Research was conducted to explore the effect of the favorable or unfavorable element of behaviors and its impact on behavior recall. Participants were introductory psychology students at Drake University receiving extra course credit for their participation. To examine this element of recall, participants were randomly assigned to 1 of 4 conditions and were asked to read a list of 18 inconsistently valenced behavior descriptions (3 behaviors exemplifying one trait, 6 behaviors exemplifying a second trait, and 9 behaviors for the third trait) of a hypothetical person, "John". When later asked to recall these behaviors, it was hypothesized that participants would be more likely to recall negative behaviors regardless of set size due to an individualís tendency to draw their attention towards negative stimuli faster than positive stimuli. However, results showed that participants were most likely to recall behaviors in the smallest set size when the behaviors were distinctive from the other 15 behaviors of the list. In other words, participants recalled a significantly higher percent of behaviors when the 3-behavior set was either positive or negative and the 6-behavior and 9-behavior set were opposite in valence. More specifically, the negative 3-behavior set was recalled significantly more often. The implications of these findings demonstrate how when a person is confronted with an individual mostly exemplifying one type of behavior (positive or negative) and is then exposed to only a few opposite behaviors, the individual is more likely to remember the few distinctive behaviors.

A small amount of previous research in social cognition has studied memory and set-size effects. Gordon & Wyer (1987) examined the relationship between category set-size and the recall of a personís behaviors. In their study, they presented participants with a list of 18 behaviors in which 3 behaviors represented one trait, 6 behaviors represented a second trait, and 9 behaviors represented the third trait. After being given a short description of a hypothetical person, "John", subjects were asked to read through their list of behavior descriptions and imagine the sort of impression they would have about John. An intervening task (a geography test) was given to the participants followed by a surprise recall task where participants were asked to recall as many behaviors as they could from their first list. Gordon & Wyer (1987) hypothesized that when the behaviors were organized in terms of trait concepts the participants would be more likely to recall behaviors the smaller the set size in which they were found. Their results showed that when the behaviors were consistent with one another (the traits they exemplified were similar in favorableness) participants were more likely to recall the smallest set size. Yet contrary to their hypothesis, when the behaviors were inconsistent with one another (the traits they exemplified were dissimilar in favorableness) they found that the likelihood of recalling the behaviors increased as the number of other behaviors in the same category increased. However, Gordon & Wyer only analyzed the set size in which the behaviors were found and not the favorableness or unfavorableness of the actual behaviors.

To take a closer look at how the favorableness or unfavorableness of a stimulus can contribute to where a participantís attention is being drawn we can take a look at a study by Hansen & Hansen (1988) on finding faces in a crowd. Previous research done in this area has discussed theories of automatic processing versus attentive processing and the filtering of information (Bargh, 1984; Logan, 1979; and Shiffrin & Schneider, 1977 as cited in Hansen & Hansen, 1988). According to this research, attentive processing is said to be less rapid than automatic processing in that automatic processing is subject to less distraction therefore, having less to slow down the processing. Automatic processing can be viewed as a way in which a person filters out information where some features are "passed through" and others are "filtered out". In processing where a feature is passed through, more attention is given to it than features that are just filtered out since the feature has come to attention. Therefore, one would assume that if an individual whose attention is being drawn to a stimulus where a feature was passed through the feature would appear to "pop out" from the entire stimulus as it catches the individuals attention more than the other features (Treisman, 1982; Treisman & Gelade, 1980; Treisman & Paterson, 1984; and Treisman & Souther, 1985 as cited in Hansen et al., 1988).

Applying these theories to face-processing, which shows evidence of angry faces being more attention-grabbing than happy faces (Schwartz, Izard & Ansul, 1985 as cited in Hansen et al., 1988), Hansen et al. hypothesized that if a participant was given a task of surveying a crowd for a discrepancy the participant would be more likely to detect an angry face in a crowd of happy faces faster than a happy face in a crowd of angry faces. Their results supported this hypothesis in that they found participants identifying the angry face in the crowd of happy faces significantly faster than the happy face in the crowd of angry faces.

