improving the academic performance of college freshmen

Journal of Personality and Social Psychology1982, Vol. 42, No. 2, 367-376

Copyright 1982 by the American Psychological Association, Inc.0022.3514/82/4202-0367$00.75

Improving the Academic Performance of College Freshmen:Attribution Therapy Revisited

Timothy D. WilsonUniversity of Virginia

Patricia W. LinvilleCarnegie-Mellon University

An attributional intervention was devised to help college freshmen who wereconcerned about their academic performance. Unlike most previous attributiontherapy attempts, an effort was made to change subjects' attributions for theirproblems from stable to unstable causes, rather than from internal to externalcauses. Freshmen were given information indicating that on the average, collegestudents improve their grades from the freshman to the upperclass years, plusthey were shown videotaped interviews of upperclassmen who reported that theirgrade point averages (GPA) had improved since their freshman year. The effectof this GPA information was dramatic. Subjaf ts who received the informationas compared to subjects who did not: (a) were significantly less apt to leavecollege by the end of the sophomore year, (b) had a significantly greater increasein GPA 1 year after the study, and (c) performed significantly better on sampleitems from the Graduate Record Exam. As in many self-attribution studies, theself-report evidence for the cognitive processes mediating these behavioralchanges was weak. None of the self-report measures of attitudes, expectancies,or mood correlated with the behavioral results. In addition, the GPA informationhad no effect on self-reports of mood. A more positive mood was reported onlyby subjects who performed a reasons analysis (i.e., who were asked to list reasonswhy their grades might improve). This divergent pattern of behavioral and self-report results is discussed in terms of the hypothesis that the determinants ofbehavioral results differ from the determinants of self-report results in self-at-tribution studies.

There have been many attempts to useprinciples of attribution theory as a thera-peutic technique. Attribution therapy hastypically been used with people who behavein a dysfunctional manner and attributetheir problems to pejorative internal causes.These people are encouraged to reattributetheir problems to an external cause, with theassumption that this will prevent additionalworrying and anxiety that exacerbate theproblem (Storms & McCaul, 1976; Valins& Nisbett, 1972). For example, insomniacsmight assume that their sleeping problem isdue to a deep-seated neurosis. This attri-

This research was supported by National ScienceFoundation Grant BNS79-21155. The authors wouldlike to thank Harry DeMik, Barbara Fawcett, and Kath-ryn Berkovsky for their assistance. Margaret Clark,Gregory Fischer, Susan Fiske, and Irene Frieze providedvaluable critiques of an earlier draft.

Requests for reprints should be sent to Timothy D.Wilson, Department of Psychology, Gilmer Hall, Uni-versity of Virginia, Charlottesville, Virginia 22901.

bution to an internal cause can lead to ad-ditional worrying and anxiety, which in turnmakes it even more difficult to get to sleep.Encouraging such people to attribute theirinsomnia to an external, nonpejorative causemight alleviate the insomnia by reducinganxiety about their mental health. Stormsand Nisbett (1970) found such results. In-somniacs who were led to attribute their anx-iety and arousal at bedtime to an externalsource (a placebo pill) reported getting tosleep quicker than did control subjects, pre-sumably because the former subjects wereno longer caught in an exacerbation cycle.

Though case reports of clinical successescorroborate the effectiveness of attributiontherapy (cf. Storms & McCaul, 1976; Valins& Nisbett, 1972), the number of successfulexperimental demonstrations is small. TheStorms and Nisbett (1970) study, the best-known source of experimental support, hasbeen embroiled in controversy regarding itsgeneralizability (Boot/in, Herman, & Ni-

367

368 TIMOTHY D. WILSON AND PATRICIA W. LINVILLE

cassio, 1976; Kellogg & Baron, 1975; Low-ery, Denney, & Storms, 1979). Attempts touse attribution therapy with other types ofproblems—notably depression—have notbeen successful (e.g., Nisbett, Borgida,Crandall, & Reed, 1976). At best, it can beconcluded that the effectiveness of attribu-tion therapy is far from proven.

The purpose of the present research wasto demonstrate that attribution therapy canbe successful when certain recent findingsin social cognition are taken into account.First, unlike most previous studies, we at-tempted to change attributions from stableto unstable causes rather than from internalto external causes. A major assumption ofprevious research has been that it is mostbeneficial to change attributions for dys-functional behavior from pejorative internalcauses to external factors. It may be morebeneficial, however, to get people to attributetheir problems to temporary factors that areapt to change rather than to permanent, un-changeable causes.

For example, it would be of little solaceto tell insomniacs that their sleeping prob-lems were due to the sound of traffic outsidetheir windows, even though this is an exter-nal cause, because it is a stable cause thatis difficult to alter. It might be more bene-ficial to convince them that their insomniais due to a temporary illness such as a cold.Although this is an internal cause, its tem-porary nature might prevent people fromexacerbating their problem by worryingabout its permanence. We are not denyingthat people who attribute their problem topejorative internal causes may worry morethan those who do not (cf. Storms, Denney,McCaul, & Lowery, 1979). The point is thatpeople are often more concerned with theprognosis of their problem than with whetherits cause is internal.

