Variables That Moderate the Attitude-Behavior Relation ...fsnagle.org/files/davidson1979variables.pdf · 2. Attitude change. In a recent review of the attitude-behavior literature,

Journal of Personality and Social Psychology1979, Vol. 37, No. 8, 1364-1376

Variables That Moderate the Attitude-Behavior Relation:Results of a Longitudinal Survey

Andrew R. Davidson James J. JaccardUniversity of Washington Purdue University

The following factors were hypothesized to moderate the attitude-behavior re-lation: (a) the behavioral sequence that must be successfully completed priorto the occurrence of the behavior, (b) the time interval between the measure-ment of attitudes and behavior, (c) attitude change, (d) the respondent's ed-ucational level, and (e) the degree of correspondence between attitudinal andbehavioral variables. The behaviors investigated were having a child and usingoral contraceptives. A stratified random sample of 244 married women in amidwestern urban area was studied during a three-wave, 2-year longitudinalstudy. Selection of attitudinal and belief measures was guided by the Fishbeinmodel of behavioral intentions. Consistent with the hypotheses, the relationsbetween behavior and both intention and the model's attitudinal and normativecomponents were substantially attenuated by (a) events in the behavioralsequence not under the volitional control of the actor, (b) an increase in thetime interval between the measurement of attitudes and behavior from 1 to 2years, and (c) changes in the model's attitudinal and normative componentsduring the first year. The respondent's educational level did not affect attitude-behavior consistency. Finally, the attitude-behavior correlation increased sig-nificantly as the degree of correspondence between the two variables increased.

Wicker (1969) concluded his comprehen- and behavioral measures. By assessing bothsive review of the attitude-behavior relation variables at corresponding levels of specificity,with the suggestion, "It is considerably more that is, measuring attitude toward the act forlikely that attitudes will be unrelated or only the prediction of a specific behavior or mea-slightly related to overt behaviors than that suring a global attitude toward an object forattitudes will be closely related to actions" the prediction of a multiple-act behavioral(p. 76). Rather than signaling a decrease in criterion, a reasonable degree of predictiveresearch on this topic, pessimistic reviews by accuracy can be obtained (Ajzen & Fishbein,Wicker and others (e.g., Deutscher, 1966, 1973, 1977; Fishbein & Ajzen, 1974; Heber-1969; Ehrlich, 1969; McGuire, 1969) appear lein & Black, 1976; Weigel & Newman, 1976;to have prompted a renewal of interest in the Weigel, Vernon, & Tognacci, 1974; Weinstein,relationship between attitude and action. One 1972; Wicker & Pomazal, 1971).encouraging line of inquiry has focused on A second line of inquiry, encouraged bymethodological refinements in the attitudinal Kelman (1974), has attempted to specify the

personal and skuational variables influencingthe extent to which people act in accord with

This research was supported by Research Grant their stated attitudes and beliefs. Individual?KD,,07^4",0,1'02'°i i°m th'Nat)onal Inst""te of difference variables that have been found toChild Health a n d Human Development. W e a r e , . 1 1 1 . ,grateful to Marilyn Cohen for her substantial aid moderate attitude-behavior correspondenceboth in supervising the interviewers and data coders include the degree of affective-cognitive COn-and in locating respondents who moved. sistency (Norman, 1975), the tendency to

Requests for reprints should be sent to Andrew R. .. ., ... ' ,, ,~ ,Davidson, Department of Psychology, NI-25, Uni- ascrlbe responsibility to the self (Schwartz,versity of Washington, Seattle, Washington 98195. 1973), and the degree of self-monitoring

Copyright 1979 by the American Psychological Association, Inc. 0022-3S14/79/3708-1364$00.7S

1364

ATTITUDE-BEHAVIOR RELATION 1365

(Snyder & Tanke, 1976). Similarly, a numberof situational variables appear to moderatethe attitude-behavior relation. For example,Warner and DeFleur (1969) found that thepublic versus private nature of the behaviorinteracts with attitude in influencing attitude-behavior consistency, and Snyder and Swann(1976) demonstrated that situations that in-crease the relevance of salient attitudes asguides to actions increase the correspondencebetween attitude and action. A recent seriesof studies (Fazio & Zanna, 1978; Fazio,Zanna, & Cooper, 1978; Regan & Fazio,1977) provide support for the notion thatattitudes formed through direct behavioralexperience with the attitude object are morepredictive of later behavior toward that ob-ject than are attitudes based upon indirect,nonbehavioral experience. In summary, sev-eral personal and situational factors have beenfound to affect the attitude-behavior relation.As Regan and Fazio (1977) commented, "Thequestion facing researchers is, therefore, nolonger whether an individual's attitudes canbe used to predict his overt behavior, butwhen" (p. 30, italics in the original).

The present research continues the study offactors that moderate the degree of attitude-behavior consistency and investigates the fol-lowing variables:

1. Sequence of prior events. As Fishbeinand Jaccard (1973) have observed, the occur-rence of a behavior under investigation is fre-quently dependent upon the successful com-pletion of a sequence of prior events. Forexample, in order to have a child a womanmust (a) have intercourse, (b) not use birthcontrol, (c) conceive, and (d) not have aspontaneous or induced abortion. While someof the sequences are potentially under thecontrol of the actor (e.g., not using birth con-trol), others are not (e.g., conception). Ifperformance of the final behavior requires thesuccessful completion of sequences not underthe volitional control of the actor, it is hy-pothesized that attitude toward the behaviorwill provide a better estimate of 'the likelihoodthat the actor will initiate the sequence thanthat the actor will complete the sequence andperform the behavior in question.

