Developmental Psychology1977, Vol. 13, No. 6, 654-665

Developmental Processes in Social Inference:Averaging of Intentions and Consequences

in Moral Judgment

COLLEEN F. SURBERUniversity of Illinois at Urbana-Champaign

Two experiments tested the applicability of Anderson's relative-weight averagingmodel to children's judgments based on intentions and consequences. Experiment1 provided support for the model over a wide age range (kindergarten through adult)using stimuli similar to Piaget's in which the intentions ranged from positive tonegative and the consequences ranged from neutral to negative. Experiment 2extended Experiment 1 by including consequences ranging from positive to nega-tive. In Experiment 2, intentions and consequences interacted suggesting that somemodification of the assumptions of the averaging model is necessary to fully accountfor the data. Data from the two experiments suggest that both intentions andconsequences influence judgments at all ages examined and that developmentalchanges in moral judgments are continuous rather than stagelike. These develop-mental differences can best be described by changes in the weights of intentions andconsequences. The major developmental change appears to be a decrease in theweight of consequences with increasing age. The possibility of using algebraicmodels to describe the results of studies that have attempted to influence therelative importance of intentions and consequences in children's moral judgments isdiscussed.

A major issue in studies of children'smoral judgment has been the extent to whichchildren of different ages use intention andoutcome information in judging another'sdeeds. Piaget (1965) proposed that childrenprogress from a stage of "moral realism" orobjectivity to a stage of "moral subjectiv-ity" between the ages of approximately 5and 12. According to Piaget, during the

This research was completed while the author held aNational Science Foundation Graduate Fellowshipand was supported in part by Grant HD-06864 to AnnL. Brown, Grant HD-05951 to Joseph C. Campione,and U.S. Public Health Service Grant HD-00244fromthe National Institutes of Child Health and HumanDevelopment.

The author thanks Michael H. Birnbaum and BobWyer for helpful advice and encouragement, SteveStegner and Judy DeLoache for helpful comments onthe manuscript, Roberta Jones and Ruth Pearl forassistance in running subjects, and Brad Enna for pro-viding adult subjects.

Requests for reprints should be addressed to Col-leen Surber, who is now in the Department of Psy-chology, University of Connecticut, Storrs, Connect-icut 06268.

stage of moral objectivity children judgedeeds primarily in terms of the outcomes orconsequences of the deeds. In contrast, amorally subjective child judges a deedprimarily in terms of the intentions of theactor and not in terms of the consequences.

Piaget seems to imply that both intentionand consequence information will be used inmaking moral evaluations, but does not pro-vide a way of representing the relative con-tribution of these cues to moral judgments.Researchers have attempted to use the pro-portion of variance accounted for, the ex-tremity of the evaluative responses pro-duced by outcome and intent information(Parsons, Ruble, Chereskin, Feldman, &Rholes, 1976), or a ratio called the "Inten-tion Judgment Quotient" (Hebble, 1971) toindex the relative importance of intentionsand consequences in determining moraljudgments. As yet, there has been no at-tempt to specify the processes involved incombining intentions and consequences informulating moral judgments or to relate

654

MORAL JUDGMENT 655

such a theory to developmental change inthe influence of intention and consequenceinformation.

The two experiments reported here rep-resent initial applications of information in-tegration theory (Anderson, 1971; 1974) tochildren's moral judgments. This approachassumes that each piece of information hasa psychological value (a scale value) on thedimension of judgment and that the scalevalues of the information presented arecombined to produce an integrated psycho-logical impression, which is then translatedinto a response on the experimenter's scale.Theories of the manner in which informa-tion is combined are expressed mathemat-ically, and the relative contribution of eachvariable is a part of the equation describingthe way in which the information is com-bined.

A particular model of information inte-gration that appears to provide a good ac-count of the manner in which adults com-bine intention and consequence informationis the relative-weight averaging model(Lane & Anderson, 1976). For moral judg-ments based on intentions and conse-quences, the averaging model can be writ-ten

, _ + WCSCJ +

+ WC +(1)

where Su and scj are the scale values of the/th andjih intentions and consequences re-spectively, and Wi and n>c are the weights ofthe intentions and consequences. Theweights can be considered to be representa-tions of the psychological importance of in-tentions and consequences for the judg-ment. The «/»« is the psychological impres-sion of the /th intention and jth conse-quence combination. The term w0s0 is usu-ally called the "initial impression"—that is,the subject's impression in the absence ofany information (or bias toward positive ornegative judgments in general). If it is as-sumed that (a) the values of w, and wc re-main constant over all levels of s, and sc, (b)the values of SH and sci are independent ofthe combinations in which they occur, and(c) that the subject's response, Ry, islinearly related to i//«, then the model pre-

diets that intentions and consequences willhave additive effects on the subject's judg-ments. There should be no interaction ofintentions and consequences in an analysisof variance, and the data should plot as aparallel set of curves.