The argument that it might help peopleto attribute their problems to unstable causesis similar to Weiner's (1974) discussion ofattributions about the causes of success andfailure. Weiner argues that the stability/in-stability dimension of causal attribution isorthogonal to the internal/external dimen-sion, and that the former type of attributionis critical in determining a person's expec-tations about future performance, affect (cf.

Weiner, 1980; Weiner, Russell, & Lerman,1978), and actual task performance. Peoplewho have done poorly on a task and whoattribute this failure to enduring, stable fac-tors such as a lack of ability or the insur-mountable difficulty of the task, should ex-pect to do poorly in the future, shouldexperience negative affect, and should per-form poorly on the task. Encouraging peopleto attribute their poor performance to tem-porary causes should increase expectationsabout future performance, reduce anxietyand feelings of helplessness, and lead to bet-ter task performance.

We chose as our target group collegefreshmen who were concerned about theiracademic performance. This group of peoplewould seem to be particularly susceptible todamaging attributions about the perma-nence of their problems. Most freshmen en-ter college with trepidation about whetherthey can handle the work, and many aredismayed to discover that the amount ofstudying required far exceeds that neededto do well in high school. When academicproblems first occur, as they do with manyfreshmen, students may see this as confirm-ing their worst fears about their inability tosucceed at college. Such attributions maycause additional worrying and anxiety, mak-ing it even more difficult to study. Thus, con-vincing freshmen that their problems arecaused by temporary rather than permanentinhibitory factors may well have beneficialeffects.

There have been very few studies thathave both encouraged people to attributetheir problems to unstable factors and ex-amined the effects of these attributions onsubsequent behavior. The few studies thathave accomplished this have found encour-aging behavioral results (e.g., Andrews &Debus, 1978; Dweck, 1975). However, thesestudies usually involved teaching people toattribute failure specifically to a lack of ef-fort. We wished to avoid such a direct at-tributional manipulation, since attributingone's failure to a lack of effort can lead toadditional negative affects such as guilt andshame (Weiner et al., 1978). We simplypointed out to subjects that many peopleexperience academic problems as freshmen,but do better in the upperclass years (i.e.,

ATTRIBUTION THERAPY REVISITED 369

in Kelley's [1967] terms we presented themwith both consensus and consistency infor-mation about academic problems). Subjectscould thus infer that their own problemswere due to unstable causes. The unstablecauses they used to explain their difficultieswere up to them, and could include such fac-tors as having to learn the ropes in a newenvironment or temporary homesickness.

Attributing one's problems to such factorsshould lead to increased expectations aboutfuture performance and enhanced affect(i.e., to a sense of relief that the problemsare not permanent). These increased expec-tations and positive affects should in turnlead to less anxiety about academic tasks,increased effort, and hence better perfor-mance. The dependent measures were thussubjects' expectations about their future per-formance, their self-reported affects, andtheir actual,performance.

Self-Report Versus Behavioral DependentVariables

Though both self-report and behavioraldependent variables were included in thedesign, there is reason to believe that changesin attributions and affect are more apt to bedetected by behavioral measures. It is verycommon in self-attribution research for be-havioral evidence for attributional changesto occur in the absence of self-reportedchanges in the internal states (e.g., attitudes,affects, traits) mediating the behavior (cf.Nisbett & Wilson, 1977; Wilson/Hull, &Johnson, 1981). Because people have im-perfect access to their cognitive processesand internal states, attributional changes aremore apt to be detected by behavioral mea-sures. Even when self-report effects arefound they are often weaker than the be-havioral effects, and the self-report measuresrarely correlate very highly with the behav-ioral measures.

For this reason, unlike most other attri-bution therapy studies, the chief dependentmeasures were behavioral: subjects' perfor-mance on sample items from the GraduateRecord Exam (GRE), whether or not theywere still enrolled in college 1 year after thestudy, and their actual performance in theircourses following the study. It was predicted

that subjects who received the informationindicating that academic problems in thefreshman year are temporary would performbetter on the GRE items, be more likely tostay in college, and perform better in theircourses than would subjects who did not re-ceive this information. Because of the un-reliability of self-report effects in previousstudies, our predictions about the self-reportmeasures were much less firm. It is not clearwhen self-report effects that are consistentwith behavioral effects will occur in self-at-tribution studies.

Wilson et al. (1981) have provided somepreliminary answers about when self-reporteffects will occur. They argued that attri-butional processes can mediate changes ininternal states that lead to behavioralchanges, but since people have poor accessto these processes and states, a verbal ex-planatory system that attempts to infer thenature of cognitive processes and internalstates also operates. If this model is correct,then it should be possible to manipulate theverbal explanatory system independently ofthe processes mediating behavior. Wilson etal. (1981) accomplished this in two self-at-tribution studies. For example, in an over-justification study some subjects were re-warded for playing with an interesting puzzleand some were not. Cross-cutting the rewardmanipulation, some subjects were asked toconsider various reasons why they were play-ing with the puzzle. Subjects in the rewardcondition subsequently played with the puz-zle less in a free-time period, regardless ofwhether they had reflected on why they ini-tially played with it. That is, rewarded sub-jects behaved as if they liked the puzzle lessthan did nonrewarded subjects. Self-re-ported differences in liking for the puzzle,however, were found only with subjects whohad reflected on why they had initiallyplayed with the puzzle (i.e., only when theverbal explanatory system was primed by thereasons manipulation).