2. Attitude change. In a recent review ofthe attitude-behavior literature, Schuman andJohnson (1976) noted that little attentionhas been paid to the possibility that attitudechange may occur between the measurementof attitude and behavior. They commented,"The extent to which A-B correlations arereduced because of real changes in underlyingattitude needs to be identified as far as pos-sible" (p. 178). Such identification is impos-sible in most studies because attitude is as-sessed at only one time. In the present re-search, attitude was measured twice prior tothe performance of behavior in order to detectattenuation attributable to attitude change.Respondents whose attitudes remain stable arehypothesized to exhibit greater attitude-be-havior consistency than respondents who ex-perience attitude change.

3. Time interval. A third variable hypoth-esized to influence the degree of relation be-tween attitude and behavior is the time inter-val between the measurement of attitude andthe performance of behavior. As Fishbein andJaccard (1973) noted, it is probably theprocesses occurring during that time interval—most notably exposure to new information—and not the passage of time per se that mod-erate the relation. This hypothesis is concep-tually similar to the one above; the longer thetime interval between the measurement ofattitude and behavior, the higher the prob-ability of exposure to new information and, inturn, attitude change.

Three previous investigations have reporteddata relevant to the influence of temporal in-stability on attitude-behavior consistency.Norman (1975) regressed behavior (volun-teering to be a subject) on attitudes towardacting as a subject, assessed both 3 and 6weeks prior to the measurement of behavior.For each of three attitudinal measures, thesecond administration was more highly cor-related with behavior than the first, but thedifferences were not statistically significant.It has been suggested (Schwartz, 1978) thatfor this behavior the 3-week interval betweenattitude assessments was too brief for an un-ambiguous effect to be detected. In studies ofvoting behavior, Kelley and Mirer (1974)

1366 ANDREW R. DAVIDSON AND JAMES J. JACCARD

have reported that for some subgroups thetime interval between the measurement ofvoting behavior and attitudes can moderatethe attitude-behavior correlation. Analyses ofdata from the 28% of the subjects judgedmost likely to change their minds found thata substantial proportion of variance in errorsof prediction was accounted for by the num-ber of days that intervened between the inter-view and the election. In the final and mostrecent investigation, Schwartz (1978) re-gressed stated willingness to tutor Wind chil-dren on perceived moral obligation to tutorblind children, measured both 3 and 6 monthsprior to the criterion measure. The correlationwith the criterion was significantly higher overthe shorter time interval. However, asSchwartz noted, it is uncertain whether theresults obtained with perceived moral obliga-tion can be generalized to more traditional at-titude scales measuring evaluation. In sum,although specific aspects of each of thesestudies—interval between attitude assess-ments, nature of the subsample investigated,nature of the predictive variable—have pre-cluded a clear demonstration of the influenceof time interval on the consistency 'betweenbehavior and 'the evaluation of the behavior,these studies together provide a sound basisfor positing the present hypothesis.

4. Respondent's education. Consistent withformulations proposed by Peak (1955), Ros-enberg (1956), and Fish'bein (1963), attitudetoward a 'behavior is viewed as the sum ofone's beliefs about the consequences of per-forming the 'behavior multiplied by the eval-uation of those consequences. Such formula-tions, with their emphasis on cognitive-affec-tive consistency, are frequently criticized ashaving greater validity for college studentsand intellectuals than for the majority of thepopulation (cf. Bern, 1970). If such critiquesare correct, education should interact withattitudes in the prediction of behavior, anddegree of attitude-behavior consistencyshould correlate positively with the respon-dent's educational level. The findings of Ro-keach (1973) also lead to the hypothesis thatgreater consistency will 'be observed for re-spondents with more education. In a nationalsample of American adults, the instrumental

value logical defined as consistent, rationalwas more positively valued 'by highly educatedrespondents than by those with less education.

5. Correspondence between attitudinal andbehavioral variables. As noted earlier, if atti-tude and behavior are both measured at a sim-ilar level of specificity, a reasonable relationis generally observed. In an extension of thespecificity hypothesis, Ajzen and Fishbein(1977) maintain that attitudinal and behav-ioral variables are defined by four elements:an action, the target the action is directedtoward, the context in which the action is tobe performed, and the time at which theaction occurs. They argue that the correlationbetween attitude and behavior is determinedin part by the degree of correspondence ormatch between the elements comprising thetwo variables. They classified past studies interms of the correspondence between attitudeand behavior on two of the four elements,target and action. Consistently significantattitude-behavior correlations were obtainedonly if there was high correspondence betweentarget and action elements. Their investiga-tion, however, examined the moderating ef-fects of target and action correspondence bycomparing attitude-behavior correlations ob-tained from different studies. In the presentresearch, the effects of target, action and, inaddition, time correspondence on attitude-behavior consistency are tested within onestudy. It is hypothesized that as the degreeof correspondence increases, the obtained at-titude-behavior correlation will increase.

Attitude Model

The selection of attitudinal measures wasguided by the Fishbein (1967) model of be-havioral intention. The model is predicated onthe assumption that most human behavior ofconcern to social scientists is to some degreevolitional in nature and hence guided by thebehavioral intent of the individual. Alge-braically, the model is expressed as follows:

B ~ El = [£ 5,-E

i-liMCAWi, (1)


where B — overt behavior, BI = the behav-vioral intention to perform the 'behavior, Bt =the belief (perceived probability) that per-forming the behavior will lead to consequenceXt, EI = the evaluation of Xit NBt = the per-ceived expectation of Referent *, MCi = themotivation to comply with Referent i, n =the number of salient consequences, m = thenumber of salient normative beliefs, and W\and Wz = empirically determined regressionweights.

As Peak (19SS), Rosenberg (1956), andFishbein (1963) have theorized, the 254£(

component is viewed as an index of the per-son's attitude toward performing the behavior(Aact). Fishbein and Ajzen (1975) maintainthat the second predictive component (2NBt-MCi) assesses the influence of the social en-vironment or the general subjective norm(SN) on behavior. Empirical support for theequivalence of 2£(£4 and Aact and of ^NBt-MCt and SN is discussed in Fishbein andAjzen (1975).