An additive model for the combination ofintentions and consequences also predictsthat there should be no interaction of inten-tions and consequences. The additivemodel,

i i i f^\

can be distinguished from the averagingmodel in Equation 1 by examining judg-ments of stimuli involving only intentioninformation or only consequence informa-tion. The averaging model predicts that theeffect of intentions will be greater whenthere is no consequence information pre-sented, while the additive model predictsthat the effect of intentions will be indepen-dent of the effect of the information withwhich it is combined. If the judgments areplotted as a function of the level of inten-tions , the averaging model predicts that theslope of the judgments of intention-onlystimuli should be steeper than the slopes ofthe judgments of intention-consequencecombinations. This follows from Equation1, since the slope is predicted to be propor-tional to the effective weight of intentions.When intentions and consequences arecombined, the effective weight of intentionsis, w'i = Wi/(wl + wc + H>0). When only in-tention information is presented, w': =H>I/(W! + w0). An analogous predictionholds for stimuli involving only conse-quence information.

Experiment 1 tests the averaging modelfor children's moral judgments with a rangeof stimuli similar to those used by Piaget(1965). Stories involving intentions rangingfrom positive to negative and accidentalconsequences ranging from slightly nega-tive to very negative were rated by childrenin kindergarten, second, and fifth grades. Inaddition, stories involving only intention in-formation and only consequence informa-tion were included to test between the addi-tive and averaging models. Experiment 2extends Experiment 1 by including conse-quences ranging from positive to negative.

656 COLLEEN F. SURBER

EXPERIMENT 1

Method

Subjects

Subjects were 25 kindergarten (14 females, 11males), 39 second-grade (19 females, 20 males), and 33fifth-grade (19 females, 14 males) children from a localelementary school who participated with parentalpermission. All children who obtained parental per-mission were included in the study. One second-gradeand one kindergarten subject were discarded becausethey failed to understand the instructions. In addition,a sample of 9 college students (5 females, 4 males) wasalso obtained.1 The mean ages of the three groups ofelementary school children were 65.5, 89.3, and 127.5months for kindergarten, second and fifth grades, re-spectively.

Stimuli

Nine stories were constructed by combining threeintentions with three different outcomes. Intentionsalways preceded the consequences. The three inten-tion sentences were:1. This little girl wanted to help set the table to make

her mother happy.2. This little girl wanted to see what kinds of things

were on the kitchen shelf. She thought she wasallowed to climb up to look on the shelf.

3. This little girl wanted to get a cookie. Her mothertold her that she couldn't have a cookie.

The three consequences were constructed by enteringthe appropriate intention ("get out the dishes,""look," or "get a cookie") in the following sentences.1. She climbed up on the kitchen cupboard to (inten-

tion), and even though she was careful, she slippedand a dish almost fell off the cupboard and broke.

2. She climbed up on the kitchen cupboard to (inten-tion), and even though she was careful, she slippedand 2 dishes fell off the cupboard and broke.

3. She climbed up on the kitchen cupboard to (inten-tion), and even though she was careful, she slippedand 10 dishes fell off the cupboard and broke.

In addition to the nine intention-consequence combi-nations, three intention-only, and three conse-quence-only stories were constructed by modifyingthe sentences above. For example, "This little girlwanted to set the table to make her mother happy, soshe climbed up on the kitchen cupboard to get out thedishes."

Rating ScaleFor the kindergarten and second grade children, a

series of four frowning faces ranging in size from 7.5 to1 This experiment was also run with a sample of 45

college subjects using a rating scale ranging from - 5 to+5, labeled "very bad" to "very good." The resultswere very similar to those reported for the smallersample here, but since the response scale differed, theresults are not reported in detail.

3.5 cm in diameter was used to represent the "bad"portion of the rating scale. The largest frowning facewas labeled "very bad" by the experimenter, and thesmallest frowning face was labeled "a little bit bad." Asimilar series of smiling faces was used to represent the"good" portion of the scale, with the largest and smal-lest smiling face labeled "very good" and "a little bitgood." For the fifth-grade and college subjects, theresponse scale consisted of the integers from 1 to 8, withI, 4, 5, and 8 labeled "very bad," "a little bit bad," "alittle bit good," and "very good," respectively.