It thus appears that self-reports of internalstates occur under different conditions thanbehavioral effects, possibly because self-re-ports and behavior are generated by inde-pendent means. Since it is important in thepresent context to improve not only perfor-mance but self-reports as well, an explora-


tory reasons analysis manipulation was in-cluded in the design. Cross-cutting theattribution manipulation, half of the subjectswere asked to list reasons why they mightimprove their grades in their upperclassyears. In a sense this was an additional at-tribution manipulation, because it forcedsubjects to consider why their problemsmight be due to temporary causes. FollowingWilson et al. (1981), however, it was pre-dicted that this verbal attribution manipu-lation would influence only the self-reportdependent measures, by priming,the verbalexplanatory system. Subjects who were in-duced to think about why their grades mightimprove were expected to report more pos-itive moods. Since this explanatory systemis hypothesized to be relatively independentof those cognitive processes mediating be-havior, the reasons analysis manipulationwas not expected to influence the behavioralmeasures of performance.

Method

Overview

Subjects were college freshmen who were concernedabout their academic performance. The experiment em-ployed a 2 (grade point average [GPA] information, noinformation) X 2 (reasons analysis, no reasons analysis)design. Subjects in the GPA information condition wereshown statistical data and videotaped interviews withupperclassmen indicating that most freshmen improvetheir GPA when they become upperclassmen. Subjectsin the reasons analysis condition were asked to list rea-sons why freshmen might improve their GPA and toindicate the factors that currently affected them. It waspredicted that the GPA information would improve theirperformance on an academic-type test, reduce the num-ber of students who left college in the following year,and improve course performance. The reasons analysismanipulation, by stimulating the verbal explanatory sys-tem, was expected to produce improved self-reports ofmood but to have no influence on behavior.

Subjects

At the beginning of the second semester of the aca-demic year, all students in the psychology departmentsubject pool at Duke University were given a question-naire asking for their GPA for the previous semester,the extent to which they worried about their academicperformance, how they thought they compared intellec-tually to their peers, and whether or not they thoughtthey had done as well as they could have in their firstsemester courses. Subjects were selected only if theywere second-semester freshmen who met all of the fol-lowing criteria: (a) Their response to the question about

worrying was equal to or above the median, (b) theirfirst semester GPA was less than or equal to 3.50, (c)they said they had not done as well as they could havein their first semester courses, and (d) they did not saythey compared intellectually much better than averageto their classmates. Of the 200 freshmen in the subjectpool, 71 (22 males, 49 females) met all of these criteria.Of these 71 freshmen, 40 were contacted several weeksafter filling out the initial questionnaire and scheduledto participate in the study. No connection was madebetween the study and the initial questionnaire. Threemales and seven females were randomly assigned to eachof the four cells of the design.

Procedure

Subjects were seen individually in a small conferenceroom. The experimenter explained to all subjects thatshe and a faculty member were conducting a targe-scalesurvey of students about their college experiences. Datahad already been collected from upperclassmen, sheexplained, and now they were beginning a survey offreshmen. Subjects were told that they were scheduledindividually because "we like to begin by interviewingpeople in depth before we conduct a mass survey." Sub-jects were not told how or on what basis they had beenselected for the study nor were they told that its purposewas to improve their academic performance. All subjectsthought they were participating in a survey.

OP A information manipulation. Half of the subjectswere randomly assigned to the GPA information con-dition. The experimenter explained to these studentsthat she wanted to familiarize them with the type ofquestions asked and the results of the survey of upper-classmen, in order to "give you a better idea of the kindsof questions we are asking, and to get your reactions towhat the upperclassmen said." She instructed them toattend to the information carefully because they wouldlater be questioned about it. The subjects first receiveda booklet containing a page of results from the surveyof upperclassmen. These results, which were selectedfrom an actual survey conducted the summer prior tothe experiment, conveyed the fact that many studentshave problems as freshmen, but that these problems getbetter (e.g., "67% said their freshman grades were lowerthan they had anticipated; 62% of the students said theirGPA had improved significantly from the first semesterof their freshman year to their upperclass years, and51% said their GPA had improved from the first se-mester of their freshman year to their sophomore year").To make this statistical data more vivid and concrete(cf. Nisbett & Borgida, 1975), GPA information sub-jects were also shown videotaped interviews of upper-classmen who reported that their GPAs had increasedsince their freshman year. The tapes were of actual in-terviews with Duke upperclassmen relating accuratepersonal information (except for GPA), thus resultingin authentic and believable presentations. The formatof each interview was identical. First the students (twojuniors, two seniors; two males, two females) reportedtheir GPAs for the first semester of their freshman year,the second semester of their freshman year, and for thesemester they had just completed. These GPAs wereselected so that each respondent reported a steady in-


crease in GPA over time. For example, one person saidhis GPA went from a 2.0 to a 2.6 to a 3.2. To increasethe plausibility of the cover story that this was a generalsurvey, each interviewee was also asked about his or hermajor, career plans, hometown, leisure-time activities,and favorite books and movies. In addition, subjects inthe GPA information condition saw a brief videotapeof a psychology professor discussing other results of thesurvey of upperclassmen. These results were filler itemsconsisting of attitudes about social issues and were un-related to academic performance.

