometimes a persuasive message will carry its own cue, as when a political party promulgates a platform urging a series of policies. In such cases, almost everyone who en counters the original message will become aware of its partisan im

plications. But in other cases, the partisan implications of a message may be obscure, perhaps deliberately so, as when a president urges a policy "for the good of the nation." In such cases, individuals may need a cueing message to perceive partisan or ideological implications in the persuasive message.

The axioms of the model may now be stated. Each is stated first

as a general theoretical posture and then operationalized or elaborated in more precise terms.

Al. Reception Axiom. The greater a person's level of attention to an

issue, the more likely he or she is to be exposed to and comprehend in a

word, to receive political messages concerning that issue.

The messages to which people are exposed may be persuasive

messages or cueing messages. A2. Resistance Axiom. People attempt to integrate incoming political

messages with their other beliefs and values.

Integration, in the context of this model, means resistance to

messages that are recognized as inconsistent with one's values (i.e.,

party attachment, religion, interests). Recognition depends on (1) prior

exposure to a relevant cueing message and (2) availability of the cueing

message in memory at the moment of determining whether to accept the persuasive message.

So if a message is consistent with one's values, it is accepted as

a consideration; if it is inconsistent, it may also be accepted, unless

the individual is able to recall information from a cueing message that

enables him to resist the persuasive message. A3. Recall Axiom. The more recently a consideration or other political

message has been called to mind or thought about, the less time it takes to

retrieve it from memory and bring it to the top of the head for use.

Conversely, the longer it has been since a consideration or message has been activated, the less likely that the consideration is accessible

128 ZALLER

at the top of the head; in the limit, a long-unused consideration may be inaccessible, which is to say, forgotten.

A4. Response Axiom. People normally respond to survey questions

by averaging across readily available considerations.

A4 indicates that individuals who have been asked a survey ques

tion do not normally canvass their minds for all considerations relevant

to the given issue; rather, they answer the question on the basis of

whatever considerations are available "at the top of the head" (Taylor & Fiske, 1978). If only a single consideration is available, individuals

answer on the basis of that consideration; if two or more considerations

come quickly to mind, people answer by averaging across available

considerations.

Altogether, then, the model consists of very general claims about

how people acquire ideas from the external environment (in the form

of persuasive arguments and cues) and transform those ideas into

survey responses. The most distinctive feature of the model is its

stipulation that people base attitude reports on what is, in effect, a

small, nonrandom sample of the considerations in their heads; in view

of this, I will, for convenience, refer to it as the sampling model of

opinion formation and change. Political awareness is central to this model. People who are, in

general, more politically aware will be generally attentive to politics and issues, thus bringing Axiom Al into play. A3 also implies awareness

effects because people who are high on political awareness will think

more often about issues, thereby increasing the availability of con

siderations. If, moreover, more-aware persons have learned a larger number of cueing messages and have these messages more available

for use in processing incoming messages, it follows from A2 that they will also be more resistant to messages that are inconsistent with their

values. Framing the model in terms of attentiveness to particular issues rather than awareness allows individuals who pay attention to

one or two issues of special personal interest to experience the effects of political awareness.

Subsequent sections of this article explore the deductive impli cations of these axioms. They will develop a series of deductions, review existing evidence pertaining to them, and present new evidence as necessary and possible to resolve outstanding issues.

STRUCTURE OF POLITICAL ATTITUDE STATEMENTS

The analysis of the implications of the model begins with an empirical observation for which there is abundant supporting evidence: Although

POLITICAL AWARENESS 129

there is a large amount of individual variability, most individuals devote little attention to politics (Kinder & Sears, 1985; Luskin, 1987). From this two things follow. First, if people pay little attention to

politics, they will be unlikely to possess the cueing messages (via Al) for selecting critically (via A2) among the persuasive messages they encounter. Thus, they will often end up accepting competing or op posing considerations (Deduction 1 [Dl]); in less technical terms, they will often find themselves ambivalent and unable to decide firmly between opposing positions.

Although few surveys are designed to detect the existence of

ambivalence, the available evidence supports Dl. Zaller and Feldman

(1989) found that, when interviewers invited respondents to discuss the issues in survey questions as a prelude to answering the questions, well over half of the sample in a national survey raised conflicting considerations that is, reasons both to favor and oppose the issue.

Hochschild (1981) provides qualitative evidence of the same phenom enon.

Dl claims that people will not, on average, be very critical in

deciding which of the messages encountered they will accept. Some

persons, however, may have been exposed to few or no persuasive

arguments on some issues; or, because they rarely think about the

arguments they have accepted, they may be unable (via A3) to call

any considerations to mind in the short time they give themselves

for answering survey questions. Such persons must presumably answer

questions with "no opinion." Because people who are low on political awareness would tend to think less about politics than other persons would, they should be less likely to offer opinion statements or, con

versely, more likely to offer no opinion responses (D2). Krosnick and

Milburn (this issue) review the considerable evidence supporting this

deduction.

If people are uncritical about the ideas they accept as considerations, and if their attitude statements depend on whatever ideas happen to

be at the top of the head at the moment of response, we should expect a fair amount of variability in people's responses to survey questions

(D3). This deduction has been strongly confirmed on numerous oc

casions (e.g., Feldman, 1989). The model also has strong implications for the structure of response

instability. First, given a stable environment, one can assume that the

balance of positive and negative considerations in people's minds will

remain stable as well. What varies, according to the model, is the

particular consideration at the top of the head at a given interview.

Hence, we should expect to find a fair amount of purely chance

variation (D4) around a stable central tendency (D5). There is wide

130 ZALLER

agreement among researchers who have investigated that this ex

pectation is correct (Achen, 1975, Erikson, 1979; Feldman, 1989; Judd,

Krosnick, & Milburn, 1981).

The model implies, in addition, that more politically aware persons will exhibit less chance variability in their survey responses (D6), for

two reasons. First, more-aware persons are more likely to possess (via Al) the cueing messages necessary to reject ideas uncongenial to their values and are more likely to have these cues available for

use (via A3); this means they will be more selective about the ideas

they internalize, which in turn means that when they select consid

erations for answering questions, they will be sampling from a more

homogeneous pool. This will increase the chances that they will sample the same type of consideration leading them to make the same

survey response each time they confront the given issue. Second, because most people pay little attention to politics, many will have

difficulty calling to mind more than one or two considerations. Yet

some highly aware persons may have several considerations (via A3) available at the top of the head; these highly aware persons end up

averaging across multiple considerations, which further reduces chance

variability.

Existing literature provides equivocal support for D6. Some re

searchers have reported that more politically aware persons exhibit

less random instability (Dean & Moran, 1977; Feldman, 1989) but

others have not (Achen, 1975; Erikson, 1979). This question will therefore

be investigated below.