According to the model, any variable otherthan the attitudinal and normative compo-nents can influence BI only indirectly. Social,demographic, and personality characteristicsof the respondent can affect intentions only ifthey influence ££<£< or ^NBiMCl or their rel-ative weights. Thus if this framework is cor-rect, such variables as education should nothave a direct effect on the magnitude of therelation between the components of the modeland intention, and hence behavior.

Fishbein and Ajzen (1975) argue that therelation between BI and B should be verystrong, provided that intention and behaviorare measured at correspondent levels of spe-cificity and that nothing intervenes to alterintention. Whenever a strong intention-behav-ior relation is observed, the behavior in ques-tion should also be predictable from the atti-tudinal and normative components. Assuminga significant BI-B correlation, it is hypoth-esized that behavior can be predicted from alinear combination of the attitudinal andnormative components. In addition, it is hy-pothesized that the variables moderating theintention-behavior relation will also moderate

the relation between attitudes, subjectivenorms, and behavior.

Behavioral Domain

The behaviors chosen for study were child-bearing and contraceptive use. While socialscientists have met with success in document-ing fertility differentials on the basis of socialand demographic variables, (cf. Rindfuss &Sweet, 1977) there has been a marked lackof success in explaining fertility behavior withpsychological constructs. Two major fertilitysurveys, the Indianapolis study (Whelpton &Riser, 1946-1958) and the Princeton study(Westoff, Potter, Sagi, & Mishler, 1961)failed to find any noteworthy relationshipsbetween psychological variables (primarilypersonality measures) and fertility and fam-ily planning behaviors.

Recently, research findings have indicatedthat a few psychological models that can becategorized as expectancy models are of someutility in the prediction of fertility decisionmaking (see, for example, Beach, Townes,Campbell, & Keating, 1976; Davidson & Jac-card, 1975; Werner, Middlestadt-Carter, &Crawford, 1975). Working with the Fishbeinmodel, Davidson and Jaccard (1975) investi-gated attitudes concerning three family plan-ning behaviors in a sample of married women.In support of the model, 2.BA and SM£4MC4

explained an average of 60% of the variancein behavioral intentions.

The data reported in Davidson and Jaccard(1975) were obtained during the first waveof a three-wave longitudinal survey. Thesample was reinterviewed both 1 year and 2years after the initial interview. The presentstudy examines the relation between inten-tions, at the first interview, of having a childduring the next 2 years and using oral contra-ceptives during the next 2 years and the cor-responding self-reports of behavior during the2-year period.

Method

Sample

Respondents in the initial survey were 270 whitemarried women, age 18-38 years, residing in a mid-


western urban area. The sample was stratified bythree levels of socioeconomic status and two levels ofreligious affiliation (Catholic and Protestant). Thesampling procedure was designed to select randomly45 women for each cell of the 2 X 3 design. For adetailed description of the sampling strategy, seeDavidson and Jaccard (1975).

Loss to Follow-Up

The respondents were reinterviewed both 1 and 2years after the initial survey. Of the original 270respondents, 244 completed all three interviews.1

Approximately one half of the respondents lost tofollow-up refused to be reinterviewed and the otherhalf could not be located. Loss to follow-up did notsignificantly vary among the six cells of the samplingdesign.

Interviews

Approximately 1 hour was required for the re-spondent to complete each questionnaire. Althoughthe interviews were self-administered, the interviewerremained in the house while the questionnaire wasbeing completed to answer any questions the re-spondent might have. The respondent received $10for each interview.

Measurement Procedures

To insure that relevant beliefs and referents for theattitudinal and normative components of the modelwere included in the preceded questionnaire, initialelicitation interviews were conducted with an inde-pendent sample of 55 women. These women wereselected from the same population as the women inthe longitudinal sample. The interviews identifiedfour referents (e.g., husband, close friends) and ninebeliefs (e.g., the effect of having a child on the hus-band-wife relationship, the woman's freedom to pur-sue other activities, the family budget, etc.) salientto the population. A list of the referents and beliefsincluded in the questionnaire is presented in Davidsonand Jaccard (1976).

All predictive components were measured on 7-point adjective scales. Examples of the measures are

1. Behavioral intention (BI) (I intend to have achild within the next 2 years) was measured on alikely-unlikely scale.

2. Belief about the act (Bi) (For me, having achild within the next two years would make my mar-riage stronger) was measured on a likely-unlikelyscale.

3. Evaluation of the consequence (Ei) (Making mymarriage stronger would be . . .) was measured on agood-bad scale.

4. Attitude toward the act (Aact) (For me, havinga child within the next two years would be . . .) wasobtained by summing the responses to three evalua-

tive scales: good-bad, nice-awful, pleasant-unpleas-ant.

5. Normative belief (NBt) (My parents think Ishould have a child within the next two years) wasmeasured on a likely-unlikely scale.

6. Motivation to comply (MCi) was measured on ascale anchored by generally speaking, I want to dowhat Referent i thinks I should do and generallyspeaking, I do not want to do what Referent i thinksI should do.

7. General subjective norm (SN) (People who areimportant to me and whose opinions I value think Ishould have a child within the next two years) wasmeasured on a likely-unlikely scale.

Scales assessing evaluations of consequences andattitude toward the act were scored —3 (bad) to+3 (good). All other scales were scored from 1 (un-likely; not motivated to comply) to 7 (likely; mo-tivated to comply).2 SBi£l and 2NB,MCt wereobtained by multiplying the score on each beliefstatement by the score on the corresponding evalua-tion or motivation to comply and then summingthese products for all beliefs. In addition to the aboveitems, the questionnaire assessed a number of fer-tility-related attitudes and contained measures ofeducation, scored from 1 (sixth-grade education orless) to 8 (PhD or other professional degree), andother demographic variables.

