By Michael J. McCormick University of Houston - Clear Lake, Jesús Tanguma University of Houston - Clear Lake, and Anita Sohn Lopez-Forment Texas A&M University

Academic Citation: McCormick, M., Tanguma, J. and Sohn, A., " Gender Differences in Beliefs About Leadership Capabilities: Exploring the Glass Ceiling Phenomenon with Self-Efficacy Theory," Kravis Leadership Institute Leadership Review, Spring 2003.

About the Authors: Michael J. McCormick is Assistant Professor, Industrial/Organizational Psychology at the University of Houston - Clear Lake where he is responsible for the masters program in I/O psychology. In addition to degrees in law and engineering, he holds a doctorate in industrial/organizational psychology from Texas A&M. His research interests include leadership training, small group leadership, and gender issues related to leadership.
Jesús Tanguma received his Ph.D. from Texas A&M University. Currently, he is an Assistant Professor of Foundations at the University of Houston Clear-Lake, where he teaches courses in research methods and statistics. For the past two years he has researched topics in applied statistics, teaching statistics online, and survey design.
Anita Sohn López-Forment received her Ph.D. from Texas A&M University. She is a bilingual school psychologist specializing in early childhood development and program evaluation.

THEORY

Women currently comprise about one-half of the national labor force in the U.S., and over 40% of all management and supervisory positions. This is a noteworthy change considering that in 1972 women occupied only 17 percent of these positions. Yet, despite this rising tide of female managers, women are not moving into higher levels of responsibility. Women hold less than five percent of executive positions in large companies (The Glass Ceiling Initiative, 1996). This lack of progress has been referred to in both the popular press and the academic literature as the glass ceiling problem (Ragins, Townsend & Mattis, 1998).

Understanding all the obstacles faced by women in their efforts to advance professionally is critical to promoting diversity in leadership. The prevailing explanation for women's slow progress to top leadership posts has focused on external factors such as sex-based discrimination, an inhospitable corporate culture, higher performance standards for women, exclusion from informal networks, and work/home conflicts (Northouse, 2001). In contrast, consideration of possible internal barriers has been limited, even though descriptive studies of successful female executives have noted the importance of certain personality factors (Morrison, 1992). Prominent among these personal variables is self-confidence, a generalized belief in one's own ability to meet the demands of any situation.

Published findings suggest there is a gender difference in beliefs about personal capabilities. Studies indicate that women in general have lower self-confidence than men, and in the business arena, female managers report lower self-confidence than their male counterparts (Morris, 1998; Morrison, 1992; Tharenou, Latimer & Conroy, 1994; Tsui, 1998). Apparently, even though there are no reported male-female differences in leadership ability (Eagly et al., 1995), women report being less self-assured about their leadership capabilities. However, since self-confidence is typically used as a criterion for screening management candidates (Northouse, 2001), expressed and/or perceived gender differences in self-confidence for the leadership task may be contributing to the modest number of females in mid and upper level management.

To raise the issue of gender differences in leadership self-confidence beliefs is NOT to shift the blame to women themselves. To do so would ignore the very real challenges confronting women in the workplace. Yet, if low task-specific self-confidence is a problem for women in management, then identifying the psychological processes sustaining this situation is warranted. This would provide researchers and practitioners with a more complete picture of the glass ceiling problem, thus leading to more effective interventions. The purpose of this study was to explore whether an internal psychological factor, low self-confidence for the leadership task (or low leadership self-efficacy), might be contributing to the lack of leadership diversity. Based upon self-efficacy theory (Bandura, 1986), a model of how self-efficacy beliefs develop, and how these self-evaluations influence leadership performance was tested. Figure 1 shows the hypothesized model, which will be developed in more detail in the following sections.