Procedures

The kindergarten and second graders were run indi-vidually. The nine intention-consequence stories, plusthe three intention-only and, three consequence-onlystories were read to the subject in one of three randomorders by a female experimenter. The subjects wereinstructed to imagine that each story pertained to a littlegirl who was 5 years old, and that each story was about adifferent little girl. The 15 stimuli were preceded by two"warm-up" stories, the positive intention-only and thenegative intention-negative consequence stories. Thesubject was asked to repeat each warm-up story to theexperimenter before he or she was asked to rate thecharacter. The experimenter did not continue until thesubject was able to accurately repeat the story. At theend of each story the subject was asked whether thelittle girl in the story was a good girl or a bad girl.Depending on the subject's answer, the experimenterpresented either the smiling or frowning faces andasked the subject to indicate "how good" or "howbad" by pointing to one of the faces.

The fifth graders were run in two classroom groups.The experimenter explained the response scale andthen read the stories one at a time to the subjects whilethey recorded their responses in individual booklets.For the college age subjects, the stories were printed ontwo pages in a random order preceded by the instruc-tions. The subjects read the stories and recorded theirjudgments.

Results

Preliminary analyses revealed no main ef-fect or interactions due to sex of subject.Therefore, sex was not included as a factorin any of the analyses reported below.

A4(Age Level) x 3(Intentions) x 3(Con-sequences) analysis of variance on the chil-dren' s ratings yielded main effects of inten-tions, F(2, 204) = 133.80, consequences,F(2, 204) = 137.12, and age, F(3, 102) =5.40. The main effect of age reflects the factthat the adults' ratings were slightly morepositive than those of the other age groups(mean ratings = 4.37, 3.36, 3.44, and 3.06,for adult through kindergarten, respec-

MORAL JUDGMENT 657

ao

7.0

6.0

5.0

4.0

3,0

2.0

1.0

/ INT./ ONLY

--•M-

2ND * INT.ONLY

H CONS.ONLY

8.0

7.0

6.0

5.0

4.0

5.0

2.0

5TH» INT.

ONLY

ADULT

H CONS. LONLY

INTENTIONS

N4M-

H CONS.ONLY

Figure 1. Mean judgments of story characters as a function of intentions (L •cookie, M = look, H = set table), with a separate curve for each level of outcome (L :

10 dishes broken, M = 2 dishes broken, N = a dish almost broken).

lively). Age level interacted with both inten-tions, F(6, 204) = 8.08, and consequences,F(6, 204) = 12.22. All other effects werenonsignificant.

Parallelism Prediction of the AveragingModel

Figure 1 presents the subjects' mean rat-ings with a separate panel for each agegroup. The mean ratings in each panel areplotted as a functions of intentions, with aseparate curve for each level of conse-quences. The levels of intentions on theabscissa are spaced according to the margi-

nal means for each age group (with the ex-ception of consequences only, which is arbi-trarily positioned). The parallelism predic-tion of the averaging model appears to holdup graphically as well as statistically, asshown by the absence of an intention byconsequences interaction. Supplementaryanalyses of each group alone snowed thatthe main effects of both intentions and con-sequences were significant, p < .01, at eachage level (with exception of adults, whoshowed no significant effect due to conse-quences, F[2, 16] = 2.04),2 and that the

The larger sample of college subjects showed a

658 COLLEEN F. SURBER

8.0

7.0

6.0

4.0

3.0

2.0

1.0

ADULT

M H

INTENTIONS

L M- N

CONSEQUENCES

Figure 2. (A) Mean judgments as a function of intentions with a separate curve foreach age level. (B) Mean judgments as a function of consequences with a separatecurve for each age level.

Intention x Consequence interaction wasnonsignificant in all cases.

Adding Versus Averaging Integration Process

In addition to predicting that the data fromthe Intention x Consequence design shouldplot as a parellel set of curves at each agelevel, the averaging model predicts that theslope of the curve representing the judg-ments of the intention-only stories should besteeper than the other curves at each agelevel. This prediction can be seen in all fourpanels in Figure 1 and is confirmed by thefinding of an Intention x Consequence in-teraction when intentions only is included asanother level of consequences in analysis ofvariance, F(6,612) = 18.81. This interactionis incompatible with the additive model ofEquation 2. An analogous prediction shouldhold for the consequences-only stories; thevertical spacing of the consequences-onlymeans should be greater than the verticalspacing of the intention-consequence combi-nations (solid curves). This prediction of theaveraging model does not appear to beconfirmed, and when consequences only(i.e., no intention) is included as a level ofintention, there is no Intention x Conse-

marginally significant consequence effect, F(2, 88) =4.40.

quence interaction, F < 1.0. The failure ofthis prediction could be accounted for bypostulating that the value of wl is small rela-tive to the value of wc + w0 at all ages. If thisis the case, it suggests that the major devel-opmental change is a decrease in the value ofwc with increasing age, rather than an in-crease in the value of wt. This possibility isdiscussed below.

