Abstract
To explain the pervasive role of humor in human social interaction and among mating partner preferences, Miller (2000a) proposed that intentional humor evolved as an indicator of intelligence. To test this, we looked at the relationships among rater-judged humor, general intelligence, and the Big Five personality traits in a sample of 185 college-age students (115 women, 70 men). General intelligence positively predicted rater-judged humor, independent of the Big Five personality traits. Extraversion also predicted rater-judged humor, although to a lesser extent than general intelligence. General intelligence did not interact with the sex of the participant in predicting rating scores on the humor production tasks. The current study lends support to the prediction that effective humor production acts as an honest indicator of intelligence in humans. In addition, extraversion, and to a lesser extent, openness, may reflect motivational traits that encourage humor production.
Introduction
The phenomenon of humor is a unique, yet universal aspect of human psychology (Caron, 2002; Gervais and Wilson, 2005; Sroufe and Waters, 1976). The ability to produce, comprehend, and appreciate various forms of humor is a distinctive hallmark of human cognition (Provine, 2000, 2004), and a number of theories have been proposed to elucidate humor's evolutionary origins (Alexander, 1986; Caron, 2002; Gervais and Wilson, 2005; Miller, 2000a; Weisfeld, 1993). In particular, Miller's account that humor evolved as a result of sexual selection via mate choice has received a fair amount of attention in recent years (Bressler and Balshine, 2006; Bressler, Martin, and Balshine, 2006; Greengross and Miller, 2008; Kaufman, Kozbelt, Bromley, and Miller, 2007). His model posits that humor appreciation is implicitly responding to various cognitive abilities required in producing effective humor. More importantly, Miller sees effective humor production as a strategy to communicate underlying genetic quality with respect to psychological traits. Of these traits, general intelligence emerges as one of the most comprehensive and heritable psychological constructs (Jensen, 1998). General intelligence has also been found to show strong correlations to other indicators of genetic quality, such as long-term health (Deary, 2005; Gottfredson and Deary, 2004; Kanazawa, 2006), physiological symmetry (Bates, 2007; Luxen and Buunk, 2006; Prokosch, Yeo, and Miller, 2004), and developmental stability (Thoma et al., 2005). Thus, Miller's model posits that some of the individual differences in humor production will reflect differences in general intelligence and its underlying genetic fitness. As a result, a preference for humor in potential mates, and a motivation to produce displays of effective humor in courtship and social settings, serves an adaptive purpose because effective humor is an honest and reliable indicator of mental ability and biological fitness (see Miller, 2000a, 2000b, 2000c).
The importance of humor in mate preference is supported in a number of self-report (Buss and Barnes, 1986; Feingold, 1992; Goodwin, 1990; Grammer and Eibl-Eibesfelt, 1990; Hansen, 1977; Regan and Joshi, 2003; Sprecher and Regan, 2002) and experimental (Bressler and Balshine, 2006; Bressler et al., 2006) studies of mating preferences. In addition, mate preferences for intelligence are also well established (Buss, 1989, 1994; Regan and Joshi, 2003). If humor is an honest signal of intelligence and Miller's model of humor is correct, then intelligence and humor should be positively correlated. However, only two studies have explicitly looked at the relationship between humor and intelligence. Feingold and Mazzella (1991) found a positive correlation between measured verbal intelligence and rater-judged humor production in a college age sample (r = .40). Masten (1986) reported substantial positive correlations among IQ, academic achievement, and rater-judged humor production in 10 to 14-year-old children (r = .50 – .53).
The current study investigates the relationship between general intelligence and rater-judged humor production, with a few key differences from the aforementioned studies. The current method provides a clearer estimation of general intelligence relative to the Feingold and Mazzella (1991) study, and unlike the pre-adolescent sample in the Masten (1986) study, the college-age sample is consistent with Miller's model that costly displays of humor would manifest primarily during peak reproductive years (Miller, 2000c). We assessed humor with open-ended verbal and drawing tasks and measure general intelligence using an 18-item version of the Raven's Advanced Progressive Matrices (Raven, Raven, and Court, 1998). In addition, we included a Big Five personality assessment to look at how personality traits are related to humor production. Previous research has found that openness has shown a significant positive relationship to general intelligence (Ashton, Lee, Vernon, and Jang, 2000; Harris, 2004; Miller and Tal, 2007; Shafer, 1999) and finding a positive relationship to humor production would lend support to Miller's model. Furthermore, these personality traits will serve as useful control variables to see if the role of intelligence is a unique predictor of humor production. These findings represent a key step in assessing a key prediction of Miller's model, and may help elucidate the evolutionary basis of humor.
