Accessibility and Stability of Predictors in the Theory of Planned Behavior.

12
Journal of Personality and Social Psychology 1992, Vol.63, No. 5, 754-765 Copyright 1992 by the American Psychological Association, Inc. 0O22-3514/92/S3.OO Accessibility and Stability of Predictors in the Theory of Planned Behavior Jorg Doll Universitat Hamburg, Federal Republic of Germany Icek Ajzen University of Massachusetts at Amherst To test the propositions that an attitude's ability to predict behavior is influenced by its temporal stability and by its accessibility in memory, 75 students were given direct or indirect experience with 6 video games and fun or skill instructions. They completed a computer-administered ques- tionnaire before and after a free-play period. On the basis of Ajzen's (1988) theory of planned behavior, time played with each game was correlated with attitudes, perceived behavioral control, and intentions with respect to playing each game. Direct experience and fun instructions improved prediction of behavior, lowered latencies of responses to questionnaire items, and increased their temporal stabilities. The improved prediction of behavior was found to be mediated by the tem- poral stabilities of the predictor variables but, contrary to expectations, not by response latencies. It is suggested that more attention be given to the role of the stability of variables in attitude-behavior models. People do not always behave in accordance with their atti- tudes. Prediction of behavior from verbal attitudes tends to improve to the extent that the attitude is based on direct experi- ence rather than on indirect experience or second-hand infor- mation (Fazio & Zanna, 1978a, 1978b; Regan & Fazio, 1977). In their review of this research, Fazio and Zanna (1981) observed that attitudes based on direct experience rely on more informa- tion, are better denned, are held with greater confidence, are more stable over time, and are more accessible in memory, and they suggested that these characteristics may be responsible for the moderating effect of direct experience on the attitude-be- havior relation. More recently, Fazio and his associates (Fazio, 1986,1990a; Fazio, Chen, McDonel, & Sherman, 1982; Fazio & Williams, 1986) have focused on accessibility in memory as the crucial factor. An attitude is said to be highly accessible if there is a strong association between the attitude object and an evalua- tive response. This associative strength is defined operationally as the time it takes to react to questions about the attitude object: The smaller the response latency, the more accessible in memory the attitude is assumed to be. Attitude accessibility is said to determine the extent to which an attitude is automati- cally activated on exposure to the attitude object, and hence the extent to which the attitude is likely to guide behavior in the presence of the object. Specifically, Fazio (1990a) proposed that accessible attitudes exert a strong biasing effect on perception of the behavioral situation, making it more likely that the behav- ior will be consistent with the attitude. It follows that attitude- behavior consistency should increase with attitude accessibility. In this account, attitude accessibility is a mediator that ex- plains the effect of type of experience (direct or indirect) on the attitude-behavior relation, and atfirstglance, research by Fazio This article has benefited from comments on an earlier draft pro- vided by Russell Fazio. We are also grateful to several anonymous reviewers who provided valuable suggestions for revisions. Correspondence concerning this article should be addressed to Icek Ajzen, Department of Psychology, Tobin Hall, University of Massa- chusetts, Amherst, Massachusetts 01003-0034. and his associates appears to provide empirical support for a mediational sequence. First, as noted earlier, direct experience has repeatedly been found to improve the predictive validity of attitudes (see Fazio & Zanna, 1981, for a review). Second, there is evidence that, in comparison with indirect experience, direct experience lowers latencies of responses to attitudinal ques- tions (Berger & Mitchell, 1989; Fazio et al., 1982). Third, and to complete the sequence, several studies have shown that attitu- dinal responses with relatively low latencies are better predic- tors of behavior (Fazio, Powell, & Williams, 1989; Fazio & Wil- liams, 1986; Houston & Fazio, 1989). Note, however, that the studies in this third set have either measured existing attitude latencies or have manipulated latencies by varying the number of times respondents were asked to express their attitudes. None of the studies that showed an effect of latency on atti- tude-behavior correspondence has examined the role of direct experience. Perhaps of greater importance, the three parts of the hypoth- esized mediational sequence were examined in different stud- ies, and—to the best of our knowledge—no direct overall test of the accessibility thesis in relation to direct experience has as yet been conducted. Three effects must be obtained in a given set of data to demonstrate mediation (Baron & Kenny, 1986). In the present context, the first two effects involve showing that type of experience influences attitude latency (the accessibility mediator) and that type of experience influences the attitude- behavior relation. We noted earlier that empirical evidence has been provided for both of these effects, albeit in separate inves- tigations. The third part of the test for mediation involves dem- onstrating that the effect of experience type on the attitude-be- havior relation is substantially reduced (ideally to nonsignifi- cance) when the mediator (attitude accessibility) is held constant. This can best be tested by regressing the attitude-be- havior relation simultaneously on type of experience and on attitude latency (Baron & Kenny, 1986). This critical test of the mediation hypothesis has apparently not been performed. One purpose of the present study was to submit the accessibility thesis to the required direct test. 754

Transcript of Accessibility and Stability of Predictors in the Theory of Planned Behavior.

Page 1: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

Journal of Personality and Social Psychology1992, Vol.63, No. 5, 754-765

Copyright 1992 by the American Psychological Association, Inc.0O22-3514/92/S3.OO

Accessibility and Stability of Predictors in the Theory of Planned Behavior

Jorg DollUniversitat Hamburg, Federal Republic of Germany

Icek AjzenUniversity of Massachusetts at Amherst

To test the propositions that an attitude's ability to predict behavior is influenced by its temporalstability and by its accessibility in memory, 75 students were given direct or indirect experiencewith 6 video games and fun or skill instructions. They completed a computer-administered ques-tionnaire before and after a free-play period. On the basis of Ajzen's (1988) theory of plannedbehavior, time played with each game was correlated with attitudes, perceived behavioral control,and intentions with respect to playing each game. Direct experience and fun instructions improvedprediction of behavior, lowered latencies of responses to questionnaire items, and increased theirtemporal stabilities. The improved prediction of behavior was found to be mediated by the tem-poral stabilities of the predictor variables but, contrary to expectations, not by response latencies. Itis suggested that more attention be given to the role of the stability of variables in attitude-behaviormodels.

People do not always behave in accordance with their atti-tudes. Prediction of behavior from verbal attitudes tends toimprove to the extent that the attitude is based on direct experi-ence rather than on indirect experience or second-hand infor-mation (Fazio & Zanna, 1978a, 1978b; Regan & Fazio, 1977). Intheir review of this research, Fazio and Zanna (1981) observedthat attitudes based on direct experience rely on more informa-tion, are better denned, are held with greater confidence, aremore stable over time, and are more accessible in memory, andthey suggested that these characteristics may be responsible forthe moderating effect of direct experience on the attitude-be-havior relation. More recently, Fazio and his associates (Fazio,1986,1990a; Fazio, Chen, McDonel, & Sherman, 1982; Fazio &Williams, 1986) have focused on accessibility in memory as thecrucial factor. An attitude is said to be highly accessible if thereis a strong association between the attitude object and an evalua-tive response. This associative strength is defined operationallyas the time it takes to react to questions about the attitudeobject: The smaller the response latency, the more accessible inmemory the attitude is assumed to be. Attitude accessibility issaid to determine the extent to which an attitude is automati-cally activated on exposure to the attitude object, and hence theextent to which the attitude is likely to guide behavior in thepresence of the object. Specifically, Fazio (1990a) proposed thataccessible attitudes exert a strong biasing effect on perceptionof the behavioral situation, making it more likely that the behav-ior will be consistent with the attitude. It follows that attitude-behavior consistency should increase with attitude accessibility.

In this account, attitude accessibility is a mediator that ex-plains the effect of type of experience (direct or indirect) on theattitude-behavior relation, and at first glance, research by Fazio

This article has benefited from comments on an earlier draft pro-vided by Russell Fazio. We are also grateful to several anonymousreviewers who provided valuable suggestions for revisions.

Correspondence concerning this article should be addressed to IcekAjzen, Department of Psychology, Tobin Hall, University of Massa-chusetts, Amherst, Massachusetts 01003-0034.

and his associates appears to provide empirical support for amediational sequence. First, as noted earlier, direct experiencehas repeatedly been found to improve the predictive validity ofattitudes (see Fazio & Zanna, 1981, for a review). Second, thereis evidence that, in comparison with indirect experience, directexperience lowers latencies of responses to attitudinal ques-tions (Berger & Mitchell, 1989; Fazio et al., 1982). Third, and tocomplete the sequence, several studies have shown that attitu-dinal responses with relatively low latencies are better predic-tors of behavior (Fazio, Powell, & Williams, 1989; Fazio & Wil-liams, 1986; Houston & Fazio, 1989). Note, however, that thestudies in this third set have either measured existing attitudelatencies or have manipulated latencies by varying the numberof times respondents were asked to express their attitudes.None of the studies that showed an effect of latency on atti-tude-behavior correspondence has examined the role of directexperience.

