This piece was first written on October 22, 2017, and first appeared on my website on April 30, 2026.
This essay is the second in a series examining sexual harassment and assault.
The first essay can be read here, which also includes more information about the study participants.
Sexual harassment and assault occur far too often. In this post, I explore what the predictors are of being a perpetrator and being a victim. In particular, I use linear regression to predict whether each person is a victim and/or perpetrator of sexual harassment based on how many (out of 6) forms of sexual harassment/assault a person said they had experienced as a victim (their 0-6 “victimization score”) and as a perpetrator (their 0-6 “perpetration score”).
Warning: this post contains extensive discussion of sexual harassment and sexual assault.
—
PREDICTORS OF BEING A PERPETRATOR OF SEXUAL HARASSMENT OR ASSAULT
Overview: risk-taking, being unamicable (i.e., the sort of person who contradicts others), and having been victimized oneself all seem to be predictors of having been a perpetrator of sexual harassment or sexual assault. However, all these relationships are fairly weak.
One can only speculate about why these three variables, in particular, are associated with perpetration. My speculative guesses are:
Risk-taking: perhaps this is linked to being a perpetrator simply because there is risk taken on by harassing or assaulting (such as risk of social ostracism or arrest). Risk-taking is also known to be correlated with some personality disorders, which may be a partial explanation as well.
Unamicability: perhaps this is linked to being a perpetrator because it relates to contradicting other people’s opinions, which happens in harassment/assault contexts (e.g. arguing against a “no”). The trait of contradicting others is also correlated with lower compassion, which may be another factor (as can be seen here on our site, PersonalityMap, which maps over 1 million human correlations).
Prior victimization: perhaps this is linked to being a perpetrator because certain contexts (e.g., where harassment and assault are more common) breed both victimization and perpetration simultaneously, or because prior victimization makes perpetration feel more acceptable to the perpetrator (e.g., a possible reflection of the phrase “hurt people hurt people”)
— details —
Looking in detail at 23 variables collected in the study ONE at a time (that is, not taking each of the others into account), the self-reported personality traits and demographic characteristics most correlated with a person’s “perpetration score” involved being:
CORRELATIONS WITH PERPETRATION (all based on self-reported information)
1. risk taking (r=0.25)
2. sex focused (r=0.19)
3. unamicable i.e., contradicting others (r=0.19)
4. promiscuous (r=0.17)
5. improvisational (r=0.15)
6. unusual or non-ordinary (r=0.15)
7. a victim of harassment/assault i.e. victimization score (r=0.14)
8. unforgiving (r=0.14)
9. selfish (r=0.13)
10. male (r=0.12)
11. uncompassionate (r=0.08)
12. sexually attractive (r=0.08)
13. more educated (r=0.08)
Each of these correlations above is statistically significantly different from 0 at roughly p<0.05, but none of them is particularly strong (i.e., even the largest correlation r=0.25 is quite a modest correlation).
Potentially more interesting is to look at a linear regression that tries to predict each person’s perpetration score from the 23 other demographic and personality variables.
Overall, the model captured only about 13% of the variance in people’s perpetration scores (i.e., adjusted R^2=0.13), which is fairly weak. In other words, it is not able to predict who is a perpetrator with reliability.
Here are the variables that came up strongest in the regression. The standardized regression coefficients (i.e., the strength of effect once each variable has been normalized to have the same mean and standard deviation as the others) are shown in parentheses.
REGRESSION COEFFICIENTS PREDICTING PERPETRATION
1. a victim of harassment/assault i.e., victimization score (0.205)
2. risk taking (0.161)
3. social conservativeness (0.149)
4. unamicable/contradicting others (0.120)
5. education (0.106)
6. emotionally aware (0.103)
7. non complexity seeking (0.101)
8. age (0.085)
All the variables shown are statistically significant at p<0.05 (the 15 other variables used in the regression are not shown). The way to interpret these variables is that each of these appears to add some additional ability to predict who will be a perpetrator, even when you know the other variables. Or, put another way, the regression coefficients are an indicator of how much each of these variables matters in predicting who is a perpetrator once you’ve controlled for all the other variables in the model. Specifically, the coefficient says how many more “yes” questions a person tends to give on the 6 sexual harassment/assault perpetration questions as the variable in question increased by 1 standard deviation, on average.
A few things to note:
Being male becomes no longer statistically significant (p=0.143, coef=0.080) once you account for these other variables, which surprised me quite a lot. In other words, it was significantly correlated on its own, but doesn’t come up as particularly relevant in the model with all the variables. We know that males commit sexual harassment and assault a lot more often than females. This result suggests that some part of the reason that males commit sexual harassment and assault more than females may be due to their higher risk-taking or unamicable (or by something related to those factors).
