Introduction

There appears to be little empirical information about sexual harassment-prone men (O’Leary-Kelly et al. 2000; Pryor 1987; Pryor and Stoller 1994). The present investigation was designed to explore the relationship between men’s social cognition and men’s propensity to sexually harass women. More specifically, we investigated the possible connection between men’s inferences of women’s critical/rejecting thoughts and feelings, men’s sexual harassment behavior, and men’s aggression against women. This study also introduces a new instrument, the Sexual Harassment Behavior Inventory (SHBI), on which men can self-report the frequency of their behaviors that women tend to find sexually harassing.

An estimated 42 to 88% of women experience sexual harassment in the workplace (Ragins and Scandura 1995). This sexual harassment creates substantial stress and fear that is likely to deter women from pursuing opportunities otherwise available to them. Sexual harassment may also reduce work productivity and precipitate expensive litigation (Hamilton et al. 1987; Loy and Stewart 1984; Paludi and Barickman 1991). In addition to occupational concerns, the potentially severe negative effect of sexual harassment on the victim’s psychological and physical health are well documented (Charney and Russell 1994; Rederstorff et al. 2007). These findings suggest a strong need for research into the causes of sexual harassment in general. Unfortunately, however, the primary focus of existing sexual harassment literature tends to focus on the perceptions of female victims (Rotundo et al. 2001). In order to reduce sexual harassment occurrences and the related negative consequences it is important for researchers to investigate the nature and causes of men’s sexual harassment behavior.

Defining Sexual Harassment

Legal definitions of sexual harassment are fairly concrete, but are generally restricted to a constellation of behaviors and consequences that occur in workplaces or schools. On the other hand, the academic literature seems to lack a clear unified definition of sexual harassment (Rotundo et al. 2001). Several authors have suggested that sexual harassment should be broadly defined for scientific purposes (Hand and Sanchez 2000; Johnson et al. 1997; O’Leary-Kelly et al. 2000; Woody and Perry 1993) and that a more scientific definition would include a broader variety of behaviors and contexts beyond workplaces and academic settings. For instance, women report experiencing more sexual harassment from strangers than from non-strangers and that sexual harassment from strangers evokes more fear in women than sexual harassment from men they know (MacMillan et al. 2000). For these reasons, the present investigation assesses men’s sexual harassment behavior in a broadly defined sense and in the wide range of contexts in which it occurs. A new self-report measure, the Sexual Harassment Behavior Inventory (SHBI), was created for this purpose and is described in the Method section.

Inferential Accuracy

Social information processing theory (McFall 1982) posits that accurate interpretation of social cues is requisite for effective social behavior (see also Dodge et al. 1986). People who less accurately infer others’ thoughts and feelings are more likely to behave inappropriately. For example, aggressive children and adults with paranoid personality traits have been found to be deficient in their ability to interpret ambiguous social situations. This social deficit frequently results in inappropriate aggressive responses (Dodge and Crick 1990; Turkat et al. 1990). Following this logic, we expected to find that men’s accuracy in inferring women’s criticism and rejection would be negatively associated with the men’s sexual harassment behavior—Hypothesis 1.

Inferential Bias

Inaccurately inferring women’s criticism and rejection could result from either over- or under-attributing criticism or rejection to women. Our second hypothesis is comprised of three competing predictions (Hypotheses 2a, 2b, & 2c) regarding the presence and direction of any inferential bias that sexual harassers may possess.