MEASUREMENT AND CONCEPTUALIZATION

OF AWARENESS

There is good reason for finding, if possible, the best measure of

political awareness. Awareness often has interactive effects, so to test for them, one must either partition one's data on levels of awareness

or use a multiplicative term of which awareness is one component. In these situations, one can normally test only one operationalization at a time. This creates a need to have strong "out of sample" precon ceptions on how awareness is best measured. One aim of this article

is to establish such preconceptions, both theoreticallv and empirically. Political awareness is most appropriately conceptualized in light

of the role it plays in the sampling model. As has become apparent, awareness is important to the model because it is associated (via Al)

POLITICAL AWARENESS 131

with reception of political messages and (via A3) with the availability of these

messages in memory. Past research has used education, level of political participation,

level of political interest, media exposure to politics, and political information as measures of political awareness. On a priori grounds, each would seem to have strengths and weaknesses.

For example, people who develop the abstract learning skills that are supposedly imparted by education would seem likely, by that

account, to be more skilled at comprehending the political messages they encounter. Yet education is correlated with media exposure (as measured below) at the level of only .37, which suggests that many educated people have little interest in politics and may therefore fail to develop much political awareness. Meanwhile, there are probably some people who regularly scan the newspaper and watch the television news but who lack the cognitive skills to fully absorb the ideas they encounter. Similarly, it is easy to imagine that some individuals describe themselves as "interested" in politics because they are fascinated by the glitter of politics rather than because they understand much about it. Thus, neither education nor interest nor media exposure would

seem ideal measures of political awareness.

Political participation has similar weaknesses as a measure of

awareness. Besides having the weaknesses of self-report measures, there exist practical difficulties: a city garbage collector who must

contribute work or money to the party machine in order to keep his

job would be indistinguishable on most political participation measures

from an activist in an issue-oriented political club, even though their

differences in political awareness (as defined here) could be very great. It is, moreover, easily possible for a person to achieve very high levels

of political awareness without ever giving money to candidates, working for parties, or otherwise participating in politics.

This brings us to political information, a measure that, to a greater extent than any of the others, captures political learning that has

actually occurred that is, political ideas that have been encountered

and comprehended and remain available for use. This is exactly what,

according to the model, we should be trying to measure.

Information is preferable on other theoretical grounds as well.

Alone among the five rival measures of awareness, tests of political

information are relatively immune to social desirability response set.

Tests of information are probably relatively immune to response effects,

such as Bishop, Oldendick, & Tuchfarber's (1984) demonstration that

expressed levels of political interest are readily affected by the context

in which questions are asked. Finally, unlike political interest and

132 ZALLER

media use, information tests do not require respondents to estimate

subjective behaviors or inner states.

OPERATIONAL MEASURES OF AWARENESS

There are, then, both practical measurement and theoretical reasons

for expecting political information to be the best available indicator

of political awareness. To provide empirical tests of this expectation, I use data from the 1,320 respondents in the 1972-1974-1976 panel

study of the National Election Studies (NES). A special advantage of

this dataset is that, over its three waves, it contains multiple, redundant

indicators of the rival measures of awareness, thereby enabling the

construction of more reliable measures than can be achieved in most

surveys.

The items used to operationalize the five measures of awareness

in this dataset are briefly summarized below; fuller descriptions may be found in the Appendix.

Political Information. The 1972 component of the panel study carries

a sufficiently rich selection of information tests that there is no need

to use data from other years. Included are a series of direct information

tests (e.g., which party controls Congress, the term of office of a U.S.

senator), ability to evaluate a variety of somewhat obscure political

figures (e.g., Henry Jackson), ability to recall the names of Congressional candidates, and ability to locate accurately the policy positions of

prominent individuals and groups. Of these several tests, only the

last requires comment.

Since 1968, the NES has routinely asked respondents to place both themselves and certain politicians and groups on 7-point policy scales. It is easy to convert these measures to tests of political infor

mation. For most partisan issues, Democratic politicians, the Democratic

party, and "liberals" should be placed to the left of most Republican

politicians, the Republican party, and "conservatives." The analyst must, to be sure, exercise some discretion, but there are numerous

cases in which the correct comparative placements are obvious to any informed observer.

The 1972 survey has two forms, and separate scales were built

in each form, then standardized and combined; their alpha reliabilities are .86 and .87.

Political Participation. The participation scale includes virtually every item in the study tapping self-reports of political behavior. The items cover such matters as attending rallies, donating money, and belonging to political organizations. The only omitted activity items are the reg-

POLITICAL AWARENESS 133

istration and voting items, which were set aside as criterion variables.

Altogether, the scale has 22 items, all scored 0-1, and an alpha reliability of .81. The measure is highly skewed, with 20% of the sample scoring 0 and 5% scoring 10 or more of a possible 22 points.

Media Exjiosure. This index measures use of media newspapers, newsmagazines, TV, and radio that carry political information. Items were culled from all waves of the panel study. Altogether, the index has 16 items and an alpha reliability of .86.

Interest in Politics. This scale contains only five items but attains an alpha reliability of .811 There are two kinds of items, one asking whether the respondent generally follows politics (all waves) and the other asking about attention to the recent campaign (presidential years only). The scale has a marked skew, with about one third of the sample scoring 13 or higher on a 15-point scale and only one sixth scoring 5 or below.

Education. Scored as years in school, education has a retest reliability

of .95.

Note that all five measures of awareness have high and roughly comparable alpha reliabilities. This will make it difficult to attribute

large differences in performance to differences in reliabilities. High levels of alpha reliability can, however, be very misleading. If some

persons exaggerate their levels of interest, media exposure, or par

ticipation across all items, and others are just as consistently modest

in these matters, it would produce misleadingly high levels of alpha

reliability. Similarly, large cross-individual differences in standards of

judgment would artificially inflate measured levels of alpha reliability. To ensure that the five measures of awareness capture more than

correlated error, each was separately tested for its ability to predict voter turnout, a variable that was measured in 1976 from official

records, thereby eliminating the possibility of overstated reports by

respondents. A probit regression was performed in which this be

havioral measure of political involvement was the dependent variable

and the given awareness measure was an independent variable (in addition to controls for age, race, and region of residence). All estimated

awareness relationships were statistically significant. Coefficients from

these regressions were then used to calculate the turnout rates shown

in Table 1. As the table shows, all five awareness variables have large effects on turnout. The effect is strongest for participation, a result

that should be expected, because turnout is similar to other behaviors

measured in the participation scale. More interestingly, interest and

media exposure also outperform information as predictors of turnout.