Perhaps of greater importance, the three parts of the hypoth-esized mediational sequence were examined in different stud-ies, and—to the best of our knowledge—no direct overall testof the accessibility thesis in relation to direct experience has asyet been conducted. Three effects must be obtained in a givenset of data to demonstrate mediation (Baron & Kenny, 1986). Inthe present context, the first two effects involve showing thattype of experience influences attitude latency (the accessibilitymediator) and that type of experience influences the attitude-behavior relation. We noted earlier that empirical evidence hasbeen provided for both of these effects, albeit in separate inves-tigations. The third part of the test for mediation involves dem-onstrating that the effect of experience type on the attitude-be-havior relation is substantially reduced (ideally to nonsignifi-cance) when the mediator (attitude accessibility) is heldconstant. This can best be tested by regressing the attitude-be-havior relation simultaneously on type of experience and onattitude latency (Baron & Kenny, 1986). This critical test of themediation hypothesis has apparently not been performed. Onepurpose of the present study was to submit the accessibilitythesis to the required direct test.

754

Page 2: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

STABILITY OF PREDICTORS 755

A second major objective of the present research was to exam-ine the viability of positing an additional mediating process toaccount for the effect of direct experience on the attitude-beha-vior relation. In the view of Fazio and his associates, such fac-tors as confidence in an expressed attitude and the attitude'stemporal stability are merely alternative indicators of attitu-dinal strength or accessibility in memory, indicators that arepresumably inferior to a measure of response latency (see Fazioet al., 1982). Contrary to this view, a recent study by Krosnick,Boninger, Chuang, and Carnot (1991) found that a multidimen-sional model was necessary to account for the relations amongsuch different indicators of attitude strength as extremity, in-volvement, affective-cognitive consistency, and accessibility.The present article takes issue with the proposition that attitudeaccessibility is the only or the primary factor mediating theinfluence of direct experience on the attitude-behavior rela-tion. We suggest that attitude stability is an equally plausiblemediating factor and that its effects cannot be explained byreference to the attitude's accessibility in memory. Because thisargument is tested in the context of Ajzen's (1988,1991) theoryof planned behavior, we begin with a brief description of thetheory. More detailed accounts, as well as reviews of empiricalsupport for the theory, can be found in Ajzen (1988,1991).

Theory of Planned Behavior

Intention and Perceived Behavioral Control

The theory of planned behavior is an extension of the theoryof reasoned action (Ajzen & Fishbein, 1980; Fishbein & Ajzen,1975) made necessary by the original model's limitations indealing with behaviors over which people have incomplete voli-tional control. According to the theory, behavioral perfor-mance can be predicted from people's intentions to perform thebehavior in question and from their perceptions of control overthe behavior. The relative importance of intention and per-ceived behavioral control in the prediction of behavior is ex-pected to vary across situations and across different behaviors.When the behavior, or situation, affords a person complete con-trol over behavioral performance, intention alone should besufficient to predict behavior, as specified in the theory of rea-soned action. The addition of perceived behavioral controlshould become increasingly important as volitional controlover the behavior declines.1

Determinants of Intentions

The theory of planned behavior postulates three concep-tually independent determinants of intention. The first is theattitude toward the behavior and refers to the degree to which aperson has a favorable or unfavorable evaluation or appraisal ofthe behavior in question. The second predictor is a social factortermed subjective norm; it refers to the perceived social pressureto perform or not to perform the behavior. The third antecedentof intention is the degree of perceived behavioral control, whichrefers to the perceived ease or difficulty of performing the be-havior and is assumed to reflect past experience as well as antici-pated impediments and obstacles. As a general rule, the morefavorable the attitude and subjective norm with respect to a

behavior, and the greater the perceived behavioral control, thestronger should be an individual's intention to perform the be-havior under consideration.

The Role of Beliefs

At the most basic level of explanation, the theory postulatesthat behavior is a function of salient information, or beliefs,relevant to the behavior. People can hold a great many beliefsabout any given behavior, but they can attend to only a relativelysmall number at any given moment (see Miller, 1965). It is thesesalient beliefs that are considered to be the prevailing determi-nants of a person's intentions and actions. Three kinds of salientbeliefs are distinguished: behavioral beliefs, which are assumedto influence attitudes toward the behavior; normative beliefs,which constitute the underlying determinants of subjectivenorms; and control beliefs, which provide the basis for percep-tions of behavioral control.

According to the theory, attitudes develop reasonably fromthe beliefs people hold about the object of the attitude. Gener-ally speaking, we form beliefs about an object by associating itwith certain attributes, that is, with other objects, characteris-tics, or events. In the case of attitudes toward a behavior, eachbelief links the behavior to a certain outcome or to some otherattribute such as the cost incurred by performing the behavior.Since the attributes that come to be linked to the behavior arealready valued positively or negatively, we simultaneously ac-quire an attitude toward the behavior. In this fashion, we learnto evaluate favorably behaviors we believe have largely desirableconsequences and we form unfavorable attitudes toward behav-iors we associate with mostly undesirable consequences.

Subjective norm is also assumed to be a function of beliefs,but in this case beliefs about the normative expectations ofsalient referent individuals or groups. Perceived social pressureis assumed to increase to the extent that salient referents withwhom a person is motivated to comply are seen as approving ofthe behavior under consideration.

Finally, perceived behavioral control can also be traced to aset of underlying beliefs, beliefs that deal with the presence or

1 The original derivation of the theory of planned behavior (Ajzen,1985) differed in two respects from the model presented here. First, itdefined intention (and the theory's other theoretical constructs) interms of trying to perform a given behavior rather than in relation toactual performance. However, early work with the model showedstrong correlations between measures of the model's variables thatasked about trying to perform a given behavior and measures thatdealt with actual performance of the behavior (Ajzen & Madden, 1986;Schifter & Ajzen, 1985). Because the latter measures are less cumber-some, they have been used in subsequent research, and the variablesare now denned more simply in relation to behavioral performance(see, however, Bagozzi & Warshaw, 1990, in press, for work on theconcept of trying to attain a behavioral goal). Second, the originalformulation of the theory postulated an interaction between perceivedbehavioral control and intention, but empirical research to date hasrevealed only main effects. The present study again obtained no evi-dence for a significant interaction between perceived control and in-tention. The absence of significant interactions can perhaps be attrib-uted to the fact that research participants rarely express highly negativeintentions, a condition necessary for strong interactions to obtain.

Page 3: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

756 JORG DOLL AND ICEK AJZEN

absence of requisite resources and opportunities. These controlbeliefs may be based in part on past experience with the behav-ior, but they will usually also be influenced by second-handinformation about the behavior, by the experiences of acquain-tances and friends, and by other factors that increase or reducethe perceived difficulty of performing the behavior in question.The more resources and opportunities individuals believe theypossess, and the fewer obstacles or impediments they antici-pate, the greater should be their perceived control over the be-havior.

Prior Experience and the Theory of Planned Behavior

The discussion of the theory of planned behavior suggeststhat information is at the root of intentions and actions. Behav-ior-relevant information, represented in memory in the form ofbehavioral, normative, and control beliefs, results in the forma-tion of attitudes, subjective norms, and perceptions of behav-ioral control, and new information may lead to the formation ofnew beliefs or alter previously formed beliefs. Now, it stands toreason that direct, behavioral experience may provide very dif-ferent information to a person than does indirect, nonbeha-vioral experience, leading to the formation of different behav-ioral, normative, and control beliefs. In support of this expecta-tion Doll and Mallii (1990) found that individuals with directexperience generated, in a free-elicitation task, significantlymore beliefs referring to behavioral consequences than did indi-viduals with indirect experience who generated significantlymore object-descriptive beliefs.

In a typical study on the effects of type of experience on theattitude-behavior relation, participants are given either directexperience with respect to a certain behavior or are indirectlyexposed to the behavioral situation. A questionnaire is thenadministered to assess the theoretical constructs of interest,followed by a test period in which the participants have anopportunity to engage in the behavior under consideration.Clearly, any new behavior-relevant information that becomesavailable in the test period may change some of the beliefs thatunderlie attitudes toward the behavior, subjective norms, andperceptions of behavioral control. Thus, participants may cometo believe that performing the behavior is less interesting thanthey thought or that it is more difficult than they anticipated.These changes, in turn, may alter intentions and subsequentbehavior. However, because the theory's constructs are assessedprior to the test period, any changes that occur will tend tolessen the observed relations with behavior (see also Ajzen &Fishbein, 1980; Schwartz, 1978; Wood, 1982).

