Heterosocial competence of rapists and child molesters: a meta-analysis.

Clinical psychologists (e.g., Abel, Blanchard, & Becker, 1976) have found that many of the sex offenders they treated had poor heterosocial skills, that is, they lacked the social skills necessary to function heterosexually (Barlow, Abel, Blanchard, Bristow, & Young, 1977). These observations have spawned a large body of research examining the social competence and heterosocial competence of rapists and child molesters.

In the research literature on social skills, various terms have been identified and sometimes used interchangeably. These include social competence, social skills, heterosocial skills, and heterosocial competence. For example, McFall (1990) distinguishes social competence from social skills in that competence is the evaluation of a person's performance of a particular task, while skills are the underlying processes that enable a person to perform that task competently. Neither of these concepts implies that the social tasks involve interactions with a member of the other sex.

On the other hand, heterosocial skills and heterosocial competence assess a person's ability to successfully interact with members of the other sex. So, based on McFall's definition, heterosocial competence could be conceptualized as the evaluation of a person's performance in heterosexual interactions, while heterosocial skills could be defined as underlying processes that enable a person to successfully interact with members of the other sex.

The purpose of this study is to examine the heterosocial competence of rapists and child molesters. For purposes of this analysis, heterosocial competence has been operationally defined as the ability to competently interact with members of the other sex. Since a number of different measures could be defined this way, a fairly broad set of measures was included in this analysis. Self-report measures of heterosocial competence indicate the person's own evaluation of his or her ability, while performance measures indicate the person's performance on a task, such as a roleplaying situation, involving a person of the other sex.

Barlow et al. (1977) argue that rape and child sexual abuse are sexual deviations and that they result, in part, from an inability to establish normal sexual relationships. Freund (1988) identifies rape as a "courtship disorder," that is, an anomalous performance of courtship behavior. However, this only applies to the minority of rapists who have an erotic preference for rape over consensual intercourse (Freund, Scher, & Hucker, 1983). Ward, McCormack, Hudson, and Polaschek (1997) suggest that sexual offenders fail to achieve and maintain close relationships and are insecurely attached to their romantic partners. Presumably, these individuals use sexual coercion to compensate for their inability to achieve intimacy.

The research evidence on the relationship between heterosocial competence and sexual offending is mixed. For rapists, some studies have found that they do not differ from nonrapists in heterosocial competence (e.g., Koralewski & Conger, 1992), while others have found that rapists show less heterosocial competence than nonrapists (e.g., Lipton, McDonel, & McFall, 1987). However, for child molesters, most studies have found that sexual abusers of children are deficient in heterosocial competence (e.g., Bumby & Hansen, 1997). Thus, there may be a stronger relationship between heterosocial competence and sexual offending for child molesters than for rapists.

Some of the studies reviewed here (e.g., Overholser & Beck, 1986) compared sex offenders and non-sex-offenders from different populations. Specifically, rapists and child molesters who were in prison (i.e., incarcerated sex offenders) were compared with college students or other volunteers who were not in prison (i.e., nonincarcerated non-sex-offenders). In some cases, these incarcerated sex offenders were specifically selected to undergo psychiatric treatment (e.g., Hudson & Ward, 1997). The problem with comparing incarcerated sex offenders with nonincarcerated sex offenders is that individuals in prison may have worse social skills in general than nonincarcerated individuals (Muehlenhard & Falcon, 1990). Also, sex offenders with poor social skills are probably more likely to be convicted of sexual crimes than those with good social skills (Clark & Lewis, 1977). Thus, incarcerated sex offenders would be expected to have lower heterosocial competence than nonincarcerated non-sex-offenders.

Other studies (e.g., Hudson & Ward, 1997) compared sex offenders and non-sex-offenders from similar populations. Some compared college students or other volunteers who indicated in self-report measures that they had committed rape (i.e., nonincarcerated rapists) with nonincarcerated non-sex-offenders. Unfortunately, there are no data on nonincarcerated child molesters. Others compared incarcerated sex offenders with prisoners who were not sex offenders (i.e., incarcerated non-sex-offenders). These studies may not find large differences between sex offenders and non-sex-offenders. Sex offenders in prison might be expected to share similar levels of heterosocial competence with other prisoners. Likewise, sex offenders who are college students might be expected to share similar levels of heterosocial competence with other college students.

I conducted a meta-analysis on the research on the heterosocial competence of sex offenders. I predicted that effect sizes would differ based on the type of sex offender sampled (rapist or child molester) and whether the samples were incarcerated or nonincarcerated.

HYPOTHESIS ONE

The first analysis compared the heterosocial competence of all rapists (including both incarcerated and nonincarcerated samples) with that of all non-sex-offenders (including both incarcerated and nonincarcerated samples). Overall, I predicted that rapists would have significantly lower heterosocial competence than non-sex-offenders. However, I anticipated this effect would be relatively small because comparisons involving sex offenders from different populations would yield large effect sizes but comparisons from similar populations would yield small effect sizes.

HYPOTHESIS TWO

Based on the reasoning that incarcerated individuals would have less heterosocial competence than nonincarcerated individuals, I predicted that the difference in heterosocial competence between incarcerated rapists and nonincarcerated non-sex-offenders would be significantly larger than the difference between incarcerated rapists and incarcerated non-sex-offenders. Based on the same reasoning, I predicted that the difference in heterosocial competence between incarcerated rapists and nonincarcerated non-sex-offenders would be significantly larger than the difference between nonincarcerated rapists and nonincarcerated non-sex-offenders.

HYPOTHESIS THREE

I predicted that child molesters would have lower heterosocial competence than both incarcerated and nonincarcerated non-sex-offenders. Since most of the studies I reviewed showed significant differences between child molesters and non-sex-offenders, I expected a significant overall effect size.

