First of all, thanks for making this extremely detailed post that addresses a tough issue in a diligent fashion. I especially admire your evenhanded tone.
I do have two points to make:
Firstly, when we discuss impact the effectiveness of an intervention, we are usually discussing things which effect huge numbers of people—perhaps all people who will ever exist—or interventions which effect hundreds of thousands of people at a very low per-person cost, like deworming. Here you’re discussing this for an intervention that targets just EAs, which is a very small number of people. There may be ~13,000 people in the facebook group, but the vast majority of them have no in-person interactions with the EA community. As such, the interventions you mention would have to be very efficacious at reducing sexual assault within the community to be competitive. Unless I’m reading this wrong, and you were more just suggesting these would be good things to do, though maybe not candidates for the best thing?
Secondly, I think Scott has a good criticism of this statistic:
What percentage of rape accusations are false? According to DiCanio, the researchers and prosecutors generally agree on a number somewhere in the range of 2% to 10%.
He suggests this is reasonable for a lower-bound, but the upper bound on false reporting is significantly higher:
Attempts to use this methodology return varying results. Lisak lists seven studies he considers credible, which find false accusation rates of 2.1%, 2.5%, 3.0%, 5.9%, 6.8%, 8.3%, 10.3%, 10.9%. The two with 10%+ mysteriously go missing and thus we get the commonly quoted number of “two to eight percent”, which is repeated by sources as diverse as Alas, A Blog, Slate, and Wikipedia (Straight Statistics keeps the original 2% – 10% number)
Feminists make one true and important critique of these numbers – sometimes real victims, in the depths of stress we can’t even imagine, do strange things and get their story hopelessly garbled. Or they suddenly lose their nerve and don’t want to continue the legal process and tell the police they were making it up in order to drop the case as quickly as possible. All of these would go down as “false allegations” under the “victim has to admit she was lying or contradict herself” criteria. No doubt this does happen.
But the opposite critique seems much stronger: that some false accusers manage tell their story without contradicting themselves, and without changing their mind and admit they were lying. We’re not talking about making it all the way through a trial – the majority of reported rapes get quietly dropped by the police for one reason or another and never make it that far. Although keeping your story halfway straight is probably harder than it sounds sitting in an armchair without any cops grilling me, it seems very easy to imagine that most false accusers manage this task, especially since they may worry that admitting their duplicity will lead to some punishment.
The research community defines false accusations as those that can be proven false beyond a reasonable doubt, and all others as true. Yet many – maybe most – false accusations are not provably false and so will not be included.
...
What is an upper bound on the number of false rape accusations? Researchers tend to find that police estimate 20%-40% of the rape accusations they get to be “unfounded”, (for example Philadelphia Police 1968, Chambers and Millar 1983, Grace et al 1992, Jordan 2004, Gregory and Lees 1996, etc, etc). Many scholars critique the police’s judgment, suggesting many police officers automatically dismiss anyone who doesn’t fit their profile of a “typical rape victim”. A police-based study that took pains to avoid this failure mode by investigating all cases very aggressively (Kanin 1994) was criticized for what I think are ideological reasons – they primarily seemed to amount to the worry that the aggressive investigations stigmatized rape victims, which would make them so flustered that they would falsely recant. Certainly possible. On the other hand, if you dismiss studies for not investigating thoroughly enough and for investigating thoroughly, there will never be any study you can’t dismiss. So while not necessarily endorsing Kanin and the similar studies in this range, I think they make a useful “not provably true” upper bound to contrast with the “near-provably false” lower bound of 2%-10%.
The second point is irrelevant—what statistic is changed by the prevalence of false rape accusations? The Lisak and Miller study cited for the 6% figure do a survey of self-reports among men on campus.
The post explicitly talks about false rape accusations, so his second point does not seem irrelevant to me? (Although it is clearly irrelevant to the 6% figure).
The ‘Strike a balance between dismissing accusations and witch-hunting people’ is about how to act given that accusations have some small (but non-negligible) chance of being false. If we instead learned that the true rate differed strongly from this, it seems reasonable that the advice would also change. (E.g. if it turned out that there were never any false accusations, we could act much more strongly on the basis of accusations).
I also think generally if a piece of writing states a fact you think is false, it is reasonable and should be encouraged to bring it up in the comments, even if it is not central to the argument’s conclusion.
There are a lot of ways in which sexual violence has an impact on effective altruism, so reducing sexual violence will help us reach our effective altruism goals in various different ways. Because it will help us do more effective altruism, and the cost-benefit ratio looks good, I believe that gives it a lot of potential to be an effective altruism cause. It seems like you may not have read the entire impact section. Here is a table of contents for the impact section:
Impact
Estimating the number of sexually violent people.
Why we should not assume that effective altruism repels sex offenders:
About 6% of men are rapists and an unknown percentage of women.
