A while back, I came across this study which claims to "vindicate" feminist criticisms of CTS (Conflict Tactics Scale) surveys on domestic violence, which usually show gender symmetry in DV. Here's a link to the study. The author seems very convinced that CTS surveys might not be accurate in their findings of gender symmetry (citing lots of feminist researchers like Kimmel and Johnson), and the author's aim is to investigate misreporting in CTS surveys and how they might affect the results.

I've read the study, and I have criticisms and want to know if they hold up. Here is the link (hope the link didn't get fucked up this time). I would recommend taking a look at the study before reading the rest of my comment.

https://www.researchgate.net/publication/279196059\_Over-Reporting\_Intimate\_Partner\_Violence\_in\_Australian\_Survey\_Research

The study asks respondents on acts perpetrated and sustained in their current and past intimate relationships, retaining the original CTS preamble and many of the original CTS items in a slightly modified form. Then it asks them about the frequency of these acts, then asks them about the context of the violence, with eight response possibilities addressing the degree of injury, the intentionally of the act and other pertinent event details.

So let's start with the criticisms I have.

Firstly this study uses un-objective criteria to determine whether a person is over-reporting or not. "The two responses relevant for the over-reporting measure of current interest are ‘accident’ and ‘joking, playful or humorous—no one took it seriously’. If the respondent endorsed one of the CTS items and subsequently indicated that either (or both) of these two event details were applicable, the endorsement was considered an over-report."

I would just like to note that we're less likely to take women's violence against men seriously than the other way around, so using perceptions of violence might be a very skewed way of determining over or under reporting.

This study noted: "We explored sex differences in perception of male versus female victims, as well as in one’s willingness to identify as a victim, to minimize or conceal victimization, and to seek help for perceived victimization in a sample of 166 (89 female, 77 male) undergraduates. Results indicated that participants held more negative attitudes toward male versus female victims. Males were less likely than females to consider hypothetical aggressive acts perpetrated against them as abusive."

If males are less likely than females to consider equivalent acts as abusive, then of course if you ask a question including the wording "no one took it seriously" clearly this will inflate the amount of female-on-male assaults which will be considered as an over-report compared to male-on-female, even if the hypothetical acts are the very same.

More than this, it's potentially possible depending on a number of factors, including survey design and respondent behaviour, that the inclusion of these questions about over-reporting in this study might actually motivate respondents to report what they consider as benign incidents so they can subsequently check off the "accident" or "no one took it seriously" question. They might not similarly do so in a normal CTS survey without these questions. The inclusion of these questions about benign incidents might make it crystal clear to the respondent that the researchers conducting the study want the respondents to provide information about benign incidents. Anecdotally speaking, I have taken online questionnaires before where I have changed my previous responses depending on what the later questions are. Thus, it is unclear whether the findings of over-reporting in this study can be so well extrapolated to other CTS surveys that do not include these questions.

But the problems with the study don't stop there. The methodology isn't just questionable, the conclusions drawn in the study seem to be misleading.

The study comes to the conclusion that "Males were significantly more likely to over-report victimizations than perpetrations. ... This can be seen in alternative form by comparing data rows 3 and 6 in Table 2. Here, we see that the rate of female over-reporting of victimization is 3.1 per cent versus 3.3 per cent for perpetration, while the rate of male over-reporting of victimization is 8.3 per cent versus 2.6 per cent for perpetration. In other words, males and females had roughly similar (non-significant) rates of perpetration-over-reporting, while males had much higher rates of victimization-over-reporting than did the females." Exactly what do the numbers here represent?

Table 2 lays out the numbers in the study.

Males Females
Victimization
Respondents reporting one or more victimisations in either current or past relationship 21.8% 14.1%
Victimization reports that are over-reports 23.4% 12.9%
Respondents over-reporting one or more IPV victimizations in either current or past relationship 8.3% 3.1%
Perpetration
Respondents reporting one or more perpetrations in either current or past relationship 7.1% 13.6%
Perpetration reports that are over-reports 47.6% 17.0%
Respondents over-reporting one or more IPV perpetrations in either current or past relationship 2.6% 3.3%

"The first data row indicates that 21.8 per cent of males reported being victims of some type of IPV during the course of the relationship. This is a far greater percentage than the 14.1 per cent of females making these reports, but consistent with prior research conducted in several countries, which indicates that males typically report victimizations at a greater rate than females."

"The second data row ... shows that almost one-quarter (23.4 per cent) of the victimizations reported by the males were subsequently classified as over-reports. Females also had substantial rates of over-reporting, but those rates were only 12.9 per cent, nearly half as much as the male over-reporting rate. Again, a chi- square test indicates that this difference is statistically significant at the p < 0.01 level. Note that the figures reported on this row do not strictly mean that 23.4 and 12.9 per cent of the reports indicated on the prior row for males and females, respectively, are over-reports because the figures on this row are based upon all victimization reports from all participants (up to seven possible victimization reports per relationship) regardless of whether each participant reported only one victimization or several."