Subjects in the no-information condition neither re-ceived any of the results of the upperclassmen surveynor viewed any videotapes. After receiving the generalintroduction to the study, they were randomly assignedto either the reasons analysis or no reasons analysis con-dition.

Reasons analysis manipulation. Half of the subjectsin both the GPA and no-information conditions wererandomly assigned to the reasons analysis condition.Just prior to filling out the dependent measures, thesesubjects were given a questionnaire that asked them tolist all the reasons they could think of why some studentsimprove their grades from their freshman to their up-perclass years, and to indicate which of these reasonsapplied to them at that time (i.e., which ones helpedexplain "any academic difficulties you might be havingthis year"). Subjects in the no-reasons analysis conditionwere given a questionnaire asking them to write abouta filler question, namely to list all the reasons they couldthink of why the divorce rate is declining in some states.The experimenter was blind to the reasons analysis con-dition of the subject.

Dependent Measures

To examine both the immediate and the long-termeffects of the manipulations, the dependent measureswere assessed at several points. At the first session, sub-jects filled out a questionnaire that contained itemsabout their attitudes toward their performance, theirexpectations about their future performance, and theirmood. They also completed two short-term behavioralmeasures. All subjects returned 1 week later and com-pleted the same questionnaire and the short-term be-havioral measures a second time. Finally, two long-termbehavioral measures were assessed: whether subjectsremained in college up to 1 year following the study andtheir performance in their courses in the year followingthe study. Each of these measures is described in detailbelow.

Attitudes toward performance. Four items wereasked about subjects' attitudes toward the freshmanyear and toward their own performance. The first ques-tion, which was a check on the GPA information ma-nipulation, asked how difficult they felt it was for the"'average' freshman to adjust to the academic and socialdemands at Duke." The next question asked how pleasedsubjects were with the way their freshman year wasgoing, while the second and third were identical to twoof the pretest questions used for subject selection;namely, how much subjects worried about their aca-demic performance and how they felt they compared"academically and intellectually" to their peers. All

questions were answered on 7-point scales with appro-priate labels at the endpoints.

Expectations about future performance. Subjectswere asked what they thought their GPA would be forthe current semester, the following semester, and whenthey graduated.

Mood scales. Four mood scales were taken fromWessman and Ricks (1966). These scales asked how theperson felt "right now" on four dimensions: Elation ver-sus depression, self-confidence versus feelings of inad-equacy, alertness versus dullness of thought, and tran-quility versus anxiety. For each dimension, subjectscircled 1 of 10 statements that best described theirmood.

Short-term behavioral measures. After finishing thequestionnaire, subjects completed two behavioral mea-sures of performance. First, subjects answered questionstaken from a study book for the GRE (Brownstein &M. Weiner, 1978). They were given 2 minutes to reada brief paragraph, after which they completed six mul-tiple-choice questions testing their reading comprehen-sion. Finally, subjects were given 4 minutes to solve 12anagrams, all names of animals.

All subjects returned 1 week later and completed thedependent measures a second time. The questions fromthe GRE and the anagrams were different from the onesthat subjects answered at Week 1.'

Long-term behavioral measures. A few weeks afterthe second session, a brief description of the study wassent to all subjects, along with an appeal to sign andreturn a consent form permitting us to examine theiracademic transcripts. Thirty-six of the 40 subjects(90%) signed and returned the consent form. The GPAswere computed for these subjects for four different se-mesters: the first semester of their freshman year (abaseline measure, since this was before the study wasconducted), the second semester of their freshman year(the time at which the study was conducted), the firstsemester of their sophomore year, and the second se-mester of their sophomore year. In addition, a recordwas kept of the percentage of all the subjects in eachcondition who left college up to 1 year after the studywas completed.

Results

Our main prediction was that the GPAinformation would significantly increase per-formance on the behavioral measures. Thisprediction was confirmed on the GRE per-

1 The questionnaire that subjects filled out at Week1 and Week 2 also included several "behavioroid" mea-sures (e.g., a question asking subjects how many classesthey cut during the previous week). In addition, at theend of Week 2, subjects were asked if they were doingas well as they should in all of their classes, the extentto which their academic problems were due to unstableand stable factors, and to estimate the percentage ofstudents at Duke who increase their GPA from theirfreshman to their upperclass years. These measures didnot yield useful information and thus will not be dis-cussed further.


formance measure as well as the long-termbehavioral measures.