A dichotomous criterion variable was created foreach behavior. For childbearing, women reporting abirth between the first and third interviews wereassigned a score of 1, and the remainder of respon-dents were assigned a score of 0. In families report-ing a birth, the child was typically observed by theinterviewer. For oral contraceptive use, womenreporting that they used oral contraceptives at anytime during the 2-year interval were scored 1, andwomen never using the pill were scored 0.

The 2Bi£( and SNBiMCi measures were obtainedfor only one of the behaviors—having a child. Dueto time constraints, only the direct measures of theattitudinal and normative components, Aact and SN,respectively, were obtained for the second behavior—using oral contraceptives. In the first of the three

1 Due to missing data and failure to follow instruc-tions, the sample sizes in the following analyses rangefrom 242 to 244.

2 In our previous studies all scales were scored —3to +3. However, we now recognize that with the ex-ception of the evaluation measures, the remainder ofscales are assessing variables that are theoreticallyunipolar (e.g., subjective probability). Hence, theunipolar scales are scored 1 to 7 in the presentanalysis. A comparison of variables formed from boththe present and previous scoring rules indicated thatthe two were substantially intercorrelated; for 2Bi£i,r = .93, and for 2#5<AfCi, r — .76. For a discussionof unipolar versus bipolar scoring of attitude andbelief measures, see Fishbein and Ajzen (1975, pp. 82-86).


interviews, questions referred to performing each be-havior during the next 2 years. In the second inter-view, the questionnaire was adjusted to query therespondents about having a child and using oralcontraceptives during the next year.

Results

The first half of this section focuses on in-tention-behavior consistency and the factorshypothesized to influence the degree of con-sistency. The second half investigates the pre-diction of behavior from the model's atti-tudinal and normative components and ex-amines the variables that moderate that rela-tion.

Prediction of Behavior jrom BehavioralIntention

The intention-behavior relation. For eachbehavior a point-biserial correlation was com-puted between behavioral intention, measuredduring the initial interview, and behavior dur-ing the 2-year period. The BI-B correlationwas .526 (p < .01) for childbearing and .678(p < .01) for contraceptive use. The relativemagnitude of these two correlations is prob-ably influenced by the extent to which eachbehavior is under volitional control. Whilecontraceptive use-nonuse is to a large degreeunder the control of the actor, childbearing isdetermined by an interaction of biological andvolitional events.

The sequence oj prior events. As discussedabove, in order to bear a child a woman musthave intercourse without using contraception,conceive, and not have a spontaneous or in-duced abortion. It was hypothesized that in-tention would provide a better estimate of theprobability that the person would initiate thesequence than of the likelihood that the per-son would complete the sequence and performthe behavior in question (bear a child). Wom-en were classified as initiating the sequence if,in the final survey, they responded that byhaving intercourse while not using contracep-tion they had been attempting for at least 9months to 'become pregnant or if they actuallygave birth during the 2-year period. Thepoint-biserial correlation of intention with

birth or attempted conception was .615 (p <.01). As hypothesized, this correlation was sig-nificantly larger than the correlation of inten-tion with birth, .615 versus .526, £(241) =4.45, p< .01.

The time interval between the assessment ojintention and behavior. It was hypothesizedthat as the time interval between the measure-ment of intention and behavior increases, theBI-B correlation decreases. This hypothesiswas tested by comparing the BI-B correlationfor the 2-year period (Year 1 to Year 3) withthe correlation for the 1-year period (Year 2to Year 3). In support of the hypothesis, forboth behaviors, the correlation for the 1-yearperiod was significantly higher than for the2-year period (oral contraceptive use: .853versus .678, z = 5.21, p < .01; birth or at-tempted conception: .829 vs. .615, z = 4.95,p < .Ol).3

Change in intention. The effect of time in-terval on intention-behavior consistency canbe attributed to processes occurring during thetime interval (e.g., exposure to new informa-tion and attitude change). To demonstratedirectly the effect of change on the BI-B cor-relation it was necessary to identify respon-dents who changed intentions between the firsttwo waves of the three-wave survey. However,as Kiesler (1977) has shown, a respondentmight change her intention to make it con-sistent with her behavior during the prioryear. For example, if a woman did not intendto have a child at the first interview but be-came pregnant during the first year of thestudy, she would probably indicate an inten-tion to have a child at the second interview.Such a case would more accurately be classi-fied as a missed prediction rather than as achange in intention. Hence, only respondentsnot performing the behavior (becoming preg-nant, bearing a child, using oral contracep-

3 As these correlations are themselves intercorre-lated, it is incorrect to use the traditional Fisher's ztransformation (Hayes, 1963, pp. 529-532) to test forthe significance of differences between them. Rather,the correlations were appropriately compared usinga test initially suggested by Pearson and Filon andreported in Peters and Van Voorhis (1940, p. 185).


tives) during the first year were eligible forthe classification of change in intention.

Eligible respondents were classified into oneof three groups based on their responses at thefirst and second interviews to the relevant in-tention items. They were classified as in-tenders if they indicated it was slightly, mod-erately, or very likely that they would per-form the behavior. Respondents indicatingthat performance of the behavior was slightly,moderately, or very unlikely were classifiedas nonintenders. Women checking the mid-point of the scale (neither likely nor unlikely)were classified as uncertain. Any eligible re-spondent who changed classification fromTime 1 to Time 2 was identified as havingchanged her intention.

Based on this procedure, 30 women wereidentified as having changed .their childbear-ing intentions, and the remaining 214 womenwere classified as nonchangers. For oral con-traceptive intentions, 36 women were classi-fied as changers and 206 as nonchangers. EachBI-B correlation was then computed sep-arately for the change and nonchange respon-dents. For both childbearing and contracep-tive use, the intention-behavior correlationwas substantially lower for women whochanged intentions than for nonchangers(birth or attempted conception: —.015 vs..742, z = 4,65, p < .01; oral contraceptiveuse: -.397 vs. .880, z = 9.57, p < .01). Thus,in support of the hypothesis, change in inten-tion attenuated the intention-behavior rela-tion over the 2-year period.

