ETA: this post was pretty much refuted by comments below.

I've noticed a situation several times that I think deserves attention.

Somebody goes around saying they've been the victim of mistreatment. But they seem mentally ill. Whether or not you know of a diagnosis, they seem "off" somehow -- highly agitated, making social faux pas, telling stories that don't quite add up. So people are very suspicious about whether their allegations are true.

Is this rational?

In general, someone who seems less trustworthy should be believed less. And, yes, mentally ill people are more likely to be delusional or exaggerating. But they are also more likely to actually be victims of crimes than the general population.

40% of women in the UK with severe mental illness are victims of rape or attempted rape.

People with severe mental illness are 6x as likely as the general population to have recently experienced sexual violence.

30% of mentally ill adults in an American study had been victims of violent crime in the previous six months.

Mentally ill adults in Sweden are 5x more likely than the general population to be murdered.

More than 25% of severely mentally ill Americans have been the victims of a violent crime in the last year, 4x the rate of the general population.

30-33% of psychiatric patients have been victims of domestic violence.

Someone being mentally ill is evidence for, not against, their being victims of a crime. And the base rates of violent crime are pretty high, so all things being equal, "someone attacked me" is not an extraordinary claim. Even when someone seems crazy and has made a lot of claims you don't believe, it can be reasonable to believe their claims of crime victimization. Don't fall into the horns effect.

New Comment
43 comments, sorted by Click to highlight new comments since:

The question isn't P(ever abused|mentally ill), the question is P(true accusation|mentally ill). Maybe 40% of mentally ill women have been victims of at least 1 attempted rape at some point in their lives, but if they go around making dozens of rape accusations, while non-mentally-ill women hardly ever do, then a mentally ill accuser is strong evidence that a specific accusation is probably false. None of the statistics you quote bear on this or prove your claim. (One might say you spend all your effort establishing a denominator but forget the numerator.)

Don't fall into the horns effect.

The horns effect is real because everything is correlated.

Thanks; I think I was just wrong here, I didn't think of that.

I think it might be good to update the title of the post with a [Edit: Updated] tag or something.

Might also be useful for the disclaimer you added to the top to include a more complete explanation of your current take on the situation (although maybe a bit more work, and depends on your what your actual current take is).

AFAICT, Ben's comment was a decent synthesis of the point you were originally intending to make and gwern's counterpoint. While I don't think the point as originally worded was correct, it was still pointing at a fact that I don't think most people really have fully integrated into their model, and that seemed like something important enough that just saying 'refuted' didn't quite seem right either.

Questions about priors:

Are you assessing that the rate of false accusations from an individual is correlated positively, negatively, or not at all with the rate of actual rapes experienced by those individuals?

Do you model the probability of an individual experiencing abuse as random, negatively correllated with prior abuse, or positively correllated with prior abuse?

Does adjusting these priors affect the assessment you made above?

Relevant: section three paragraph 7 on serial abuse victims. There's probably more to say, but I think that some combinations of answers to the above questions lend support to the OP.

Likewise, just because an accusation of abuse is true, doesn't mean we will gain anything by believing it / defending it. Sometimes it's actually to our advantage to let someone be abused, if the abuser can more consistently reward us than the abused.

Just to clarify: am I being downvoted for being factually wrong, or for being uncomfortable?

It's worse than "being uncomfortable." The denotation of your comment isn't incorrect, but it degrades the always fragile common knowledge that we will coordinate to resist abusers. If it's already the case that "we" gain advantages by openly letting abusers run unchecked, then those of "us" who can should consider abandoning being part of this "we" and attempting to join healthier social groups.

Further-semi-aside: "common knowledge that we will coordinate to resist abusers" is actively bad and dangerous to victims if it isn't true. If we won't coordinate to resist abusers, making that fact (/ a model of when we will or won't) common knowledge is doing good in the short run by not creating a false sense of security, and in the long run by allowing the pattern to be deliberately changed.

