|Home | About | Journals | Submit | Contact Us | Français|
Is possession of desirable personality characteristics the only predictor that someone will be well-liked in a group of acquaintances, or will similarity to others in the group also matter? We tested participants (n=844) who had been randomly assigned to peer groups and had spent 6 weeks together. Participants assessed self and peer personalities in a round-robin design. We found that after controlling for attributions of desirable and undesirable personality characteristics, individuals with similar personality patterns liked each other more than individuals with dissimilar patterns. Further analysis revealed similarity of undesirable traits mattered more for liking than similarity of desirable traits. Results provide the first comprehensive analysis of relations between personality similarity and liking among acquaintances in a randomized, naturalistic design.
In every social network, individuals may find they connect with some colleagues and peers, but remain distant from others. Given opportunities to meet and interact with everyone, what determines who will be liked and who will not? At a fundamental level, people like others who they believe have desirable traits, and dislike others who they believe have undesirable traits. We call this tendency the fundamental principle of liking (FPL). Over and above the FPL, there may be more subtle factors, such as similarity of personality, that draw certain acquaintances together and push others apart. Researchers have investigated the role of personality similarity for over half a century, updating what was known about the magnitude of its influence (which appears to be small) as new theoretical frameworks (e.g., Social Relations Model, Kenny & Kashy, 1994) and statistical techniques became available (Cooper & Sheldon, 2002). The present study is the first to investigate the role of similarity in liking while controlling for the FPL using a randomized, naturalistic design.
The FPL is the notion that people like others to whom they attribute desirable personality traits (e.g., generosity, kindness) and dislike others to whom they attribute undesirable traits (e.g., arrogance, rudeness). In previous studies, there was often no way to know if similarity between personality characteristics predicted liking independent of the positive and negative personality traits people attributed to each other, perhaps because peer attributions of personality are rarely collected (Vazire, 2006). For example, there might be two good-humored individuals who like each other simply because funny people are easy to like, not because of their similarity per se. As a result, the effects of personality similarity may have been confounded with the effects of personality itself. In the current study, we circumvent this limitation by assessing personality similarity after controlling for the desirable and undesirable traits peers attribute to each other. We can thus ask the question: conditional on whether other people think individuals have desirable and undesirable traits, does similarity in participants’ self-reports predict more liking? Although a number of studies have inquired into the nature of personality similarity and liking, none that we know have been able to control for the confounding influence of the FPL.
When researchers investigate whether people with similar personalities like each other more, what exactly do they mean by similar personality? Researchers have typically defined similarity as either a difference between (or interaction of) mean scores on self-reports of a single personality dimension or, much less commonly, as a high positive correlation between the personality patterns of two persons across many traits. It is often not recognized that these two aspects of similarity in personality are potentially independent of each other (as noted by, e.g., Cronbach & Gleser, 1953 and Luo & Klohnen, 2005). Recent research that directly and extensively compared the two approaches—absolute mean differences and patterns—among a population of newlyweds found that the former approach was a much weaker predictor of relationship satisfaction than the latter (Luo & Klohnen, 2005). Thus, personality similarity plays a role in determining the relationship satisfaction of newlyweds, and prior research that failed to find this effect (see Klohnen & Mendelsohn, 1998 for a review) may have done so in part because it used a less consequential operationalization of similarity. Although studies of romantic relationships have shown that similarity of personality patterns predicts relationship satisfaction, romantic relationships are qualitatively different from other types of relationships; these findings will not necessarily generalize to groups of acquaintances (Eastwick, Finkel, Mochon, & Ariely, 2007).
In the current study, we attempt to disentangle the effect of pattern similarity from mean level similarity by measuring both in the same population of randomly assigned acquaintances. To measure similarity of patterns, we compute correlations between the profiles of pairs of individuals, so the magnitude of the correlation is independent of the mean levels of either member of the pair. To measure mean levels of similarity, we compute absolute differences between people’s self-reports of personality traits (cf. Luo & Klohnen, 2005).