The present study is based on the proposition that the type ofprior experience participants have had will affect the degree towhich their beliefs, attitudes, and intentions remain stable inthe test period. An influential source of information for thedevelopment of attitudes, subjective norms, and perceived be-havioral control is the feedback one obtains when performingthe behavior. By performing the behavior, one can learn aboutits positive or negative consequences, about requisite resources,and about the reactions of other people. In the direct experi-ence condition of an experiment, this information is availableto the participants during their initial encounter with the behav-ior; little new information is added in the test period. In con-

trast, participants in the indirect experience condition actuallyperform the behavior for the first time in the test period. Muchmore of the feedback they receive will be new to them, leadingto more change in their beliefs, attitudes, and intentions. Thetheory's predictive validity should thus be greater in the directas opposed to the indirect experience condition.

Note that this hypothesis is independent of the latency oraccessibility of the beliefs, attitudes, and intentions. Even whendirect and indirect experience produce equally accessible dis-positions, because of their differential stability, we expect bet-ter prediction of behavior under direct experience (see Doll &Mallii, 1990).

Hypotheses

In light of the foregoing discussion we now formulate severalresearch hypotheses. First, it is predicted that attitudes, percep-tions of behavioral control, and intentions are more stable andmore accessible under conditions of direct experience thanunder conditions of indirect experience.2 This can be tested bycomparing measures of the different variables obtained beforeand after the test period and by assessing the latencies of thesevariables. Second, because of the differential stability and ac-cessibility of intentions and perceived behavioral control, pre-diction of behavior from these variables is more accurate underdirect- as compared with indirect-experience conditions. Be-cause past research has focused on the attitude-behavior rela-tion, we also examine how type of prior experience affects thisrelation. Third, if response latency and temporal stability ofpredictors do indeed mediate the effect of type of experienceon accuracy of prediction, we expect that this effect is dimin-ished or eliminated when response latency, temporal stability,or both are statistically controlled. Furthermore, we hypothe-size that the effect of the predictors' stabilities on accuracy ofbehavioral prediction is at least partly independent of the pre-dictors' accessibilities in memory.

The next hypothesis is derived from the stability explanationof the moderating effect of direct experience; it does not followfrom the accessibility theorem. According to the stabilitytheorem, prediction of intentions from attitudes, subjectivenorms, and perceptions of behavioral control should not differsignificantly between the two experience conditions. This hy-pothesis is based on the fact that all of the variables involved areassessed before the behavior. Changes that occur during the testperiod can thus have no impact on observed relations. In con-trast, because attitudes (as well as subjective norms and percep-tions of behavioral control) formed under direct-experienceconditions are more accessible in memory, they should providebetter guides for the development of behavioral plans (inten-tions), when compared with indirect-experience conditions.The accessibility theorem thus leads to the expectation thatprediction of intentions from attitudes (and from subjectivenorms and perceived behavioral control) is better in the direct-than in the indirect-experience condition.

2 In the context of the present study, there was no reason to expectthat type of experience would influence the stability or accessibility ofsubjective norms.

Page 4: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

STABILITY OF PREDICTORS 757

Finally, to further explore the role of attitudinal accessibilityand stability, the present study varied the motivational orienta-tion induced in participants. A person's motivational orienta-tion may produce selective attention to different aspects of thebehavioral situation. Thus, an orientation that views the pur-pose of the behavior as having fun may focus attention on atti-tudinal aspects of the behavior, whereas an orientation con-cerned with the development of skills may focus attention onaspects of the behavior that have to do with behavioral control.Such selective attention could influence the strength of the pre-dictors in the theory of planned behavior. Specifically, it is hy-pothesized that attitudes are established more firmly under afun orientation, whereas perceptions of behavioral control arestronger under a skill orientation. Because stronger attitudes orperceptions of behavioral control are likely to be more stableand more accessible, we would expect that attitudes predictbehavior better under fun instructions and that perceptions ofbehavioral control are better predictors of behavior under skillinstructions. Furthermore, the effects of motivational orienta-tion on predictive accuracy are expected to be mediated bytemporal stability and accessibility in a manner comparablewith the effects of direct experience. Hence, the same media-tional hypotheses apply.

Method

Participants

Forty-three female and 32 male undergraduate students at the Uni-versity of Massachusetts participated in the experiment for extra classcredit. The participants were randomly assigned to each of four be-tween-groups experimental conditions in a 2 (direct vs. indirect experi-ence) X 2 (fun vs. skill orientation) design. The design was unbalanced,with 19 participants in the indirect-experience, skill condition; 18 par-ticipants in the indirect-experience, fun condition; 18 participants inthe direct-experience, skill condition; and 20 participants in the di-rect-experience, fun condition.

Stimulus Material

Six video games, varying in complexity and objectives, were playedon a Commodore 64 with a color monitor. They were (a) a game inwhich a pilot had to shoot down other planes and to bomb objects onthe ground, (b) a game in which all unconnected parts of a maze had tobe connected, (c) a game in which a cave man had to jump over variousobstacles, (d) a game in which a tube was moved along a track and wasused to destroy another tube with balls it could fire at the other tube, (e)a game in which a racing car was to be driven around the track as oftenas possible while scratching the other cars in the race, and (0 a game inwhich an engineer was to set up an assembly line to produce a product.

Procedure and Measures

On entering the experimental laboratory the participants signed theconsent form, which explained the purpose of the experiment.

This study will examine attitudes and opinions regarding playingwith video games. You are given the opportunity to play with sixvideo games, and then you have to answer certain questions aboutplaying with video games. The questionnaire is given to you on apersonal computer. The first part of the questionnaire containsgeneral questions regarding your attitudes toward playing withvideo games. Then you have a total of about 30 minutes to get to

know six video games we selected for this study. The second partof the questionnaire contains specific questions regarding yourattitudes and opinions toward playing with the six games. Thenyou are free to play for 45 minutes with the six video games.

The experimenter seated the participant at an IBM personal com-puter. A program, written for this purpose, controlled the experimen-tal session and registered the participants' ratings and reaction timesfor each rating. The computer instructed participants to respond to the7-point rating scales that were used for most questions by striking therespective number keys. They were told to respond quickly but accu-rately. Participants then received a warm-up question, rated their atti-tudes toward playing video games in general on a 7-point scale rangingfrom extremely enjoyable to extremely unenjoyable, reported how oftenthey had played video games in the past (at home, at a friend's home, ina video arcade, and so forth: 1 = never, 2 = 1-10 times, . . . , 6 =100-200 times, 7 = more than 200 limes) and reported how often andwith how many different games they had played (1 = never, 2 = 1-10games,. . . , 6 = 40-50 games, 7 = more than 50 games).

Next, the computer provided the participant with an instructiondesigned to manipulate motivational orientation. Participants in thefun condition were informed that "Video games are meant to be en-joyed. All we want you to do is to try to have fun playing the games."Participants in the skill condition were informed that "It takes skill toplay video games. What we want you to do is to demonstrate your skillsin playing the games. This means that you should try to keep the gamesgoing as long as possible without making a mistake."

The participants were then asked to move over to the video gamecomputer and to read the description of the first video game, whichwas written on a single sheet together with the name of the game. Atthis point the second experimental manipulation was introduced. Par-ticipants in the direct-experience condition played 4 min with eachgame while participants in the indirect-experience condition watcheda video recording of the same games, each played 4 min. The videorecording was the same for all participants, and it showed the actualplaying sequence of another person who had played the games withoutbeing given any motivational orientation. The game sequences wererecorded from the video output of the Commodore 64.

After having been introduced to the six video games the participantsmoved back to the IBM personal computer. Beside the computer theyfound a list with the names of the six video games given in alphabeticalorder. Beneath each name were written characteristic elements of thegame (e.g., fighter pilot or turn a maze green) to help the participantsidentify each game. To assess their subjective knowledge, the partici-pants were asked to rate, on a 7-point scale, how informed they feltabout the six video games {pxtremely informed-extremely uninformed).This was followed by a more objective knowledge test in the form of sixmultiple-choice items. Each item involved a certain feature of a game,and the respondents were asked to identify the game to which thefeature applied (e.g., "In which game can you duck?"). Next, the partici-pants were instructed a second time that they would be given the oppor-tunity to play 45 min with the games and that they would be free todecide how long and how often they would play with each game. Theonly restriction was that the participant would have to play at least 2min with each game. The instruction, designed to make salient the twomotivational orientations, was also repeated a second time.

