1 __________________________________________________________________________________________
Gender Differences in Strategic Confidence and Self-promotion in
the Labor Market. Two Experiments
Student: Sebastián Arechaga Garcés Professor: Dr. J. J. (Joël) van der Weele
University of Amsterdam
FACULTY OF ECONOMICS AND BUSINESS MASTER IN ECONOMICS
2
Index
Gender Differences in Strategic Confidence and Self-promotion in the Labor Market. Two Experiments... 3
I. Introduction ... 3
II. Theories and Literature ... 4
Overconfidence and self-promotion as strategic behaviors ... 4
Theories of gender differences on behavior ... 5
III. Data ... 6 IV. Hypotheses ... 7 Hypothesis 1 ... 7 Hypothesis 2 ... 8 Hypothesis 3 ... 8 Hypothesis 4 ... 9 V. Analysis of Experiments ... 9 V.1 Experiment 1 ... 9 Description... 9
Empirical Strategy of Experiment 1 ... 11
Results of Experiment 1 ... 13
Results of the Regression Analysis ... 18
V.2 Experiment 2 ... 24
Description... 24
Empirical Strategy of Experiment 2 ... 25
Results of Experiment 2 ... 26
Results of the Regression Analysis ... 29
VI. Conclusions ... 32
VII. Discussion ... 33
VIII. References ... 34
3
Gender Differences in Strategic Confidence and Self-promotion
in the Labor Market. Two Experiments
Sebastián Arechaga Garcés University of Amsterdam
Abstract: I test experimentally if there are gender differences in self-confidence and self-promotion during labor interviews. In order to do this, I examined two experiments which allow me to test several hypotheses. Main results indicate that there is no consistent evidence that evaluators weight differently self-promotion when it is stated by a male or female applicant. Nevertheless, female applicants incur more social costs for promoting themselves. Also, men demonstrate different levels of self-promotion depending on the gender of the evaluator. Finally, one of the two experiments shows that women are more responsive to feedback than men when they estimate their level of confidence.
I. Introduction
Evidence in behavioral sciences has found that people tend to overestimate their abilities. This behavior leads to negative consequences, such as misallocation of resources (Camerer & Lovallo, 1999) (Malmendier & Tate, 2005) or the disability to enjoy positive outcomes (McGraw, Mellers, & Ilana, 2004). On the other hand, some positive consequences have been found; like increments on effort (Gervais & Goldstein, 2007) and self-worth (Blanton, Pelham, DeHart, & Carvallo, 2001).
Controlled experiments have found that the level of confidence varies depending on the context or task (Klaymana, Soll, González-Vallejo, & Barlas, 1999). People increase their confidence under situations where it is convenient to promote a good self-image. This evidence suggests social roots in self-confidence, because it varies depending on the strategic environment (Mobius & Rosenblat, 2006) (Trivers, 2011) (Charness, Rustichini, & Van de Ven, 2013) (Schwardmann & Van der Weele, 2016).
Other studies have found differences in self-confidence and self-promotion between men and women (Barber & Odean, 2001) (Bengtsson, Persson, & Willenhag, 2005) (Santos-Pinto, 2012). Theoretically, gender differences in self-confidence and self-promotion could explain part of the gender gap in situations such as the labor market (Santos-Pinto, 2012). Popular books and social media have focused on the relevance of confidence, self-promotion and women's empowerment as tools to reduce the gender gap in the labor market. A different approach argues that promoting a positive self-image could lead to other social reprisals. Policy makers have used affirmative action policies to eliminate the gender gap in the labor market. Nevertheless, researchers have found mixed effects of this policy (Coate & Loury, 1993) (Parker, Baltes, & Christiansen, 1997)1.
Also, women who behave overconfidently do not necessarily reduce the gender gap, because women face higher social costs than men when they behave overconfidently (Rudman, 1998) (Ludwig & Thoma, 2014) (Thoma, 2016). Gender role theory states that women who show behaviors associated with male stereotypes could be punished because they are not behaving according to what is expected of women (Eagly & Wood, 1999) (Eagly & Karau, 2002). Also, the fear of backlash inhibits self-promotion in women
1
There are many definitions about affirmative action's. In this literature review, it was considered as a policy that require employers to achieve the same rate of jobs and/or assignments for both groups. In our case, both genders.
4
(Moss-Racusin & Rudman, 2010). This evidence suggests that gender roles may rule the different levels of biased behaviors -such as overconfidence and self-promotion- between men and women.
An interesting question is evaluating whether gender differences in confidence and self-promotion are ruled by differences in costs faced by each gender when they are in strategic environments.
In this research, a strategic environment is defined as a situation where applicants have monetary incentives to convince employers that they had (or will have) a successful performance in a certain task. In
this study I am going to test through two experiments where hypothetical job applicants have the opportunity to deceive evaluators and whether men and women have different strategic behavior in overconfidence and self-promotion. I will also test whether these differences are actually justified by incentives associated with gender roles. If overconfidence and self-promotion are masculine characteristics, we would expect that women who promote themselves and show high levels of confidence are evaluated worse than men with the same characteristics. It would lead women to be less overconfident.
In the following section, I will present the theory and evidence that support overconfidence as strategic behavior and also the theory and evidence that explain gender differences in social behavior. Then, I will formalize the hypotheses.
From section five, the study analyses the two experiments one by one. First, explaining the experiments; secondly, explaining the empirical method used to test hypotheses; and finally, presenting the results. The research finalizes with a conclusion and further discussions.
II. Theories and Literature
This research is based around two main areas of study: a) the study of over-confidence and self-promotion as strategic behaviors, and b) the study of gender differences in human behavior, which is needed to explain differences in the strategic behavior of men and women2.
Overconfidence and self-promotion as strategic behaviors
In strategic situations, overconfidence could be useful to promote a positive self-image without incurring technical or moral costs of lying (Schwardmann & Van der Weele, 2016). There are many strategic situations where deceiving others brings private benefits; like labor interviews (Schwardmann & Van der Weele, 2016), entry to tournaments (Niederle & Vesterlund, 2007) or deterrence of competitors (Charness, Rustichini, & Van de Ven, 2013). This positive self-image would bring private benefits if others have difficulties to estimate the actual performance of the self-promoted subject. When deceiving others is not possible, under-confidence is preferred to overconfidence (Thoma, 2016). Another cost, called paralyzed self-deception, is that being overconfident makes us more vulnerable to deception by others (Trivers, 2009).
Another characteristic of strategic overconfidence and self-promotion is that subjects have interests in maintaining high expectations about their own performance, ignoring negative feedback. Therefore,
2
In this research, gender differences are limited to sex comparison. Sex differences are a simplified way to understand gender, but not synonyms. Critics says that some gender differences can overlap between sexes, and sex comparison could overlook dissimilarities between sexes. Responses to those critics say that sex comparisons allows us to understand differences with a defined method, and not only differences must be studied, but also similarities, and the mechanisms that create them (Eagly, Beall, & Sternberg, 2004).