Materials and Methods
Participants and procedures
One hundred and eighty-five undergraduate students (115 women, 70 men; Mean age = 22 years, SD = 4.11 years, Range = 18 – 45; 45% Caucasian, 55% Hispanic) from California State University, Long Beach (CSULB), and Saddleback Community College (SCC) voluntarily completed a paper and pencil questionnaire. All students participated in partial fulfillment of course credit. Participants completed the questionnaires under conditions of complete confidentiality and anonymity. The CSULB Institutional Review Board approved all materials and procedures in the study.
Individual difference measures
All participants provided general demographic information such as age, sex, ethnicity, sexual orientation, height, weight, and basic academic history. General intelligence was measured using an 18-item version of the Raven's Advanced Progressive Matrices (Miller and Tal, 2007; Raven et al., 1998). The Raven's Progressive Matrices test has been shown to have a high factor loading (.60) with general intelligence (Marshalek, Lohman, and Snow, 1983), and published test reliability scores on the 12-item short form of the Raven's Advanced Progressive Matrices range from α = .66 to .73, compared to α = .84 for the original 36-item version (Arthur and Day, 1994; Arthur et al., 1999; Raven, Raven, and Court, 1998). Score distributions of the 18-item Raven's Advanced Progressive Matrices from the samples studied are as follows: CSULB participants (n = 139; Mean score = 10.61, SD = 3.08 years, range = 3 – 18) and SCC participants (n = 46; Mean score = 11.67, SD = 2.86 years, range = 6 – 17). These distributions are very similar to those reported in the normative sample of Arthur, Tubre, Paul, and Sanchez-Ku (1999).
Personality was measured using a 50-item version taken from the International Personality Item Pool (IPIP; Goldberg et al., 2006). Drawing from constructs similar to those in the Revised NEO Personality Inventory (NEO-PI-R; Costa and McCrae, 1982), the IPIP scale assessed the Big-Five personality traits (with corresponding internal consistencies in the current sample): openness (α = .69), conscientiousness (α = .85), extraversion (α = .89), agreeableness (α = .75), and neuroticism (α = .85). All IPIP items used can be found at http://ipip.ori.org/newNEODomainsKey.htm.
Humor production tasks
To measure humor, we used an open-ended response format designed to give the participant the opportunity to display humorous content in a loosely-restricted template. Before engaging in the tasks, participants were instructed to produce humorous content with the knowledge that it would be read and judged by anonymous raters. Participants were asked to complete three different tasks, each designed to measure different aspects of humor production. The latter two tasks were adopted and revised from creativity measures used in Miller and Tal (2007).
The first task consisted of six character profile forms to be filled in by the participant (Figure 1). This task was designed to measure verbal humor in the context of witty stereotypes and references. Each profile contained an individual portrait photograph (three male and three female portraits in all) along with general headings (e.g., name, occupation, typical day) to help create a mock description of the individual in the photograph. All six individuals portrayed in the character profile task were friends of the author, and gave written consent to use their photograph. None of the individuals portrayed had any knowledge of or relationship to any potential participant currently attending CSULB or SCC.
Sample character profile humor task
The second task consisted of three questions to be answered by the participant (Figure 2). This task was designed to measure verbal humor in the context of sarcastic and/or disparaging responses. The instructions on this subset present the questions as a hypothetical e-mail from a friend asking for humorous responses geared at finding differences in types of humor.
E-mail response humor task
The third task measured non-verbal humor. Participant's created a total of eight humorous drawings of different types of animals and people (Figure 3).
Drawing humor task
Humor judges
Twenty-eight undergraduate and graduate student volunteers (14 women, 14 men; Mean age = 23.38 years, SD = 3.6 years, range = 18 – 30) from California State University, Long Beach rated the humor production tasks. Students recruited from Introductory Psychology classes did so in partial fulfillment of psychology course credit requirements. Judges completed all ratings under conditions of complete confidentiality and anonymity. In addition, all ratings were done independently, blindly, and without any explicit knowledge of the participant's sex, intelligence, personality, or other information.
Humor ratings
Each participant's six character profiles, three e-mail response answers, and two pages of humorous drawings (four drawings per page) were rated independently by four judges (two women and two men) on a 1–7 humor scale (e.g., 1 = “not funny at all,” 4 = “Moderately Funny,” and 7 = “Hilarious”). To keep the number of ratings done per judge manageable, ratings were done in blocks, such that each judge provided humor scores for blocks of 22 – 35 participants. All judges were randomly assigned to fill the requisite slots in each rating block. In all, seven rating blocks were used to provide humor scores for the total amount of participants. Humor was not formally defined for the judges.
