Journal of Abnormal and Social Psychology 1961, Vol. 63, No. 2, 346-360

PRIMACY EFFECTS IN PERSONALITY IMPRESSION FORMATION1 NORMAN H. ANDERSON AND ALFRED A. BARRIOS University of California, Los Angeles

TUDIES in communication research have reached no consensus as to the effect of order of presentation. The pro-con vs. con-pro paradigm leads sometimes to primacy effects (greater effect of the first communication) and sometimes to recency effects. Although some progress has been made in the study of the relevant variables (Anderson, 1959; Hovland et al, 1957a; Luchins, 1958; Miller & Campbell, 1959), the situation is far from clear. One limitation of most work is the small number, typically two, of communications involved. Two experiments (Anderson, 1959; Weld & Roff, 1938) have used reasonably long sequences of communications on a single topic. However, no work has been done to assess practice effects over a sequence of communications on separate topics. Both lines of attack are desirable, not only because they simulate everyday situations more closely, but also for theoretical reasons. Thus, Hovland (1957b) has emphasized some of the ways in which prior familiarity with the topic might influence reception and acceptance of the communications. Analogous considerations would apply when a sequence of different topics is used, and various practice effects might also be important. The two cited paradigms would thus be expected to be useful in bringing the relevant psychological processes under closer experimental scrutiny. The present experiment was designed to study order effects over a sequence of communications on separate issues. The classic paper of Asch (1946) suggested the use of personality adjectives in order to get a large body of relatively homogeneous material. Asch's results, as well as those of Luchins (1957, 1958), indicated that under the specific experimental conditions employed here, primacy effects would be obtained for the initial trials. However, it was thought that with continued practice the primacy effect would decrease and perhaps become a recency effect.

S

1 This research was supported by Grant M-3502(A) from the National Institute of Mental Health.

346

METHOD Subjects. Volunteers who were fulfilling a class requirement in introductory psychology were assigned randomly to the various conditions within each experiment. Sex was balanced in Experiment 1 but not in Experiment 2. The JVs for the two experiments were 64 and 24, respectively. Subjects were examined individually. Stimuli. A set of 328 adjectives descriptive of personality characteristics were picked on the basis of familiarity by the experimenter and were then given a rough scaling based on the responses of 10 judges. The judges rated each adjective on a six-point FavorableUnfavorable scale which had an additional "X" rating for unfamiliar words. The 25 adjectives that received an X were not used. The mean ratings of the remaining adjectives ranged from 0.1 to 4.8, with a rating of 5 corresponding to Very Favorable. The 48 adjectives in the range 4.0-4.3 will be called H; the 48 adjectives in the range 1.2-2.S will be called L. Table 2 shows some of the adjectives that were used. Procedure. Subjects were told that they would be read a number of sets of adjectives, each set describing a different person, and that they should try to form an impression of the kind of person the adjectives described. The instructions indicated that they were to think of the six adjectives as having been given by six different people who knew that person well. After the experimenter read each set, the subject indicated his response numerically in terms of a rating scale typed on a card in front of him. This eight-step scale ranged from +4 to —4, the neutral response being disallowed. An identifying label was placed by each number using the various combinations of Favorable and Unfavorable, with Highly, Considerably, Moderately, and Slightly as modifiers. Experiment 1. Each subject judged 61 or 62 sets of six adjectives each. One of these sets was the same as that used by Asch (1946, Experiment VI). It was used as the first set for half the subjects, and as the last set for the other half. Within each of these halves, the set was given in forward order to half the subjects and in reverse order to the other half. Those subjects who received this set as their first set also received it as their last set. The remaining sets were of five types. Type HL consisted of three adjectives in the H range followed by three adjectives in the L range of scale values. Type LH had three L adjectives followed by three H adjectives. Type GD was a sequence of six adjectives gradually descending from H to L in scale value. Type GA gradually ascended from L to H. Type R consisted of six adjectives chosen similarly to the other sets but arranged in random order within a set. Twelve sets of each type were constructed randomly subject to the restriction that no adjective appear twice in any set and that each adjective be used about equally often.

