Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=Transcript of the talk, from the video's captions. Auto-generated: speaker names in particular are unreliable.
=

# The Consequences of Faculty Sexual Misconduct

Authors: Sarah Cohodes, Katherine B. Leu  
Discussant: None  
Video: https://www.youtube.com/watch?v=1kb3a99sA-E&t=632s


## Talk (00:10:32 – 00:56:30)

[00:10:32] Okay, thank you. >> All right. Uh, thank you for having me.  
[00:10:39] Uh, my name is Sarah Hodus. I'm presenting this paper today, which is joint with my student, uh, Katie Lou.  
[00:10:45] Um, I'm going to let you read this uh quote uh while I do a little uh intro and um and then you'll see uh how it becomes relevant. Uh the first thing I want to let you know is that today I'm going to be talking about uh the  
[00:10:59] consequences of faculty sexual misconduct and it's possible that people in this room have experienced uh faculty sexual misconduct. So I wanted to let everybody know that while we are discussing the topic um I'm not going to be going into any details about the  
[00:11:13] specifics of any incidents. uh or exactly what went on. Uh and I uh also just wanted to acknowledge the shared experiences of people in this room.  
[00:11:24] Um so as I said uh this uh work is joint with uh Katy Louu. Uh and uh let's get going. Uh as you all are probably well aware academic sexual misconduct is  
[00:11:36] widespread. Uh specific uh fields uh archaeology, anthropology, uh social work uh there's been surveys that show very high levels of uh misconduct especially for cases in the  
[00:11:50] field or with trainees. Uh a recent survey that tried to be comprehensive of over 30 campuses found that uh 19% of young people in college experienced some form of sexual misconduct and faculty  
[00:12:05] perpetrators accounted for 11% of those incidents. So obviously that's a small proportion of any uh of the misconduct that college students experience. But to me, I think that is particularly egregious because of the abuse of power  
[00:12:18] that comes with a person in authority taking advantage of that position. Uh in the lab sciences where somebody comes in to work in a lab with a a PI, uh this is a particularly uh relevant problem  
[00:12:33] because there's less ways to to get out of a a situation with a particular supervisor. uh and uh women in the sciences at the graduate level report experiencing sexual harassment typically  
[00:12:46] from their PI at levels of 20 to 50%. Uh lest you think I would leave my own institution out, uh this is some reporting from the Michigan Daily uh talking about uh an incident in the  
[00:12:59] music and uh theater and dance uh department. Um uh and uh something to highlight is that the news of this story was broken by the Michigan Daily and student newspapers are really a uh a  
[00:13:12] hero uh in uh this story bringing a lot of these incidents to light.  
[00:13:18] So uh what do we know about institutions and how students respond and how uh institutions respond? There's been some research out there that doesn't necessarily look only at faculty sexual misconduct but looks at scandals in  
[00:13:32] general which includes uh faculty sexual misconduct as well as uh cheating uh crime other things like that. Uh in general people have found that there is  
[00:13:44] a small decline in applications that then picks back up after a few years and there's no difference in enrollment. Uh there is a paper that looks at title nine cases in uh specifically finding  
[00:13:58] that after a title nine case this is typically studentto student misconduct uh that becomes known uh applications actually increase uh which they interpret as uh the idea that any press is good press just getting your name out  
[00:14:12] there uh gets people more aware of the institution using the same source of data that I'm going to talk about shortly uh There is um uh some research on what happens to  
[00:14:24] perpetrators. Uh finding that uh probably shocking nobody uh that uh there is no difference uh that senior scholars have fewer employment consequences than junior scholars if  
[00:14:37] they have been uh accused uh and found uh uh responsible for one of these incidents. Uh and also that perpetrators work is less cited after the incident.  
[00:14:47] Is that iPad supposed to tell me how much time I have left? I it's it's not telling me anything. So, uh I I I'm gonna interpret that as uh I have infinite time. Um >> Okay. Okay.  
[00:15:02] >> Okay. All right. So, um this brings me back to um uh the the quote that I brought up at the beginning of the talk uh which is the idea that in if you look only at the institution level, you're  
[00:15:16] missing a margin by which students can respond to this type of misconduct.  