Respondent's educational level. It was hy-pothesized that the respondent's educationallevel would be positively related to the degreeof intention-behavior consistency. Hence in-tention should interact with education in theprediction of behavior. To test this hypoth-esis, behavior was regressed hierarchically(Allison, 1977; Cohen, 1978) against inten-tion, education and the multiplicative inter-action term of Intention X Education. Forneither behavior did the inclusion of the inter-action term lead to a significant increment inR2; both Fs < 1. Therefore the hypothesisthat educational level moderates the inten-tion-behavior relation was not supported.

In summary, there was a reasonably strongcorrelation between intentions and behaviorduring the subsequent 2 years. Consistentwith the hypotheses, however, this relationwas attenuated by (a) events in the behav-ioral sequence not under the volitional controlof the actor, (b) changes in intention, and(c) the time interval between the assessmentof intention and behavior. The respondent'seducation level did not moderate the inten-tion-behavior relation,

A demonstrated relation between what peo-ple say they are going to do (intention) andwhat they actually do (behavior) is probablyof greater applied than theoretical signif-icance. Once strong intention-behavior con-sistency has been observed, however, the rela-tion between the attitudinal and normativecomponents of the model and behavior is ofconsiderable theoretical interest. The subse-quent analyses examine this relation.

Prediction of Behavior From the Model'sA ttitudinal and Normative Components

The regression of behavior on attitudes andnorms. For both contraceptive use and child-bearing, the dichotomous index of behaviorduring the 2-year period was regressed onattitudes and norms, measured during the in-itial interview. As indicated in Rows 1 and 3of Table 1, for each behavior both of themodel's components received significant re-gression weights in the prediction of behavior.For the prediction of birth the multiple cor-relation was .508 (p < .01), and R2 adjustedfor shrinkage was .252. For the prediction oforal contraceptive use, the multiple correla-tion was .606 (p < .01) and R2 adjusted forshrinkage was .362.

The sequence of prior events. It was hy-pothesized that events in the behavioral se-quence not under the volitional control of theactor would attenuate the prediction of be-havior from the model's components. The datapresented in row 2 of Table 1 support thishypothesis. The multiple correlation withbirth or attempted conception is significantlyhigher than the correlation with birth, .595versus .508; t(241) = 4.33, p < .01.


Table 1Regression of Fertility and Contraceptive Behavior on Attitudes and Norms

Standardized regressioncoefficients

Behavior

Birth»'b

Birth or attemptedconception"'0

Use of oralcontraceptives'1'8

ZBiEt

.187

.203

Aact

.392

£ NBiMd

.361

.438

SN

.268

R

.508

.595

.606

adjustedfor

R2 shrinkage

.258 .252

.354 .349

.367 .362

Note. All regression coefficients and multiple correlations are significant (p < .01). Abbreviations are asfollows: £ Bi-Ei = one's beliefs about the consequences of performing the behavior multiplied by theevaluation of those consequences; £ NBiMd = one's normative beliefs about performing the behaviormultiplied by one's motivation to comply with those perceived norms; Aact = one's attitude towardperforming the behavior; SN = one's subjective norm (generalized normative belief) about performingthe behavior.• N = 244.b Coded; birth during the 2-year period = 1; no birth = 0.0 Coded: birth or attempted conception during the 2-year period = 1; no birth or attempted conception = 0.d N = 242.6 Coded: use of oral contraceptives during the 2-year period = 1; nonuse = 0.

Time interval between the assessment ofthe model's predictive components and behav-ior. For both behaviors, the multiple correla-tion with the model's predictive componentswas higher for the 1-year period (Year 2 toYear 3) than for the 2-year period (Year 1 toYear 3). However, for birth or attemptedconception the difference between the Rs,failed to achieve the traditional significancelevel (oral contraceptive use: .704 vs. .606,z = 2.48, p < .01; birth or attempted concep-tion: .649 vs. .595, z = 1.39, p < .09).

Attitudinal and normative change. As pre-viously noted, the effect of time interval onattitude-behavior consistency is attributable,in part, to attitude and belief change occur-ring during the interval. To demonstrate theeffect of change on the correlation betweenbehavior and the model's components, it wasnecessary to identify respondents whose atti-tudes or norms changed from Interview 1 toInterview 2. The procedure used here is sim-ilar to the one described above for assessingchange in BI. First, to differentiate norm andattitude change from post hoc justification ofprior behavior, only respondents not perform-ing the behavior (becoming pregnant, bearinga child, using oral contraceptives) during the

first year were eligible for the classificationof change in attitudes or norms. Next, sep-arately for each behavior and each of themodel's two predictive components, eligiblerespondents who changed from Time 1 toTime 2 were identified as changers, and theremaining respondents were classified as non-changers. The correlation between the rel-evant component and behavior was then com-puted separately in each group. Consistentwith the hypothesis, lower correlations wereexpected in the change than in the nonchangegroup.

Attitude toward the act of using oral con-traceptives (Aact) was assessed by summingthe responses to three evaluative scales, eachscored +3 to —3. A score of 0 was categorizedas a neutral attitude, a score greater than 0was a positive attitude, and less than 0 con-stituted a negative attitude. Sixty-three elig-ible respondents changed attitude categoriesbetween Interview 1 and Interview 2. As pre-dicted, the Aact-B correlation was substan-tially higher in the nonchange than in thechange group (.733 vs. -.238; z = 7.88, p <.01).

Responses to the general subjective normconcerning the use of oral contraceptives


(SN) were also divided into three categories:uncertain (a score of 4, the midpoint of thescale), positive norm (a score greater than 4),and negative norm (a score less than 4). Sev-enty-four eligible respondents changed cat-egories from Interview 1 to Interview 2. Insupport of the hypothesis, the SN-B correla-tion was significantly higher in the nonchangethan in the change group (.682 vs. —.039;2 = 6.14, p < .01).