5
when subjects receive new information to update their beliefs, those that are interested in maintaining a good self-image would be prone to deceive themselves. Experiments have found a positive association between overconfidence and demand for new information when individuals receive a positive impression of their performance (Burks, Carpenter, Goette, & Rustichini, 2012). Also, subjects that have the opportunity to deceive others are less responsive to new information (Schwardmann & Van der Weele, 2016). Furthermore, when people obtain new information, Bayesian updating of confidence is commonly rejected (Mobius, Niederle, Niehaus, & Rosenblat, 2011). Research in self-serving bias also supports that individuals pay more attention to positive signals (Babcock & Loewenstein, 1997) (Shepperd, Malone, & Sweeny, 2008).
Theories of gender differences on behavior
Since we already explained theories and literature that have researched the strategic characteristics of overconfidence, it is necessary to explain main theories about gender differences in human behaviors.
There are two main explanations about gender differences in human behavior: evolutionary psychology theories and social structure theories. Evolutionary psychology theories argue that reproductive differences between sexes and primitive economies are the key features that created dissimilarities in adaptive behaviors between genders. Due to the fact that men competed against other men for reproduction, characteristics such as violence, competition or risk taking are favored; women have developed behaviors which demonstrate a more nurturing and caring approach due to the benefits which come from long term relationships (Buss, 1995). On the other hand, social structure theories argue that behavioral differences arise from the sexual division of labor. Men's accommodation to roles with greater strength and power produces more dominant behaviors compared to women. Also, these differences are preserved over time by a balance of activities associated with typical roles and the individual convenience of imitating social structure (Eagly & Wood, 1999) (Archer, 1996).
Both theories explain sex differences clearly and offer a channel that maintains these differences over time. Nevertheless, we will focus on a theoretical framework based on social structure theory instead of an evolutionary psychology approach. This is because evolutionary psychology theory focuses on reproduction, and the path of adaptation is commonly slower than in social structure theories, who can explain sex-typed behavior when people face different environments or situations. In other words, social changes analyzed in this paper emerge, not from individuals' tendencies to maximize their inclusive fitness, but instead for their effort to maximize their personal benefits and minimize their personal costs in their social settings (Eagly & Wood, 1999).
In social structure theories, differentiating roles are mediated by many processes. Two of the main processes are explained by the Role Congruity Theory, developed by Eagly and Karau (2005). In this theory subjects, when manifesting a behavior, are evaluated by "social evaluators" or "social perceivers" who hold a stereotypes about a social group (i.e. men and women.) When "social evaluators" perceive the behavior of the subject, two ways of discrimination could emerge: a) the behavior of the evaluated subject could be incongruent with the requirements of the task but congruent with the gender role (prescriptive stereotype) or b) the behavior of the evaluated subject could be congruent with the requirements of the task
6
but incongruent with the gender role (prescriptive stereotype)34 (Eagly & Karau, 2002) (Reuben, Rey-Biel, Sapienza, & Zingales, 2012).
Considering this, the Role Congruity Theory predicts that women will present less confident attitudes than men because it is a masculine characteristic (i.e. agentic characteristic.) Therefore, women are not expected to be confident (descriptive stereotype), and confident women are punished because it is incongruent with gender role (injunctive stereotype).
Predictions of the Role Congruity Theory are also consistent with experiments which find that men inflate their self-images more than women when there are incentives to repeal people from a tournament (Charness, Rustichini, & Van de Ven, 2013); and women are more reluctant than men to initiate negotiations (Bowles, Riley, & Lai, 2007).
Vicki S. Helgeson (2009) summarizes some findings related with achievement in a complete book about the psychology of gender. It says that higher social costs for women are not only in overconfidence, but in public success in general. Women avoid appearing confident and promoting their abilities, believing that others' self-esteem would be threatened by such self-promotion. Furthermore, gender differences in self-confidence seem to be limited to masculine tasks, or tasks with underrepresentation of women. Finally, Vicki S. Helgeson (2009) explains that women are more responsive to feedback than men; partially because women view public information as more informative of their abilities, and partially because men protect their self-esteem. This is caused because men base their sense of self on independence, and women base it on interdependence.
Contrary to our predictions, one experiment shows that men tend to reduce their confidence level more than women when they are in a strategic setting with monetary incentives, compared with a situation without monetary incentives. Nevertheless, this behavior would be driven by non-monetary image concerns (Thoma, 2016). Also, it would be possible that overconfidence is useful to deceive others, and there were no chances to deceive employers in the experiment of Thoma (2016). In this paper, we will analyze gender differences in strategic overconfidence when deceiving employers is possible (Schwardmann & Van der Weele, 2016) (Mobius & Rosenblat, 2006).
III. Data
I used data from two experiments that simulate hypothetical labor interviews. Data comes from two papers: "Deception and Self-deception" (Schwardmann & Van der Weele, 2016), and "Why beauty matters" (Mobius & Rosenblat, 2006). In both experiments, applicants to a job estimate their performance in a certain task (reporting their confidence level). Then, applicants are in a labor interview, and they have the opportunity and incentive to deceive employers about their real performance. In both experiments, employers have to estimate the performance of applicants without knowing their confidence levels nor their
3
Eagly and Karau (2005) focus on the incongruity between female gender role and leadership tasks. Nevertheless, it could be extrapolated to any task that is more or less congruent with one gender role instead of other.
4
Descriptive stereotypes (descriptive norm): expectation about what members actually do. There is evidence that people expect “communal characteristics” (affection, helpful, kind, sympathetic, etc) in women and “agentic characteristics” (aggression, ambition, independence, self-confidence) in men. Prescriptive stereotype (Injunctive norm): expectation about what a group should do. There is evidence that people approve “communal characteristics” in women and “agentic characteristics” in men.
7
performances. These experiments simulate labor interviews where applicants for a job have incentives to show more abilities than other applicants in order to increase their payoff.
These experiments have different settings, and some variables differ which makes results and empirical methods not comparable. Also, some hypotheses can be tested with both experiments, but other hypotheses can only be tested with one of the two experiments because they don't collect the same data. Nevertheless, both experiments contain the main dependent variables that are of interest, like the evaluation of employers and the confidence level of applicants (only "Deception and Self-deception" register the self-promotion of applicants during the interviews.) Also, both experiments collect exogenous variables which are relevant for my hypotheses, like feedback about their performance, different strategic treatments and the gender of subjects. Due to these reasons, both experiments are explained and analyzed separately.
"Deception and Self-deception" and "Why beauty matters" are not thought to test our hypotheses, but these experiments make it possible to test hypotheses related with gender role in strategic environments. Using public data from previous research has the advantage of being cheaper than running a new experiment. The disadvantage is that researchers are blinded about what was going on in the application of the experiments5(unless the previous experiment is run for the current researchers, which is not this case.) From now, we will refer to "Deception and Self-deception" and "Why beauty matters" as Experiment 1 and Experiment 2 respectively. I would like to thanks in advance the researchers of "Deception and Self-deception" and "Why beauty matters" for allowing me to obtain and use their data.