[00:15:20] namely changing their area of study. So, so leaving uh their chosen path to go to a different one or never entering that path in the first place because they know uh uh that they might encounter misconduct either they're worried about  
[00:15:35] experiencing it directly or worried about reputational consequences or something like that. So the idea here is how do we shift from thinking about okay what about the research that this so-called genius produced and what happens if we have consequences for that  
[00:15:49] person and to shift and start thinking about okay what does the existence of this misconduct do to the pipeline of future people who could be involved in this field who may be deterred because  
[00:16:01] of the the incident of misconduct. So, uh, I'm going to ask, uh, to what extent do reports of faculty sexual misconduct affect degree completion within the field where the perpetrator, uh, is, uh,  
[00:16:15] housed and does that differ over time and by gender? And then also, do those shifts in major, uh, which are induced, uh, end up, uh, increasing gender segregation in majors and does this affect potential earnings for the young  
[00:16:30] people who are affected? I'm going to bring together two data sources to do that. So, one is iPads, the integrated post-secary education data system from the US Department of Education. Uh, just a shout out that  
[00:16:44] this was a data set that has been produced for many, many years and hopefully will be continued to be produced, but you never know. Um, and this has the number of graduates by institution, gender, and major uh from  
[00:16:56] the 80s until basically the present. It also has institution characteristics and I'm able to link that to the college scorecard and other uh sources of information on earnings associated with institutions and majors in order to get  
[00:17:11] some measure of predicted earnings. I'm going to link that to the academic sexual misconduct database ASMD. Uh this is a database uh begun by uh Julie Learin who is a professor at MSU and  
[00:17:25] it's a systematic catalog of internet searches of misconduct from the 70s to the present. It only uses publicly available information information and has been used uh in research elsewhere.  
[00:17:37] So a little bit about this because I'm sure you're interested. Um I found out about this by reading uh inside higher ed. So, you know, read your trade press.  
[00:17:46] Uh, and, uh, the ASMD began as a catalog of anecdotes sort of in the, uh, uh, me too era of of individuals reporting their own experiences. Uh, and then probably because of legal liability  
[00:18:00] reasons, it became a systematic search protocol uh, with only verified publicly known cases involving faculty, administration, staff, and coaches.  
[00:18:10] So into [cough and clears throat] this database which anybody can go find on the internet you can download uh the link to the information source the name of the individual who was the perpetrator the institution their academic field the resolution and the  
[00:18:24] resolution year so uh uh I went and downloaded that over the years uh but it uh turns out that's not all the information that you would need uh so we had to supplement that. But first, I will tell you uh that it's important to  
[00:18:37] acknowledge that uh the ASMD is only going to be the tip of the iceberg. This has verified, publicly known reports.  
[00:18:45] There's going to be more misconduct going on. Um I we argue that this is the right thing to focus on because students need to know that the misconduct has occurred in order to respond to it. Of  
[00:18:59] course, there may be whisper networks and other ways of information getting conveyed, but a public report is the the um uh most convincing way for for people to to learn about this. So, um we should  
[00:19:12] say that our estimate is in reference to the public report of an incident, not the existence of an incident itself. Uh in a few cases we do not in a few cases in a majority of cases uh we know the  
[00:19:26] timing of the incident itself and so we can test to see if people respond to the incident rather than the public report.  
[00:19:32] Though often it is somewhat vague in the spring of such and such uh a few years ago as reported in the news article as opposed to having a specific date for a news article. We don't know the extent  
[00:19:45] to which uh public reports underount the actual incidents uh instances of uh uh misconduct. Um just to put it in perspective though again we have no way  
[00:19:58] of knowing uh there is a study of one state in 2015 that had about a thousand reports to title nine offices. So that's going to not just include faculty misconduct but also studentto student misconduct.  
[00:20:11] of that about a quarter of those com um interactions with the title n office became official complaints of which less than half were substantiated.  
[00:20:22] uh and so uh many people who contact even though so and those are only the cases that get elevated to somebody contacting the title n office in many cases uh the situation is resolved uh  
[00:20:36] with counseling or referral to resources as opposed to an official report and in some cases that is the uh desire of of the the person who is is reporting so um I can't say how much but I am confident  
[00:20:51] that there is more mont conduct going on then rises to the public report.  
[00:20:56] >> Substantiated >> I'm I'm going to talk about that uh uh right now. Uh so so what gets uh included in this database? It has to be a person in a supervisory role. So faculty administrator, researcher  
[00:21:11] engaging in sexual harassment, assault, misconduct, stalking, violation of campus policies around dating or pornography interactions with students.  
[00:21:20] And then there needs to be qualifying evidence. And so that can be a finding of uh uh responsibility. Uh so uh a decision-making process in uh these cases does not often involve a finding  
[00:21:34] of guilt as in a court of law. But uh an administrative a process can find somebody responsible. So if there is a finding of responsibility uh if the person uh uh leaves their job  
[00:21:48] after this report that is taken to mean that there is uh uh qualifying evidence.  