FIGURE 1
Study Variables and Predicted Relationships

Self-Confidence, Self-Efficacy and Leadership Diversity

Self-confidence is an admired personal quality about which people often express interest. And, as it is in so many other human endeavors, self-confidence is essential for success in the leadership role as well (Locke, 1991). However, for model building purposes, self-confidence is not a construct that is part of a validated theoretical system specifying its determinants, processes, and effects. Being a trait, it has predictive value but no explanatory power. Researchers have typically addressed the situation by substituting Bandura's (1986) self-efficacy construct for the self-confidence variable. They have done so for two reasons. First, the two are conceptually similar. For example, self-efficacy has been defined as: "Confidence in one's capabilities" (Spreitzer, McCall, &Mahoney, 1997, p.8), or "Task specific self-confidence" (Locke & Latham, 1994, p.17). A second reason is that self-efficacy is the central variable in a comprehensive and well-supported model of human functioning called social cognitive theory developed by Bandura (1997).

Research on social cognitive theory has found that personal efficacy beliefs influence the goals people choose, their aspirations, how much effort they exert on a given task and how long they will persist in the face of difficulties, obstacles and setbacks. Also, findings have linked self-efficacy with whether a person experiences self-hindering or self-aiding thought patterns, how well a person responds to taxing and threatening situations, and their resilience to adversity and setbacks. (See Bandura, 1997, for a review). Furthermore, a recent meta-analysis by Stajkovic and Luthans (1998) indicated that an employee's task specific self-confidence is an important causal variable in the job performance process.

Coincidentally, studies of effective leaders have described them as being highly motivated, persistent, resilient, confident, and effective problem solvers (Locke, 1991). Viewed from a social cognitive perspective, this suggests that what leadership researchers have been describing for years is a person with high self-efficacy for the leadership task. This implies that belief in one's leadership capabilities is involved in the leadership performance.

The idea that a healthy sense of personal efficacy as a leader (leadership self-efficacy) is a necessary factor for successful leadership is especially relevant for developing a comprehensive explanation for the glass ceiling phenomenon. As noted previously, consideration of possible gender differences in task-related cognitions has received very limited attention in the research community. Yet, the reported differences in self-confidence between men and women may offer an important but overlooked clue regarding an additional reason for the glass ceiling situation. Low leadership self-efficacy may be acting to constrain women's managerial progression. Therefore, based on the previous discussion, the following two hypotheses are made.

Hypothesis 1: Females will report significantly lower leadership self-efficacy than males of similar age and experience.

Hypothesis 2: An individual's belief in their leadership capabilities, that is, their leadership self-efficacy, will be positively associated with leadership behavior.

The Efficacy Development Model

The line of research begun by Hoyt (2002) and her colleagues highlights the singular importance of a woman's belief in her leadership capabilities, especially in non-traditional leadership roles. They found that self-efficacy for leadership influenced how a woman cognitively appraised a leadership situation. Highly confident women regarded a stressful leadership situation (for example, a woman leading a team comprised mostly of men) as a challenge, whereas less confident women regarded a similar situation as threatening. As a result, the less confident women performed poorly as group leaders, were more anxious prior to leading a group, and identified less with the leadership role. All this because they lacked faith in their leadership capabilities. Hoyt's work also lends support to Murphy's (2001) self-regulation model of leadership, which posits leadership self-efficacy as a central cognitive factor in the leadership process.

Given the potential importance of leadership self-efficacy, understanding the antecedents and processes involved in developing efficacious leadership beliefs becomes critical to designing interventions more likely to promote the advancement of women. According to Bandura (1986), perceived efficacy is the foundation of human actions, and efficacy beliefs are derived from four kinds of experiences. The first, personal performance accomplishments, is the most influential. The more successful experiences a person has had, the more confident he or she will be. The second most influential is role models. Exemplars are particularly important for developing interpersonal skills and for providing inspiration, especially when the model is a highly regarded other. The third is performance feedback and the opinions of credible and knowledgeable people such as bosses, mentors, coaches, or parents. The fourth is physiological condition and mood states. The better one feels physically and emotionally, the more confident one will be. And, the more efficacy enhancing activities and information a person has experienced, the greater will be their task-specific self-efficacy beliefs, which in turn positively impact leadership performance. For purposes of further discussion, the four sources of efficacy beliefs posited by Bandura will be combined into a single variable referred to hereafter as developmental experiences. Based upon the previous review, the following hypothesis is presented.

