Ajzen - Knowledge and the Prediction of Behavior

Seediscussions,stats,andauthorprofilesforthispublicationat:https://www.researchgate.net/publication/232941314

KnowledgeandthePredictionofBehavior:TheRoleofInformationAccuracyintheTheoryofPlannedBehaviorARTICLEinBASICANDAPPLIEDSOCIALPSYCHOLOGYAPRIL2011ImpactFactor:1.09DOI:10.1080/01973533.2011.568834

CITATIONS58

READS769

4AUTHORS,INCLUDING:

IcekAjzenUniversityofMassachusettsAmherst113PUBLICATIONS51,080CITATIONS

SEEPROFILE

NicholasMJoyceUniversityofMaryland,CollegePark5PUBLICATIONS374CITATIONS

SEEPROFILE

Allin-textreferencesunderlinedinbluearelinkedtopublicationsonResearchGate,lettingyouaccessandreadthemimmediately.

Availablefrom:IcekAjzenRetrievedon:14January2016

Knowledge and the Prediction of Behavior: The Role ofInformation Accuracy in the Theory of Planned Behavior

Icek Ajzen, Nicholas Joyce, Sana Sheikh, and Nicole Gilbert Cote

University of MassachusettsAmherst

The results of the present research question the common assumption that being wellinformed is a prerequisite for effective action to produce desired outcomes. In Study1 (N 79), environmental knowledge had no effect on energy conservation, and inStudy 2 (N 79), alcohol knowledge was unrelated to drinking behavior. Such disap-pointing correlations may result from an inappropriate focus on accuracy of infor-mation at the expense of its relevance to and support for the behavior. Study 3(N 85) obtained a positive correlation between knowledge and pro-Muslim behavior,but Study 4 (N 89) conrmed the proposition that this correlation arose becauseresponses on the knowledge test reected underlying attitudes. Study 4 also showed thatthe correlation could become positive or negative by appropriate selection of questionsfor the knowledge test. The theory of planned behavior (Ajzen, 1991), with its focus onspecic actions, predicted intentions and behavior in all four studies.

Only when well informed can we act effectively to pro-duce desired outcomes. This article of faith is frequentlyinvoked to explain a wide range of detrimental lifestylebehaviors. Uninformed or misinformed, people eat anunhealthy diet and dont exercise enough, engage inunsafe sex, abuse drugs and alcohol, fail to protectthemselves from the harmful rays of the sun, and pollutethe environment. Beyond detrimental lifestyles, whenpoorly informed, people fearfully avoid members of out-groups, investors succumb to irrational exuberance,and a nations leader embarks on misguided policies.

A well-informed citizenry is the essential backbone ofa free society, and few would dispute the value of moreand better information. Yet the possession of accurateinformation is no guarantor of wise judgments, nor ismisinformation necessarily a precursor of bad decisions.It is an empirical question as to whether accurate infor-mation encourages people to act in their own orsocietys best interests, and whether lack of informationhas detrimental implications for effective action, that is,action that produces desired outcomes. This article

examines the effects of accurate information or knowl-edge on peoples attitudes and decisions.

One indication that there may be no simple relationbetween level of accurate information and behaviorcomes from research on the prevention of AIDS. Surveysindicate that in many at-risk populations, knowledgeabout AIDS, its causes, and transmission routes is quitehigh while adoption of safer sex practices in these popula-tions remains low (e.g., Calsyn, Saxon, Freeman, &Whittaker, 1992; DiClemente, Forrest, & Mickler,1990; see Helweg-Larsen & Collins, 1997). In their reviewof this literature, Helweg-Larsen and Collins furtherobserved that the many research efforts aimed speci-cally at examining the relation between knowledge aboutAIDS . . . and preventive behaviors suggests overwhelm-ingly that this relation is weak or nonexistent (p. 23).Consistent with this conclusion, a meta-analysis of sevendata sets dealing with general AIDS knowledge andintentions to use condoms (Sheeran & Taylor, 1999)found a relatively low mean weighted correlation of .21,and general AIDS knowledge has not been found toaccount for much variance in actual condom use beha-vior (e.g., Ananth & Koopman, 2003).

Dissociations between knowledge and behavior havebeen documented in other domains as well. Forinstance, weak or nonsignicant ndings have been

Correspondence should be sent to Icek Ajzen, Department of

Psychology, University of Massachusetts, Tobin Hall-135 Hicks

Way, Amherst, MA 01003-9271. E-mail: [email protected]

BASIC AND APPLIED SOCIAL PSYCHOLOGY, 33:101117, 2011

Copyright # Taylor & Francis Group, LLCISSN: 0197-3533 print=1532-4834 online

DOI: 10.1080/01973533.2011.568834

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

reported with respect to the relation between knowledgeabout colorectal cancer screening and actually getting ascreening (Guerra, Dominguez, & Shea, 2005), knowl-edge about breast cancer and performing breastself-examinations (Schlueter, 1982), knowledge aboutorgan and tissue donation and signing a donor card orregistry (Feeley & Servoss, 2005), and knowledge aboutdiabetes and compliance with a diabetes control regimen(Spirito, Ruggiero, Duckworth, & Low, 1993). Simi-larly, in two studies (Ievers-Landis et al., 2003; SilverWallace, 2002), knowledge about osteoporosis amongyoung women was found to be largely unrelated to exer-cising and calcium intake, two recommended preventivebehaviors. According to Silver Wallace (2002), knowl-edge has been consistently shown to be noninuential inpredicting behavior (p. 170).

Of interest, these negative ndings have made barelya dent in the conviction that accurate information is acritical determinant of self-protective behavior. Thus,while reporting that HIV=AIDS knowledge was notsignicantly related to use of condoms (p. 538),Ananth and Koopman (2003) recommended that inter-vention efforts may benet from dispelling misconcep-tions about AIDS (p. 529). Indeed, many healthprofessionals continue to believe that ignorance is theprimary reason for the spread of HIV=AIDS (seeHelweg-Larsen & Collins, 1997) and for other health-related problems.

In critical analyses of this literature and in theirtheoretical formulations, investigators typically con-clude that knowledge, although necessary, is not suf-cient to produce the desired behavior (see, e.g.,DiClemente, 1989; Fisher & Fisher, 1992). It is usuallyargued that, in addition to having the required knowl-edge, people must also be motivated to perform thebehavior in question. In Fisher and Fishers (Fisher &Fisher, 1992; Fisher, Fisher, Williams, & Malloy,1994) information-motivation-behavioral skills model,for example, knowledge and motivation jointly inuencebehavior, either directly or indirectly via their effects ofbehavioral skills. However, research with this model hasshown that knowledge does not consistently inuencebehavior, and when it does, its effects on behavior tendto be relatively small and mediated by behavioral skills(e.g., Fisher et al., 1994; Misovich, Martinez, Fisher,Bryan, & Catapano, 2003).

KNOWLEDGE, INFORMATION, AND BELIEFS

In this article we take issue with the proposition thatknowledge is a prerequisite for effective action. Indeed,relying on work with the theory of planned behavior(TPB; Ajzen, 1991), we propose that knowledge isneither sufcient nor necessary and try to show why

knowledge, as typically conceptualized and assessed,fails to predict behavior. Knowledge tests consist of aseries of factual assertions and participants are asked,for each assertion, whether they believe that it is trueor false. Degree of knowledge is ascertained by countingthe number of responses considered to be correct bysome objective standard.

It is important to distinguish between knowledge, asjust dened, and amount of information. For example,to assess amount of information, Wood (1982; Kallgren& Wood, 1986; Wood & Kallgren, 1988) asked parti-cipants to indicate how well informed they were withrespect to a given topic or counted the number of beliefstatements participants could list in a short period. Inthe current research we were not interested in theamount of information people possess. Instead, weexamined the role of knowledge, that is, the extent towhich the information people have is accurate.

In the TPB, beliefs constitute the informational foun-dation that ultimately determines behavior. Note, how-ever, that the theory deals neither with amount ofinformation (i.e., with the number of beliefs peoplehold) nor with the accuracy of that information. Unlikeknowledge, the beliefs in the TPB may be incorrect, theymay reect wishful thinking or be biased in other ways,and they may be unrepresentative of the informationthat is considered important in a given behavioraldomain. Nevertheless, these beliefs are assumed to guideintentions and behavior. Specically, beliefs about abehaviors likely consequences (behavioral beliefs) areassumed to determine attitudes toward the behavior,beliefs about the expectations and behaviors of others(normative beliefs) are assumed to determine subjectivenorms, and beliefs about potential facilitating or inhibit-ing factors (control beliefs) are assumed to determineperceived behavioral control. Attitudes, subjective norms,and perceptions of control in turn combine to produceintentions which, together with actual control, deter-mine performance of the behavior. (For a discussionof the theory, see Ajzen, 2005.)