HYPOTHESIS FOUR

I predicted that the difference in heterosocial competence between child molesters and both incarcerated and nonincarcerated non-sex-offenders would be significantly larger than the difference between rapists and both incarcerated and nonincarcerated non-sex-offenders. Research suggests that heterosocial competence is more of a problem for child molesters than for rapists (e.g., Bumby & Hansen, 1997).

HYPOTHESIS FIVE

Based on the reasoning that incarcerated individuals would have lower heterosocial competence than nonincarcerated individuals, I predicted that the difference in heterosocial competence between child molesters (all of whom were incarcerated) and nonincarcerated non-sex-offenders would be significantly larger than the difference between child molesters and incarcerated non-sex-offenders.

METHOD

Literature Search

I conducted an exhaustive literature search for studies that examined the social competence of people who engaged in sexual coercion. Past research in this area has only examined child sexual abuse (familial and nonfamilial) and rape. I conducted online searches on the PsycINFO database using the keywords sexual coercion and rape. I scanned all relevant articles for appropriate references and conducted manual searches of Archives of Sexual Behavior (1971-December 2000), Criminal Justice and Behavior (1986-September 2000), Journal of Consulting and Clinical Psychology (1980-June 2000), The Journal of Sex Research (1975-August 2000), Journal of Interpersonal Violence (1986-October 2000), and Sex Roles (1975-September 2000).

Inclusion Criteria

Studies must have assessed participants' heterosocial competence using either behavioral or self-report measures. I also collected effect size data from measures of similar constructs such as social anxiety and fear of intimacy as long as the context specifically involved a heterosexual relationship.

Studies must also have had sex offenders in their samples. The offenders in these samples need not have been convicted. College students, for example, who reported engaging in physical sexual coercion would be included in a sample of rapists.

Studies Not Included in This Meta-Analysis

Some studies reported that there were "no significant differences" between groups but did not provide any other statistical information. For example, Spaccarelli, Bowden, Coatsworth, and Kim (1997) found no significant differences between sexually aggressive, violent non-sex-offending and less violent non-sex-offending delinquents in perceived competence. However, because the authors did not report means and standard deviations, these findings could not be included. The null results were not included in the analysis, thus inflating the overall effect size. (1)

Other studies (Lipton et al., 1987; Overholser & Beck, 1986; Stahl & Sacco, 1995) provided adequate statistical information for some measures of heterosocial competence, but not for others. So, only some measures from these studies were entered into the analysis.

Studies Included in This Meta-Analysis

In total, I obtained 14 published papers fitting the study criteria (Bumby & Hansen, 1997; Hudson & Ward, 1997; Katz, 1990; Koralewski & Conger, 1992; Lipton et al., 1987; Lisak & Ivan, 1995; Malamuth & Brown, 1994; Muehlenhard & Falcon, 1990; Murphy, Coleman, & Haynes, 1986; Overholser & Beck, 1986; Segal & Marshall, 1985, 1986; Stahl & Sacco, 1995; Ward, Hudson, & Marshall, 1996). These papers yielded a total of 119 effect sizes representing the responses of 5,295 participants. Table 1 presents a summary of these studies.

All participants were adult males, except for Katz's (1990) sample, which consisted of adolescent males averaging 15 years of age. Rapist samples were either incarcerated or nonincarcerated. One study (Bumby & Hansen, 1997) used incarcerated rapists who were selected to participate in a treatment program for sex offenders.

All child molester samples were incarcerated. However, in four studies (Bumby & Hansen, 1997; Hudson & Ward, 1997; Katz, 1990; Ward et al., 1996) the child molesters had been selected to enter treatment programs. Some studies did not specify whether the child molesters in their samples included incest offenders or nonfamilial child molesters or whether they victimized boys or girls. (2) Therefore, in this meta-analysis child molesters were placed in a single category.

Each test was coded for the following predictors: type of sex offender sample (1 = students or other volunteers, 2 = incarcerated rapists, 3 = child molesters); type of non-sex-offender sample (1 = students or other volunteers, 2 = incarcerated non-sex-offenders); and type of measure (1 = self-report, 2 = performance).

Computation of Effect Sizes

I used the effect size index g, representing the difference between the rapists' group and the non-sex-offenders' group divided by the pooled standard deviation (SD). These gs were converted to ds to correct for sample size bias (Hedges & Olkin, 1985).

In this meta-analysis, a positive effect size indicates that the sex offender group has a higher level of social competence than the non-sex-offender group. Pooled SDs were based on individual cell SDs whenever possible, or else were derived from analysis of variance results.

Multiple Effect Sizes

A number of studies yielded multiple effect sizes because they employed multiple measures of heterosocial competence. Other studies yielded multiple effect sizes because they used multiple group comparisons, such as comparing both rapists and child molesters with more than one group of non-sex-offenders, such as incarcerated non-sex-offenders and college students.

Since they are no longer independent, these multiple effect sizes bias the results of the meta-analysis. However, it is of critical theoretical interest to compare the results of rapists and child molesters with both incarcerated and non-incarcerated groups. Therefore, I decided that every group of rapists and child molesters would each be contrasted with every group of non-sex-offenders.

Some studies yielded multiple effect sizes because they used both performance and self-report measures of heterosocial competence. I conducted a set of categorical analyses to determine whether self-report and performance measures yielded significantly different effect size data. If there were no differences, then I could assume that the overall difference in heterosocial competence between sex offenders and nonoffenders was the same for both self-report and performance measures. If there was a difference, then I could assume that self-report and performance measures assessed heterosocial competence to different degrees or that they assessed different constructs.