Due to the nature of close relationships, many prior studies of personality similarity did not randomly assign participants to peer groups; instead, they assessed the similarity of already-established couples or friends and investigated whether their similarity predicted relationship satisfaction. In these naturalistic studies, it was always a possibility that people liked similar others merely because they were more likely to meet and spend time with them and not because of an active preference (Carli, Ganley, & Pierce-Otay, 1991). Other studies have attempted to address this issue by statistically controlling for background variables like age and education, but this method does not completely rule out the potential confounds (Feng & Baker, 1994; Watson et al., 2004). When random assignment was implemented, it was usually at the expense of external validity; participants often rated the qualities and likeability of hypothetical or fictional characters (e.g., Ajzen, 1974; Byrne, Clore, & Worchel, 1966; Byrne, Griffitt, & Stefaniak, 1967; Horton, 2003; Stapel & Van der Zee, 2006). Thus, most studies were limited in that they had either random assignment with a contrived paradigm—using information about personality that people might not rely on in natural interactions—or a naturalistic design without random assignment.
The current study is one of only a few to use data from well-acquainted, randomly assigned individuals in a naturalistic setting to assess personality similarity and liking (see also Bernieri et al., 1994; Carli, Ganley, & Pierce-Otay, 1991; Kurtz & Sherker, 2003) and the only one of these to use participants who were not college roommates. We use data from a large sample of individuals whom the military had randomly assigned to groups (mean group size = 38.4) to complete 6 weeks of basic military training. Within each group, members had relatively equal opportunity to meet and spend time with each other, so associations between personality similarity and liking would not be due to mere propinquity, but would suggest an active preference for similar others.
In addition to controlling for the FPL and comparing pattern and mean level similarity in a naturalistic experiment, this study assesses the relationship between personality similarity and liking using round-robin information about how participants perceive themselves and each other (Kenny, 1994). Using hierarchical linear regression models, we account for non-independence in responses arising from raters (some participants like people more than others) or targets (some participants are better liked across peers).
Researchers have compared congruence of individuals’ desirable and undesirable traits as separate predictors of liking, and results have been mixed (e.g., Robins, Caspi, & Moffitt, 2000; Rosenblatt & Greenberg, 1988). For example, in one study, undergraduate participants who scored particularly high or low on a depression inventory read about the personality of other classmates who had scored similarly or dissimilarly to them on the inventory. Results suggested that the positive effect of similarity on liking was significant for only nondepressed (i.e., low scoring) participants; depressed participants did not show a preference for other depressed participants (Rosenblatt & Greenberg, 1988). In later research, the actual interpersonal interactions between combinations of individuals with and without depression were explored. Researchers found that during a tape-recorded conversation about 3 suggested topics, individuals with depression did report less discomfort after interacting with other individuals with depression, but still there were no differences in ratings of general liking (Rosenblatt & Greenberg, 1991). In contrast, in a study of newlyweds, the similarity of negative emotionality (comprised of stress, alienation, and aggression scales) significantly predicted relationship satisfaction for men and women, whereas similarity of positive emotionality (comprised of well-being, social potency, social closeness, and achievement scales) did not (Robins, Caspi, & Moffitt, 2000). We believe that exploring the similarity of desirable and undesirable personality characteristics separately after controlling for the FPL could shed light on this discrepancy. In the studies involving individuals with depression, the absence of an effect of similarity of depression on liking may have occurred because the effect was masked by the tendency for raters, regardless of their own level of depression, to perceive nondepressed others as easier to like.
The current study extends previous literature on personality similarity and relationships in several ways. 1. We use random assignment and a naturalistic design with groups of acquaintances 2. We compare similarity of personality patterns and similarity of levels of traits to predict peer liking; 3. We compare the effects of positive and negative personality dimensions; 4. We control for the fundamental principle of liking; and 5. We use random effect models to account for correlations among observations arising from the same target or the same rater.
The major goals of the study are to determine whether similarity in patterns or mean levels predicts peer liking after controlling for peer attributions of personality traits, and whether personality similarity findings apply equally to positive and negative traits. Based on the findings from the romantic couples literature, we predict that patterns rather than mean levels will be positively related to liking, and that similarity of negative traits will matter more than similarity of positive traits.
Participants were 844 Air Force recruits (292 women; 552 men; median age 19) who were several days from completion of 6 weeks of basic military training at Lackland Air Force Base. The sample was part of a larger study of self- and peer-report of personality and personality pathology (see Oltmanns & Turkheimer, 2006). Only the 844 participants whose data were collected during the last year of the study, about 42% of the total, were used in this study, because ratings of liking were only initiated during the final year.
The participants were enlisted personnel (not pilots), who were going to be trained for a wide variety of jobs (e.g., positions in security, cooking, or electronics). On the first day of basic training, the Air Force assigned participants to groups, known as “flights.” For 6 weeks, members of a flight worked, lived, and ate meals together, and they had the opportunity to observe each other’s behavior during many challenging situations. The median number of participants in each group, or “flight,” was 36.5 (range = 27–54). There were twenty-two flights included in this study; 16 of the flights were mixed-sex and 6 were all male. Participants were predominantly White (64.5%), followed by Black (16.1%), Other (11.4%), Bi-racial (4.2%), Asian (3.2%), and Native American (.7%).