(IN)VALIDITY OF KNOWLEDGE TESTS

We now consider why scores on common knowledgetests often fail to predict behavior. First, and foremost,such tests may actually fail to assess knowledge. Facedwith questions about cancer, AIDS, the environment,energy conservation, or the health care systemissuesabout which they may have only limited informationindividuals may simply guess at the correct response.Such guesses, however, are unlikely to be random. Asearly as 1946, Newcomb demonstrated that whenrespondents are uncertain, their responses are consistentwith their own attitudes toward the issue in question. In

102 AJZEN ET AL.

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

other words, rather than assessing knowledge, responsesto items on a knowledge test may often reect the part-icipants attitude. In fact, Hammonds (1948; see alsoBlumenfeld, 1966; Weschler, 1950) error-choice method,a disguised attitude measurement technique, relies onthe assumption that answers to factual questions arebiased in such a way as to reveal the respondents atti-tude (see Kidder & Campbell, 1970).

If responses to the factual questions on a knowledgetest are indeed biased to be consistent with the parti-cipants attitudes, then the obtained knowledge scoreswill correlate positively with attitudes when correctresponses have positive implications for the issue underconsideration. However, the same bias should producea negative correlation with attitudes when correctresponses have largely negative implications. When theknowledge questionnaire is balanced such that equalnumbers of correct responses have positive and negativeimplications, there should be little correlation betweenattitudes and knowledge. It follows that if there is acorrelation between attitudes and behavior, we wouldexpect a positive correlation between knowledge andbehavior when correct responses reect a favorable atti-tude, a negative correlation between knowledge andbehavior when they imply an unfavorable attitude, andno signicant correlation when an equal number of cor-rect responses have positive and negative implications.We report a direct test of this idea in the fourth and nalstudy reported here in a later section.

A second problem related to knowledge tests is thatthe factual items they contain generally deal with broadconcepts such as breast cancer, HIV=AIDS, or theenvironment, not with the specic behavior under con-sideration. Empirical research has shown that broadattitudes or other general dispositions are usually poorpredictors of specic actions (Ajzen, 2005; Ajzen &Fishbein, 1977). The same is likely to be true for generalknowledge. Thus, a measure of general knowledge ina given domain is unlikely to predict any particularbehavior in that domain. Dispositional measures thatare compatible with the behavioral criterion in terms ofthe action involved, the target at which the action isdirected, and in terms of the context and time of beha-vioral performance tend to predict behavior much betterthan global dispositions (see Ajzen, 2005; Fishbein &Ajzen, 1974; Kraus, 1995).

A third issue related to use of knowledge tests to pre-dict behavior has to do with the content of items on suchtests. Judging factual items on knowledge tests to beeither true or false often has no clear implications forbehavior. Consider knowledge in relation to breast can-cer. One of the items on the knowledge inventory usedby Misovich et al. (2003) is the following: It is normalfor some womens breasts to feel lumpy or uneven. Thebehavioral criterion in this study was performance of

breast self-examinations. It is not at all clear whetheragreement with the statement that it is normal for somewomens breasts to feel lumpy or uneven would encour-age or discourage breast self-examinations.

Whats more, agreements or disagreements withassertions on knowledge tests are not scored in termsof their support or lack of support for the behavior ofinterest. Instead, they are scored for their accuracy inrelation to an objective criterion. A response scored ascorrect will therefore not necessarily indicate a cognitionin support of the desired behavior. Consider, forexample, the relation between knowledge about heroinand other illicit drugs and actual drug use. Many inves-tigators would assume that heroin use should declinewith accurate knowledge about this drug and its effects.Now, imagine that a heroin knowledge test includes thefollowing assertions: The street price of heroin hasdeclined in the past few years and Heroin use resultsin hair loss. Agreement with the rst assertion wouldseem to reect a belief that encourages heroin use eventhough, being factually correct, it would be scored asrevealing accurate knowledge about the topic. Con-versely, agreement with the second assertion, eventhough incorrect, would be indicative of a belief that dis-courages heroin use. Clearly, then, there is no reason toexpect a simple, direct relation between amount of her-oin knowledge and avoidance of heroin.

The aforementioned considerations lead to the con-clusion that we can expect a positive correlation betweenknowledge and behavior only if correct responses on theknowledge test consistently imply support for perform-ance of the behavior and incorrect responses consistentlyimply lack of support. The present article reports theresults of four studies that examined the relation betweenknowledge and behavior in three domains: knowledgeabout the environment and energy conservation beha-vior, knowledge about drinking alcohol and drinkingbehavior, and knowledge about Islam and Muslims andbehavior directed at Islam and Muslims.

STUDY 1: ENERGY CONSERVATION

The rst study assessed knowledge about the environ-ment to predict a category of behaviorenergy conser-vation. The discussion in the introduction suggests,however, that we cannot expect a strong associationbetween knowledge about the environment in generaland the specic behavioral category of energy conser-vation. Much better prediction of intentions to conserveenergy, and actual energy conservation, should be pro-vided by measures of the behavior-specic constructsin the theory of planned behavior (i.e., attitudes, subjec-tive norms, and perceptions of control with respect toconserving energy).

KNOWLEDGE, ATTITUDES, AND BEHAVIOR 103

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

Method

A total of 79 undergraduate college students (58.2%female) participated in the study for partial coursecredit. They ranged in age from 18 to 26, with a medianage of 20. The study was described as an environmentalknowledge survey. A self-contained questionnaire wasadministered in small groups of 5 to 12 participants.A pilot study with 30 students was conducted to developthe questionnaire which contained an environmentalknowledge test followed by items to assess TPB con-structs, and a scale to measure general attitudes towardenvironmental protection.

Environmental Knowledge Test

The 33-item environmental knowledge test containeditems translated from a German version of an extensiveenvironmental knowledge survey developed by Kaiserand Frick (2002; Frick, Kaiser, & Wilson, 2004) foruse in Switzerland. Some of the items were reformulatedto suit local circumstances. Among the items includedwere, If the amount of carbon dioxide (CO2) doubled,the average global temperature would increase by about10 Fahrenheit, Nuclear energy and fossil fuels are 2types of renewable energy, Using aerosol spray cansnegatively affects the environment by adding to theproblem of global warming, and On average, recycl-able beer bottles are recycled and reused 10 times. Part-icipants rated each statement as true or false and thenindicated their condence on a 5-point scale: How con-dent are you that your answer is correct? Condenceratings were provided on 5-point scales labeledextremely condent, very condent, moderately condent,somewhat condent, and not at all condent. A totalknowledge score was computed by counting the numberof correct responses without reference to the certaintyrating which was used to compute an estimate of sup-port for environmentally friendly behavior, describednext.

Support for Proenvironmental Behavior

Responding to a given item on the knowledge test astrue or false was judged on an intuitive basis to implyeither support, or lack of support, for behavior friendlyto the environment. Internal consistency analyses (i.e.,item-total correlations) were used to conrm the intuit-ive classication. If responses to an item correlated posi-tively with the total score, agreement with the item wasconsidered to be supportive of environmental protec-tion; if they correlated negatively with the total score,agreement with the item was considered to indicatelack of support for the environment. This classicationwas used in the small number of cases where itdisagreed with our intuitive classication. Support for

environmentally friendly behavior was indicated byacceptance of such items as, If we were to stop allozone depleting emissions, the ozone layer would beable to completely regenerate and Air conditionersin cars raise environmental concerns because they leakchlorouorocarbon (CFC) into the air. Consideringthese statements to be false was scored as lack of sup-port for proenvironmental behavior. A supportiveresponse was given a score of 1 and a nonsupportiveresponse a score of 1. These scores were then multi-plied by the corresponding condence rating (scored 1to 5)1 and the resulting products were averaged to pro-duce an estimate of support for environmentally friendlybehavior.2 The environmental support scale had a .51alpha reliability coefcient.3

Environmental Attitude Scale

A nine-item environmental attitude scale included theeight items from the abbreviated version of the NewEcological Paradigm scale (Cordano, Welcomer, &Scherer, 2003; see also Dunlap, Van Liere, Mertig, &Emmet Jones, 2000), and one additional item to addressa salient contemporary concernevidence for globalwarming. This additional item read as follows: Thereis no scientic proof for global warming. Among theother items on the scale were, The balance of natureis very delicate and easily upset, If things continueon their present course, we will soon experience a majorecological catastrophe, and Plants and animals existprimarily to be used by humans. Responses were pro-vided on a 5-point scale with endpoints labeled eitherstrongly disagree and strongly agree or never and always.Negative items were reverse scored, and the mean acrossthe nine items constituted the attitude score (a .77).

TPB Measures

The behavior of interest was dened as conservingenergy this semester. Participants were told that thisrefers to regular performance of energy-saving behaviorsincluding, but not limited to, turning off lights and com-puters when not in use, walking or using bikes=publictransportation instead of an automobile, car pooling,and limiting the duration of hot showers or shampoo-ing. The questionnaire contained 25 items designed to

1We also tried to score the condence scale from 0 to 4, but the

1-to-5 scoring produced somewhat better results.2Because the condence scale was inadvertently omitted for one of

the items, the estimate of support for environmentally friendly beha-

vior is based on 32 items.3The relatively low reliability of this scale is not unexpected because

items were not selected for their support or opposition to protection of

the environment but because, in previous research, they were con-

sidered to assess important knowledge about the environment.