IV. Hypotheses
I would like to find gender characteristics that affect applicant´s and employer´s dependent variables in the context of a labor interview. These dependent variables are: employer's evaluation about the applicant performance, the self-promotion of the applicant and confidence level of subjects.
The evaluation is the belief of the employer about the performance of the applicant when the interview is finished.
The self-promotion (or "message") is the belief that people state in public (a job interview), where applicants have incentives to promote themselves.
The confidence level is measured with a variable that indicates the private belief about their own performance in a certain task.
In this section, I present the main four hypotheses that I intend to test with the two experiments. As both experiments have different settings, not all of the hypothesis can be tested with both experiments. I will also explain which experiment is able to test each hypothesis.
Hypothesis 1
If differences in confidence and self-promotion of applicants have strategic origins; the asymmetry in confidence and promotion between sexes is justified only if the confidence level and the self-promotion showed by the applicant are weighted differently by the employer's evaluation depending on the
5
Thanks in advance to the researchers of "Deception and Self-deception" and "Why beauty matters" for allowing me to obtain their data (or as above).
8
gender of the applicant. Also, if we expect that self-promotion is a behavior associated with masculine characteristics suggested by gender role theory, women that report high self-promotion would be seen as less honest and less sympathetic. I interpret this result as evidence of discrimination through gender role.
H1: Employers evaluate differently the message if it comes from a man or a woman.
This hypothesis can be directly tested only with data from Experiment 1, because Experiment 2 does not have any data on self-promotion. Nevertheless, we can evaluate this hypothesis indirectly with Experiment 2, measuring how private confidence affects the evaluation of each gender.
Another desired feature of Experiment 1 is that, in this experiment, employers did not only
estimate applicants performances in the task, but they also evaluated other characteristics of applicants, like honesty, confidence, attractiveness and sympathy. It allows us to test whether gender role affects not only the employer's evaluation of the applicant in the task, but also employer's social evaluations.
Hypothesis 2
Another variable of interest is self-promotion. If, as hypothesis 1 says, self-promotion is evaluated differently between genders; therefore, there would be differences in the strategy used by each applicant depending on their gender. Nevertheless, the message will not only depend on their own characteristics but also the characteristics of the employer (who interviews the applicant). The characteristics of the employer that are public information (e.g. the gender) could be used to estimate how employers are going to perceive the credibility of a message. Therefore, sending different messages to male and female employers may be strategically convenient.
Also, if applicants of one gender behave more strategically than the other gender, both genders are going to promote themselves asymmetrically in front of a man or a woman.
H2: Male and female applicants promote themselves differently if the employer is a man or a woman.
This hypothesis can only be tested with Experiment 1, because Experiment 2 does not have any data on self-promotion during the job interview.
Hypothesis 3
In both experiments, people do not know their performance in the test, but they receive feedback in order to make a better prediction of their performance.
If subjects under strategic situations are interested in maintaining a high self-image, they will be less sensitive to new information about their real performance. Furthermore, findings on gender role suggest that self-image of women is based on interdependence and being more responsive to external feedback; whereas men base their self-esteem on independence. Then, if men want to continue deceiving themselves (in order to deceive employers) they will update their beliefs more conservatively than women when they receive new and useful information to estimate their real performance.
H3a: Valuable feedback to estimate their real performance affects more the confidence level in women than men.
9
This Hypothesis can be tested with data from Experiment 1 and Experiment 2, because both experiments have data on feedback and confidence. The difference between both experiments is that Experiment 1 has information about a) the feedback received and b) the confidence level of the subjects before and after receiving the feedback. Experiment 2 has data on a) the feedback and b) beliefs of participants only after receiving the feedback.
It is possible that gender differences on reluctance to update beliefs emerge only when people have the possibility to deceive others. That is, people resist updating their believes when the possibility of deceiving others is known.
H3b: Valuable feedback to estimate their real performance affects more the confidence level in women than men. And this effect is higher on strategic situations where subjects have the possibility of deceiving others.
This hypothesis can be tested only with Experiment 1 because in this experiment both applicants and employers report their confidence level. In this case, applicants have the opportunity to deceive others, but employers do not have that chance. It makes employers a control group where subjects are not under a strategic situation.
Hypothesis 4
Finally, theories on gender role say that confidence is perceived as a male characteristic (stereotype). This makes overconfidence less convenient for women than men when the employer perceives high levels of confidence. Therefore, women adjust their confidence less than men in a strategic interaction.
H4: Men and women adjust their confidence level differently if they are in a strategic situation (men adjust confidence more than women).
This hypothesis is only testable with "Deception and Self-deception." In this database, both applicants and employers report their level of confidence. When employers and applicants have to reveal their private beliefs, the only difference between them is that applicants know that they will have the opportunity to deceive employers. Therefore, applicants know that they are in a strategic situation.
V. Analysis of Experiments
In this section, I present each experiment in detail, focusing on the variables used to measure my hypotheses, how experiments proceeded, and the incentives that people have to bring reliable answers. In the analysis of each experiment, after the description of experiments, I explain the empirical strategy used to evaluate each hypothesis, the preliminary results and the regression analysis.
V.1 Experiment 1
Description
Authors of "Deception and Self-deception" (Experiment 1) run the experiment with students from München on 2015. There were 18 sessions, with 16 subjects each. The 16 subjects were divided into two groups of 8 participants, and, in each group, 4 subjects were assigned to the role of applicants and 4 to the role of employers.
10
The experiment is divided in two parts, the self-deception stage and the deception stage. At the beginning of the self-deception stage, subjects completed an intelligence task. Then, they were divided into applicants and employers. Applicants were informed that they were going to receive 2 euros if the employer ranks them as the top 2 (out of four) of the applicants. Employers were not informed about their future role as evaluators. This means that applicants knew that they were in a strategic situation where they have the chance to deceive others in the future, but employers did not know it.
After subjects received information about the strategic setting, both applicants and employers submitted their beliefs under a variation of the Becker-DeGoot-Marshak mechanism. In this method, participants choose a probability that it is interpreted as the probability of being in the top 2 (out of four); I will call this variable "prior belief". They were paid according to how accurately they estimate their beliefs6.
Then, participants received information (feedback) about their performance. Participants that were in the top 2 (out of 4) drew a ball from an urn with 15 black balls and 5 red balls; participants in the bottom 2 drew a ball from an urn with 5 black balls and 15 red balls. It makes the feedback noisy but unbiased. After receiving the feedback, participants update their private believes again, this variable is going to be called "posterior belief," and this is the end of the self-deception stage.
