Over the years, many studies have examined the effect of gender on leadership. Since the 1980s, two analyses have made particularly important contributions to this line of research. In the first study,Deaux (1984)argued that there was no biological basis in gender differences in social behavior.
In the second contribution, the meta-analyses of Eagly and colleagues
(Eagly & Johnson, 1990; Eagly & Karau, 1991; Eagly, Makhijani, &
Klonsky, 1992) demonstrated that some weak effects are present in terms of gender differences in leadership effectiveness and perception of leadership.
These differences are mainly due to social categorization. In particular, men are more likely to emerge as leaders when measures of general or task leadership are used, whereas women are more likely to emerge as leaders if socioemotional measures are used (Eagly & Karau, 1991). It seems that when the situational requirements match those associated with the feminine stereotype, women emerge as leaders. According to the social-role theory of gender differences (Eagly, 1987), these stereotypes tend to work against women by creating negative expectations, which mitigate the likelihood that women will be perceived as leaders in task-oriented situations.
Another relevant issue is the effect of solo status or tokenism.
Of particular concern is the extent to which women’s contributions might be devalued when the group includes only one woman (Fuegen & Biernat, 2002).
Using the SRM can clarify the empirical and theoretical meanings of gender differences in the study of leadership, answering long-standing questions and proposing other interesting and intriguing novel queries.
We begin by noting that there is not just one gender-related issue to be addressed in the study of leadership. Among the questions that have been considered are the following:
Are men or women more likely to be seen as leaders?
Do men or women see more leadership?
Are men more likely to be seen as leaders when men (rather than women) are judges?
How are men and women affected by the gender composition of the group? For instance, are women less likely to be seen as leaders when the group includes fewer women?
We examined each of these questions simultaneously using the SRM.
In so doing, we used the data gathered byLord et al. (1978). In that study, there were 24 groups, each with 4 persons in laboratory settings. Specifically, groups of Carnegie–Mellon students worked together on several tasks.
To simplify the analysis, we averaged the five measures of leadership and created three variables: gender of the perceiver, called Pgender (1 ¼ female;
1 ¼ male); target gender, called Tgender (1 ¼ female; 1 ¼ male); and number of females, called Nfem(0–4). Note that the interaction of Pgender
with Tgender captures the difference between same-gendered pairs (þ1) and mixed-gendered pairs (1). We also allowed Nfem to interact with
Pgender, Tgender, and their interaction. In terms of levels of analysis, Pgender
and Tgenderare at the individual level of analysis, their interaction is at the dyadic or relationship level, and Nfemis at the group level.
So far in our discussion of the SRM, we have focused on the components of perceiver, target, and relationship. Each of these is a random variable.
Considering the target effect, our interest in that component lies not in determining which person in particular is seen as a leader in the group, but rather in estimating the population variance due to target. The gender variables are all fixed variables. We are interested in particular values (i.e., the means) of the levels of that variable. For instance, we want to know if men are more likely to be seen as leaders than are women.
The addition of the fixed variables to the SRM greatly complicates the statistical model and the analysis. For technical reasons, the complica-tions are especially problematic when the fixed variables are at the level of relationship (e.g., PgenderTgender). Standard SRM software cannot properly handle such a model. Fortunately, advances in multi-level modeling have been made and computer programs such as SAS can accomplish such analyses. We used SAS to estimate the model and controlled for nonindependence due to the SRM variances and covariances.
In a later section of this chapter, we describe the details of the estimation method.
The model includes seven fixed effects; and the four SRM variances of group, perceiver, target, and relationship; and the two covariances, perceiver–
target and relationship. Table 3 shows the results for the fixed effects.
Table 3. Solution for the Fixed Effects of Gender fromLord et al.
(1978)Study.