Short-Term Behavioral Effects

The GPA information improved perfor-mance on the sample GRE items. Each sub-ject was assigned a score from 0 to 6 ac-cording to the number of reading com-prehension questions answered correctly ateach session. A 2 (GPA information) X 2(reasons analysis) X 2 (Week 1 versus Week2) bet ween-within analysis of variance (AN-OVA) revealed a significant main effect ofthe GPA information, F(l, 36) = 4.48,p < .05, reflecting the fact that subjects inthe GPA information condition answeredmore of the questions correctly than did sub-jects in the no-information condition (cf.Table 1). No other effects were significant.Subjects who received the GPA informationanswered 70% of the GRE items correctlyon the average, whereas those in the no-in-formation condition answered only 58% cor-rectly. There were no effects of the manip-ulations on the number of anagrams solved.This may have been due to a ceiling effect,since subjects solved an average of 9.0 of the12 anagrams.

Long-Term Behavioral Measures

The effects of the GPA information on thelong-term behavioral measures were dra-matic. As seen in Table 1, 25% (5 of 20) ofthe students in the no-information conditionhad left college by the end of their sopho-

Table 1Behavioral Results

Condition ORE* Dropouts" GPAC

No informationGPA information

3.504.18

255

-.05.34

Note. GRE = Graduate Record Examination; GPA =grade point average." Average number of sample GRE questions answeredcorrectly, averaged over Weeks 1 and 2.b Percent no longer enrolled as of the second semesterof the sophomore year.0 Average increase in GPA between the second semesterof the sophomore year and the first semester of the fresh-man year. These figures do not include those whodropped out by the second semester of the sophomoreyear.

more year, but only 5% (1 of 20) had leftcollege in the GPA information condition.The difference between these percentages issignificant (z = 1.89; p = .059, two-tailed)using the arcsin transformation for propor-tions (cf. Langer & Abelson, 1972). Thus,the GPA information reduced the percent-age of subjects who left Duke by 80%. Thereasons analysis manipulation had no de-tectable effect on the the dropout rates.

Furthermore, of those students who re-mained at Duke, those in the GPA infor-mation condition improved their grades more1 year following the study. There were nodifferences between the GPA and no-infor-mation conditions on the baseline measure(i.e., their GPAs for the first semester of thefreshman year were not significantly differ-ent: Ms = 2.58 and 2.87 for the two con-ditions, respectively, p> .10).2 Thereforethree GPA increase scores were computedby subtracting each subject's baseline GPAfrom his or her GPA for (a) the second se-mester of the freshman year, the time atwhich the study was conducted; (b) the firstsemester of the sophomore year; and (c) thesecond semester of the sophomore year. Bythe end of the second semester, freshmanyear subjects in the GPA information con-dition had improved their grades slightly (Mincrease = .11), whereas subjects in the no-information condition showed a slight dropin their grades (M = -.14). These increaseswere not significantly different, which is notsurprising since the study was not conducteduntil midway through the second semesterof the freshman year. Subjects in both con-ditions received better grades in the first se-mester of their sophomore year as comparedto the first semester of their freshman year.The increase in the GPA information con-dition was larger (M = .35) than it was inthe no-information condition (M = .17), butagain this difference was nonsignificant.

By the second semester of their sophomoreyear, subjects in the GPA information con-

2 These mean first semester GPAs were computedonly on those students who stayed in college throughthe second semester of their sophomore year and whogave us permission to examine their transcripts (n = 30).The mean first semester GPA for all students who gaveus permission to examine their transcripts (« = 36) was2.63 in the GPA information condition and 2.82 in theno-information condition, p = .25.


dition obtained GPA increases that were sig-nificantly higher than they were in the no-information condition, Ms = .34 and -.05,respectively, F(i, 26) = 4.27,p < .05. Thus,1 year after the completion of the study, sub-jects who had received information indicat-ing that grades increase after the freshmanyear did injact increase their grades morethan did Subjects who did not receive thisinformation. The reasons analysis manipu-lation had no effect on any of the GPA in-creases.

It should be noted that the increases inGPA are conservative estimates of the dif-ference between the information conditions.These calculations do not include those sub-jects who left college by the end of the soph-omore year, and more subjects left in the no-information condition. If it is assumed thatstudents who left would not have performedas well in the sophomore year as those whostayed, then the difference between the con-ditions in GPA increases is smaller than itwould be had all students remained in col-lege.3

Self-Report Results

Expectations. Two measures were cal-culated to reflect subjects' expectations abouttheir future academic improvement. Short-term expectations were computed by takingthe difference between subjects' predictedGPA for the current semester and for thefollowing semester (first semester of thesophomore year). Long-term expectationswere computed by taking the difference be-tween subjects' predicted GPA for the cur-rent semester and their predicted GPA atgraduation. 2 (GPA information) X 2 (rea-sons analysis) X 2 (time at which the ques-tions were answered; i.e., Week 1 vs. Week2) ANOVAS were performed on both mea-sures. There were no significant effects ofthe manipulations or of time on the short-term predictions. There was, however, a sig-nificant main effect of the GPA informationon long-term expectations, F(l, 34) = 5.20,p < .05. Over the long run, subjects in theGPA information condition expected to im-prove their GPA more than did subjects inthe no-information condition (Ms = .45 and.24, respectively).