Only one comparison--nonincarcerated rapists versus nonincarcerated nonrapists--yielded a significant difference between self-report and performance measures. For this comparison, performance measures yielded a significantly greater effect size than did self-report measures, [Q.sub.B](I) = 9.69, p < 0.01. Therefore, for this comparison I analyzed self-report and performance measures separately. All other effect size data reflect the combined effects of self-report and performance measures. If studies used multiple performance measures or multiple self-report measures, I averaged the effect sizes within these categories, when possible using methods specified by Rosenthal and Rubin (1986).

I dealt with the problem of using multiple effect sizes from the same study by using methods specified by Erel and Burman (1995). First, I calculated the correlation between the number of effect sizes reported per study and the magnitude of the composite mean weighted effect sizes reported by the studies. This correlation indicated that the number of effect sizes obtained from a study was not correlated with the magnitude of effect sizes, r = .05, p = .87. Second, I calculated the correlation between the number of effect sizes obtained from a study and the homogeneity statistics of the effect sizes obtained from the study. This correlation indicated that the number of effect sizes obtained from a study was not correlated with homogeneity statistics, r = -.02, p = .94. In other words, multiple effect sizes from single studies did not bias results toward greater homogeneity or greater magnitude of effect.

RESULTS

Single Effect Size Model

The first analysis examined the heterosocial competence of all sex offenders compared with all non-sex-offenders. All measures of heterosocial competence were included in the analysis.

The combined effect of 119 tests of heterosocial competence, weighting each test by sample size, revealed a significant effect size indicated by Hedges' d = -.36, 95% CI = -.41 / -0.30, r = -.18, p < .001. Sex offenders (rapists and child molesters inclusive) had less heterosocial competence than non-sex-offenders (incarcerated and nonincarcerated inclusive).

To address the "file drawer problem" (Rosenthal, 1979), defined as the likelihood that a number of relevant studies were never published due to null results, I conducted a file drawer analysis (Wolf, 1986). The analysis indicated that 95 additional null effects would have to be added to the data set before the overall effect size would be reduced to .2, which Cohen (1977) has identified as a small effect size.

I expected significant variance heterogeneity because child molesters and rapists would presumably have differing levels of heterosocial competence. I found this to be the case ([Q.sub.T] = 332.43); therefore I deemed it appropriate to conduct further categorical analyses (see Table 2).

First Residual Model

The first residual model divided sex offenders into two categories: rapists and child molesters.

Rapists versus non-sex-offenders. I predicted in hypothesis one that rapists would have significantly lower heterosocial competence than non-sex-offenders. This hypothesis was supported.

The combined effects of 63 tests of heterosocial competence, weighting each test by sample size, revealed a significant overall effect size, d = -.23, 95% CI = -.30 / -0.16, r = -.11, p < .001 (see Table 2). Overall, rapists had less heterosocial competence than non-sex-offenders.

Child molesters versus non-sex-offenders. I predicted in hypothesis three that child molesters would have significantly lower heterosocial competence than non-sex-offenders. This hypothesis was supported.

The combined effects of 56 tests of heterosocial competence, weighting each test by sample size, revealed a significant effect size, d = -.54, 95% CI = -0.62 / -0.45, r = -.26, p < .001 (see Table 2). Child molesters had a significantly lower level of heterosocial competence than did non-sex-offenders.

I predicted in hypothesis four that the difference in heterosocial competence between child molesters and non-sex-offenders would be significantly larger than the difference in heterosocial competence between rapists and non-sex-offenders. This hypothesis was supported.

The overall effect size obtained from comparing rapists with non-sex-offenders was compared with the overall effect size obtained from comparing child molesters with non-sex-offenders. As predicted, I found a significant difference between these two effects, [Q.sub.B](1) = 30.35, p < 0.001 (see Table 2).

As expected, this first categorical model still had variance heterogeneity since rapists and child molesters were compared with both incarcerated and nonincarcerated non-sex-offenders. Significant variance heterogeneity was indicated for both rapists ([Q.sub.W] = 168.39) and child molesters ([Q.sub.W] = 133.69) compared with non-sex-offenders. Each of these [Q.sub.W] statistics remained significant after subtracting [Q.sub.B]: Thus, it was appropriate to continue with a second residual model.

Second Residual Model

The second residual model categorized sex offenders and non-sex-offenders based on whether or not they were incarcerated.

Nonincarcerated rapists versus nonincarcerated non-sex-offenders. For self-report measures, the combined effects of five tests of heterosocial competence with each test weighted by sample size revealed a non-significant effect size, d = +0.07, 95% CI = -0.23 / +0.37, r = +.04, p = .63. For performance measures, the combined effects of four tests of heterosocial competence with each test weighted by sample size revealed a significant effect size, d = -0.45, 95% CI = -0.58 / -0.31, r = -.18, p < .001 (see Table 2). For performance measures only, nonincarcerated rapists had significantly lower heterosocial competence than nonincarcerated non-sex-offenders.

Incarcerated rapists versus incarcerated non-sex-offenders. The combined effects of 36 tests of heterosocial competence, weighting each test by sample size, revealed a nonsignificant effect size, d = +0.05, 95% CI = -0.07 / +0.17, r = +.02, p = .44 (see Table 2). Incarcerated rapists did not differ from incarcerated non-sex-offenders in heterosocial competence.

Incarcerated rapists versus nonincarcerated non-sex-offenders. The combined effects of 29 tests of heterosocial competence, weighting each test by sample size, revealed a significant effect size, d = -0.38, 95% CI = -0.51 / -0.26, r = -.19, p < .001 (see Table 2). Incarcerated rapists had significantly lower heterosocial competence than nonincarcerated non-sex-offenders.