One possibility is that harassment prone men may be biased to over-infer women’s criticism or rejection. This would be consistent with recent reports suggesting that sexual harassment is actually a form of hostility. Mild sexual harassment may represent one end of a continuum of sexual aggression with severe sexual violence, e.g. rape, at the other extreme (Begany and Milburn 2002; Long 1998; see also Terpstra and Baker 1987). If sexual harassment is a form of aggression against women, then harassment-prone men may be psychologically similar to men who aggress against women in other ways, e.g. wife-directed aggression or sexual assault. Several empirical findings indicate that more aggressive husbands tend to be biased toward over-inferring their wives’ criticism and rejection (Holtzworth-Munroe and Smutzler 1996; Moore et al. 2000; Schweinle and Ickes 2007; Schweinle et al. 2002). Schweinle, Ickes, and Bernstein called this social-perceptual phenomenon the Critical/Rejecting Overattribution Bias (C/R-O Bias). In addition, Lisak and Roth (1988) found that men who had sexually assaulted women reported perceptions of belittlement by women more frequently than non-assaultive men. Given the growing body of literature suggesting the aggressive underpinnings of sexual harassment and evidence suggesting a relationship between men’s maltreatment of women and men’s bias to over-infer women’s criticism and rejection, we could expect to find that sexual harassment-prone men tend to be biased to overattribute criticism and rejection to women—Hypothesis 2a.

On the other hand, it is possible that men who exhibit sexual harassment-related behaviors are actually biased against inferring criticism or rejection. When observing a woman’s social interaction behavior, men are more likely than women to perceive sexuality in her behavior. Women are more likely see the same behavior as general friendliness (Abbey 1982; Saal et al. 1989). Baker et al. (1990) have suggested that this perceptual deficiency may underlie men’s sexual harassment behavior and that men who are prone to sexually harass women tend to incorrectly infer women’s sexuality or seduction when women intend criticism or rejection. Following this line of reasoning, we might expect to find that sexual harassment-prone men are biased against inferring women’s criticism and rejection—Hypothesis 2b.

There is yet another possibility worth considering—the possibility that there are two types of sexual harassers. One type is more aggressive and will tend to over-infer women’s criticism or rejection, similar to aggressive husbands and sexually assaultive men. The other type of sexual harasser, who is focused on seduction, will be biased toward under-inferring women’s criticism or rejection. If there are indeed two types of sexual harassers we could plausibly expect to find that more harassment-prone men will tend to be biased either toward over-inferring or toward under-inferring women’s criticism or rejection. The men who are least likely to sexually harass women would be those men who are the least biased in either direction. Hypothesis 2c proposes that there are actually two types of sexual harassers, the aggressive type who over-infer women’s criticism and rejection and the “narcissistic” type who under-infer women’s criticism and rejection.

Hypotheses

Based on the theoretical and empirical precedents cited above, we proposed the following hypotheses:

  1. Hypothesis 1:

    Men’s ability to accurately infer women’s criticism and rejection will be negatively related to their self-reported sexual harassment behavior towards women.

  2. Hypothesis 2a:

    Men who exhibit more sexual harassment-related behavior are biased toward over inferring criticism or rejection in women’s thoughts and feelings.

  3. Hypothesis 2b:

    Men who exhibit more sexual harassment-related behavior are biased against inferring criticism or rejection in women’s thoughts and feelings.

  4. Hypothesis 2c:

    A curvilinear relationship exists between men’s bias to infer women’s criticism or rejection and the men’s sexual harassment of women.

We tested these four hypotheses using a newly created self-report measure of men’s sexual harassment behavior, an adaptation of the empathic accuracy paradigm (Ickes et al. 1990) and signal detection methods, which have been effectively used in social psychology research and differentiate accuracy (sensitivity in signal detection terms) from bias (MacMillan and Creelman 1991; Gable et al. 2003; Schweinle and Ickes 2007; Schweinle et al. 2002).

The measures of men’s sexual harassment behavior and wife-directed aggression used in this study are face-valid self-reports. And, men tend to under-report their abuse of women (Dutton and Hemphill 1992; Jouriles and O’Leary 1985; Szinovacz 1983). For these reasons we used a shortened version of the Marlowe-Crowne socially desirable response scale, the M-C 1(10) (Strahan and Gerbasi 1972), and regression methods to adjust participants’ scores on all self-report measures of sexual harassment behavior and wife-directed aggression.