One may ask what the implications of the Hansen et al. (1988) study would be on set-size effects shown by Gordon & Wyer (1987). If a majority of the behaviors were positive and only a few were negative would the few negative behaviors be more likely to be recalled since they may "pop out" of the behavior list? Or does the favorable or unfavorable element of the behavior have no effect on what behaviors a participant will recall? To examine this more closely the Gordon & Wyer study was re-created however, the statistical analysis focused on the positive or negative element of the behaviors recalled. I hypothesized that when a list of 18 inconsistent behaviors is presented to participants a negative set-size effect will occur when the smallest set size consists of negative behaviors due to the findings that the negative stimuli should be recognized more quickly than positive stimuli.

Subjects

A total of 63 introductory psychology students at Drake University participated in this study to receive extra credit towards their introductory course.

Stimulus Materials

Subjects were given a list of 18 behavior descriptions with the instructions to form an impression of a person described by the experimenter. The hypothetical person was John Robertson, a 20-year-old at Harrison Community College who was working at the campus bookstore. Subjects were told that John was required as an employee of the school to show around a prospective student. The list of behaviors the participants were presented with were behaviors that the prospective student observed John participating in throughout the day.

The behavior descriptions were representative of four different traits (dishonest, lazy, helpful and generous), yet each list only contained three of the four traits. In Condition 1, participants were presented with a behavior description list of 3 helpful behaviors, 6 generous behaviors, and 9 dishonest behaviors. Participants in Condition 2 were given a list of 3 helpful behaviors, 6 lazy behaviors, and 9 dishonest behaviors. Condition 3 contained 3 dishonest behaviors, 6 lazy behaviors, and 9 helpful behaviors. Condition 4 consisted of a list of 3 dishonest behaviors, 6 generous behaviors, and 9 helpful behaviors. In all conditions, the group of 9 behaviors was always inconsistent with the 3 behaviors (i.e., if the 3 behaviors were negative then the 9 behaviors were positive, and vice versa). The experimenter created each of the behavior descriptions. There was no pre-testing done on how well each of the behavior descriptions represented the trait it was assumed to exemplify. Examples of each trait are provided in Table 1.

A filler task was used after the presentation of the behavior description lists. This task was a 15-word standard anagram.

Procedure

Participants were randomly assigned to 1 of the 4 conditions. As each subject entered the classroom they were handed 3 sheets of paper. Their Informed Consent was face up on the top, and underneath (face down) was their Behavior List and Word Jumble. Subjects were seated every other seat in every other row around the classroom so that the difference in behavior lists was not noticed among the participants. After the experimenter read aloud the brief description of John Robertson, all participants were asked to turn over the top sheet on their desk which was the list of 18 different behavior descriptions of behaviors that John was observed participating throughout the day by the prospective student. On the top of each of the lists were the instructions, "Throughout the day, the prospective student observed John participate in the following behaviors. Please think of the sort of impression you would form of John based on these 18 behaviors." All participants were given 2_ minutes to read through the behavior list and form an impression.

When the 2_ minutes were finished, participants were asked to turn their lists over and hand them back to the experimenter. This was to avoid any participants referring back to the list during the Behavior Recall task. After all behavior description lists were turned in, participants were asked to turn the next piece of paper on their desk over, which was a Word Jumble. All participants were given 8 minutes to complete as many of the 15 scrambled words as they could. When the 8 minutes had passed participants were asked to turn their Word Jumble face down and to recall as many of the behaviors from their original behavior description list as they could on the back of their Word Jumble. Subjects were given as much time as needed to recall the behaviors.

Scoring

The free-recall behavior lists were scored on the basis of "gist" by the experimenter. That is, the responses were scored correct if the participant communicated the basic idea of the original behavior description even if the wording of the recall was different. The recall items that were scored as inaccurate were ones characterized in abstract terms, or by recalling the recognized trait of the behavior (i.e., "helpful" instead of "Helped his roommate put a new stacking shelf together"). Recall items that were not an actual behavior or the "gist" of a behavior from the original list were also counted as inaccurate. Of the 63 participants, 5 participantís data were not included in the analysis either due to not following directions on the behavior recall task or completing the tasks out of order, which in turn did not give the participants the same amount of time between the stimulus presentation and the behavior recall. Therefore, of the 63 participants, only 58 participantís data were used for the statistical analysis.

