Into the Black Box: Sex and Gender in the Study on Decision-Making – An Evidence from a Slovak Sample

1.1 Research on Differences Between Men and Women

The psychological research on differences between men and women has a long history, with the oldest studies being more than a hundred years old and dating back to Freud’s time at least (Spence & Buckner, 2000). The simple questions these studies asked provided simple answers. Women turned out to be less intelligent, have a different brain structure and, in the male-dominated economic world, were considered farther from the rational ideal than men (see Hyde, 2005; Hyde et al., 2019 for an overview). However, despite this long history, there is still great debate over what we understood by gender differences and how the issue should be approached.

As gender became a buzzword, many scholars claimed to be studying gender differences, when in fact they were mistaking gender for sex, coding it as a dummy variable, and—as Maney (2016) and Westbrook and Saperstein (2015) observed—tacitly implying that the differences are biological in origin and immutable. Connell (2006) concluded that treating gender as a fixed and dichotomous category leads to understanding statistical differences between the sexes as an explanation of thereof. Furthermore, Westbrook and Saperstein (2015, p. 534) noticed that “essentialist practices that treat sex and gender as synonymous (…) will continue to reproduce statistical representations that erase important dimensions of variation and likely limit understanding of the processes that perpetuate social inequality”. Even today, in the literature, there are few studies in which the relation between gender and performance have been investigated seriously for instance in association with entrepreneurial intentions (Adamus, et al., 2021a; Gupta et al., 2009), risk aversion (Adamus, et al., 2021b; Meier-Pesti & Penz, 2008) and political behaviour (Bittner & Goodyear-Grant, 2017). Nevertheless, despite being scarce, all these studies have suggested the observed differences between men and women are better explained by variables other than biological sex, including masculinity and femininity.

1.2 Measurement Issues

This gap in research may be caused by the lack of commonly accepted parsimonious and reliable measures. Although we expect that a significant portion of an individual’s identity is shaped by their gender-related self, measuring it is a problem because gender remains an abstract and ephemeral concept. Smiler and Epstein (Smiler & Epstein, 2010) observed that in the late 1980s, over 1400 gender-related measures were published—nearly six times as many as in the previous decade. However, only a small proportion of these newly published scales have been used or referred to more than five times (Smiler & Epstein, 2010). This shows two important things: there was (and still is) enormous interest in studying gender and even greater disagreement about the appropriate methods and approaches for capturing it.

So what are the characteristics of a good gender-related measure? Constantinople (1973) believed that it should serve as a reliable predictor of an individual’s sex, others have claimed that it should also predict sexual orientation (Kachel et al., 2016). Today, however, such an approach seems redundant. If the gender measure is to be an approximation of biological sex, then we have no need for it—it would be simpler to ask individuals about their self-ascribed biological sex. Moreover, Bem (1974) hinted that identifying traits as masculine or feminine based on the frequency with which they are considered typical of men or women would be biased. If we want an instrument that can be used to understand the gender phenomenon, we should focus on socially desirable traits rather than typical traits (Bem, 1974, p. 155). In our view, a good gender-related measure should meet two requirements: (i) reflect at least part of the gender-related self-concept (construct validity) and (ii) predict some forms of gender-related behaviour or social action (explanatory power) (Athenstaedt, 2003; Unger, 1979).

Although none of the existing measures seems to provide a comprehensive picture of the global gender-related self, quantitative studies do not shy away from asking questions about the scope and sources of gender differences. However, confounding sex and gender in quantitative studies limits our understanding of gendered phenomena and their potential underpinnings. To investigate differences between men and women properly, such research should be equipped with proper tools: both informative and concise. Consequently, we decided to explore whether any of the two personality traits-based questionnaires could become convenient candidates to be used by quantitative researchers. Importantly, Smiler and Epstein (Smiler & Epstein, 2010) observed the BSRI and PAQ are the most frequently used gender measures. Therefore, employing one of the two questionnaires could foster comparability of results delivered by the older literature with those yet to come. Furthermore, aware of the limitations of quantitative research, we decided to use the short versions of the BSRI and PAQ questionnaires because they have good reliability and because in more extensive studies it is pragmatic to use short forms rather than the more time-consuming 60-item questionnaires.

