Attitudes and attitude change.

KEY WORDS: persuasion, evaluation, bias, attitude structure, affect

CONTENTS

INTRODUCTION ATTITUDE BASES AND STRUCTURE

Structural and Functional Bases of Attitudes

Individual Differences as a Basis of Attitudes ATTITUDE CHANGE

High-Elaboration Processes: Focus on Biased Information Processing

Peripheral Processes

Multiple Roles for Persuasion Variables: Focus on Effects of Mood

on Persuasion

Continuing Research on Persuasion Variables CONSEQUENCES OF ATTITUDES

The Impact of Attitudes on Behavior

The Impact of Attitudes on Information Processing and Social Judgments CONCLUSIONS

INTRODUCTION

Because of the sheer amount of published research from 1992 to 1995, Allport's (1935) statement that "attitude" is the single most indispensable construct in social psychology may again be true. In addition, a plethora of new books (e.g. Eagly & Chaiken 1993, Perloff 1993, Petty & Krosnick 1995, Shavitt & Brock 1994, Stiff 1994) provided further testament to the vitality of the field.

Because of space limitations, many interesting applications of attitude change theory, especially in the areas of counseling and consumer psychology (e.g. Heesacker et al 1995), cannot be included in this review. We note that interest in attitude measurement remains strong. In recent years, a small cottage industry has developed around studying the best ways to ask questions in attitude surveys (e.g. Schwarz & Sudman 1996, Tanur 1992). Notable developments in attitude measurement included Roese & Jamieson's (1993) review of 20 years of research on the bogus pipeline procedure, and continuing work to develop an actual pipeline for assessing controversial attitudes. For example, Fazio et al (1995) developed an unobtrusive measure of racial prejudice based on the automatic attitude activation effect (Fazio et al 1986). Cacioppo et al (1994) developed a procedure in which the amplitude of late positive brain potentials is used to detect the extent of evaluative consistency between a target and prior stimuli. These late positive potentials were present regardless of the accuracy of participants' attitude reports (Crises et al 1995).

Two interesting new research areas emerged. One is the dynamical systems approach to attitudes (e.g. Eiser 1994). For example, Latane & Nowak (1994) derived from catastrophe theory the notion that attitudes should become more categorical with increases in involvement (see also Sherif & Sherif 1967). Another addressed implicit theories of persuasion. One study, for instance, found that although people have beliefs consistent with classical conditioning theory, they do not show a general belief in cognitive dissonance effects (Snell et al 1995). Although there was some past work on this topic (e.g. Rule & Bisanz 1987), there now appears to be a critical mass of researchers interested in this issue (e.g. Broniarczyk & Alba 1994, Friestad & Wright 1994, Kover 1995, Trafimow & Davis 1993).

Of course, most of the work during our review period continued themes that were dominant in earlier periods, falling into three traditional areas: the structure and bases of attitudes, attitude change, and the consequences of attitudes.

ATTITUDE BASES AND STRUCTURE

Attitudes have been defined in a variety of ways, but at the core is the notion of evaluation. Thus, attitudes are commonly viewed as summary evaluations of objects (e.g. oneself, other people, issues, etc) along a dimension ranging from positive to negative (e.g. Petty et al 1994). One traditional theme in attitude research is the investigation of the underlying bases and structure of these evaluations. Much work on the bases and structure of attitudes was carried out under the label of attitude strength because differences in the underlying structure of attitudes is thought to produce differences in strength. A recent edited book (Petty & Krosnick 1995) contains reviews of the many variables thought to make attitudes strong (i.e. persist over time, resist counterpersuasion, and have an impact on judgments and behavior) (Krosnick & Petty 1995).