Valence Effect

There was no main effect found for the valence of the 3 behaviors (valence 3) on percent of overall behaviors recalled, F(1,54)=.14, p=.71, power=.05 or the valence of the 6 behaviors (valence 6) on percent of overall behaviors recalled, F(1,54)=.04, p=.85, power=.04. Yet there was a marginally significant effect of the 2x2 interaction between valence 3 and valence 6, F(1,54)=3.60, p=.06. Although there was this medium-size effect (Cohen, 1988), power = 0.46 shows that with this particular effect size and sample size the test was close to being a fair test, but just missed the mark.

To examine this effect more closely, the means of valence 3 and the means of valence 6 were collapsed across set size (see Table 2). Prior to the t-test evaluation, the means showed that when valence 6 was negative and valence 3 was positive the highest percent of behaviors were recalled. The second highest percent of behaviors recalled occurred when valence 6 was positive and valence 3 was negative. In both cells where valence 6 and valence 3 were the same (either both positive or both negative) the smallest percent of behaviors were recalled. However, t-tests were conducted among means to locate the significant differences and showed that the difference between the inconsistent valences (valence 6 was positive and valence 3 was negative, and vice versa) were not significant (t=.89).

Set Size Effect

To examine the possible effect of the number of behaviors in each set size on the percent of behaviors recalled, an added independent variable ("number") was created which was a within subjects variable comparing the percent of the 3 behaviors recalled (percent 3) by the percent of the 6 behaviors recalled (percent 6) by the percent of the 9 behaviors recalled (percent 9). Analysis confirmed a significant main effect of number of behaviors in a set on percent of behaviors recalled, F(2,108)=23.09, p<.0001, partial eta squared = .30, power =1.00. T-tests between each pair of means (which are shown as a percent of behaviors recalled in the specific cell) were conducted to locate the significant differences. All differences were found to be significant, such that percent 3 was recalled more than percent 6 (t=12.50, p<.01), percent 3 was recalled more than percent 9 (t=9.96, p<.01), and percent 9 was recalled more than percent 6 (t=2.54, p<.05). Therefore, overall the behaviors recalled the most in comparison with all 3 set sizes was percent 3.

Results also showed a significant main effect of valence 3 by number of behaviors in a set on percent of behaviors recalled, F(2,108)=3.84, p<.024, partial eta squared = .06, power=.69 (see Table 3). In this 2x3 the means are shown for valence 3 in each set size. T-tests were conducted to assess the differences among means. Results showed that when valence 3 was negative (M=73.81) the percent of behaviors recalled in set size 3 was significantly more than when valence 3 was positive (M=66.67) in the same set size (t=4.03, p<.01). While valence 3 was still negative the percent of behaviors recalled in set size 3 were also recalled significantly more than when in set size 6 (M=47.02, t=15.14, p<.01) and set size 9 (M=46.83, t=15.24, p<.01). The second highest percent of recalled behaviors was when valence 3 was positive and in set size 3 (M=66.67). They were recalled significantly more than when valence 3 was positive and in set size 6 (M=48.89, t=10.05, p<.01) and set size 9 (M=57.78, t=5.02, p<.01). Significant effects were also found when valence 3 was positive and in set size 6 (M=48.89). The percent of behaviors recalled were less than when in set size 9 (M=57.78, t=5.02, p<.01). The percent of behaviors recalled in set size 9 were significantly greater when valence was positive rather than negative (M=46.83, t=6.19, p<.01). All other effects were found to not be significant.

A statistical analysis was also performed on valence 6 by number of behaviors in a set on percent of behaviors recalled. There was not a significant effect on the behaviors recalled, F(2,108)=1.04, p<.358, partial eta squared = .02, power=.23.