Because the sampling model has no implications for voting turnout, the data in Table 1 indicate nothing about the comparative performance

134 ZALLER

TABLE 1

Political Awareness and Vote Turnout

PERCENTILE SCORE ON AWARENESS SCALE

5th 20th 83rd 95th 20 to 95 DIFF.

Information 48 60 82 86 26 pts

Education 50 61 81 86 25 pts

Participation a

58 85 93 36 pts

Media exposure 49 59 85 90 31 pts

Interest in politics 44 60 84 87 27 pts

Note: Cell entries are estimated percentage of respondents who voted in the 1976 election, as

validated by NES.

a209r of all respondents scored in the lowest participation category; hence there exists no 5th

percentile on this scale

of the five measures in the model. Table 1 has been included only as

a reliability check on the rival measures. The conclusion to draw from

it is that, despite the concern over correlated error, the measures of

interest, media exposure, and participation are measuring something

quite real. Thus if, as anticipated, information performs in later tests

as the best measure of political awareness, the results in Table 1 will

make it difficult to claim that information's success is due to the

weakness of the competing measures.

ATTITUDE CRYSTALLIZATION

As shown, the model implies that more politically aware persons will

be more likely to express opinions (D2) and more likelv to be stable

in their attitude reports (D6). In these particular senses, more-aware

persons should have more crystallized attitudes.

A test of this expectation has been conducted as follows: For each

of seven attitude scales that were carried on the 1974-1976 portion of the panel, the original policy scale was recoded to a 3-point scale

(two polar positions and a middle position) plus "no opinion." Re

spondents who changed positions on a 3-point scale between 1974

and 1976, or gave no opinion in either wave, were coded as 0 (i.e.,

having uncrystallized attitudes). Respondents giving the same sub

stantive opinion on both waves were coded as 1. The seven 0-1

crystallization variables created in this fashion were aggregated into

a summary index, giving an overall measure of "stable opinionation" across seven issues. This index was regressed on the five measures

POLITICAL AWARENESS 135

of awareness to produce a multivariate test of the effects of the rival

measures on a single criterion variable. The standardized coefficients from this test indicate information

is the single most important determinant of crystallization ((3 = .225,

p < .01), though interest (p = .116, p < .01) and education (f3 = .118,

p < .01) also have effects. If one squares the coefficients to obtain a

measure of variance explained, information is approximately 3.6 times more important than its nearest competitor.

The model also implies that people's attitude statements should

vary randomly around a stable central tendency and that the amount of this chance variation will be lower for more aware persons (D5, 6). The Wiley-Wiley (1970) measurement model for three-wave panel data permits testing of these expectations. That is, it permits separation of "error variance" from changes in the central tendency of people's responses. One can, in addition, estimate the effect of political awareness on error variance by estimating the Wiley-Wiley model separately within both high- and low-awareness groups. The advantage of this

procedure is that, in contrast to the crystallization test, it produces

separate measures for error variance and for stability of the underlying attitude. A disadvantage is that it does not permit direct multivariate

testing; another is that it accommodates only persons who offer opinions on all three waves of the panel, a form of selective inclusion that is

likely to lead to underestimates of the overall effect of awareness.

The estimated error variances from the Wiley-Wiley tests are

shown in Table 2 for the six attitude items that were carried full-

sample on the three-wave NES survey. The estimates are shown sep

arately for respondents in the lower third and upper third of each

awareness scale. As can be seen, persons scoring in the bottom third

of the several awareness scales had higher error variances on all six

issues that did persons scoring in the top third of these scales.

As expected, information outperformed the other measures of

awareness in the sense that it produced larger high-low differences

in response variability than did any other measure. Education was,

however, a very close second, producing an average reduction in error

variance of 46% compared to 49% for information. These effects in

reducing error variance easily achieve statistical significance, as does

the effect of participation. Media exposure and interest do not have

a statistically significant tendency to reduce error variance.

Comparative performance of the five measures was tested via

the following model:

Error Var. =

A0 + A^Info.) + A2(Dum x ALT) +

B3(Dum) + B4(Job) . . . + B8(Bus)

00 (N 00 (N

O O 1-1

uj O

UJ

y >

o

C

c

o

J3 ns

C

& "33 0)

XI

> m

QJ (0 o

T5 0)

XI J

60

C

in 3 X)

01

>

S3

C O

it It1 qj

> >

O ofj 2 ^ M o o

-

o Jl ooJ X x Jl ^ jI

U c2 3

136

POLITICAL AWARENESS 137

where ALT is the alternative measure of awareness and where Dum

takes the value of 0 for information and 1 for the alternative measure.

The dependent variable consists of 36 error variances (6 issues x 3

awareness levels x 2 awareness measures). The expectation is that the sign on B3 will be negative, indicating that the slope for the

alternative measure is less than that for information. This interaction

was obtained and achieved statistical significance for separate tests of information against participation, interest, and media exposure (t s= 3.0 in all three models, df

=

30, p < .01). The effect ran in the

expected direction for education but did not approach statistical sig nificance.

From D5, the model implies that, despite the large effects of awareness on chance variability, awareness should have no association with systematic change in the central tendencies of attitude statements.2 This expectation is also supported. The path coefficients describing attitude stability average about 0.9 (where 1.0 would indicate perfect stability) and show no significant tendency to vary linearly by political awareness. In the case of information, stability coefficients average .91 in the low-awareness group and .89 in the high-awareness group.

Altogether, then, these results uphold the sampling model on a

point on which prior evidence has been unclear: awareness is associated

with a statistically significant reduction in random variance on typical attitude items. The results also favor information over alternative

measures of awareness, though the margin of information over ed

ucation is somewhat unclear: it is large in the multivariate crystallization test but modest in serial tests in Table 2.

The discussion of this section has focused on phenomena that

are sometimes considered methodological artifacts. The next section

broadens the range of the model by showing that it can also explain the substance of people's attitudes.

SUBSTANCE OF POLITICAL ATTITUDES: THE

MAINSTREAM AND POLARIZATION EFFECTS

As indicated below, there is evidence that political elites in the United

States sometimes achieve a consensus in support of particular "main

stream" policies. What, one may ask, would be the expected effect

on public opinion of such an elite consensus? Or, to ask the same

question in the language of the model: What would be the effect on

public opinion if virtually all of the persuasive messages carried in

political media on a particular policy were favorable to that policy and

138 ZALLER

if there were no cueing messages to alert people that the policy was inconsistent with their values?

Axiom Al suggests, first of all, that higher levels of awareness

will be associated with higher levels of exposure to persuasive messages

favoring the mainstream issue. Also, if all of the cueing messages on

the issue were favorable, no one would have any basis via A2 for

resisting it. From this we can deduce that the greater a person's level

of political awareness, the greater the number of pro-mainstream

messages the person would internalize in the form of considerations

and hence, all else equal, the greater the person's level of expressed

support for the mainstream policy on surveys (D7). The notion that exposure to elite discourse automatically increases

support for mainstream policies obviously cannot apply to every type of policy issue. Many policies are heatedly contested by elites, so that

no mainstream exists.