104 AJZEN ET AL.

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

assess the four major TPB constructs in relation to con-serving energy this semester: attitude toward the beha-vior, subjective norm, perceived behavioral control,and intention. After eliminating one subjective normitem to increase internal consistency, 6 items were avail-able to assess each of the theorys predictors. Responsesto all items were provided on 5-point scales. The itemsassessing a given construct were interspersed amongitems assessing the other constructs.

Attitude toward the behavior. Conserving energythis semester was rated on six bipolar adjective scalesthat ranged from very unpleasant to very pleasant,strongly dislike to strongly like, very negative to verypositive, extremely undesirable to extremely desirable,extremely unwise to extremely wise, and extremely badto extremely good. The mean score across the six itemsconstituted our measures of attitude toward conservingenergy this semester, with an alpha coefcient of .88.

Subjective norm. Four items were used to assessinjunctive norms, whereas another two items assesseddescriptive norms. The injunctive norm items asked aboutthe perceived expectations of important others: Peoplewhose opinions I care about approve of my conservingenergy this semester, People I care about encourageme to conserve energy this semester, I feel social press-ure to conserve energy this semester, and People whoare close to me would approve of my conserving energythis semester. Responses were provided on 5-pointstrongly disagree to strongly agree scales. The two descrip-tive norm items referred to the perceived behavior ofothers: Most people like me are going to conserve energythis semester (extremely unlikelyextremely likely) andMost people who are important tome currently conserveenergy (strongly disagree strongly agree). The injunctiveand descriptive norm items were combined to produce anoverall subjective norm measure (a .83).

Perceived behavioral control. The following sixitems provided a measure of perceived behavior control.If I wanted to, I could easily conserve energy this sem-ester (strongly disagreestrongly agree), Whether Iconserve energy this semester is entirely up to me(strongly disagreestrongly agree), For me to conserveenergy this semester is (completely impossibledenitelypossible), Conserving energy this semester is (de-nitely beyond my controldenitely under my control),It will be difcult for me to conserve energy this sem-ester (strongly disagreestrongly agree), and I shouldhave no trouble conserving energy this semester(strongly disagreestrongly agree). The coefcient alphafor this scale was .73.

Intention. The following six items were used to assessintentions. I am planning to conserve energy this sem-ester (strongly disagreestrongly agree), I am likely toconserve energy this semester (strongly disagreestrongly agree), I intend to conserve energy this sem-ester (denitely notdenitely yes), I will probablyconserve energy this semester (denitely will notde-nitely will), I have decided to conserve energy this sem-ester (strongly disagreestrongly agree), and I expect Iwill conserve energy this semester (strongly disagreestrongly agree). The intention scale had an alpha coef-cient of .97.

Energy-saving behavior. Finally, participants wereasked to report, on 5-point scales (from never to always),how frequently they engage in seven specic behaviorsrelated to the environment and energy use and to ratetheir overall energy conservation behavior on twoadditional 5-point scales. After eliminating one behaviorto increase internal consistency, the following specicbehaviors were used to construct an environmentalbehavior index: I walk, ride a bicycle, or take publictransportation to work or school, I wait until I havea full load before doing my laundry, When shopping,I ask for paper bags rather than plastic ones, I regu-larly read at least one environmental journal=magazine(hard-copy or online), I make sure to recycle regularly(e.g., glass bottles, paper, and plastic), and I make agenuine effort to turn off electricity and appliances whennot in use. The mean score was used as a measure ofenvironmental behavior, with an alpha coefcient of .68.

The two general energy conservation items were for-mulated as follows: Generally speaking, do you makean effort to conserve energy in your daily living?(neveralways) and Thinking back over the past fewweeks, how much energy have you been conserving?(none at alla great deal). The internal reliability forthe scale composed of these two items was .77.

Because the scores for the specic and general beha-vior measures were highly correlated (r .67, p< .01),all eight items were combined to produce an overallbehavior score, with a coefcient alpha of .77. This over-all measure of energy conservation behavior was used inthe analyses reported below. It represents a self-reportof current and past energy-saving behavior.

Results

Environmental Knowledge and Support for theEnvironment

As can be seen in Table 1, on average the participantsanswered a little more than 19 questions correctly, for a58% hit rate. Although signicantly different fromthe 50% chance level of 16.5 correct responses,


Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

t(78) 9.98, p< .01, this indicates relatively little accu-rate information about the environment, at least asassessed by the items on our knowledge test. Theenvironmental knowledge score also failed to correlatesignicantly with support for environmentally friendlybehavior as implied by responses on the knowledge test,4

with general proenvironment attitudes, with any of theTPB variables, or with self-reported environmentalbehavior (see Table 1).

In contrast, the environment support implied byresponses to the items on the knowledge test did corre-late signicantly with attitudes toward saving energy,with intentions to save energy, and with reported energysaving behavior. These results suggest that the accuracyof factual information (i.e., knowledge) regarding theenvironment was largely irrelevant for determiningenvironmentally friendly intentions and behavior. Whatmattered more was the extent to which the participantsresponses to the knowledge items tended to be support-ive or nonsupportive of environmentally friendly beha-vior, regardless of whether these responses werecorrect or incorrect. Of course, even the estimate of sup-port for proenvironmental behavioras expressed onthe knowledge testwas a relatively poor predictor. Ithad a correlation of only .26 with intentions to saveenergy and a correlation of .23 with reported energyconservation. These correlations were similar in magni-tude to the correlations obtained for general environ-mental attitudes (see Table 1).

Predicting Energy Saving Intentions andBehavior: TPB

Much better prediction of energy-saving intentionsand behavior was afforded by the TPB. As shown in

Table 1, the best single predictor of intentions to saveenergy was the attitude toward this behavior (r .79,p< .01), and the best single predictor of current or pastenergy-saving behavior was the intention to do so(r .62, p< .01). The results of hierarchical multipleregression analyses are shown in Table 2. In the predic-tion of energy-saving intentions, all three components ofthe model made signicant contributions, accountingfor 69% of the variance. Adding the number of correctanswers on the knowledge test to the model on thesecond step of the analysis failed to account for anyadditional variance in intentions. Perceived control overenergy saving inuenced performance of this behavioronly indirectly by its effect on intentions. Once inten-tions were formed, they were the sole signicant predic-tor of behavior and perceived control failed to make asignicant additional contribution. As can be seen inTable 2, intentions and perceived behavioral controlaccounted for 40% of the variance in reportedenergy-saving behavior.5 Once again, addition of knowl-edge on the second step of the hierarchical regressionanalysis failed to produce a signicant increase in theproportion of explained variance.

Because support for environmentally friendly beha-vior implied by responses to the knowledge test wasfound to correlate signicantly with energy-saving inten-tions and behavior, additional hierarchical regressionswere performed in which support was included asanother predictor on the second step of the analyses.As can be seen in Table 2, neither in the prediction ofintentions nor in the prediction of behavior did theaddition of support for environmentally friendly beha-vior produce a signicant increase in the proportion ofexplained variance suggesting that the effect of supportwas mediated by the TPB variables.

4The relatively low correlation (r .30) between attitudes towardthe environment and support for the environment (as implied by

responses to the knowledge items) may be attributable to the fact that

items on the knowledge test were not designed to assess attitudes and

that the support scores that were derived had relatively low internal

consistency (reliability).

5Strictly speaking, the TPB posits that perceived behavioral control

moderates the effect of intentions on behavior. However, addition of

the Control Intention term to the prediction equation failed toaccount for additional variance in behavior in any of the three studies

reported here.

TABLE 1

Means, Standard Deviations, and Correlations Among Main Variables: Study 1Conserving Energy

M SD KNOW SUP ENV ATT SN PBC INT

Knowledge (KNOW) 19.31 2.48

Support for environment (SUP) 0.71 0.77 .07 Environmental attitude (ENV) 3.97 0.56 .14 .02 Attitude toward saving energy (ATT) 4.02 0.76 .12 .30 .29 Subjective norm (SN) 3.00 0.82 .12 .19 .29 .65 Perceived behavioral control (PBC) 4.30 0.62 .03 .15 .15 .56 .42 Intention to save energy (INT) 3.67 1.00 .04 .26 .12 .79 .65 .63 Energy-saving behavior 3.04 0.67 .05 .23 .33 .57 .57 .47 .62

Note. N 79. r> .18 signicant at p< .05; r> .28 signicant at p< .01.

106 AJZEN ET AL.

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

Discussion

The results of the rst study showed that knowledgeabout the environment was virtually unrelated to gen-eral attitudes regarding the environment, and it had noinuence on attitudes, subjective norms, or perceptionsof control with respect to engaging in energy-savingbehaviors. Consequently, it also had no effect on inten-tions to engage in such behaviors or on reported actionsof an environmentally friendly nature. Nevertheless,support for environmentally friendly behavior asimplied by endorsement of statements on the environ-mental knowledge inventory did have a signicant,albeit rather low correlation (r .23, p< .05) with atti-tudes toward engaging in energy-saving behaviors, andthis effect carried over to intentions and self-reportedbehavior. However, as would be expected in the contextof the TPB, the effect of implied support was fullymediated by the theorys components, and the additionof support to the regression equation failed to accountfor any additional variance in intentions or behavior.

