I interested Tee Barnett and Peter Hurford in adding sexual violence questions to the survey. Therefore sexual violence definitions need to be created. There are various challenges involved. Below I have listed the challenges and proposed a set of questions that I think will help work around them. (This does not mean Tee and Peter will include my particular questions. I am making a proposal and outlining specific challenges that I am aware of.)

    1. Survivorship bias. The survey will only be taken by people who are still around, not by people who have left their EA job or been driven out of the community by sexual violence. Sexual violence does increase turnover. Given what happens to female lawyers in the legal profession, which has a similar gender ratio, we should be very concerned about this. According to The American Bar Association, between half and two thirds of female lawyers experienced or observed sexual harassment by male superiors, colleagues, or clients during the two years prior to the survey. [1] There are some other references on turnover as well. [2] [3] [4] [5] [6]

      If you want a solid example, I am leaving you. That’s right. I plan to exit the effective altruism network, the LessWrong diaspora, and all of my male-dominated interests including programming. I am sick and tired of sexual harassment, sick and tired of people behaving as if they don’t understand or care about my physical safety concerns, and I’m sick and tired of encountering dismissive and insensitive attitudes about something that is causing me suffering and harm. I am leaving.

      A lot of other women have left you. They’ve complained in the women’s group. They’ve warned me in private messages. They’ve commented and urged me to join them in this that and the other totally unrelated community. They were doing these kinds of things *before* I published my article “Exploration of Sexual Violence Reduction for Effective Altruism Potential”. Some people seem to want to avoid talking about this to prevent women from being scared off. Women are being targeted by sex offenders. They are going to notice that. Sex offenders are scaring people off whether we talk about it or not.

      If we want to be rigorous enough to have meaningful survey results, we *need* to compensate for survivorship bias. I have no doubts about that.

      Lists of EA Global attendees from different years can be compared to see how many people returned and how many people did not. CEA won’t release their data. There is a quick and dirty way to accomplish a rough estimate without the data. CEA told me they’d run a script that:

      1. Compares the full names of attendees of EA Global from two different years to see which people returned to the conference and which people did not. (Keeping the names a secret.)

      2. Counts the number of “male” and “female” names (and ambiguous names) on each side. (Based on a database of baby names which includes the gender they’re usually assigned to.)

      3. Outputs a table containing the total number of names in each category that did and did not attend both events.

An r or python script can be run in an installation of r statistics on one of the CEA computers. *Only* the table containing the completely anonymized numerical totals needs to be shared. So yes, this is quick and dirty and would only output a rough estimate. The only way to get really accurate data is for CEA to have an internal person write the script.

    1. Definitional quagmires. Sadly, the definitions of sexual violence vary a lot from one source to the next, and this even applies to the sources that are supposed to be credible like legal definitions, psychology definitions, and research definitions. Worse, all of the definitions I have seen so far are completely insane in at least one respect. For instance, the CDC definition classifies male rape via penile envelopment as a different kind of sex offence, as if it is not rape. It is *not* included in the number of rapes! For some reason, a lot of definitions of stalking require that the person stalks you more than once. What if the stalker follows you around and successfully attacks you on the first try? I guess someone decided that when stalkers are especially effective at what they do, they aren’t really stalking after all? Insane. A lot of the definitions I’ve seen contain something nutty like this.

      1. No standardized definitions.
      2. Lots of definitions include at least one really insane loophole.
      3. All the terms are ambiguous because a lot of people refuse to report sex offences if they have to be specific. This is reminiscent of what happened to user agent values in the browser wars though it happens for a different reason. (They started having browsers lie to websites about which browser they are as a chaotic way to manage competitions over features and compatibility, so the identifiers became pretty muddled.)

      4. Humans respond to incentives with motivated reasoning, cognitive biases and other forms of irrationality. This means that the definitions we use must be resistant to people coming along and to tear them down simply because a state of denial is a lot more pleasant.