Child molesters versus nonincarcerated non-sex-offenders. The combined effects of 30 tests of heterosocial competence, weighting each test by sample size, revealed a nonsignificant overall effect size, d = -0.82, 95% CI = -0.94 / -0.69, r = -.38, p < .001 (see Table 2). Child molesters had significantly lower heterosocial competence than nonincarcerated non-sex-offenders.

Child molesters versus incarcerated non-sex-offenders. The combined effects of 26 tests of heterosocial competence, weighting each test by sample size, revealed a significant overall effect size, d = -0.30, 95% CI = -0.41 / -0.19, r = -.15, p < .001 (see Table 2). Child molesters had significantly lower heterosocial competence than incarcerated non-sex-offenders.

This second residual model still had variance heterogeneity. However, when [Q.sub.B] was subtracted from [Q.sub.W], none of the comparisons were significantly heterogeneous (see Table 2).

Child molesters versus rapists. I included an additional analysis to assess differences between child molesters and rapists. The combined effects of 20 tests of heterosocial competence, weighting each test by sample size, revealed a significant overall effect size, d = -0.33, 95% CI = -0.46 / -0.19, r = -.16, p < .001. Child molesters had significantly lower heterosocial competence than rapists.

Comparison Tests

I predicted in hypothesis two that the difference in heterosocial competence between incarcerated rapists and nonincarcerated non-sex-offenders would be significantly larger than the difference between incarcerated rapists and incarcerated non-sex-offenders. This hypothesis was supported.

The effect size obtained from comparing incarcerated rapists with nonincarcerated non-sex-offenders was compared with the effect size obtained from comparing incarcerated rapists with incarcerated non-sex-offenders. I found a significant difference between these two effects, [Q.sub.B](1) = 23.32, p < 0.001. The difference in heterosocial competence between incarcerated rapists and nonincarcerated non-sex-offenders was significantly larger than the difference in heterosocial competence between incarcerated rapists and incarcerated non-sex-offenders (see Table 3).

I also predicted in hypothesis two that the difference in heterosocial competence between incarcerated rapists and nonincarcerated non-sex-offenders would be significantly larger than the difference between nonincarcerated rapists and nonincarcerated non-sex-offenders. This hypothesis was partially supported.

Because self-report measures and performance measures of heterosocial competence yielded significantly different overall effect sizes for the comparison of nonincarcerated rapists with nonincarcerated non-sex-offenders, I conducted the following analyses separately for self-report and performance measures. The overall effect size comparing incarcerated rapists with nonincarcerated non-sex-offenders was compared with the effect size comparing nonincarcerated rapists with nonincarcerated non-sex-offenders. For self-report measures, I found a significant difference between these two effects, [Q.sub.B](1) = 6.23, p < 0.02. The difference in heterosocial competence between incarcerated rapists and nonincarcerated non-sex-offenders was significantly larger than the difference in heterosocial competence between nonincarcerated rapists and nonincarcerated non-sex-offenders. However, for performance measures, I found no significant difference between these two effects, [Q.sub.B](1) = .54, p = 0.46. (see Table 3).

I predicted in hypothesis five that the difference in heterosocial competence between child molesters and nonincarcerated non-sex-offenders would be significantly larger than the difference between child molesters and incarcerated non-sex-offenders. This hypothesis was supported.

The effect size obtained from comparing child molesters with nonincarcerated non-sex-offenders was compared with the effect size obtained from comparing child molesters with incarcerated non-sex-offenders. I found a significant difference between these two effects, [Q.sub.B](1) = 36.90, p < 0.001. The difference in heterosocial competence between child molesters and nonincarcerated non-sex-offenders was significantly larger than the difference between child molesters and incarcerated non-sex-offenders (see Table 3).

Study Quality

I examined studies to determine if there were any specific methodological flaws that should also be taken into account. There were no flagrant methodological flaws in any of the studies. However, some studies used measures with fairly poor reliability or did not report reliability statistics at all. For example, the hostility- and seduction-discrimination indices (Malamuth & Brown, 1994; Murphy et al., 1986) were reported without any reliability data. Also, Segal and Marshall (1986) only reported reliability data on some of their measures.

I conducted another meta-analysis in which effect sizes derived from measures with questionable reliability were removed from the analysis. I found that when these effect sizes were removed, the results were identical to the results obtained when they were left in, with one exception: Nonincarcerated rapists no longer differed from nonincarcerated non-sex-offenders in heterosocial competence. Thus, after the unreliable measures were removed, all hypotheses were supported.

DISCUSSION

Each group of sex offenders was found to have significantly lower heterosocial competence than nonincarcerated non-sex-offenders. The relative levels of heterosocial competence among sex offenders are displayed in Table 4.

Rapists

Predictions regarding the heterosocial competence of rapists were well supported. Rapists had lower heterosocial competence than non-sex-offenders, yet the effect size (-.24) of this relationship was low. According to Cohen (1977), an effect size of .2 is considered low. Therefore, in distinguishing rapists from non-sex-offenders or predicting who might be likely to commit rape, heterosocial competence should not be ignored, but should not be emphasized as a primary factor in sexual coercion.

I predicted that incarcerated rapists would have lower heterosocial competence than nonincarcerated non-sex-offenders, and that this effect size would be significantly larger than the effect size for rapists compared with non-sex-offenders from similar populations. This hypothesis was partially supported.

As hypothesized, the effect size for the difference between incarcerated rapists and nonincarcerated non-sex-offenders was significantly larger than the effect size for the difference between incarcerated rapists and incarcerated non-sex-offenders. In fact, incarcerated rapists did not differ at all from incarcerated non-sex-offenders in heterosocial competence (see Table 2).

Contrary to prediction, the effect size for the difference between incarcerated rapists and nonincarcerated non-sex-offenders was no different from the effect size for the difference between nonincarcerated rapists and nonincarcerated non-sex-offenders. Also, nonincarcerated rapists had significantly lower heterosocial competence than nonincarcerated non-sex-offenders. However, when studies using measures of questionable reliability were eliminated from the analysis, this difference disappeared.