After the self-deception stage, participants acquire the role of applicant or employer. Employers receive a sheet where they had to evaluate the performance, honesty, confidence, sympathy and attractiveness of applicants. Then, employers and applicants simulated a job interview. During the interview, the applicant could only send one message indicating the probability of being in the top 2 of the group. This message is our measure of self-promotion. Then, the employers proceed to evaluate the applicant in the 5 dimensions previously explained (performance, honesty, confidence, sympathy and attractiveness.)
In each group of each session, the 4 employers interviewed the 4 applicants. Considering that the experiment consisted on 18 sessions with 16 participants, there are 288 participants; 144 applicants and 144 employers. Each applicant and employer was in 4 interviews, that is, the dataset has 576 interviews.
It is important to say that in between the self-deception and the deception stage, groups were divided into 3 treatments. A "Tutorial and Warning" (TW) treatment, where employers receive a lie-detection tutorial, and applicants were warned about it; a "Tutorial and Warning" (TnW) tutorial, where employers receive a lie-detection tutorial, and applicants were not warned about it; and a "No Tutorial" (NT) treatment. The lie-detection tutorial was a simple computer screen with four tips of detecting lies. I suggest that this treatment does not necessarily have to be interpreted as a treatment with employers with higher abilities on lie-detection, but also as a treatment where employers were previously warned about the possibility of facing lying applicants. These conditions are not the focus of this research; nevertheless, it is important to control for these variables.
11
Empirical Strategy of Experiment 1
As I described in section IV, we are going to test gender differences in the determinants of three variables; the evaluation of the employer, the message that the applicant sent to the employer and the confidence level of the subjects.
Evaluation
When employers evaluate applicants during Experiment 1, they have to evaluate the performance on the task, the honesty, the confidence, the sympathy and the attractiveness of applicants. In our model, evaluations depend on the information available to the employer, and their own characteristics. That is, the employer receives the message sent by the applicants, the characteristics of the person who sent the message (e.g. the gender, attractiveness, etc.), but not the private information of the applicant (private beliefs.) With that information and her own characteristics, the employer is going to evaluate the applicant in several dimensions. The empirical strategy is then:
where is the evaluation of the employer to the applicant in the dimension k=(performance, honesty, confidence, attractiveness, sympathy, etc.) the evaluation of the performance is the main evaluations of our concern, but we are also interested in testing whether gender differences emerge in other kinds of evaluations. We are going to understand employer's evaluation of applicant's honesty, confidence, attractiveness and sympathy as "social evaluations" that do not have a direct monetary cost for employers and applicants. I included an interaction between the gender of the applicant and the message. This interaction indicates whether employers evaluate differently a high message (self-promotion) if it comes from a male or a female applicant. The model considers a fixed effect within employer in order to control for omitted variables that can affect the outcome (e.g. skepticism or exigency of the employer.)
I assume that employers do not consider the internal beliefs of the applicants to evaluate them. That effect is captured through the characteristics of the applicant that are public information, such as gender, attractiveness and the message. Nevertheless, I include characteristics of the applicant that are not known for the employer in some regressions, but I interpret these variables as proxies of omitted variables that can affect the evaluation. For example, the conviction (voice, facial expression, etc.) with which an applicant sends a message can be captured by variables as the confidence level of the participant.
All the parameters of equation (1) are of interest, but most important is , which tests hypothesis 1 (Employers evaluate differently the message if it comes from a man or a woman, also self-promotion in women is evaluated worse in sympathy or honesty, but men are evaluated better in confidence because of gender roles).
Another empirical model that I want to test assumes that employers use the evaluation of honesty, confidence, sympathy and attractiveness as input to evaluate the performance in the task. Therefore, the model would be:
12
This model considers a fixed effect by the employer, due to the possibility of omitted variables that affect the final evaluation and are correlated with the evaluation of honesty, confidence, sympathy and attractiveness. The model is run through a seemingly unrelated regression, due to the social evaluations are outcome and explanatory variables at the same time which would result in the error terms being correlated7. In this case, the parameters of interest are: , that tests the direct effect of the message in the evaluation, controlled by others channels, like the perception of honesty, confidence, sympathy and attractiveness; together with , which considers the gender discrimination in the message due to the asymmetrical perception of honesty between men and women; together with , which considers the gender discrimination through the confidence channel; and measure the sympathy channel; and and the attractiveness channel. Other control variables are also included.
Message
Applicants have to promote themselves in order to be well evaluated by employers, but a high self-promotion can also be seen as a lie by the employer. Then the message (self-self-promotion) of the applicant will depend on her confidence level of the applicant, her personal characteristics, but also on the characteristics of the employer that the applicant knows. The message will depend on the characteristics of the employer because those characteristics are part of the public information at the moment of the interview.
Employers with different characteristics can evaluate a message as being more or less truthful. Applicants would react to this, diminishing or raising their self-promotion.
where is the message sent from applicant to employer , and is the gender of the employer. We include an interaction between the gender of the employer and applicant to test whether different sexes change their strategies asymmetrically in front of someone of the same or other sex. An interaction between the confidence level and the gender is included to test whether applicants of different sexes actually use the confidence level with different intensities when they send a message. To correct for omitted variables that affect the applicants' message, I should include a fixed effect. However, this model does not include a fixed effect because the interaction between genders and fixed effect are linearly dependent.
I assume that applicants do not consider directly the internal beliefs of the employers when they send the message. They only estimate the right message according to the exogenous characteristics of the
7
13
employer, which are public information. Nevertheless, I include private characteristics of the employer as control variables, but I interpret them as proxies of public characteristics that are not measured.
Our parameters of interest are , , and . But more specifically , which tests hypothesis 2 (Male and female applicants promote themselves differently if the employer is a man or a woman).
Confidence
At job interviews, applicants have an idea about their own productivity, and it will be higher or lower depending on their level of confidence. But the level of confidence could also be affected by other incentives, like the intention to convince the employer that he is a highly qualified applicant. Then, the level of confidence will be determined by the strategic situation that applicants are going to face and their own characteristics.
where are the characteristics of the subject, is the sex, indicates whether participants are in a context where they have incentives to rise their private confidence level (i.e. whether they are in the role of applicants or employers,) and is a variable that indicates the information that the participant has in order to make a more accurate estimation of her performance.
I allowed for an interaction between and because we want to test whether men and women update their level of confidence differently depending on the strategic situation. I also included an interaction between and to test if men and women adjust their beliefs differently depending on the new information. Finally, I included an interaction between and to test whether the gender differences in beliefs adjustment after receiving information emerges from strategic situations.
Parameters of interest are , and , which test hypothesis 4 (Men and women adjust their confidence level differently if they are in a strategic situation), hypothesis 3a (Men and women update their confidence level differently if they receive valuable feedback) and hypothesis 3b (Men and women update their confidence level differently if they receive valuable feedback when they have the opportunity to deceive others), respectively.