Structural and Functional Bases of Attitudes

ACCESSIBILITY Research suggests that the strength of object-evaluation associations (i.e. the accessibility of attitudes) has important implications for understanding the functioning of attitudes (for a review, see Fazio 1995). One recent controversy has centered around the automatic attitude activation effect. Fazio et al (1986) argued that attitudes can sometimes be automatically activated from memory upon merely encountering an attitude object. They found that the strength of this effect increased as the accessibility of people's attitudes toward the attitude-object increased. Bargh et al (1992) challenged this conceptualization. First, they presented evidence that automatic activation of attitudes occurred for relatively inaccessible attitudes. Second, contrary to the model proposed by Fazio et al, they reported results suggesting that normative accessibility (i.e. differences across objects) rather than idiosyncratic accessibility (i.e. differences across individuals) determined the strength of the automatic activation effect. However, Fazio (1993, 1995) noted that although Bargh et al (1992) obtained consistent evidence for automatic activation even at low levels of accessibility, whereas his original studies did not, the strength of the automatic activation effect in Bargh et al (1992) was still moderated by attitude accessibility. Fazio (1993, 1995) also criticized evidence supporting the superiority of normative measures of accessibility by arguing that these analyses failed to adequately control for individual differences in baseline speed of responding. He noted that reanalyses of the Bargh et al (1992) data found that idiosyncratic measures of accessibility were superior to normative measures when stronger controls for individual differences in baseline responding were included. More recently, Chaiken & Bargh (1993) have suggested that although accessibility moderates automatic activation using Fazio's traditional paradigm, it does not do so under certain procedural conditions or when the task is made nonevaluative (Bargh et al 1996). Thus, accessibility moderates automatic activation in Fazio's traditional paradigm, but perhaps not in other paradigms. Future research must clarify the psychological mechanisms responsible for regulating when accessibility does or does not moderate the automatic activation of attitudes.

Another relevant topic concerns the use of repeated attitude expressions as a manipulation of the strength of the object-evaluation association. Manipulating how often a person expresses his or her attitude has generally been thought to influence the accessibility of the attitude without changing other properties of the attitude (e.g. Powell & Fazio 1984). However, Judd and colleagues found that repeated expression manipulations also influenced the extremity of attitudes (Brauer et al 1995, Downing et al 1992; see also Judd & Brauer 1995). This effect was also obtained for nonevaluative judgments (Downing et al 1992; cf Mandler et al 1987). However, the exact mechanisms underlying these effects and their generality remain a matter of considerable speculation (Fazio 1995, Judd & Brauer 1995).

Research also examined the conditions under which repeated expression leads to enhanced attitude accessibility. Breckler & Fried (1993) investigated the moderating role of object representation. They found that responses were faster when preceded by previous ratings of objects in the same representational format (i.e. odor-odor or verbal label of the odor-verbal label of the odor) but not when preceded by previous ratings in a different representational format (i.e. odor-verbal label or verbal label-odor). In addition, Maio & Olson (1995a) found that both truthful and untruthful repeated attitude expression led to enhanced attitude accessibility as long as the untruthful responses required subjects to consciously recall their true attitude.

Although the importance of an attitude issue is determined largely by its perceived self-relevance (Boninger et al 1995a, Petty et al 1992), evidence that the importance of an attitude increases as the mere number of times a person expresses an attitude increases was obtained by Roese & Olson (1994). Mediational analyses suggested that the impact of repeated expression on importance was mediated by its influence on the accessibility of the attitude. Roese & Olson suggested that the association between accessibility and importance might occur because people use ease of retrieval as a cue for inferring importance.

AMBIVALENCE Another structural property of attitudes that has been the focus of recent attention is the extent to which attitudes are ambivalent (i.e. based on evaluatively inconsistent information). An interesting development in this literature is Cacioppo & Berntson's (1994) Bivariate Evaluative Space Model. Cacioppo & Berntson suggested that researchers have often assumed that positive and negative evaluative reactions are reciprocally activated (i.e. increases in one will be associated with decreases in the other), but that findings from research literatures as diverse as attitude research and animal learning suggest this assumption is often not tenable. Although explicit adoption of this assumption in the attitude literature may be rare, it is possible that some researchers improperly interpreted the ubiquitous negative correlation between positive and negative evaluative responses as implying reciprocal activation. Cacioppo & Berntson argued that the relation between positive and negative responses should be viewed as a bivariate evaluative plane in which reciprocal or coactive activation can occur. They noted the inability of traditional bipolar attitude scales to fully differentiate among these possibilities and suggested that future research use separate measures of the positive and negative bases of attitudes (perhaps in addition to the traditional bipolar assessment of attitudes; see also Kaplan 1972, Thompson et al 1995).

