Incomplete answers and scattered thoughts on school violence, civil rights. With graphs.

Also something about AI that you don't have to read.

I.

Some parents in Wauwatosa worry about violence in schools. Others think this concern is overblown, or they think it’s just a “coded” way to be racist and disparage Black students, or they say that it’s not a problem unique to Wauwatosa but part of a statewide or national increase in student misbehavior, or (somewhat confusingly) all three.

Among the worried, some feel the problem has become sharply worse following the pandemic but that perhaps it began in earnest even earlier. Others attribute the uptick in violence to ubiquitous social media use, relaxed disciplinary standards partially motivated by administrator and school board members’ ideological aversion to punishing students who break rules, or something something unequal distribution of non-resident students.

Some support for the claim that this is not a Wauwatosa problem but a regional or nationwide problem include articles like this plus 80,000 articles about how kids don’t know how to interact with other human beings anymore. Some support for the claim about relaxed disciplinary standards include public comments from current and former teachers during school board meetings that they were instructed not to report infractions and school board member Mike Meier’s claim that Superintendent Demond Means had been encouraged by another board member to stop bringing so many students to the board for expulsion. Some support for the idea that people think it’s all ideologically driven include the numerous denials (presumably prompted by something) from Mr. Means that actually, no, he and other administrators don’t hate the police and do wish there were more of them in schools than there currently are.

I can’t really resolve any of the questions implied above. Instead, I’ll try to make things a little more confusing by looking at Wisconsin Department of Public Instruction data on pre-pandemic suspensions and federal civil rights laws.

II.

The Wisconsin Department of Public Instruction hosts Wisconsin's Information System for Education Data Dashboard (WISEdash) with information on test scores, graduation rates, school finances, attendance, special education, something called Digital Equity that I didn't click on, and discipline for every school district in the state. Unfortunately, the online interface is not great and they make it almost impossible to look at trends over time. WISEdash kindly acknowledges this problem but their proposed solution is just to open up multiple browser windows with different years of data and look at them side-by-side.

Another problem is that the State changed its data entry system in 2016 from something else that was slightly incompatible, and you can’t even view data prior to that. However, you can download giant excel spreadsheets, so I downloaded the last 13 years of disciplinary data submitted by each school district on suspensions, expulsions, and various infractions and tried to fit them together and graph them.

[Note: I am not a great programmer and had originally written a long digression extolling the benefits of a recently released AI chatbot called ChatGPT that saved me probably a dozen hours of fiddling and screwing around, but I’m not sure most people are interested, so I put it in a footnote.¹]

I was able to combine yearly disciplinary from 2007-2021. I focused on out of school suspensions (OSS) because expulsions are rare, there would be a lot of zeroes, and suspensions are severe enough that I think they serve as a meaningful proxy for some underlying level of misbehavior and discord within schools.

Some caveats:

• There can be lots of a reasons to suspend a student that don’t involve fighting or violence. There is separate DPI data on specific infractions, and maybe I’ll look at that next.

• Suspensions obviously decreased in 2020 because students weren’t in school for part of the year.

• People probably care more about the data from 2021-2022 and 2022-2023 but it’s not available yet.

• There are holes in the data although it seems to have gotten more consistent and complete since 2016.

Nevertheless, we can see if suspensions had been rising or falling prior to the pandemic and ask what this might mean. Falling numbers of suspensions might indicate fewer suspension-worthy offenses or that previously suspension-worthy offenses were considered less suspension-worthy. Rising numbers might indicate the opposite (although I don’t think anyone is claiming that schools have become more strict or more likely to suspend students). I plotted out-of-school suspensions (OSS) by gender, race, and disability status, compared the middle and high schools in Wauwatosa, and compared Wauwatosa’s suspension rates to surrounding districts.

Compared to nearby districts, Wauwatosa suspended a middling percentage of its students—more than Elmbrook, less than West Allis, similar to Waukesha (though with less variation)—and this remained fairly constant over 13 years.

Boys get in trouble more often than girls. News at 11:

Wauwatosa suspends students with disabilities at higher rates than those without:

It seems like suspensions declined among both Black and White students especially in high school but I wouldn’t attach much significance to this.

Of course, absolute numbers don’t account for the relative proportion of the student body comprised of Hispanic, White, and Black students. Between 2016-2019, about 63% of students were White, 18% Black, and 8% Hispanic. So you can divide number of suspensions by enrollment numbers to get rates at which each group were suspended. There is maybe a slight decline in the rate of suspensions at the high school level and a lot of variation at the middle school level.

You’d think data for the district as a whole would be more complete, but there are strange holes in the data for reasons I don’t entirely understand. Black students are suspended at 2-3x the rate of Hispanic students and 4-5x the rate of White students.

Overall, I find this mostly unsurprising. Suspension rates bounce around a lot but seem mostly constant. If you squint hard enough you might notice a decline but it’s small and gradual rather than large and sudden. And it could just be noise.

Despite the lack of exciting trends or smoking guns, I think it makes certain theories less likely to be true. For instance, if you thought there was, like, a conspiracy among administrators to just stop enforcing rules, you might expect a sudden, large drop in suspension rates. But in order to fit the data you’d have to posit that the overall number of suspension-worthy offenses increased at exactly the same time and in exactly the right amount to leave the relative trend unchanged. I suppose this is possible but it seems more likely that things just didn’t change much.