Also, Experiment 1 has data of the confidence level before and after the feedback. It makes possible to test this hypothesis, not in the level of confidence, but in the variation of confidence within a person. Therefore, the model would be:
Finally, I also estimated a Bayesian regression model proposed by Möbius et al. (2011) and replicated in the paper of Schwardmann and Van der Weele (2016), but including interactions between variables that indicate whether the participant is a man or woman in a strategic situation or not.
Results of Experiment 1
Preliminary results on Evaluations
Before running the regressions previously explained, some descriptive graphs would help us to understand the dynamic of gender role, confidence and self-promotion under this experiment. Figure 1.A shows a
14
scatter plot and a simple linear regression between the employer's evaluation of the applicant's performance and the score obtained by the applicant. This figure indicates that women with low scores are evaluated better than men, but the difference is reversed in high scores. As a result, the relation between the applicant's score and the employer's evaluation is higher for men than for women. Nevertheless, this asymmetry does not emerge because of different evaluation of the applicant's self-promotion. Figure 1.B shows that, at the same level of self-promotion, men and women are evaluated equally; showing almost parallel fitted lines.
Figure 1.C suggests that gender differences of Figure 1.A emerge because women that have lower scores present higher levels of self-promotion than men, and this difference is reversed for higher scores. An interesting fact is that women only required a score over nine to state a higher message than men (Figure 1.C,) and 66% of women and 68% of men achieved it; however, they require a score over eighteen to got a better evaluation (Figure 1.A,) and it was achieved only by 19% of women and 24% of men.
Figure 1.D shows that, for this task, confidence level is the same for men and women when we control for the score. These graphs do not give strong support to hypothesis 1; nevertheless, it suggests asymmetries between gender in the relation between employer's evaluation and score.
Figure 1. Relations between Score, Evaluations and Self-confidence. Fig 1.A shows the scatter plot and linear regression between the employer's evaluation of the applicant's performance for men and women. This relation is different for each gender./ Figure 1.B shows the linear relation of evaluation and self-confidence. Hypothesis 1 would predict a steeper curve for men in Fig 1.B, but the graphs show that slopes are similar./ Fig1.C shows gender differences in the relation between self-promotion and the score, similarly to the relation between employer's evaluation and score, but the intersection of both curves is at a higher score./ Fig 1.D shows the relation between self-confidence and the score. It does not show evidence that at the same score men are more confident.
15
Another interesting result is given by the Figure 2. Each graph of the Figure 2 shows the average employer's evaluation of honesty (Fig. 2.A,) confidence (Fig. 2.A,) sympathy (Fig. 2.A,) and attractiveness (Fig. 2.A,) depending on the quartile of applicant's self-promotion by sex8. Figure 2 shows that people who send higher messages are evaluated as less honest, more confident, slightly less attractive and sympathetic. The negative relationship between the evaluation of honesty and self-promotion is more pronounced in women than men (Fig.2.A). The gender difference in the evaluation of honesty is mainly because of the differences in the average evaluation of honesty in quartile 3, composed mostly of people who report self-promotion of 75, 80 and 85.
Figure 2. Relations between social evaluations and quartile of self-promotion. Fig 2.A shows that the higher the self-promotion, the
lower the evaluation. This relation is more clear for women./ Fig 2.B shows a positive relation between self-promotion and the evaluation of confidence./ Fig 2.C shows a slightly negative relation between sympathy and self-promotion. This negative relation is caused mainly because of the high average confidence of quartile two./ Fig 2.D shows a slightly negative relation between attractiveness and self-promotion. Also, average attractiveness is lower for men.
Preliminary results on evaluations suggest that the effect of the self-promotion in the evaluation of the performance is not different between genders (Fig 1.B). Nevertheless, the impact of self-promotion on the evaluation of honesty differs between genders (Fig. 2.A). This is relevant because, as I will show in the regression result, the evaluation of honesty has a significant impact on the evaluation of performance.
8
The horizontal axis is the quartile of self-promotion and not the level of self-promotion, because most of the people report high levels of self-promotion. Making a graph by quartile ensure almost the same amount of data by quartile.
16
Preliminary results on Message
Hypothesis 2 tests whether male and female applicants promote themselves differently when they face a male or female employer. Figure 3 shows the average message (self-promotion) of male and female applicants depending on the gender of the employer. It also shows the distribution of the interviews by the gender of applicants and employers9. The bar graph indicates that the average self-promotion of men, when they face female employers, is the lowest. The difference in the average self-promotion is statistically different that the average self-promotion of female applicants who face female employers (p-value=0.007), and comparing with male applicants who face male employers (p-value=0.031).
Figure 3. Average self-promotion by gender of applicants and employers. Bar MA-ME is the average self-promotion of male applicants
when employers are men./ Bar FA-ME is the average self-promotion of female applicants when employers are men./ Bar MA-FE is the average self-promotion of male applicants when the employer is a woman, and it is the only statistically significant bar that is lower than the others./ Bar FA-FE is the average self-promotion of female applicants when the employer is a women.
Figure 3 suggests that hypothesis 2 would apply for men: they apparently reduce their level of self-promotion in front of women.
Preliminary results on confidence
Hypotheses 3 and 4 test whether men and women report different levels of confidence when subjects a) receive feedback and b) are under a strategic situation that (potentially) encourages them to behave more confidently, respectively. Figure 4 shows the confidence level of men and women before and after receiving feedback. This figure is divided into four graphs which differ in the strategic situation (employers or applicants) and two types of possible feedback (positive or negative). In this dataset, employers are those who do not have incentives to deceive others.
As we would expect, the confidence level after the feedback is significantly lower than the prior confidence average for those who received negative feedback, and significantly higher for those who
9
53% of male applicant's interviews were with a male employer, and 53,7% of the female applicant's interviews were with a female employer, suggesting that applicant's gender is not randomly distributed between employer's gender, but these differences are not significant. It is because some sessions have more men, or more women; rising the probability of matching with someone of the same gender.
17
receive positive feedback10. As a result, those who received negative information do not have a statistically significant lower level of confidence than those who receive positive feedback before feedback; but their confidence level differs after receiving feedback, this applies for both genders.
Furthermore, after receiving feedback, the only average confidence level which is higher for women than men is in the group with incentives to deceive (applicants) and received negative feedback (Fig. 4.C). In all the other groups, average confidence is higher for men. Nevertheless, all the differences are not statistically significant. When we compare the confidence variation (posterior minus prior belief), the reduction of confidence in women who received negative feedback is higher for employers than applicants (p-value=0.071.)
Figure 4. Average confidence before (Prior) and after (Post) feedback for men (M) and women (F). Figure 4.A and 4.B shows the
average prior and post information confidence level by gender for employers; Fig. 4.A shows those who receive a negative feedback and Fig. 4.B those with positive feedback. Figure 4.C and 4.D shows the average prior and post information confidence level by gender for applicants; Fig. 4.C shows those who receive a negative feedback and Fig. 4.D those with positive feedback.