The model, however, leads us to expect quite different patterns of mass attitudes for cases in which policies are contested. To see

why, let us assume a situation in which elites are roughlv evenly divided on a partisan issue, with one partisan camp sponsoring per suasive and cueing messages favoring the liberal position and the

other sponsoring messages in support of the conservative position.

Among liberals in the general public, increases in awareness will

lead to increased exposure to persuasive messages favoring both the

liberal and the conservative position (from Al) and also increased

exposure to cueing messages concerning the issue. Reception of the

cueing messages will enable politically aware liberals (via A2 and A3) to reject conservative arguments on the issue but will not impede their internalization of liberal messages. Less politicallv aware liberals, on the other hand, will be exposed to few persuasive messages and,

owing to their low exposure to and the availabilitv of cueing messages, will be unselective about the persuasive messages thev internalize.

As a result of all this, the most aware liberals will fill up their

minds with a large number of considerations that are, on balance, favorable to the liberal side of the issue. Less-aware liberals, for their

part, will form a small number of considerations, and these consid

erations will not consistently favor the liberal side of the issue. Thus, with increases in political awareness, the ratio of ideologically consistent

considerations to ideologically inconsistent ones should increase (D8). Zaller and Feldman (1989), using the open-ended data of the ideas

underlying people's attitude statements, have reported support for this deduction for the three issues they examine (federal job guarantees, level of government services, and aid to blacks).

POLITICAL AWARENESS 139

One may expect that an increasing ratio of ideologically consistent

to ideologically inconsistent considerations should translate

into dif

ferences in people's attitude statements: more-aware liberals will be

more likely to call to mind considerations favorable to the liberal

position and hence will be more likely to support it. Less-aware liberals

will be less likely to call to mind considerations of any kind, which

will lead to higher no-opinion rates, and less likely to endorse the

liberal position when they do offer an opinion. The logic of this argument

applies equally to conservatives; that is, increases in awareness make

mass conservatives increasingly likely to make conservative attitude

statements when asked about the issue.

Thus, in the case of a divided partisan elite, the effect of political

awareness is to promote the polarization of attitude reports, as more-

aware liberals and conservatives gravitate more reliably to the con

ventional liberal and conservative poles (D9). The data in Figure 1 illustrate both the mainstream and polarization

effects of political awareness. When, in 1964, American elites nearly all supported the Vietnam War, increases in awareness led nonelite

FIGURE 1

An illustration of the mainstream and polarization models.

80

70 -

Percent

Supporting American

Involvement go -

In Vietnam

50

1964

Conservatives

Liberals

1970

Level ol Political Awareness

86 179 153

Conservatives

140 ZALLER

liberals and conservatives to become more supportive of the mainstream

war policy. Yet when, in 1970, American elites had become deeply divided about the war, increases in awareness are associated with

greater polarization of the attitudes of mass liberals and conservatives.4

(In Figure 1, awareness has been measured by simple tests of factual

information. Liberals and conservatives are persons who rated liberals

higher than conservatives, or vice versa, on 100-point feeling ther

mometers. Persons supporting American involvement are those who

said either that the United States should "keep troops in Vietnam but

try to end the fighting," which was the administration position [Gelb with Betts, 1979], or that the United States should take a stronger stand by invading North Vietnam.)

The key idea here is that political awareness has different effects

on mass attitudes, depending on whether elites are united on a main

stream policy or divided over partisan policies. To test this idea more

systematically, I selected items from the 1972-1974-1976 survey that

seemed, on their face, to exemplify mainstream policies and partisan

policies of the early 1970s.

The five mainstream issues were whether the United States should

actively involve itself in the world affairs or keep to itself (v291), whether China should be a member of the United Nations (v62), whether farmers and business should be allowed to trade with Com

munist nations (v55), whether the government should support the

right of blacks to go to any hotel or restaurant they can afford (vllO), and whether women should have an equal role with men (v232). The

partisan issues were whether children should be bused to schools

outside their neighborhoods to achieve racial integration (v202), whether

the government in Washington should see to it that every person has a job and a good standard of living (vl72), and whether the United States did the right thing insetting into the fighting in Vietnam (v50).

Selection of the items was based on my judgment of the positions of liberal and conservative elites, political party elites, and the mass media at the time of the NES study. To partially check these judgments, I asked a research assistant to read the platforms of the Democratic and Republican parties in 1972 and to rate each party on each issue.

Party platforms were used not because they were assumed to be the most important source of elite cues on these issues but because they represented a convenient means of measuring the extent of elite agree ment or disagreement on issues. The research assistant was unaware of my expectations and did the ratings on the basis of instructions that were conveyed in writing. I expected that both parties would

explicitly endorse the policies I had identified as mainstream policies and would take sharply opposing positions on policies I had judged

POLITICAL AWARENESS 141

as partisan policies. These expectations were largely confirmed. The one

exception involved an item about whether the federal government should

guarantee blacks the right to equal treatment in hotels and restaurants. A provision on equal accommodations was a key part of 1964 Civil Rights Act, which, Senator Barry Goldwater notwithstanding, passed the Congress with majority support from both the Democratic and Republican parties. The Democratic platform, as expected, explicitly endorsed this policy, but the Republican party, although expressing general support for equal rights, made no direct reference to it. I continue, in light of the bipartisan history of the Civil Rights Act, to consider equal accommodations in hotels and restaurants a mainstream

government policy. The model used to estimate the effect of awareness on each of

these items is

Prob.(Lib. Response) = Prob. (Opinionation) x

Prob.(Lib. | Opinionation)

That is, the probability of a liberal response is the probability of offering any opinion at all, times the probability of making a liberal response, given that an opinion has been offered. The two parts of the model have been estimated separately.

The probability of a liberal opinion, given that an opinion statement has been made, has been modeled as a logit function of awareness,

ideological self-designation (7-point scale), party identification (5-point scale), and standard demographic variables (race, age, income, and residence in a Southern state). In addition to these variables, the initial

specification of each equation contained an interaction term for

Awareness x Ideology and Awareness x Party. This equation was

estimated separately by nonlinear least squares for each of the five

mainstream issues and three partisan issues. To maximize comparability of results across different item formats, each item was recoded to a

3-point scale running from 0 to 0.5 to 1.0; in doing the recoding, I

attempted to equalize variance across the several dependent variables.

The expectation from the model is that for partisan policies, the

two Awareness x Values interaction terms will be strong but that for

mainstream policies these interactions will be anemic or zero. The

second expectation is that awareness will have an important positive effect on support for mainstream policies.