Comparing incarcerated rapists with nonincarcerated non-sex-offenders inflates the estimate of the true difference in heterosocial competence between rapists and non-sex-offenders. Rapists did not differ from non-sex-offenders in heterosocial competence when both groups were incarcerated.

The problem with relying on prison samples to study sex offenders is that they are unrepresentative of the population of sex offenders in general. For example, Russell (1984) points out that most rapes are not reported, and when these rapes are reported, most accused rapists do not get convicted (Koss, 1988). Rapists who do get convicted are usually poor and uneducated (Marolla & Scully, 1986). Since most rapists are not incarcerated, drawing conclusions from prison samples does not tell us enough about rapists in general.

Men who get sent to prison for violent and nonviolent offenses, including rape, seem to share similar levels of heterosocial competence. In addition, individuals who have been in prison for a long period of time may have developed poor heterosocial skills. The relationship between social skills and sexual coercion may be based primarily on the relationship between social skills and crime in general.

Other variables may also contribute the relationship between heterosocial competence and sexual coercion. For example, the relationship may be stronger for individuals with a low socioeconomic or educational level. Also, there may be cultural differences in defining what behaviors are appropriate when interacting with the other sex. Thus, measures of heterosocial competence may have a cultural bias. Future research should take these variables into account.

Many men and women engage in psychological sexual manipulation, in which they lie, deceive, falsely proclaim love, or threaten breakups to coerce their partners to have sex (O'Sullivan, Byers, & Finkelman, 1998). Muehlenhard and Falcon (1990) found that heterosocially skilled men were more likely than unskilled men to have engaged in verbal sexual coercion. Perhaps rapists with poor heterosocial skills use physical coercion and end up in prison, while rapists with effective heterosocial skills use verbal or psychological sexual coercion and avoid breaking the law.

Child Molesters

The studies reviewed in this article have consistently shown that child molesters have poorer heterosocial skills than either rapists, incarcerated non-sex-offenders, or nonoffenders.

I predicted that child molesters would have less heterosocial competence than non-sex-offenders. This hypothesis was supported, and is consistent with previous research.

I also predicted that the difference in heterosocial competence between child molesters and non-sex-offenders would be significantly larger than the difference between rapists and non-sex-offenders. This hypothesis was also supported. Since child molesters have less heterosocial competence than rapists, treatment programs that include improving offenders' heterosocial competence may be more appropriate for child molesters than for rapists.

Finally, I predicted that the difference in heterosocial competence between child molesters and nonincarcerated non-sex-offenders would be significantly larger than the difference between child molesters and incarcerated non-sex-offenders (see Table 3). This hypothesis was supported. Clearly, comparing sex offenders with nonincarcerated non-sex-offenders will yield larger effect sizes than comparing them with incarcerated non-sex-offenders.

Limitations

Heterosocial competence is a broadly defined construct. Therefore, I used a broad set of measures assessing heterosocial competence in this meta-analysis. One of the most difficult tasks in conducting a meta-analysis is deciding what measures are or are not appropriate to include. Researchers who want to create a "new" measure often fail to clearly distinguish it from other highly similar measures. Thus, many measures exist in various bodies of literature that share a great deal of common variance. In a meta-analysis, a balance must be achieved so that the data set does not contain measures that are too diverse or too narrowly defined. All measures in this meta-analysis involve a person's ability to interact with a person of the opposite sex in an actual, hypothetical, or potentially romantic situation, or the ability to correctly interpret a woman's affective cues in a romantic situation. It is possible that the measures I selected were too divergent. Analyzing self-report and performance measures separately dealt with this problem, but it is still possible that this meta-analysis was cluttered with measures that really did not share a great deal of similarity.

Another limitation of this study is that there are a large number of undetected rapists and child sexual abusers in the community (Finkelhor & Lewis, 1988; Russell, 1984) and these individuals could not be included in the analysis. Thus, we do not have enough information about the heterosocial competence of a large proportion of the population of sex offenders. Further, it seems likely that men with good heterosocial skills would successfully avoid detection.

Implications for Treatment

A number of clinicians recommend social skills training as an integral component of treatment for both rapists and child molesters (e.g., Marshall, Anderson, & Champagne, 1997). The results from this meta-analysis suggest that this may be a mistake. Improving rapists' social skills may actually allow them greater access to their victims and, thus, greater opportunity to engage in sexual coercion (Muehlenhard & Falcon, 1990). Thus, it is questionable whether social skills training should be the key component of treatment for rapists.

This meta-analysis has shown that rapists do not have a great deficiency in heterosocial competence, as compared with non-sex-offenders. Therefore, the argument that sex offenders engage in sexual coercion because they have difficulty establishing consensual relationships is probably false. In fact, sexually coercive college men reported having more sexual experience and a greater preference for partner variety and casual sex than noncoercive men (Lalumiere, Chalmers, Quinsey, & Seto, 1996).

On the other hand, this meta-analysis has shown that sexual abusers of children have substantial deficiencies in heterosocial skills. Therefore, social skills training may be useful in the treatment of these offenders. Nevertheless, treatment programs for child molesters should not exclusively focus on improving social skills, since this does not address child molesters' sexual arousal to children. Hall's (1995) treatment program, for example, is designed to reduce child molesters' deviant sexual arousal as well as increase empathy for the victim and improve anger management. Social skills training may be helpful as one component of a comprehensive treatment program for child molesters.