One cause of this non-significant result may be the lack of data on each group. Making a more general analysis, the average confidence level after receiving feedback is statistically higher for men than women, in those people who receive positive feedback (p-value=0.089) -including both employers and applicants. Furthermore, men have more confidence than women after feedback in those who are employers (p-value=0.034) -without discerning between the type of feedback. This suggests that higher
10
These data exclude those who upload their believes in the wrong direction as well as those who didn't update their confidence level. As an important consideration for our hypothesis, the percentage of men that didn't update believe is higher than women. Also, the women that update in the wrong direction are mostly women who receive a positive feedback and reduce their confidence. On the other hand, the amount of men that update in the wrong direction are equal in those who receive positive and negative feedback.
18
posterior confidence in men compared to women would not be motivated by strategic reasons. We also have to consider that most of women who were employers received a negative feedback (65%); therefore, a regression analysis is needed.
Results of the Regression Analysis
Results on Evaluations
Running a regression of the employer's evaluation of the applicant performance on the self-promotion, gender and an interaction variable allows us to test directly hypothesis 1. Additionally, a regression of these explanatory variables on the employer's evaluation of honesty, confidence, attractiveness and sympathy allows us to test if self-promotion has different social costs for men and women. Table 1 shows the effect of these explanatory variables in each of the dependent variables; Table 2 controls for other variables.
In our model, we assumed that the employers cannot use the private information of the applicant to evaluate them (due to fact that they are not available as information for employers); however, Table 2 includes some of those variables. Given our assumptions, these variables are interpreted as proxies of possible omitted variables that are public information between the applicant and the employer but don't exist in the dataset, like the voice tone or body expressions. Nevertheless, regressions of Table 2 have the drawback of including variables that affect the evaluation indirectly through the message. Given that the message is a variable that depends on these characteristics, it can lead to inconsistent results. For these reasons, I would use Table 1 as the main results, and Table 2 gives us complementary information about potential omissions. Other variables that would affect the employer's evaluations are the employer's characteristics, which is why I use a fixed effect regression by employers.
Regressions of Table 1 confirm preliminary results. As we expected, self-promotion has a statistically significant effect on employer's evaluation of the performance, honesty and confidence. However, the effect of self-promotion on the performance is not different between men and women (p-value=0.986)11. An interesting result is provided by the second column: the interaction variable between gender and self-promotion has a significant effect at 90% confidence level on the evaluation of honesty (p-value=0.088). It means that, at high levels of self-promotion, the perception of honesty is more affected when applicants are women compared to men. These results suggest that high levels of self-promotion have higher social cost for women than men, as some literature points out. The magnitude of the effect is not considerable for men, but more than twice for women. The evaluation of honesty is measured from 1 to 5 and the self-promotion from 0 to 100. Then, increasing 10 points (out of 100) of self-promotion leads to a reduction of 0.09 (out of 5) points for men and 0.2 for women. Gender differences in self-promotion are not significant any more when I controlled for more variables.
Furthermore, if the evaluation of honesty has an effect on the evaluation of performance; then, self-promotion would have a different indirect effect on the final performance for men and women. To confirm this hypothesis, it is necessary to test equation 2, which is a system of equations that considers the effect of the evaluation of honesty, confidence, attractiveness and sympathy on the evaluation of performance.
11
After evaluation, employers decided which of the applicants was going to be hired. I run a probit regression that provide same results that the equation of the first column: self-promotion rise the probability of being hired, but there is not differences between genders.
19
TABLE 1- IMPACT OF SELF-PROMOTION IN EACH GENDER ON EMPLOYER'S EVALUATIONS EMPLOYER'S EVALUATION ON:
VARIABLES PERFORMANCE HONESTY CONFIDENCE ATTRACT. SYMPATHY
SELF-PROMOTION 0.615*** -0.009* 0.026*** 0.001 -0.002
(0.000) (0.055) (0.000) (0.886) (0.640)
FEMALE -2.001 0.800 0.242 0.654 0.434
(0.862) (0.123) (0.643) (0.183) (0.300)
FEMALE EMPLOYER*MALE APP. -1.110 -0.095 0.259 0.270 0.027
(0.799) (0.628) (0.191) (0.146) (0.864) FEMALE*SELF-PROMOTION -0.003 -0.011* -0.003 -0.002 -0.003 (0.986) (0.088) (0.687) (0.713) (0.585) CONSTANT 5.316 4.127*** 1.446*** 2.633*** 3.829*** (0.544) (0.000) (0.000) (0.000) (0.000) N 576 576 576 576 576 R2_W 0.131 0.049 0.102 0.038 0.020 R2_O 0.089 0.026 0.077 0.033 0.015 p-values in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01
TABLE 2- IMPACT OF SELF-PROMOTION IN EACH GENDER ON EMPLOYER'S EVALUATIONS EMPLOYER'S EVALUATION ON:
VARIABLES PERFORMANCE HONESTY CONFIDENCE ATTRACT. SYMPATHY
SELF-PROMOTION 0.527*** -0.010* 0.023*** -0.001 -0.000
(0.000) (0.082) (0.000) (0.811) (0.947)
FEMALE -18.054 0.416 0.282 0.637 0.405
(0.136) (0.458) (0.604) (0.199) (0.350)
FEMALE EMPLOYER*MALE APP. -3.080 -0.176 0.181 0.217 0.020
(0.490) (0.393) (0.367) (0.237) (0.899) FEMALE*SELF-PROMOTION 0.199 -0.008 -0.003 -0.001 -0.000 (0.225) (0.272) (0.690) (0.894) (0.943) VALUEHONESTY 14.545* 0.050 0.445 -0.267 -0.235 (0.012) (0.850) (0.086) (0.257) (0.253) DOMINANCE -5.120 -0.139 0.236 0.156 -0.127 (0.385) (0.610) (0.374) (0.518) (0.548) CONFIDENCE -0.019 0.002 0.001 0.001 -0.001 (0.789) (0.505) (0.860) (0.840) (0.833) SCORE 0.385 -0.004 -0.000 0.007 0.004 (0.125) (0.750) (0.991) (0.475) (0.645) FEMALE*CONFIDENCE -0.019 -0.004 -0.004 -0.003 -0.003 (0.850) (0.416) (0.399) (0.495) (0.359) FEMALE*SCORE -0.085 0.024* 0.010 -0.008 -0.002 (0.781) (0.091) (0.461) (0.523) (0.821)
AVERAGE HONESTY EVAL. 1.557 0.046 -0.097 -0.025 0.083
(0.417) (0.608) (0.263) (0.753) (0.224)
AVERAGE CONFIDENCE EVAL. -1.268 -0.069 0.251** 0.064 -0.049
(0.469) (0.396) (0.002) (0.376) (0.430)
AVERAGE ATTRACT. EVAL. 3.631* -0.001 0.177** 0.461*** 0.171**
(0.070) (0.994) (0.049) (0.000) (0.017)
AVERAGE SYMPHATY EVAL. 3.258 0.185 -0.032 0.060 0.128
(0.225) (0.137) (0.794) (0.588) (0.185) CONSTANT -20.173 3.559*** 0.344 1.088 2.844*** (0.165) (0.000) (0.599) (0.068) (0.000) N 544 544 544 544 544 R2_W 0.178 0.083 0.187 0.186 0.091 R2_O 0.115 0.052 0.148 0.171 0.060
Impact of social evaluations and self-promotion on the evaluation of task performance
Equations 2.a to 2.e hypothesize that the evaluation of honesty, confidence, attractiveness and sympathy impact the final evaluation of the task performance. To examine this hypothesis, I run a seemingly unrelated regression considering variables of equations 2.a to 2.e. It is worth remarking that control variables are still considered proxies of omitted variables. Regressions that include these variables have to
20
be analyzed with caution, because they would be explained by our variables of interest, like the gender of the applicant.