      5. Sexual violence definitions do not have a utilitarian basis like how much suffering they cause. This is probably because there can be many differences between an event that is upsetting vs. an event which does not. If a dog sniffs your crotch, you will probably be annoyed but not traumatized. If an adult human did that, it would be very upsetting. Why? The dog is smaller. The dog probably isn’t expressing interest in you. The dog has no political power whatsoever. The dog may be expressing friendly emotions rather than sadism, dominance, etc. Probably other reasons. There are many contextual things like this which result in a huge amount of confusion and also make it extremely hard to define sexual violence in a utilitarian way. A lot of people in our network are interested in whether every type of behavior that qualifies as a sex offence causes harm.

      6. The main benefit of utilizing a common set of sexual violence definitions is that we’ll easily be able to compare our rate of sex offences to external rates of sex offences to know whether it is elevated or not. The main downside of utilizing a commonly used set of sexual violence definitions is that most of them include something insane. Unless we can find a credible source of sexual violence definitions which has no insane loopholes, we will need to choose between comparability and sanity.
    2. Insufficient understanding of sexual responses and preferences: Some people never experience sexual trauma, or are not as likely to experience it due to some kind of trait. Perhaps they did not receive the genes for trauma at all, or the genes for sexual trauma specifically, or they experience sexual traumas under a totally different set of conditions from others.

      From what I’ve been seeing, I am really starting to suspect that there are multiple “orientations” around consent needs and what situations are experienced as sexual violence. Some people vehemently deny that sex offences cause harm or act like harm is rare when research shows it is not. I think what’s happening is that people with non-standard sexual responses and sexual preferences tend to construct bubbles based on these traits. They may not realize they are living in a bubble and that the rest of us really do experience sexual trauma.

      There isn’t a definition that such people will find credible because the crux is not about verbiage for them. What they would find credible is an explanation as to why some people claim X is harmful while others claim X was enjoyable or did not cause harm.

      My best answer for them is that most people probably experience sexual trauma, while those who don’t experience it (or who have different conditions for it) probably at least sometimes end up in social bubbles with others who have different sexual responses and behaviors.

      I don’t know all the details, and I may only see the tip of the iceberg. Right now, I’m pretty sure that some people need both explicit verbal consent and monitoring for enthusiasm in order to function sexually and avoid trauma. Others require body language (Or perhaps other cues? I am not one of these people.) to function sexually, and have literally reported that using explicit verbal consent actually causes them dysfunction. One person reported that this is *in spite of* the risk of trauma, *not* because the risk of trauma is absent.

      To make it more complicated, I think there are kinky people in both the explicit consent and body language camps. That is to say I think some kinky people sexually dysfunction if they can’t use body language and must use explicit verbal consent.

      Such people are not going to cease their sexual behaviors because others function a different way. They wouldn’t function at all then. At this time, I think it is safest for everyone to encourage people to form different bubbles that all members are comfortable with. Vilifying comfortable bubble creation doesn’t make much sense. I think the people in each bubble just need to be aware that there are people in various other bubbles who genuinely have different responses and preferences.

      I think right now there are a lot of misunderstandings because the people in the different bubbles don’t even believe that the other bubbles really exist for sure, let alone do people from each bubble know how the other people work. We’ve all been viewing each other’s bubbles as if they are non-existent, imaginary or dysfunctional.

      I think if people shared openly the consent philosophy that they prefer others to use on them, this would help immensely with preventing confusion when people in multiple bubbles mix. I think this has the potential to reduce sexual violence because it will reduce confusion between those who need explicit verbal consent and those who need to use body language. There may be a lot of other differences I am unaware of, so I will leave off with a hope that people will communicate more clearly in the future to prevent misunderstandings.


      Bubbles are *not* a complete solution. This is because some people literally have preferences like “doing stuff you’re not supposed to do” / “doing unwanted things”. Yes, I do mean that there are people who do this *without* hoping to delight the other person. (Some sadists and dominants prefer to delight masochists and submissives while others just go for anybody without consideration, and some actually intend to cause people harm.)

      Others have problems along the lines of a strong mind projection fallacy which results in them believing others reciprocate their interest when this is not true. This bias might be bad enough to cause a sex offence. Additionally, it seems that many people experience so much motivated reasoning that they rape people even though they experience empathy and should theoretically be able to tell that they’re doing something harmful. I guess, for some, the motivated reasoning and bias that can happen during mating experiences is deeply confusing.