When information is used as the measure of awareness, the first

expectation is strongly confirmed. The Ideology x Awareness term

is strongly and significantly positive for the busing, job guarantees, and Vietnam items but nonsignificant in the other equations. The

142 ZALLER

Party x Awareness interaction is not significant for any of the partisan issues, but this would not seem to damage the model. It indicates

only that ideology is the better measure of the political values that

affect policy preferences. The substantive import of these Awareness x Ideology interactions

is best conveyed by means of graphical analvsis, as shown for the

three partisan issues in the top half of Figure 2. (The coefficient estimates

cannot be published for reasons of space but are available from the

author.)

The second expectation, concerning the effect of awareness on

support for mainstream policies, was difficult to confirm because of

the presence of multicollinearity. For all five mainstream policies, the

estimated coefficient for awareness was correlated with the estimated

coefficient for Awareness x Ideology at the level of .95 or higher. Because of this, neither awareness, ideology, nor their interaction

achieved statistical significance in several equations. Howrever, mul

ticollinearity can greatly reduce the precision of estimates even when

the true effect of one of the collinear variables is zero (Hanushek &

Jackson, 1977; Rao & Miller, 1971). To test whether the Awareness x

Value interaction terms had a significant effect on the mainstream

policies, I reestimated each equation without the interaction terms

FIGURE 2

The effect of awareness on political attitudes.

Gov't Guaranteed Jobs Wrong on Vietnam

Liberals

Busing for Integration

Probability 0 50 -

ol Liberal

Response

Political Awarness

Admit China to UN

Liberal

Role ol Women

Liberals

Equal Accommodations

Liberals

Probability

ol Liberal

Response

Political Awareness

Note Slopes based on regression coetlicients. as described in text

POLITICAL AWARENESS 143

and did an F-test to see if the omissions had a significant effect on the residual sum of squares. For anti-isolationism, women's rights, trade with Communist nations, and admission of China to the UN, the F-test indicated that the interaction terms had no importance. Moreover, with the interactions omitted, awareness took on a statistically significant positive coefficient in all four cases, as expected. In the fifth case, equal accommodations, the Party x Awareness coefficient remained significant, but with the nonsignificant Ideology x Awareness term omitted, awareness had the expected positive effect on support for this mainstream race policy.5

The bottom half of Figure 2 graphically illustrates the results for the mainstream issues. Two mainstream issues anti-isolationism and

trade with Communist nations are not shown in Figure 2 but very closely resemble the results for the item on admission of China to the United Nations, shown in the figure.

Of the mainstream issues, only the women's role item poses any problem for the model, which is that, despite the mainstream norm,

ideology clearly has an important effect on people's attitudes. Note, however, that awareness does significantly increase support for equal rights, as it is expected to do, and that, in contrast with the partisan issues, it functions to dampen rather than to magnify differences between liberals and conservatives, thereby increasing consensus among the most-aware segments of society.

Altogether, then, the mainstream results, along with those for

the three partisan issues, support the two basic deductions of the

model (D7,9) concerning the effects of a divided elite and of a united

elite on mass opinion.6 Information was used as the measure of awareness in the tests

in Figure 2 because preliminary investigation showed that it specified the effects of awareness better than any of the alternative measures.

That is, the direct effects of awareness on support for the mainstream

issues, and the interactive effects of awareness on the partisan issues,

were larger when an information scale was used than when other

measures were used.

To construct a test of the statistical significance of these performance differences, I reasoned as follows: If information has the same true

effect as a given alternative measure, the residual sums of squares from the eight equations should be roughly equally large, whether

one is using information or an alternative measure of awareness. This

will be the null hypothesis. In light of the theoretical reasons adduced

earlier, the alternative hypothesis will be that the residual sums of

squares are smaller across the eight issue equations when information

is used as the measure of awareness. Assuming independence across

144 ZALLER

the eight items, a simple t test can be used to decide between the

null and alternative hypotheses. For serial tests of information against

participation, interest, and media exposure, it was possible to reject the null hypothesis of no difference at the .01 level of confidence. For

the test of information against education, the null hypothesis could

be rejected at the level of .05. Thus here, as in the case of response

stability, information appears to be somewhat more effective than

education and clearly more effective than participation, interest, and

media exposure.

The dynamic that gives rise to the polarization of mass liberals

and conservatives in cases of elite division can also explain the ap

pearance of attitude constraint that is, the tendency for persons who

are liberal (or conservative) on one issue to be similarly liberal (or

conservative) on a whole range of other issues. In fact, the logic of

the polarization effect is identical to Converse's (1964) account of

attitude constraint.

According to Converse (1964), ideologies originate among a "min

uscule" number of "creative elites" and subsequently diffuse through the public. Converse argues, however, that the diffusion of elite-

created belief systems is highly imperfect. Only the politically aware

pay enough attention to elite discourse to find out the ideological

implications of different policies in Converse's terms, to learn "what

goes with what." As a result, attitude constraint of the conventional

liberal-conservative type develops mainly among the more politically aware strata.

Compare this argument to the representation of the polarization model in Figures 1 and 2: Highly aware liberals and conservatives (or Democrats and Republicans) look to appropriate partisan elites to

acquire "cueing messages" concerning "what goes with what." Having

acquired this information, they are able to become consistently liberal

or consistently conservative across a range of issues. The less-aware, as shown in these figures, are less likely to acquire the attitude that

is conventionally appropriate to their partisan orientation and hence

less likely to develop "attitude constraint" across issues. The typical finding of belief-systems studies that average interitem correlations

among issues are higher among more aware citizens is just a gen eralization of this pattern to a cluster of ideologically charged issues.

Thus, the existence of consistency across issues, and the well-docu mented tendency for consistency to become stronger with increases in awareness, can both be explained by the sampling model (D10,ll).

Two additional inferences can be drawn from the polarization effect. First, suppose that more-aware persons are, by some means,

given extra time to think about their survey responses time, that

POLITICAL AWARENESS 145

is, to call to the "top of the head" a larger than usual number of

considerations and to form their survey response by averaging across

this larger set of ideas. Given that, as the polarization argument con

tends, they are sampling from a fairly ideologically homogeneous pool, the effect ought to be to make their attitude reports more reliable

indicators of their underlying ideology because multi-element samples have less random variance than one-element samples. The effect of extra time on the attitudes of less-aware persons, however,

should

be different. Because less-aware persons possess, according to the

model, an ideologically heterogeneous pool of considerations, averaging should not make their responses more ideologically reliable.

Extra time to form survey responses, in sum, should make the

attitude statements of politically aware persons more ideologically consistent but should have less effect on attitude consistency among the less-aware (D12).