Developmental Changes in Weights

According to the averaging model inEquation 1, the slopes of the curves in Fig-ure 1 are proportional to the effectiveweight of intentions, w\ = w\/(wi + wc +w0), and the vertical spacing of the curves isproportional to the effective weight of con-sequences, w'c = wc/(wi + wc + w0). InFigure 1, it is apparent that with increasingage the slopes become steeper, while thevertical spacing of the curves decreases.The changes in effective weight can be seenmore clearly in Figure 2, which plots theAge x Intention and and Age x Conse-quence interactions for the present data.The slope of each curve in Figure 2 A isproportional to the effective weight of in-tentions for each age group, and the slopeof each curve in Figure 2B is proportionalto the effective weight of consequences. Itcan be seen that the effective weight of

MORAL JUDGMENT 659

intentions increases with age, while the ef-fective weight of consequences decreases.

Another possible way of accounting forthe Age x Intention and Age x Conse-quence interactions in the present experi-ment might be to postulate that the scalevalues of the cues change with age. If oneassumes that the weights of both intentionsand consequences remain constant over agelevels, then in order to account for the data,the range of the scale values of intentionsmust increase with age. Under these as-sumptions it would be difficult to accountfor the observation that the slopes of theintention-only curves in Figure 1 shows lit-tle developmental change, while the slopesof the intention-consequence combina-tions show rather large age variation. Thisobservation can be accounted for by pos-tulating a small value of wl for all ages.Thus, the hypothesis that the weightschange developmentally seems to provide abetter account of the data of Experiment 1.

The data were fit to the averaging modelof Equation 1 with the aid of a computersubroutine (Chandler, 1969), using a leastsquares criterion, and assuming that thevariables wi, wc, and s0 depend on agelevel. The scale values were assumed to beequal across all ages and the value of w0was arbitrarily set equal to 1.0. The correla-tion of the predictions of the model and theobtained values was .985. The estimatedvalues of the weights and s0 are shown inTable 1. It can be seen that the estimates ofw, increase only slightly with age, while theestimated wc show a great decrease withage. This follows from the observation thatthe slopes of the intention-only curvesshow little developmental change, while thespacing of the consequence-only meansshows a marked decline with age. The es-timated scale values of the three intentions

Table 1: Estimated Values of wc, w, and sa inExperiment 1

Age group

KindergartenSecond gradeFifth gradeAdult

wc

6.4185.608

.735

.201

HI

.8941.389

.7191.083

so

5.229.293.815.01

Note. The value of w0 = 1.0.

were 8.098, 4.124, and -.302, and the esti-mated scale values of the consequenceswere 4.489, 2.255, and .697.

Discussion

The averaging model of Equation 1 ap-pears to provide a good account of the de-velopmental trends in moral judgmentsbased on intention and consequence infor-mation. The process for making moralevaluations appears to be the same across abroad age range, and the theory provides ameaningful way of discussing develop-mental change in the relative contributionof intentions and consequences. It appearsthat the major developmental change is inthe absolute weight of consequences, wc,rather than in the weight of intentions, Wj.These results suggest that developmentalchange in moral judgment proceeds partlyas a result of the child's learning that con-sequences are not relevant to moral evalua-tions, rather than his learning that inten-tions are relevant. Clearly, even the kin-dergarten children are aware of the evalua-tive implications of the intention informa-tion.

The data also appear to support the notionthat moral judgments change gradually overthe age range, rather than following astagelike progression. However, the data ofFigure 1 could be generated by four agegroups composed of different proportions ofsubjects who respond either totally on thebasis of intentions or totally on the basis ofconsequences. Inspection of the individualpatterns of responses does not support thisinterpretation, since 76% of the kindergar-ten children, 87% of the second-grade chil-dren, 94% of the fifth-grade children, and56% of the adults showed response variationdue to both intentions and consequences.

EXPERIMENT 2

Experiment 1 leaves an important issueunexamined. Previous researchers havefound that when the consequence informa-tion is positive, even 5-6-year-old childrenappear to place more importance on inten-tions in their judgments than when the con-

660 COLLEEN F. SURBER

sequences are negative (Costanzo, Coie,Grumet, & Farnill, 1973; Feldman, Klos-son, Parsons, Rholes, & Ruble, 1976; Par-sons, et al., 1976; Weiner, & Peter, 1973).Thus, in Experiment 2 the range of conse-quences was extended to include positiveconsequences as well as negative ones. Inaddition, it seemed desirable to replicate theresults of Experiment 1 with a different storytheme to enhance the generality of thefindings.