It is important to recognize that any measure of attitudes, be it bipolar or otherwise, will fail to capture fully the multitude of potentially important differences in the structure and bases of attitudes. Even measures that assess positivity and negativity separately will be unable to distinguish among attitudes that differ in other meaningful ways (e.g. accessibility, affective/cognitive bases, etc). Thus, although there are benefits of examining the positive and negative bases of attitudes, it is unclear whether failure to do so should be considered a fundamental measurement flaw any more than failure to assess other bases and structural properties of attitudes.

Other ambivalence research focused on a variety of empirical issues. Some research examined the validity of different formulas for combining positive and negative responses to arrive at an overall index of attitudinal ambivalence (Breckler 1994, Thompson et al 1995). Thompson & Zanna (1995) examined the role of personality dispositions (i.e. need for cognition, personal fear of invalidity) and domain-specific factors (i.e. issue involvement) as antecedents of ambivalence. Leippe & Eisenstadt (1994) found that compliance with a counterattitudinal request in a dissonance paradigm increased as attitudinal ambivalence increased. Finally, Vallacher et al (1994) used a measure that assessed moment-to-moment shifts in evaluation to demonstrate that increased ambivalence was associated with lower attitude stability.

AFFECTIVE/COGNITIVE BASES OF ATTITUDES Conceptualizing attitudes as having affective (emotional) and cognitive (belief) bases has been one of the most popular means of classifying the different types of information upon which attitudes are based. One theme that has emerged recently is an increased concern with appropriate measurement of attitude-relevant affect and cognition. Eagly et al (1994) criticized past research for relying on close-ended measures of affect and cognition (e.g. rating scales, checklists), which they suggested suffered from methodological limitations. To correct these problems, Eagly et al (1994) used open-ended measures in which participants were asked to list their emotions and beliefs separately and found that these measures of affect and cognition often contributed unique explanatory power to the prediction of attitudes. Unfortunately, no empirical comparisons were made between the open-ended measures and traditional close-ended measures, so it was impossible to confirm that the new measures were an improvement over past measures.

A different approach was reported by Crites et al (1994). In an initial study, evidence was obtained suggesting that many measures used in past affect/cognition research lacked important psychometric properties (e.g. reliability, convergent validity, and discriminant validity). Two subsequent studies showed that new scales designed to assess the affective and cognitive bases of attitudes had high levels of reliability across attitude objects as diverse as social issues, academic subjects, and animals. Factor analyses of the scales also suggested that the scales had good convergent and discriminant validity. Finally, the scales could detect an experimental manipulation of the affective and cognitive bases of attitudes.

Research also continued to explore the extent to which attitudes are based on affect and cognition in various domains. Haddock et al (1993) found that, for people high in authoritarianism, attitudes toward homosexuals were driven primarily by symbolic beliefs and past experiences. In contrast, among people low in authoritarianism, attitudes were determined primarily by stereotype beliefs and affect. Rosselli et al (1995) found that postmessage attitudes were based on affective and cognitive responses when the message was emotional in nature but only on cognitive responses when the message was fact based. These results only occurred when participants were in a neutral mood. In positive moods, cognitive and affective responses had no impact on postmessage attitudes.

