Taking risks for personal gain: An investigation of self-construal and testosterone responses to competition

ABSTRACT

Recent research on testosterone and risk-taking behavior is beginning to focus on the role of context-dependent changes in testosterone. Extending this research, our study investigated the association between testosterone reactivity to competitive outcomes and risk-taking in the context of a video game based competition. The study also examined whether self-construal moderated this relationship. Results indicated that a rise in testosterone during competition did not predict subsequent risk-taking behavior. However, a rise in testosterone during competition predicted subsequent risk-taking behaviors within winners with independent self-construals. Nevertheless, results did not reveal an association between basal testosterone and risk-taking, nor did competitive outcomes modulate a differential testosterone response. Overall, we treat these findings as preliminary, as there were multiple analyses conducted and effect sizes were relatively small. We discuss these results in the context of recent animal findings that testosterone facilitates success at future competitions after winning a competition, as well as recent research suggesting self-construal moderates associations between testosterone and aggression.

KEYWORDS:

• Testosterone
• competition
• risk-taking
• self-construal

Acknowledgments

We thank Shyneth Galicia, Jordan Liphardt, Elliana Lozoya, Nicholas Jones, and Brian Tyminsky for their assistance with data collection, as well as Benjamin Moreau for help with salivary assays. Funding of this research was supported by faculty start-up funds at Wayne State University and Nipissing University, and a Thomas C. Rumble Graduate Fellowship at Wayne State University. JMC is funded by a Natural Sciences and Engineering Research Council of Canada (NSERC) Discovery Grant and by a Northern Ontario Heritage Fund Corporation (NOHFC) grant.

Contributions

KMW designed the study, collected the data, performed the analyses, and drafted and revised the paper. AR, SG, and SK edited and revised the paper. JMC oversaw the hormone assays and edited and revised the paper.

Disclosure statement

No potential conflict of interest was reported by the authors.

Supplemental Material

Supplemental data can be accessed here.

Notes

¹ Alternatively, one could conduct power analyses with effect size for the T reactivity x self-construal interactions reported by Welker and colleagues (2017). Using a generic effect size of r and a two-tailed alpha of .05, we have 49% power (power = .49) to assess effects of the magnitude of the T-reactivity x Self-construal effect size reported by the integrated data analysis of Welker and colleagues (2017, partial r = .15). We also provide power analyses for our 3-way interaction in our linear multiple regression models for this effect size. For this effect size in a 7-predictor moderated regression model (f² = .023), our sample size achieved 49% statistical power (power = .49).

² Other researchers have calculated this score by subtracting the sum of the interdependence scale from the independent scale (e.g., Kitayama et al., 2014). Although we did not happen to use that approach in this paper, since the subscales have equal items, this approach is a linear transformation of reverse-scoring the interdependent items and computing an average. Choosing the alternate approach does not change any inferential statistics and conclusions in this paper.

³ Team condition significantly moderated a 3-way outcome X basal testosterone X self-construal interaction (i.e., a 4-way interaction). Although our sample is very underpowered to examine 4-way interactions and there are difficulties interpreting 4-way interactions, we nevertheless explored this interaction in our supplemental materials for interested readers.

⁴ Because we recorded the amount of money points participants earned in the BART, our analyses invited assessing whether the interaction effects used in Aims 1 and 2 replicated when predicting money points earned on the BART. Money points earned were strongly related to risk-taking behavior (r = .74, p < .001), although this relationship was curvilinear with some heteroscedasticity (See Supplemental Materials). However, when we ran the models associated with Aims 1 and 2, there was no significant competition outcome X testosterone reactivity interaction predicting money points (B = .01, t(152) = 1.66, p = .100, Aim 1 Model) or a significant three-way outcome X testosterone reactivity X self-construal interaction (B = .02, t(148) = 1.26, p = .211, Aim 2 Model). Although nonsignificant in the Aim 1 Model, the testosterone reactivity X outcome interaction was significant in the Aim 2 model, with a trend hinting at a positive direction between testosterone reactivity and points earned in winners (B = 1.04, t(148) = 1.69, p = .093), but not losers (B = -.96, t(148) = −1.28, p = .202). Altogether, these interaction effects on money points earned were not particularly robust compared to the standard BART measure of risk-taking.

⁵ We previously examined dual effects of basal cortisol and testosterone on risk-taking within this data (Mehta, van Son et al., 2015 Study 2), as well as the effects of facial width-to-height ratio (fWHR) and status (Welker, Goetz et al., 2015). For robustness, we also examined the analyses for Aims 1 and 2 (Presented in Tables 2 and 3) controlling for basal testosterone, basal cortisol, and a basal testosterone X cortisol interaction term, as well as fWHR, status, and a fWHR status interaction term. With these covariates, the testosterone reactivity x outcome interaction presented in Tables 2 and 3 remained significant in both models (ps < .006). Within the analyses in Table 3, the three-way testosterone reactivity X self-construal X outcome interaction became nonsignificant (p = .262), but the general pattern of slopes held. Moreover, the outcome x testosterone reactivity conditional interaction was significant when people were more independent (p = .008) but not more interdependent (p = .266). Although the three-way interaction was attenuated with these covariates, it is important to note that the sample size of these models dropped (N = 144) in these follow-up analyses due to missing data in the covariates, contributing to decreased statistical power.

Additional information

Funding

This work was supported by the Natural Sciences and Engineering Research Council of Canada [1502]; Northern Ontario Heritage Fund Corporation; Wayne State University.