For childbearing behavior, the midpoints ofthe scales for the 'S,BiEl and ^NBtMCi compo-nents had to be empirically identified. Usingdata from the first interview, the mean scoreson 2.Bt£« and 'S.NBiMCi were determinedseparately for those intending and not intend-ing to have a child. The score midway be-tween the means was designated the neutralpoint on the scale. In the interval betweenInterview 1 and Interview 2, 34 eligible re-spondents moved across the neutral point ofthe S.Bi-Ei measure, and 2 5 eligible respondentsmoved across the neutral point of the 2,NBt-MCt measure. Consistent with the prediction,the correlation of both ^,BiEi and ^NBiMdwith birth or attempted conception was higherin the nonchange than in the change group

.580 vs. .011, z = 3.38, p < .01;c. .648 vs. -.061, z = 3.71).

Respondent's educational level. It was hy-pothesized that the respondent's educationallevel would interact with attitudes and normsin the prediction of behavior. To test this hy-pothesis, behavior was regressed hierarchicallyagainst the model's attitudinal and normativecomponents, education, and the multiplicativeinteractions of Education x Each Component.For neither behavior did the inclusion of theinteraction terms lead to a significant incre-ment in R-; both Fs < 1. Hence, the hypoth-esis was not supported.

In summary, there was a reasonably strongcorrelation between the model's attitudinaland normative components and behavior dur-ing the subsequent 2 years. Similar to the re-sults obtained with behavioral intention, theprediction of behavior from attitudes andnorms was attenuated by (a) events in thebehavioral sequence not under the volitionalcontrol of the actor, (b) the time interval be-tween the measurement of the model's predic-

Table 2Correlations Between Selected AttitudinalVariables and Behavior

Attitudinal variable r

Use of birth control pills during the2-year period0

Attitude toward birth control .083Attitude toward birth control pills .323*Attitude toward using birth control pills .525*Attitude toward using birth control pills

during the next 2 years .572*

Birth or attempted conception during the2-year periodb

Attitude toward children —.007Attitude toward having children .187*Attitude toward having a child in the

next 2 years .535*

• N = 244.» N = 242.* p < .01.

live components and behavior (for childbear-ing there was only a trend in the hypothesizeddirection), and (c) changes in attitudes andnorms. The respondent's educational level didnot moderate the strength of the relation.

Correspondence Between Attitudinal andBehavioral Variables

It was hypothesized that the degree of cor-respondence between the elements of the at-titudinal and behavioral variables would bepositively related to attitude-behavior con-sistency. The present analysis investigated theeffect of three elements on the attitude-be-havior correlation: action, the target theaction is directed toward, and the time atwhich the action occurs. As indicated in Table2, four attitudinal variables, measured duringthe first interview, were correlated with thedichotomous behavioral measure of birth con-trol pill use during the 2-year period. The at-titudinal measures systematically varied withregard to the number of elements they had incorrespondence with the behavioral measure,as follows: zero elements in correspondence(attitude toward birth control, r = .083, ns);one element in correspondence—target (atti-tude toward birth control pills, r = .323, p <.01); two elements in correspondence—target


and action—(attitude toward using birth con-trol pills, r = .525, p < .01); three elementsin correspondence—target, action, and time—(attitude toward using birth control pills dur-ing the next 2 years, r — .572, p < .01). Insupport of the hypothesis, for each successiveincrease in correspondence between attitudeand behavior variables there was a significantincrease (ranging from p < .01 to p < .06) inthe magnitude of the attitude-behavior corre-lation.

Following a similar pattern, the dichoto-mous behavioral measure of birth or at-tempted conception during the 2-year periodcorrelated —.007 (ns) with attitude towardchildren, .187 (p < .01) with attitude towardhaving children, and .535 (p < .01) with atti-tude toward having a child in the next 2 years.For each successive increase in correspondencethere was a significant increase (all ps < .01)in the attitude-behavior correlation.

Discussion

The primary purpose of the present re-search was to investigate five factors hypoth-esized to influence the extent to which peopleact in accord with their stated attitudes. Theinvestigation of the first of these factors, thesequence of prior events, directed attentionto the behavioral variable. The majority ofprevious attitude-behavior research hasfocused exclusively on problems associatedwith attitudinal variables during attempts toaccount for low attitude-behavior correla-tions. However, the performance of a be-havior is often dependent upon the successfulcompletion of a series of behavioral events.As the present research has demonstrated, ifall of these steps are not under the volitionalcontrol of the actor, intention will provide abetter estimate of the probability that theactor will initiate the sequence than of thelikelihood that the actor will complete thesequence and perform the behavior in ques-tion. These results suggest that future studiesof attitude-behavior consistency should seekto identify and quantify the behavioral se-quences that must be completed prior to theperformance of the behavior of interest.Analyses can then isolate the sequences at

which serious disruptions occur between atti-tudes and actions. From an applied perspec-tive, such knowledge would be of considerableuse in the selection of behavior-change strat-egies. This approach would lead to the iden-tification of situations in which the most ap-propriate strategy is not to change attitudesbut rather to institute policy changes makingit easier for people to act in accord with theirexisting attitudes.

