Saturday, December 09, 2017

"Believe the women" and Bayesian Priors

This is where I'll likely get schooled by someone who truly understands Bayesian stats, unlike my lawyer understanding, and that's okay. I think a statistical framework to the debate of "who do you believe" would be a useful contribution.

The prevailing feminist position is to believe the women when they accuse men of sexual harassment or assault. The reason for this position depends on the individual feminist but for many is derived from feminist ideology that not everyone else shares. When a non-feminist sees this and thinks why should I simply believe the woman instead of the man, and sees a reasoning based on ideology of female oppression the person doesn't agree with, then the non-feminist dismisses the argument as non-scientific (add a varying mix of bias to this as well). That's where you get the silly, anti-"Believe all women" backlash that the NYTimes enabled.

"'Believe All Women' Isn't a Thing" says Katie McDonough, and she has a point. I'd like to see any link to any feminist actually saying believe all women in spite of contrary evidence. (There's one exception where many feminists come close to saying that - in one-on-one conversation with women or girls that come to a friend or especially a counselor with an accusation of harassment or assault. It's appropriate in that case to not be impartial or fair because your role is to help her, so an almost-but-not-quite immovable belief in what she says is fine.*)

So what about the rest of us, considering the issue of who to believe as a general concept or a particular allegation you hear about? If we accept that we live in a world of probabilities and incomplete information rather than certainty, then we have options. Some non-feminists say why should we believe the woman rather than the man when we don't know either of them? The answer would be that we do have information in the media about not-powerful women accusing powerful men of harassment and assault, and it seems in the vast majority of cases they are telling the truth.

So that's your prior - believe the woman. More information can adjust your assessment of that particular allegation, or I suppose more information in general can adjust your prior. What seems to be happening now with the flood we're seeing is the prior is getting strengthened.

*Importing this attitude from college crisis counseling centers to college disciplinary proceedings is a huge mistake, however, and people are now realizing this. I'll stand up for lawyers, including feminist lawyers, for pushing on the issue of appropriate process.


Tom said...

I think there's too much shorthand involved in this discussion.

'Believe the women' should be a longer sentence, something like 'Don't you dare dismiss accusations out of hand.'

It should not involve a presumption of guilt on the part of the harasser. But more importantly it should not privilege the defendant because of power, position or place in society.

Doesn't fit on a bumper sticker.

jgnfld said...

Re. comment on discipline proceedings, I may disagree to some degree (and speak from experience from sitting on panels). One needs to weigh the degree of proof required against the proposed range of penalties. If the proposed penalty is to essentially require an on-campus peace bond for a period of time or, say, to block the student from certain jobs (e.g., RA on the complainants floor) where there will be contact, the level of evidence need not be terribly high. If it ranges to exclusions from zones on campus or getting kicked out of residences, the level is higher. If it ranges to expulsion it is higher still. Yet even expulsion doesn't rise to the criminal standard of "beyond a reasonable doubt". It is much more the "balance of probabilities" standard as the college cannot do a single thing to the potential offender more than ban him from campus property. You would find HR departments using the same standard "balance of probabilites" standard to fire a person and expel them from coming on to company property.


A century of totalitarianism stands as the Bayesian prologue to the exponentially growth of authoritarian interpretations of texts like Title IX , by a generation indoctrinated to view as priveleged the only mode of discourse many campuses permit.

The hegemony of campus PC has excluded from the present conversation what may be its most disturbing cultural prior -- present post-structuralists don't like to be reminded that the uncivil deconsruction of civil liberty recapitulates the struggle sessions of the Great Proletarian Cutural Revolution.

Jan Galkowski said...

Well, sexual abuse and harassment of all kinds is much more prevalent than the culture in the USA cares to admit. I think, frankly, this is a special case of a larger problem here, where if someone thinks they can reasonably get away with something, the barrier to doing that something is much lower than would be expected if they saw themselves as a member of a greater community.

The most prevalent form of this is in driving habits which, despite these jerky behaviors, still somehow manage to keep most vehicular slaughter down. But it's getting worse, as the frequenters of San Francisco's Taraval Street are aware, but not limited to there. And this is despite strict laws on vehicle-pedestrian relationships, in both directions on the relationship.

Bullying and sexual harassment and abuse fall clearly in this category, especially on the part of males who get encouragement and sometimes pressure to engage in this from peers. There is a gender asymmetry, in the same way the evidence shows males have a propensity for accidental or deliberate use of guns: A presentation at the Boston chapter of the American Statistical Association showed that accidental killings and injuries from unlocked guns in homes by children, often under 5 years old, is overwhelmingly initiated by males.

So, yes, on certain attributes, like maleness and risk-taking behavior, I'd say there are strong priors. But others contribute. Age of participants is another ... Risk-taking is a young person's characteristic, which is why, in part, those who voluntarily enroll in the military tend to be young. But, also, the absence of trustable, believable authority to enforce social standards must also contribute.

That, in the context of a new awakening, where victims of abuse are riding a wave of acknowledgement that such occurs much more than we collectively want to accept, it is inevitable that there will be a fraction of opportunists who will jump on to further their own situation. But it is downright improbable to think this can explain the majority of reports. I think the new awakening is kind of like the good, flip side of a phenomenon well known from sociology, whether it is group sickness or propensity for self-harm: If enough peers do it, then it becomes a thing, in a much more fundamental or organic way than is meant by the word fad.

I guess there are exceptions and I could be wrong: This of the genesis of the Salem witch trials for example, or the accusations which flew about the Reign of Terror during the French Revolution of the 1780s. But even that only happened because the society about it was set up with a propensity to believe these things.

elspi said...

Conservative women have a very ugly and bloody history of making false accusations. Both for racist and political purposes. So this needs to be factored in to the prior.

Jan Galkowski said...


Regarding "Conservative women have a very ugly and bloody history of making false accusations", do you have concrete, quotable, referenceable evidence of this, or is it just your personal opinion?

And, since you are suggesting it needs to be "factored into the prior", surely you understand by suggesting that, you are embracing the responsibility to give a confidence range for your assertion, as well as a falloff rate around your quantitative point estimate. So, what is it?

elspi said...

Jan, we kinda expect people to google things themselves, but just this once:

Emmett Till

Jaime Phillips

Tulsa race riot

Rosewood massacre

Roger Stone and Leeann Tweeden

Is 5 enough or do I need to go on, because I could do this literally forever. Just say the word and I will dig up the corpses.

Jan Galkowski said...


This not consistent with the necessary prerequisites of the discussion, as asked,

"And, since you are suggesting it needs to be "factored into the prior", surely you understand by suggesting that, you are embracing the responsibility to give a confidence range for your assertion, as well as a falloff rate around your quantitative point estimate. So, what is it?"

Accordingly, your comment and use of the Royal We will be ignored, at least by this analyst.

elspi said...

Sorry, I didn't see the "no true scotsman" warnings in you first comment. You are right, no true scotsman would do such a thing.