Method

Participants

As part of a larger investigation of men’s aggression toward women, the present sample was recruited through newspaper advertisements asking for married men in the Dallas–Fort Worth, Texas, USA area who were interested in a marital conflict study. Eighty-five men were paid $35 for their participation. However, two participants declined to participate shortly after the procedure began and data from three other participants were tainted due to local construction noise. The study sample consisted of the remaining 80 men, ranging in age from 21 to 73 years old (M = 43.4, SD = 12.5). At the time of testing they had been married from 4.5 months to 50 years (M = 14.6, SD = 12.5). Seventy percent of participants identified themselves as White/European-American, 16% as Black/African-American, 9% Hispanic/Latino, and 5% reported their ethnicity as “other.”

Measure

At the end of the procedure described below the participants were given a final battery of questionnaires which included the Sexual Harassment Behavior Inventory (also described below) and two widely used, well-validated and reliable measures of the men’s wife-directed aggression: the Conflict Tactics Scale (CTS, Straus 1979) and the Psychological Maltreatment of Women Inventory (PMWI, Tolman 1989). The CTS reliably measures self-reported aggression or abuse between marital partners (alphas > .80; Avakame 1998; Downey et al. 2000; Dutton 1998; Ehrensaft and Vivian 1999; Hanley and O’Neill 1997; Ryan 1998). The PMWI is designed to measure men’s psychological maltreatment of women in a close relationship. Dutton (1998) reports several studies using the PMWI that have yielded alphas greater than .80. In the present sample the alphas for the CTS and PMWI scales were .80 and .88, respectively. Scores from the CTS and the PMWI were computed by summing the individual item responses. These scores were used in exploratory analyses to help us better understand the implications of our findings.

The men’s scores on a shortened version of the Marlowe-Crowne social desirable response scale, the M-C 1(10) (alpha = .54 in the present sample; Strahan and Gerbasi 1972), were used with a regression method described by Saunders (1991) to adjust the men’s scores on face-valid self-report measures of sexual harassment behavior and wife-directed aggression.

Sexual Harassment Behavior Inventory (SHBI)

The Sexual Harassment Behavior Inventory, developed specifically for this study, is a reliable and face-valid self-report instrument in which men report the frequency of their behaviors that women tend to find sexually harassing. To create the 23 items on the SHBI, 24 female undergraduate psychology students were asked to recall and write down instances in which they felt sexually harassed (and were willing to disclose). These women varied in age and included both traditional and nontraditional students. Their responses were transcribed into a complete list and behaviors appearing more than once were collapsed into single entries. After several group discussions, the women further consolidated the list and agreed on 23 different behaviors that the majority of the women regarded as sexual harassment. These 23 behavior descriptions were assembled into a self-report scale (0 = never to 5 = always) designed to assess the relative frequency of male respondents’ self-reported sexual harassment behavior in the preceding year. Scores from all items, which are listed in Appendix A, were summed to calculate the SHBI score.

Since each behavior item in the SHBI represented sexual harassment to a majority of the women polled, the SHBI had reasonable construct validity and was internally reliable (α = .87 in the present sample of men). Because sexual harassment is a function of both the actor’s behavior and the target’s perception, it is difficult to measure the actual occurrence of sexual harassment from actors’ self reports alone. We therefore make the logical assumption that men who more frequently exhibit harassing behaviors are more likely to make women feel sexually harassed.

Despite the care with which this instrument was constructed there is an important limitation inherent to any results obtained from the SHBI. The women who reported men’s sexual harassment behaviors were college students, which may limit the generalizability of these findings to other populations of women.

Procedure

Each participant was met at the lab by the experimenter, seated in a cubicle, and given a consent form to read and sign. After the participant signed the consent form, the experimenter gave the participant an initial questionnaire battery to complete and then left the cubicle. The initial questionnaire battery included some demographic questions and the M-C 1(10) (Strahan and Gerbasi 1972).