I don't think it's that simple. First, if abusers and victims exist then the situation just is actively dangerous. Hypocrisy is unavoidable but it's less bad if non-abusers can operate openly and abusers need to keep secrets than vice versa. Second, I don't think the pattern can be deliberately changed except by creating a sense of security that starts out false but becomes true once enough people have it.

The thing ialdabaoth and Nick are both pointing towards is something like 'the capacity to actually point out abuse (or any kind of unethical conduct) when someone you respect is involved (or someone all your friends respect) is really hard. It is something that requires practice, awareness of your self, and awareness of how the web of social connections works, and often literal sacrifice.

And it's something that is very easy to tell a story about how you'd totally speak out, or go to the cops or something. But if it turns out that your mom/dad/favorite-teacher/favorite-writer /favorite-startup-founder/boss-that-is-paying-your-salary/venture-capitalist-who-can-offer-you-money is doing something wrong, the overwhelming default course of action is to just fail at speaking out, and if all you have is a vague narrative that you're a good person or that your community is good people, you will be woefully unprepared.

Of course, all of that is true. I'd go further and say there's basically no way a vague narrative that you're a good person would be enough. What I'm trying to say is that the fear of the consequences of speaking out should be balanced with the fear of the consequences of looking complicit if the truth comes out and you didn't speak out. Talking about how advantageous it can be to ignore abuse, or how hard it is to speak out (and implicitly, how forgivable it would be) is tipping that scale in the wrong direction.

I do agree with that (As mentioned in my earlier comment, I still lean in the direction of deleting comments like the initial downvoted one in the future).

But as worded I think I (at least sort of) disagree with your comment, in particular:

Second, I don't think the pattern can be deliberately changed except by creating a sense of security that starts out false but becomes true once enough people have it.

I don't think this is the mechanism by which anyone becomes safe. I think the sense of security doesn't add up to "actual safety" even if 100% of the people have it.

I should've been more clear—by safety I meant safety of making a (true) accusation, rather than direct safety from actual abuse. I think the latter can only follow from the former.

Ah, I think that makes sense.

I don't see how we can fight entropic systems without understanding them.

I don't think your original comment actually contributes to understanding them – I've talked to you enough to have some idea what you meant, but it's buried beneath layers of different frames and inferential distance, which adds up to:

a) the comment mostly just getting parsed 'yay abuse' rather than anything nuanced or important.

b) sort of a drive-by change-in-topic/frame/hammering-on-pet-issue. (i.e. sort of like if we periodically have discussion of videogames, and someone keeps jumping in to say 'have you considered that videogames are superstimulus lotus eating?' And maybe they are, and maybe it's important to talk about, but that doesn't mean it's a productive dynamic to keep bringing them up in random threads)

I do think there's an important thing you're actually trying to say, and you're welcome to say it either in your own top level posts (that actually spell out the inferential distance and make it clear what your actual goals are), or in the comments of people who opt into it. (I have no idea of Sarah's take on that and whether she'd consider this subthread off-topic or on-topic).

But I think making comments like this whenever the topic veers in the direction of abuse or mental illness makes it harder, not easier, to talk seriously about it. And, yes, is generally uncomfortable and makes the site a place people are less excited to hang out, and that is something that actually matters.

I haven't chatted with other moderators about this yet and haven't formed an official longterm stance on this, but am leaning towards deleting comments similar to the initial downvoted one except on threads where the OP author has opted into it.

I've talked to you enough to have some idea what you meant, but it's buried beneath layers of different frames and inferential distance

FWIW, I don't think that I've talked to ialdabaoth a huge amount, but the comment seemed pretty comprehensible to me. The grammar isn't very complicated, and all the words are pretty common and mean what they usually mean*. Maybe I'm just spoiled by reading a lot of Robin Hanson? I sort of agree that point (b) applies and is sort of annoying.

*except maybe 'abuse', which is one that I've never been super clear on what exactly people mean by it, but that's a word that's basically unavoidable here. [I'm not very interested in people explaining it to me in this thread]

I have more thoughts (short answer: yeah, I think reading and internalizing Hanson is a quite reasonable way to bridge most of the inferential distance, although not all of it). Will respond in more detail later.