Method

Subjects

Subjects were 28 kindergarten (18 female, 10 male),20 second-grade (10 female, 10 male), 12 fifth and sixthgraders (5 female, 7 male), and 19 adults (15 female, 4male). The grade school children were recruited from alocal elementary school and participated with parentalpermission. The mean ages of the three groups of chil-dren were 71.5, 97.4, and 139.2 months.

Stimuli

Twelve stories were constructed by combining threeintentions with four consequences. Intentions alwayspreceded consequences. The subjects were instructedto imagine that each story pertained to a different littleboy who was 5 years old. The three intentions were:1. This little boy went outside to rake the leaves up to

make his mother happy.2. This little boy went outside to play, and he thought

that it was all right to go outside.3. This little boy went outside to play, even though his

mother told him to stay in and clean up his room.The four consequences were:1. After he went outside, he looked down and he found

50 cents.2. After he went outside, he looked down and he found

a penny.3. While he was outside, he accidentally stepped in his

mother's flower garden and killed two of the plants.4. While he was outside, he accidentally stepped in his

mother's flower garden and killed all of the plants.In addition to the 12 intention-consequence combina-tions, the 3 intention-only and 4 consequence-onlystories were also presented. The consequence-onlysentences were modified to read, for example, "Thislittle boy was outside, and he looked down and found 50cents."

Procedures

The procedures of Experiment 2 were similar to Ex-periment 1 with a few exceptions. Three warm-upstories were used; the positive intention-only, the nega-tive intention-negative consequence, and the moderate

intention-moderately positive consequence stories.For the kindergarten and second graders, the storieswere tape-recorded in two different random orders, andthe subject was allowed to activate the tape recorder topresent each story. Both portions of the rating scalewere presented simultaneously to the subject followingeach story, and the experimenter asked,'' How good orhow bad was the little boy in the story?"

The fifth and sixth graders were from a mixedhomeroom class and were run in one group using thesame procedures as in Experiment 1.

Results and Discussion

Preliminary analyses revealed no main ef-fect or interactions due to sex of subject.Therefore, sex was not included as a factorin any of the analyses reported below.

A 4(Age) x 3(Intentions) x 4(Conse-quences) analysis of variance revealed maineffects of intentions, F(2, 150) = 199.55,consequences, F(3, 225) = 120.09, and agelevel, F(3, 75) = 12.72. The main effect ofage is again due to the fact that the adults'ratings were slightly more positive thanthose of the other age groups (mean ratings= 5.14, 4.28, 4.02, and 4.15, for adultthrough kindergarten, respectively). As inExperiment 1, age level interacted with bothintentions, F(6, 150) = 9.21, and conse-quences, F(9, 225) = 7.71. In addition, theIntention x Consequence interaction wassignificant, F(6, 450) = 12.97.

Figure 3 presents the mean ratings, plottedas a function of the level of intentions with aseparate curve for each level of conse-quences. Qualitatively, the Age x Intentionand Age x Consequence interactions repli-cate Experiment 1. With increasing age, theeffect of consequences (vertical separationof the curves) decreases and the effect ofintentions (slope) increases. Supplementaryanalyses of each age group separatelyshowed that the main effects of intentionsand consequences were significant,p < .01,at each age level replicating the results ofExperiment 1 that both intention and conse-quences are important in determining moraljudgments at all ages examined.

It can again be seen that the intention-onlycurves are steeper than the curves for theintention-consequence combinations, aspredicted by the averaging model.

MORAL JUDGMENT 661

O

8.0

7.0

6.0

5.0

4.0

_ 3.0

H< 2.0

1.0

8.0

Z 7.0

,- 6.0

5.0

4.0

3.0

2.0

1.0 -

2NDr INT.

/' ONLY

• M-

H CONS. LONLY

H CONS.ONLY

5TH ADULTINT. ONLY

,\ • HiM+

M H CONS. LONLY

M H CONS.ONLY

INTENTIONS

Figure 3. Mean judgments of story characters as a function of intentions (L = play,not supposed to, M = play, thought it was okay, H = rake leaves), with a separate curvefor each level of consequence (L = killed all the plants, M- = killed 2 of the plants,M+ = found a penny, H = found 50 cents).