This deduction from the model has been confirmed in two ex

periments. One is the study by Judd and Downing in this volume.

Greater practice in thinking about the relations among different issues

was used to experimentally induce greater attitude consistency, but

the effect registered only among the more politically aware subjects. The other confirmation is in Zaller and Feldman (1989). In a national

survey, they experimentally asked half of the sample to say what

came to mind when they thought about a standard issue question and then to answer the question; controls answered the question in

the usual way. Among highly aware persons, the effect of this "stop and think" treatment was to increase the correlation between the

attitude statement and a measure of the person's ideology; among less-aware persons, the treatment did not increase this correlation.

The notion that individuals form attitude statements on the basis

of ideas that are at the tops of their heads can explain one other well-

established empirical regularity. It has often been shown that the

order in which questions are asked can affect responses (Schuman &

Presser, 1981). This may be explained by A3, which holds that the

more recently a consideration has been activated, the more available

it is for use in answering questions. Different question orders bring different considerations to the tops of people's heads, thereby changing the responses given (D13).

ALTERNATIVE EXPLANATIONS

There are, to be sure, alternative explanations for the empirical reg ularities noted in this article, but these explanations tend to focus

146 ZALLER

narrowly on particular substantive problems. For example, Stouffer

(1954) argues that better-educated persons are more supportive of

civil liberties because they are more likely to give a "sober second

thought" to freedom issues; Allport (1954) cites the greater personal

security of educated persons as an explanation for their greater racial

liberalism; Bobo and Licari (1989) contend that education promotes

political tolerance because it is associated with "more sophisticated modes of reasoning"; and Hamilton (1968) and Hahn (1970) suggest that higher levels of support for the Vietnam War among educated

persons may be due to "upper-middle class authoritarianism" (Ham

ilton, p. 446). Each of these arguments is perhaps plausible on its

face, but none generalizes easily to other issues. The more parsimonious

explanation of the sampling model would be that better-educated

persons had in each case been more heavily exposed to mainstream

elite values.

In a careful piece of work, Jackman (1978) notes that education

is associated with liberalism on some race items (e.g., "strict segregation" of the races) but not on others (e.g., busing). She concludes from this

that, contrary to much past research, education promotes onlv "su

perficial" learning of democratic values rather than genuine commitment

to them. The alternative argument of this article, of course, is that

the first type of item taps a mainstream policy and the second a

partisan policy. There also exist alternative explanations for attitude constraint.

Most stress the internal sources of attitude constraint that is, as

sociations that develop within the mind as a result of thought about

issues. These explanations also stress that more politically aware persons are more ideologically consistent than are less-aware persons because

they think more about politics (Fiske & Kinder, 1981; Judd & Krosnick,

1989; see also Luskin, 1987).

Though not denying the importance of intrapsvchic connections

and the individual's own thought in developing them, the sampling model manages to explain attitude consistency' without referring to

such mental organization. It instead stresses external sources of con

straint, namely, the pattern of elite division or nondivision on the

given issue. In maintaining this focus on external rather than internal

sources of constraint, the sampling model is able to explain the quite different effects that awareness has in different contexts, as in Figures 1 and 2.

In sum, scholars have proposed numerous alternative explanations for the empirical regularities explained by the sampling model, but

each applies to only one or perhaps two phenomena rather than to

the range of phenomena covered by the sampling model.

POLITICAL AWARENESS 147

SUMMARY AND DISCUSSION

This article has tried to do two things: first, to show that a simple model that affords a central place to political awareness can explain a wide

range of empirical regularities, and second, to establish that

political awareness is best measured by tests of political information.

MEASUREMENT AND CONCEPTUALIZATION

OF AWARENESS

One is politically aware to the degree that one chronically exposes oneself to and comprehends media reports of political events, issues, and personages. It has been argued that political awareness, understood in this way, is best measured by tests of political information. Infor mation is preferable on theoretical grounds because, alone among the rival measures of awareness, it captures political ideas that have actually gotten into people's heads while being, at the same time, relatively immune to social desirability response set and individual differences in standards of self-description. Information is preferable on empirical grounds because it specifies, better than alternative measures can,

relationships that political awareness ought to specify. The performance

advantage for information is not always large, but it shows up across

three quite different types of tests and achieves statistical significance in two of the cases.

THE SAMPLING MODEL

The strengths of the sampling model are breadth and parsimony. The

model consists of four axioms that are at least modestly plausible, and it is able to explain 13 varied empirical regularities, many of which

are well established in the research literature. These include attitude

constraint, response stability, support for mainstream policies such

as internationalism and racial equality, and question order effects,

among others. Additional deductions from the four basic axioms have

been developed elsewhere (Zaller, 1990; Zaller & Feldman, 1989),

including priming effects, generational differences in persuasibility, and certain nonobvious patterns of attitude change.

In my opinion, the greatest weakness of the political psychology

field is that it has proliferated theoretical mechanisms that are crafted

to particular empirical problems while neglecting to develop mech

anisms that span multiple domains. Researchers focus on the single

148 ZALLER

problem that interests them while ignoring links between seemingly similar phenomena (e.g., racial and political tolerance, or response

instability and question order effects). The greatest value of the sampling model, I believe, is that it takes concrete steps toward redressing this

dominant practice. The model, however, has the vice of its virtues: Breadth and

parsimony have been achieved at the expense of oversimplification. It is possible to argue, with justice, that each of the axioms fails to

capture fully what actually occurs. Let me review some of the most

important difficulties.

The assertion in Al that greater attention to an issue is associated

with greater reception of messages concerning that issue would seem

initially unexceptionable. Yet what is it to which people are exposed? Almost certainly, different people can be exposed to the same message and yet receive quite different messages, or perhaps no intelligible

message, depending on their prior knowledge. For example, suppose the Air Force stages an elaborate spectacle for the maiden flight of

the B2, its new flying-wing bomber, to show that the plane really works; some members of the public may then receive this message, but others may "see" in the staged spectacle a "batmobile" that is

"obviously useless." Such presumed differences in reception (given

equal attentiveness to the same message) will depend on people's

previously existing schema, which may differ both in content and

degree of development across individuals (Fiske & Kinder, 1981). Yet

the sampling model, as presently constituted, makes no allowance

for schema effects of this type. Consider next the limitations of the resistance axiom, A2, which

claims that people can resist persuasion only to the extent that they have acquired an appropriate cueing message. It is not clear that

resistance to a message is separate from reception of the message. For example, someone who "sees" in the maiden flight of the B2 a comical Batmobile and who sees this, incidentally, because he or

she is an avowed pacifist does not need to engage in a separate evaluation to see whether the message should be internalized as a

pro-B2 consideration. Perception and evaluation may constitute a single, schema-driven step.

