The first challenge in creating a model for human mate choice was to decide which factors would be used to determine who was matched with whom. My research on this topic led me to a comprehensive study of mating determinants performed by Buss and Barnes in 1986. Their study surveyed two groups—married couples and unmarried college students—and asked which factors were most important to them in choosing a mate. The results showed a surprising amount of agreement between the two groups, with kindness, intelligence, and exciting personality as the factors with greatest influence, closely followed by professional status, easygoing-adaptable, political stance, and liking of children. In total, nine factors were found to have a significant effect on mate selection. At first, it seemed logical to weight these factors according to their importance and use these weightings to assign mates. After considering this decision more closely, I decided that a smaller number of more quantifiable factors were necessary to provide a meaningful model. Otherwise, it would be unlikely to achieve any insight on the amount of predictability contributed by any one factor over another. Also, it would be difficult, if not impossible, to assign numerical values to characteristics like “religiosity” and “ideology.” Therefore, I chose to continue researching to find a more parsimonious system to use in the simulation.
Ultimately, my literature review uncovered a 1997 article by Botwin, Buss, and Shackelford that investigated the importance of the Big Five personality traits in mate selection. This study required dating and married couples to fill out personality questionnaires for themselves and for the personality dimensions desired in an ideal mate. Results for the partner’s scores were then correlated with the ideal scores. The only significant coefficient for assortative mating was conscientiousness in married couples. However, all five of the dimensions showed weak-moderate correlations (0.2 to 0.4) for assortative mating, and a larger sample may have produced more significant relationships. Due to its popularity and ease of measure, I decided to use the Big Five personality traits to predict mate choice in my model.
For the basic design of the model, I chose to follow the structure of mating with adolescence and courtship proposed by Todd and Simao (2002). These authors outlined many of the decision rules and constants employed in the simulation, though some details were omitted. As a result, my design is very similar to this original model. The greatest alteration I made was to assign values for each of the Big Five dimensions to every person in the simulation. The differences between each person’s personality attributes and the ideal attributes for his or her sex were squared to compute the mating quality of that person.
If the squared deviation is high, the person is assigned a lower value for quality. A zero value for squared deviation gives the highest possible quality. The mean quality value for males was 5.28 with and standard deviation of 2.45; the mean quality for females was 3.94 with a standard deviation of 2.01. Note that the quality distributions for each sex are different. Since each person only interacts with the opposite sex and adjusts aspiration level accordingly, these differences are not problematic. Since I could not find any strong research that supported individual-specific differences (for example, an extraverted person preferring a more introverted person), there was no ideal match for one person that was less than ideal for another. Also, unlike some previous studies, all of the traits that were randomly assigned to individuals in the simulation were drawn from a normal distribution. The normality of these traits in the population has been confirmed in several cross-cultural studies (McCrae, 1998).
I wrote my model in the object-oriented programming language Python using a Person class that contained data on the individual’s personality, quality, and relationship status. Since females are known to be the more discriminating sex in humans, they were selected to be the ones in charge of choosing to enter courtship with specific males. At the beginning of the simulation, 100 males and 100 females with random personalities are generated. Their qualities are computed based on these values and their aspiration levels are set to 0. Next, both males and females go through a three-year period of adolescence during which they modify their aspiration levels. This step is important because no member of either sex will accept a potential mate who is below this aspiration level. After adolescence, the mating period begins. During this time, females gradually meet males (the rate is controlled by a constant) and add them to their alternate lists. Eventually, potential courtships will arise from members of these lists. The attached diagram shows the status of a female and her alternate list at one point in the simulation. In this figure, only male #76 will be a compatible mate for the female, because his quality is above her aspiration level and his aspiration level is below her quality. The table below this diagram shows the constants used in the simulation runs that yielded the results described in the next section. Once the time courting a single individual reaches 30 (years), then both that male and female are removed from the pool of availables and they become mated. During courtship, it is possible for the female to choose another male to court if his quality is significantly greater than the current partner to offset the female’s time investment in the current relationship. If a rejection occurs during the courtship period (either for the male or the female), aspiration levels are adjusted again.