Developmental Changes in Weight

Figure 4 shows the Age x Intention andAge x Consequence interactions for Exper-iment 2. The changes in slope with age showthat the effective weight of intentions in-creases while the effective weight of conse-quences decreases. Estimates of the valuesof wi and wc (presented below) replicate thefinding of Experiment 1 that the major de-velopmental change is in the absolute weightof consequences rather than in the weight ofintentions. Inspection of the response pat-terns of individual subjects showed that theAge x Intentions and Age x Consequences

interactions were not produced primarily bydifferent proportions of subjects who re-sponded either totally on the basis of inten-tions or totally on the basis of consequences.Eighty-two percent of the kindergarten, allof the second and fifth graders, and 58% ofthe adults showed variation due to bothcues, replicating Experiment 1 fairlyclosely.

Interaction of Intentions and Consequences

The interaction of intentions and conse-quences in Experiment 2 can be seen in thatthe curves in each panel of Figure 3 diverge

662 COLLEEN F. SURBER

8.0,

7.0

6.0

5.0

4.0

3.0

2.0

1.0

ADULT

52

K

L M H

INTENTIONSL M- M+ H

CONSEQUENCES

Figure 4. (A) Mean judgment as a function of intentions with a separate curve foreach level. (B) Mean judgment as a function of consequences with a separate curve foreach age level.

to the right. This interaction was probablynot detected in Experiment 1 because theconsequences were all in the negativerange. In research with adults on bothmoral evaluations and personality im-pressions where stimuli of both negativeand positive value are used, interactionssimilar to that of Experiment 2 are found(Birnbaum, 1972, 1974; Kanouse & Han-son, 1971). Thus, the results of the presentexperiment seem to support the idea thatthe mechanisms behind such evaluativeprocesses are qualitatively similar across awide age range.

There are several possible ways of mod-ifying the assumptions outlined in the intro-duction to account for the interaction foundin Experiment 2. One possible modificationof the averaging model discards the as-sumption that the scale values of the infor-mation are independent of the combinationsin which they occur. To account for thepresent data, one would have to assumethat the scale value of the highest piece ofinformation in a set shifts toward the lowerpiece. For example, in the context of anegative consequence, a positive intentionmight be viewed as less positive than oth-erwise.

Another possible way of modifying theaveraging model is to discard the assump-

tion that the weights are constant over alllevels of the scale values of intention andconseqeuences. One possible alternative isthe differential-weight averaging model inwhich the weights depend on the scale val-ues of the items (Anderson, 1972). In Ex-periment 2, items with lower scale valueswould receive greater weight. This modelrequires that two parameters be estimatedfor each stimulus, the weight and the scalevalue. Another possibility is Birnbaum's(1972, 1974) configural-weight averagingmodel in which the weights depend partlyon the configuration of the scale values ofthe information presented. This model re-quires estimation of only one additional pa-rameter and provides a good account ofinteractions similar to those of Figure 3.3

3 Another assumption that might be questioned isthat the responses, Ru, are linearly related to the psy-chological impressions, i|iu. That is, perhaps the in-teractions in Figure 3 are due to a nonlinear responsefunction. This interpretation would suggest that thedata in Figure 3 should be monotonically rescaled toeliminate the interaction. Although the magnitude ofthe interaction can be greatly reduced by monotonicrescaling, one must still address the question ofwhether the data should be rescaled to parallelism. Insupport of the interpretation that the interaction ofintentions and consequences reflects differences in thei|)u rather than nonlinearity in the response function,Birnbaum (1972) has found that adults' judgments of

MORAL JUDGMENT 663

Table 2: Estimated Values of wc, w,, \0 and i|« inExperiment 2Age group wc w, .s0 i|i

KindergartenSecond gradeFifth gradeAdult

2.6351.2721.115.296

.751

.8761.1381.046

2.783.745.325.75

-.470-.477-.579-.093

Note. The value of wa = 1.0.

Birnbaum's configural-weight model canbe written as follows:

,1, _ + WjSu + WCSCJ_

+ + Wc

<a\Su - SCj|- (3)

This model was fit to the data of Experi-ment 2 assuming that s0,

wi, wc, and <u de-pend on age. The value of w0 was again setto 1.0 and the scale values were assumed tobe the same for all ages. The correlation ofthe predicted and obtained values was .987.The estimated values of s0, wlt wc, w arepresented in Table 2. As in Experiment 1,the values of wc decline with age, while thevalues of wt increase only slightly. The es-timated scale values of the intentions were9.354, 5.793, and -.528. The estimatedscale values of the consequences were8.293, 7.425, 3.344, and 1.511.

GENERAL DISCUSSION

The averaging model of Equation 1 ap-pears to provide a satisfactory account ofthe developmental trends in moral judgmentobtained in the present study. The majordevelopmental change appears to be in theabsolute weight of consequences, ratherthan in the weight of intentions. The devia-tions from the model in Experiment 2 arequalitatively similar to the findings in re-search on personality impressions withadults and appear to be accounted for by a

combinations of deeds show a similar interaction, andhis results did not seem to be due to a nonlinear re-sponse function. Nevertheless, future research shouldexamine the possibility that the interactions in thepresent experiment are due to a nonlinear responsefunction. Methods for examining this hypothesis havebeen developed by Birnbaum (1974), but have not yetbeen applied to children's judgments.

simple modification of the theory. Overall,the results support the notion that evaluativesocial inferences involve a similar processover a wide age range.