The response axiom, which claims that individuals answer survey

questions by averaging across considerations recalled from memory, is at first glance the most contestable of the axioms. It is openly at odds with a considerable research literature, which argues that, instead

of making "memory-based" judgments from available ideas, as in A4,

people's judgments are normally made at the point of initial processing of incoming information (Hastie & Park, 1986; Lichtenstein & Srull,

POLITICAL AWARENESS 149

1987; Lodge, McGraw, & Stroh, 1989; McGraw & Pinney, this issue). That is, people update their evaluations "on line" as they encounter new information, storing the updated evaluation in long-term memory but discarding most of the information that was used for the update. When asked to express an attitude, people simply retrieve their updated evaluation and report it. As Hastie and Park maintain, on-line attitude

formation is the norm in social cognition.

Notwithstanding all of this, the response axiom used here is defensible as a first approximation of how political attitude statements are made. First, researchers do not contend that on-line judgments occur automatically; they occur only when a person is personally involved in a subject and interested in forming attitudes about it. The

question thus arises: how important is it to most citizens to maintain

updated attitudes on the extensive range of subjects on which pollsters routinely inquire? To judge by what leading political researchers say about the centrality of politics in the public mind, it would seem not

very likely (Kinder & Sears, 1985). The attitudes of most people on most political issues almost certainly fall into a category of personal

unimportance a category to which the on-line model is not especially

applicable. Second, the impetus to the current emphasis on on-line processing

was the repeated disappointment in finding negligible or even negative correlations between what people could recall from memory at the

moment of judgment and the judgments actually expressed. This and

other evidence caused researchers to conclude that people were for

getting most of the information on which judgments were made, so

that what was later remembered bore no particular relation to the

attitude. But, contrary to this general pattern, correlations between

attitude statements and recalled ideas in the domain ofpolitical attitudes

are neither negative nor negligible. According to the available evidence,

they range between about +.40 and about +.80, depending on how

they are assessed (Zaller & Feldman, 1989). But although the memory-based judgment process embodied in

A4 is defensible, it is defensible only as a first approximation of what

must actually occur. It seems likely that people engage in what might be called bounded on-line processing of political information; that is,

they make on-line evaluations of limited subjects but do not integrate each updated evaluation into a fresh global evaluation. So if, for

example, people encounter information about third-generation welfare

families, they make an on-line update of their judgment of the value

of social spending for the alleviation of poverty. If, sometime later,

they encounter information about the needs of homeless persons,

they update their judgment on the unmet welfare needs of the nation.

150 ZALLER

But what they do not do, unless called to do so by an unusual event, is to update their global attitudes on welfare spending. So, when

asked to make a global evaluation of the welfare system in a survey,

they will have no up-to-date global evaluation to retrieve but will

instead have to make an on-the-spot judgment from the (updated) considerations that come most readily to mind. By this account, A4

is not so much wrong as underspecified: memory-based evaluation

occurs, but so, at another level, does on-line processing. In view of these and no doubt other limitations, the sampling

model must be considered provisional. It performs useful work in

forging theoretical links between diverse research problems, but it is

far from giving a complete account of how political attitudes form

and change. Future work should attempt to retain the range and

simplicity of the present model while amending it to be in closer

agreement with our best understanding of reality.

A draft of this article was presented at the Third Annual Meetings of the Political

Methodology Society. I gratefully acknowledge helpful comments on this research by

Joel Aberbach, Larry Bartels, Stanely Feldman, Barbara Geddes, John Jackson, Robert

Luskin, Dwaine Marvick, John Petrocik, Jon Krosnick, and Lee Sigelman. I would also

like to thank Hiroaki Minato and Matt Lyons for excellent research assistance. I alone

am responsible for any errors the article may contain. Data for the article were originally collected by the National Election Studies and were made available through the Inter-

University Consortium for Political and Social Research at the University of Michigan and the Institute for Social Research at UCLA. Correspondence should be addressed

to John Zaller, Department of Political Science, UCLA, Los Angeles, CA 90024.

NOTES

1. Variables measuring concern with election outcomes scale poorly with the interest

items and are too few to form their own scale. Hence, I do not analvze them.

2. The model has very strong implications about the effect of awareness on attitude

change for cases in which the political environment is unstable. This effect, however,

is strongly nonlinear and so cannot be captured bv the stabilitv coefficients in the

Wiley-Wiley (1970) model. Space precludes the examination of these effects in the

present article.

3. For the interest and participation measures, there was an unexpectedlv strong tendency'

for the middle awareness groups to have the highest stability coefficients. The tendency did not achieve statistical significance in either case with a two-tailed test, as is appropriate for unexpected results, but was close to significance. I can think of no simple theory to account for this effect.

4. To validate these claims concerning elite consensus and division, I asked my research

assistant to classify cover stories on Vietnam in Newsweek and Time. In 1964, prowar

POLITICAL AWARENESS 151

cover stones outnumber antiwar ones by a margin of approximately 10 to 1; in 1970,

the ratio was about 1 to 1.

5. It is not permissible to do a parallel test omitting the direct awareness term because

awareness must be in the equation if Awareness x Ideology is included.

6. Although theoretical vocabularies differ from that used here, Gamson and Modigliani (1966) and Chong et al. (1985) also provide empirical support for D7, 9.

REFERENCES

Achen, C. H. (1975). Mass political attitudes and the survey response. American Political

Science Review, 69, 1218-1231.

Allport, G. (1954). The nature of prejudice. New York: Doubleday. Belknap, G., & Campbell, A. (1951-1952). Party identification and attitudes toward

foreign policy. Public Opinion Quarterly, 15, 601-623.

Bishop, G. F, Oldendick, R., &Tuchfarber, A. (1984). What must my interest in politics be if I just told you "I don't know?" Public Opinion Quarterly, 48, 210-219.

Bobo, L., & Lican, F. C. (1989). Education and tolerance: Testing the effects of cognitive

sophistication and target group affect. Public Opinion Quarterly, 53, 285-318.

Chong, D., McClosky, H., & Zaller, J. (1985). Social learning and the acquisition of

political norms. In H. McClosky & J. Zaller (Eds), The American ethos (chap. 8).

Cambridge, MA: Harvard.

Converse, P. E. (1962). Information flow and the stability of partisan attitudes. Public

Opinion Quarterly, 26, 578-599.

Converse, P. E. (1964). The nature of belief systems in mass publics. In David Apter (Ed.), Ideology and discontent (pp. 206-261). New York: Free Press.

Dean, G., & Moran, T. (1977). Measuring mass political attitudes: Change and uncertainty. Political Methodology, 4, 383-424.

Erikson, R. (1979). The SRC Panel data and mass political attitudes. British Journal of Political Science, 9, 89-114.