STUDY 2: DRINKING ALCOHOL

The rst study had to do with knowledge about a broaddomainthe environmentnot knowledge related toany particular intention or behavior. Like general atti-tudes or personality traits, such broadmeasures of knowl-edge would not be expected to be good predictors ofparticular behaviors. The second study, therefore,assessedmore narrowly dened knowledge, that is, knowl-edge about alcohol and drinking. Usually, the assumptionwould be made that the more accurate peoples

information about alcohol and drinking, the more nega-tive their attitudes toward drinking should be and, there-fore, the lower should be their intentions to drink andthe less alcohol they should consume. In contrast, we pre-dicted that knowledge (i.e., accurate information) aboutalcohol and drinking will lower intentions to drink onlyto the extent that accurate beliefs about drinking havenegative implications for performance of this behavior.

Method

The methods and procedures were similar to thoseemployed in the rst study. A total of 91 undergraduatecollege students (82% female) participated in the studyfor partial course credit. They ranged in age from 18to 33, with a median age of 20. The study was describedas a survey of attitudes toward alcohol and alcohol use.Formative research was conducted with students fromthe same population (N 32) to construct questionnairemeasures of alcohol knowledge, general attitudestoward alcohol, and TPB variables. Participants in themain study completed the self-contained questionnairein small groups of 5 to 12 students.

Alcohol Knowledge Test

The rst part of the questionnaire contained a31-item alcohol knowledge inventory. The items on thistest were modeled after knowledge questions commonlyfound in the informational literature on alcohol abuse.Examples are Nearly 18 million adult Americans abusealcohol or are alcoholics, Alcohol intoxication pro-duces hearing and vision problems, The leading causeof death among Americans 1524 years old is due toalcohol-related automobile accidents, and One ounceof beer has the same amount of alcohol as one ounce ofwine. In addition, the knowledge test also included vequestions of relevance to the local student population,such as, According to the University of MassachusettsAlcohol Policy, students are allowed to use alcohol con-tainers as decorations within their residence hall room aslong as the containers are not lled with alcohol andOne consequence for violating the campus alcohol pol-icy by drinking or possessing alcohol includes housingprobation for one year. Participants responded bychecking either the true or false option for each item,followed immediately by the same 5-point condencescale used in the rst study. Knowledge scores werecomputed by counting the number of correct responsesto the 31 knowledge items.

Support for Drinking

The true or false responses, together with the con-dence ratings, were used to derive a support for drinking

TABLE 2

Hierarchical Regression Analyses for the Prediction of

Intentions and Behavior: Study 1

Step 1 Step 2

Predictors b R2 b R2 DR2

Intention to save energy

Attitude .51 .52

Subjective norm .21 .22

Perceived behavioral

control

.25 .69 .25

Knowledge .05

Support .03 .69 .00Energy-saving behavior

Intention .54 .52


control

.14 .40 .14

Knowledge .07

Support .07 .41 .01

Note. N 78.p< .05. p< .01.


Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

score by the method described in the rst study. As inthe rst study, initial classication of items as supportiveor nonsupportive of drinking was tested by examiningitem-total correlations. The support for alcohol scalehad a reliability coefcient alpha of .55 (see footnote 3).

Alcohol Attitude Scale

A 20-item scale to assess attitudes toward alcohol andalcoholics was modeled after the measure developed byStrassburger and Strassburger (1965). Based on ourpilot work, many of the items were reformulated to suitcurrent times and our student population (e.g., Somepeople may need to drink in order to feel gay was refor-mulated as Some people need to drink in order to beoutgoing and sociable). A few items were replaced alto-gether. Thus, the item A drunk makes me feel dis-gusted was omitted and replaced by You cant trustpeople who are heavy drinkers. Each item was followedby a 5-point strongly disagree to strongly agree responsescale. Negative items were reverse scored, and the meanacross all 20 items constituted the attitude score. Thisscale had an internal consistency coefcient alpha of .70.

TPB Measures

The behavior of interest was dened as drinking alco-hol this semester. A set of 22 items were formulated toassess the four major constructs in the TPB: attitude,subjective norm, perceived behavioral control, andintention with respect to drinking alcohol this semester.These items were very similar to those used in the rststudy. Preliminary analyses led to the elimination of 2items that reduced internal consistency, one assessingsubjective norms and the other perceived behavioralcontrol. The nal measure included 5 items each asses-sing attitudes, subjective norms, and intentions, and 4items for perceived behavioral control. Reliability coef-cients alpha were .92 for attitudes, .85 for subjectivenorms, .71 for perceived control, and .98 for intentions.

Alcohol Drinking Behavior

A series of eight questions assessed the participantscurrent drinking behavior. After item analysis, four ofthese questions with high internal consistency wereretained as a measure of behavior. The rst behavioralitem asked participants to rate how often they drinkalcohol on a 7-point scale with endpoints labeled neverand virtually every day. The second question was a quan-titative measure: How many drinks do you typicallyconsume on one occasion (e.g., at a party or when youare hanging out with friends)? A drink was dened asa 12-ounce can or bottle of beer, a 4-ounce glass of wine,a 12-ounce bottle or can of wine cooler, or a shot ofliquor straight or in a mixed drink. Participants enteredthe number of drinks in a blank space. Next they wereasked, How would you describe yourself in terms ofyour current use of alcohol? Abstainer, infrequent drin-ker, light drinker, moderate drinker, heavy drinker,chronic alcohol abuser. They were asked to check oneof the alternatives. Finally, they indicated on how manyoccasions they had had one or more drinks in the past30 days, with response alternatives of none, 1 to 2 occa-sions, 3 to 5 occasions, 6 to 9 occasions, 10 to 15 occa-sions, and more than 15 occasions. The reliabilitycoefcient alpha for the measure of behavior was .79.

Results

Alcohol Knowledge and Support for Drinking

Table 3 presents descriptive data and correlationsamong the major variables assessed in the study. Onaverage, the participants provided correct answers toabout 24 of the 31 alcohol knowledge items (77% cor-rect). Interestingly, the knowledge score had a strongnegative correlation (r.57) with support for drink-ing. This nding shows that having correct informationabout drinking and its effects tends to undermine sup-port for drinking, at least as reected in the implicationsof responses to the knowledge test. However, as can beseen in Table 3, knowledge did not correlate signicantly

TABLE 3

Means, Standard Deviations, and Correlations Among Main Variables: Study 2Drinking Alcohol

M SD KNOW SUP ALC ATT SN PBC INT


Support for drinking (SUP) 1.02 0.46 .07 Alcohol attitude (ALC) 2.85 0.43 .16 .21 Attitude toward drinking (ATT) 3.60 0.92 .08 .03 .48 Subjective norm (SN) 3.66 0.95 .09 .02 .53 .61 Perceived behavioral control (PBC) 4.48 0.72 .02 .05 .56 .72 .65 Intention to drink (INT) 4.29 1.10 .06 .01 .50 .89 .69 .84 Drinking behavior 3.06 0.46 .02 .02 .59 .74 .59 .67 .76


108 AJZEN ET AL.

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

with the general alcohol attitude scale (r .16) or withany of the TPB variables. Not surprisingly, therefore,it also did not predict reported drinking behavior(r.02).

Predicting Drinking Intentions and Behavior: TPB

Hierarchical multiple regression analyses were per-formed to examine the degree to which the TPB couldpredict drinking intentions and behavior. As can be seenin Table 4, intentions to drink alcohol were predictedwith a high degree of accuracy. Inspection of theregression coefcients shows that each of the three pre-dictors made a signicant contribution although atti-tudes and perceptions of control appeared to be moreimportant than subjective norms. The model accountedfor 87% of the variance in intentions. The addition ofknowledge on the second step of the analysis had nosignicant effect and did not explain any additionalvariance.

A very similar pattern of results was obtained for theprediction of self-reported drinking behavior. Intentionsand perceptions of behavioral control accounted for58% of the variance in drinking behavior, but onlyintentions made a signicant contribution to the predic-tion. This nding suggests that perceived behavioralcontrol exerted its effect on behavior indirectly by, aspreviously noted, inuencing intentions to drink alco-hol. Adding alcohol-related knowledge to the equationfailed to improve prediction of behavior.

Discussion

The second study showed that knowledge relatedto alcohol and its effects does not seem to discourage

intentions to drink alcohol or inuence reported alcoholconsumption. This conclusion emerged despite the factthat the knowledge test dealt with issues closely relatedto, and thus to some degree compatible with, the beha-vioral criterion. Even when answers to the knowledgequestions were scored in terms of their apparent supportor lack of support for drinking alcohol, the resultingscores predicted neither attitudes toward drinking nordrinking intentions or behavior. The critical issue, fromour perspective, has to do with the contents of theknowledge test. According to the TPB, decisions regard-ing alcohol use are based on readily accessible beha-vioral, normative, and control beliefs regarding thisbehavior. We would argue that the knowledge test, eventhough assessing what is commonly considered to beimportant information about drinking, failed to reectthe participants own beliefs about drinking, beliefs thatactually guided their decisions.