It needs to be reemphasized that all of the child molesters included in this meta-analysis were incarcerated. Therefore, sexual abusers of children who have not been caught may have good social skills and be able to get away with their crimes. Men who travel abroad to seek underage sexual partners may not have any heterosocial skills deficits. Rather, they may be taking advantage of another country's permissiveness toward child prostitution. Thus, it is very important in the future to study the characteristics of child molesters who are not imprisoned.

Child sexual abuse does not only include the acts of a tiny minority of sexual deviants. According to Finkelhor and Lewis (1988), there are vast numbers of undetected child molesters. However, there are no data on the heterosocial competence of these individuals. It is quite possible that these individuals have no difficulty establishing heterosexual relationships and that their sexual abuse of children is not merely a substitute for unattainable adult sexual relationships.

Implications for Future Research

Results from this meta-analysis suggest that men who commit child sexual abuse, but not men who commit rape, have some difficulties interacting with women or interpreting women's emotional responses. Clearly, there are other factors that account for sexual coercion. Sex offenders do not necessarily engage in sexual coercion because they are unsuccessful in heterosocial situations.

Studies that classify participants based on simple categories such as rapist or child molester miss out on the important distinctions within these groups. Rapists include college students who attack their dates and career criminals who commit rape in conjunction with burglary or robbery. Child molesters include those who only attack family members as well as those who seek out anonymous encounters with children from foreign countries. There is very little research on self-reported, unidentified child molesters. For example, Fromuth, Burkhart, and Jones (1991) found that 3% of a sample of college students reported committing child sexual abuse. Future research should examine the characteristics of men and women who sexually abuse children and do not get caught.

This study demonstrates how a meta-analysis can help clarify mixed results found in a particular body of research. Qualitative literature reviews report the mixed results and the authors draw their own conclusions. In contrast, meta-analyses quantitatively examine studies' results, taking methodological differences into account. The results of a meta-analysis indicate the strength of a relationship, rather than statistically significant differences.

There are few meta-analyses on the characteristics of sex offenders, and of sex research in general. Future meta-analyses can examine many other predictors of sexual coercion such as alcohol use, hostility toward women, or deviant sexual arousal. These meta-analyses would be useful in distinguishing date or acquaintance rapists from stranger rapists, and incest offenders from nonfamilial child molesters. I hope that this study will encourage other sex researchers to use meta-analytic procedures more often.

Table 1. Summary of Studies
                                  Offender   Nonoffender
Study                             samples      samples

 1. Bumby & Hansen (1997)           I, C        I, N
 2. Hudson & Ward (1997)            I, C        I
 3. Katz (1990)                     C           I, N
 4. Koralewski & Conger (1992)      N           N
 5. Lipton et al. (1987)            I           I
 6. Lisak & Ivan (1995)             N           N
 7. Malamuth & Brown (1994)         N           N
 8. Muehlenhard & Falcon (1990)     N           N
 9. Murphy et al. (1986)            N           N
10. Overholser & Beck (1986)        I, C        I, N
11. Segal & Marshall (1985)         I, C        I, N
12. Segal & Marshall (1986)         I, C        I, N
13. (Stahl & Sacco (1995)           I, C        I
14. Ward et al. (1996)              I, C        I

                                                        Number of
Study                             Measures used          effects

 1. Bumby & Hansen (1997)         FIS                       8
 2. Hudson & Ward (1997)          FIS                       8
 3. Katz (1990)                   SHI                       8
 4. Koralewski & Conger (1992)    SHI                       4
 5. Lipton et al. (1987)          TRAC                      2
 6. Lisak & Ivan (1995)           FAR                       1
 7. Malamuth & Brown (1994)       HDI, SDI                  2
 8. Muehlenhard & Falcon (1990)   SHI                       4
 9. Murphy et al. (1986)          HDI, SDI                  1
10. Overholser & Beck (1986)      HBCL, TBCL               30
11. Segal & Marshall (1985)       CAX, CSS, JAX, JSS,
                                  SAX, SHI, SSS            48
12. Segal & Marshall (1986)       DIS                      18
13. (Stahl & Sacco (1995)         ACR                       6
14. Ward et al. (1996)            RQG, RSQF                 8

Note. Offender Sample: C = child molesters incarcerated or in
treatment; I = rapists incarcerated; N = nonincarcerated
non-sex-offenders. Nonoffender sample: I = incarcerated
non-sex-offenders; N = nonincarcerated non-sex-offenders. Measures
used: ACR = affect categorization in a romantic scenario (Stahl &
Sacco, 1995); CAX = confederates' ratings of anxiety; CSS =
confederates' ratings of social skills; DIS = discrepancy between
predicted and actual self-efficacy in heterosocial situation (Segal
& Marshall, 1986); FAR = Facial Affect Recognition (Lisak & Ivan,
1995); FIS = Fear of Intimacy Scale (Descutner & Thelen, 1991); HBCL
= Heterosocial Skills Behavior Checklist (Barlow et al., 1977); HDI
= Hostility Discrimination Index (Murphy et al., 1986); JAX = judges'
ratings of anxiety; JSS = judges' ratings of social skill; RQF =
Relationship Questionnaire, Fearful attachment; RSQF = Relationship
Scales Questionnaire, Fearful attachment; SAX = self-ratings of
anxiety; SDI = Seduction Discrimination Index (Murphy et al., 1986);
SHI = Survey of Heterosexual Interactions (Twentyman, Boland, & McFall,
1981); SSS = self-ratings of social skills; TBCL = Timed Behavior
Checklist for Anxiety (Kern, 1982); TRAC = Test of Reading Affective
Cues (Lipton et al., 1987).