Results
Humor ratings
Looking at the distribution of raw rating scores, the majority of aggregated ratings per participant were quite low (Mean score = 2.31, SD = .625 years, range = 1.27 – 4.25). Distributional properties of the humor ratings showed a slight positive skew and no clear kurtosis. Our first concern was to see if judges were interpreting the 1–7 humor scale in a consistent manner. We first compared the seven rating blocks using a one-way ANOVA, finding a significant difference in raw mean humor scores between rating blocks, F(6, 178) = 19.03 (p = .000). We subsequently looked at raw mean humor scores of the four judges within each rating block, finding significant differences in all seven rating blocks (all ps < .01). In order to correct for differences in scale usage, we standardized all of the ratings given by each judge.
Inter-rater reliabilities were then computed for the four judges within each rating block. We first computed the inter-rater reliability for each of the eleven humor task items individually, finding an average Cronbachs Alpha (α) = .58 and average inter-rater correlation (r̅ i ) = .28 across all seven rating blocks. Although these reliabilities are below the commonly accepted reliability criterion for psychological measurements, they are consistent with past humor production measures (Feingold and Mazzella, 1991; Köhler and Ruch, 1996; Masten, 1986), and possibly reflect the varied nature of humor preferences. Next, we looked at inter-rater reliabilities among mean scores of each humor subtask across all seven rating blocks, with α = .72 and r̅ i = .39 for the six character profiles, α = .63 and r̅ i = .32 for the three e-mail responses, and α = .71 and r̅ i = .40 for the two pages of humorous drawings. Finally, we looked at entire humor task including all 11 items, with α = .74 and r̅ i = .42.
A factor analysis on all 11 items was analyzed using principal components and promax rotation in SPSS. Looking strictly at factor loadings, a verbal and drawing factor seemed to emerge. Upon further analysis, indications from the scree plot (Figure 4) suggested only one predominant factor, where the first factor accounted for 49% of the variance while the second accounted for 10%.
Scree Plot on humor item factor loadings (Using Principal Components and Promax rotation)
In addition, the component correlation matrix showed an r = .61 between the two factors. Thus, the mean score of all 11 items (which we refer to as “humor production”) was used as the quantitative basis for the humor construct, as it accounted for the most variance in the factor analysis and showed the highest and most consistent inter-rater reliability across the seven rating blocks. In Table 1, we have included a factor loading plot with both factors because we believe that if the drawing tasks were given equal weighting to the other tasks, a two factor solution would have emerged. We have also included item correlations with our main predictor of intelligence.
Humor task item loadings from Principal Components factor analysis pattern matrix
= significant at p < .05
= significant at p < .01
Bivariate correlations among key variables
Table 2 shows the bivariate correlations of humor production and general intelligence to openness, conscientiousness, extraversion, agreeableness, neuroticism, age, height, weight, and number of semesters in college. Of note, humor scores were significantly and positively correlated with general intelligence, openness, extraversion, height, and weight. None of the variables in Table 2 showed any significant non-linear relationships.
Bivariate correlations with humor production and general intelligence
= significant at p < .05
= significant at p < .01
We were interested in the significant correlations of weight and height to humor production, and decided to calculate a body mass index (BMI) variable. Despite the significant correlations in height and weight, BMI did not correlate with humor production, r(183) = .10, p = .21.
Humor production multiple regression
To analyze key predictors of humor production, we entered ten variables into multiple regression: general intelligence, openness, conscientiousness, extraversion, agreeableness, neuroticism, age, height, weight, and number of semesters in college. Standardized beta weights and significance levels are shown in Table 3, with both general intelligence and extraversion being significant and unique predictors of humor production.
Predictors of humor production in multiple regression reported with standardized beta weights (and significance levels)
Sex differences in humor production
Independent samples t-test indicated that standardized male humor scores (M = .19) were rated significantly higher than standardized female humor scores (M = −.11), t(183) = 2.75, p = .007 (two-tailed), d = .40. This result is consistent with previous studies on sex differences in humor (Thorson and Powell, 1993b; Ziv, 1984) as well as Miller's sexual selection model that reproductive benefits for humorous displays would be greater for males than females (Miller, 2000c). Table 4 displays the bivariate correlations showing sex differences with regard to humor production.
Sex differences in bivariate correlations (All male correlations are shown in bold)
= significant at p < .05
= significant at p < .01
We ran a moderated regression to understand if the relationship of general intelligence to humor production depended on sex. Counter to Miller's prediction that male displays of humor, being more effective, will have a stronger relationship with general intelligence (Prediction 5 in Miller, 2000c), female intelligence (β = .33, t(113) = 3.70, p = .000) was a stronger predictor of humor than male intelligence (β = .15, t(68) = 1.23, p = .22), although this interaction was not statistically significant (β = −.11, t(181) = −.70, p = .49). We also ran a moderated regression on openness, as it also showed sex differences in the bivariate correlations. Male openness (β = .38, t(68) = 3.33, p = .001) was a stronger predictor of humor than female openness (β = .11, t(113) = 1.13, p = .26), and this interaction was statistically significant (β = .18, t(181) = 2.36, p = .02). This result is more consistent with Miller's prediction, although openness can only be considered an indirect assessment of intelligence. We should also note that there were no significant interactions with respect to the sex of the judges on any of the relationships to humor production.