347

PRIMACY EFFECTS IN IMPRESSION FORMATION Six ordered blocks of 10 sets each were constructed randomly subject to the restriction that two sets of each type appear in each block. Four sequences of 60 sets were obtained using four different permutations of the six blocks. From these four sequences, four additional sequences were formed by reversing the order of the adjectives in the 60 sets. The GD, GA, and R sets were completely reversed in order. However, the H and L subsets in the HL and LH sets were simply interchanged without disturbing the order of the three adjectives within either subset. The adjectives of each set were read by the experimenter at an approximate rate of one adjective each 3 seconds. Three trials were given each minute until all the sets had been judged. Experiment 2. Each subject judged 90 sets of two adjectives each, 30 HL, 30 LH, 15 LL, and 15 HH. These sets were constructed as in Experiment 1 and balanced for type in ordered blocks of 30 sets each. Half the subjects received the sets in one sequence; the other half received the same sequence but with the order of the two adjectives in the LH and in the HL sets reversed. The main independent variable was the time between the reading of the first and second adjectives of a set. Each subject judged one block of 30 sets at each of three time intervals: 0, 2, and 4 seconds. All six possible sequences of time intervals were used in a latin square design. Successive blocks were separated by 1 minute during which the subject was told that a new tempo would be used for the following block.

1

2

3

20 T R I A L B L O C K S

FIG. 1. Mean primacy scores as a function of sex, type of set, and trial blocks. (Experiment 1.)

TABLE 1 ANALYSIS OF VARIANCE OF ORDER EFFECT SCORES SUMMED OVER ALL TRIALS Source

Mean

Sex

Error (b) Type Sex X Type Error (w)

V

P

1 1 62 1

48.38* 2.23 (183.81)*

1 62

8.03* (83.77)

.02

Note. —Error mean squares in parentheses. * # < .05.

RESULTS Experiment 1. If an HL set produces a higher response than the corresponding LH set, then the first subset of three adjectives had a stronger effect than the second subset, within at least one of the two sets. Positive HL-LH differences, and positive GD-GA differences thus represent primacy effects. These two difference scores were computed for each subject using the several adjectives of each type in a given block of trials. The results are shown in Figure 1 which plots mean difference scores for the four main experimental conditions as a function of trial blocks. It is seen that there is a strong primacy effect which equals 0.69 averaged over all conditions. Although the effect decreases somewhat over trials, it does not appear to be approaching zero. Females show more primacy than males but this sex difference resides largely in the HL-LH sets. Indeed, the sexes are not too far apart on the GD-GA curves and, compared to these, the males are lower and the females are higher on the HLLH curves. The analysis of these data was performed on the difference scores summed over all 60 trials and is given in Table 1. Since a difference score was used, the F for Mean shows a significant primacy effect. Although the main effect of Sex is not significant, the significant Sex X Type interaction verifies the comments on sex differences made in the preceding paragraph. A trend test showed that the decline of the primacy effect with trials was significant (F = 12.27, df = 1/60). The absolute responses, as distinguished from the difference scores, are also of interest. The mean response over all descending sets

348

NORMAN H. ANDERSON AND ALFRED A. BARRIOS TABLE 2 MEAN RESPONSE TO Six SETS SHOWING STRONGEST PRIMACY EFFECTS Descending Ascending order order

Words

GD HL GD HL GD HL

smart, artistic, sentimental, cool, awkward, faultfinding determined, tolerant, gentle, stubborn, forgetful, tricky orderly, entertaining, humble, cool, calculating, moody efficient, scholarly, smart, crafty, faultfinding, unruly warm, independent, optimistic, reserved, tough, faultfinding patriotic, perfective, generous, softhearted, uninhibited, humorless