The averaging model can also probablyaccount for the results of many studies thathave attempted to manipulate the relativeimportance of intentions and consequencesin moral judgments. For example, two re-cent experiments have shown that the use ofintention and consequence information bychildren younger than 9 or 10 years of agedepends greatly on the order in which theintention and consequence information ispresented (Feldman et al., 1976; Parsons etal., 1976). Feldman et al. found that whenintentions were presented last, 4-5-year-oldchildren's judgments showed a greater ef-fect due to intentions, and the subjects'judgments appeared to reflect what they re-membered. By postulating developmentalchange in the degree to which the weight of acue depends on the order of presentation,the averaging model can account forFeldman et al.'s findings that the judgmentsof 8-9-year-olds also showed order effects,but that the order effects were not as pro-nounced as with the younger children. An-derson (1965) has shown that changes in theweights can account for order effects in per-sonality impressions.

Feldman et al. (1976, p. 559) also con-cluded that "younger subjects' relative ne-glect of intent in the normal order of infor-mation may be based in part on their failureto remember the material correctly ratherthan on the differential weighting of the in-formation." It is worth noting that to mea-sure the children's memory for the intentionand consequence components of a story,Feldman et al. asked the children to recallthe valence of the intention and conse-quence. After judging a story, the subjectwas asked whether what the actor tried to doand what actually happened were good orbad. This is not really a measure of the chil-dren's memory for the intention or conse-quence per se, but is ajudgment of the inten-tion or consequence component of the story.One would expect contextual shifts in theweights and/or scale values to be reflected injudgments of the component stimuli even ifthe subject remembered the components

664 COLLEEN F. SURBER

perfectly. Feldman et al.'s data support thenotion that some sort of contextual changesoccur in either the weights or scale values,that such contextual shifts are order depen-dent, and that such changes are greater foryounger children than for the older children.The degree to which these contextual shiftsin the judgments of the component stimuliare due to failure to remember the compo-nents accurately remains a topic for futureresearch.

Although it is probably true that memorydifficulties contribute to the younger sub-jects' neglect of the intention information,the results of the consequence-only ratingsin both Experiments 1 and 2 show that theage differences in the children's responsesmust be due in part to different values of theweights independent of memory differ-ences. In the context of the rest of the dataof Figures 1 and 3, it is possible to accountfor the fact that the spread of theconsequence-only means decreases withage by postulating that the value of wc de-creases with age. Thus, it appears that evenif young children remember the informationperfectly (it would be difficult to argue formemory differences in the consequence-only ratings), there is some developmentalchange in the weight assigned to the infor-mation. That is, an important factor in de-velopmental change in moral judgment ap-pears to be a change in the psychologicalimportance of consequences for such eval-uations . The children already know that in-tentions are important for moral evalua-tions; what they must learn is that acciden-tal consequences are irrelevant.

The results of Experiments 1 and 2 alsoprovide some insight into the inadequacy ofthe "moral dilemma" technique for provid-ing information about the relative impor-tance of intentions and consequences inmoral judgments. In the moral dilemmatechnique, the child is asked which of twocharacters is the naughtiest in a pair ofstories in which the intentions and conse-quences conflict. For example, in Experi-ment 1, if the experimenter had asked thesecond-grade children whether the child inthe "Set the table-10 dishes broken" (Hintention-L consequence) story wasnaughtier than the child in the "Cookie-no

dishes broken" (L intention-N conse-quence) story, Figure IB suggests that thechildren would have chosen the characterin the set table-10 dishes story. Thus, theexperimenter would have concluded thatthe children were normally objective intheir judgments. If the experimenter hadchosen the "Cookie-2 dishes broken" (Lintention-N consequence) story for com-parison, the data suggest that he wouldhave come to the opposite conclusion; thesecond-grade children's judgments wouldnow appear to be morally subjective. Actu-ally, these children rely on both intentionsand consequences in formulating theirjudgments. From data collected using themoral dilemma technique, it is impossibleto determine the relative importance of in-tentions and consequences.