At this point, the variables contained in the theory of planned behav-ior were assessed. The computer presented one question at a time,together with the 7-point rating scale. Each question was repeated sixtimes in a row, once for each video game, but the order of the videogames was randomized for each set of questions. Each variable wasmeasured by two questions, separated by items concerning other vari-ables. There was no systematic order to the sequence.

The behavior to which the questions referred was defined as "play-ing with Game X longer than the required 2 minutes." Attitude toward

Page 5: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

758 JORG DOLL AND ICEK AJZEN

this behavior was assessed by means of two evaluative scales ("Myplaying with Game X more than the required 2 minutes is" extremelypleasant-extremely unpleasant and extremely interesting-extremelyboring). The two normative questions referred to the perceived expec-tations of the experimenter, because this seemed to be the most rele-vant referent in the experimental situation ("The experimenter wouldapprove of my playing with Game X for more than the required 2minutes": extremely likely-extremely unlikely; "The experimenterthinks I should play with Game X longer than the required 2 minutes":extremely likely-extremely unlikely). Perceived behavioral control wasmeasured as follows: (a) "For me, playing with Game X longer than therequired 2 minutes is" extremely easy-extremely difficult; (b) "I canplay with Game X successfully longer than the required 2 minutes":extremely likely-extremely unlikely. Finally, intention was also assessedby means of two items ("I intend to play with Game X longer than therequired 2 minutes": extremely likely-extremely unlikely; "I will try toplay with Game X longer than the required 2 minutes": extremelylikely-extremely unlikely). For reasons described in the Results sec-tion, the pretest questionnaire contained nine filler items, separatingthe questions dealing with the different variables. The filler items gavea one-sentence description of the subsequent set of questions and askedrespondents to press one of two keys to continue. At the end of thequestionnaire, that is, before the free-play period, the motivationalorientation instruction was repeated.

After moving back to the video game computer, the participantsstarted playing the first game of their choice. The video game wasloaded by the experimenter into the computer. When the participantsbegan playing the game, the experimenter started a clock standing onthe TV monitor. The clock was stopped when the participants finishedplaying with one game, and it was started again when they proceededto the next game. The presence of the clock was explained to the partici-pants as an aid to help them gauge how long they had already playedwith a given game. If playing behavior indicated that participants hadneglected to play at least 2 min with each game, the experimenterreminded them of the instruction and asked them to play the last 2 minwith the game that was omitted.

After having played 45 min with the six games, the participantsreturned to the IBM personal computer. There the motivational orien-tation instruction was repeated a third time and the participants wereasked to judge the games again "now that they had some experiencewith the games." They were asked to rate how informed they felt aboutthe video games, followed by questions measuring the variables in thetheory of planned behavior. Although the questions pertaining to thedifferent variables were presented in the same order as in the pregamequestionnaire, the order of the six games within each section was newlyrandomized. The items concerning the theory of planned behaviordiffered only in time perspective from those asked in the pretest ques-tionnaire. Whereas the items in the pretest questionnaire had beenformulated with respect to the immediately following free-play period,the items in the posttest questionnaire were formulated with respect tothe preceding free-play period. For example, the item, "My playingwith Game X was" (extremely pleasant-extremely unpleasant) consti-tuted one measure of attitude. The only exception occurred with re-spect to the measures of intention, which were asked in relation to animagined second free-play period of 45 min (e.g., "I would play withGame X longer than the required 2 minutes": extremely likely-ex-tremely unlikely). The posttest questionnaire contained eight filleritems of the same kind as in the pretest questionnaire.

Finally, the participants reported their age and sex and were thor-oughly debriefed. They were informed about the four experimentalconditions and about the purpose of the experiment.

The participants' game behavior was recorded by a video recorderconnected to the video output of the Commodore 64. The video record-ing was analyzed with respect to the time played with each game, the

order in which the games were played, and the cumulated scores re-ceived in playing each game.

Results and Discussion

Preliminary Analyses

To rule out possible artifacts due to prior direct experience,we performed 2 (direct vs. indirect experience) X 2 (fun vs. skillorientation) analyses of variance (ANOVAs) on the number oftimes participants reported having played video games in thepast and on the number of games they had played. There wereno significant main effects or interactions. The average ratingscale category in the total sample for the number of timesplayed was 5.0, representing 50-100 times played, and the aver-age rating scale category for the number of games played was3.7, representing 20-30 games.

Next, we analyzed reports of how informed participants feltabout the video games. These reports were submitted to a three-way ANOVA, which, in addition to the experience and orienta-tion factors, included time of measurement (pre- vs. posttest) asa repeated measures factor. The only significant effect was theExperience X Orientation interaction, F{\, 70) = 6.2, p < .05. Inthe indirect-experience condition, participants in the fun orien-tation felt better informed than participants in the skill orienta-tion {M = 2.1 vs. 3.1), whereas the order was reversed in thedirect-experience condition (M = 3.2 vs. 2.7). However, thisinteraction was significant only in the pretest; it did not reachstatistical significance in the posttest measure. The multiple-choice knowledge test, submitted to an Experience X Orienta-tion ANOVA, revealed no significant effects. The mean numberof correctly identified game features was 5 out of 6.

Next, we tested the internal consistency of each of the twoindicators used to measure the variables of the theory ofplanned behavior. This analysis, as most of the analyses re-ported later, relied on within-subjects procedures. Pearsoncorrelations between the two measures, across the six games,were computed for each participant. These correlations weretransformed by means of Fisher's r-to-z transformation, aver-aged, and the averages were retransformed to correlation coeffi-cients. As expected, the average correlations were found to behigh and significant for attitude, perceived behavioral control,and intention (r = .82 to .92 in the pretest, r = .84 to .95 in theposttest); responses to the two items were therefore averaged toyield a measure of each variable. Subjective norms, however,seemed to be largely irrelevant in the context of this experi-ment. Most respondents rated the experimenter's expectationsidentically on all six games, making it impossible to computewithin-subjects correlations for these respondents. Among theremaining 34 participants, the two subjective norm items corre-lated, on the average, .55 in the pretest and .73 in the posttest.Because, as is shown later, subjective norms were also found tomake only minor contributions to the prediction of intentions,this construct was omitted from most analyses.

Finally, we examined the pre- and posttest means and vari-ances of the variables in the theory of planned behavior as afunction of experience and orientation. Means and variances ofattitude, perceived behavioral control, and intention over thesix games were computed separately for each of the 75 partici-

Page 6: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

STABILITY OF PREDICTORS 759

pants. The obtained values were then submitted to ANOVAs.The ANOVAs revealed only one significant effect, due to moti-vational orientation: Intentions measured on the pretest hadsignificantly greater variances under the fun than under theskill orientation, F(l, 70) = 6.60, p < .05. No other differenceswere significant on the pretest or on the posttest.

Accessibility and Temporal Stability

Accessibility. Because the distributions of response latencieswere found to be positively skewed, we followed the practice ofFazio and his associates (Fazio et al., 1982; Powell & Fazio,1984) and subjected the recorded latencies to reciprocal trans-formations. All data analyses were performed on the trans-formed scores, but for ease of discussion we report the retrans-formed latencies. We also considered the possibility that mea-sures of response latency can contain error variance due toindividual differences in speed of responding that are unrelatedto accessibility (see Fazio, 1990b). The filler items in the pre-and posttest questionnaires were included to control for thispossibility.

To obtain an overall measure of attitude accessibility, the(inverse) response latencies for the two indicators of attitudewere averaged across the six games. Similar latency measureswere computed for perceived behavioral control and for inten-tion. However, before averaging the latencies for the two indica-tors of a given variable, we tested for internal consistency. Herewe had to use a between-subjects analysis because only oneaveraged response latency measure was available for each re-spondent. On the pretest, the correlations between the two la-tencies (averaged across the six behaviors) were .55 for attitude,.65 for perceived control, and .65 for intention (p < .01 in eachcase). The corresponding correlations on the posttest were .64,.75, and .57 (all significant at p < .01).

A 2 X 2 X 2 ANOVA on the filler latencies, with type ofexperience and motivational orientation as between-subjectsfactors and time of measurement as a within-subjects factor,revealed no significant main effects or interactions. However,the latencies of the filler items were found to correlate signifi-cantly with the latencies of the measures concerning the threetheoretical constructs, with correlations ranging from .49 to .52(p < .01) on the pretest and from .32 to .37 (p < .05) on theposttest. The filler latencies were therefore statistically con-trolled in all subsequent analyses involving response latencymeasures.