(HL and GD) was 0.94, and the mean response over all ascending sets (LH and GA) was 0.25. This latter value is quite close to the mean response of 0.32 for the R sets. Since the adjectives for the R sets were from the same pool as for the other sets, the asymmetry of these results suggests that the primacy effect has its source in the initial adjectives of the descending sets. However, this suggestion must be viewed with caution since, although the HL-LH and GD-GA order effects were almost identical, the mean response to GA and LH sets was 0.48 and 0.02, respectively. Results from the set used by Asch (1946) were consistent with the above. When this set was given first, the means were 0.87 and —0.94 for the GD and GA orders, respectively. The primacy effect of 1.81 was significant (F = 6.87, df = 1/30). The primacy effect was just half as large when the set was given last and this was nonsignificant, even when all 64 subjects were used. Of the 48 critical sets used, all but 5 gave a primacy effect. The descending versions of the 6 sets which showed the greatest overall mean primacy effect are listed in Table 2. Experiment 2. The overall mean scores for the HH, HL, LH, and LL sets were 2.68, 0.80, 0.76, and —1.27, respectively. The analysis of variance was performed on the HL-LH scores computed for the three successive blocks of 30 trials. Neither time interval nor trial blocks approached significance. The 95 % confidence interval for the overall mean order effect was 0.04 ± 0.12. It thus seems reasonable to conclude that the true order effect is rather small when only two adjectives are used. Supporting evidence is found by comparing Error(b) to Error(w) which tests for individual differences. The resulting

1.38 1.62

1.84 0.66 1.69 1.84

-0.72

-0.16 0.44 -0.66 0.41 0.59

Mean primacy

2.10 1.78 1.40 1.32 1.28 1.25

F (df = 17/36) was 1.19. Because of the great power of the F test with this many df, it seems reasonable to conclude that true individual differences in the order effect are also quite small when only two adjectives are used. DISCUSSION The primacy effect that was found is striking. Over the first block of 10 trials, the mean difference in response obtained from simply reversing the order of the six adjectives was 1.12. This is a respectable part of the eightpoint scale, especially since few subjects used the full scale range. Moreover, the result is not due to some peculiarity of the particular sets used since they were constructed randomly and since a primacy effect was observed in 43 of the 48 sets. However, it is not clear how far the results may be generalized beyond the experimental situation used here. The need for some caution in this respect is indicated by the work of Luchins (1957, 1958) who finds that the order effect changes with variations in the procedure. The decline in primacy over the later trials has a number of possible causes. It may be that, despite the use of the random sets, the pattern of good and bad words in the remaining sets brings about an increased tendency to take account of all the words in each set. This possibility could perhaps be tested by reducing the relative frequency of the critical sets. A progressive loss of interest in the task might also influence the primacy effect, a hypothesis which would be testable by inserting a rest and motivating instructions partway through the session. A third possibility is that adaptation to the experimental situation and practice in integrating the material are the governing factors. If so, the decrement in primacy found here indicates

PRIMACY EFFECTS IN IMPRESSION FORMATION that caution should be used in generalizing results of studies employing only one or two communications. The results of an earlier experiment (Anderson, 1959) were interpreted as suggesting the existence of two opinion components. The basal component, once formed, was quite resistant to change whereas the surface component was readily influenced by the successive communications. The present data would be consistent with the two-component hypothesis if it were assumed that a strong basal component was formed over the first three but not the last three adjectives within a given set. Thus for an HL set, the surface component induced by the three H adjectives would be approximately canceled by the surface component induced by the three L adjectives. However, the basal component induced by the three H adjectives would persist to produce a response toward the Favorable end of the scale. For the corresponding LH set, the basal component would tend to produce a response toward the Unfavorable end of the scale, and the difference between the two would constitute the primacy effect. The two-component hypothesis is thus not inconsistent with the present data but other explanations cannot be ruled out. In particular, the primacy effect may result from a progressive decrease in attention over the adjectives of a given set. Attention decrement could presumably be reduced by more stringent control or at least tested by asking subjects to recall the adjectives (McGuire, 1957). It should also be noted that in the present situation the basal component would be equivalent in effect to and perhaps explainable in terms of Luchins' (1957) Einstellung, or Asch's (1946) concept of direction. However, a two-component interpretation suggests a test which the other two formulations do not. For consider the order effect paradigm defined by the two sets, HLH'L' vs. HLL'H', in which the first six adjectives are the same in both sets and the last six differ only in order of presentation. According to hypothesis, only the surface component is affected over the last six adjectives and hence (Anderson, 1959) the paradigm should yield a recency effect. Although this prediction must be considered rather speculative, the use of this and similar