In addition, there was a significant main effect of valence 3 by valence 6 by number, F(2,108)=7.37, p<.001, partial eta squared = .12, power=.93 (see Table 4). The means (which are shown as a percent of behaviors recalled in the specific cell) are shown for each set size for valence 3 and valence 6. Similar to the 2x2 interaction between valence 3 and valence 6 the highest percent of behaviors recalled was found when in set size 3, valence 6 was positive, and valence 3 was negative (M=85.71). The second highest percent was found in the same set size when valence 6 was negative and valence 3 was positive (M=77.08). T-tests were conducted to assess the differences between the means. Results showed that when examining the highest and second highest percents, when valence 3 was negative and valence 6 was positive a significantly higher percent of behaviors was recalled than when valence 3 was positive and valence 6 was negative (t=4.88, p<.01). More extensively, when valence 3 was positive and valence 6 was negative a significantly higher percent of behaviors were recalled in set size 3 than when in set size 6 (t=22.86, p<.01) and set size 9 (t=21.52, p<.01). The same condition was found to contain a higher percent of recalled behaviors than when both valence 6 and valence 3 were negative and found in set size 3 (t=13.45, p<.01), in set size 6 (t=20.85, p<.01) and in set size 9 (t=22.42, p<.01).

As previously stated, the second highest percent of behaviors recalled was found in set size 3 when valence 6 was negative and valence 3 was positive. T-tests were once again conducted to assess the differences in this condition with the other means. Results showed that in set size 3 when valence 6 was negative and valence 3 was positive a significantly higher percent of behaviors were recalled than when in set size 6 (t=13.53, p<.01) and in set size 9 (t=12.44, p<.01). This same condition was also found to contain a higher percent of recalled behaviors than when valence 6 was positive and valence 3 was negative in set size 3 (t=12.61, p<.01), in set size 6 (t=18.66, p<.01), and in set size 9 (t=9.47, p<.01).

The results of this study showed a set-size effect when a list of 18 inconsistent behavior descriptions were presented to participants and later recalled. These results conflict with that of Gordon & Wyer (1987) who concluded that when the 18 behaviors were recalled, the behaviors in the largest set size were most likely to be recalled. In my analyses, a marginally significant 2-way interaction was found between the valence of the 3 behaviors and the valence of the 6 behaviors on the percent of behaviors recalled overall. According to the t-tests conducted, when the valences are inconsistent between the 3 and the 6 (and 9), a larger percent of behaviors are recalled. In these conditions, the valence of the 3 behaviors is the opposite of both the 6 and the 9 behaviors of the same list of 18 behavior descriptions. On the other hand, when the valences are consistent with one another a smaller percent of the behaviors are remembered. In these conditions, the valence of the 3 and 6 behaviors would be the same and the 9 behaviors would be inconsistent. Yet a negative set-size effect was not found in these conditions since the 3 behaviors were recalled the most and the 6 behaviors were recalled the least. This interaction has the potential to demonstrate that when the smallest set size consists of behaviors opposite in valence of the 6 set size and 9 set size of the same behavior list the distinctiveness of the 3 behaviors are most likely to result in the highest percent of recall. Unfortunately, due to a low power (power=.46) this effect with this sample size did not reach significance.

Therefore, we must take a closer look at the set size findings for the significance. The first significant effect found was that the highest percent of behaviors recalled over all 3 set sizes was found in the smallest set consisting of 3 behaviors. The second highest percent of recalled behaviors was found in the 9-behavior set leaving the 6-behavior set as having the least percent of recalled behaviors. Yet, there remains a question as to which valence of the 3-behavior set is causing this significant effect. For that reason a 2x3 was analyzed in which the percent of behaviors recalled were laid out in all three set sizes according to the valence of 3. The analysis showed that in the condition where valence 3 was negative in the 3-behavior set size the highest percent of behaviors were recalled. In comparison with all of the other means (in the 6-behavior set size and the 9-behavior set size) the second highest percent was also found in the smallest set size in which valence 3 was positive. These results are beginning to show a negative set-size effect in comparison with the results of the Gordon & Wyer (1987) study.

However, an even more extensive analysis needed to be conducted to display in which conditions (of the 4 stated in the Stimulus Materials) the 3 behaviors were recalled the most. Therefore, a 2x2x3 analysis was examined consisting of valence 6, valence 3, and the number of behaviors in a set. Similar to the 2x2 interaction between valence 3 and valence 6 the highest percent of behaviors recalled were found when in set size 3, valence 6 (and 9) was positive, and valence 3 was negative. In addition, the second highest percent was found in the same set size when valence 6 (and 9) was negative and valence 3 was positive. As proposed earlier, these results tend to show that the behaviors recalled the most are distinctive from the other 15 behaviors in the original behavior description list. Therefore, we can concluded that in a group of inconsistent behaviors (more specifically a group of behaviors in which the majority is either positive or negative and the minority is the opposite), the behaviors most likely to be recalled are the ones distinctive from the rest of the group (see Condition 1 and Condition 4 in Stimulus Materials). This supports the original hypothesis in that the smallest negative set size contained the highest percent of recalled behaviors. Yet, this finding may not have been a result of the unfavorableness of the behavior (as it was hypothesized) but rather it was a result of the behaviorís distinctiveness among the entire list of behaviors.

