Cassell’s Empirical Call To Be Unconstitutional

There are few things that makes us feel more justified in being authoritarian than an empirical study, all wrapped up statistical-ish words. We don’t know quite what he means, and hang on the technical jargon so the weasel words connecting the parts that don’t move well together can be easily overlooked. This is no more true than when there is an evil villain to blame for the death of human being, and Paul Cassell doesn’t disappoint.

As the Chicago Tribune reported this morning, University of Utah Economics Professor Richard Fowles and I have just completed an important article on the 2016 Chicago homicide spike. Through multiple regression analysis and other tools, we conclude that an ACLU consent decree trigged [sic] a sharp reduction in stop and frisks by the Chicago Police Department, which in turn caused homicides to spike. Sadly, what Chicago police officers dubbed the “ACLU effect” was real—and more homicides and shootings were the consequence.

It’s like a perfect storm, homicide spike, “multiple regression analysis and other tools,” and of all the people who could possibly be responsible, it’s the ACLU! The ACLU? Who doesn’t hate them? It’s even got a cool name, the “ACLU effect.” How could this not be right?

The analysis is relatively straightforward. It is well known that homicides increased dramatically in Chicago in 2016. In 2015, 480 Chicago residents were killed. The next year, 754 were killed—274 more homicide victims, tragically producing an extraordinary 58% increase in a single year. What happened?

What happened is, indeed, straightforward. There was a spike in homicide. That, however, isn’t analysis, just a fact. A fact worthy of a front page headline.

Well, that won’t outrage the unwashed much. So where’s the beef?

In our paper, Professor Fowles and I bring empirical research tools to bear in an attempt to identify what changed in Chicago during that time.

Appeal to expertise, plus “empirical research tools” (don’t worry your silly head about what these are. They’re “empirical,” so they’re science!) and then, the big wiggle.

While such analysis may be unable to provide absolutely definitive answers, it can suggest which factors are more likely than others to have been responsible.

By using the hyperbole of “absolutely definitive,” Cassell seeks to upend criticism as if only a complete hater would demand perfection. But his alternative to “absolutely definition” turns out to be “factors…more likely than others.” Not absolute. Not definitive. Not even remotely provable. Merely “more likely.” Maybe his empirical tools were rusty?

We next attempt to pinpoint the time when things changed in Chicago—what might be called the “inflection” or “break” point in the data series. We begin by seasonally adjusting Chicago homicide and shooting data, which show significant seasonal fluctuation from cold weather months to warm weather months. Once the data are seasonally adjusted, a change or “break” in the data series can be statistically detected around November 2015.

In less pretentious terms, when did the killings start? And this will surprise no one.

Good reasons exist for believing that the decline in stop and frisks caused the spike. Simple visual observation of the data suggests a cause-and-effect change. In the chart below, we depict the (seasonally unadjusted) monthly number of stop and frisks (in blue) and the monthly number of homicides (in gold). The vertical line is placed at November 2015—the break point in the homicide data. This is precisely when stop and frisks declined in Chicago. (Emphasis added.)

Don’t mutter correlation doesn’t prove causation when a “simple visual observation” does the trick. Cassell is using science, which you shouldn’t try at home or you could sprain something.

We then qualitatively search for other possible factors that might be responsible for the Chicago homicide spike. For various reasons, none of these other candidates fit the data as well as the decline in stop and frisks. In addition, Bayesian Model Averaging (“BMA”) provide strong statistical evidence that our findings are robust in the sense that they are not due to inclusion or exclusion of any particular variables in our equations.

Don’t ask what other factors, or why they don’t fit. He’s using Bayesian Model Averaging, and isn’t that good enough for you? And if that’s not persuasive enough, it not only costs lives, but money.

Our equations permit us to quantify the costs of the decline in stop and frisks, both in human and financial terms. We conclude that, because of fewer stop and frisks in 2016, a conservative estimate is that approximately 236 additional homicides and 1115 additional shootings occurred during that year. A reasonable estimate of the social costs associated with these additional homicides and shootings is about $1,500,000,000. And these costs are heavily concentrated in Chicago’s African-American and Hispanic communities.

And, indeed, even the ACLU says it’s to blame.

In our penultimate section, we explain why the ACLU settlement agreement with the Chicago Police Department is the most likely cause of the decline in stop and frisks. Indeed, the ACLU took credit for the decline in stop and frisks at the start of 2016.

As much as there is little in Cassell’s highly scientific study that’s persuasive, there is one rather significant “factor” that he neglects to mention. The tactic of “stop and frisk” was stopped because it was flagrantly unconstitutional. It was nothing more than cops deciding to randomly toss people for looking suspicious or for kicks. There’s no mention by Cassell of how many people were stopped, how many guns were seized, and how many constitutional rights were violated.