The fact that Black students are suspended at higher rates than Hispanic students who are suspended at higher rates than White students also isn’t so surprising, because people have been pointing this out for years, including the federal government.

III.

One reason I was curious to see if there had been a sudden drop in suspension rates was because in the beginning of 2014, the Department of Justice’s Civil Rights Division and the Department of Education’s Office for Civil Rights issued a “Dear Colleague Letter on the Nondiscriminatory Administration of School Discipline” noting that “African-American students without disabilities are more than three times as likely as their white peers without disabilities to be expelled or suspended” and that “significant and unexplained racial disparities in student discipline give rise to concerns that schools may be engaging in racial discrimination that violates the Federal civil rights laws.”

I imagine that school districts really don’t want to violate or even appear to potentially violate Federal civil rights laws. For instance, Milwaukee Public Schools suspended Black students at much higher rates than White students and this led to a multi-year investigation and lots of corrective actions that are probably even more unpleasant to implement when the federal government is looking over your shoulder.

The letter cites some research to support the claim that these disparities are at least partly due to discrimination although it seems like as studies become more rigorous and better account for prior disciplinary history, poverty, and severity of infractions, the effect of discrimination becomes smaller and smaller. An even more recent and more high quality study also finds relatively small effects from discrimination, and they estimate that racial discrimination adds about .05 days to an average suspension of 2.9 days [Although, maybe some contrary evidence here.]

Importantly, even if a policy is applied equally to everyone, schools can still be found guilty of “disparate impact” if some races are effected more than others.

Schools also violate Federal law when they evenhandedly implement facially neutral policies and practices that, although not adopted with the intent to discriminate, nonetheless have an unjustified effect of discriminating against students on the basis of race. The resulting discriminatory effect is commonly referred to as “disparate impact.”²¹

The letter then offers numerous “suggestions”—like implementing restorative justice practices or social emotional learning programs—for schools to avoid civil rights violations and multi-year investigations. At the same time, they also make it clear that you can’t just assume that if you do these things, everything will be okay:

They are not intended to be exhaustive or exclusive; do not address recommendations specifically targeted at preventing discriminatory discipline that is based on prohibited grounds other than race, color, or national origin; and may not be applicable to every specific factual setting in a particular school.¹ Additionally, these recommendations do not constitute legal advice, and schools that choose to implement one or more of these recommendations might still be found to be in violation of Federal law(s).

I suspect this type of legal ambiguity is a recipe for risk-averse bureaucrats to implement policies that inevitably err in the direction of “doing everything and more.” If the federal government says, “You should really try to prevent this thing from occurring and here are some suggestions we think might work and show you care but also it’s not legal advice and we don’t guarantee you won’t get in trouble anyway,” you will do all the suggestions and more to put yourself as far away from any potential legal liability as possible.

I’m not saying this is what is happening (I have no idea), but I do think people sometimes overestimate the ability of bureaucrats, functionaries, and administrators to enact their personal preferences—especially when we disagree with those preferences!—and underestimate the extent to which they’re often captive to forces beyond their control and are just trying their damnedest not to get in trouble.

1. I am a very slow programmer and do it only sporadically. Every time I do it again it takes forever because I've forgotten everything I learned last time, have to look up commands, trawl through ten year old forum posts when I get an error I don't understand, and read a bunch of tutorials trying to find something that approximates the problem I'm trying to solve. It's all very painful.

ChatGPT is the newest AI from the not-so-non-profit OpenAI. Scientists gave it a large chunk of the internet and created a giant statistical model that helps it predict the next word in a sentence. This doesn’t sound like it should help you with something like programming but apparently if the model is large enough it becomes capable of doing, frankly, kind of astonishing things. It can pass the bar exam. It can write articles, describe the plot of a book, synthesize arguments, and I guess write code given somewhat nebulous natural-language prompts.

My prompt:

I have a number of csv files, each for a different year, and they all have these columns: SCHOOL_YEAR AGENCY_TYPE CESA COUNTY DISTRICT_CODE SCHOOL_CODE GRADE_GROUP CHARTER_IND DISTRICT_NAME SCHOOL_NAME GROUP_BY GROUP_BY_VALUE REMOVAL_TYPE_DESCRIPTION TFS_ENROLLMENT_COUNT REMOVAL_COUNT and I would like to combine them all. How do I do that?

And it said, "Sure! Here's some code for one way to do this.”

And it wrote a short program that did exactly that. I ran it and it worked flawlessly. This isn't a complicated programming task. But it would have taken me a few hours as I sorted through different potential solutions and tried to think about the best way to do it so I didn't go down the wrong path and spin my wheels for hours. With ChatGPT it took me a few minutes to formulate the question and about 30 seconds for the software to generate a response.

Another example: instead of searching for a command online and translating it to my own situation, I just asked ChatGPT:

Importantly, this probably won't help someone who doesn't know anything about programming but it will help idiot programmers like myself.

Anyway, it probably saved me at least 10 or 12 hours. I find it hard to overstate how remarkable this is. I can think of half a dozen things that I wouldn’t have had the time to tackle before but could probably accomplish with the help of my AI friend.