Multivariate analyses of variance (MANOVAs), with fillerlatencies treated as a covariate, were performed to test the ef-fects of the experimental manipulations on the accessibility ofattitudes, perceptions of behavioral control, and intentions.The MANOVAs revealed three main effects and a significantinteraction between motivational orientation and time of mea-surement. Mean adjusted response latencies3 were lower follow-ing direct as opposed to indirect experience (M= 4.12 vs. 5.02s), F(3,68) = 3.01, p < .05; they were lower under fun (M = 4.15)than under skill (M= 5.01) orientation, F(3,68)= 4.35, p<. 01;and they were lower on the posttest (M = 3.67) than on thepretest (M = 5.54), F(3,69) = 17.05, p < .001. The Orientation XTime of Measurement interaction, although significant, F(3,68) = 4.67, p < .01, was of very small magnitude. As is shown

later, the effect of motivational orientation was somewhatsmaller on the pretest than on the posttest.

Univariate ANOVAs produced virtually the same pattern ofresults for attitudes, perceived behavioral control, and inten-tions: three main effects and significant Orientation X Time ofMeasurement interactions for attitudes and perceived control,but not for intentions. Separate analyses for pre- and posttestalso showed that the effects of the experimental manipulationswere generally comparable. In the multivariate analyses of thepretest data, the mean latencies (adjusted for the filler latencies)in the direct- and indirect-experience conditions were 4.97 ver-sus 6.13, F(3,68) = 5.79, p < .01. Univariate tests showed thisdifference to be significant for attitudes and perceived behav-ioral control (p < .01 in each case), but only marginally signifi-cant for intentions, F(l, 70) = 3.08, p < .09. On the posttest, theaverage adjusted latency was 3.40 for direct experience and 3.93for indirect experience. In the same direction as on the pretest,this difference was marginally significant in the multivariatetest, F(3, 68) = 2.56, p < .07. The univariate tests revealedsignificant differences for all three dependent variables, atti-tudes, perceived behavioral control, and intentions (p < .05 ineach case).

With respect to the effect of motivational orientation, sepa-rate analyses on pre- and posttest measures produced the fol-lowing results. On the pretest, the average adjusted latencyunder the fun orientation was 5.20, compared with an averagelatency of 5.90 under the skill orientation, F(3,68) = 5.79, p <.01, and the univariate tests for attitudes, perceived control, andintentions were significant (/? < .05) for perceived behavioralcontrol and intentions, and marginally significant for attitudes,F(l, 70) = 3.14, p < .09. The corresponding means on the post-test were 3.27 versus 4.07, F(3, 68) = 5.89, p < .001, and allthree univariate tests produced statistically significant results(p<.05).

To summarize briefly, the main effect due to type of experi-ence replicates results of earlier research, suggesting that directexperience increases accessibility of attitudes, perceived behav-ioral control, and intentions. The second main effect shows thataccessibility was generally greater in the fun than in the skillorientation, contradicting the hypothesis that accessibility ofperceived behavioral control is greater under skill than underfun instructions. Somewhat unexpectedly, the effects of type ofexperience and motivational orientation were found to persisteven after all participants had spent 45 min playing the videogames. Finally, the third main effect due to time of measure-ment shows, as might be expected, a practice effect, such thatresponse latencies declined on the second administration of thequestionnaire.

Temporal stability. The temporal stabilities of attitudes, per-ceptions of control, and intentions were assessed by computingthe within-subjects correlations between pre- and posttest mea-sures of these variables, across the six games.4 All statistical

3 Although the analyses were done on the inverse latencies, for clar-ity we are reporting the means of the reinverted latencies.

4 Temporal stability was also assessed by computing the sum of thesquared differences between pre- and posttest measures across games.These indices showed results very similar to the correlations.

Page 7: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

760 JORG DOLL AND ICEK AJZEN

analyses involving these stabilities were performed after thestability correlations had been submitted to Fisher's r-to-ztransformation. Reported later are the mean retransformedcorrelations. A 2 (type of experience) X 2 (motivational orienta-tion) MANOVA revealed significant main effects of experienceand motivational orientation, and no significant interactions.As expected, direct experience resulted in greater temporal sta-bility, that is, in higher correlations between pre- and posttestmeasures (mean r = .85 for direct experience vs. r = .66 forindirect experience), F(3,69) = 10.27, p < .001. Furthermore, asin the case of latencies, the fun orientation produced greatertemporal stability (mean r = .82) than did the skill orientation(mean r = .71), F(3, 69) = 3.44, p < .05.

The univariate tests produced the same general results, al-though the differences were not always significant. Attitudesand perceptions of control were significantly more stable underdirect experience (mean r = .86 and r = .84) than under indirectexperience (.65 and .55, p < .05), but the difference in the stabil-ity of intentions (.85 vs. .75) was only marginally significant(p < .09). Similarly, motivational orientation had significant ef-fects on the stability of attitudes and intentions (p < .05), suchthat measures of these variables were more stable in the funcondition (mean r = .85 and r = .86) than in the skill condition(.73 and .68), but the effect of this manipulation on the stabilityof perceived behavioral control was not significant (mean r =.74 vs. .73, F= 0.04).

In sum, the two experimental manipulations, type of experi-ence and motivational orientation, influenced the accessibilityof attitudes, perceptions of control, and intentions in the ex-pected direction, and they also influenced, as expected, thetemporal stabilities of these variables.

Prediction of Intentions and Behavior

We now turn to predictions of intentions and game behaviorin the context of the theory of planned behavior. Although notdirectly relevant to our hypotheses, we first performed an over-all test of the theory's ability to predict game behavior across allconditions of the experiment. Two average within-subjectscorrelation matrices, one for behavior and one for intentions,were submitted to multiple regression analyses. As required bythe theory, behavior was regressed on the pretest measures ofperceived behavioral control and intentions, and pretest inten-tions were regressed on pretest attitudes, subjective norms, andperceptions of behavioral control.5

Three behavioral measures were available: time played witheach game, order of play, and accumulated scores for eachgame. Only results using time played with each game as thecriterion are reported in this article. Order was largely unre-lated to attitudes because some respondents played their leastliked games first (saving the best for later), whereas othersstarted with their most liked games. Number of points accumu-lated did correlate with the theory's variables in cross-sectionalanalyses, but these scores were not comparable across games,making it difficult to perform within-subjects analyses. Timeplayed with each game turned out to be the best measure ofplaying behavior for our purposes.

The results of the multiple regression analyses provided goodsupport for the theory of planned behavior. In the prediction of

game behavior, the multiple correlation was .50 (p < .01), withintentions as well as perceptions of behavioral control makingsignificant contributions to the prediction. The standardizedregression coefficients for the two predictors were .30 and .22,respectively (p < .05 in each case). When intention was thecriterion, the multiple correlation was extremely high (R = .94,p < .001), and all three predictors entered significantly into theregression equation. The standardized regression coefficientswere .48 for attitude, .15 for subjective norm, and .41 for per-ceived behavioral control (p < .05 in each case).

Effects of Experience and Motivational OrientationWe now turn to an examination of the effects of our experi-

mental manipulations on accuracy of prediction. Consideringthe accessibility theorem, the findings reported earlier con-cerning the effects of the manipulations on response latencylead to the following expectations. Because attitudes, percep-tions of control, and intentions were more accessible followingdirect rather than indirect experience, and in the fun ratherthan skill orientation, prediction of intentions and behaviorshould be better under conditions of direct experience and funorientation. By comparison, the stability theorem also leads tothe expectation that accuracy of prediction in the theory ofplanned behavior will be affected by type of experience and bymotivational orientation because both influenced the temporalstabilities of attitudes, perceptions of control, and intentions.However, improved prediction is expected only with respect togame behavior, not with respect to intentions.

Prediction of game behavior. Within our theoretical frame-work, game behavior is predicted from intentions and per-ceived behavioral control. However, consistent with the focus ofmost past research, we also examined the relation between atti-tudes and game behavior. Thus, for each participant, we com-puted three correlations across the six games: game behaviorwith attitudes toward the games, with perceptions of behavioralcontrol, and with intentions. Following an r-to-z transforma-tion, these correlations served as the dependent variable inthree regression analyses in which type of experience and moti-vational orientation served as independent dummy variables,coded — 1 and +1. In the first analysis, the correlations wereregressed on type of experience, motivational orientation, andthe product of these two independent variables. In the secondregression analysis, stability of the predictor (attitude, per-ceived control, or intention) was entered as an additional inde-pendent variable.6 Finally, in the third analysis, stability was

5 It would have been possible to compute a multiple correlation foreach participant and to submit the multiple correlations to ANOVAs.However, multiple correlations are insensitive to the direction of rela-tions between independent and dependent variables. A number of par-ticipants were found to have negative correlations between one or moreof the predictor variables and the criterion. Such negative correlationswill reduce the average zero-order correlation, but they will increasethe multiple correlation.