Some evidence in support of this supposition comesfrom research that directly assessed beliefs about drink-ing in the context of the TPB. For example, Armitage,Conner, Loach, and Willetts (1999) elicited accessiblebehavioral, normative, and control beliefs about drink-ing and about cannabis use in a pilot study and thenused the most frequently listed beliefs in the main studyto explain intentions to use these substances and theiractual use. We focus here only on the behavioral andcontrol beliefs regarding alcohol use. Among the likelyoutcomes of alcohol consumption (behavioral beliefs)mentioned by the participants were the beliefs thatdrinking makes me more sociable, will result in mybecoming dependent on it, and will result in my gettinginto trouble with authority. Because beliefs of this kinddo not represent objectively veriable factual infor-mation about drinking, such items of informationare not included in tests that assess knowledge aboutdrinking. Similarly, among the control factors men-tioned most frequently were the beliefs that I haveopportunities to drink and that drinking alcohol coststoo much money. Again, for obvious reasons, thesekinds of issues did not appear on the alcohol knowledgetest. In a sense, then, a typical alcohol knowledge testeven when specically dealing with the behavior of inter-estmay simply assess the wrong kinds of beliefs,beliefs that are not readily accessible when consideringalcohol use and that do not underlie the motivation toperform this behavior.

STUDY 3: ATTENDING A MOSQUE SERVICE

In the rst two studies, knowledge had virtually noeffects on intentions or behavior in relation to conserv-ing energy and drinking alcohol. Even though endorse-ment of factual statements on knowledge tests can, in

TABLE 4

Hierarchical Regression Analyses for the Prediction of

Intentions and Behavior: Study 2

Step 1 Step 2


Intention to drink

Attitude .55 .55



control

.36 .87 .36

Knowledge .02 .87 .00Drinking behavior

Intention .66 .67


control

.12 .58 .11

Knowledge .03 .58 .00

Note. N 90.p< .05. p< .01.


Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

principle, be used to infer support for a behavior ofinterest, only in the second study was there a signicantrelation between knowledge (i.e., accuracy of responseson the alcohol knowledge test) and inferred supportfor drinking. Even there, however, the measure of sup-port for drinking did not predict drinking intentionsor behavior, and neither did the measure of knowledge.We have proposed that the kind of information calledfor on knowledge tests often has little to do with the sali-ent information that actually guides peoples intentionsand actions. Our third study provides evidence thatknowledge can be correlated with behavior-relevantattitudes and with behavior.

The third study dealt with information aboutMuslims and Islam and its effect on willingness toattend a Muslim worship service. We expected that, inthis socially sensitive domain, attitudes toward Islamand Muslims may well have a strong biasing effect onresponses to factual questions, thereby creating a corre-lation between attitudes and knowledge. Knowledgewould then be expected to correlate with the behaviorof interest (willingness to attend a worship service) tothe extent that the attitude measure itself is predictiveof the behavior. Of course, the domain of knowledgein question is very broad, whereas the behavior is quitespecic, and we would therefore expect the correlationbetween knowledge and behavior to be modest at best.In contrast, we hypothesized that the behavior-specicTPB variables afford good prediction of intentions toattend a Muslim worship service and that accurateinformation about Islam and Muslims, if it has an effect,exerts its effect indirectly by inuencing attitudes towardthe behavior, subjective norms, or perceptions of beha-vioral control.

Method

The methods and procedures employed in the rst twostudies served as a model for the third study. Theself-contained questionnaire was developed in formativeresearch with a sample of 31 undergraduate college stu-dents. A sample of 85 students from the same popu-lation participated in the main study in exchange forpartial course credit. The study was described as a sur-vey of knowledge about Muslims and Islam. The ques-tionnaire contained a knowledge test, a scale to assessgeneral attitudes toward Muslim and Islam, and mea-sures of the theory of planned behavior constructs.The participants, who completed the questionnaire insmall groups of 5 to 12, were 76% female and rangedin age from 18 to 31 with a median age of 20. Fifty-two(61%) described themselves as Christian, 12 (14%) asJewish, 5 (6%) as agnostic, 4 (5%) as atheist, and 12(14%) as other. One Muslim participant was excludedfrom the analyses.

Islam Knowledge Test

Items of factual information about Muslims andIslam were constructed as part of the formativeresearch. The information used came from a variety ofsources, including books about Islam and Muslimsand various Internet sites. After eliminating or reformu-lating items that were ambiguous or posed other dif-culties in the formative research, the nal knowledgetest contained 29 items. Examples are Islam, likeChristianity and Judaism, traces its roots back to thePatriarch Abraham, Converting non-Muslims throughforce is one of the ve pillars of Islam, The majority ofMuslims live in the Arabian Peninsula, Giving charita-bly is mandated by Islam, and Muslims started the rstcrusade by attacking Jerusalem. Response alternativeswere true or false, and a total knowledge score wascomputed by counting the number of correct responses.As in the rst two studies, each true=false judgment wasfollowed by a 5-point condence scale.

Support for Islam and Muslims

As in the previous two studies, the true=falseresponses, in combination with the condence rating,were used to construct a support for Islam=Muslimsscale. Again, item-total correlations were used to testthe initial intuitive categorization of all items as support-ive or nonsupportive of Islam or Muslims. The alphareliability coefcient for the support scale was .75 (seefootnote 3).

It should be noted that correct responses to 20 of the29 items on our knowledge test were found to imply afavorable attitude toward Islam and Muslims. Thiswas due to the fact that any unbiased sampling of factsabout Islam and Muslims will contain more positivethan negative information. Moreover, the factual itemsfor the test were culled primarily from Muslim sourcesthat, naturally, tend to present the positive sides orIslam.

Muslim Prejudice Scale

Bushman and Bonaccis (2004) 11-item scale to assessprejudice against ethnic groups was adapted to assessattitudes toward Muslims. Among the items were thefollowing: Muslims have moral standards that theyapply in their dealings with each other, but withnon-Muslims, they are unscrupulous, ruthless, andundependable, It is wrong for Muslims andnon-Muslims to intermarry, and You just cant trusta group of young Muslim men together because theyare probably up to criminal or delinquent activity.Responses were provided on 5-point strongly agree tostrongly disagree scales, and the mean response served

110 AJZEN ET AL.

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

as the measure of attitude toward Muslims. The scalehad a .90 alpha reliability coefcient.

TPB Measures

The studys behavioral focus was on attending a shortmosque service. Participants were given the followinginstructions.

As part of an attempt to bridge relationships betweendifferent ethnic groups on campus, a local nearby mos-que is opening its doors to students interested in attend-ing a brief mosque service. There would also be peopleavailable to answer questions at the end of the service.Transportation would be provided. In order to help usdetermine the interest level of UMASS students, pleasethink about your interest level and answer the followingquestions by circling your choice.

The participants then responded to a series of 20 TPBitems designed to assess their attitudes, subjectivenorms, perceptions of behavioral control, and intentionswith respect to attending the mosque service. Five items,very similar to those used in the rst two studies,assessed each of the four constructs. The reliabilities ofthese measures were .87 for attitudes, .81 for subjectivenorms, .63 for perceived behavioral control, and .95for intentions.

Behavior. To assess interest in attending a brief mos-que service, a form was attached to the questionnaire onwhich participants could sign up for a mosque service bylisting their names and e-mail addresses. They were toldthat this information would be passed on to the commit-tee that is organizing visits to the mosque. The behaviorwas coded as 1 for participants who completed theform and handed it in and coded 0 for participantswho did not. During the debrieng, participants wereinformed that there was actually no organized effort toencourage attending a mosque service, but they were

given information about opportunities to do so if theywere interested.

The questionnaire assessed 16 additional behaviorswith respect to Muslims. Among other things, parti-cipants were asked to indicate how often they discussedmatters pertaining to Muslims or Islam with theirfriends or family; whether they had ever taken a classon Islam, read a book about Islam, or attended a lectureabout Islam; how often they read newspaper articles orwatched TV news stories about Muslims; and howoften, alone or in a group, they have had lunch with aMuslim student. The total set of 16 behaviors had analpha reliability coefcient of .72.

Results

Knowledge About and Support for Islam andMuslims

As assessed by our instrument, participants in thestudy exhibited a relatively low level of knowledge aboutIslam and Muslims. On average, the number of correctresponses on the 29-item test was 17.47, representing a60% hit rate (see Table 5). Still, the average knowledgescore was signicantly higher than the 14.5 scoreexpected by chance, t(84) 8.66, p< .01. Moreover, incontrast to the rst two studies, the degree to whichparticipants had accurate information (or less inaccurateinformation) correlated with the other measuresobtained in this study. First, as can be seen in Table 5,there was a negative correlation (r.39, p< .01)between knowledge regarding Islam and Muslims andprejudice toward Muslims. This nding is consistentwith our expectation that responses to the factual itemson the knowledge test may have been biased by the part-icipants attitudes toward Muslims. The more favorabletheir attitudes (i.e., the lower their prejudice), the morethe participants tended to provide responses with posi-tive implications for Islam and Muslims. In fact, knowl-edge had a .80 correlation with support for Islam andMuslims, as implied by responses on the knowledge test.