Table 2. Categorical Models of Effect Size Data

Comparison groups    k     d+        95% CI          r+

AS - AN             119   -0.36   -0.41 / -0.30   -.18 ***
  AR - AN            63   -0.23   -0.30 / -0.16   -.11 ***
  CM - AN            56   -0.54   -0.62 / -0.45   -.27 ***
    NR - NN           9   -0.36   -0.48 / -0.24   -.18 ***
    IR - IN          25   +0.05   -0.07 / +0.17   +.02
    IR - NN          29   -0.38   -0.51 / -0.26   -.19 ***
    CM - NN          30   -0.82   -0.94 / -0.69   -.38 ***
    CM - IN          26   -0.30   -0.41 / -0.19   -.15 ***

Comparison groups   [Q.sub.W]    [Q.sub.B](1)   [Q.sub.T]

AS - AN                                         332.43 ***
  AR - AN           168.39 ***   30.35 ***
  CM - AN           133.69 ***
    NR - NN          15.70 *     97.15 ***
    IR - IN          55.92 ***
    IR - NN          66.86 ***
    CM - NN          48.87 **
    CM - IN          47.92 ***

Note. IR = incarcerated rapists; IN = incarcerated non-sex-offenders;
NR = nonincarcerated rapists; NN = nonincarcerated non-sex-offenders;
AR = all rapists; AN = all non-sex-offenders; CM = child molesters.

* p < .05. ** p < .01. *** p < .001.

Table 3. Comparison Tests

Comparison groups      k     d+        95% CI        r+

All measures
  IR - NN              29   -0.38   -0.51 / -0.26   -.19
  IR - IN              25   +0.05   -0.07 / +0.17   +.02
Self-report measures
  IR - NN               7   -0.41   -0.65 / +0.18   -.20
  NR - NN               5   +0.07   -0.23 / +0.37   +.03
Performance measures
  IR - NN              22   -0.37   -0.52 / -0.22   -.18
  NR - NN               4   -0.45   -0.58 / -0.31   -.22
All measures
  CM - NN              30   -0.82   -0.94 / -0.69   -.38
  CM - IN              26   -0.30   -0.41 / -0.19   -.15

Comparison groups      [Q.sub.W]   [Q.sub.B](1)

All measures
  IR - NN              66.86 ***     19.72 ***
  IR - IN              55.92 ***
Self-report measures
  IR - NN               6.57          6.23 *
  NR - NN               1.51
Performance measures
  IR - NN              60.22 ***      0.54
  NR - NN               4.51
All measures
  CM - NN              48.87 *       36.90 ***
  CM - IN              47.92 ***

Note. IR = incarcerated rapists; IN = incarcerated non-sex-offenders;
NR = nonincarcerated rapists; NN = nonincarcerated non-sex-offenders;
AR = all rapists; AN = all non-sex-offenders; CM = child molesters.

* p < .05. ** p < .01. *** p < .001.

Table 4. The Heterosocial Competence of Sex Offenders
Compared With Nonincarcerated Non-Sex-Offenders

  Sex
offender
 group     k     d+        95% CI        r+      P     [Q.sub.W]

NR         12   -0.34   -0.46 / -0.22   -.17   <.001   18.25
IR         36   -0.38   -0.49 / -0.27   -.19   <.001   78.22 ***
CM         49   -0.79   -0.88 / -0.69   -.37   <.001   76.30 **

Note. IR = incarcerated rapists; NR = nonincarcerated rapists;
CM = child molesters.

* p < .05. ** p < .01. *** p < .001.

(1) Nonsignificant differences between groups can have a wide range of effect-size magnitudes. Entering all of these effect sizes as zero would be inaccurate and would decrease the true overall effect size. On the other hand, leaving these effect sizes out artificially inflates the true overall effect size. A "file drawer" analysis (Wolf, 1986) deals with the problem of artificial inflation by identifying how many null effects would have to be added to the meta-analysis to bring the overall effect size to .2, which Cohen (1977) has identified as a small effect size.

(2) Only one study (Stahl & Sacco, 1995) distinguished child molesters who victimized only girls from those who victimized only boys. Therefore, I could not complete a categorical analysis to compare homosexual child molesters with heterosexual child molesters.

REFERENCES

Abel, G. G., Blanchard, E. B., & Becket, J. V. (1976). Psychological treatment of rapists. In M. Walker & S. Brodsky (Eds.), Sexual assault: The victim and the rapist (pp. 99-115). Lexington, MA: Lexington Books.

Barlow, D. H., Abel, G. G., Blanchard, E. B., Bristow, A. R., & Young, L. D. (1977). A heterosocial skills behavior checklist for males. Behavior Therapy, 8, 229-239.

Bumby, K. M., & Hansen, D. J. (1997). Intimacy deficits, fear of intimacy, and loneliness among sexual offenders. Criminal Justice and Behavior, 24, 315-331.

Clark, L., & Lewis, D. (1977). Rape: The price of coercive sexuality. Toronto: Woman's Press.

Cohen, J. (1977). Statistical power analysis for the behavioral sciences (Rev. ed.). New York: Academic Press.

Descutner, C. J., & Thelen, M. H. (1991). Development and validation of a Fear-of-Intimacy Scale. Psychological Assessment, 3, 218-225.

Erel, O., & Burman, B. (1995). Interrelatedness of marital relations and parent-child relations: A meta-analytic review. Psychological Bulletin, 118, 108-132.

Finkelhor, D., & Lewis, I. A. (1988). An epidemiologic approach to the study of child molestation. Annals of the New York Academy of Sciences, 528, 64-78.

Freund, K. (1988). Courtship disorder: Is this hypothesis valid? Annals of the New York Academy of Sciences, 528, 172-182.

Freund, K., Scher, H., & Hucker, S. (1983). The courtship disorders. Archives of Sexual Behavior, 12, 369-379.

Fromuth, M. E., Burkhart, B. R., & Jones, C. W. (1991). Hidden child molestation: An investigation of adolescent perpetrators in a nonclinical sample. Journal of Interpersonal Violence, 6, 376-384.