Age appearing here may not mean what you expect; it could just be due to more years alive, meaning more opportunities to be a perpetrator (rather than someone who is older being more likely to be a perpetrator due to age itself)
Having been the victim of sexual harassment/assault is found here to be predictive of having been a perpetrator. We can’t, from this data, tell what the causality for this looks like. It could be that there is direct causation (i.e., victimization causally increases the chance of future perpetration), but it also could be that there is indirect causation through some other variables (e.g., if in your social world sexual harassment and assault are common, that might increase your chance both of becoming a victim and of becoming a perpetrator).
Now, onto the other major question we’re exploring here.
WHAT PREDICTS BEING A VICTIM OF SEXUAL HARASSMENT/ASSAULT?
Overview: variables associated with being a victim of sexual harassment or assault included being female, worried, a prior perpetrator of sexual harassment/assault, unusual (i.e., self-identified as non-ordinary), compassionate, less educated, improvisational, and unselfish.
One can only speculate about why these variables are associated with victimization. My speculative guesses for reasons for some of the associations are:
Worried: perhaps this is associated with being a victim because having been a victim of harassment/assault or being currently surrounded by risks of it can substantially raise a person’s anxiety level.
Unusual: perhaps because “not fitting in” or “not being like others” may make a person a more noticeable or an easier target of harassment/assault
Compassionate/unselfish: perhaps because more compassionate and unselfish people may be taken advantage of by would-be harassers/assailants or seem like easier targets (e.g., a perpetrator may guilt the soon-to-be victim into spending time with them)
— details —
Looking at in detail at the 23 variables one at a time (that is, not taking each of the others into account), the self-reported personality traits and demographic characteristics most correlated with a person’s “victimization score” involved being:
CORRELATIONS WITH VICTIMIZATION
1. female (r=0.42)
2. worried (r=0.26)
3. thin-skinned (r=0.22)
4. compassionate (r=0.19)
5. unselfish (r=.16)
6. a perpetrator of harassment/assault, i.e., perpetration score (r=0.14)
7. unusual or non-ordinary (r=0.13)
8. warm (r=0.13)
9. less educated (r=0.10)
10. emotionally aware (r=0.10)
11. unamicable (r=0.08)
12. improvisational (r=0.08)
All of these correlations are statistically significantly different from 0 at about p<0.05. Though all are quite small correlations, except being female, which has a moderately strong correlation.
We can also again look at a linear regression, this time trying to predict each person’s victimization score from the 23 demographic and personality variables.
This model overall is more accurate than our previous one for predicting perpetration, capturing about 32% of the variance in victimization scores (i.e., adjusted R^2=0.32) rather than only 13%, though this is still not an especially accurate model.
Here are the self-reported personality traits and demographic variables that come up in the regression as being statistically significant at the p<0.05 level, along with the standardized coefficients (i.e., strength of effect).
REGRESSION COEFFICIENTS PREDICTING VICTIMIZATION
1. female (0.443)
2. worried (0.173)
3. a perpetrator of harassment/assault i.e., perpetration score (0.160)
4. unusual or non-ordinary (0.143)
5. socially liberal (0.141)
6. economically conservative (0.123)
7. compassionate (0.115)
8. less educated (0.108)
9. improvisational (0.106)
10. promiscuous (0.104)
11. complexity seeking (0.099)
12. unselfish (0.090)
The interpretation of this is that the regression coefficients are an indicator of how much each of these variables matters in predicting who is a victim once you’ve controlled for all the other variables in the model.
A few things to note:
Being worried is the second strongest predictor here, but keep in mind it’s unclear which way the causality goes. It could be that being a worried or anxious person makes someone more of a target, but it also could be that those who have been victims become more worried (as would make sense), or it could be that some other variables cause both increased victimization and increased worry (such as growing up in a non-supportive environment).
It’s weird to see that social liberalness is associated with more victimization, but economic conservatives as well. I have no idea what to make of that.
—
APPENDIX
If you’re interested, here is the full set of variables analyzed here:
Attributes
1. age
2. education level
3. household income
4. self-identified gender
5. economic liberalness/conservativeness
6. social liberalness/conservativeness
7. self-rated sexual attractiveness
8. self-rated sexiness of clothing
Self-rated personality traits:
9. promiscuous
10. amicable (i.e. non-contradictory of others)
11. compassionate
12. complexity seeking
13. emotionally aware
14. forgiving
15. improvisational
16. risk averse
17. sex focussed
18. thick skinned
19. unselfish
20. unusual
21. unworried
22. warm
Dependent variables
23. victimization score (0-6 counting yes responses to which forms of sexual harassment/assault they have been a victim of)
24. perpetration score (0-6 counting yes responses to which forms of sexual harassment/assault they have been a perpetrator of)
—
In the next post, I will explore what perpetrators said about why they harassed or assaulted others.
Comments