Hypothesis 3: Leadership-related developmental experiences will be positively related to leadership self-efficacy which, in turn, will be positively associated with leadership behavior.

Self-efficacy beliefs develop over time through life events, and it is during these many experiences that the four types of efficacy information (developmental experiences) discussed previously are encountered. The number and type of potentially efficacy enhancing developmental experiences an individual encounters is affected by both external and internal factors. In the case of the glass ceiling phenomenon, data-based explanations (which represents the dominant research paradigm) have focused exclusively on externally erected barriers to women's managerial advancement based upon sex (see Northouse, 2001 for a review). For example, women report having had fewer challenging leadership role experiences, less leadership training and development, and fewer mentoring experiences than their male co-workers, all because of external barriers such as male prejudice, gender stereotyping, and culture-related preconceptions. Thus the external perspective would predict that an individual's sex would be related to their level of leadership-related developmental experiences, and this would account for the observed gender differences in self-confidence for the leadership task. This line of reasoning leads to the next hypothesis.

Hypothesis 4: An individual's sex will be associated with leadership-related developmental experiences.

On the other hand, self-efficacy theory says that the efficacy development process is also influenced by internal psychological variables. Various personality variables have been found to affect the types of activities a person chooses to experience and what information is attended to in the social environment (Bandura, 1997). An individual difference characteristic that appears to be particularly relevant to women, leadership and the glass ceiling phenomenon is sex-role identity, the socially communicated expectations regarding the appropriate behaviors, personal qualities, and social roles for men and women. As currently conceptualized, sex-role identity is comprised of two independent dimensions each encompassing a cluster of socially endorsed masculine or feminine attributes and behaviors (Bem, 1993).

Research into what underlies sex differences in perceived efficacy for different careers reveals sex-role identity rather than sex per se to be the critical factor (Hackett, 1995; Matsui & Onglatco, 1991). It is women with highly stereotypic feminine orientations (high femininity scores and low masculinity score) who display self-doubts about their capabilities for non-traditional careers. This would be especially true for the business leadership role. Since group leadership is considered to be a male activity by both men and women (Eagly & Karau, 1991; Lord, DeVader, & Alliger, 1986), women with a highly feminine identity orientation would avoid leadership self-efficacy enhancing opportunities because it would be inconsistent with their sex-role identity. And having fewer leadership role experiences would lead to low leadership self-efficacy, that is, low self-confidence for the leadership task. However, those individuals, both men and women, who have endorsed many of the stereotypically masculine characteristics will be more comfortable accepting a leadership role. This suggests that an important internal factor contributing to the lower number of females in the management pipeline could be the masculine dimension of a person's sex-role identity, because of its potential impact on the development of leadership self-efficacy perceptions. The feminine dimension, on the other hand, would have little or no impact. Therefore, the following two hypotheses are proposed.

Hypothesis 5: The masculine dimension score of an individual's sex-role identity will be positively associated with leadership-related developmental experiences.

Hypothesis 6: The feminine dimension score of an individual's sex-role identity will minimally related with leadership-related developmental experiences.

METHOD

A survey was employed to gather information about study participants such as their age, sex, and class rank. Data was also obtained regarding their sex-role identity, previous leadership role experiences, prior leadership training, leadership self-efficacy perceptions, and frequency of attempts to assume a leadership role in a group setting.

Participants

The sample was comprised of 404 male and female volunteers recruited from undergraduate classes at a major southwest university. The males and females did not differ significantly in terms of age and education level. The mean age of the group was 22 years (SD= 2.38), and 53% were females. Study participants signed informed consent forms after being assured their responses would be confidential, and then completed the questionnaire. Afterwards, they were debriefed, and their participation in this study reported to their instructors.