[00:21:53] If there is um uh electronic communications that substantiate the case. Um if there is a court uh uh finding of guilt or a court settlement and if the perpetrator dies. Those are  
[00:22:07] all taken as uh qualifying evidence. Um it the um uh on there are several cases of uh perpetrators dying of suicide um  
[00:22:19] during the the time that these uh cases uh take place. Um so if none of those uh sources of information arise uh then uh you do not get put in the database. And  
[00:22:33] there are some cases that are in the database but have not become sort of official entries until uh the resolution happens because a resolution could happen and uh the person be found not  
[00:22:45] responsible. Um uh so these are the the sort of cases any or uh or more than one of these criterion means that you qualify uh and then you're included in that the database.  
[00:22:59] Uh so what did we do with those uh cases? We basically we we searched again for every single one of these cases because what we needed was not just did this happen but we wanted the public revelation of when it happened. So when  
[00:23:13] was the first uh case uh made known when was this case first made known as opposed to the date of the resolution which was what was collected in the ASMD. Uh so we confirm or correct everything. We use the exact same uh  
[00:23:26] information uh that was uh used to do the search in the first place. Though we do some extra web soouththing in cases where it wasn't apparent uh what uh academic field the perpetrator was associated with. We find that uh and then we so we are uh collecting the date  
[00:23:41] of public revelation, the date of the incident if it is known, the gender of the perpetrator and the victim, whether or not the victim was a student and if so were they an undergraduate or a graduate. This also includes some cases  
[00:23:54] uh where other faculty or staff are the the victims and then how much news coverage uh uh took place and we'll talk a little bit more about each of those things.  
[00:24:05] Here are some uh sample articles. One from the Hartford Krant uh that's from uh 1996 and then this one is from BuzzFeed um from 2016. And so uh you will see that uh articles like this are  
[00:24:19] um um you you it's not uncommon to see an article like this uh in in the press uh both in local press and in uh more specialized press.  
[00:24:32] Uh as you can see during the time period that the ASMD covers uh incidents have increased over time. Uh that is likely a factor of multiple things. So not necessarily the um existence of the  
[00:24:46] incidents in the first place but whether or not is reported and whether or not it has that verified evidence. Uh the big uh peak that you see there comes from uh 2018 sort of post uh the Harvey Weinstein revelation. Uh and then you  
[00:25:00] see the decline afterwards. Part of that probably has something to do with COVID.  
[00:25:05] So, fewer uh in-person interactions, maybe a breakdown of reporting systems, um and uh also sort of this sort of pent up uh um demand for reporting that came out at that time. It's also the case  
[00:25:19] that there are still cases that are pending that are going to feed into those more recent years but have not yet been verified.  
[00:25:27] Um what are the majors where cases come from? uh there are two things that seems to really um uh come behind a a major having a lot of incidents. So one is uh being a large major. So things like  
[00:25:41] biology, English, psychology, if you have a lot of humans, there are a lot of interactions and a potential for interactions to go badly. Everybody wants to know where economics is. It's 11 incidents. It's in the it's in the  
[00:25:54] middle there. So, uh, while we are not, uh, not known for, uh, covering ourselves in glory when it comes to gender relations, I guess we're sort of, uh, in the middle of the field when it  
[00:26:06] comes to the number of incidents. Uh, >> that's true. Um, the other uh, the other case where there are a lot of, you know, very high reports, uh, theater, music,  
[00:26:20] anthropology. So these are cases where there is a lot of one-on-one interaction between uh uh faculty members and students and in some case there's interactions that happen outside of the classroom in a practice room in a  
[00:26:33] theater in a on site on field and so then you can think about there being the potential for more misconduct to happen um in those cases. Yes.  
[00:26:45] >> Why are you ex Oh, thank you. Why uh why exclude the faculty on faculty misconduct? Because if it was to be made public, it could also affect me to to think, you know, that >> um this is a place where misconduct  
[00:26:59] happens and I could leave the major. >> So um uh we I mean I'll show you the line in my robustness table that that includes it. So So we do include it. Um I don't know if you all remember being a college student but like the activities  
[00:27:14] of faculty were in general sort of distant from the daily experiences of students in school. And so our thought was to sort of focus on like what is the  
[00:27:24] most salient to a student what um uh would be relevant to them. Uh and that's why we end up focusing on the cases with the students as well. And then it's not the case that somebody can say, "Oh,  
[00:27:38] this is just dating gone wrong." There's like a clear uh violation of a faculty student uh relationship.  