Mood results. It was predicted that sub-

jects in the reasons analysis condition wouldreport a more positive mood, since they hadbeen induced to think about why their gradesmight improve. This prediction was sup-ported by the mood results, at least at Week1. An overall mood index was formed bysumming each subject's score on the fourindividual mood scales. A 2 (GPA infor-mation) X 2 (reasons analysis) X 2 (Week1 versus Week 2) between-within ANOVArevealed a significant Reasons Analysis XTime interaction, F(l, 36) = 4.47, p < .05.This interaction reflected the fact that atWeek 1, subjects in the reasons analysis con-dition reported a significantly more positivemood than did subjects in the no-reasonsanalysis condition, but that there were nodifferences between the conditions at Week2 (cf. Table 2). Separate 2 (GPA informa-tion) X 2 (reasons analyis) ANOVAS on themood index at Week 1 and at Week 2 re-vealed a strong main effect of the reasonsanalysis manipulation at Week 1, F(l,

3 A possible alternative explanation for the behavioralresults is that random assignment failed (i.e., that thosein the GPA information condition were initially brighterthan those in the no-information condition). An ex-amination of the students' Scholastic Aptitude Test(SAT) scores (obtained from their transcripts) and theirGPAs from the first semester of their freshman yearindicates that this was not the case. There were no sig-nificant differences between the conditions on thesemeasures. Furthermore, analyses of covariance on thebehavioral measures (sample GRE items, dropout rates,GPA increases 1 year after the study), using SAT scoresas covariates, left the results virtually unchanged. Theseanalyses revealed that (a) in no case did the effects ofthe manipulations interact significantly with SAT scores(as computed with regression techniques, cf. Nie et al.,1975, pp. 381-383); (b) in no case did SAT scores cor-relate significantly with the behavioral measures, withthe exception of the correlation between verbal SATscores and sample GRE items, K34) = .45, p < .01; and(c) after adjusting for SAT scores, the effects of theGPA information on the behavioral results were virtu-ally identical to those reported above. We should men-tion, however, one other alternative explanation for theincrease in grades in the GPA information condition.Since subjects in this condition started out at a lowerpoint than did no-information subjects (i.e., their first-semester freshman-year GPA was lower), the fact thatthey increased their grades, but no-information subjectsdid not, could simply reflect regression to the mean. Thisalternative seems unlikely, since (a) the difference infirst semester GPAs was not significant and (b) the firstsemester GPA in the no-information condition was notvery high (M = 2.87), allowing ample room for an in-crease. However, this interpretation can not be entirelyruled out.


36) = 10.08, p = .003, but no effect of thereasons analysis manipulation at Week 2,F(l, 36) < 1, ns. There were no significantmain effects or interactions involving theGPA information manipulation on any ofthese ANOVAS. Thus, the reasons analysismanipulation improved the subjects' moodat Week 1, but this effect had worn off byWeek 2.

Other questionnaire responses. Subjectsanswered four questions about their .atti-tudes toward their freshman year and theiracademic performance. The first questionasked how difficult they thought it was forthe average freshman to adjust to collegelife. A 2 X 2 X 2 between-within ANOVA re-vealed that subjects who received the GPAinformation thought that it was more diffi-cult for freshmen to adjust than did thosewho did not receive this information, Ms =3.08 and 3.86, respectively; F(l, 36) = 8.79,p < .01. This result may be viewed as a con-firmation of the effectiveness of the GPAinformation manipulation. Telling subjectsthat freshmen typically improve their gradesonce they become upperclassmen convincedthem that the freshman year is a difficultone. There were no other significant effectson this measure.

Subjects also rated how pleased they werewith their own freshman year, how muchthey worried about their performance, andhow they felt they compared academicallyto their peers. These were intended as ex-ploratory items on which no firm predictionswere made. Convincing people that theiracademic problems are caused by temporaryfactors might cause a sense of relief. On theother hand, this does not change the factthat they are having problems. In fact, theGPA or reasons analysis manipulations mightfocus attention on current problems and thusmake people feel temporarily worse abouttheir work.

Table 2Self-Reported Mood

Condition Week 1 Week 2

No reasons analysis 23.25 25.10Reasons analysis 26.30 24.95

Note. These scores represent the sum of responses tofour separate mood scales, A higher number indicatesa more positive mood.

The results of the question about howpleased subjects were with the way theirfreshman year was going suggests that thelatter effect occurred, at least at Week 1. A2 X 2 X 2 between-within ANOVA revealeda significant main effect of the GPA infor-mation manipulation, F(l, 36) = 4.80, p <.05, which was qualified by a GPA Infor-mation X Time interaction, F(l, 36) = 4.38,p<.05. At Week 1, subjects in the GPAinformation condition reported being lesspleased than did those in the no-informationcondition (Ms = 3.65 and 4.85, respec-tively). This difference was less pronouncedat Week 2 (Ms = 4.20 and 4.75, respec-tively).