For the most part, the results of my simulation mirrored those found by Todd and Simao (2002). This outcome is understandable because of the similarity in the two model designs. Note that the statistics reported in this section represent averages across one hundred runs of the simulation program. As shown in Table 2, the correlation between overall male and female quality was 0.445, which replicates findings of trait correlations in the general population (and in previous models). Furthermore, the number of dates to match around 90% of the population in my simulation ranged from 5.0 to 6.0, which was higher than 1.4 to 3.2 (Todd & Simao, 2003) but much lower than the 40 dates required in Kalick and Hamilton (1986). Five or six lifetime partners seems to be a reasonable estimate, though actual data does not seem to exist for this statistic. Like earlier models, most individuals were matched within the first few years after adolescence, followed by more gradual matching as the simulation approached 100 percent matching. Figure 2 demonstrates this trend for a simulation run that matched the entire population. Although twenty years were allotted for reproductive lifetime, only seventeen were required to match everyone. There was a preference for higher quality individuals to be matched earlier than lower quality individuals.
All traits except for intellect/openness had statistically significant (p < .05) correlations between matched males and females. However, all of these correlations were only weakly positive. Quantitatively, it appears that neuroticism was the greatest predictor of matching, followed by conscientiousness, extraversion, and agreeableness to nearly equivalent degrees. Intellect/openness had a correlation of almost exactly 0, illustrating a limited effect on controlling who was matched to whom. These results differed from those discovered in personality surveys of people and the actual mates they obtained in Botwin, Buss, and Shackelford (1997). In that study, intellect/openness had the highest correlation between partners (0.39, significant at p < .001). The only other significant correlation was for extraversion (0.32, significant at p < .01). Correlations for agreeableness, conscientiousness, and neuroticism were respectively 0.25, 0.24, and 0.12. As an aside, there was a strong correlation (0.75 to 0.80) between quality and aspiration levels for both males and females; therefore, higher quality individuals generally had greater quality expectations. Also, changing any of the constants (Table 1) did little to alter the outcomes of the simulation unless the change was completely unrealistic.
In analyzing my data, it was clear that the correlations I found were lower than those observed in samples of human participants. There are several possible explanations for this result. First, there may be type preferences for individuals with certain characteristics. In other words, the “ideal” person of the opposite sex will probably have unique characteristics specific to the desires of the individual in question. Because of limited data, I was forced to assume that every male has the same ideal female with certain personality characteristics. In reality, this portrait is incorrect; everyone views personality characteristics differently. If I had designed my model so that people assigned higher qualities to potential mates based on how similar they were to themselves, I would have obtained higher correlations across the Big Five. Another explanation relates to the potential interaction between personality traits and courtship. It is possible that certain traits are more like to exist in paired males and females because these traits support successful courtship. I assumed that each of the Big Five contributes equally to quality determination, though this may not be the case. Finally, it is possible that there were errors in the simulation programming, and that higher correlations would have been found if it had been created perfectly.
Even considering these limitations, the results of this study do have potential implications for studying human mate choice. Neuroticism had the highest differential between the level participants desired and the average level within the population. Thus, it makes sense that neuroticism is a crucial determinant for predicting mate choice: if a potential mate is close to the ideal level of neuroticism, he will be assigned a high quality in the pool of male and likely achieve a high-quality female. Also, intellect/openness had the lowest correlation in the results from this model, which indicates that it may not be a critical factor in mate selection despite it having the highest correlation (0.39) among married couples.
Though numerous articles have been written about the many factors that influence human mate choice, attempts to find the most salient factors have been limited. Computational studies on mate choice have been even more rare. More research needs to be done to define this decision that affects virtually every member of the human race.