I'll be moving this subthread to meta, but I'm taking this opportunity to force myself to actually build a "move subthread to meta" tool instead of doing it manually.

I know the mod team have been busy in the past while, but I'd like to remind you of this.

Lol, thanks. I certainly did not end up getting around to this on the day I was planning to.

1. It's not a change in topic. It's an explicit focus on the topic-in-question, and an attempt to explain - in a way that people's guts will *get* - WHY the current equilibrium is preferred to the one being proposed by the author.

2. At no point does it even connotationally say "yay abuse". It DOES connotationally call out humans-as-a-process for consistently performing actions that signal "yay abuse", however. Connotationally saying "yay abuse" would have been phrased very differently, and I think we all know that.

3. Controversiality has less to do with opt-in/opt-out, and more to do with... who we think the connotations are making look bad. I'd really like that to stop.

… an attempt to explain—in a way that people’s guts will get - WHY the current equilibrium is preferred to the one being proposed by the author.

Though it’s likely that what you said is true in some cases, if you think that the model you propose is of comparable explanatory importance to what gwern said, then you’re simply mistaken—so framing your point as “I’m just explaining it in a way people will get” is not appropriate.

Even though it had equally suspect connotation?

What difference does that make…?

It hilights problematic assumptions that lead to problematic voting patterns.

Aaaaand now we really ARE meta.

That has nothing to do with what I said.

I disagree. How do we resolve who's right, within the current trust environment?

From things you've written before I suspect the point you were intending to make is "in practice, people will act in this shitty way because it's advantageous to them". But the actual words you wrote - and the place you put the comment - clearly denote "we should act in this shitty way out of self-interest". I more or less agree with the first thing and disagree strenuously with the latter. (And I really hate the thing you do where you conflate is and ought on things like this.)


Guess: Antisocial opinion. Potentially promoting/endorsing causing suffering to the "victim".

The promotion is already happening in revealed preference, to lethal consequence. I'm just keeping score.

Didn't downvote, but I am confused about what point you were trying to make.

Did you mean that we don't gain anything from a short-term, selfish perspective? In which case, of course -- who's advocating believing victims for personal gain?

Or did you mean that in the aggregate, overall, as a society, we're not better off believing and defending true claims of abuse? In which case, that seems questionable at best.

If I take "gain anything" literally, then it seems clearly wrong. At the very least we gain knowledge by believing true claims.

And if I take "gain anything" to mean, "are better off overall", then I'm still a bit skeptical. For example, in the short run it only hurts me to learn that farm animals suffer. But in the long run, that's something I want to know about and hope that we could ultimately fix. The situation seems similar with abuse of humans.

What am I missing -- do you disagree with any of the above? Are we just using words differently?

Just looking at that first link, the methodology was apparently

They were interviewed using the British Crime Survey questionnaire for domestic and sexual violence, and their responses were compared to those from 22,606 respondents to the 2011/12 national crime survey.

I find it totally plausible that there's a much higher base rate of crime, abuse, etc among the mentally ill. But if we want to argue against the model "mentally ill people are more likely to be delusional or exaggerating", then a study which just asks them is not going to lend much evidence.

I think this post has helped me be less confused on this matter. In my life, I have been seriously mislead by people close to me with mental illness about traumas that they have faced - people who had experienced significant abuse, telling me that they experienced different abuses which they had not experienced.

I think previously I would've expressed it in a more confused way, that their being mentally ill was not evidence against their having experienced abuse. But I would now phrase it that both my prior on them having experienced severe trauma is higher, and also the standard of evidence I require to believe a particular instance of abuse is also higher.

I guess there's a desire that these two things would cancel out. One might hope to say that it doesn't take more effort to discover abuse in the life of a person with severe mental illness than otherwise, but I think that the truth is that it just does take more effort to help prevent any abuse in this situation. I'll phrase it this way: if I am taking care of / in a significant relationship with someone who is severely mentally ill, while I might not trust their direct reports as much, I will increase my prior that there is, and increase my desire for evidence for the lack-of, abuse in the various parts of their life.