"The third data row indicates over-reporting rates in an alternative manner. While the prior data row provides over-reporting percentages based upon the number of reports as the denominator, this third row provides over-reporting percentages based upon the total number of participants in each gender category as the denominator. Here, we also see that the male participants were over 2.5 times more likely to over-report one or more IPV events than were the female participants. The chi-square test indicates that this difference is statistically significant at the p < 0.01 level."

"Perpetration rates are presented in data rows three through six. Row 3 indicates that 7.1 per cent of the males reported perpetrating at least one of the seven types of IPV, while 13.6 per cent of the females reported perpetration. Again, this difference is statistically significant at the p < 0.01 level."

"The largest of the gender differences are shown in the fifth data row. Here, we see that of the male perpetrations reported on the prior data row, 47.6 per cent were subsequently classified as over-reports, while only 17 per cent of the female endorsements were so classified. As before, this difference is statistically significant at the p < 0.01 level."

"The only gender difference not statistically significant at the 0.05 level or better is shown on the sixth data row. Here, we see that 2.6 per cent of the male versus 3.3 per cent of the female participants in the sample had one or more perpetration over-reports."

So this study is relying on row 3 and 6 in order to draw these conclusions, and what they essentially represent is: Of the ENTIRE samples of men and women - not of those who reported perpetration, mind you - 2.6% of men and 3.3% of women over-reported perpetration. And of the ENTIRE samples of men and women - not of those who reported victimisation - 8.3% of men and 3.1% of women over-reported victimisation.

Now, using these numbers makes no sense. This is like trying to find out how likely is a person to fall off a horse, right? You have a total sample of X people, only Y of whom have ever been on a horse, and Z fell off. We know how likely a person in the sub-sample Y is to fall off a horse, that's Z/Y, but we don't know anything about the subsample of (X-Y) because they have never been on a horse and we don't know how good they are at riding. And this paper says, well, we found out what Z/X is. Well, Z/X represents both people who have been tested in horse-falling, and people who have never even been on a horse.

If you want to determine likelihood of over-reporting, the numbers you need to use are the over-reports as a percentage of those who reported perpetration and victimisation respectively, not as a percentage of the entire sample. 

Fewer men than women reported perpetration, with 7.1% and 13.6% respectively reporting perpetration. More men than women reported victimisation, with 21.8% and 14.1% respectively reporting victimisation.

So, we can combine the numbers to determine likelihood of over-reporting.

2.6/7.1 = 37% of men who reported perpetration over-reported perpetration. 3.3/13.6 = 24.2% of women who reported perpetration over-reported perpetration.

8.3/21.8 = 38% of men who reported victimisation over-reported victimisation. 3.1/14.1 = 22% of women who reported victimisation over-reported victimisation.

So. It is not the case that males and females had roughly similar rates of perpetration over-reporting when you express it as a percentage of those that reported one or more acts of perpetration, instead of as a percentage of the entire sample. The only reason why the amount of men and women who over-report perpetration as a proportion of the entire sample look similar is because men as a whole report perpetration much, much less. Men's reports of perpetration are more likely to be over-reported, but they report perpetration at lower rates in the first place.

CTS studies already show that women perpetrate domestic abuse at higher rates than men. If we take what this study found at face value and take it into account, the rates of female perpetration are going to be even higher relative to male perpetration than previously thought.

What do the rates look like after the people who over-reported one or more acts are taken into account? Here:

7.1 - 2.6 = 4.5%. After subtracting off the men who overreported, 4.5% of men perpetrated IPV.

13.6 - 3.3 = 10.3%. After subtracting off the women who overreported, 10.3% of women perpetrated IPV.

21.8 - 8.3 = 13.8%. After subtracting off the men who overreported, 13.8% of men were victimised.

14.1 - 3.1 = 11%. After subtracting off the women who overreported, 11% of women were victimised.

However, as far as I can tell in this study one can't make the assumption that everyone who over-reported an instance of perpetration or victimisation hasn't perpetrated or hasn't been victimised. They could have reported more than one act of perpetration or victimisation, and only over-reported one, for example. So people who over-reported an instance of perpetration can still be a perpetrator, and people who over-reported an instance of victimisation can still be a victim.

But even if we try to make things look as favourable for women as possible, only selectively deducting off female over-reporting of perpetrations and male over-reporting of victimisations, 7.1% of men and 10.3% of women perpetrate IPV. 13.8% of men and 14.1% of women are victimised by IPV. Even under these extreme assumptions made to make women look as good as possible, the perpetration figures still show that females perpetrate more whereas the victimisation figures show gender symmetry.