After completing the initial questionnaire battery the participant was given several specially prepared inference forms. The participant then viewed an instructional videotape that described the basic procedure of viewing a videotape of a woman in a psychotherapy session and each time the tape was paused by the experimenter (1) writing, in the form of a sentence, the inferred content of the thought or feeling that the female therapy client had when the tape was stopped, (2) indicating whether he thought that the client’s thought or feeling was critical/rejecting or not critical/not rejecting of her ex-husband by checking the appropriate box on the inference form, and (3) restarting the tape by means of a remote control. The participant then viewed the videotape until the experimenter paused the tape again. The participant followed this process for all 30 thought-feeling inferences in the stimulus tape. The men’s critical/rejecting inferences were scored using signal detection indices of sensitivity (d′ ) and bias (B”D). These scores were then correlated (Pearson and regression) with the men’s SHBI scores to investigate the presence and direction of the men’s biases as proposed in hypotheses 1, 2a, 2b, and 2c.

The stimulus videotape was a slightly edited version of a videotape developed by Marangoni et al. (1995). The tape depicts a White/European-American, college-educated, 24 year old woman, who was participating in a videotaped psychotherapy session with a male therapist. She had previously consented to having her session videotaped and used as stimulus material in subsequent research. Although both the client and the therapist knew the therapy session would be viewed by others, it was genuine in nearly every other respect. The female client came to the therapy session prepared to discuss personal issues that were of real concern to her. The session was videotaped “live” from beginning to end without any prior rehearsal, with the genuineness and spontaneity of the session being evident in the range of emotional expressions the client displayed. The licensed male therapist, trained in the Rogerian tradition, used a nondirective approach by helping the client to clarify and explore the implications of her own statements while refraining from giving advice. Immediately after her therapy session, the female client viewed the videotape of the session and made a complete list of all the specific thoughts and feelings that she remembered having during the session as well as the time the thought or feeling occurred (using a running timer that appeared as an overlay on the video image).

Five female undergraduate students independently viewed the entire therapy session tape and the client’s reported thoughts and feelings to determine whether each thought or feeling was critical/rejecting of the client’s former husband. Fifteen of the therapy client’s 30 thoughts and feelings were rated as critical or rejecting and 15 were rated as not critical or rejecting. The overall Kappa coefficient (Fleiss 1981) for these ratings was .31 (Ho: κ = 0, z = 5.3, p < .001) which represents a “fair” level of interrater agreement (Landis and Koch 1977). Cronbach’s alpha for these ratings was .71.

Results

Table 1 contains descriptive statistics including reliability alphas and Pearson correlations for the measures used in this study.

Table 1 Descriptive statistics, reliability coefficients and Pearson intercorrelations for the variables of interest (n = 80).
Full size table

Psychometric Properties of the SHBI

The average raw SHBI score was 10.13 (SD = 9.63, range: 0 to 53). After adjustment for socially desirable response bias, the average SHBI score was 13.52 (SD = 9.52, range: 3.79 to 55.27). The SHBI exhibited good internal reliability in this sample (alpha = .87). A principal component analysis yielded a single factor with eigenvalue = 7.35. Five additional factors emerged with eigenvalues greater than one (range = 2.20 to 1.20). However, these factors were clearly beyond the “crook of the elbow” on the scree plot, suggesting that the SHBI measures a single factor, despite varying item reactivity or response “difficulty.” Final communality estimates for all 23 items ranged from .45 to .90. Because each behavior item in the SHBI represented sexual harassment to the majority of the 24 women polled, the SHBI possesses construct-validity.

Analyses of the Men’s Inferences

Signal detection analyses assessed the accuracy with which the men inferred the theme (Critical/Rejecting vs. Non-Critical/Non-Rejecting) as well as the degree and direction of any bias when inferring criticism or rejection in the female client’s thoughts and feelings. The present study used d’ as a measure of thematic accuracy (M = 1.21, SD = 1.06). A bias measure (B” D , Donaldson 1992) was computed to indicate bias on a scale of 1 to −1, with 1 being the strongest possible bias to overattribute criticism or rejection to the therapy client. The present sample had an average bias score of .004 (SD = .53) suggesting that the sample included a good mix of men who were biased toward and away from inferring women’s criticism and rejection.