A rough estimate of rapists in EA:
Sexual violence reduction as a life saver:
Sexual violence reduction as suffering reduction:
Sexual violence reduction for diversity and disadvantage reduction:
Comparing sexual violence rates by gender:
Greatly multiplied risk to women due to the gender ratio in EA:
Gay and bisexual people have around twice the sexual violence risk:
List of specific disadvantages that EA women, bisexuals and homosexuals face:
Potential of sexual violence reduction to prevent productivity loss:
The low estimate:
The high estimate:
Sexual violence reduction as part of movement building:
The male sex offenders studied are shockingly prolific:
I read the whole post, and though I saw a lot of good points about why sexual violence is bad, I didn’t see much about how efficacious the interventions you suggested were. It might be the case that things which increase EA productivity in a cost-effective manner are EA causes—though it seems a little strange to consider something like “getting enough sleep” to be an EA cause—but I don’t think you’ve really made the case that these interventions do have a particularly high effectiveness.
“There are a lot of options that have a chance to succeed. The impact could be many times greater than the effort it takes to use the options explored herein. Testing is needed to determine the effectiveness of the options. Given the human rights concerns and the potential for a large productivity impact, testing options could turn out to be very worthwhile.”—from my conclusion section.
This is my honest conclusion, which I made as accurate as possible. We do not know how effective all the methods are, but it looks like it’s worth testing them to find out.
Having this information is a valid kind of progress.
First of all, thanks for making this extremely detailed post that addresses a tough issue in a diligent fashion. I especially admire your evenhanded tone.
I do have two points to make:
Firstly, when we discuss impact the effectiveness of an intervention, we are usually discussing things which effect huge numbers of people—perhaps all people who will ever exist—or interventions which effect hundreds of thousands of people at a very low per-person cost, like deworming. Here you’re discussing this for an intervention that targets just EAs, which is a very small number of people. There may be ~13,000 people in the facebook group, but the vast majority of them have no in-person interactions with the EA community. As such, the interventions you mention would have to be very efficacious at reducing sexual assault within the community to be competitive. Unless I’m reading this wrong, and you were more just suggesting these would be good things to do, though maybe not candidates for the best thing?
Secondly, I think Scott has a good criticism of this statistic:
He suggests this is reasonable for a lower-bound, but the upper bound on false reporting is significantly higher:
source
The second point is irrelevant—what statistic is changed by the prevalence of false rape accusations? The Lisak and Miller study cited for the 6% figure do a survey of self-reports among men on campus.
The post explicitly talks about false rape accusations, so his second point does not seem irrelevant to me? (Although it is clearly irrelevant to the 6% figure).
It mentions them, but does it make any points based on the assumption that there are too few of them?
The ‘Strike a balance between dismissing accusations and witch-hunting people’ is about how to act given that accusations have some small (but non-negligible) chance of being false. If we instead learned that the true rate differed strongly from this, it seems reasonable that the advice would also change. (E.g. if it turned out that there were never any false accusations, we could act much more strongly on the basis of accusations).
I also think generally if a piece of writing states a fact you think is false, it is reasonable and should be encouraged to bring it up in the comments, even if it is not central to the argument’s conclusion.
There are a lot of ways in which sexual violence has an impact on effective altruism, so reducing sexual violence will help us reach our effective altruism goals in various different ways. Because it will help us do more effective altruism, and the cost-benefit ratio looks good, I believe that gives it a lot of potential to be an effective altruism cause. It seems like you may not have read the entire impact section. Here is a table of contents for the impact section:
Impact
Estimating the number of sexually violent people.
Why we should not assume that effective altruism repels sex offenders:
About 6% of men are rapists and an unknown percentage of women.
A rough estimate of rapists in EA:
Sexual violence reduction as a life saver:
Sexual violence reduction as suffering reduction:
Sexual violence reduction for diversity and disadvantage reduction:
Comparing sexual violence rates by gender:
Greatly multiplied risk to women due to the gender ratio in EA:
Gay and bisexual people have around twice the sexual violence risk:
List of specific disadvantages that EA women, bisexuals and homosexuals face:
Potential of sexual violence reduction to prevent productivity loss:
The low estimate:
The high estimate:
Sexual violence reduction as part of movement building:
The male sex offenders studied are shockingly prolific:
Sex offenders increase turnover in workplaces:
Sexual violence reduction for lawsuit prevention:
I read the whole post, and though I saw a lot of good points about why sexual violence is bad, I didn’t see much about how efficacious the interventions you suggested were. It might be the case that things which increase EA productivity in a cost-effective manner are EA causes—though it seems a little strange to consider something like “getting enough sleep” to be an EA cause—but I don’t think you’ve really made the case that these interventions do have a particularly high effectiveness.
“There are a lot of options that have a chance to succeed. The impact could be many times greater than the effort it takes to use the options explored herein. Testing is needed to determine the effectiveness of the options. Given the human rights concerns and the potential for a large productivity impact, testing options could turn out to be very worthwhile.”—from my conclusion section.
This is my honest conclusion, which I made as accurate as possible. We do not know how effective all the methods are, but it looks like it’s worth testing them to find out.
Having this information is a valid kind of progress.