The second and third factors, attitudechange and the time interval between the mea-surement of attitude and behavior, focused oninstability in the attitude and belief variables.As the time interval decreased from 2 years to1 year, there was a 58.3% increase in theamount of variance in oral contraceptive useaccounted for by BI. The corresponding in-crease for childbearing was 81.7%. Substan-tial improvements in the prediction of be-havior were also achieved when analyses werelimited to the sample of respondents who didnot exhibit attitude change. The magnitudeof these effects directs attention to the prob-lem of using static models in the analysis ofa dynamic process. When stating attitudes andintentions about future behaviors, a respon-dent is probably assuming either that the en-vironment will not change or that it willchange in an anticipated manner. Unexpectedchanges (e.g., loss of employment, exposure tonew information) can lead to a change in atti-tude and, if such changes are not monitored,to an apparent lack of consistency betweenattitudes and behavior. Future researchshould attempt to specify the personal andsituational variables influencing attitudinalstability. From the studies reviewed above,some of the factors that might initially behypothesized to contribute to stability includethe extent to which the attitude was formedthrough direct behavioral experience with theattitude object (Fazio & Zanna, 1978; Regan& Fazio, 1977) and the individual differencevariables of self-monitoring (Snyder & Tanke,1976) and tendency to ascribe responsibilityto the self (Schwartz, 1973).

The fourth factor hypothesized to moderatethe attitude-behavior relation was the re-spondent's educational level. It was hypoth-esized that as completed years of schooling


increased, so would the magnitude of theattitude-behavior correlation. However, aninteraction between education and attitude-behavior consistency was not observed. Hence,these results lend no support to the contentionthat consistency is especially prevalent amongthe highly educated. It appears unlikely thatnonsupport for this hypothesis is attributableto lack of variance on the education variable;years of education ranged from 8 to 16. Inaddition, attitudinal models similar to thepresent one have provided good prediction ofbehavioral intentions and behavior in cross-cultural samples with considerably lower ed-ucational levels than the present sample(Davidson, Jaccard, Triandis, Morales, &Diaz-Guerrero, 1976; Davidson & Thomson,1980). It should be noted that the presentfinding provides indirect support for the theo-retical framework of Fishbein. As notedabove, if the framework is correct, soci'al,demographic, and personality characteristicsof the respondent should not have a directeffect on the magnitude of the relation be-tween the components of the model and in-tentions, and hence behavior.

The fifth factor hypothesized to moderateattitude-behavior consistency is the degree ofcorrespondence between the elements of theattitudinal and behavioral variables. As theattitudinal measures moved along a con-tinuum of correspondence in terms of thenumber of elements they had in agreementwith the behavioral measure (from zero tothree elements), their correlations with be-havior significantly increased. The presentresults concerning the impact of action andtarget correspondence are consistent with theconclusions of Ajzen and Fishbein (1977). Inaddition, the findings highlight the impact of athird element, not previously investigated, thetime during which the action occurs. To ob-tain reasonable predictions of behavior fromattitudinal variables, it appears important toensure correspondence in target, action, andtime elements.

The findings from the present study alsoprovide support for the predictive validity ofthe components of the model of Fishbein.Even prior to the adjustments made for atti-tude change and the sequence of events, both

intention and the attitudinal and normativecomponents measured at the first interviewprovided reasonable correlations with behaviorduring the subsequent 2 years ('the lowestvalidity coefficient was .508).

A weakness in the present research is thatthe data on oral contraceptive use were ob-tained solely through self-reports. Hence it isnot possible to determine if the respondentsmisrepresented their behavior in order tomake it consistent with their previously statedattitudes and intentions. It appears doubtfulthat there was extensive misrepresentation,because the findings concerning oral contra-ceptive use were very similar to those forchildbearing in terms of both the percentageof subjects correctly classified and the vari-ables found to moderate the degree of atti-tude-behavior consistency. There was also areliance on self-report data concerning theattempt to become pregnant. Respondentsmight have falsely indicated that they hadbeen unsuccessfully trying for at least 9months to become pregnant in an attempt toappear consistent with previously stated in-tentions. However, this rival hypothesis re-quires that the respondents admit somethingthat is socially undesirable about themselvesfor the purpose of appearing consistent.

In conclusion, this research indicates thatthe contraceptive and fertility behavior ofmarried women is quite predictable from be-havioral intentions, attitudes, and normativebeliefs. However, the degree of the relation isattenuated by events in the behavioral se-quence not under the volitional control of theactor, the time interval between the measure-ment of behavior and the model's predictivecomponents, changes in attitudes and beliefs,and the degree of correspondence between theattitudinal and behavioral variables.

References

Ajzen, I., & Fishbein, M. Attitudinal and normativevariables as predictors of specific behaviors. Jour-nal of Personality and Social Psychology, 1973,27, 41-57.

Ajzen, I., & Fishbein, M. Attitude-behavior rela-tions: A theoretical analysis and review of em-pirical research. Psychological Bulletin, 1977, 84,888-919.


Allison, D. Testing for interaction in multiple regres-sion. American Journal of Sociology, 1977, 83, 144-153.

Beach, L., Townes, B., Campbell, F., & Keating, G.Developing and testing a decision aid for birthplanning decisions. Organizational Behavior andHuman Performance, 1976, IS, 99-116.

Bern, D. Beliefs, attitudes, and human affairs. Bel-mont, Calif.: Brooks/Cole, 1970.

Cohen, J. Partialed products are interactions: Par-tialed powers are curve components. PsychologicalBulletin, 1978, 85, 858-866.

Davidson, A., & Jaccard, J. Population psychology:A new look at an old problem. Journal of Per-sonality and Social Psychology, 1975, 31, 1073-1082.

Davidson, A., and Jaccard, J. Determinants of fer-tility intentions. In S. Newman & V. Thompson(Eds.), Population psychology: Research and ed-ucational issues. Washington, D.C.: U.S. Govern-ment Printing Office, 1976.

Davidson, A., Jaccard, J., Triandis, H., Morales,M. L., & Diaz-Guerrero, R. Cross-cultural modeltesting: Toward a solution of the etic-emic dilem-ma. International Journal of Psychology, 1976, 11,1-13.

Davidson, A., & Thomson, E. Cross-cultural studiesof attitudes and beliefs. In H. Triandis & R. Brislin(Eds.), Handbook of cross-cultural psychology(Vol. 5: Social Psychology). New York: Allyn &Bacon, 1980.