[00:27:45] >> What's the coaches? >> We don't have coaches because we can't associate them with an academic field.  
[00:27:51] And so we're looking at the major specific response. And so um uh and University of Michigan certainly has its uh issues in that that area. Uh so because we can't associate that with the one department is why we've excluded um  
[00:28:05] coaches and administrators as well where University of Michigan has also not distinguished itself. So there was I saw the mic go out for another question.  
[00:28:14] Anna um >> oh we can go for it.  
[00:28:16] >> Okay go ahead please. >> So reporting is likely to be likely to internalize the culture in the place. In other words, in departments where you think your report won't be taken seriously, you're less likely to report.  
[00:28:30] Are you able to account for this? Because this could flip the whole thing around.  
[00:28:34] >> Yeah. So, um, we only know of the reports that exist. We we don't know of reports that have not been verified, have not gone through this process. So it, you know, it certainly  
[00:28:49] is a selected sample and the the uh institutions that have reports are different from institutions that don't have reports. I'll I'll show you that uh in a little bit. And so we do some matching to account for the type of  
[00:29:03] institutions where we're likely to see a report. Um and all of our analysis looks at sort of within major changes. So if we think that's associated with a specific major, we're accounting for that there. we can't account for the  
[00:29:18] sort of idioc idiosyncrasies of this department as opposed to to that department.  
[00:29:24] >> Um I just have a clarifying question. Do you know if like the more prominent cases of people being shamed for sexual misconduct leads college students to maybe be more empowered to report on it?  
[00:29:37] Is that something you guys can look at at the back end?  
[00:29:39] >> So we haven't looked at that directly. I think the just the descriptive fact that we saw a lot of cases post the Harvey Weinstein, you know, 2017 is an indication uh that um there is like a  
[00:29:53] public, you know, this rising to public attention does make a difference. a downside of what we're doing by comparing within major across institutions means if there's like a national thing something that makes it  
[00:30:07] to the front of the New York Times that's going to depress both the institution specific but potentially also the neighboring institution if there's an incident at Harvard then maybe the folks at TUS also say huh you  
[00:30:20] know neuroscience isn't for me or or whatever it is so that if anything we think is depressing our findings and as I'll show you later um we actually find that the magnitude of the response is larger in cases where there is less news  
[00:30:34] which actually implies that you know uh in the higher news cases there's a level of national response that we can't pick up because we're doing this based on the internal >> Sarah with response to the one of the  
[00:30:46] previous questions you should say that these are after some period they're all title nine cases so they're not going to a particular department. They're going to the Title N office.  
[00:31:00] >> They are not necessarily title N cases. >> They're not necessarily [clears throat] necessary.  
[00:31:05] >> That's right. But in many cases, they are. And therefore, they're not going to a particular department.  
[00:31:10] >> Yeah. Right. Okay. Yes. I see where you're who is adjudicating this. It's not the individual department that adjudicates it. It is the the university as a whole. And then if you make a a police report that that process as a  
[00:31:23] whole. >> Yeah. So um uh 57% of our cases have multiple victims that are reported. It's highly likely that those where there's  
[00:31:37] only one reported case have um also have additional victims. All right. So let me tell you more and and and then we'll we'll take a breather in in a second and come back to questions. Uh shocking.  
[00:31:49] Nobody in this room. Over 96% of the perpetrators are men. Uh but there is a small number of female perpetrators in the data set. The vast majority of the victims are women. So 89% of them. But there are some cases where there are  
[00:32:04] male victims. Uh 6.6% and some cases where there are both male and female victims.  
[00:32:10] uh the majority of cases that are reported and are in the database over 80 cases involve student victims and then uh about 17% are faculty or staff are the the victims and we show uh uh what  
[00:32:23] happens when you add those or not. Uh as I mentioned uh the majority of cases have multiple reported victims and uh uh a little bit more than 40% have only one identified victim but the likelihood of there being other victims in that case  
[00:32:36] is very high. It's also the case that not only do uh perpetrators have more than one victim, there's also cases of departments having incidents of uh clustered bad behavior, either repeated incidents in the same  
[00:32:50] department or a number of incidents that come to light at the same time in the same department. Uh you might might remember this happening uh at Dartmouth uh recently. Berkeley School of Music has had that uh uh University of  
[00:33:03] Colorado Philosophies department. Uh so about um uh 17% of the cases have some sort of clustered incident.  