Since the questions about how much sub-jects worried and how they thought theycompared to their classmates were identicalto ones asked several weeks earlier, 2 (GPAinformation) X 2 (reasons analysis) X 3 (ini-tial ratings, Week 1 ratings, Week 2 ratings)between-within ANOVAS were performed onthese questions. These ANOVAS failed to re-veal any effects of the manipulations on ei-ther measure. There was a significant maineffect of time on the question about howmuch subjects worried about their perfor-mance, F(2, 72) = 6.56, p < .005, reflectingthe fact that all subjects tended to worry lessover time.

Discussion

The results of the present study were noth-ing less than dramatic. By giving freshmeninformation indicating that academic prob-lems during the freshman year are tempo-rary, we succeeded in (a) improving theirperformance on sample GRE items, (b) low-ering the percentage of these students wholeft college, and (c) improving their GPA 1year after the completion of the study. Themagnitude of these behavioral effects mayseem large, especially considering that theGPA information was a one-time manipu-lation. It should be noted, however, that themagnitude of these effects are similar to thefindings of other studies that have used one-time interventions to help college students.Hanusa and Weiss (1979), for example, hada group of incoming freshmen participate ina one-session social skills training program.They found that these students significantly


improved their grades over the course oftheir first semester, whereas a control groupdid not. In the present study, a simple andstraightforward attributional interventionhas been identified that had similar long-term effects.

As in most other self-attribution studies,evidence for the attributions and internalstates that were mediating the behavioralresults is hard to come by (Nisbett & Wil-son, 1977). Consistent with Weiner's (1974)model of attribution, subjects who receivedthe GPA information reported higher ex-pectations of future academic success, which,according tp Weiner (1974), can mediateactual improvement in performance. TheGPA information did not, however, producereports of enhanced mood, which might alsohave mediated the behavioral results.

On the face of it, this pattern of resultsseems to support the notion that subjects'expectations mediated the changes in aca-demic performance but their moods did not.The correlations between the self-report andbehavioral measures, however, indicate thateven the self-reported expectations may nothave had a causal influence on behavior. Ofthe dozens of correlations computed betweenthe self-report (i.e., the expectations, mood,and all other questionnaire measures) andthe performance measures (i.e., GRE scores,dropout rates, GPA increases), no more weresignificant than would be expected by chance,and the significant correlations were evenlydivided in a positive and negative direction.Even the correlations between the expec-tancy and the performance measures werelow. Subjects' long-term expectations aboutgrade improvement, for example, correlated—.03 with the scores on the sample GREitems, .21 with whether or not the studentwas enrolled 1 year later, and —.31 with theincrease in grades 1 year later. None of thesecorrelations was significant. Thus, the pres-ent study, along with dozens of other self-attribution studies, found powerful behav-ioral effects in the absence of self-report ev-idence for the mediating cognitive processesand internal states.

The fact that none of the self-report mea-sures correlated with the performance mea-sures is consistent with the idea that self-reports are generated by an explanatory sys-tem that is at least partially independent of

the processes mediating behavior. Further-more, the fact that the reasons analysis ma-nipulation improved subjects' mood (at leastat Week 1), but had no effect on the be-havioral results at any time, suggests thatthe explanatory system can be manipulatedindependently of the behavioral system. Self*reports of new internal states (i.e., traits andattitudes in Wilson et al. [1981], mood inthe present study) are more likely to befound when subjects are asked to deliberateabout the reasons for their actions.

Unfortunately it is not yet clear whichtypes of self-reports are most apt to bechanged by reasons analysis manipulations.Asking subjects to think about reasons ledto reports of more positive moods in the pres-ent study but did not influence expectationsabout future performance or reports of at-titudes about performance. It is becomingincreasingly clear, however, that the condi-tions under which people will change theirbehavior in accord with attributional pre-dictions differ from the conditions underwhich people will change self-reports of in-ternal states. Thus, practitioners interestedin using attribution therapy should decidewhether their goal is to change behavior orself-reports and choose their techniques ac-cordingly.

This position bears some similarity to Lev-enthal's (1970) parallel response model ofattitude change. The traditional attitudemodel holds that a stimulus such as a per-suasive communication triggers emotionaland attitudinal responses which then causechanges in behavior. In contrast, Leventhal(1970) argues that attitudinal,' emotional,and behavioral responses are not linked ina sequential chain but instead occur in par-allel, each with its own set of antecedents.We are proposing a similar alternative to thetraditional view of self-attribution processes.The traditional attribution model states thatattributional manipulations (e.g., the GPAinformation in the present study) lead tocausal inferences about the self (e.g., "Thecauses of my academic problems are unsta-ble and temporary"), which lead to new in-ternal states (e.g., increased expectations orimproved moods), which lead to changes inbehavior (e.g., improved academic perfor-mance). We are not disputing that such aprocess occurs; indeed, we hold that such a


process accounts for our behavioral results.However, we argue that people's ability toreport these attributions and internal statesis limited, and that people thus rely on anexplanatory system to generate verbal re-ports. As implied by the present results,these reports by the explanatory system donot always correctly represent the attribu-tions and the states mediating behavior.