Is it likely that a facially unconstitutional tactic was effective, at least to some extent? Of course. If cops were allowed to make random house and car searches, there is a likelihood they would find guns and drugs as well. Randomly search enough people and you’re sure to find something eventually.

But wrapping up an unconstitutional tactic, and blaming the ACLU for pursuing an end to an unconstitutional tactic, in statistical jargon is asinine. And it’s doubly so when the best he can come up with is “more likely.” There was no “control” for Chicago cops being compelled to do their job while respecting the Constitution.

Or their sad feelings when they were told that they couldn’t toss black kids against walls anymore for kicks, so they decided to cash their paychecks and do as little policing as possible, ceding the streets to the gangs that festered for self-protection against the biggest chicago gang of all, the one that murdered Laquan McDonald and covered it up. Talk about regression analysis.

29 thoughts on “Cassell’s Empirical Call To Be Unconstitutional

  1. Skink

    I’m very heartened to see the results of this study, as it confirms my recent work along the same lines. My dog (yes, she still occasionally shits on the floor) and I undertook some rather elaborate testing to determine why I tried fewer cases in recent, as opposed to past, years. Using a radiator from a ’77 Chevette and a spoon, we found the cause to be flowers. Actually, the prime generator seemed to be rain, but that was considered a statistical anomaly. After all, rain makes flowers, right?

  2. Ahcuah

    Cassell also seems to be assuming that when the Chicago Police were ordered to cease doing the stop and frisks that they actually did so. Based upon what we’ve seen before about their behavior, is it not likely that they simply stopped reporting the stop and frisks while still conducting them? That would rather screw up his beautiful vertical line.

    1. SHG Post author

      But he said it’s “precisely” when they stopped. Like throwing a light switch, or it wouldn’t be precise at all.

  3. LocoYokel

    Why do you hate science so much?

    He’s got graphs and everything. How much more proof do you need? We obviously need to start having cops start tossing all blacks and hispanics twice a day, every day. While you’re at it, let’s bring back phrenology. Now there was a sound, scientific technique to know who was a criminal.

    Commie pinkos like you and your “Bill of Rights” are why the police can’t properly defend our cities from the criminals running amok.

  4. DaveL

    Notice he says “seasonally adjusted” i.e. adjusted for season, not adjusted for weather. I checked, and as it turns out, November 2015 was 10 degrees warmer on average than November 2014. However, under “seasonal adjustment”, both would have had the same correction applied.

    But of course, such mathematical number-torture miss the point. Justice is not the same thing as statistically optimal crime reduction. The latter could be achieved simply by eliminating all criminal laws, or it could be approximated by keeping everyone in cages full-time, but neither would be justice. Similarly, throwing young men against the wall for being black is unjust, no matter how statistically useful it may be.

    1. MonitorsMost

      You can read the paper. He adjusts for season and then tests temperature as a possible reason for the spike.

      I agree with the second point though. Fine, tossing young black hoodlums in high-crime areas might lower gun homicides. The Constitution prohibits a whole bunch of effective law enforcement tactics. Cost of doing business.

  5. Richard Kopf


    I have not fully studied the paper published by the professors in the SSRN and recounted in your post and the Chicago paper. I don’t intend to do so either.

    However, the big red flag to me is this statement in the abstract of the SSRN paper that reads “[t]he “contrary experience in New York City may be an anomaly.” That is, New York reduced stop and frisk and so did Chicago. If the reduction in stop and frisk were causal one would expect both cities to suffer a rise in homicides. Apparently, that did not happen in New York. To call that difference an “anomaly” understates a potentially devastating flaw in the analysis in and the conclusion of the paper.

    The foregoing said I agree with the underlying sentiment of the research. The ACLU and the courts are not well-suited to imposing system-wide public safety policies on huge police departments. On the other hand, the problem, of course, is the one you stress. Improperly run and supervised stop and frisk policies–those than run amock–grievously violate the Fourth Amendment and the Fourteenth Amendments rights of our fellow citizens.

    All the best.


    1. SHG Post author

      Maybe NY is the anomaly. Maybe the second city is the anomaly. Maybe both. Maybe neither. That’s statistics for ya.

    2. Boffin

      “The only possible interpretation of any research whatever in the ‘social sciences’ is: some do, some don’t.”
      ― Ernest Rutherford

  6. Jim Ryan

    Now if only Cassell had a Law Degree, then he could have been both statistical-ish and legal-ish, my head hurts thinking about it.
    Oh and I could use that ’77 Chevette radiator.

    1. Skink

      That dog is my pal–she really digs me. To keep me happy, she turned the radiator into a still. It makes fine rot-gut.

        1. Skink

          Being without me brings sadness, and the batches have been smaller lately. It’s probably time for another empirical study. I’ll begin with the conclusion that batches are smaller because pythons from the Everglades are stealing my hooch. That’s easily provable.