Due to the extent of the results, results of the system of equations are in Table A.2 of the appendix. These regressions include fixed effect by employer, given the inner characteristics of each employer that would affect their evaluations. Results of the first column show that the evaluation of the task performance is significantly and positively affected by the self-promotion value=0.000), the evaluation of honesty (p-value=0.000) and the evaluation of confidence (p-value=0.004). The evaluation of sympathy, being female and the value of honesty become significant at 95% in column 2 and 312.
The magnitude of the impact of the evaluation of honesty and confidence on the evaluation of the task remains almost the same in the three columns. However, the impact of the evaluation of honesty is clearly higher. The evaluation of honesty and confidence are from 1 to 5, and the evaluation of the task is from 0 to 100. Therefore, considering the first column, increasing 1 point the evaluation of honesty increases the evaluation of the task performance by 11,7 points, but increasing the evaluation of confidence by one point increases the evaluation of the task performance by only 2,6 points.
Considering the predictions, self-promotion has a higher impact on women, but this effect is not statistically significant in the first column (p-value=0.171). Nevertheless, this effect starts to became significant at 95% confidence when I control for more variables (column 2 and 3, value=0.024 and p-value=0.009 respectively). Furthermore, together with this effect, being female has a negative and significant impact in these two columns (p-value=0.012 and p-value=0.005 respectively). It means that self-promotion would have a higher impact on employers evaluation for women than men; but, at the starting point (self-promotion=zero) women are evaluated worse than men. This evidence doesn't support hypothesis one.
Regarding the employer's evaluation of social characteristics (i.e. honesty, confidence, attractiveness and sympathy), table A.2 (appendix 2) shows that self-promotion has a statistically significant positive impact on the evaluation of confidence and negative on the evaluation of honesty. Furthermore, the negative effect of self-promotion in honesty is statistically lower for women at the 95% confidence level (p-value=0.047).
To illustrate the mechanism of the effect of self-promotion on the employer's evaluation on the task performance, we will consider an exogenous rise of self-promotion by 10 points. Using only the statistically significant results of column 1, the effect for men and women are the following: men would have a positive direct effect (0.669*10 points), minus the effect on honesty (-11.695*0.009*10 points), plus the effect on the evaluation of confidence (2.3*0.026*10 points), which brings a total effect of 6.24. On the other hand, women have the same net effect (6.24) minus the higher punishment in honesty evaluation (-11.695*0.011*10 points), having a total effect of 4.95.
If we make this exercise considering the second column, we also have to add the positive effect of the interaction variable between female and self-promotion, subtract the effect of being a woman and omit the negative effect of female self-promotion on the evaluation of honesty. According to these parameters, women would have a higher employer's evaluation at a self-promotion higher than 81.3. Considering that
12
21
the median self- promotion of men and women is 75, it should be only a slightly better evaluation for men than women, as in fact is (50,8 for men and 49,6 for women.)
Table A.3 divides these results by male and female employers. Results show that most gender differences remain only with female employers. Analyzing only female employers, the interaction variable between gender and self-promotion has a significant positive direct effect on the evaluation of performance value=0.014), a negative effect on the evaluation of honesty value=0.061), confidence (p-value=0.007) and sympathy (p-value=0.087). On the other hand, for male employers, there is only a positive effect of the interaction variable on confidence (p-value=0.02). Results don't change if I run the regression only using employers with the lie detection tutorial, or in applicants not warned about the lie detection tutorial.
With all of this, results indicate that: there are gender differences in favor of men because women who send high messages are seen as less honest than men. However, including more variables makes these effects disappear, and the positive effect of self-promotion on the employer's evaluation is higher in women; also, this effect is crowding out because of the effect of being a woman. Also, the gender differences on social costs because of self-promotion are significant with female but not male employers. This is, hypothesis one only holds for the evaluation of honesty, but not for the direct effect of promotion on the evaluation of the performance, where some regressions indicate that female self-promotion is actually more weighted, but from a lower starting point.
Gender differences on self-promotion
Hypothesis 2 states that applicants of different genders will state different levels of self-promotion, and it will change depending on the gender of the employers. In this case, I assumed that applicants state a self-promotion level depending on the public characteristics of the employer (e.g. gender) and their own characteristics, which are not all available in the dataset. I also used other private characteristics of the employer as proxies of public characteristics that are not available.
To test this hypothesis, I ran two types of regressions on self-promotion. The first one does not include fixed effect. It has the advantage of capturing the effect of the applicant's confidence on the self-promotion, and whether this effect is different for men and women. The disadvantage is that applicants could have some unmeasured characteristics that could affect the self-promotion. If these characteristics are correlated with the gender of the applicant, the parameter is biased. The second regression includes fixed effect by applicants, it has the advantage of controlling for omitted variables, but it only captures the effect of gender interaction within an applicant, and not between applicants.
Results of the regression without fixed effect are shown in Table 3. As the preliminary results suggested, regressions found that male applicants reduce their message in front of female employers. Also, as would be expected, the level of confidence affects positively the self-promotion, without any differences between genders.
The effects of the gender interaction disappear when average evaluation of social characteristics is included (column 7.) The average evaluation of honesty and sympathy become significant. The average evaluation is the average employer's evaluation of the social characteristic of the applicant, excluding the employer of the specific interaction. Then, it is the evaluation of honesty and sympathy that other employers gave to the applicant. A possible explanation is that average honesty and sympathy becomes
22
significant because of reverse causation; this means that people who send higher messages have worse evaluation of honesty and sympathy, not only for the employer, but also for other employers. It is highly likely because most of the variation in self-promotion is between participants, not within participants; then, it is a consistency between the self-promotion stated by one applicant to different employers13.
