Number averaging behavior: A primacy effect

Number averaging behavior: A primacy effect CLYDE HENDRlCK and ARTHUR F. COSTANTINI, Kent State University, Kent, Ohio 44240 In two experiments, Ss guessed averages of serially presented numbers. Guesses were made only at the end of a stimulus sequence. The results of both experiments showed strong primacy effects in the guesses; earlier information in the stimulus sequence was weighted more heavily than later information. The results for number averaging were comparable to results obtained for personality traits in previous research, suggesting that both tasks represent a more general problem of integration of serial information. All the studies discussed above, except Anderson & Norman (1964), used trait adjectives as stimuli. Anderson (1964) introduced a number-averaging task that modeled the trait impression paradigm. Two-digit numbers were read to Ss in serial order. Ss were required to guess the average of the numbers to that point as each number was presented. This "continuous mode" of response was analogous to the continuous mode used by Stewart (1965) to obtain evaluation ratings after each trait was presented. Anderson (1964) obtained a strong recency effect for the number stimuli with this mode of response, as did Stewart (1965) with the trait stimuli. In the context of information integration, numbers have a distinct advantage as stimuli. In asense, numbers have their own "buiIt-in" scale values. This attribute offers the possibility that m ultivariate or analysis of variance methods may be used to determine conveniently weights for serial positions of stimulus sequences. Determination of weights wiII eventually become important to help choose between competing models of information integration. The two experiments that are reported had more Iimited objectives. The fust objective was to determine if a primacy effect would be obtained with a number-averaging task when a final mode of responding at the end of the sequence was used. Since primacy effects have always been obtained with trait stimuli when the final mode is used, a primacy effec! was predicted for the number stimuli also. The second objective was to determine what effect, if any, variation in Since Asch's (1946) c1assic work on impression formation, several experiments have found a primacy effect when Ss are asked to evaluate a person described by serially presented trait adjectives (e.g., Anderson, 1965). In a typical experiment, the E reads several sets of six or eight adjectives to a S and obtains a numerical evaluation after each set. A more favorable evaluation is usually obtained when the adjectives are presented in a desirable-undesirable sequence than when presented in the reverse sequence. The primacy effect has also been obtained with other c1asses of verbal stimuli, inc1uding food words (Anderson & Norman, 1964). Such results suggest that the personality impression task should be considered as an instance of the integration of serial information (Anderson, 1968a), rather than the study of personality impression formation per se. This broader conception directs attention toward the process of serial information processing and its determinants. The research to date indicates that the primacy effect is obtained only under a Sets special set of experimental conditions. Hendrick & Costantini (in press) conc1uded A that a primacy effect may be generally ("B expected only when one final evaluative D response (called the "fmal mode" of E response) is given to the stimulus set after F it is presented. When other response requirements are imposed, either no effect or a recency effect is obtained. When evaluative responding was required after each trait word (Stewart, 1965), when Sets recall of the stimuli was required A (Anderson & Hubert, 1963), or when B pronunciation of the stimuli by the Ss was D E required (Anderson, I 968b; Hendrick & F Costantini, in press), recency effects were obtained. Psyehon. Sei., 1970, Vol. 19 (2) number size within stimulus sets had on the magnitude of the primacy (or recency) effec!. EXPERIMENT I The stimuli consisted of six sets of numbers shown in the top panel of Table I. Each set consisted of three high (H) numbers between 100 and 200, and three low (L) numbers between 5 and 50. The numbers were not selected entirely at random, because a requirement was imposed that all six sets have the same me an in order to ensure relatively simiIar sets. By random choice, a sum of 575 for the six numbers was se1ected, yielding a mean of 95.8. Given this requirement, the numbers were selected at random insofar as was possible. Each stimulus set was presented in both an HL and an LH order. In addition, there were eight filler sets consisting of Hand L numbers mixed in random order. The filler sets prevented Ss from developing an expectancy that the numbers would always be presented in an orderly HL or LH sequence. The Ss were 10 males and 7 females from a section of introductory psychology. The data were collected from each S individually. The experiment was presented to the Ss as a study in human information processing. Their job was to listen to the E read off sets of six numbers and to form an impression of the "psychological average." It was stressed that accuracy was not important-the impression of what the average might be was the desired response. The E read the numbers at a steady rate of one number every 2 sec. The S was required to respond within 5 sec after the E read the sixth number. Each set of numbers was printed on aseparate eard. The deek of cards, including both fiDer and experimental sets, was shuffied randomly f~ each S, except that an HL and LH Table I Number Sets, Guessed Means, and Primacy-Recency in the Two Experiments Stimuli Experiment Guessed Means HI H2 H3 LI L2 L3 HL LH Diff. 151 183 189 146 163 191 128 169 146 161 188 159 198 175 125 154 199 153 35 20 26 36 07 17 46 18 40 32 13 46 17 10 49 46 05 09 104.7 101.2 95.0 93.8 108.9 106.6 90.7 80.6 73.6 90.8 89.3 85.5 +14.0 +20.6 +21.4 +3.0 +19.6 +21.1 Mean 101.7 85.1 +16.6 Experiment 2 High D iscrepancy Low Discrepancy HL LH Diff. HL LH Diff. 97.2 102.9 90.5 105.4 105.7 88.8 89.8 82.7 85.4 84.9 +8.4 +13.1 +7.8 +20.0 +20.8 95.9 92.5 91.9 93.5 94.3 86.2 81.8 88.3 88.1 82.7 +9.7 +10.7 +3.6 +5.4 +11.6 100.3 86.3 +14.0 93.6 85.4 +8.. 2 121 order of a given set could not follow each other consecutively. EXPERIMENT 2 The general procedures were the same as in Experiment I. The main difference was in the stimulus sets. The size of the numbers used to construct sets was systematically varied. The main concern was whether the difference between the sum (or average) of the three H numbers and the three L numbers in a set might affect the magnitude or even the direction of the order effect. In order to vary size of the numbers, two types of number sets were constructed. One type contained numbers with a relatively large difference between the average of the H an (...truncated)