VALUES AND ATTITUDE FUNCTIONS AS BASES OF ATTITUDES Researchers continued to explore the extent to which values and attitude functions influence attitudes. Feather (1995) found that the importance people placed in specific values influenced their attitudes toward behavioral choices designed to reflect different value orientations (see also Stern et al 1995). Gastil (1992) found that the level of support for democracy was related to the extent to which democracy was seen as fulfilling a value-expressive function and an ego-defensive function. Other research examined the extent to which the functional basis of an attitude moderated the relation between values and attitudes. For instance, Mellema & Bassili (1995) found that values were more predictive of attitudes for people low in self-monitoring than for people high in self-monitoring. Similarly, Maio & Olson demonstrated by measuring (Maio & Olson 1994) and by manipulating attitude functions (Maio & Olson 1995b) that the impact of values on attitudes and behavior increased as the extent to which attitudes served a value-expressive function increased.

Individual Differences as a Basis of Attitudes

One interesting recent development has been the recognition among psychologists that attitudes can have some genetic basis (e.g. Lykken et al 1993) and the implications of this. For instance, Tesser (1993) argued that attitudes that have a substantial genetic basis will tend to be stronger than attitudes with little genetic basis. To support this assertion, he conducted a series of studies comparing attitudes that past research indicated varied in their genetic basis. Tesser found that as the amount of variance in attitudes attributable to genetic factors increased, so did the accessibility of attitudes in memory, the resistance of attitudes to conformity pressures, and the impact of the attitudes on interpersonal attraction.

Research on the genetic basis of attitudes challenges traditional attitude theories that have stressed the role of experience as the basis of attitudes (e.g. McGuire 1969). However, there are important limitations to this research. First, methodological challenges make exact estimates of the genetic versus environmental basis of attitudes controversial (e.g. Bouchard et al 1992, Olson & Zanna 1993, see Cropanzano & James 1990). A second limitation is a lack of clearly articulated or empirically verified mediating processes. Discussions of the possible mechanisms by which genetic factors might influence attitudes have been brief and speculative, though highly interesting (e.g. Tesser 1993).

ATTITUDE CHANGE

In the 1950s, researchers assigned a critical role for memory of message arguments as a mediator of persuasion (e.g. Hovland et al 1953). Following Greenwald et al's (1968) comparison of message memory versus cognitive responses as mediators of attitude change, however, research on memory and persuasion waned. Over the past decade, however, a limited but important role for message memory has been established. Specifically, as might be anticipated from Hastie & Park's (1986) research, current work indicates that message memory is most important in predicting attitudes when elaboration of message arguments at the time of exposure is unlikely, an unexpected judgment is requested sometime after message exposure, and simple cues to message validity are relatively unavailable at the time of judgment (Frey & Eagly 1993, Haugtvedt & Petty 1992, Haugtvedt & Wegener 1994, Mackie & Asuncion 1990; for a discussion, see Petty et al 1994). In such circumstances, people apparently judge the advocacy by retrieving whatever message substance they can, and then either evaluating these recalled arguments or making an inference of validity based on the number of arguments remembered.

Much persuasion work continues to be guided by the elaboration likelihood model (ELM) (Petty & Cacioppo 1986) and the heuristic-systematic model (HSM) (Chaiken et al 1989). These models have likely maintained their popularity over the past five review periods in part because these theories encompass the effects of a multitude of persuasion variables, processes, and outcomes. Although differences between these frameworks have been noted that could be important in certain circumstances, the theories are generally more similar than different, and typically they can accommodate the same empirical results, though the explanatory language and sometimes the assumed mediating processes vary (see Eagly & Chaiken 1993, Petty 1994, Petty & Wegener 1997). In the ELM, the central route (high-effort scrutiny of attitude-relevant information) and peripheral route (less effortful shortcuts to evaluating attitude objects) anchor opposite ends of an elaboration likelihood continuum. Even though "central" processes increase in impact as elaboration increases across the continuum and "peripheral" processes decrease in impact as elaboration increases, attitude change is often determined by both central and peripheral processes (though much early research attempted to capture one or the other by …