  7. Jyjon

    ‘University of Utah Economics Professor Richard Fowles and I have just completed an important article’

    Delusions of grandeur much?

    1. SHG Post author

      If it’s important to him, shouldn’t that make it important to the world? It’s important, damnit.

      Still waiting for somebody to notice whose name appears on the top right of the Marsy’s law website. I expected that to be in the first comment, second at the latest.

      1. Charles

        So did you cancel your subscription to the Times now that you have a new, apparently limitless source of inspiration?

        1. SHG Post author

          Hardly new, Charles. I’ve had an eye on Cassell for quite a while, and there are numerous posts here about his ideas over the years.

  8. Fubar

    Cassell is almost excusable for not knowing from Shinola about statistics. But an economics professor, not so much.

    The following, and countless other spurious correlations, can be shown in data from impeccable sources:

    US spending on science, space, and technology correlates with Suicides by hanging, strangulation and suffocation. Correlation: 99.79% (coefficient of determination r=0.99789126). Data sources: U.S. Office of Management and Budget and Centers for Disease Control & Prevention.

    Number of people who drowned by falling into a pool correlates with Films Nicolas Cage appeared in. Correlation: 66.6% (coefficient of determination r=0.666004). Data sources: Centers for Disease Control & Prevention and Internet Movie Database.

    Per capita cheese consumption correlates with Number of people who died by becoming tangled in their bedsheets. Correlation: 94.71% (coefficient of determination r=0.947091). Data sources: U.S. Department of Agriculture and Centers for Disease Control & Prevention.

    I won’t provide the URL for these and countless other spurious correlations (complete with neat graphs), because rulz.

  9. Richard Kopf


    I suspect that your “spurious correlations” are based on simple linear regressions using only one explanatory variable. In fairness to the professors, their analysis is quite a bit more sophisticated.

    As for Nick Cage, the only movie worth seeing involving Nick is “Leaving Las Vegas.” Cage stars as a suicidal alcoholic who has ended his personal and professional life to drink himself to death in Las Vegas. This worthy goal is ultimately achieved. But before he kicks the bucket, he develops a relationship with a hardened prostitute played by Elisabeth Shue, and that relationship forms the center of the film.

    Cage won an Academy Award for Best Actor for Leaving Las Vegas. Now, to be clear, any attempt to correlate that award and the quality of Cage’s 72-movie career would, in fact, be spurious. My coefficient of reproducibility (r-squared) for that statement approaches .9999.

    As qualitative (not statistical) proof, please watch “Left Behind.” Plot: An airline pilot (Nicolas Cage) and a journalist (Chad Michael Murray) are among those who must try to make sense of a prophetic event causing millions of people to suddenly vanish. So bad, it is almost good.

    All the best.


    1. Fubar

      I suspect that your “spurious correlations” are based on simple linear regressions using only one explanatory variable.

      You are correct that my examples were simple linear regressions. They illustrate a principle.

      In fairness to the professors, their analysis is quite a bit more sophisticated.

      The principle is simple. Extended to “more sophisticated” multivariate linear or even nonlinear regressions: pick your independent variables carefully, and pick your regression function (if nonlinear regression) carefully, and you can show all manner of oddball correlations. But nobody in his right mind would consider the correlations to demonstrate anything more meaningful than (cherry)picking variables or functions.

      In fact, the cited Chicago Tribune article noted that NYC “homicides remained low even when the number of stop-and-frisks fell sharply.” That, of course prompted the Utah researchers to argue that “that comparisons with New York seemed “dubious,” in part because of Chicago’s far higher rate of gun crimes.”

      That, at least, is implicit admission that the Utah researchers consider their results to be meaningful only in Chicago, and not possible to generalize to “stop and frisk” policies elsewhere.

  10. LTMG

    About “absolutely definitive”. Statistical science is unable to prove anything definitively or absolutely. There is always mathematical error in the result. What statistical science Can do is to positively state that with some precise probability an assertion is true or false. For example, the results of a study might conclude, “With 98.2% reliability, the hypothesis is true.”

    In summary, mathematical statistics cannot absolutely say that an assertion is true, but can Very strongly say that an assertion is true with at least a certain probability.

    As a side note, if testing the DNA of a child versus a parent, the test results Will have a certain tiny error that is possible to calculate.

  11. B. McLeod

    Clearly, the Chicago police need to throw off the yoke of the evil ACLU and get back down to business. They can simply cordon off all the high crime areas, and establish checkpoints where citizens entering or leaving strip down and pass through a metal detector. The murder spike should be flattened and citizens should be grateful that their neighborhoods will be statistically happier places.

  12. CAB

    It’s almost as though Cassell and Fowles have never read Weber’s science essay; there’s a reason we still teach that to undergrads. (As a silver lining to this mess, though, they’ve given us theory instructors a fantastic new illustration of his point.)

Comments are closed.