6 Note that the measures of attitude, perceived behavior control, andintention enter into this analysis in two places: as predictors of behav-ior and in the stability index that serves as the mediating variable.However, there was no evidence for confounding. Pretest measures ofattitude, perception of behavioral control, and intention had correla-tions of .12, -.05, and .06 with the respective stabilities of these vari-ables. None of these correlations approached statistical significance.

Page 8: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

STABILITY OF PREDICTORS 761

replaced by latency as the additional independent variable, andlatency of responses to the filler items was also entered. Theresults in each analysis revealed significant main effects, and nosignificant interactions. Table 1 presents the mean correlationsby condition, and Table 2 the standardized regression coeffi-cients and / tests for the main effects of experience and motiva-tional orientation, with and without the stability and latencymediators.

Examining first the effect of type of experience, the resultsshowed that direct experience produced stronger correlationsthan did indirect experience. In the direct experience condi-tion, behavior correlated .59 with attitude, .51 with perceivedbehavioral control, and .61 with intention. This compares, re-spectively, with correlations of .37, .40, and .39 in the indirectexperience condition. As can be seen in Table 2, without theintroduction of mediating variables, the regression coefficientsfor these effects were significant for the attitude-behavior andthe intention-behavior correlations (p < .05). Similarly, predic-tion of game behavior was found to be better under the fun thanunder the skill orientation. In the fun condition, the meancorrelations between behavior and attitudes, perceived control,and intentions were .58, .57, and .60, respectively, whereas thecorresponding correlations in the skill condition were .38, .34,and .41. As shown in Table 2, the effect of motivational orienta-tion was significant in two of the three comparisons, that is, forthe prediction of behavior from attitude and from perceivedbehavioral control (p < .05); in the case of the intention-beha-vior correlation, the effect was of marginal significance (t =1.74, p<.09).

The remaining regression analyses were performed to testthe extent to which the observed effects of experience and orien-tation were mediated by response latency (i.e., accessibility) ofattitudes, perceptions of control, and intentions or by the tem-poral stabilities of these variables. The results in Table 2 showthat entering response latency into the regression equation didnot eliminate or even diminish the observed effects. In con-trast, when temporal stability was entered into the regressionequation, virtually all significant differences were reduced tononsignificance.7 The only exception occurred with respect tothe effect of motivational orientation on the correlations be-tween perceived behavioral control and behavior. In this case,the effect of motivational orientation remained significant (p <.01). In a further analysis not shown in Table 2, we included, inaddition to the manipulated variables, both mediators by firstentering response latency, followed on the next step by tem-poral stability. The results were essentially identical: The effectsof the experimental manipulations were eliminated only aftertemporal stability had been entered into the equation, againwith the exception of the effect of motivational orientation onthe control-behavior correlation.

Clearly, the findings displayed in Table 2 provide no supportfor the moderating role of response latency, but they are gener-ally consistent with the stability theorem. To gain a better un-derstanding of the differential results for response latencies andstabilities, it is instructive to examine the effects of these media-tors in the regression analyses. It can be seen in Table 2 thatwhereas the stabilities of the predictors had strong and signifi-cant positive effects on predictive accuracy, their latencies hadnegative effects, although significant only for the prediction of

behavior from attitude. This could account for the finding inTable 2 that entering response latency into the regression equa-tion tended to accentuate the effects of the experimental manip-ulations on the prediction of behavior.

It is important to exercise caution in interpreting these re-sults because it is difficult to ascertain the reliability of ourlatency measures. Lack of reliability may produce misleadingconclusions in that the regression analysis will tend to overesti-mate the impact of the independent variables and underesti-mate the role of the accessibility mediator (Baron & Kenny,1986). It has been suggested, however, that response latenciesmay accurately reflect differences in accessibility at the grouplevel even when they are poor indicators of attitude accessibil-ity at the individual level (Fazio, 1990b). To test for this possibil-ity, we followed past practice (e.g., Fazio & Williams, 1986) andsplit the sample at the median into low- and high-latency sub-groups. The results showed no significant differences betweenthe two groups in terms of attitude-behavior correlations(mean r = .43 in the low-latency group vs. mean r = .54 in thehigh-latency group, r(72) = -1.11), in terms of control-be-havior correlations (mean r = .47 vs. .48, t(72) = —.12), or interms of intention-behavior correlations (mean r = .54 vs. .43,?(72)=1.06).

Prediction of intentions. We turn now to the effects of theexperimental manipulations on the prediction of intentionsfrom attitudes and from perceived behavioral control. Giventhe significant effects of the manipulations on response laten-cies, the accessibility theorem would suggest that attitudes andperceptions of control should predict intentions better underdirect than under indirect experience and better under funthan under skill instructions. The stability theorem, however,would predict no significant differences in correlations due totype of experience or motivational orientation. Attitude-inten-tion and control-intention correlations were regressed on typeof experience, motivational orientation, and the product ofthese two variables. Consistent with the stability theorem, theresults revealed no significant main effects or interactions. Themean within-subjects correlations were uniformly high, rang-ing from .82 to .95, with no evidence for systematic differencesbetween experimental conditions.8

To summarize briefly, type of experience and motivationalorientation influenced prediction of behavior, and these effectswere found to be mediated by the temporal stability of thepredictor but not by its latency. Moreover, and in further sup-port of the stability theorem, the experimental manipulationshad little effect on the prediction of intentions.

7 When controlling for the mediator reduces to nonsignificance theeffect of the independent variable on the dependent variable, this istaken as evidence for the hypothesized mediating process (see Baron &Kenny, 1986). A more stringent test would require demonstration thatintroduction of the mediating variable significantly reduces the effectof the independent variable, that is, that the effect of the independentvariable on the dependent variable is smaller after the mediator hasbeen entered than without it. Unfortunately, no procedure appropriatefor testing this difference appears to be available.

8 In defense of the accessibility theorem it could be argued that thesehigh correlations made it very difficult to observe any effect due totype of experience or motivational orientation.

Page 9: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

762 JORG DOLL AND ICEK AJZEN

Table 1Mean Within-Subject Correlations by Experimental Condition

Correlation

Attitude-behaviorControl-behaviorIntention-behavior

Direct experience

Funorientation

.67

.61

.68

Skillorientation

.49

.39

.53

Indirect experience

Funorientation

.47

.51

.50

Skillorientation

.27

.29

.26

Conclusions

Past research has revealed a number of situational and per-sonal factors that tend to moderate the attitude-behavior rela-tion (see Ajzen, 1988, for a review). As in previous research, thepresent study found that direct experience produces strongerattitude-behavior correlations than does indirect experience.In addition, we saw that direct experience also increases thepredictive validity of intentions and of perceived behavioralcontrol. In a similar fashion, the present study demonstratedthat motivational orientation can moderate the relations be-tween verbal responses and observed actions. Attitudes, per-ceptions of control, and intentions were found to predict play-ing of video games better under a fun than under a skill orienta-tion.

Two mediating processes were examined as potential expla-nations for the observed effects: accessibility and temporal sta-bility of attitudes, perceptions of control, and intentions. Asexpected, the experimental manipulations were found to influ-ence accessibility and stability in much the same way as theyinfluenced the prediction of behavior from attitudes, perceivedbehavioral control, and intentions. That is, accessibility andstability of attitudes, perceived behavioral control, and inten-tions were generally higher under direct-experience than under

indirect-experience conditions and higher under fun thanunder skill orientations.

In the past, parallel findings concerning the effects of suchvariables as type of experience on response latency and on ob-served attitude-behavior relations have been taken as evidencein support of the mediating role of accessibility (e.g., Fazio et al.,1982). A moment's reflection reveals, however, that direct expe-rience, or any other variable such as the number of attitudeexpressions or motivational orientation, can influence re-sponse latency as well as attitude-behavior correlations, even ifaccessibility is not the important mediating process. The pres-ent study was based on the idea that accessibility is only onepossible mediator and that temporal stability may serve as an-other.

As suggested by Baron and Kenny (1986), this idea wastested by controlling statistically for response latency and fortemporal stability of attitudes, perceptions of behavioral con-trol, and intentions. These analyses were supportive of the sta-bility theorem, but unexpectedly, they revealed no mediatingrole due to response latency. When response latency was en-tered into the regression equation, type of experience and moti-vational orientation retained their significant effects on ob-served relations between attitudes, perceived behavioral con-

Table 2Regression Coefficients and t Tests for the Effects of Type of Experience and MotivationalOrientation on the Prediction of Game Behavior From Attitude,Perceived Behavioral Control, and Intention

Correlation

Attitude-behaviorType of experienceMotivational orientation

MediatorControl-behavior

Type of experienceMotivational orientation

MediatorIntention-behavior

Type of experienceMotivational orientation

Mediator

b

.24

.22

.15

.29

.23

.20—

Withoutmediator

t

2.12*1.98*

1.302.58**

2.02*1.74

Latency as mediator

b

.31

.28- .35

.18

.27- .17

.25

.25- .13

t

2.81**2.46*

-2.75**

1.522.32*

-1.39

2.17*2.06*- .95

Stability asmediator

b

.09

.08

.42

- .03.29.37

.18

.11

.31

t

.87

.733 .53"

- .332.70**3.05**

1.681.012.65**

Note. N= 75. b = standardized regression coefficient.*p<.05. **p<.0l.