[00:33:12] >> About half the cases have um uh one to four news articles and then about 20% end up with 10 or more news articles and there's a small number of cases usually early cases that only have uh court  
[00:33:26] documentation that's available. So it's still a public record but maybe something that is uh uh less wellknown.  
[00:33:33] The most common outcome for a perpetrator is that they're no longer employed. And many things and talking with people who have been in administrative uh positions often this ends with a you are leaving and there's  
[00:33:46] a non-disclosure agreement and uh you know nobody is going to say anything about this. So that's the most common thing that happens. There are cases where uh you know there's a warning or training or something like that. Uh and  
[00:33:59] in some cases uh uh about 8% of cases there's a court judgment.  
[00:34:07] >> Yes. All of this is for the substantiated cases that are in the database. So we don't I we don't know what isn't there in in the database. If you have a question, raise your hand so that my assistant  
[00:34:21] can give you the microphone or else the people in the the millions of people listening to this will not hear you.  
[00:34:31] >> Uh, okay. So um as I mentioned the instit institutions that ever have a report in the ASMD that's column three here are different uh um than uh the uh  
[00:34:45] institutions overall. They're more likely to be selective. Um they are more likely to have graduate programs. Uh there really isn't a difference though in terms of the percent of women on campus. Uh and uh they tend to be larger  
[00:34:59] institutions. Uh the differences here could be both about um uh where these institute uh incidents happen. It can be about uh where they get reported and uh  
[00:35:12] also sort of the likelihood of reporting uh uh given that they happen. Um as you can see I have some comparison institutions which look more like the ever ASMD sample than the all or the never and we'll talk more about where  
[00:35:26] those come from in a couple of slides. All right. So what am I we going to do here? We're going to use variation in when and where misconduct happens uh when it becomes public I should say when it is revealed to look at the how  
[00:35:41] exposure to knowledge of that uh incident happening affects degree completion within the field of the perpetrator. So again this is about the public report of uh misconduct not the existence of misconduct and uh per se.  
[00:35:56] So students can only respond to what they know. It does mean that the comparison departments are both going to include uh departments uh uh that have no public report. So that's going to be departments where there is private misconduct going on as well as departments where there is no  
[00:36:10] misconduct. Uh a challenge in doing this is uh the literature that uh everybody has been exposed to over the past five or six years in difference and differences.  
[00:36:20] This is a long panel. We're worried about how responses evolve over time uh and uh how uh the estimators that are traditionally used uh can be biased uh when you see that difference in response over time. So we're trying to avoid  
[00:36:34] contaminated comparisons comparing to institutions that also have public re revelations. So the solution and there are many solutions we can talk about but the one that we focus on here is a stacked event study and I'll tell you  
[00:36:48] how we do that. um we start with an institution that has a reported incident and we take the data that happens eight years before that and uh eight years after that. So we have this sort of  
[00:37:01] chunk of data for that institution that uh had the reported incident in a specific major and then we sort of have as the pool for potential comparison institutions all the neverreated  
[00:37:15] institutions that have the same major and same years. Again never treated means no public report.  
[00:37:22] And then we're going to do some propensity score matching to make the institutions in the comparison group look more similar to the institutions uh uh in the treated group. So we're going to restrict to these high propensity comparisons.  
[00:37:36] So we're only going to keep those and then we're going to sort of create a what we call a stack uh uh that is a data set for that particular report. And then we're going to create those for every single report uh each with their  
[00:37:50] own field and own timing. Uh uh they each get their own data set. And then we're going to aggregate all of those together to be our stacked data data set. So each case gets its own set of comparisons that are uh matched  
[00:38:05] comparisons that are never treated. So there's no chance for uh sort of contamination in the uh comparison group. And we're going to have institution by data set and year by data set fixed effects. Each stack is weighted by the share of treated  
[00:38:20] observations. That's just going to uh because each uh stack is an individual incident, that's just going to weight by the length of the panel. And our standard errors are going to be clustered by institution by field. And by construction, all of those comparisons happen within the same  
[00:38:34] major. So if for example we're worried about the rise of computer science over time and that obscuring uh uh what is going on in some of these departments all of the comparisons are comparing within major. We don't have a major uh  
[00:38:49] fixed effect because of how this is constructed. The data set fixed effect sort of is that major fixed effect.  