Feldman, S. (1989). Reliability and stability of policy positions: Evidence from a five-

wave panel. Political Analysis, 1.

Fiske, S. T., & Kinder, D. R. (1981). Involvement, expertise, and schema use: Evidence

from political cognition. In N. Cantor & J. Kihlstrom (Eds.), Personality, Social

cognition, and social interaction. Hillsdale, NJ: Erlbaum.

Gamson, W., & Modigliani, A. (1966). Knowledge and foreign policy opinions: Some

models for consideration. Public Opinion Quarterly, 30, 187-199.

Gelb, L., with Betts, R. (1979). The irony of Vietnam: The system worked. Washington, DC: Brookings.

Hahn, H. (1970). Correlates of public support for war: Local referenda on the Vietnam

issue. American Political Science Review, 64, 1186-1198.

Hamilton, R. F. (1968). A research note on public support for "tough-minded" military

policies. American Sociological Review, 33, 439-449.

Hanushek, E., & Jackson, J. (1977). Statistical methods for social scientists. New York:

Wiley. Hastie, R., & Park, B. (1986). The relationship between memory and judgement depends

whether the task is memory-based or on-line. Psychological Review, 93, 258-268.

Hochschild, J. (1981). What's fair? Princeton, NJ: Princeton University Press.

Jackman, M. (1978). General and applied tolerance: Does education increase commitment

152 ZALLER

to racial integration? American journal of Political Science, 22, 302-324.

Judd, C. M., & Krosnick, J. A. (1989). The structural bases of consistency among political attitudes: The effects of political expertise and attitude importance. In A. R.

Pratkanis, S. J. Breckler, & A. G. Greenwald (Eds.), Attitude structure and function.

Hillsdale, NJ: Erlbaum.

Judd, C, Krosnick, J. A., & Milburn, M. A. (1981). Political involvement and attitude

structure in the general public. American Sociological Review, 46, 660-669.

Key, V. O. (1961). Public opinion and American democracy. New York: Knopf. Kinder, D. R., & Sears, D. O. (1985). Public opinion and political action. In G. Linzey

& E. Aronson (Eds), Handbook of social psychology . Reading, MA: Addison-Wesley.

Lichtenstein, M., & Srull, T. K. (1987). Processing objectives as a determinant of the

relationship between memory and recall. Journal of Experimental Psychology, 23,

93-118.

Lodge, M., McGraw, K. M., & Stroh, P. (1989). An impression-driven model of candidate

formation. American Political Science Rezneiv, 83, 399-420.

Luskin, R. (1987). Measuring political sophistication. American Journal of Political Science,

31, 856-899.

MacKuen, M. (1984). Exposure to information, belief integration, and individual re

sponsiveness to agenda change. American Political Science Review, 78, 372-391.

Rao, R., & Miller, R. (1971). Applied econometrics. Belmont, CA: Wadsworth.

Schuman, H., & Presser, S. (1981). Questions and answers in attitude surveys. New York:

Wiley. Stimson, J. (1975). Belief systems, constraint and the 1972 election. American Journal of

Political Science, 19, 393-417.

Stouffer, S. (1954). Communism, conformity, and civil liberties. New York: Wiley.

Taylor, S. E., & Fiske, S. (1978). Salience, attention, and attribution: Top of the head

phenomena. In L. Berkowitz (Ed.), Advances in experimeyital social psychology (pp.

249-288). New York: Academic Press.

Wiley, D., & Wiley, J. (1970). The estimation of measurement error in panel data,

American Sociological Reineiv, 35, 1122-1127.

Zaller, J. (1987). The diffusion of political attitudes. Journal of Personality and Social

Psychology. Zaller, J. (1989). Bringing Converse back in: Information flow in political campaigns.

Political Analysis, 1 .

Zaller, J. (1990). Political competition and public opinion. Unpublished manuscript, UCLA.

Zaller, J., & Feldman, S. (1989, Julv). A simple model of the survey response. Revision of

paper delivered at 1988 meetings of Political Methodologv Society in Los Angeles.

(Available from authors)

APPENDIX

AWARENESS MEASURES

Information Scale. The panel information scale contains several tests of

respondents' ability to correctly locate candidates or groups on a 7-point policy scale. Each test is based on a combination of three variables: an initial 7-point

policy self-placement scale and two variables asking the respondent to place a group or candidate on the same scale. In order to be counted correct, a

POLITICAL AWARENESS 153

respondent must state an opinion and also locate the two candidates or groups

in appropriate left- right order. The location tests common to both forms are McGovern/Nixon on pro

gressive tax scale (vl79 and vl80, or v662 and v663, depending on form of

interview); McGovern/Wallace on aid to minorities (v632 and v631); Democrats/

Republicans on school busing (v207 and v206); McGovern/Nixon on job guar antees (v614 and v615, or vl73 and vl74); McGovern/Nixon on rights of

accommodation (v622 and v623); McGovern/Nixon on Vietnam (vl85 and

vl86, or v591 and v592); McGovern/Nixon on liberal-conservative scale (v653 and v654); Wallace/Democrats on liberalism-conservatism (v655 and v656). On the last three tests, respondents had to place the candidates at least two units apart in order to receive credit for a correct answer.

The additional Form I items are recognition of (i.e., willingness to rate)

Henry "Scoop" Jackson (v259) and Pete McClosky (v263) (dk = 0, else

=

1);

which party is more conservative (v500); how many times a president can be

reelected (v943); terms of U.S. senator and representative (v944 and v949);

party control of Congress prior to and after recent election (v950, v951); recall

names of House candidates (based on v945 and v946). The additional Form

II items were China's form of government (v59) and China's UN status (v60);

Democrats/Republicans on Vietnam (vl89 and vl88); liberals/conservatives

on student unrest (v685 and v684).

The final scales were created by adding up all 0-1 variables, averaging across missing data in the standard way.

Participation Scale. The variables are v392, v396, v468 through v472, v474,

v475; v2195 through v2199; v3529, v3530, v3532 through v3535, v3537, v3538.

Media Exposure Scale. The media exposure scale contains 16 items, each

coded to a 0-3 range. All but three of these items are based on two separate

variables the first asking whether an individual attends to a particular medium, and the second, asked only of those replying affirmatively to the first, inquiring how much attention is paid to that medium. The item pairs are v456 and v457;

v459 and v460; v461 and v462; v463 and v464; v2036 and v2037; v2046 and

each of v2050, v2053, v2056; v3600 and v3601; v3602 and v3603; v3604 and

v3605; v3606 and v3607; v3639 and v3646. The following items were used

individually: v2031, v3648, and v3651.

Political Interest Scale. The variables are vl63, v476, v2027, v3031, v3599.

Reproduced with permission of the copyright owner. Further reproduction prohibited without permission.