This post may not have been quite correct Bayesianism (... though I don't think I see any false statements in its body?), but regardless there are one or more steel versions of it that are important to say, including:

  • persistent abuse can harm people in ways that make them more volatile, less careful, more likely to say things that are false in some details, etc.; this needs to be corrected for if you want to reach accurate beliefs about what's happened to someone
  • arguments are soldiers; if there are legitimate reasons (that people are responding to) to argue against someone or see them as dangerous, this is likely to bleed over to dismissing other things they say more than is justified, especially if there are other motivations to do so
  • the intelligent social web makes some people both more likely to be abused, and less likely to be believed
    • whether someone seems "off" depends to some extent on how the social web wants them to be perceived, independent of what they're doing
    • seriously I don't know how to communicate using words just how powerful (I claim) this class of effects is
  • there are all kinds of reasons that not believing claims about abuse is often just really convenient; this sounds obvious but I don't see people accounting for it well; this motivation will take advantage of whatever rationalizations it can

Charities that work with victims of torture rarely put actual victims in front of cameras to try to drive donations, as they're rarely sympathetic, in large part due to visible mental and physical consequences of the abuse they suffered. Adults who were victims of severe and prolonged child abuse are a good example of this as well.

Additional datapoint, researchers studying 'ability to read emotions in faces' found that incarcerated serial rapists were on average the best at it, and victims of sexual assault were among the worst. If I remember correctly, the paper contained a categorical refusal to speculate further about a predator-prey dynamic.

I think that this analysis would benefit greatly from distinguishing between different types of mental illness. Mental illnesses can be vastly different. After all, the only thing in common between them is that they are mental conditions that are considered "not normal" for some definition of "normal". I suspect that different illnesses might have very different values of

  • The rate of being targeted by abusers.
  • The rate of reporting abuse which did not really happen.
  • The rate of not reporting abuse which did happen.
  • The credibility that other people lend to reports by such a person (possibly for irrational reasons).

I expect much of these effects comes from mentally ill people being in worse circumstances, and disappears if you condition on circumstances, which it seems like you can usually do in practice.


Can you clarify which claim you are making? Is it this:

Somebody goes around saying they’ve been the victim of mistreatment. But they seem mentally ill.

people are very suspicious about whether their allegations are true.

Someone being mentally ill is evidence for, not against, their being victims of a crime.


they are more likely to actually be victims of crimes.


they seem “off” somehow—highly agitated, making social faux pas, telling stories that don’t quite add up.

For no reason.

I don't think they are more or less likely to be telling the truth. I'd be more inclined to propose that the mental health makes for a higher hurdle to be taken seriously because people may be thinking, "how am I supposed to help this person?" which is a hard question to answer for a not-mentally-unwell person.

I don't think "people with mental illness are more likely to say in surveys that they are victim of crimes" should lead one to conclude "if someone with a mental illness tells me that they are a victim of a crime, I should believe them even if there story seems fishy".

The first empiric data towards which you point has that quality and I haven't checked it for the other links. What Korzybski called consciousness of abstraction is important when making arguments like this.

This seems well into 'Politics is the Mind Killer' territory.

this isn't obvious to me, can you elaborate?

As gwern pointed out, this post commits one of the basic fallacies of conditional probability. Since this is a somewhat politically charged topic, and sarahconstantin has been around a while and so could be expected not to make such 101-level mistakes under normal circumstances, someone could reasonably surmise that she stopped looking for possible counterarguments once she had reached a conclusion she liked for political reasons.

It definitely overlaps with politically-charged topics of how "society" treats various subgroups. I'm not sure I'd have gone so far as to call it mind-killing as written, but it could be taken that way (and I'll admit that I did initially take it badly, and gave it more thought primarily because of trust built from the author's previous posts).