Effect Estimate df t p
Pgender
a 0.1503 118 2.01 0.0469
Tgender
b 0.1286 199 1.55 0.1231
PgenderTgender
d 0.0338 141 0.54 0.5879
Nfem
c 0.0384 77.3 0.45 0.6538
PgenderNfem 0.0528 153 0.75 0.4564
TgenderNfem 0.0434 129 0.53 0.5971
PgenderTgenderNfem 0.0669 106 0.88 0.3805
aPerceiver gender coded 1 ¼ female and 1 ¼ male.
bTarget gender coded 1 ¼ female and 1 ¼ male.
cNumber of females in the group, 0–4.
dSame-gender (1) versus mixed-gender (1) dyads.
Note that some of the degrees of freedom in the t tests are fractional;
fractional degrees of freedom can occur within multi-level modeling because error variances are pooled across levels. Interestingly, we found only one statistically significant fixed effect: Female perceivers see more leadership than do male perceivers. The gender of target effect was not significant, and we even found that women tend to be seen as leaders more than men. If men devalue women’s contributions more than women do (Eagly et al., 1992), there should be an interaction between gender of the perceiver and gender of target – something we did not find. Moreover, unlike Fuegen and Biernat (2002), we did not find any statistically significant effects of the gender composition of the group. The analysis in this case resembles a multiple regression, so all effects are controlling for all other effects in the analysis.
Unlike a traditional multiple regression, the analysis allows for all sorts of nonindependence.
Of course, our study had some serious limitations. First, it took place at Carnegie–Mellon University, a school with a reputation for strong quantitative ability. Second, of the 96 participants in the study, only 42 were female. Thus, there were very few groups with large numbers of females. Nonetheless, our work illustrates how many interesting and important questions about leadership and gender can be analyzed.
The results emerging from this study confirm the results from previous studies showing a lack of target gender differences in leadership when measured in a university context (Eagly & Karau, 1991). Moreover, the absence of the interaction effects of the actor’s gender with the gender of target, even when controlling for the composition of the group, confirms a substantial fairness in leadership perceptions. A more intriguing and perhaps original finding is the significant effect due to the perceiver:
It seems, at least in this study, that the bias of a gendered evaluation of leadership is mainly in the eye of the beholders. We know of no prior studies or theories that explicitly analyze gender differences in the perceiver effect of leadership perceptions. Nevertheless, we can speculate that, following the social-role theory of gender differences, women more than men see more influence in the group.
Another explanation of the gender differences in the perceiver effect could be an extension of the ‘‘contrastive effects’’ of stereotyped judgment (Biernat & Kobrynowicz, 1997): Members seem to judge leadership expressed in groups by comparing others to the expectations of their own gender. Thus, women perceive a greater extent of leadership in groups because they judge the influence of other members to be stronger in
comparison to their negative leadership expectations and not in comparison to their actual expressed influence.
In most of the previous studies, this gender difference in the perceiver effect was not measured and controlled. If a study included more men than women, and if women also tended to see more leadership than men, then it would falsely appear that men would be seen more as leaders than women because of the hidden gender difference in the perceiver effect.
One last issue should be addressed. When we study a fixed effect, such as Pgender, we might ask how much of the variance in the perceiver effect gender explains. To answer this question, we conducted the run twice – once with gender variables in the model and once without them – and each time we determined the perceiver variance. Conducting such an analysis, we found that the variance due to the perceiver declined by 12.8% when we included gender variables in the model. Thus, we conclude that gender variables, mainly Pgender, explain 12.8% of the variance in the perceiver effect. Because the perceiver effect itself explains approximately 12.2% of the total variance (seeTable 1), gender explains only 1.6% of the total variance, at least when only the perceiver effect is considered. Including the fixed effects of gender in the model reduced the perceiver variance while having little or no effect on the target or relationship variance. This result bolsters the conclusion that the effect of gender in this study is at the level of the perceiver.
The strategy illustrated here could also be applied to the study of variables that predict the target effect. If such research were carried out, we would have a much better understanding of the percentage of variance that traits explain in the random effect of target than we currently have. That is, because error variance is measured in SRM analyses, a much larger proportion of variance would almost certainly be explained than is possible when using conventional methods.