      Then there are people who take more risk with sex or kink than they intend to and someone ends up unnecessarily harmed through a lack of education. So, unfortunately, there are a bunch of things that can result in people getting hurt even if everyone aims to communicate really well and people stick to dating/hookups/play with others who share a compatible consent philosophy.

      If we use a mainstream source of sexual violence definitions, it will stigmatize people who need to use body language rather than explicit verbal consent. They probably aren’t going to change their behavior, since that would result in sexual dysfunction. Also, I’m concerned that there probably is not enough research on people with this preference for us to have any understanding of them. Without thorough research, stigma is unjustified.

      If we don’t use definitions that require explicit verbal consent, this may open the floodgates for a lot of people who are confused by motivated reasoning and bias. It is probably safer if people prone to motivated reasoning and bias do *not* try to use body language to figure out what others want. They need to work extra hard to make distinctions and I suspect it helps them to be told clearly and explicitly what is wanted.

    3. Cognitive biases, motivated reasoning, and incentives.
      1. Optimism bias:
        The tendency to expect things to be more pleasant than they are.

      2. Normalcy bias:
        The tendency to fail to plan in advance for disasters and to respond inappropriately when one is underway.

      3. Ostrich effect bias:
        Ignoring an obvious negative situation.

      4. Affect heuristic:
        Being more likely to believe information that is pleasant than information that is unpleasant.

      5. Status quo bias:
        Being more likely to support the status quo.

      6. Survivorship bias:
        Focusing on who or what is left / failing to take into account who or what is gone.
      7. Sour grapes bias:
        Telling yourself that what you’ve lost wasn’t desirable anyway.

Proposed definitions:

The things I care about the most are things like whether someone is traumatized, whether they become suicidal and whether they leave or intend to leave the movement or their job.

What would be *really* interesting is to ask *both* “did things A, B, or C happen to you” *and* “did it cause you to experience X, Y, Z harm”?

This would be long because there’d be a subset of questions *about* each of the questions in the main set. Also, since multiple offences can happen to one person, you’d need to be able to answer each question and subset question more than one time. This sounds complicated and I’m not sure it will happen since surveys have length constraints.

This would be awesome because we would have an opportunity to understand this so much better by taking a peek at which behaviors cause the most harm.

If it needs to be concise, I’m imagining something like this:

      1. What consent philosophy do you prefer others use on you? (If you use a particular variation on a consent philosophy, please specify.)

      2. Where is your consent philosophy documented? (Please link to the specific variation if possible.)

      3. How many times did any EA violate your consent philosophy with speech that appeared to be motivated by sexual or romantic desire?

      4. How many times did any EA violate your consent philosophy with exhibitionism that appeared to be motivated by sexual or romantic desire?

      5. How many times did any EA violate your consent philosophy with voyeurism that appeared to be motivated by sexual or romantic desire?

      6. How many times did any EA violate your consent philosophy with touching that appeared to be motivated by sexual or romantic desire?

      7. How many times did any EA violate your consent philosophy with following, intruding or spying (non voyeurism) that appeared to be motivated by sexual or romantic desire?

      8. How many of the aforementioned instances of violations had a significant negative impact on your psychological health?

      9. How many of the aforementioned instances of violations resulted in suicidal thoughts?

      10. How many of the aforementioned instances of violations resulted in a desire to leave an EA job?

      11. How many of the aforementioned instances of violations resulted in a desire to leave the EA movement?

      12. If applicable, please list all the types of EA spaces where the aforementioned instances of sexual harassment or sexual violence happened. (examples: EA share house, EA organization, EA conference, EA work party, an unrelated cafe, a park, etc.).

      13. What do you think might have prevented the violation(s) of your consent philosophy?

This wouldn’t give us the sort of result that’s directly comparable to external sexual violence statistics. However, it would give us the sort of result we care about: whether people’s preferred consent philosophies are being violated, and whether the violations are causing harm. Most importantly, we’d actually have some data about the impact on specific areas including psychological health, suicidal thoughts, turnover intentions, and intentions to leave the community.

The results would be harder to compare from one year to the next because tying sexual violence definitions to people’s individual ideas about consent adds more complexity than brutalizing all the subtleties to create standardized verbiage. It *is* possible to make survey results comparable *if* people put in the effort to thoroughly document all the different sets of preferences they have, name each consent philosophy or variation, and publish these. I think encouraging people to do this is a good thing because it encourages us to deepen our understanding and communicate more clearly.