Hypothesis 1 proposed that men’s sexual harassment of women is inversely related to the men’s ability to accurately infer women’s critical or rejecting thoughts and feelings. The correlation between the men’s adjusted SHBI scores and the men’s accuracy for inferring the therapy client’s criticism or rejection (d’) approached significance in a negative direction (r = −.22, p < .055) suggesting the possibility that men who reported more sexual harassment behavior may be less sensitive, i.e. less able to accurately infer women’s critical or rejecting sentiments.

The second set of hypotheses (2a, 2b, & 2c) proposed three competing arguments regarding harassment prone men’s possible inferential bias. The initial finding supported Hypothesis 2a; the correlation between the men’s SHBI scores and their bias to overinfer the stimulus target’s criticism or rejection was positive and significant, r = .22, p < .05.

The positive correlation between men’s C/R-O bias and their sexual harassment ruled against Hypothesis 2b, which proposed that more harassment-prone men are biased against inferring women’s criticism and rejection.

However, these results do not rule out Hypothesis 2c, which posits a quadratic relationship between men’s sexual harassment of women and their bias to infer women’s sexual harassment. We tested Hypothesis 2c with a quadratic regression analysis (Adjusted SHBI Score = b 0 + b 1(C/R Bias) + b 3(C/R Bias2)). The fit of the quadratic model was not significant (F(2, 77) = 2.82, p > .075) , which does not support the conclusion that there are two types of sexual harassers.

In summary, the overall pattern of results suggests that men who are more prone to sexual harassment-prone tend to be biased toward overinferring criticism or rejection in women and suggest the possibility that these men tend to be inaccurate when making this type of inference.

Other Correlates of the Men’s SHBI Scores

In addition to having a positive relationship with the strength of the men’s C/R-O bias, the men’s scores on the SHBI were significantly related to their adjusted scores on the CTS, r = .35, p < .01. Similarly, the men’s adjusted PMWI scores were significantly related to the men’s adjusted SHBI scores, r = .38, p < .01. These findings are consistent with the assertion by O’Leary-Kelly et al. (2000) that men’s sexual harassment of women is primarily aggressive/abusive behavior rather than seductive behavior. Further, this pattern of results suggests that more harassment-prone men tend to aggress more against their own wives than less harassment-prone men.

Discussion

The Sexual Harassment Behavior Inventory, a modified version of the empathic accuracy paradigm (Ickes 1997; Schweinle et al. 2002) and signal detection methods were used in this investigation to test four hypotheses about the relationships between men’s sexual harassment of women and their accuracy and bias when inferring women’s criticism and rejection. The following discussion will focus first on the SHBI and follow with a discussion of the hypothesis results.

The Sexual Harassment Behavior Inventory

The SHBI was developed for this study and includes 23 items—each describing men’s behaviors—that a group of college women had experienced and tended to find sexually harassing. The SHBI exhibited good internal reliability and has content and construct validity within a college population. One important shortcoming of the SHBI is that the women who contributed the SHBI items, although ranging widely in age and work experience, were college students, which may result in less generalizability to other populations. The reader should also note that the SHBI was not developed to test for specific types of sexual harassment. Fitzgerald, Shullman, Bailey and Richards (1988) suggested that sexual harassment takes many different forms, including gender harassment, seductive behavior, sexual bribery, sexual coercion, and sexual assault. Future development of the SHBI could result in subscales designed to differentially measure these specific types of harassment.

Further research is also needed to more thoroughly assess the measurement and predictive properties of the SHBI. In particular, a validation study should be conducted to compare men’s self-reported sexual harassment with the perceived harassment that is reported by women who know or work with the men. Future research should also obtain potential items from a larger and more diverse sample of women than the group that contributed to the present version of the SHBI. New items could be integrated into the scale or perhaps used to replace some items in the current version of this measure.

However, the results obtained with the SHBI are in line with prior reports and can lend insight into the social cognition of sexual harassment-prone men.