TABLE 3 - IMPACT OF GENDER INTERACTION ON APPLICANT'S SELF-PROMOTION APPLICANT'S SELF-PROMOTION
VARIABLES (1) (2) (3) (4) (5) (6) (7)
MALE EMP.*FEMALE APP. -0.681 -0.751 -0.388 -0.652 -0.606 -0.482 2.979
(0.873) (0.862) (0.927) (0.878) (0.887) (0.910) (0.458)
FEMALE EMP.*MALE APP. -4.919* -4.958* -5.082* -4.877* -4.876* -4.783* -3.158
(0.042) (0.041) (0.033) (0.044) (0.044) (0.046) (0.177)
FEMALE EMP.*FEMALE APP. -1.418 -1.472 -1.280 -1.350 -1.327 -1.232 3.107
(0.740) (0.731) (0.762) (0.750) (0.755) (0.772) (0.435) CONFIDENCE 0.224*** 0.224*** 0.201*** 0.185*** 0.185*** 0.186*** 0.150** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.002) FEMALE*CONFIDENCE 0.008 0.008 0.004 0.005 0.005 0.004 -0.012 (0.891) (0.888) (0.944) (0.929) (0.928) (0.952) (0.832) EMPOYER'S CONFIDENCE 0.010 0.010 0.007 0.008 0.007 0.007 0.007 (0.692) (0.701) (0.791) (0.768) (0.781) (0.777) (0.783) EMPLOYER SCORE -0.030 -0.031 -0.037 -0.034 -0.033 -0.025 0.042 (0.703) (0.697) (0.637) (0.670) (0.680) (0.753) (0.571)
EMPLOYER'S VALUE HONESTY 1.307 1.269 1.048 0.965 0.977 0.897 2.855
(0.695) (0.704) (0.752) (0.771) (0.768) (0.787) (0.334) EMPLOYER'S DOMINANCE 1.105 1.035 1.045 1.076 1.129 -1.002 (0.745) (0.763) (0.761) (0.754) (0.744) (0.750) SCORE 0.207* 0.187* 0.192* 0.196* 0.161 (0.021) (0.044) (0.040) (0.037) (0.067) BLACKBALL 1.600 1.588 1.775 2.300 (0.291) (0.294) (0.243) (0.101) VALUE HONESTY -0.979 -1.026 -4.146 (0.702) (0.689) (0.138) DOMINANCE -3.629 -5.690 (0.237) (0.056) HONESTY AVERAGE -3.827*** (0.000) CONFIDENCE AVERAGE 6.744*** (0.000) ATTRACTIV. AVERAGE -1.798 (0.053) SYMPHATY AVERAGE -2.787* (0.035) N 544 544 544 544 544 544 544 R2 0.159 0.159 0.170 0.172 0.172 0.174 0.305 p-values in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01
The results of the fixed effect regression by applicants are presented in the Table A.4 (appendix 4.) It shows that the effect of the gender interaction disappears. The fixed effect model also shows a low R-squared within sample. This means that most of the explanation of the gender interaction is caused because of the variation between applicants. Another way to see it is looking at Table A.5 (appendix 5) where I run a regression with the average self-promotion of the applicant as the dependent variable and the percentage of female employers faced by each applicant as an explanatory variable. I also included an interaction between the gender of the applicant and the percentage of female employers. Results show that applicants
13
42.4% of applicants did not change their self-promotion level during the whole experiment. 66.7% changed it less than 10 points between the maximum and the minimum, and 84% change it equal or less than 10 points between the maximum and minimum level. I keep the average social evaluations between the variables to maintain variables in common with the regressions of the evaluations; however, the effects of these variables have to be interpreted with caution.
23
that face more female employers send lower messages (p-value=0.045), but this effect is crowded out when the applicant is a woman (p-value=0.086). Male applicants send lower messages when they face more female employers.
I interpret this result as a more strategic self-promotion in men than women, because women do not change their level of self promotion depending on the gender of the employer, but men do. This strategic behavior would be motivated by gender roles. The aim of this hypothesis was not to test the precise motivation behind the behaviors of men and women, but only if they react differently in front of employers with different genders. Results indicate that it is true for men, but not for women.
Impact of being in a strategic situation and receiving feedback on self-confidence
Hypotheses 3 and 4 are related to strategic roots of self-confidence. Table 4 provides results of the regression proposed in equation 3. Different columns indicate different groups of participants. The first column indicates the general regression for all the participants in the experiments. The variable “applicant” indicates if the subject is under a strategic situation. Variable "blackball" is one if the participant receives positive feedback and zero if the participant received a negative feedback. Results indicate that the level of confidence after receiving information is determined mostly by the level of confidence that individuals had before the feedback and the quality of the feedback. The interaction variables that were of our interest do not show any statistically significant results. With these data, it is not possible to ensure that men and women have different behaviors under a strategic situation.
TABLE 4- DETERMINANTS OF CONFIDENCE UNDER STRATEGIC SITUATIONS BY GENDER
VARIABLES ALL APPLICANTS EMPLOYERS BLACKBALL REDBALL FEMALE MALE
FEMALE -5.547 3.004 -4.532 -4.518 -5.777 (0.153) (0.505) (0.265) (0.130) (0.171) SCORE 0.065 -0.012 0.189 -0.030 0.185 0.048 0.106 (0.636) (0.943) (0.371) (0.863) (0.353) (0.756) (0.724) DOMINANCE 4.550 -3.353 12.661* 2.675 7.509 0.070 10.659 (0.385) (0.663) (0.077) (0.613) (0.427) (0.991) (0.232) VALUEHONESTY 0.877 -4.280 4.731 6.529** -4.265 -0.688 3.428 (0.812) (0.409) (0.382) (0.038) (0.534) (0.873) (0.609) PRIOR 0.752*** 0.873*** 0.656*** 0.647*** 0.855*** 0.763*** 0.740*** (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) BLACKBALL 35.601*** 37.795*** 36.005*** 37.324*** 35.745*** (0.000) (0.000) (0.000) (0.000) (0.000) APPLICANT -5.704 -3.191 -5.904 2.765 -5.153 (0.256) (0.231) (0.241) (0.408) (0.337) APPLICANT*BLACKBALL 2.077 -2.525 1.438 (0.722) (0.510) (0.817) FEMALE*BLACKBALL 1.763 -2.506 0.761 (0.723) (0.610) (0.882) FEMALE*APPLICANT 8.685 3.178 9.337 (0.145) (0.356) (0.117) FEMALE* APPLICANT*BLACKBALL -4.946 (0.475) CONSTANT -7.303 -12.032* -11.019* 34.318*** -13.443 -9.737* -12.514 (0.178) (0.092) (0.062) (0.000) (0.109) (0.070) (0.131) N 192 89 103 93 99 119 73 R2 0.815 0.843 0.801 0.678 0.582 0.824 0.802 R2_A 0.804 0.830 0.787 0.652 0.550 0.812 0.781 p-values in parentheses * p < 0.1, ** p < 0.05, *** p < 0.01