Measures

Sex-Role identity. The Bem Sex-Role Inventory (BSRI) was used to measure the masculine and feminine dimensions of each participant's sex-role identity (Bem, 1993). Each participant's masculine and feminine sex-role dimension score was calculated by averaging their responses across the twenty items for each scale. The internal consistency reliability for the masculinity scale (MSR) was .88, and the test-retest reliability was .86. The femininity scale (FSR) demonstrated a coefficient alpha reliability of .80 and a test-retest reliability of .79. The test-retest reliability values were based on a sub-sample of 280 participants who completed the BSRI twice. The interval between test administrations was three months.

Developmental experiences. Respondents were asked to recall the number of times they had acted in the role of a group leader responsible for organizing, directing, and motivating the actions of others in a variety of leadership settings (i.e., clubs, class projects, service groups, church, community, ROTC, etc.) during their high school and college careers. Responses were made on a five-point scale ranging from "No leadership experiences" (1) to "More than six experiences" (5). Also, participants were asked to recall how many formal leadership training classes and seminars they had previously attended during their high school and college careers. Responses were made on a 5-point scale ranging from "No training" (1) to "More than 6 courses" (5). This item was included because many leadership development programs include role plays, skills training, viewing of videos depicting effective leaders, and positive encouragement from bosses, peers, direct reports and trainers, all sources of efficacy information (Conger, 1992). The developmental experiences variable was calculated by summing the leadership role item with the leadership training item.

Leadership behaviors. Next, participants were asked how often they had sought to be a group leader when they were given the opportunity to assume a leadership role during their high school and college carriers. A 5-point scale was used with the following behavioral anchors: "Never" (1), "Less than half the time" (2), "About half the time" (3), "More than half the time" (4), and "Always" (5).

Leadership self-efficacy. Participants' leadership self-efficacy was assessed using a measure developed by Kane and Baltes (1998). It is an 8-item questionnaire which asks respondents to report their ability to perform the following functional leadership activities: (1) perform well as a leader across different group settings, (2) motivate group members, (3) build group members' confidence, (4) develop teamwork, (5) "take charge" when necessary, (6) communicate effectively, (7) develop effective task strategies, and (8) assess the strengths and weaknesses of the group. Response options ranged from 1 (no confidence) to 7 (100% confident). Item responses were summed and an average was calculated to arrive at an overall leadership efficacy score.

FIGURE 2
Results of Model Fit Analysis

RESULTS

A two-sample, independent t-test was computed using SPSS 10.0 to determine if a statistically significant difference existed between the males and the females on their leadership self-efficacy. The level of significance was set a priori at a = 0.05. Descriptive statistics indicated that the mean for the males was 5.67 (SD = 0.81) whereas the mean for the females was 5.24 (0.98). Similarly, results of the independent t-test indicated a significant difference between the males and females (t(404) = 4.72, p < 0.01). In addition, the calculated effect size (d = 0.47) was medium (Cohen, 1992).

Data were analyzed using the analysis of moment structures (AMOS; Arbuckle, 1999) software package and the models tested were covariance structural models with six observed (manifest) variables. AMOS is one of several (e.g., SAS PROC CALIS, EQS, PRELIS) latent variable structural equations modeling (SEM) programs commonly used to create and evaluate theoretical models. Standard deviations and intercorrelations for the study's six variables are presented in Table 1.

Table 1 (in PDF format)

The present analysis followed a two-step procedure based in part on an approach recommended by Anderson and Gerbin (1988). In the first step, confirmatory factor analysis was used to develop a measurement model that demonstrated an acceptable fit to the data. In step two, the measurement model was modified (based on modification indices as well as on theoretical reasons). The resulting model was used as the theoretical (causal) model of interest. This theoretical model was then tested and revised until a statistically/theoretically acceptable model was found.