TABLE 5

Means, Standard Deviations, and Correlations Among Main Variables: Study 3Mosque Service

M SD KNOW SUP MUS ATT SN PBC INT


Support for Muslims (SUP) 0.54 0.67 .80 Muslim prejudice (MUS) 2.18 0.81 .39 .48 Attitude toward attending (ATT) 3.11 0.77 .47 .39 .43 Subjective norm (SN) 2.49 0.84 .35 .34 .44 .63 Perceived behavioral control (PBC) 3.76 0.75 .35 .34 .48 .50 .50 Intention to attend (INT) 1.96 0.97 .45 .45 .29 .67 .50 .32 Willing to attend mosque service 0.15 0.36 .31 .31 .14 .37 .13 .29 .50



Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

However, as we saw earlier, for the majority of items,responses with positive implications for Islam andMuslims were also the correct responses, thereby pro-ducing the observed negative correlation betweenknowledge scores and the measure of prejudice.

Knowledge exhibited correlations of moderate magni-tude (r .35.47) with attitudes, subjective norms, per-ceived control, and intentions in regard to attending amosque service, and had a point-biserial correlation of.31 (all ps< .01) with signing up for attending a mosqueservice. The same correlation (r .31) obtained betweensupport for Muslims as inferred from responses to itemson the knowledge test and the dichotomous behavior mea-sure. This contrasts with a very low and nonsignicant cor-relation (r.14) between general prejudice towardMuslims and interest in attending the mosque service. Itshould be noted that only 13 of the 85 participants (15%)expressed such an interest, a oor effect that limits the pre-dictive validity of any potential antecedent variable.

Signing up for a mosque service did not correlate sig-nicantly (r .14) with the index of 16 behaviors relatedto Muslims and Islam, but knowledge, support forMuslims and Islam, and prejudice toward Muslimsdid. The respective correlations were .35, .46, and -.28,all signicant at p< .01. Thus, the three dispositionalmeasures that dealt generally with Islam and Muslimshad moderate predictive validity in relation to a broadmeasure of behavior, a nding consistent with the pat-tern of attitude-behavior correlations observed in theliterature (see Ajzen & Fishbein, 1977). In contrast,the TPB measures, which addressed the specic behaviorof attending a mosque service, were poor predictors ofthe general behavior index. The correlations with thebehavior index ranged from .11 (ns) for subjective normto .25 for intention (p< .05).

Predicting Mosque Attendance Intentions andBehavior: TPB

The moderate though signicant effect of knowledgeon mosque attendance intentions and behavior permit atest of the hypothesis that these effects are mediated bythe proximal antecedents specied in the TPB. Note,rst, that the best single predictor of intentions to attenda mosque service was the attitude toward this behavior(r .67, p< .01), and that the best single predictor ofsigning up for a mosque service was the intention todo so (r .50, p< .01). A hierarchical regression analysisshowed that the TPB accounted for 46% of the variancein intentions. Only attitude toward attending a mosqueservice had a signicant regression coefcient in thisanalysis (see Table 6). Even though the proportion ofvariance accounted for increased from 46% to 48%,neither the addition of knowledge about nor theaddition of support for Islam and Muslims made a

signicant contribution to the prediction of intentionsonce the TPB constructs were taken into consideration.Further analyses revealed that only attitudes toward thebehavior mediated the effects of knowledge and supporton intentions. This would be expected in light of thending that only attitudes made a signicant contri-bution to the prediction of intentions.

Because of the dichotomous nature of the behavioralcriterion, a logistic regression analysis was performed forthe prediction of signing up for a mosque service. As canbe seen in Table 6, intention and perceived behavioral con-trol accounted for 40% of the variance in this behavioralcriterion (Nagelkerle R2), but only intention made a sig-nicant contribution to the prediction. This correspondsto an 87% rate of correct prediction. When knowledgeabout and support for Islam and Muslims were added tothe prediction equation, neither made a signicant contri-bution. The proportion of explained variance increasedto 41%, a nonsignicant increase, and the percentage ofcorrect prediction remained at 87%.

Discussion

The third study showed strong correspondence betweenknowledge about and support for Muslims and Islam.That is, when responses to items on the knowledge testwere coded for factual accuracy, they correlatedstrongly with the extent to which acceptance of thatinformation implied support for Islam and Muslims.This nding may be explained by the possible biasingeffect of prejudice on endorsement of questions with lar-gely positive implications for Islam and Muslims. Con-sistent with this argument, knowledge scores also

TABLE 6

Hierarchical Regression Analyses for the Prediction of Intentions and

Behavior: Study 3

Step 1 Step 2


Intention to attend mosque service

Attitude .61 .55



control

.05 .46 .09

Knowledge .02

Support .21 .48 .02Signing up for mosque service (logistic regression)

Intention 1.35 1.25


control

.91 .40 .82

Knowledge .09

Support .04 .41 .01

Note. N 85.p< .01.

112 AJZEN ET AL.

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

correlated negatively with prejudice toward Muslims:the less prejudice they expressed the higher the knowl-edge scores of our participants and the more willing theywere to engage behaviorally with Muslims, as assessedby a 16-item behavioral index. In addition, knowledgealso correlated signicantly with the particular behaviorof interest in the present study: attending a mosque ser-vice. However, because the knowledge test dealt withIslam and Muslims in general, not with any particularbehavior directed at Muslims, and because of a possibleoor effect, the correlations with intentions to attend amosque service and with actually signing up for a mos-que service were of only moderate magnitude. Further,when considered in the context of the TPB, the effectof knowledge on intentions was indirect, mediated lar-gely by attitudes toward attending a mosque service;the effect of knowledge on behavior was also indirect,mediated largely by intention. Knowledge did not mod-erate the effect of intentions on behavior.

STUDY 4: VOTING TO SUPPORT MUSLIMSTUDENT ACTIVITIES

The correlations observed in the third study betweenknowledge about Islam and Muslims on one hand and,on the other, attitudes toward Muslims and support forIslam and Muslims as reected in answers to the knowl-edge test were, arguably, fortuitous. They were likely theresult of the fact that correct responses on the knowledgetest had, for the most part, positive implications for Islamand Muslims. In the fourth study we tested this idea bysystematically manipulating the evaluative implicationsof correct responses on a test of knowledge about IslamandMuslims. As discussed in the introduction, we expecta positive correlation between knowledge and behaviorwhen correct responses to items on the knowledge testimply support for Islam and Muslims, a negative corre-lation when correct responses imply opposition to Islamand Muslims, and no signicant correlation when anequal number of correct responses have positive andnegative implications.

Method

The methods and procedures closely followed those of thethird study. Participantswere 89undergraduate college stu-dents (69% female) who received partial course credit fortheir participation. They had a mean age of 20 and com-pleted the self-contained questionnaire in small groups of5 to 10 students. Fifty-two (58%) indicated that they wereChristian, 8 (9%) were Jewish, 25 (28%) chose atheist=agnostic or other, and 2 (2.2%) did not state their religion.

This study differed from the previous study in twoimportant respects. First, the participants completed

one of three forms of the knowledge test, each consistingof 40 questions. In the positive form, 32 of the knowledgequestions (80%) were selected such that a correctresponse implied support for Islam and Muslims. Exam-ples are Islam teaches that there is no compulsion in reli-gion (the correct response is true, with a positiveimplication) and Historically, Muslims have spreadIslam though military force (false, positive). For theremaining 8 questions the correct answers had negativeimplications. In the second form, 80% of the questionswere selected for correct responses to have negative impli-cations, for example, In strict Islamic tradition, theft ispunished by cutting off a hand (true, negative) andMost predominantly Muslim countries are democra-cies (false, negative). For the remaining questions, cor-rect answers had positive implications. In the neutralform, the questions were balanced such that correctresponses had positive implications for Islam andMuslims for 50% of the items and negative implicationsfor the other 50%. Thirty-two participants completedthe positive form, 25 the negative form, and 32 the neu-tral form. As in the previous studies, the participants rstindicated whether they thought each statement was trueor false, followed by a 5-point condence rating.

The second difference between this and the pre-vious study had to do with the behavioral criterion.Because relatively few participants in Study 3 hadindicated an interest in attending a Muslim worshipservice, we chose as our criterion a different behavior.After completing the knowledge test, the participantswere informed that, to encourage diversity on thecampus, a proposal has been brought before the uni-versity administration to increase funding for Muslimstudent activities. To obtain student input, the admin-istration was said to consider holding a referendumand the participants were told that, later on, theywould have an opportunity to vote their support oropposition to the proposal.

Consistent with this, the TPB constructs wereassessed with respect to voting in favor of the proposalto increase funding for Muslim student activities. Thequestionnaire contained the same items used in the pre-vious study to assess attitudes toward the behavior, sub-jective norms, perceptions of behavioral control, andintentions. One subjective norm item was eliminated toincrease internal consistency. Coefcients alpha were.83 for attitudes, .68 for subjective norms, .69 for per-ceived control, and .95 for intentions.

As in the previous studies, we computed knowledgescores by counting the number of correct answers as wellas scores indicating support for Islam and Muslims, asreected in the responses to the knowledge questions(see the Methods section of Study 3).

In the nal part of the questionnaire, participantswere asked to vote for or against increasing funding


Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

for Muslim student activities at the university. This voteconstituted our measure of behavior.