Hall, G. C. N. (1995). The preliminary development of theory-based community treatment for sexual offenders. Professional Psychology: Research & Practice, 26, 478-483.

Hedges, L. V., & Olkin, I. (1985). Statistical methods for meta-analysis. Orlando, FL: Academic Press.

Hudson, S. M., & Ward, T. (1997). Intimacy, loneliness, and attachment style in sexual offenders. Journal of Interpersonal Violence, 12, 323-339.

Katz, R. C. (1990). Psychosocial adjustment in adolescent child molesters. Child Abuse and Neglect, 14, 567-575.

Kern, J. (1982). The comparative external and concurrent alidity of three role plays for assessing heterosocial performance. Behavior Therapy, 13, 666-680.

Koralewski, M. A., & Conger, J. C. (1992). The assessment of social skills among sexually coercive college males. The Journal of Sex Research, 29, 169-188.

Koss, M. P. (1988). Hidden rape: Sexual aggression and victimization in a national sample of students in higher education. In A. W. Burgess (Ed.), Rape and sexual assault II (pp. 3-25). New York: Garland.

Lalumiere, M. L., Chalmers, L. J., Quinsey, V. L., & Seto, M. C. (1996). A test of the mate deprivation hypothesis of sexual coercion. Ethology and Sociobiology, 17, 299-318.

Lipton, D. N., McDonel, E. C., & McFall, R. M. (1987). Heterosocial perception in rapists. Journal of Consulting and Clinical Psychology, 55, 17-21.

Lisak, D., & Ivan, C. (1995). Deficits in intimacy and empathy in sexually aggressive men. Journal of Interpersonal Violence, 10, 296-308.

Malamuth, N. M., & Brown, L. M. (1994). Sexually aggressive men's perceptions of women's communications: Testing three explanations. Journal of Personality and Social Psychology, 67, 699-712.

Marolla, J. A., & Scully, D. (1986). Attitudes toward women, violence, and rape: A comparison of convicted rapists and other felons. Deviant Behavior, 7, 337-355.

Marshall, W. L., Anderson, D., & Champagne, F. (1997). Self-esteem and its relationship to sexual offending. Psychology, Crime, and Law, 3, 81-106.

McFall, R. M. (1990). The enhancement of social skills: An information-processing analysis. In W. L. Marshall, D. R. Laws, & H. E. Barbaree (Eds.), Handbook of sexual assault: Issues, theories, and treatment of the offender (pp. 311-330). New York, NY: Plenum.

Muehlenhard, C. L,, & Falcon, P. L. (1990). Men's heterosocial skill and attitudes toward women as predictors of verbal sexual coercion and forceful rape. Sex Roles, 23, 241-260.

Murphy, W. D., Coleman, E. M., & Haynes, M. R. (1986). Factors related to coercive sexual behavior in a nonclinical sample of males. Violence and Victims, 1, 255-278.

O'Sullivan, L. F., Byers, E. S., & Finkelman, L. (1998). A comparison of male and female college students' experiences of sexual coercion. Psychology of Women Quarterly, 22, 177-195.

Overholser, J. C., & Beck, S. (1986). Multimethod assessment of rapists, child molesters, and three control groups on behavioral and psychological measures. Journal of Consulting and Clinical Psychology, 54, 682-687.

Rosenthal, R. (1979). The "file drawer" problem and the tolerance for null results. Psychological Bulletin, 86, 638-641.

Rosenthal, R., & Rubin, D. B. (1986). Meta-analytic procedures for combining studies with multiple effect sizes. Psychological Bulletin, 99, 400-406.

Russell, D. E. H. (1984). Sexual exploitation: Rape, child sexual abuse, and workplace harassment. Beverly Hills, CA: Sage.

Segal, Z. V., & Marshall, W. L. (1985). Heterosexual social skills in a population of rapists and child molesters. Journal of Consulting and Clinical Psychology, 53, 55-63.

Segal, Z. V., & Marshall, W. L. (1986). Discrepancies between self-efficacy predictions and actual performance in a population of rapists and child molesters. Cognitive Therapy and Research, 10, 363-375.

Spaccarelli, S., Bowden, B., Coatsworth, J. D., & Kim, S. (1997). Psychosocial correlates of male sexual aggression in a chronic delinquent sample. Criminal Justice and Behavior, 24, 71-95.

Stahl, S. S., & Sacco, W. P. (1995). Heterosocial perception in child molesters and rapists. Cognitive Therapy and Research, 19, 695-706.

Twentyman, C., Boland, T., & McFall, R. M. (1981). Heterosocial avoidance in college males. Behavior Modification, 5, 523-552.

Ward, T., Hudson, S. M., & Marshall, W. L. (1996). Attachment style in sex offenders: A preliminary study. The Journal of Sex Research, 33, 17-26.

Ward, T., McCormack, J., Hudson, S. M., & Polaschek, D. (1997). Rape: Assessment and treatment. In D. R. Laws & W. O'Donohue (Eds.), Sexual deviance: Theory, assessment, and treatment. New York: Guilford.

Wolf, E M. (1986). Meta-analysis: Quantitative methods for research synthesis. Newbury Park, CA: Sage.

Manuscript accepted January 15, 2003

Michael T. Dreznick is now at the Department of Psychology, Randolph-Macon Woman's College.

The author would like to thank Caroline K. Waterman, Donn Byrne, and Glenn Sanders for their helpful advice.

Address correspondence to Michael T. Dreznick, Department of Psychology, Randolph-Macon Woman's College, 2500 Rivermont Ave., Lynchburg, VA 24503; e-mail: [e-mail].

Michael T. Dreznick

University at Albany, State University of New York