[00:38:56] And then uh we have uh several different comparison groups that we are considering. So one is uh from this propensity score at the institution by major level we include all the characteristics that were in this uh  
[00:39:10] table here as well as uh sort of the um uh growth of the major and and uh in that field. Um we can also we have an alternative comparison group which is where we do something very similar but  
[00:39:24] at the institution rather than the institution by major level. And then uh we uh also look at the not yet treated and never treated same majors and evertreated uh institutions. So if you  
[00:39:38] have an institution that has an incident in philosophy, we throw out your philosophy but we'll take your other um uh incidents in other departments and then we do a little matching in there to  
[00:39:50] uh uh uh compare between similar groups. The idea here is not that incidents are random just that these comparisons form a plausible counterfactual for how growth of that major would have evolved  
[00:40:03] around over time or decline of that major would have evolved over time. in the absence of the public report of misconduct. If we think that comparison academics fields also have incidents that people respond to and are moving away from, that's going to bias our  
[00:40:17] estimate towards zero because that's going to be depressing uh uh degree completions in that field in a comparison institution.  
[00:40:26] Uh here is just the event study that we are going to estimate. I'm not going to belabor it uh uh given the time. So uh the same thing that we saw before we're going to look at how the effects evolve uh uh before and after. Uh something  
[00:40:40] that we do do uh is we aggregate uh the estimates using a comparative interrupted time series. So that's a little different from a difference and difference. Basically we fit a trend beforehand and a tread trend after time  
[00:40:53] and look at how these evolve over time. This is for statistical power reasons.  
[00:40:57] These are rare events. the uh event study on its own is underpowered and and aggregating for the uh uh SIPs uh gives us some more statistical power.  
[00:41:13] >> Uh I'm wondering how you think about pre-trends in this context because you it wouldn't be surprising if through word of mouth you did see some anticipatory effect but at the same time of course and it's hard to diagnose  
[00:41:24] whether there's also a pre issue. So um so what I can say is empirically we don't see pre-trends and then when we reestimate using the incident year for the I think it's like 93% of cases we  
[00:41:37] there from reading the news articles we were able to extract something about the timing of when it h the incident happened. We basically the effects disappear if you do it based on the incident year. So we don't think these  
[00:41:50] whisper networks are are super active that it really is the public revelation and I think the whisper networks probably are active but just at a small scale that we can't pick up in in uh our  
[00:42:03] our data. Um uh we don't have a jump term in uh this sit because we think the response evolves linearly. Uh that's because uh people who are just about to graduate have very little opportunity to  
[00:42:18] change their academic field and so it takes some time uh to see the effects uh uh come in as you'll see shortly. Though uh for a few of our outcomes at the institution level we do introduce that jump term just because it seems like it  
[00:42:31] fits it fits the data level uh better. Um the institution level outcomes aren't differentiated by major. So we collapse to the institution level uh and we have to use our alternative comparison group of uh propensity matched institutions  
[00:42:45] because we are at uh uh the institution level other otherwise we have uh the sort of the same event study uh and same fixed effects uh though we aren't com limiting comparison to the same academic field and we're going to cluster  
[00:42:59] standard errors at the institution level. Okay, so the first thing I'm going to show you is a big nothing. uh this is institution level outcomes for application enrollment and degree completion. There's no change uh in any  
[00:43:13] of them uh that's statistically significant either with the sits or with the individual event studies. This is totally consistent with what we've seen in prior research where we don't see a response uh uh from the public  
[00:43:26] revelation at the institution level. Um but we do see a decline when we start to look at the infield degrees. So here I have the institution level degrees on  
[00:43:39] your left uh and the infield degrees on your right. This is uh the whole time period. And just zooming in on those uh infield degrees is where we start to see that decline over time. Uh and if you  
[00:43:52] look at as a summary statistic, I'm looking at the impact four years after the incident is publicly revealed and so there's a decline of 4.8% which is marginally statistically significant.  
[00:44:04] Yes, Emily. >> So two questions. One is are the standards kind of expanding on the ends because it's an unbalanced unbalanced panel? Yes, >> just sanity check. And then question number two, I'm wondering to help with precision is if you could get almost a  
[00:44:18] weighted estimate where you weight by how often the faculty member has interacted with say freshman year students. So you can imagine some random dude who advises five students who cares, but if they teach a 300 person intro class, I've heard of the guy, I'm much more likely to switch. And so it'll be interesting to see if that exposure  
[00:44:33] matters. >> Yeah. So we we don't have that like fine grain information about about faculty in our data, but we could think about ways to proxy for it.  
[00:44:43] Um I'm going to continue for forward just because of my time being limited.  
[00:44:47] What happens next is we were thinking about okay like uh what is driving this?  