349

paradigms should prove valuable in the study of opinion and impression formation. Experiment 2 was designed on the assumption that a considerable crystallization of impression had occurred in the first half of the sets in Experiment 1. It was thought that the primacy effect would increase as increased time was allowed for the first adjective to sink in. Even though the reduction from six to two adjectives was expected to decrease the overall effect, the finding of negligible order effects in Experiment 2 came as a surprise in view of the strong primacy obtained in Experiment 1. The results of Experiment 2 thus give no evident support to Asch's (1946) hypothesis that the first adjective sets up a directed impression in terms of which the later adjectives are interpreted. However, the combined results of the two experiments give some basis for speculating that the critical events leading to primacy in Experiment 1 occurred at the second and third adjectives. SUMMARY Sets of Favorable and Unfavorable adjectives descriptive of general personality characteristics were constructed so as to test for effect of order of presentation of the adjectives. The sets were read to subjects who were asked how favorable an impression they had of the person described by the set of adjectives. In Experiment 1, 64 subjects each judged some 60 sets of six adjectives each. Strong primacy effects were found although there was some decrement over trials. Females showed greater primacy than males for sets in which the change from Favorable to Unfavorable (and vice versa) was abrupt. However, there was little sex difference when the change was gradual. In Experiment 2, 24 subjects each judged 90 sets of two adjectives each, with intervals of 0, 2, and 4 seconds between the adjectives. Time interval had no observable effect and order of presentation was also nonsignificant. REFERENCES ANDERSON, N. H. Test of a model for opinion change. /. abnorm. soc. Psychol, 1959, 69, 371-381. ASCH, S. E. Forming impressions of personality. /. abnorm. soc. Psychol, 1946, 41, 258-290. HOVIAND, C. I., MANDELL, W., CAMPBELL, ENID, H., BROCK, T., LUCHINS, A. S., COHEN, A. R., McGUIRE, W. J., JANIS, I. L., FEEERABEND, ROSALIND, L., & ANDERSON, N. H. The order of pre-

350

NORMAN H. ANDERSON AND ALFRED A. BARRIOS

sentation in persuasion. New Haven: Yale Univer. Press, 1957. (a) HOVLAND, C. I. Summary and implications. In C. I. Hovland et al., The order of presentation in persuasion. New Haven: Yale Univer. Press, 1957. Pp. 129-157. (b) LUCHINS, A. S. Experimental attempts to minimize the impact of first impressions. In C. I. Hovland et al., The order of presentation in persuasion. New Haven: Yale Univer. Press, 1957. Pp. 62-75. LUCHINS, A. S. Definitiveness of impression and primacy-recency in communications. /. soc. Psychol, 1958, 48, 275-290.

McGuiKE, W. J. Order of presentation as a factor in "conditioning" persuasiveness. In C. I. Hovland et al., The order of presentation in persuasion. New Haven: Yale Univer. Press, 1957. Pp. 98114. MILLER, N., & CAMPBELL, D. T. Recency and primacy in persuasion as a function of the timing of speeches and measurements. /. abnorm. soc. Psychol., 1959, 59,1-9. WELD, H. P., & ROIT, M. A study in the formation of opinion based upon legal evidence. Amer. J. Psychol., 1938, 51, 609-628. (Received October 6, 1960)