The theoretical model was estimated using the maximum likelihood method, and the resulting chi-square for the model was statistically significant, x²= 389.977, df = 10, and p< 0.001, see Table 2. Additionally, Table 2 presents the goodness of fit indices for the theoretical model. Goodness of fit indices (e.g., NFI, IFI, CFI) are derived to range in value from zero to unity, with values close to 0.95 indicative of superior fit (Hu & Bentler, 1999). As evidenced from Table 2, all fit indices are in the acceptable range.

Table 2 (in PDF format)

Although the fit indices are indicative of good fit and the chi-square value is statistically significant, the chi-square statistic is well known to be dependent on sample size and have very stringent assumptions. If the assumptions (e.g., multivariate normality) hold, this statistic may be used to test the null hypothesis that the model fits the data. According to Bollen (1989a), the higher the probability value associated with the chi-square statistic, the closer the fit is between the hypothesized model and the perfect fit. However, researchers (e.g., Bearden, Sharma, & Teel, 1982; Fan, Thompson, & Wang, 1999; Tanguma, 2001) have repeatedly stated that the chi-square statistic is very sensitive to sample size and departures from multivariate normality, thus, very often resulting in a rejection of a well-fitting model. Consequently, researchers (James, Mulaik, & Brett, 1982; Joreskog & Sorbom, 1989) have recommended that the model chi-square statistic be used as a goodness of fit index where smaller chi-square values are indicative of a better model fit. Similarly, Carmines and McIver (1981) have indicated that models with chi-square ratios of less than three are indicative of better model fit.

A number of other results, however, indicated that there was in fact a problem with the model's fit. For example, one of the paths linking two constructs proved to be nonsignificant. Table 3 shows that the standardized path coefficient for the path from sex to developmental experiences was only -0.05. The critical ratio (estimate/standard error of the estimate) for this path was only -1.000. Thus, this path was dropped from the model in subsequent analyses.

Table 3 (in PDF format)

A Wald test (Bentler, 1989) suggested that if sex were correlated with masculine sex-role identity as well as with feminine sex-role identity, the value of the chi-square will be reduced. Because the addition of the paths could be justified on theoretical grounds, the paths were added and the resulting model (revised model) was then estimated. Fit indices for this model are presented in Table 2. Once again, overall goodness of fit indices for the model were indicative of good fit, with values on the NFI, IFI, and CFI in excess of 0.900. However, the value of the chi-square statistics (x² = 295.804) is still very large compared to the degrees of freedom. But since x² = (N-1)Fmin(sample size minus one, multiplied by the minimum fit function), this value tends to be large when the model does not hold and the sample is large (Joreskjog & Sorbom, 1993). Thus, findings of a large x² relative to degrees of freedom are common in most SEM empirical research (Byrne, 2001). Table 3 shows that the path coefficients in the revised model were all statistically significant. Adding such paths would be consistent with the findings from studies on career choice by Hackett (1995), which suggested that although gender does not have a direct effect on whether or not females pursue traditionally male occupations such as managerial positions, gender does have an indirect effect on career choice through its influence on self-efficacy development.

Fit indices for all models are presented in Table 2. It can be seen that the fit indices (i.e., NFI, IFI, and CFI) were not only above the 0.900 but were also higher than those displayed by the initial theoretical model. A chi-square difference test comparing the initial theoretical model to the revised model revealed a statistically significant difference value of 389.977 -295.804 = 94.173 (df = 1, p <0.001). This finding shows that the revised model provided a better fit to the data that was significantly better than the fit provided by the initial theoretical model. Thus, the addition of the new paths and deletion of the path from sex to total developmental experience are justified.

Figure 2 displays standardized path coefficients for the revised model. It can be readily seen that all coefficients were significant and in the predicted direction. R²values showed that leadership self-efficacy accounted for 40% of the variance in the frequency of attempting to assume leadership roles, while total developmental experience accounted for 20% of the variance in leadership self-efficacy. Similarly, 16% of the variance in total developmental experiences was explained by male sex-role identity. However, participant's sex had no direct impact on the leadership self-efficacy development process.