Tying the definition to people’s individual preferences just seems a lot more sane than trying to come up with sexual violence definitions that work for everybody at a time when so much more research needs to be done to understand all the variation in responses and preferences, not to mention solving whatever it is that has been resulting in insane loopholes.

Nonetheless, I think something in this general direction would be much better at tracking the things we care about the most: whether preferences are being respected, and whether harm is being done.

I’m not an expert on designing surveys. There will be people on the survey team looking this over and it may need to be changed. This is step one in the process: post a draft to get feedback on!

Please provide your feedback.


    1. Laband, David N., and Bernard F. Lentz. "The effects of sexual harassment on job satisfaction, earnings, and turnover among female lawyers." ILR Review 51.4 (1998): 594-607.


    2. Sims, Carra S., Fritz Drasgow, and Louise F. Fitzgerald. "The effects of sexual harassment on turnover in the military: time-dependent modeling." Journal of Applied Psychology 90.6 (2005): 1141.


    3. Merkin, Rebecca S. "The impact of sexual harassment on turnover intentions, absenteeism, and job satisfaction: Findings from Argentina, Brazil and Chile." Journal of International Women's Studies 10.2 (2008): 73.


    4. Merkin, Rebecca S., and Muhammad Kamal Shah. "The impact of sexual harassment on job satisfaction, turnover intentions, and absenteeism: findings from Pakistan compared to the United States." SpringerPlus 3.1 (2014): 215.


    5. Salman, Maheen, Fahad Abdullah, and Afia Saleem. "Sexual Harassment at Workplace and its Impact on Employee Turnover Intentions." Business & Economic Review 8.1 (2016): 87-102.


    6. Sandada, Maxwell. "The influences of sexual harassment on health, psychological condition, work withdrawal and turnover intention in South Africa." Journal of Business 1.2 (2013): 84-72.


11 comments, sorted by Click to highlight new comments since: Today at 4:06 PM
New Comment

I interested Tee Barnett and Peter Hurford in adding sexual violence questions to the survey. Therefore sexual violence definitions need to be created.

Thanks for your dedication to this issue. I'm compelled to point out that that briefly speaking about a particular issue in an informal manner should not be seen as an endorsement on behalf of myself or Rethink Charity.

In response to the comment that was deleted below, we do not intend to ignore this issue.

I fucking hope not. RIP Kathy <3

Can we put a lid on all these disclaimers being thrown everywhere by everyone who does anything in EA? It is getting stifling and degrading. Just state what you believe.

[-][anonymous]4y 1

Is the "collaboration" mentioned here referring to the same brief informal conversation?

Posting these links here for cross-reference:

As I said previously, I'm glad you are working on this problem. However, I suspect you are defaulting too strongly to public communication and coordination on this forum. My impression is that many EA orgs tend to keep their analysis private and publish only if there's a compelling reason. For this particular issue, I see a few reasons to default even more strongly to privacy: First, there are reputational risks to the EA movement. Second, public online discussion of topics like this can degrade notoriously quickly. Therefore, I suggest that instead of soliciting public comments, you put together a small group of advisors, people with a few different perspectives from a few different stakeholders, and ask them to provide views on things like what the right set of questions to ask is. I suspect that this approach will get you better advice than drive-by internet commenting, while also mitigating the risks I described.

I have some more thoughts, but I'm following my own advice and sending them via a personal message :)

I am also not an expert on designing surveys, but it seems really hard to get meaningful data on something like the 'consent philosophies' that you describe, at least without broadly-understood and theorised examples of different versions of them. Imagine trying to get an idea of the different ethical views of EAs without being able to rely on terms like 'utilitarianism' and 'condequentialism' and others that have a well-developed meaning from the philosophical literature.

Asking people to describe their ethical views in such a situation seems like a really bad idea - not only would it require a ton of work in reading through all the responses to get even a vague idea of the common threads, it requires the survey-taker to do a ton of work in writing up their views; not only to describe them, which itself might take some time, but also to reflect and come up with verbal descriptions of things they don't neccessarily think explicitly about.