1.3 Sex, Gender and Risk-Taking

Measuring gender should not only involve describing at least part of the gender-related self but also the different forms of behaviour stereotypically associated with each of the sexes. One of such economically important and greatly gendered phenomena is risk-taking. Its importance stems primarily from the associations between risk-taking and favourable economic outcomes. For decades, the psychological and economic literature had indicated that women were more risk-averse and consequently, fared worse in terms of economic performance. Until a seminal paper by Nelson (2015) cast doubt on this. Since her results were published, an increasing number of studies have concluded that there are no significant differences between men and women (e.g., Csermely & Rabas, 2016; Filippin & Crosetto, 2016). In the Slovak context, Adamus (2021b) showed that regardless of the measure, men and women tend to express similar attitudes towards risk in economic contexts. The authors pointed that in the strictly financial domain risk attitudes may be better explained by factors other than biological sex including the BSRI femininity subscale. Therefore, to avoid setting too narrow a definition, in this paper we decided to define risky behaviour in line with the common-sense definition, namely as a situation involving potential negative consequences, losses or injury. Such forms of behaviour may be also well-described as reckless or careless. To do so, we employed Decision Outcome Inventory (Bruine de Bruin et al., 2007) that captures potentially harmful outcomes of every-day decisions. In other words, it measures how an individual behave in situations involving risk of a negative, maladaptive or ineffective outcome.

2.1 Objectives and Research Questions

In this study we address the issues encountered in domain-specific research when attempts are made to incorporate gender-related considerations into the design. The first essential step in gender-sensitive research is to obtain a proper understanding of the analytical distinction between sex and gender. The second is to ensure the appropriate instruments are used to obtain reliable measurements of the phenomena being investigated. Although the study is explorative and so there are no hypotheses, we put forward three research questions and stipulated explanations of possible results. First, we asked whether individuals scoring high on expressive (instrumental) traits have the notion of being feminine (masculine) as well. The consistency between those self-ascriptions would indicate that by measuring expressive and instrumental traits, BSRI and PAQ tap into the subjective notion of being feminine and/or masculine. Then, we sought answer to the question of the measures explanatory power. Are BSRI and PAQ capable of explaining gendered risky behaviour above and beyond biological sex? The measure that would show best construct validity and, at the same time, better explain gendered behaviour could be considered a reliable – among currently available – concise measure of gender-related self with a potential to provide meaningful information about antecedents of gendered behaviour.

2.2 Participants and Procedure

A total of 470 Slovak adults (238 men, 232 women) consented to and participated in Study 1 (M_age = 42.35, SD_age = 13.12). Participants were 470 adults. Participation was voluntary and anonymous. Data collection was through an online survey hosted on Qualtrics and data were collected by an external agency (chosen in a tender and compliant with the international ESOMAR code). Participants were selected according to the following criteria: non-student, aged 18 and over, gender balance, education level—a balanced of secondary school education and university education, sample size 440 (the sample size was calculated using an alpha of 0.05, a confidence interval 95% and effect size 0.14; yielding a sample of 400 + 10%). Data collection was to be terminated after reaching the desired number (440); finally, the external agency collected data from 470 participants.

After reading and signing the informed consent form, the participants completed demographic questions on age and self-stated biological sex, the BSRI, PAQ, TMF and Decision Outcome Inventory, as well as a further six questionnaires (not reported here) that were intedended to form an independent study. The materials included two control questions to eliminate random responses (e.g., participants had to click on a specific option: ‘Now click response 4.’). Participants giving a wrong response to the control questions were not allowed to participate further in the research.

2.3 Measures

The Traditional Masculinity-Femininity questionnaire (TMF, Kachel et al., 2016) is a measure of self-ascribed femininity and masculinity. The instrument consists of six items measuring gender-role adoption (1 item), gender-role preference (1 item) and gender-role identity (4 items). Participants had to rate each item on a continuous, 7-point scale ranging from very masculine (1) to very feminine (7). Accordingly, participants choosing the middle value (i.e., 4) can be considered moderately feminine and masculine.

Bem’s Sex Role Inventory (BSRI, Bem, 1974) and the Personal Attributes Questionnaire (PAQ, Spence et al., 1975) are two older instruments reflecting a trait approach that posits that masculinity and femininity are closely related to the individual’s instrumental and expressive traits. Studies have shown that the two subscales in the two instruments are weakly correlated, indicating that masculinity and femininity are two independent, unipolar dimensions. Based on results regarding instrument reliability reported by Fernández and Coello (2010), we decided to apply the short versions of the questionnaires (12 items BSRI and 16 items PAQ, the short version of the PAQ is now considered standard). Participants rated themselves on each item, on a scale from 1 (never or almost never true) to 7 (almost always true); a higher score indicated a higher masculinity/femininity.

Due to sufficient sample size we used random 66% of the sample to conduct Exploratory factor analysis (EFA) and 33% of the sample to conduct Confirmatory factor analysis (CFA) to verify the factor structures of BSRI and PAQ.