[00:44:52] What is different uh over time? Is there a difference between early incidents and later incidents? We end up splitting our time to 2015 and after uh and uh 2014 and before. Some of you might remember  
[00:45:06] presidential election in 2015, the campaign in 2015 where uh misconduct uh towards women became very prominent and then the me too movement uh uh happening publicly. Uh so that's where we got our split. It also is convenient because  
[00:45:21] then we have a long enough panel uh in the second period to look at that full-time period. But we end up seeing basically no impact in uh the beginning of our uh panel. But when you focus on  
[00:45:34] the what we're calling the me too era is where you see a larger decline and the sort of total impact after four years comes to a decline of 9% in terms of degree completions. I should say all of  
[00:45:46] this is in uh log uh log points. So uh in the in log terms um women and men tend to experience this overall decline in similar ways. My prior was that women would have a bigger  
[00:45:59] decline than men but that turned out not to be true. The decline for women is 9%, the decline for men is 8.6 uh% though a a little less precise uh there. Okay. So what is driving this greater response in  
[00:46:14] recent years? Has something sort of changed or is it about sort of a change in mores that is happening with the me too movement? A change in uh how people respond and how consider how people  
[00:46:25] consider uh sexual misconduct. Um it is the case that consequence uh severity has increased over time. So there's more likely to be serious consequences. It's also the case that more serious incidents have more serious  
[00:46:39] consequences. It is the case that students respond more to the more serious uh uh consequence cases. But that only happens in the recent period.  
[00:46:50] So that points us towards maybe there is something different going on in that that recent period in terms of how people respond.  
[00:46:57] Uh it is the case that there's more news coverage in the recent period but something that we observe is that and though this is not statistically significant uh if anything we see greater response to the lower news  
[00:47:10] coverage uh events which again we think is because the high news coverage events are suppressing completions in that field uh more globally and so our comparison institutions are suppressed  
[00:47:22] as well and we don't see a difference in terms of response by the number of victims. So um another thing that we look at is whether this is sort of driven by the department and like maybe this is a mechanical thing that is  
[00:47:37] happening because the department has fewer faculty than it did before. So basically we look at the difference between small departments and large departments. We define a small department as under 40 degree completions. That's about the 75th  
[00:47:51] percentile of average graduates across years. um and uh in the early period we see a larger response in the smaller departments. So this is in the early period is when we see no overall response there is a response in in these  
[00:48:05] uh uh smaller departments which is consistent with either there being more active information sharing at that time or there being this mechanical capacity constraint. In the recent period we don't see any differences between large  
[00:48:18] and small departments. The response there is the same in terms of declines in degree degree completion which to us is suggesting that whatever is driving this difference where the uh more recent period is where we see more response  
[00:48:32] it's not just a a factor of uh faculty capacity.  
[00:48:37] Okay. So where do students go if they are leaving their chosen majors? we see that they are more likely to exit majors dominated by men. Uh and uh uh uh that  
[00:48:50] is uh the panel on the right there. Interestingly, the people who leave the male dominated majors are men more so than women. We see a greater response among men in terms of leaving male dominated majors. And our data can't  
[00:49:03] speak to why that that happens. Is this about reputational consequences and not wanting to get lumped in? Is this about uh awareness? Um uh is this about um you know just consciousness raising unclear  
[00:49:18] what is happening >> the Oh please.  
[00:49:21] >> Yeah. So >> it might be that it's easier to switch from majors dominated by men like okay you're switching say from physics >> to maybe I don't know physical chemistry  
[00:49:35] or whatever than it is maybe if you're doing history to want to switch to English I'm just maybe >> or um or from history to physics yeah yeah so so that um that's actually an an excellent point that I hadn't  
[00:49:50] thought of before so thank you for that uh I think that is likely true and we can probably uh get some traction on that. Um because I only have six minutes left, I'm going to bust on forward.  
[00:50:00] Um so we think that this should have consequences in terms of the predicted earnings of young people because it is the case that um male dominant majors tend to have higher expected earnings.  
[00:50:14] You can see computer science and engineering there. Uh I highly recommend nursing as a career for women. My husband is a nurse and just random fact, male nurses actually get paid more than female nurses. Um uh even in even in nursing uh but that's a great career.  
[00:50:28] This one dot up there is like basically like medical school but we have very few medical programs at the undergrad level which is why it's there. So you see this pretty strong gradient. But when you look at predicted earnings, whether we either look at five years out using the  
[00:50:43] college scorecard or at predicted lifetime earnings using Doug Weber has like a great table of of this that looks by major, there's no impact on earnings.  