Oltmanns and Turkheimer (2006) developed the Multi-source Assessment of Personality Pathology (MAPP) which is composed of 103 items, including 79 items based on the features of 10 personality disorders listed in DSM-IV as well as 24 supplementary items based on additional personality traits (mostly positive characteristics, such as “trustworthy and reliable,” “agreeable and cooperative,” and “articulate and persuasive”). All of the 79 DSM-IV PD features on the MAPP were rewritten into words that avoided the use of technical psychopathological terms and psychiatric jargon. The MAPP instrument was written in the third person for peer-report and translated into the second person for self-report.
Participants rated each member of the group for how much they liked them on a scale of (0) do not like at all to (3) like extremely well.
Participants signed informed consent statements and participated on a voluntary basis. All measures were presented on a computer monitor, one participant to a computer. Members of each flight were tested simultaneously, in a single 2-hour session. Participants rated how much they liked members of their group before completing the MAPP. Then participants read the following instructions: “We are interested in your perceptions of other people in your group. You will be presented with descriptions of various personal characteristics. For each characteristic, you will be asked to click the mouse button when the cursor is pointing to the names of the people in your group who best fit that description. You may click as many names as you want, but you must select at least one person for each characteristic …”
The MAPP items were presented in a quasi-random order and were listed one at a time on the top of the computer screen. To obtain peer-reports of personality, the names of all members of the flight group (excluding the name of the participant completing the MAPP) appeared below the item. A scale containing the numbers (0) never like this, (1) sometimes like this, (2) usually like this, and (3) always like this was listed to the right of each person’s name, with the default selection being (0). Participants nominated from one to as many people in their group as they saw fit for each particular item and used the scale from 0 to 3 to indicate the extent to which those people exhibited the characteristic in question.
All of the regression analyses in the study were conducted with mixed model regression, using PROC MIXED in SAS. Mixed model regression controls for clustering of observations within samples. In the round-robin design of the current study, in which each target was rated by multiple raters and each rater rated multiple targets, observations arising from the same rater or pertaining to the same target were correlated with each other. For a rating of liking yijk, in which target i is rated by rater j, with both i and j in flight k, we estimated the equation:
in which λijFPLij is the fixed effect of the FPL (i.e., the tendency for raters to like targets more when they have rated them positively), and λijSij is the fixed effect of the similarity between target i and rater j. The crossed random variances τi and τj are the effects of ratings by the same target and rater, respectively, nested within the effect of flights, τk. σ;ij is the residual error variance of liking ratings after the other effects have been accounted for (Kenny, 1994).
The 79 items in the MAPP corresponding to DSM criteria for PD were initially classified as negative traits, and the 24 additional items were initially classified as positive traits. We then examined the relation between each item and liking using a mixed model in which liking was predicted from an individual item on the MAPP as a fixed effect, with random variability attributable to flights and to targets and raters within flights. Twenty-one non-DSM items were positively associated with liking and were considered positive traits in subsequent analyses. Forty-six of the DSM items were significantly negatively associated with liking and were considered negative traits in subsequent analyses.
We computed coefficients of similarity for each pair of rater and target using the MAPP. The pattern similarity coefficient was computed as a correlation between the self-reports of the pair, computed across the items. The mean level similarity coefficients were computed as absolute differences between the self-reports of the pair summed across items. We then estimated a series of random effects models, with liking as the dependent variable, predicted by peer attributions of positive and negative traits (representing the FPL) and one of the similarity coefficients (either pattern similarity or absolute mean level similarity) using the equation given above. Random variances were estimated for the effects of raters, targets and flights.
There were 32,700 pair-wise ratings of liking using the scale from (0) do not like at all to (3) like extremely well (M = 1.48, SD = 1.01). 18.39% of the ratings were (0), 35.45% were (1), 26.33% were (2), and 19.83% were (3).
Models were developed to predict liking based on the following factors: Model 1: liking predicted from peer reports of positive and negative traits and the pattern correlation of self-reports of positive and negative traits. Model 2: liking predicted from peer reports of positive and negative traits and the pattern correlation of only negative traits. Model 3: liking predicted from peer reports of positive and negative traits and the pattern correlation of only positive traits. Model 4: liking predicted from peer reports of positive and negative traits and the mean difference of self-reports of only negative traits. Model 5: liking predicted from peer reports of positive and negative traits and the mean difference of self-reports of only positive traits. Results of the mixed effects models are provided in Table 1.
