https://projects.fivethirtyeight.com/sexual-assault-victims/
Among various numbers from the National Crime Victimization Survey:
84% women, 16% men
44% income < $25,000 (versus 22% of households)
59% in urban areas, 28% in suburban areas, 13% in rural areas
33% in intimate relationship, 39% acquaintance, 19% did not know offender
77% not reported to police, 23% reported to police
11% received assistance from victim services agency (other than law enforcement)
The NCVS may be the best data we have for sexual assault victims. But it’s rather poor data, as is noted in the 538 article. For one thing, the interviews of the people are conducted in full view of other family members, which might well induce reticence in many victims. Moreover, as the article notes, the interview techniques most likely don’t elicit descriptions of all or even most or possibly even the majority of sexual assaults experienced by the interviewees.
The NCVS is the best data on crime victimization we have, based on sample size and sample methodology. While the initial interview is done in person, later interviews are to be done by phone whenever possible.
The National Academy of Science, National Academy of Engineering, Institute of Medicine and National Research Council put out an entire book on why the NVCS is inaccurate in measuring sexual assault. They conclude it undercounts sexual assault.
The book can be downloaded here, for free: https://www.nap.edu/download/18605
^It’s far from clear why the panel attached so much weight to the other surveys, given their own statements:
It’s far from clear to me why you say they attach so much weight to other surveys, when they don’t. They say the NVCS is inaccurate and undercounts rape.
If a survey has a high response rate and a large sample size, but gets incorrect information from the people it samples, the large sample size isn’t doing any good.
Regarding the issues identified in #3, issue 1 is just as likely to overestimate as underestimate, there’s not much credible evidence issue 2 affects survey results in a meaningful way, and the NCVS was revamped to have less ambiguous and more specific questions, so issue 4 has been fixed. To the extent issue 3 is important, it should show up in intra-survey differences(in person vs telephone interviews for the NCVS) and (in person alone vs in person with others present for the NCVS).
Is there a point to this dispute? Does it affect any of the relative percentages 538 was highlighting? (I thought this was an especially weak 538 piece, by the way. It just throws some of the relative percentages out there with little critical discussion.)
Two things caught my eye:
(1) The very significant disparity for sexual assaults between victims in the lowest income quintile and all other victims. Is that because (a) poor people are more likely to suffer sexual assault, (b) young people – who earn less – are more likely to suffer sexual assault, or © poor and/or young people are more likely to call what happened to them a crime?
(2) Even given the factors leading people to think this survey undercounts sexual assaults, in large part by emphasizing their criminality, still less than 25% of these sexual assaults were reported to police.
In other words, you’re saying that in sampling, you’re just as likely to see fewer than expected of your event as more than expected. That’s not precisely true for rare events.
Moreover, the problem is that with rare events, you don’t always detect them when you see them. So for example if a doctor rarely sees a disease, they might not diagnose it when the patient presents with it. Similarly, if the interviewers don’t know how to elicit the description of a sexual assault from a woman who in truth has been assaulted, they will undercount the incidence of sexual assault.
Why might the interviewer not count the woman who was, by law, assaulted? Maybe she doesn’t want to talk about her rape-- many women don’t. Maybe she doesn’t think of it as a sexual assault, even though by law, she was the victim of a sexual assault. Maybe her assaulter is listening in on the phone call. Maybe she erroneously thinks that her report of her assault will be passed on to the police, and she does not want to report her assault to the police.
This is ridiculous. We know that different surveys get order-of-magnitude different answers, so obviously the context of “crime” DOES affect survey results.