Page 10: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

STABILITY OF PREDICTORS 763

trol, and intentions on one hand and behavior on the other. Incontrast, the observed relations were greatly attenuated whentemporal stability was entered, generally reducing the impactof the experimental manipulations to nonsignificance. Addi-tional evidence supporting the stability theorem was providedby the finding that prediction of intentions was unaffected bytype of experience and by motivational orientation.

The results also showed that instructions to focus on funraised the accessibility and stability of attitudes, perceptions ofbehavioral control, and intentions. Contrary to expectations,however, there was no evidence that instructions to focus onskill would raise the accessibility or stability of perceived behav-ioral control. It may be argued that the fun orientation is gener-ally consistent with the "natural" motivational orientation forplaying video games, whereas the skill instructions invoked aninappropriate achievement-related orientation. The conflict inthe skill condition between the inherent fun orientation andthe induced achievement orientation may be responsible forlowering the accessibilities and stabilities of the measured vari-ables (see also Millar & Tesser, 1986, 1989; Shavitt & Fazio,1991, for the effects of incompatible orientations on attitude-behavior correspondence).

The unexpected finding that response latency did not medi-ate the effects of direct versus indirect experience, nor the ef-fects of fun versus skill orientation, are inconsistent with theidea that an attitude's accessibility in memory determines itsability to guide behavior. Of course, no single study is sufficientto disprove the hypothesized mediating effect of accessibility,and future research may well reaffirm the importance of acces-sibility as a determinant of attitude-behavior correspondence.Nevertheless, our findings raise questions about the conclu-sions that have been reached on the basis of past research. Re-call that the mediating role of accessibility has not been directlytested. The second experiment reported by Houston and Fazio(1989) is perhaps most directly comparable with our research.In that experiment, the investigators manipulated frequency ofattitudinal expression, measured response latency, and thenshowed that attitude accessibility as well as attitude-behaviorcorrespondence were greater for subjects in the high-frequency-of-expression condition. Although consistent with the idea thataccessibility may have mediated the effect of multiple attitudeexpressions on the attitude-behavior relation, it should be clearthat the mediation hypothesis was not subjected to a direct test.

Fazio and Williams (1986), Fazio et al. (1989), and Houstonand Fazio (1989), Experiment 1, measured latency of attitudinalresponses and showed that attitudes with lower latencies arebetter predictors of judgments and overt behaviors. At firstglance, these findings appear to differ from the results of ourstudy in which splitting respondents into low- and high-latencysubgroups had no significant effect on the predictive validity ofthe attitudinal variables. Note, however, that the studies byFazio and his associates assessed previously formed attitudestoward issues with which respondents were well familiar: presi-dential candidates, commonly available commercial products,and the policy of capital punishment. Response latencies withrespect to previously formed attitudes may well be confoundedwith such uncontrolled variables as amount of information ortype of experience on which the attitude is based, involvementwith the attitude object, and so forth. Factors of this kind would

tend to influence attitude stability, which could account for theobserved relation between response latency and the attitude'spredictive validity. In contrast, the present study assessed atti-tudes toward video games to which respondents were exposedfor the first time. Aside from baseline speed of responding,which was statistically controlled, response latencies in thissituation should be primarily a function of the manipulatedvariables and should not be confounded with uncontrolled fac-tors. Past studies involving different types of intellectual puz-zles (e.g, Regan & Fazio, 1977) have also assessed newly formedattitudes. However, to the best of our knowledge, those studiesdid not measure response latency, and it is thus not clearwhether they would have shown the mediating effect of re-sponse latency with respect to attitude-behavior correspon-dence.

A more general question can perhaps be raised with respectto the construct validity of the latency measures in our study.Fazio (1990b) has alerted investigators to the difficulties of se-curing response latency measures that are valid indicators of anattitude's accessibility in memory. However, the finding thatthe experimental manipulations of the present study affectedresponse latencies as expected argues against the propositionthat our latency measures lacked validity. Instead, it suggeststhat at least at the group level, our response latency measureswere reflective of attitude strength or accessibility. Neverthe-less, comparison of low and high latency subgroups revealed nosignificant differences in attitude-behavior correlations.

Putting aside for a moment the finding that measures of atti-tude accessibility failed to mediate the effects of our experimen-tal manipulations, the attempt to explain the predictive validityof attitudes in terms of their accessibility in memory seems tous unnecessarily complex. We agree that such factors as directexperience, amount of information, or repeated attitude expres-sion can strengthen a person's attitude. Furthermore, we wouldalso agree that in comparison with relatively weak attitudes,strong attitudes are more easily accessible in memory; that is,they can be recalled faster from memory than can weak atti-tudes. However, this does not necessarily mean that ease ofaccessibility is responsible for the greater predictive validity ofstrong attitudes. To explain why a relatively accessible attitude isa better predictor of behavior, Fazio (1990a) proposed a modelthat involves a variety of psychological processes. Specifically,the model stipulates that to influence behavior in a given situa-tion, an attitude must first be activated; that is, it must becomesalient in the behavioral situation. Once activated, the attitudeis assumed to influence the way that the object and situation areperceived or interpreted. This selective or biased perceptionthen predisposes favorable or unfavorable behavior toward theobject of the attitude. Finally, for attitude accessibility to influ-ence its predictive validity, the model must also assume that theattitude is activated automatically on mere observation of theattitude object and that only strong attitudes are likely to beautomatically activated. This last assumption is necessary be-cause if attitudes were activated following conscious delibera-tion rather than automatically, then even weak attitudes wouldbe activated, would bias perception, and could guide behaviorjust as well as strong attitudes.

Compare this elaborate account with the explanation pro-vided by the stability theorem. The starting point is the same:Type of experience, amount of information, repeated attitude

Page 11: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

764 JORG DOLL AND ICEK AJZEN

expressions, motivational orientation, and so on can all influ-ence the strength of attitudinal variables. The only other as-sumption we have to make is that strong attitudes are morestable, that is, are less likely to change, than weak attitudes. Theeffect of attitude strength on attitude-behavior correlations fol-lows logically from this assumption. Clearly, if people's atti-tudes change after they have been assessed, the measured atti-tudes are less likely to permit accurate prediction of subsequentbehavior. No assumptions need to be made about automaticactivation of attitudes or about their biasing impact on the per-ception of attitude object and situation. It can be seen that thestability theorem is clearly more parsimonious than the accessi-bility model.

It is interesting to note in this context that the results of arecent series of studies by Bargh, Chaiken, Govender, andPratto (1992) call into question one of the fundamental as-sumptions of the accessibility model, namely, the assumptionthat only strong attitudes are activated automatically on mereobservation of the attitude object. These investigators foundthat weak attitudes can also be automatically activated. Further-more, there are indications that the notion of accessibility inFazio's model may not be sufficiently differentiated. In theirsocio-cognitive model, Pratkanis and Greenwald (1989) distin-guished between three manners of attitude activation: retrievalof the evaluative summary judgment (akin to Fazio's view ofattitude activation), retrieval of the underlying beliefs or cogni-tions, or joint retrieval of evaluative summary and cognitivestructure. It is conceivable that activation by retrieval of theunderlying cognitive knowledge structure involves greater elabo-ration and that, in this case, longer response latencies may indi-cate stronger rather than weaker attitudes because the attitudeis based on a richer and better articulated set of beliefs. In anyevent, it seems to us that attitude accessibility, as denned inFazio's model, cannot easily be assumed to represent all thedifferent aspects of an attitude's strength.

The view that temporal stability mediates the effect of directexperience on attitude-behavior correspondence may actuallynot be inconsistent with Fazio's (1986) model. An attitude, onceactivated, is assumed to bias perception of the situation andtherefore guide behavior in an attitude-consistent direction.However, once the behavior is initiated in accordance with theattitude, its persistence or continuation is predicated on theassumption that the perception of the situation does notchange. If new information becomes available that changes per-ception of the situation or, in our terms, the new informationchanges beliefs about the behavior, then the behavior is alsolikely to change in the direction of the new perceptions.