The overall pattern similarity coefficients ranged from −.45 to .93, with a mean of .46, a median of .49, and a standard deviation of .21. The mean similarity coefficient is greater than zero because some items are more frequently endorsed than others across the entire sample. Controlling for the FPL, pattern similarity between rater and target significantly predicted increased liking (b = .14, se = .03, p < .001).
The pattern similarity coefficients for just the negative traits ranged from −.54 to .98, with a mean of .05, a median of .03, and a standard deviation of .17. Controlling for the FPL, pattern similarity between rater and target for negative traits predicted increased liking with marginal significance (b = .05, se = .02, p = .06).
The pattern similarity coefficients for just the positive traits ranged from −1.22 to 1.22, with a mean of .08, a median of .00, and a standard deviation of .40. Controlling for the FPL, pattern similarity between rater and target for positive traits did not significantly predict increased liking (b = .001, se = .02, ns).
The mean level matching coefficients for the negative traits ranged from 0.00 to 2.98, with a mean of .32, a median of .20, and a standard deviation of .37. Unlike the marginally significant effects of similar patterns of negative traits, after controlling for the FPL, there were no effects of similarity of self-reported negative traits using absolute difference scores (b = −.02, se = .02, ns). The negative level similarity coefficient was a significant predictor of liking only if the FPL was not included in the model (b = −.11, se = .02, p < .001).
The mean level similarity coefficients for the positive traits ranged from 0.00 to 2.43, with a mean of .51, a median of .43, and a standard deviation of .38. After controlling for the FPL, there were no effects of similarity of self-reported positive traits using absolute difference scores (b = −0.0, se = .01, ns). Once again, the positive level similarity coefficient was a significant predictor of liking only if the FPL was not included in the model (b = −.04, se = .02, p < .01).
The current study investigated whether similarity between self ratings of positive and negative personality characteristics in a group of peers predicted liking. The results suggest that, even after controlling for the tendency for people to like others who have desirable traits and dislike others who have undesirable traits (i.e. the FPL), greater similarity between the self-reports of pairs of peers was associated with mutual increased liking. Echoing research with romantic couples, this effect was due primarily to similarity of patterns of negative traits rather than positive traits (cf. Robins, Caspi, & Moffitt, 2000), and the effect was stronger in assessments of patterns of personality rather than in absolute mean differences of isolated traits (cf. Luo & Klohnen, 2005). Unlike in other research, we can claim that the effects of personality similarity on liking were not due to propinquity of individuals, artifacts of individual rating strategies, or the attributions peers make about the desirability of each others’ personality.
The finding that the absolute mean difference in levels of self-reported positive and negative traits was negatively associated with liking only when the FPL was not controlled provides an interesting example of why controlling for the FPL is important in naturalistic studies of personality similarity and liking. Because most people rated themselves as having positive traits, the participants who rated themselves positively had the smallest average differences in level (because they were similar to the majority of people who rated themselves positively), thereby inducing a positive correlation between self-reported positive traits and similarity in level. Self-reported positive traits were correlated with peer-reported positive traits, which in turn predicted liking via the FPL. Therefore, similarity in level and liking were related as long as the FPL was not controlled statistically, but the relation disappeared once peer-reported positive traits were in the model. The crucial test for any hypothesis about predictors of liking is whether it predicts liking over and above the established tendency to have greater liking for others to whom we attribute positive characteristics.
The findings of the present research lead to the question of why people with similar personality patterns tend to like each other more than people with dissimilar personality patterns. Implicit egotism (Pelham, Mirenberg, & Jones, 2002) offers one explanation. People generally see themselves in a positive manner, and as a result, anything reminiscent of the self could prompt automatic positive associations. Researchers have found support for this hypothesis in person perception and in important life decisions. For example, as implicit egotism predicted, participants reported being more attracted to others whose experimenter-given code name contained the numbers of their own birthday than those whose code name did not contain those numbers. Also, according to archival data, people were disproportionately likely to marry someone else who had a similar sounding name (Jones, Pelham, Carvallo, & Mirenberg, 2004). Perhaps people are drawn to others who exhibit the same pattern of personality traits because they automatically like whatever reminds them of themselves.