Exploratory factor analysis (EFA) was conducted on all items of BSRI (n = 313) with oblique rotation (oblimin). Bartlett’s test of sphericity, χ² (66) = 2188.46, p < 0.001 and KMO = 0.86, indicated that correlations between items were sufficiently large for EFA. An initial analysis was run to obtain eigenvalues for each component in the data. Two components had eigenvalues over Kaiser’s criterion of 1 and in combination explained 58% of the variance. The scree plot showed inflexions that would justify retaining these 2 factors. Component 1 represented masculinity and consisted of all original BSRI items of masculinity; Component 2 represented femininity and consisted of all original BSRI items of femininity. Internal consistency of components is in Table 1.

We applied the Confirmatory Factor Analysis (CFA) to assess the fit of two factors model of BSRI on a 157 randomly selected participants. The model showed acceptable fit (GFI = 0.85; NFI = 0.85; CFI = 0.89; RMSEA = 0.11; SRMR = 0.18; RFI = 0.82; IFI = 0.90; PNFI = 0.70). Moreover, the factors had excellent composite reliability (CR) and convergent reliability (average variance extracted – AVE). For masculinity CR = 1.00 and AVE = 1.00, for femininity CR = 0.98 and AVE = 0.91.

Similarly, exploratory factor analysis (EFA) was conducted on all items of PAQ (n = 313) with oblique rotation (oblimin). Bartlett’s test of sphericity, χ²(120) = 1633.29, p < 0.001 and KMO = 0.88, indicated that correlations between items were sufficiently large for EFA. Two components had eigenvalues over Kaiser’s criterion of 1 and in combination explained 39% of the variance. The scree plot showed inflexions that would justify retaining 2 factors: Component 1 represented masculinity and consisted of all original PAQ items of masculinity and Component 2 represented femininity and consisted of all original PAQ items of femininity. Internal consistency of components is in Table 1.

Next, we the Confirmatory Factor Analysis (CFA) was conducted to assess the fit of two factors model of PAQ on a 157 randomly selected participants. The model showed acceptable fit (GFI = 0.86; NFI = 0.77; CFI = 0.86; RMSEA = 0.09; SRMR = 0.14; RFI = 0.73; IFI = 0.86; PNFI = 0.67). The factors had good composite reliability (CR) and weaker convergent reliability (average variance extracted – AVE). For masculinity CR = 0.75 and AVE = 0.28, for femininity CR = 0.79 and AVE = 0.32.

The Decision Outcome Inventory (DOI, Bruine de Bruin et al., 2007) measures experiences of life events that are the result of decision-making. Participants had to choose from a list of life events where individuals may make the wrong decision (for example buying food, weddings) and those who had experienced these life situations responded to the items concerning the negative consequences of the decision (for example, having to throw out food they had bought). The last six items (being imprisoned, involved in a public fight or argument; being bankrupt; forgetting someone close’s birthday; being diagnosed with Type 2 diabetes; having a fracture due to a fall, slip, or a misstep) were answered by all the participants regardless of whether they had experienced the situation or not. The total score from the negative-consequence items was calculated as follows: the last six items were automatically awarded 1 point (for example, a participant who had been in prison earned 1 point); for the other items, we calculated the value using the following formula:

For example, participants who threw out food they had bought earned 0.12 points:

The lower the score, the less negative the real-life outcomes the participant had experienced in life, the less risky behaviour they reported.

4.1 Descriptive Statistics and Sex Differences

The descriptive statistics of the variables including internal consistency of instruments are given in Table 1. Men expressed stronger tendency to risky behaviour than women, and there were sex differences in all variables except masculinity (BSRI, PAQ): women had higher femininity and identified themselves more as feminine than men (sex differences in characteristics from BSRI and PAQ are displayed in the Figure 1).

Figure 1:

Masculine and feminine traits – sex differences. Note: BSRI, Bem’s sex role inventory; PAQ, personal attributes questionnaire.

4.2 Gender and Sex Overlap

To examine the self-ascribed femininity and masculinity and traits femininity and masculinity overlap and gender and biological sex overlap, we divided participants into three types according to mean and standard deviation in the TMF: as masculine were labelled those with a low score (M − 1SD; range 1–1.40), as feminine those with a high score (M + 1SD; range 5.92–7) and as unspecific or moderately feminine and masculine those between low and high score (range 1.41–5.91). Next, using K-mean cluster analysis (with 4 number of clusters, silhouette measure of cohesion and separation was 0.26), we divided participants into four groups according to their BSRI and PAQ scores: feminine (high in femininity and low in masculinity), masculine (high in masculinity and low in femininity), androgyny (high in femininity and masculinity) and unspecified (low in femininity and masculinity).

An association between the TMF typology and BSRI/PAQ typology could be observed (χ² (6) = 72.04, p < 0.000). For more detail see Table 2; 36.4 percent of participants identified themselves as masculine had also traits masculinity (BSRI/PAQ) and 33.9 percent of those identified themselves as feminine had trait femininity (BSRI/PAQ). There were more feminine participants that had trait masculinity (14.8%) than masculine participants with trait femininity (9.0%). Those who identified themselves as feminine are more specified than masculine: only 9.5 percent of feminine, but up to 23.2 percent of masculine ones, were unspecific (low both in masculinity and femininity).