In short, temporal stability of attitudes seems to play an im-portant role in determining the strength of observed attitude-behavior relations. Specifically, our findings are consistent withthe idea that temporal stabilities of attitudes, perceptions ofcontrol, and intentions mediate the predictive validity of thesevariables. It thus seems reasonable to suggest that, along withaccessibility, more attention should be paid to temporal stabil-ity as an important determinant of an attitude's predictivevalidity.

ReferencesAjzen, I. (1985). From intentions to actions: A theory of planned behav-

ior. In J. Kuhl & J. Beckmann (Eds.), Action-control: From cognitionto behavior (pp. 11-39). New York: Springer.

Ajzen, I. (1988). Attitudes, personality, and behavior. Homewood, IL:Dorsey Press.

Ajzen, I. (1991). The theory of planned behavior. Organizational Behav-ior and Human Decision Processes, 50, 179-211.

Ajzen, I., & Fishbein, M. (1980). Understanding attitudes and predictingsocial behavior. Englewood Cliffs, NJ: Prentice-Hall.

Ajzen, I., & Madden, T. J. (1986). Prediction of goal-directed behavior:Attitudes, intentions, and perceived behavioral control. Journal ofExperimental Social Psychology, 22, 453-474.

Bagozzi, R. P., & Warshaw, P. R. (1990). Trying to consume. Journal ofConsumer Research, 17, 127-140.

Bagozzi, R. P., & Warshaw, R. R. (in press). An examination of theetiology of the attitude-behavior relation for goal-directed and mind-less behaviors. Multivariate Behavioral Research.

Bargh, J. A., Chaiken, S., Govender, R., & Pratto, F. (1992). The general-ity of the automatic attitude activation effect. Journal of Personalityand Social Psychology, 62, 893-912.

Baron, R. M., & Kenny, D. A. (1986). The moderator-mediator vari-able distinction in social psychological research: Conceptual, strate-gic, and statistical considerations. Journal of Personality and SocialPsychology, 57,1173-1182.

Berger, I. E., & Mitchell, A. A. (1989). The effect of advertising onattitude accessibility, attitude confidence, and the attitude-behaviorrelationship. Journal of Consumer Research, 16, 269-279.

Doll, J., & Mallii, R. (1990). Individuierte Einstellungsformation, Ein-stellungsstruktur und Einstellungs-Verhaltens-Konsistenz [Individu-alized attitude formation, attitude structure, and attitude-behaviorconsistency]. Zeitschrift fur Sozialpsychologie, 21, 2-14.

Fazio, R. H. (1986). How do attitudes guide behavior? In R. M. Sorren-tino & E. T. Higgins (Eds.), The handbook of motivation and cogni-tion: Foundations of social behavior (pp. 204-243). New York: Guil-ford Press.

Fazio, R. H. (1990a). Multiple processes by which attitudes guide be-havior: The mode model as an integrative framework. In M. P.Zanna (Ed.), Advances in experimental social psychology (Vol. 23,pp. 75-109). San Diego, CA: Academic Press.

Fazio, R. H. (1990b). A practical guide to the use of response latency inpsychological research. In C. Hendrick & M. S. Clark (Eds.), Reviewof personality and social psychology, Vol. 11: Research methods inpersonality and social psychology (pp. 74-97). Newbury Park, CA:Sage.

Fazio, R. H., Chen, J., McDonel, E. C, & Sherman, S. J. (1982). Attitudeaccessibility, attitude-behavior consistency, and the strength of theobject-evaluation association. Journal of Experimental Social Psy-chology, 18, 339-357.

Fazio, R. H., Powell, M. C, & Williams, C. J. (1989). The role of atti-tude accessibility in the attitude-to-behavior process. Journal ofConsumer Research, 16, 280-288.

Fazio, R. H., & Williams, C. J. (1986). Attitude accessibility as a moder-ator of the attitude-perception and attitude-behavior relations: Aninvestigation of the 1984 presidential election. Journal of Personalityand Social Psychology, 51, 505-514.

Fazio, R. H., & Zanna, M. (1978a). Attitudinal qualities relating to thestrength of the attitude-behavior relationship. Journal of Experimen-tal Social Psychology, 14, 398-408.

Fazio, R. H., & Zanna, M. (1978b). On the predictive validity of atti-tudes: The roles of direct experience and confidence. Journal of Per-sonality, 46, 228-243.

Fazio, R. H., & Zanna, M. P. (1981). Direct experience and attitude-be-havior consistency. In L. Berkowitz (Ed.), Advances in experimentalsocial psychology (Vol. 14, pp. 161-202). San Diego, CA: AcademicPress.

Fishbein, M., & Ajzen, I. (1975). Belief, attitude, intention, andbehavior:An introduction to theory and research. Reading, MA: Addison-Wes-ley.

Page 12: Accessibility and Stability of Predictors in the Theory of Planned Behavior.

STABILITY OF PREDICTORS 765

Houston, D. A., & Fazio, R. H. (1989). Biased processing as a functionof attitude accessibility: Making objective judgments subjectively.Social Cognition, 7, 51-66.

Krosnick, J. A., Boninger, D. S., Chuang, Y. C, & Carnot, C. G. (1991).Attitude strength: One construct or many related constructs? Unpub-lished manuscript, Department of Psychology, Ohio State Univer-sity, Columbus.

Millar, M. G., & Tesser, A. (1986). Effects of affective and cognitivefocus on the attitude-behavior relation. Journal of Personality andSocial Psychology, 51, 270-276.

Millar, M. G, & Tesser, A. (1989). The effects of affective-cognitiveconsistency and thought on the attitude-behavior relation. Journalof Experimental Social Psychology, 25,189-202.

Miller, G. A. (1965). The magical number seven plus or minus two:Some limits on our capacity for processing information. Psychologi-cal Review, 63, 81-97.

Powell, M. C, & Fazio, R. H. (1984). Attitude accessibility as a functionof repeated attitudinal expression. Personality and Social PsychologyBulletin, 10,139-148.

Pratkanis, A. R., & Greenwald, A. G. (1989). A socio-cognitive modelof attitude structure and function. In L. Berkowitz (Ed.), Advances in

experimental social psychology (Vol. 22, pp. 245-285). San Diego,CA: Academic Press.

Regan, D. T, & Fazio, R. H. (1977). On the consistency between atti-tudes and behavior: Look to the method of attitude formation. Jour-nal of Experimental Social Psychology, 13, 38-45.

Schifter, D. B., & Ajzen, I. (1985). Intention, perceived control, andweight loss: An application of the theory of planned behavior. Jour-nal of Personality and Social Psychology, 49, 843-851.

Schwartz, S. H. (1978). Temporal instability as a moderator of theattitude-behavior relationship. Journal of Personality and Social Psy-chology, 36, 715-724.

Shavitt, S., & Fazio, R. H. (1991). Effects of attribute salience on theconsistency between attitudes and behavior predictions. Personalityand Social Psychology Bulletin, 17, 507-516.

Wood, W. (1982). Retrieval of attitude-relevant information from mem-ory: Effects on susceptibility to persuasion and on intrinsic motiva-tion. Journal of Personality and Social Psychology, 42, 798-810.

Received November 29,1990Revision received December 3,1991

Accepted April 24,1992 •

AMERICAN PSYCHOLOGICAL ASSOCIATION

SUBSCRIPTION CLAIMS INFORMATION Today's Date:_

We provide this form to assist members, institutions, and nonmember individuals with any subscription problems. With theappropriate information we can begin a resolution. If you use the services of an agent, please do NOT duplicate claims throughthem and directly to us. PLEASE PRINT CLEARLY AND IN INK IF POSSIBLE.

PRINT FULL NAME OR KEY NAME OF INSTITUTION

ADDRESS

MEMBER OR CUSTOMER NUMBER (MAYBE FOUND ON ANYPASTISSUB LABEL)

DATE YOUR ORDER WAS MAILED (OR PHONED):

P.O. NUMBER:

PREPAID CHECK ___CHARGECITY STATE/COUNTRY ZIP CHECK/CARD CLEARED DATE:_

YOUR NAME AND PHONE NUMBER

TITLE

(If possible, send * copy, front and back, of jour cancelled check to help us In ourresearcb of your claim.)

ISSUES: ___MISSING DAMAGED

VOLUME OR YEAR NUMBER OR MONTH

Thank you. Once a claim is received and resolved, delivery of replacement issues routinely lakes 4-6 weeks.

DATE RECEIVED:ACTION TAKEN:STAFF NAME:

DATE OF ACTION:INV.NO.&DATE:LABEL NO. & DATE:

SEND THIS FORM TO: APA Subscription Claims, 750 First Street, N.E., Washington, DC 20002

PLEASE DO NOT REMOVE. A PHOTOCOPY MAY BE USED.