Another hypothesis that could explain why people with similar personality patterns are especially inclined to like each other extends from the “mere exposure” effect. The mere exposure effect is the finding that all else being equal, familiar stimuli (e.g., Chinese characters presented often to non-Chinese speakers) are liked more than novel stimuli (e.g., Chinese characters presented infrequently) (Zajonc, 1968). Familiar stimuli are easier to perceive and interpret than novel because information about the familiar has already been processed at least once before. In general, processing stimuli with ease (called perceptual fluency or processing fluency) is experienced as pleasurable; and when people encounter a familiar stimulus, they might attribute the pleasure they feel from perceptual fluency to the stimulus itself (Reber, Schwarz, & Winkielman, 2004). The mere exposure effect has been extended to person perception previously. Research has shown that people judge familiar people (encountered often) to be more attractive than less familiar people (Moreland & Beach, 1992), and familiar faces to be happier and less angry than unfamiliar faces (Claypool, Hugenberg, Housley, & Mackie, 2007). Thus, perhaps for the same reason that people like designs and faces more the second time they encounter them, people who have familiar personality traits might be easier to understand and therefore easier to like.
Although across both positive and negative traits we found that people tend to like others with similar personality patterns, this effect can be attributed mostly to similarity of negative rather than positive traits. Why would similarity of patterns of positive traits not affect liking? One explanation is that positive traits might be less unusual and thus more familiar than other types of traits. If positive traits are often encountered and rarely hidden, then they would be easy for everyone to perceive and understand regardless of whether people possessed the traits themselves. There would be no added benefit to being familiar with someone’s positive traits because everyone would be familiar with them.
A second, slightly different explanation is that negative traits are viewed in degrees of negativity, but positive traits are viewed as equally positive by everyone. Although implicit egotism and other research about self-enhancement suggest that in general people view all of their own traits as more positive than other people view those traits, some traits are viewed extremely positively by everyone—making it possible that there is a ceiling effect of how desirable those positive traits appear. There might be no equivalent floor effect for negative traits. As an example, most people believe that being patient is highly desirable, regardless of whether they have that trait themselves, but not everyone views narcissistic behavior as equally negative. People who are themselves narcissistic might believe that narcissism is less problematic than people who are not narcissistic. This line of reasoning could explain why we found that similar patterns of positive traits alone was not sufficient to increase liking after controlling for the desirability of positive and negative traits.
A third explanation is that people with similar patterns of positive traits do understand each other better and do see their own positive traits in others in an especially positive light, but feel competitive with each other at times, thereby negating any added benefit to similarity. When two people in the same group of acquaintances have overlapping positive qualities and abilities, the indispensability of each person to the group is mitigated. As a result, people who share positive attributes might feel a little bit threatened by each other. Research on competitiveness supports this idea by showing that people prefer dissimilar rather than similar others when they are in competitive mind frames (Glaman, Jones, & Rozelle, 2002) and actively attempt to differentiate themselves from others when the uniqueness of their identity is threatened (Brewer, 1991; Lemaine, 1974).
Currently, both implicit egotism and the mere exposure effect are plausible explanations for why similarity of personality predicts increased peer liking. We believe future research should pit these explanations against each another to better determine the mechanism of people’s preference for others with similar personalities. Prior research has shown that the theories can be differentiated in terms of their differential predictions about unusual characteristics of the self. According to implicit egotism theory, people with uncommon names will be especially prone to prefer others with similar names because the distinctiveness of the name imparts greater relevance to self-identity, whereas mere exposure theory predicts that people with uncommon names would be less prone to prefer others with similar names, because they are exposed more frequently to common names (Jones, Pelham, Mirenberg, & Hetts, 2002).
Quantifying exactly what is meant by an “unusual personality configuration,” however, will not be a simple task. The dynamics underlying human preferences for other humans in uncontrolled environments are certain to be more complex than those determining preferences for somewhat artificial, albeit better controlled, stimuli in the laboratory. Nevertheless, the ability to be liked by others and to choose felicitous others to like is crucial to well-being. We expect that the importance of human liking will merit the methodological difficulties inherent in studying it.
This research was supported in part by NIMH Grant 1R01MH51187-06 to the second and third authors. We thank Simine Vazire, Paige Harden, and Erik Petterson for thoughtful comments on an earlier draft of the manuscript.
Elizabeth R. Tenney, Department of Psychology, University of Virginia.
Eric Turkheimer, Department of Psychology, University of Virginia.
Thomas F. Oltmanns, Department of Psychiatry, Washington University in St. Louis.