Discussion
Consistent with the previous research on humor and intelligence (Feingold and Mazzella, 1991; Masten, 1986), we found that humorous pictures and stories created by individuals higher on general intelligence are rated as being funnier. This result holds after controlling for Big Five personality factors. Although females showed a stronger relationship of humor production to general intelligence than males, which ran counter to our expectations, this difference was not significant.
The relationship of humor production to extraversion is also consistent with previous research (Köhler and Ruch, 1992; Thorson and Powell, 1993b). The current study showed that extraversion corresponds to facets of sociability (e.g., “have a lot to say”, “feel comfortable around people”), dominance (e.g., “like to draw attention to myself”), and active sensation seeking (e.g., “am the life of the party”). With respect to reproductive fitness, recent evidence found that extraversion has a consistent cross-cultural relationship with short-term mating success (Schmitt and Shackelford, 2008). Although openness showed a significant bivariate correlation with humor production, particularly in males, this effect was diminished when observed in the multiple regression analysis and did not show any significant relationship with intelligence. Despite this, the similar bivariate correlations of humor production to both extraversion and openness may be indicative of tendencies for exploratory behaviors, cognitive flexibility, and desire for novelty found in the higher-order “plasticity” dimension (see DeYoung, Peterson, and Higgins, 2002). This suggests that humor production has a social component distinct from its cognitive relationship, and we speculate that this social component is tied to the motivation to produce humor displays in social contexts.
Our preliminary findings support Miller's contention that humor evolved as an honest display of general intelligence in humans. In turn, individual differences in humor production taps into the large portion of heritable mental abilities that make up general intelligence, with humor appreciation (as rated by the humor judges in this study) reflecting this association (Miller, 2000a, 2000b, 2000c). However, the low reliability of agreement in rating humor production has been an issue here and in other studies (see Feingold and Mazzella, 1991; Köhler and Ruch, 1996; Masten, 1986). This presents a challenge to Miller's model that humor appreciation is consistent enough between individuals to reliably indicate any other individual differences in humor producers.
Limitations and future directions
The current study only assessed one particular facet of humor production—written and drawn forms of creative stereotypes and descriptive statements. To preserve participant confidentiality and anonymity, as well as methodological feasibility, ecologically valid aspects of humor production, such as social setting, interpersonal relations, shared knowledge, and spontaneity had to be sacrificed. As a result, participants and judges may have been less motivated to produce and appreciate humor, respectively, leading to decreased power to detect actual individual differences in humor production. Thus, the construct of humor investigated here may give only a partial glimpse into the phenomenon of humor and how it relates to other patterns of individual differences. Future studies investigating more ecologically valid measures of humor (e.g., observational studies) and self-report measures of humor production (see Thorson and Powell, 1993a) would broaden the scope of the current study. In addition, our study did not investigate heritable biological fitness in terms of genetic quality or overall health, and further research on the relationship of humor production to other fitness-related constructs will be necessary to understand the full breadth of humor production as a fitness indicator.
Another limitation came from partitioning humor ratings into blocks instead of having each rater judge the entire sample of humor tasks, which may have weakened the consistency and reliability of the rating process. However, the alternative method of having fewer raters judge the entire sample of humor tasks also has a potential cost of not representing the varied nature of humor preferences, making this issue difficult to resolve without extensive resources. Furthermore, due to the pencil and paper format of the humor tasks, there is a possibility that raters may have inferred the sex of the participant by their penmanship, which may have biased their subsequent rating for the humor task.
With respect to Miller's notion that the primary function of humor is to display intelligence during courtship (Miller, 2000a), the humor tasks in the current study were not designed to assess courtship-motivated humor. In our case, the types of humor used may reflect content that extends outside of a courtship context. Distinguishing the types of humor used for courtship from those used in non-courtship social interactions may reveal differences and similarities between social and sexual selection strategies, and how humor may be strategically manipulated to serve different adaptive functions (see Griskevicius et al., 2007).
Summary
The current study serves as an exploratory glimpse into the evolutionary basis of humor. Our findings are consistent with the idea that humor is a signal—an honest advertisement of underlying cognitive ability and fitness—and that it continues to serve this function today. However, much more research will be needed to understand its evolutionary origins and why it is so pervasive in human social interactions.
Footnotes
Acknowledgements
The authors would like to thank Anthony Huntley, Farnaz Kaighobadi, Matthew Keller, Geoffrey Miller, William Pedersen, and one anonymous reviewer for their valuable insights and suggestions on earlier drafts of this paper. Also, the authors would like to thank Nuria Giralt and Steven Teh for their help with data collection.