[00:50:54] And that's because while we do see people shifting out of these high earning maledominated majors, they go in some cases to relatively high earning uh majors. So we see a decline in STEM, but  
[00:51:08] we see an increase in business and econ. The other place we see a very clear decline is in the arts. Uh and as you might remember, that's where we see sort of a disproportionate response. It's also the case that the arts are very  
[00:51:22] poorly paying uh uh fields. So basically what we see is people are willing to give up both highly compensated careers and also their creative pursuits to avoid the potential of encountering this  
[00:51:34] type of misconduct. Um did you have a question?  
[00:51:41] >> No, no, no. >> Oh, quick question. So you had something about Oh, I'm loud. Um you asked about male-dominated majors. Yeah. Did you at all think about like the composition of the faculty themselves? Like  
[00:51:54] >> we did and you um um but we don't have that information at a granular granular level for um uh fields by institutions.  
[00:52:06] We could probably do something at the field that just categorize the fields.  
[00:52:10] >> You might get like a arc data. Yes. I have that.  
[00:52:14] >> Yeah. Yeah. Yeah. So no, that would be that would be great. Um, it might also be that we don't see a lot of variation there because even, you know, I taught at an ed school for many years and somehow we still managed to have uh more men in the room than women. Um, all  
[00:52:27] right. So, uh, just to make you a little more confident in uh, uh, what I've showed you here, um, a few robustness checks at the end. We do a randomization inference procedure where we basically take each of our stacks. We remove the  
[00:52:42] treated case and then we randomly choose another case to be um the uh the incident you so uh I should say there's the idea that uh uh standard errors might overreject when there's a small  
[00:52:55] number of treated institutions. Um uh so we do that we per uh we do a 100 permutations of doing that. uh and it turns out randomization inference uh confirms the cluster robust results and if anything our results are sort of more  
[00:53:09] statistically significant when we do the the randomization inference you'll see that our uh estimates are really are in the the tail of the distribution this is the estimate and that's the uh t statistic as I've mentioned before uh  
[00:53:23] it's not known for 892 it's known for 92% of incidents uh so timing is known for 92% of the incidents um uh about uh 8% occur in the same year as the report, 20% one year prior and another quarter  
[00:53:38] uh 2 to 3 years prior and then there's just a long tale of uh when the incidents become known otherwise some become known like a decade later.  
[00:53:48] >> So we just shift >> so fast.  
[00:53:50] >> Yeah. >> So can you rule out that this is because it's driven by the timing of when it becomes publicly known which is closer to the timing of consequences. Can you rule out that this is just like losing a faculty member is bad for departmental health?  
[00:54:02] >> That's what we tried to do with the department size idea that um you know seeing the same response in small and large departments means that um a large department has a better chance of recovering from that. A small department doesn't and in recent years we didn't  
[00:54:16] see a difference. Um so we don't see any response to uh the incident itself only to the public revelation.  
[00:54:25] So you you mentioned uh previously something about the seriousness of the consequences. That's a seriousness of the consequence to the perpetrator.  
[00:54:34] >> Yes. >> Uh because one view of the world is that oh we figured out who the bad apple is.  
[00:54:39] You know the person left the department now everything is good. Of course, the other >> interpretation is that, you know, this is a bad environment and uh you know, even if we got rid of one bad apple, there could still be a lot of other >> and so another interpretation also like  
[00:54:53] about the difference in gender that we see is maybe that like women have a greater expectation of misconduct going on and thus they're more resilient to it it happening. So they >> but I didn't understand what the you know the heterogeneity by serious and of the conference.  
[00:55:07] >> Okay. Well, the I mean uh we do see like greater response to the more severe uh incidents in or se severe outcomes in the more recent period. Um it's not statistically significant. It  
[00:55:21] you know it is is present. I I have 24 seconds. So um uh the thing two things I want to say. So one is uh the muted response to the incident itself. I don't want anybody to take that to mean we  
[00:55:35] should not have public reports. We should have public reports because the direct harm to individuals is more important than any changes to the majors that that young people might have. Uh we do a bunch of different robustness checks and they're they're awesome. You  
[00:55:49] can look at them in the paper. Uh and we just wanted to end with the idea that by expanding sort of a vision of the consequences of these incidents, we can see consequences that are not apparent when you look at the macro level and  
[00:56:04] that it is like very important to think about those consequences and the futures that young people aren't going to have because of the incidents of misconduct out there. And I just want to end by saying that all of us are in a position of power. If you're sitting in this  
[00:56:17] room, you're here because of uh likely being a faculty member um uh or potentially a future faculty member and and we have the power to sort of change uh the environments in our departments