These theoretical matters, though impor-tant, should not obscure the practical sig-nificance of the present results. An inexpen-sive, simple procedure has been found to helppeople concerned about their academic per-formance. Videotaped interviews with up-perclassmen who report that they had tem-porary problems as freshmen could easily beshown as part of a freshman advising pro-gram, or even shown on a mass scale at fresh-man orientations. The present results implythat such viewings would lead to consider-able academic improvement.

References

Andrews, G. R., & Debus, R. L. Persistence and thecausal perception of failure: Modifying cognitive at-tributions. Journal of Educational Psychology, 1978,70, 154-166.

Bootzin, R. R., Herman, C. P., & Nicassio, P. Thepower of suggestion: Another examination of misat-tribution and insomnia. Journal of Personality andSocial Psychology. 1976, 34, 673-679.

Brownstein, S. C., & Weiner, M. Barren's how to pre-pare for the Graduate Record Examination, (3rded.). Woodbury, N.Y.: Barren's Educational Series,1978.

Dweck, C. S. The role of expectations and attributionsin the alleviation of learned helplessness. Journal ofPersonality and Social Psychology, 1975, 31, 674-685.

Hanusa, B. H., & Weiss, P. Importance of perceivedconfidence and perceived control in adaptation to col-lege. Pittsburgh, Pa.: Carnegie-Mellon University,1979. (ERIC Document Reproduction Service No.ED 174 867)

Kelley, H. H. Attribution theory in social psychology.In D. Levine (Ed.), Nebraska Symposium on Mo-tivation (Vol. 15). Lincoln: University of NebraskaPress, 1967.

Kellogg, R., & Baron, R. S. Attribution theory, insom-nia, and the reverse placebo effect: A reversal ofStorms and Nisbett's findings. Journal of Personalityand Social Psychology, 1975, 32, 231-236.

Langer, E. J., & Abelson, R. P. The semantics of askinga favor: How to succeed in getting help without really

dying. Journal of Personality and Social Psychology,1972, 24, 26-32.

Leventhal, H. Findings and theory in the study of fearcommunications. In L. Berkowitz (Ed.), Advances inexperimental social psychology (Vol. 5). New York:Academic Press, 1970.

Lowery, C. R., Denney, D. R., & Storms, M. D. Thetreatment of insomnia: Pill attributions and nonpe-jorative self-attributions. Cognitive Therapy and Re-search, 1979, 3, 161-164,

Nie, N. N., Hull, C. H., Jenkins, J. G., Steinbrenner,K., & Bent, D. H. Statistical package for the socialsciences (2nd ed.). New York: McGraw-Hill, 1975.

Nisbett, R. E., & Borgida, E. Attribution and the psy-chology of prediction. Journal of Personality andSocial Psychology, 1975, 32, 932-943.

Nisbett, R. E., Borgida, E., Crandall, R., & Reed, H.Popular induction: Information is not necessarily in-formative. In J. S. Carroll & J. W. Payne (Eds.),Cognition and social behavior. Hillsdale, N.J.: Erl-baum, 1976.

Nisbett, R. E., & Wilson, T. D, Telling more than wecan know: Verbal reports on mental processes. Psy-chological Review, 1977, 84, 231-259.

Storms, M. D., Denney, D. R., McCaul, K. D., & Low-ery, C. R. Treating insomnia. In I. H. Frieze, D. Bar-Tal, & J. S. Carroll (Eds.), New approaches to socialproblems. San Francisco: Jossey-Bass, 1979.

Storms, M. D., & McCaul, K. D. Attribution processesand emotional exacerbation of dysfunctional behav-ior. In J. H. Harvey, W. J. Ickes, & R. F. Kidd (Eds.),New directions in attribution research (Vol. 1). Hills-dale, N.J.: Erlbaum, 1976.

Storms, M. D., & Nisbett, R. E. Insomnia and the at-tribution process. Journal of Personality and SocialPsychology, 1970, 16, 319-328.

Valins, S., & Nisbett, R. E. Attribution processes in thedevelopment and treatment of emotional disorders.In E. E. Jones, D. E. Kanouse, H. H. Kelley, R. E.Nisbett, S. Valins, & B. Weiner (Eds.), Attribution:Perceiving the causes of behavior. Morristown, N.J.:General Learning Press, 1972.

Weiner, B. Achievement motivation and attributiontheory. Morristown, N.J.: General Learning Press,1974.

Weiner, B. The role of affect in rational (attributional)approaches to human motivation. Educational • Re-searcher, 19W, 9, 4-11.

Weiner, B., Russell, D., & Lerrhan, D. Affective con-sequences of causal ascriptions. In J. H. Harvey,W. J. Ickes, & R. F. Kidd (Eds.), New directions inattribution research (Vol. 2). Hillsdale, N.J.: Erl-baum, 1978.

Wessman, A. E., & Ricks, D. F. Mood and personality.New York: Holt, Rinehart& Winston, 1966.

Wilson, T. D., Hull, J. G., & Johnson, J. Awarenessand self-perception: Verbal reports on internal states.Journal of Personality and Social Psychology, 1981,40, 53-71.

Received January 28, 1981 •

improving the academic performance of college freshmen

Documents