Results

Although signicantly higher than chance, t(86) 15.92,p< .01, for the total sample, the mean number of correctresponses on the 40-item knowledge test was 26.54,representing a 66% hit rate. The knowledge scores didnot differ appreciably from each other in the three var-iations of the test. The mean number of correct answerswas 26.81 in the positive form, 26.24 in the negativeform, and 26.50 in the neutral form.

Predicting Voting Intentions and Behavior: TPB

TheTPB afforded accurate prediction of intentions andbehavior. In the total sample of participants, the TPBexplained 69% of the variance in intentions to vote in favorof the proposal to increase nancial support for Muslimstudent activities. Of interest, these intentions were con-trolled almost exclusively by subjective norms that had astandardized regression coefcient of .80 (p< .001).Neither attitudes toward the behavior (b .06) nor per-ceived behavioral control (b .09) made signicant con-tributions to the prediction. Such ndings are notinconsistent with the TPB, because in the TPB the relativeimportance of the three predictors of intentions can vary,and it is not impossible for only one predictor to accountfor most or all of the variance in intentions. Very similarresults were obtained when regression analyses were per-formed separately for participants in the three conditionsof the experiment. The proportion of explained variance inintentions was .69 in the positive condition, .67 in thenegative condition, and .71 in the neutral condition (allp< .001), and only the regression coefcients of subjectivenorms were statistically signicant.

A logistic regression for the prediction of the binaryvoting criterion from intentions and perceived beha-vioral control resulted in a Nagelkerle R2 of .79 in thetotal sample of participants, with a 93% hit rate. Onlyintentions made a signicant contribution to the predic-tion, with an unstandardized regression coefcient of3.78 (p< .001). Very similar results were obtained inthe positive condition (Nagelkerle R2 .83; 94% hitrate), in the negative condition (Nagelkerle R2 .70;92% hit rate), and in the neutral condition (NagelkerleR2 .84; 93% hit rate). In each case, only intentionscontributed signicantly to the prediction.

Prejudice, Knowledge, Support for Muslims, andVoting Behavior

The results displayed in Table 7 show that, whendata were collapsed across the three conditions of the

experiment, knowledge did not correlate signicantlywith support for Islam and Muslims, nor did it correlatesignicantly with voting intentions or behavior. Knowl-edge did have a signicant, albeit low correlation of .22(p< .05) with attitudes toward Muslims.

Of greater interest are the comparisons betweenexperimental conditions. As expected, when knowledgewas assessed by means of the positive form, it showedsignicant positive correlations with attitudes towardMuslims, with support for Islam and Muslims, with vot-ing intentions, and with voting behavior. In markedcontrast, when knowledge was assessed by means ofthe negative form, all correlations were negative and,except in the case of intentions, statistically signicant.In the neutral condition, the correlations were low andnone reached statistical signicance.

Between-condition comparisons showed that thecritical differences in correlations were statistically sig-nicant. Thus, the positive .65 correlation betweenknowledge and attitudes in the positive condition wassignicantly greater (z 3.50, p< .001) than the negative.47 correlation in the negative condition. There wasalso a signicant difference between the negative andthe neutral condition (z 2.11, p< .05); the differencebetween positive and neutral conditions was not signi-cant (see Table 7). Similarly, the correlations betweenknowledge and support for Islam and Muslims, asimplied by responses to the knowledge questionnaire,differed signicantly across conditions (z 6.93,p< .001, for the comparison of positive and negative con-ditions; z 2.46, p< .05, for positive vs. neutral con-ditions; and z 4.29, p< .001, for negative vs. neutralconditions).

The knowledgeintention correlations differed signi-cantly in the comparison between positive and negativeconditions. When the positive form was used,knowledge showed a positive correlation of .48 with

TABLE 7

Correlations of Knowledge With Attitudes Toward Muslims, Support,

Voting Intentions, and Behavior: Study 4

Knowledge FormCorrelations of

Knowledge With . . .Total

Samplea Positivea Negativeb Neutralc

Attitudes .22 .65a .47b .28aSupport .14 .71a

.78b .15cVoting intentions .19 .48a

.25b .21a,bVoting behavior .10 .45a

.45b .14bNote. Correlations with different subscripts within rows differ sig-

nicantly from each other at p< .05.aN 89.bn 32.cn 25.dn 32.p< .05. p< .01.

114 AJZEN ET AL.

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

intentions, compared to a negative correlations of .25in the negative condition (z 2.79, p< .01). Finally, andperhaps of greatest interest, knowledge correlated posi-tively (r .45) with actual voting behavior when thepositive form was used to assess knowledge, but it hadan equally strong negative correlation with behavior(r.45) when the negative knowledge form was used.The difference between these two correlations is highlysignicant (z 3.50, p< .001). There was also a signi-cant difference between negative and neutral conditions(z 2.11, p< .05); the difference between positive andneutral conditions was not signicant.

To test the hypothesis that the knowledgebehaviorcorrelations in the positive and negative conditions werelargely due to the effect of attitudes toward Muslims onboth knowledge and behavior, we computed partial cor-relations between knowledge and behavior while control-ling for attitudes. The results conrmed this hypothesis.In the positive condition, attitudes toward Muslims cor-related .65 with knowledge aboutMuslims and Islam (seeTable 7), and it had a correlation of .62 with behavior.Controlling for attitudes toward Muslims reduced the.45 knowledgebehavior correlation to .09 (ns). In thenegative condition, attitudes correlated .47 withknowledge and .55 with behavior. When attitudes werecontrolled, the knowledge-behavior correlation wasreduced from .45 to .13 (ns). (In the neutral condi-tion, the correlation between knowledge and behaviorremained nonsignicant, going from .14 to .03.)

Discussion

The fourth study clearly demonstrated the problematicnature of the relation between accurate informationand behavior. When responses to factual questions ona knowledge test have no evaluative implications, thetest may in fact assess little more than the degree towhich respondents have accurate information in thedomain of interest. However, in many cases answers tofactual questions have favorable or unfavorable implica-tions for the topic under consideration, and responses tothe questions, rather than reecting factual knowledge,may be viewed as an indirect expression of an underly-ing attitude. When the knowledge test is balanced suchthat correct answers have an approximately equal num-ber of positive and negative implications, the resultingmeasure may have little to do with accurate infor-mation, but it will not be biased in either a positive ornegative direction. However, when correct responseshave predominantly positive or predominantly negativeimplications for the topic under consideration, we willobserve a positive or a negative correlation, respectively,between attitudes and knowledge. As a result, we willalso obtain either a positive or a negative correlationbetween knowledge and behavior.

Consistent with these arguments, in Study 4 there wasa positive correlation between knowledge and behaviorwith respect to Muslims when the positive form of theknowledge test was used, and this correlation virtuallydisappeared after attitudes toward Muslims were stat-istically controlled. Conversely, there was a negativecorrelation between knowledge and behavior when thenegative form of the knowledge test was used, and thiscorrelation was also reduced to nonsignicance bycontrolling for attitudes toward Muslims.

GENERAL DISCUSSION AND CONCLUSIONS

It stands to reason that the information individuals havein a certain behavioral domain is of central importancefor the decisions they make. More problematic is theproposition that appropriate or desirable behavioraldecisions require that this information be accurate.The frequently observed lack of correlation betweenknowledge and behavior effective to produce desiredoutcomes has led many investigators to conclude thatknowledge is a necessary but not a sufcient condition(DiClemente, 1989; Fisher & Fisher, 1992). However,from the perspective of the TPB, information accuracyis neither necessary nor sufcient; indeed, it can be irrel-evant to decision making. Instead, what determinesintentions and actions is subjectively held information(i.e., beliefs) that links a behavior of interest to positiveor negative outcomes, to the normative expectations ofimportant referent individuals or groups, and to controlfactors that can facilitate or inhibit performance of thebehavior. Whether that information is accurate or inac-curate is immaterial. To be sure, the TPB leaves openthe possibility that the extent to which peoples infor-mation is accurate will, on occasion, correspond to thebehavioral, normative, or control beliefs they hold.However, this will often not be the case, because itemson a knowledge test rarely deal with the particular beha-vior of interest and, even when they do, they often haveno clear implications for behavioral performance.Furthermore, knowledge tests may reect attitudesrather than assess accurate information.

The results of the four studies reported in this articleare consistent with these arguments. The knowledge testin the rst study had to do with general informationabout the environment, rather than with the particularbehavior of interest (conserving energy), and parti-cipants knowledge scores were indeed not predictiveof their support for conserving energy. In fact, even sup-port for the environment implied by acceptance of theinformation on the knowledge test was unrelated tothe accuracy of peoples responses. The knowledge testin the second study did deal with the specic behaviorof interest, namely, drinking alcohol. Even though at


Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

least one study (de Nooijer, Lechner, & de Vries, 2003)has found a signicant correlation between behavior-focused knowledge and behavior, the items on mostknowledge testseven if dealing with the behavior ofinterestare largely factual in nature, having no clearimplications for behavior, and are thus unlikely to cor-respond to the behavioral, normative, or control beliefsthat actually guide peoples behavior. This was also thecase in our knowledge test. Consequently, alcoholknowledge predicted neither intentions to drink norreports of actual drinking behavior.