These results may have implications for real-life situations in that if a person is generally a positive person but is seen participating in a few negative behaviors, it is those few negative behaviors that others will remember the most. This possible implication is drawn from the idea that while the overall impression others may have of them would most likely be positive since they were the most prevalent behaviors, according to these results, it is the distinctive behaviors that will be recalled significantly more. The same can be said for generally negative individuals observed participating in a few positive behaviors. Yet much research still needs to be conducted so not to generalize these findings. One must take into consideration how stereotypes may play a role in the behavior recall, or even the possibility of preconceived sex roles contributing to the recall (i.e., John helped his mom plant flowers in her garden; Ann helped her dad fix the engine in their old car). Another possible area to be examined would be how the distinctive behaviors recalled would effect impression formation if the original overall impression was positive and the target person was participating in negative behaviors, and vice versa, as shown above.

For future consideration, a few confounds were found in this study which would need to be eliminated for stronger significant effects. There was no pre-testing done on how well each of the behavior descriptions represented the trait it was assumed to exemplify. To confirm that the behaviors were in fact completely negative or completely positive, raters should be used to identify the trait and valence relevance. Another confound identified was using the experimenter as the coder of all data due to a limit in time. Since this procedure may lead to a bias in coding, blind judges should be used, along with the examination of interjudge reliability, in scoring responses. These results have strong implications for behavior recall and possibly impression formation. Eliminating these confounds, in addition to improving significance through gathering more participants, would greatly improve the generalizability of these findings. Furthermore, examining the impact of stereotyping on behavior recall and set-size effects and exploring the relationship these findings may have with impression formation would also increase the validity of these results and its application to social interactions.

Table 1

Table 2

Means of Percent Behaviors Recalled of Valence 3 By Valence 6
 

		Valence 3
Valence 6		Negative	Positive
	Negative	52.25	61.92
	Positive	58.52	53.04

*all means are percents of behaviors recalled

Table 3

Table 4

Bargh, J.A. (1984). Automatic and conscious processing of social information. In R. S. Wyer, Jr., & T. K. Srull (Eds.), Handbook of social cognition (Vol. 3, pp. 1-43). Hillsdale, NJ: Erlbaum

Cohen, J. (1988). Statistical Power Analyses for the Behavioral Sciences. Hillsdale, NJ: Erlbaum.

Gordon, S. A. & Wyer, R. S. (1987). Person Memory: Category-Set-Size Effects on the Recall of a Personís Behaviors. Journal of Personality and Social Psychology, 53(4), 648-662.

Hansen, C. H. & Hansen, R. D. (1988). Finding the Face in the Crowd: An Anger Superiority Effect. Journal of Personality and Social Psychology, 54(6), 917-924.

Logan, G. D. (1979). On the use of a concurrent memory load to measure attention and automaticity. Journal of Experimental Psychology: Human Perception and Performance, 5, 189-207.

Schwartz, G. M., Izard, C. E., & Ansul, S. E. (1985). The five-month-oldís ability to discriminate facial expressions of emotion. Infant Behavior and Development, 8, 65-77.

Shiffrin, R. M., & Schneider, W. (1977). Controlled and automatic human information processing : II. Detection, search, and attention. Psychological Review, 84, 127-190.

Treisman, A. (1982). Perceptual grouping and attention in visual search for features and for objects. Journal of Experimental Psychology: Human Perception and Performance, 8, 194-214.

Treisman, A., & Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology, 12, 97-136.

Treisman, A., & Paterson, R. (1984). Emergent features, attention, and object perception. Journal of Experimental Psychology: Human Perception and Performance,10, 12-31.

Treisman, A., & Souther, J. (1985). Search asymmetry: A diagnostic for preattentive processing of separable features. Journal of Experimental Psychology: General, 114, 285-310.