Since many training studies that have at-tempted to manipulate the relative impor-tance of intentions and consequences inchildren's moral judgments have used chil-dren's responses to moral dilemmas as theircriterion measure, it should be apparent thatinterpretation of their results is difficult(Bandura & McDonald, 1963; Bearison &Isaacs, 1975). It is possible that the proce-dures in such studies have manipulated onlythe child's responses to similarly con-structed story pairs and not the relative im-portance of intentions and consequences indetermining moral evaluations. Even if suchstudies have successfully manipulated theimportance of intentions in the children'sjudgments, one cannot determine how muchthe relative importance of intentions hasbeen manipulated simply by examining thechildren's responses in the moral dilemmaprocedure. The data presented in Figure 1and 3 should be sufficient to convince thereader that "morally subjective" responsesto a pair of moral dilemmas could representwidely differing patterns of evaluative re-sponses to intention-consequence combina-tions. Thus, to fully evaluate the results ofany training study, it is necessary that judg-ments of more than just moral dilemmas beobtained.

Finally, the present study represents anew approach to the study of developmentalprocesses in social inference. Recently,Shantz (Note 1) suggested that develop-

MORAL JUDGMENT 665

mental researchers attempt to apply theoriesdeveloped with adults to social inferenceprocesses in children. The experiments pre-sented above suggest that algebraic mod-els developed in research with adults pro-vide a useful approach to the description ofdevelopmental processes in social inferencetasks. Moral judgments are just one of manytypes of social inference in which peopleengage. The major concern of develop-mental research is often the extent to whichchildren of different ages use various typesof information. Since algebraic models ofinformation integration are well suited to an-swer such questions, future research mightprofitably examine developmental trends ina variety of social inference tasks using al-gebraic models that have been successful inadult research.

REFERENCE NOTE

1. Shantz, C. U. Views on social-cognitive develop-ment of children and adults. Paper presented at theannual meeting of the American Psychological As-sociation, Chicago, September 1975.

REFERENCES

Anderson, N. H. Primacy effect in personality impres-sion using a generalized order effect paradigm. Jour-nal of Personality and Social Psychology, 1965, 2,1-9.

Anderson, N. H. Integration theory and attitudechange. Psychological Review, 1971,75, 171-206.

Anderson, N. H. Looking for configurality in clinicaljudgment. Psychological Bulletin, 1972,78, 93-102.

Anderson, N. H. Information integration theory: Abrief survey. In D. H. Krantz, R. C. Atkinson, R. D.Luce, & P. Suppes (Eds.), Contemporary develop-

ments in mathematical psychology (Vol. 2). SanFrancisco: Freeman, 1974.

Bandura, A., & McDonald, F. Influence of social rein-forcement and the behavior of models in shapingchildren's moral judgments. Journal of Abnormaland Social Psychology, 1963,67, 274-281.

Bearison, D., & Isaacs, L. Production deficiency inchildren's moral judgments. Developmental Psy-chology, 1975,//, 732-737.

Birnbaum, M. H. Morality judgments: Tests of an av-eraging model. Journal of Experimental Psychology,1972, 93, 35-42.

Birnbaum, M. H. The nonadditivity of personality im-pressions. Journal of Experimental Psychology,1974, 102, 543-561.

Chandler, J. D. Subroutine STEPIT—Finds localminima of a smooth function of several parameters.Behavioral Science, 1969, 14, 81-82.

Costanzo, P. R., Coie, J. D., Grumet, J. F., & Farnill,D. A reexamination of the effects of intent and con-sequence on children's moral judgments. Child De-velopment, 1973,44, 154-161.

Feldman, N. S., Klosson, E. C., Parsons, J. E., Rholes,W. S., & Ruble, D. N. Order of information presenta-tion and children's moral judgments. Child Develop-ment, 1976, 47, 556-559.

Hebble, P. W. The development of elementary schoolchildren's judgment of intent. Child Development,1971,42, 1203-1215.

Kanouse, D., & Hanson, L. R. Negativity in evalua-tions. In E. Jones, D. Kanouse, H. Kelley, R. Nis-bett, S. Valins, & B. Weiner (Eds.),Attribution: Per-ceiving the causes of behavior. Morristown, N.J.:General Learning Press, 1971.

Lane, J., & Anderson, N. H. Integration of intentionand outcome in moral judgment. Memory & Cogni-tion, 1976,4, 1-5.

Parsons, J. E., Ruble, D. N. Chereskin, E. P.,Feldman, N. S., & Rholes, W. S. Order effects onchildren's moral and achievement judgments. Devel-opmental Psychology, 1976,12, 357-358.

Piaget, J. The moral judgment of the child. New York:Free Press, 1965.

Weiner, B., & Peter, N. A cognitive-development anal-ysis of achievement and moral judgments. Develop-mental Psychology, 1973,9, 290-309.

(Received December 28, 1976)