Deutscher, I. Words and deeds. Social Problems, 1966,13, 235-254.

Deutscher, I. Looking backward: Case studies on theprogress of methodology in sociological research.American Sociologist, 1969, 4, 35-41.

Ehrlich, H. Attitudes, behavior and the interveningvariables. American Sociologist, 1969, 4, 29-34.

Fazio, R., & Zanna, M. Attitudinal qualities relatingto the strength of the attitude-behavior relation-ship. Journal of Experimental Social Psychology,1978, 14, 398-408.

Fazio, R., Zanna, M., & Cooper, J. Direct experienceand attitude-behavior consistency: An informationprocessing analysis. Personality and Social Psychol-ogy Bulletin, 1978, 4, 48-51.

Fishbein, M. An investigation of the relationshipbetween beliefs about an object and attitudetoward that object. Human Relations, 1963, 16,233-240.

Fishbein, M. Attitude and the prediction of behavior.In M. Fishbein (Ed.), Readings in attitude theoryand measurement. New York: Wiley, 1967.

Fishbein, M., & Ajzen, I. Attitudes toward objectsas predictors of single and multiple behavioralcriteria. Psychological Review, 1974, 81, 59-74.

Fishbein, M., & Ajzen, I. Belief, attitude, intentionand behavior. Reading, Mass.: Addison-Wesley,1975.

Fishbein, M., & Jaccard, J. Theoretical and method-ological considerations in the prediction of family

intentions and behavior. Representative Research inSocial Psychology, 1973, 4, 37-51.

Hayes, W. Statistics. New York: Holt, Rinehart &Winston, 1963.

Heberlein, T., & Black, J. Attitudinal specificity andthe prediction of behavior in a field setting. Journalof Personality and Social Psychology, 1976, 33,474-479.

Kelley, S., & Mirer, P. W. The simple act of voting.American Political Science Review, 1974, 68, 572-591.

Kelman, H. Attitudes are alive and well and gainfullyemployed in the sphere of action. American Psy-chologist, 1974, 29, 310-324.

Kiesler, S. Post-hoc justification of family size.Sociometry, 1977, 40, 59-67.

McGuire, W. The nature of attitudes and attitudechange. In G. Lindzey & E. Aronson (Eds.), Thehandbook of social psychology (2nd ed., Vol. 3).Reading, Mass.: Addison-Wesley, 1969.

Norman, R. Affective-cognitive consistency, attitudes,conformity, and behavior. Journal of Personalityand Social Psychology, 1975, 32, 83-91.

Peak, H. Attitude and motivation. Nebraska Sym-posium on Motivation (Vol. 3). Lincoln: Uni-versity of Nebraska Press, 1955.

Peters, C., & Van Voorhis, W. Statistical proceduresand their mathematical bases. New York: McGraw-Hill, 1940.

Regan, D., & Fazio, R. On the consistency betweenattitudes and behavior: Look to the method ofattitude formation. Journal of Experimental SocialPsychology, 1977, 13, 2&-4S.

Rindfuss, R., & Sweet, J. Postwar fertility trends anddifferentials in the United States. New York: Aca-demic Press, 1977.

Rokeach, M. The nature of human values. NewYork: Free Press, 1973.

Rosenberg, M. Cognitive structure and attitudinalaffect. Journal of Abnormal and Social Psychology,1956, S3, 367-372.

Schuman, H., & Johnson, M. Attitudes and behavior.Annual Review of Sociology, 1976, 2, 161-207.

Schwartz, S. Normative explanations of helping be-havior: A critique, proposal and empirical test.Journal of Experimental Social Psychology, 1973,9, 349-364.

Schwartz, S. Temporal instability as a moderator ofthe attitude-behavior relationship. Journal of Per-sonality and Social Psychology, 1978, 36, 715-724.

Snyder, M., & Swann, W. When actions reflect atti-tudes: The politics of impression management.Journal of Personality and Social Psychology, 1976,34, 1034-1042.

Snyder, M., & Tanke, E. Behavior and attitudes:Some people are more consistent than others. Jour-nal of Personality, 1976, 44, 501-517.

Warner, L., & DeFleur, M. Attitude as an inter-actional concept: Social constraint and social dis-tance as intervening variables between attitudesand action. American Sociological Review, 1969,34, 153-169.


Wcigel, R., & Newman, L. Increasing attitude-be-havior correspondence by broadening the scope ofthe behavioral measure. Journal of Personality andSocial Psychology, 1976, 33, 793-802.

Weigel, R., Vernon, D., & Tognacci, L. The specificityof the attitude as a determinant of attitude-be-havior congruence. Journal of Personality and So-cial Psychology, 1974, 30, 724-728.

Weinstein, A. Predicting behavior from attitudes.Public Opinion Quarterly, 1972, 36, 355-360.

Werner, P., Middlestadt-Carter, S., & Crawford, T.Having a third child: Predicting behavioral inten-tions. Journal of Marriage and the Family, 1975,37, 348-358.

Westoff, C., Potter, R., Sagi, P., & Mishler, E. Family

growth in metropolitan America. Princeton, N.J.:Princeton University Press, 1961.

Whelpton, P. K., & Kiser, C. V. (Eds.). Social andpsychological factors affecting fertility (5 vols.).New York: Milbank Memorial Fund, 1946-1958.

Wicker, A. Attitudes versus actions: The relationshipof verbal and overt behavioral responses to attitudeobjects. Journal of Social Issues, 1969, 25, 41-78.

Wicker, A., & Pomazal, R. The relationship betweenattitudes and behavior as a function of specificityof attitude object and presence of a significant per-son during assessment conditions. RepresentativeResearch in Social Psychology, 1971, 2, 26-31.

Received August 1, 1978 •

Documents

Variables That Moderate the Attitude-Behavior Relation ...fsnagle.org/files/davidson1979variables.pdf · 2. Attitude change. In a recent review of the attitude-behavior literature,