Next, comparing gender and biological sex (Table 3); an association between gender and biological sex could be observed (TMF: χ² (2) = 238.07, p < 0.000; BSRI/PAQ: χ² (3) = 32.35, p < 0.001). We found that 50.8 percent of men described themselves as masculine (TMF), while 33.2 percent of men had masculinity traits in the BSRI/PAQ. Similarly, 51.7 percent of women identified themselves as feminine, while 33.2 percent of women had femininity traits in the BSRI/PAQ.

4.3 Correlation Analysis

Correlations between all variables are in Table 4. The age related positively to femininity (BSRI, PAQ) and weakly with masculinity (PAQ).

The BSRI correlated with the PAQ in the expected way: there was a strong positive correlation between the masculinity subscales of the BSRI and the PAQ as well as between the femininity subscales of the BSRI and the PAQ. Moreover, there was a weak positive correlation between the TMF and the femininity subscales of the BSRI and the PAQ (and no correlation between the TMF and the masculinity subscales of the BSRI and the PAQ).

4.4 Construct Validity of BSRI, PAQ and TMF

The construct validity of instruments BSRI, PAQ and TMF was examined through mutual correlations and paired comparison. A weak correlation between masculinity and femininity as traits (BSRI and PAQ) and the identification as masculine/feminine (TMF) were due to continuous scale of rating TMF items (ranging from very masculine to very feminine). To verify whether the subjective gender-related self-concept (TMF) reflects personality traits (BSRI, TMF) we divided participants into types using the TMF score, as mentioned above (Gender and sex overlap section). We compare these three types of participants in their traits masculinity and femininity. Analysis of variance (ANOVA) revealed significant differences both in masculinity and femininity (BSRI, PAQ). Descriptive statistics and post-hoc tests are in Table 5. Participants who identified themselves as moderately feminine and masculine (unspecific) had lowest score of masculinity and femininity also in BSRI and PAQ. Interestingly, there were significant differences between masculine and feminine types of participants in femininity but not in masculinity (BSRI, PAQ): feminine participant had significantly higher trait femininity than masculine participants but masculine participant did not have significantly higher trait masculinity than feminine participants.

The relationships between BSRI and PAQ trait masculinity and BSRI and PAQ trait femininity were examined by paired t-tests. There were significant differences (t = 35.414; p < 0.001; d = 1.401) between BSRI masculinity (M = 4.42; SD = 1.03) and PAQ masculinity (M = 3.39; SD = 0.59) and also between BSRI femininity (M = 5.10; SD = 0.96) and PAQ femininity (M = 3.78; SD = 0.57), t = 53.4475; p < 0.001; d = 2.115. To sum up, participants scored higher in BSRI than in PAQ.

4.5 Prediction of Gender-Related Behaviour

We found that risky behaviour correlated negatively with the femininity subscales of the BSRI and PAQ and TMF (Table 4). To examine the effect of biological sex and gender on specific risky behaviours (DOI), we calculated the hierarchical regression analysis.

A hierarchical regression was calculated to predict risky behaviour based on gender (BSRI, PAQ, TMF) and biological sex. The biological sex (dummy variable, man = 0, woman = 1) was entered in the first step, self-ascribed femininity and masculinity (TMF) in the second step, masculinity and femininity (BSRI, PAQ) entered in the third step of regression. Regression statistics are in Table 6.

The hierarchical multiple regression revealed that at step 1, the biological sex contribute significantly to the regression model, F (1468) = 21.635, p < 0.001 and explained 4.4% of the variation in risky behaviour. Introducing the self-ascribed femininity and masculinity (TMF) explained only an additional 1.0% of the variation in risky behaviour and this change in R² was significant, F (2467) = 13.312, p < 0.001. Finally, introducing masculinity and femininity (BSRI, PAQ) explained an additional 3.1% of the variation in risky behaviour and this change in R² was significant, F (6463) = 7.148, p < 0.001. Only masculinity was significant predictor of risky behaviour; surprisingly, BSRI masculinity was positive predictor and PAQ masculinity was negative predictor of risky behaviour.

5.1 The Study Limitations and Future Directions

Despite our best efforts, the study is not free from limitations. Primarily, our sample was too small and homogenous for us to be able to address the issue of intersectionality, which can play a significant role in gender differences research and particularly the role of stereotypes in behaviour. Second, as all the variables were elicited through self-reports, it is possible that more instrumental individuals presented themselves as more risk-seeking, believing that both risk-seeking and instrumentality are desirable characteristics (self-serving attributional bias).