That knowledge can sometimes be predictive of atti-tudes and behavior was shown in the third study. Here,knowledge about Islam and Muslims corresponded toacceptance of items that supported a pro-Muslim pos-ition. As a result, knowledge correlated positively withattitudes toward Muslims, it predicted the general pat-tern of behavior in relation to Muslims and, to a lesserextent, the specic behavior of signing up for a mosqueservice. In the fourth study we were able to show that thedirection of the correlation between knowledge andbehavior can be systematically manipulated. By select-ing factual questions in such a way that correct answersimplied either predominantly favorable attitudes or pre-dominantly unfavorable attitudes toward Muslims, wewere able to produce a positive knowledgebehaviorcorrelation or a negative knowledgebehavior corre-lation, respectively.

In contrast to relying on general knowledge toexplain intentions and behavior, the TPB focuses onthe proximal antecedents of the behavior in question.Relying on this approach, it was possible to achieve ahigh degree of predictive accuracy in all four studies.Attitudes, subjective norms, and perceptions of controlwere found to predict intentions to drink alcohol, toconserve energy, to attend a mosque service, and to votesupport for Muslim student activities; these intentionswere generally good predictors of the correspondingbehavior. Also consistent with the theory, when knowl-edge did correlate positively with intentions and beha-vior in Study 3, these relations were mediated byattitudes in the case of intentions and by intentions inthe case of behavior.

The implications of the current research arefar-reaching and call into question current behavioralintervention strategies. Many educational campaigns,especially in the health domain, are focused on impart-ing accurate factual information of a general nature. Itis expected that once people have a good understandingof the issues, they will engage in socially or personallydesirable behavior. Unfortunately, more often thannot, this approach results in failure, and people continueto take unnecessary risks or engage in socially undesir-able behavior. The present research helps explain whythe focus on knowledge is misplaced and suggests an

alternative approach that is more likely to be effective.Instead of trying to make sure that people have accurateinformation, we need to nd out what information theyactually possess and how this information affects inten-tions and actions, irrespective of whether the infor-mation is accurate or not. Furthermore, we need to beconcerned not about general knowledge in a behavioraldomain but rather with information or knowledge thatguides the behavior of interest (i.e., with beliefs aboutthe behavior). Once we have identied the behavioral,normative, and control beliefs that are readily accessiblefor individuals in the population of interest, we can pro-vide them with information to challenge beliefs that arecontrary to the desired behavior, with information thatstrengthens their existing supportive beliefs, or withinformation that leads to the formation of new beliefssupportive of the desired behavior.

ACKNOWLEDGMENTS

Nicholas Joyce is now at the Department of Communi-cation, University of Arizona.

REFERENCES

Ajzen, I. (1991). The theory of planned behavior. Organizational

Behavior and Human Decision Processes, 50, 179211.

Ajzen, I. (2005). Attitudes, personality, and behavior (2nd ed.).

Maidenhead, UK: Open University Press.

Ajzen, I., & Fishbein, M. (1977). Attitudebehavior relations: A theor-

etical analysis and review of empirical research. Psychological

Bulletin, 84, 888918.

Ananth, P., & Koopman, C. (2003). HIV=AIDS knowledge, beliefs,

and behavior among women of childbearing age in India. AIDS

Education and Prevention, 15, 529546.

Armitage, C. J., Conner, M., Loach, J., & Willetts, D. (1999). Different

perceptions of control: Applying an extended theory of planned

behavior to legal and illegal drug use. Basic and Applied Social

Psychology, 21, 301316.

Blumenfeld, W. S. (1966). A research note on the method of

error-choice. Educational and Psychological Measurement, 26,

847851.

Bushman, B. J., & Bonacci, A. M. (2004). Youve got mail: Using

e-mail to examine the effect of prejudiced attitudes on discrimi-

nation against Arabs. Journal of Experimental Social Psychology,

40, 753759.

Calsyn, D. A., Saxon, A. J., Freeman, G., & Whittaker, S. (1992). Inef-

fectiveness of AIDS education and HIV antibody testing in reducing

high-risk behaviors among injection drug users. American Journal of

Public Health, 82, 573575.

Cordano, M., Welcomer, S. A., & Scherer, R. F. (2003). An analysis

of the predictive validity of the New Ecological Paradigm Scale.

Journal of Environmental Education, 34, 2228.

de Nooijer, J., Lechner, L., & de Vries, H. (2003). Social psychological

correlates of paying attention to cancer symptoms and seeking

medical help. Social Science and Medicine, 56, 915920.

DiClemente, R. J. (1989). Prevention of human immunodeciency

virus infection among adolescents: The interplay of health education

116 AJZEN ET AL.

Downlo

aded

By:

[Un

iver

sity

Of

Mass

achu

sett

s] A

t: 1

2:59

17

May

2011

and public policy in the development and implementation of

school-based AIDS education programs. AIDS Education and

Prevention, 1, 7078.

DiClemente, R. J., Forrest, K. A., & Mickler, S. (1990). College

students knowledge and attitudes about AIDS and changes in

HIV-preventive behaviors. AIDS Education and Prevention, 2,

201212.

Dunlap, R. E., Van Liere, K. D., Mertig, A. G., & Emmet Jones, R.

(2000). Measuring endorsement of the new ecological paradigm:

A revised NEP scale. Journal of Social Issues, 56, 425442.

Feeley, T. H., & Servoss, T. J. (2005). Examining college students

intentions to become organ donors. Journal of Health Communi-

cation, 10, 237249.

Fishbein, M., & Ajzen, I. (1974). Attitudes towards objects as predic-

tors of single and multiple behavioral criteria. Psychological Review,

81, 5974.

Fisher, J. D., & Fisher, W. A. (1992). Changing AIDS-risk behavior.

Psychological Bulletin, 111, 455474.

Fisher, J. D., Fisher, W. A., Williams, S. S., & Malloy, T. E. (1994).

Empirical tests of an information-motivation-behavioral skills

model of AIDS-preventive behavior with gay men and heterosexual

university students. Health Psychology, 13, 238250.

Frick, J., Kaiser, F. G., & Wilson, M. (2004). Environmental knowl-

edge and conservation behavior: Exploring prevalence and structure

in a representative sample. Personality and Individual Differences,

37, 15971613.

Guerra, C. E., Dominguez, F., & Shea, J. A. (2005). Literacy and

knowledge, attitudes, and behavior about colorectal cancer screen-

ing. Journal of Health Communication, 10, 651663.

Hammond, K. R. (1948). Measuring attitudes by error choice: An

indirect method. Journal of Abnormal and Social Psychology, 43,

3848.

Helweg-Larsen, M., & Collins, B. E. (1997). A social psychological

perspective on the role of knowledge about AIDS in AIDS preven-

tion. Current Directions in Psychological Science, 6, 2326.

Ievers-Landis, C. E., Burant, C., Drotar, D., Morgan, L., Trapl, E. S.,

& Kwoh, C. K. (2003). Social support, knowledge, and self-efcacy

as correlates of osteoporosis preventive behaviors among preadoles-

cent females. Journal of Pediatric Psychology, 28, 335345.

Kaiser, F. G., & Frick, J. (2002). Entwicklung eines Messinstrumentes

zur Erfassung von Umweltwissen auf der Basis des MRCML

Modells. Diagnostica, 48, 181189.

Kallgren, C. A., & Wood, W. (1986). Access to attitude-relevant infor-

mation in memory as a determinant of attitudebehavior consist-

ency. Journal of Experimental Social Psychology, 22, 328338.

Kidder, L. H., & Campbell, D. T. (1970). The indirect testing of atti-

tudes. In G. F. Summers (Ed.), Attitude measurement (pp. 333385).

Chicago, IL: Rand McNally.

Kraus, S. J. (1995). Attitudes and the prediction of behavior: A

meta-analysis of the empirical literature. Personality and Social Psy-

chology Bulletin, 21, 5875.

Misovich, S. J., Martinez, T., Fisher, J. D., Bryan, A., & Catapano, N.

(2003). Predicting breast self-examination: A test of the information-

motivation-behavioral skills model. Journal of Applied Social


Newcomb, T. M. (1946). The inuence of attitude climate upon some

determinants of information. The Journal of Abnormal and Social


Schlueter, L. A. (1982). Knowledge and beliefs about breast cancer and

breast self-examination among athletic and nonathletic women.

Nursing Research, 31, 348353.

Sheeran, P., & Taylor, S. (1999). Predicting intentions to use condoms:

A meta-analysis and comparison of the theories of reasoned action

and planned behavior. Journal of Applied Social Psychology, 29,

16241675.

Silver Wallace, L. (2002). Osteoporosis prevention in college women:

Application of the expanded health belief model. American Journal

of Health Behavior, 26, 163172.

Spirito, A., Ruggiero, L., Duckworth, M., & Low, K. G. (1993). The

relationship of diabetes knowledge to regimen compliance

and metabolic control during pregnancy. Psychology & Health, 8,

345353.

Strassburger, F., & Strassburger, Z. (1965). Measurement of attitudes

toward alcohol and their relation to personality variables. Journal of

Consulting Psychology, 29, 440445.

Weschler, I. R. (1950). An investigation of attitudes toward labor and

management by means of the error-choice method:

Documents

Ajzen - Knowledge and the Prediction of Behavior