Hypothesis 1

The results did not clearly support our first hypothesis that men’s ability to accurately infer women’s criticism and rejection will be negatively related to their self-reported sexual harassment behavior towards women. However this correlation did approach significance. Future research with larger samples, additional stimulus tapes, and additional measures of men’s sexual harassment propensity is needed to better test this relationship.

Hypotheses 2a, 2b, and 2c

Recall that hypotheses 2a, 2b, and 2c were mutually exclusive predictions about the relationship between men’s bias to infer women’s criticism or rejection. The results supported hypothesis 2a: men who report behaving in sexually harassing ways tend to over-infer criticism or rejection in women’s thoughts and feelings. This social cognitive characteristic has been called the Critical/Rejecting Overattribution bias by Schweinle et al. (2002). Because men’s C/R-O bias has also been found among more aggressive husbands (Schweinle and Ickes 2007; Schweinle et al. 2002), this finding is consistent with the emerging literature suggesting that sexual harassment is a form of aggression against women (Begany and Milburn 2002; Long 1998; O’Leary-Kelly et al. 2000; Terpstra and Baker 1989). This implication is further supported by the positive correlation between men’s SHBI scores and the men’s scores on two wife-directed aggression measures: the Conflict Tactics Scale (Straus 1979) and the Psychological Maltreatment of Women Inventory (Tolman 1989). This overall pattern of results is perhaps the most important finding in this study, and it makes an especially strong argument that men’s sexual harassment of women has aggressive or hostile underpinnings. On the other hand, these findings are correlational, and as such are subject to more skeptical interpretation than results from stronger empirical designs.

Until recently, sexual harassment research tended to focus on the perceptions of sexually harassed people, primarily women (O’Leary-Kelly et al. 2000). The current study shifts the focus to the psychology of the male sexual harasser, and the data support the conclusion that sexual harassers are biased to overattribute criticism and rejection when inferring the specific content of a woman’s thoughts and feelings. The present data further hint that sexual-harassment prone men may be less accurate when they infer women’s criticism or rejection.

A practical implication of these findings is that interventions designed to reduce male-to-female sexual harassment should incorporate features that decrease the strength of the men’s C/R-O bias. The present data tend to suggest that increasing men’s critical/rejecting inferential accuracy may also reduce the men’s sexual harassment behavior, though this assertion could stand more thorough testing.

Teaching empathy in both psychotherapy and in sexual harassment prevention training is particularly effective in reducing repeated occurrences of sexual harassment (Brunswig and O’Donohue 2002). Dobrich and Dranoff (2000) suggest a similar empathy-based approach to sexual harassment prevention, stressing the need for empathy between coworkers and between management and workers to decrease the incidences of sexual harassment in organizations. It is possible that increased empathy results in greater critical/rejecting inferential accuracy, lower C/R-O bias and, ultimately, less harassing behavior. Future research should test these causal relationships empirically.

The pattern of correlations between participants’ sexual harassment scores, their C/R-O Bias and their aggression against their own wives has even broader implications. Perhaps most striking, the results support previous reports that sexual harassment is a form of aggression against women (O’Leary-Kelly et al. 2000). These findings also raise the possibility that men’s C/R-O bias may be related to other forms of men’s aggression toward women, forms that do not specifically qualify as wife-directed aggression or sexual harassment, per se.

Conclusions

The present study developed a reasonably valid and reliable instrument to explore the socio-cognitive characteristics of male perpetrators of sexual harassment. Signal detection analyses, which are ideal for differentiating recognition from interpretation, were used to assess how men infer women’s critical or rejecting thoughts and feelings. These methods revealed that more harassment-prone men tend to be biased to over-infer women’s criticism or rejection, i.e. have a C/R-O bias, and to make these inferences at the wrong times. The present findings demonstrate that, with respect to the C/R-O bias, sexual harassers are similar to, if not actually, men who are prone to aggress against their wives. This finding supports recent findings linking men’s sexual harassment of women to aggression rather than seduction.