DISCUSSION

The prevailing explanation for the glass ceiling problem has centered almost exclusively on external barriers to women's managerial advancement. Consideration of possible internal psychological constraints has received little attention in the mainstream literature. Drawing upon self-efficacy theory and previous findings indicating that female managers report being less confident than their male co-workers, we argued that leadership self-efficacy, an individual's confidence in their leadership capabilities, may provide a valuable new perspective on the glass ceiling situation.

In our sample females reported significantly lower leadership self-efficacy perceptions than the males. In fact, the calculated effect size of d = .47 indicates that this gender difference is "an effect likely to be visible to the naked eye of a careful observer" (Cohen, 1992, p. 156). In addition, leadership self-efficacy affected functional leadership behavior. It had a direct impact on the frequency with which a person attempted to take on a leadership role accounting for 40% of the variance. This finding is consistent with the large body of research on self-efficacy that has shown that the more confident one is that one can successfully perform a task, the more frequently one will engage in the activity (Bandura, 1997; Stajkovic & Luthans, 1998). Furthermore, these findings also are consistent with the empirical results presented by Hoyt (2002) and the proposed importance of leadership self-efficacy in Murphy's (2001) model of leadership self-regulation.

Proposed antecedents of leadership self-efficacy perceptions were also examined. Consistent with self-efficacy theory, the number of previous leadership role experiences and leadership training classes (or developmental experiences) a person had had directly influenced the leadership self-efficacy estimate. This means that leadership-related developmental experiences contribute to personal confidence for the leadership task. However, the amount of variance accounted for (only 20%) suggests other factors and processes are probably involved. Bandura (1997) contends that personal, social and situational variables influence what efficacy-related information is attended to and how it is evaluated and integrated into the efficacy estimate. Given the potential significance of the leadership self-efficacy construct, future research should be directed towards identifying possible moderators and determining their role in the efficacy estimation process. A particularly noteworthy one is attributional gender effects. Women and men tend to see different causes for identical levels of their own performance (Rosenthal, 1995). Males usually attribute successful performance to their ability (which is efficacy enhancing), while females usually attribute their successes to hard work or the help of others (which contributes little to efficacy enhancement).

The final set of findings revealed what part study participants' sex played in the leadership self-efficacy development process. The dominant glass ceiling explanation, which focuses on external barriers to women's management development, would argue that gender differences in leadership self-efficacy is likely the result of sex-based differences in leadership-related development experiences. Interestingly enough, sex had no influence on the level of leadership-related developmental experiences (i.e., leader roles and leader training) the females and males in this study had previously experienced. In short, sex did not explain the gender differences in leadership self-efficacy. Instead, the individual's sex-role identity was the key factor, particularly the masculinity component. The more masculine-type behaviors a person had incorporated into their self-concept, the greater was the number of leadership-related developmental activities engaged in. And the more leadership-related developmental experiences a person had encountered, the greater was his or her leadership self-efficacy.

These results highlight how the gender-role socialization process can act to constrain women's managerial advancement through its influence on the self-efficacy development process. As previous research has revealed, women base their occupational choices more heavily on their perceived efficacy than on the potential benefits the vocation offers (Hackett, 1995). Thus, young women may be self-selecting out of the business management career path because of a weaker sense of efficacy that they can master the educational requirements and job functions of a traditionally male-gendered occupation.

Implications

The finding that leadership self-efficacy is involved in the leadership process has an important implication for glass ceiling research. For leadership to occur in any group, someone must attempt to assume the leadership role. It is the essential first step to becoming recognized as someone with leadership potential. In a business setting, showing such initiative typically garners the attention of important decision-makers, and is one reason why young female professionals are advised to volunteer for special projects and high profile assignments as a strategy for management promotion (Northouse, 2001). However, if a woman has little faith in her leadership abilities (or low leadership self-efficacy), then the previously discussed results suggest that she would be reluctant to take on the very types of activities that could lead to higher levels of responsibility. Thus, she would never be identified and placed into the management development pipeline. Evidence presented by Tannen (1995) and reported in Wells (2001) lends support to this scenario, as does the work of Hoyt (2002).

Tannen (1995) observed the decision-making process at a large, multinational corporation involving identification of high potential young managers. When none of the female candidates were singled out for promotion, the reason given by senior managers was that the women were not ready for advancement because they lacked the necessary confidence. Wells (2001) reported that the only competency on which women managers scored lower than men was self-confidence, and that this lack of confidence was self-limiting. When asked about their career aspirations, the women were looking only about two positions up from their current one.

A strategy for addressing the glass ceiling problem which has not been explored is the use of training specifically designed to increase women's beliefs in their leadership capabilities. Fortunately, self-efficacy theory provides a framework for structuring activities that will increase trainees' task-specific self-confidence. Published studies support its usefulness (see, Cole & Latham, 1997; Eden & Aviran, 1993; Gist, 1989). A training program that includes content instruction, guided enactive mastery, role-modeling, and positive persuasory messages is the most effective, especially if the training program concludes with a module on developing a post-training, self-management plan.

Another strategy for raising the leadership self-efficacy beliefs of women is attributional re-training. Such interventions have been successful in persuading women to interpret the reasons for their effective performance differently, with resulting improvements in self-confidence, motivation, and job performance (Rosenthal, 1995). Perhaps re-attribution training could be included as part of a company-wide initiative to enhance the leadership capabilities of female employees, since it could result in increased self-efficacy for the leadership task.

Limitations

Every study has its limitations, and this one has at least five. First is drawing causal relationships based upon cross-sectional data. While in general this is problematic, the hypothesized causal relationships were theory driven, and findings here were consistent with results from prior investigations that tested parts of the proposed model (Bandura, 1997). Longitudinal studies involving these constructs are needed.

Second is common method variance. Measuring individual level variables at the same time and with the same survey can inflate the relationships among the variables. Furthermore, attempted leadership, prior leadership role experiences, and leadership training experiences were assessed with single item measures. Different methods for evaluating these constructs need to be developed and used in any future investigations.

Third, is the characteristics of the sample. Because male and female college students comprised the sample, the merits of the study may be questioned. However, studies of career choice and development summarized by Hackett (1995) point out that decisions to pursue a management career occur during the college years. A low sense of efficacy for the leadership task would likely deter females from positioning themselves in jobs that lead to management progression. Nevertheless, the proposed model must also be tested using data gathered in different settings involving different participants, such as older more experienced men and women employed in corporate settings, before the generalizability of the model can be established.

Fourth is the leadership self-efficacy measure. It assesses a person's confidence on only eight leadership activities. The instrument needs to be more fully developed to reflect the complexity of the leadership process by basing it on an acceptable, competency-based model of leadership. In addition, the psychometric characteristics of the instrument need to be explored further.

Fifth, while the Chi-square fit index is large for the revised model suggesting that the model is a poor fit to the data, that is a result of the large sample size. The other four goodness of fit indices (see Table 2) approach one in value reflecting a model that is a good fit to the data.

Conclusion

There is now a consensus among policy makers, business leaders and the public that obstacles to the professional advancement of women do exist. Identifying all the factors contributing to the glass ceiling phenomenon is essential before lasting progress in achieving leadership diversity in the workplace can be made. The present study contributes (1) by demonstrating that leadership self-efficacy is related to leadership behavior, (2) by revealing that females report significantly lower leadership self-efficacy than males and (3), by demonstrating that socialization factors may be of more importance than situational constraints. This all suggests that low leadership self-efficacy among women (a personal belief that can be altered under certain conditions) may be a factor contributing to the glass ceiling problem. Further research is warranted.

REFERENCES

Wells, S. J. (2001). A female executive is hard to find, HR Magazine, 46, 40-49.