Michael LaCour in 20 years

In case you were wondering what “Bruno” Lacour will be doing a couple decades from now . . .

James Delaney pointed me to this CNN news article, “Connecticut’s strict gun law linked to large homicide drop” by Carina Storrs:

The rate of gun-related murders fell sharply in the 10 years after Connecticut implemented a law requiring people buying firearms to have a license, according to a study. . . . To assess the effect of this law, researchers identified states that had levels of gun-related homicide similar to Connecticut before 1995. These include Rhode Island, New Hampshire and Maryland. When the researchers compared these states to Connecticut between 1995 and 2005, they found the level of gun-related homicide in Connecticut dropped below that of comparable states.

Based on the rates in these comparable states, the researchers estimated Connecticut would have had 740 gun murders if the law had not been enacted. Instead, the state had 444, representing a 40% decrease.

Wow—40%, that’s a lot! And, indeed, Storrs has a quote on it:

“I did expect a reduction [but] 40% is probably a little higher than I would have guessed,” said Daniel Webster, director of the Johns Hopkins Center for Gun Policy and Research who led the study, which was published Friday in the American Journal of Public Health.

A legal expert named Daniel Webster, huh? And I guess it’s good they have someone named Storrs writing articles about Connecticut.

Anyway, that’s a funny quote from the leader of the study! Perhaps the reporter should push a bit, maybe ask something like: Do you really believe the effect is 40%?? Or do you think that 40% is an overestimate coming from the statistical significance filter and the garden of forking paths?

OK, this is all important stuff. But it’s not the subject of today’s post.

Here’s the deal. Storrs continues her article:

Ten states have laws similar to Connecticut’s, including background check requirements. It is hard to know what effect permit-to-purchase laws have without looking in these other states, said John R Lott Jr., president of the Crime Prevention Research Center, a gun rights advocate and columnist for Fox News. “If 10 states passed a law, eight could increase and two could fall, and how do I know that it was because of the gun law?” he said.

Wha??? John Lott? CNN can’t find any real expert to interview? Why not just follow up with a quote from Mary Rosh, endorsing Lott as “the best professor I ever had”???

For those of you who don’t remember, John Lott shares with Michael LaCour the distinction of having announced, with great publicity, controversial data from a survey that he said he conducted, but then for which he could supply no evidence of its existence. Damn! I hate when that happens. As I wrote last month, Lott represents a possible role model for LaCour in that he seems to continue to be employed in some capacity doing research of an advocacy nature. And, like LaCour, Lott never admitted to fabrication nor did he apologize. (I guess that last part makes sense: if there’s nothing to admit, there’s nothing to apologize for.)

Ok, just on the statistics for a moment, Lott’s argument is terrible. First, “Ten states have laws similar to Connecticut’s” is not so relevant, given that the causal identification comes from the change in the law, not the existence of the law. Indeed, Storrs gets a good quote dismissing Lott’s argument:

Although Webster said he would like to study the effect of gun laws in other states, that research is not practical. Most states passed meaningful gun laws, such as laws requiring background checks, long ago, “frankly before I was born,” and it would be hard to know how those laws were enforced back then, and how society responded to them, he explained. In addition, information from death certificates was less readily available from the Centers for Disease Control and Prevention before 1980, he said.

Second, Lott says, “If 10 states passed a law, eight could increase and two could fall.” But that’s just ridiculous. Why suppose that introducing this law, which the data indicated was associated with a drop in homicides, would lead to an increase in 8 states out of 10?

It’s not that the Webster et al. claims are airtight. I’ve already expressed my concern that the estimated effect is too high, also Storrs alludes to evidence from other states that send mixed messages. And Delaney has a point when he writes, “One concern about the construction of the synthetic control is Connecticut’s proximity to and interconnections with NYC, which experienced a dramatic decrease in overall homicides from 1177 in 1995 to 539 in 2005 (according to Wikipedia). Whereas, from what I can tell, homicide totals while decreasing across the nation during this period, happened to be closer to constant in this period in New Hampshire, Rhode Island, and Maryland.”

But Lott’s criticisms are uninspiring. Let’s hope that Bruno Lacour can do better in his future career as an advocate and pundit, and let’s hope that news outlets can do better when looking for a quote. I heard John Yoo is available. . . .

30 thoughts on “Michael LaCour in 20 years

  1. Comparing to states with similar gun homicide rates at the time of the change in law is not a good idea.

    I think it’s pretty clear that homicide varies from state to state for various long-term cultural reasons. There will also be variation due to short term changes, or just Poisson-style chance. Suppose that Connecticut has a lower-than average long-term homicide rate (I haven’t checked to see whether this is actually the case, but the argument applies one way or the other unless Connecticut is right on the average). When it is matched with other states with similar homicide rates, many of those will be states with higher long term homicide rates, that have a lower rate at the moment due to short-term or chance variation. These states will then tend to move upwards to their long-term mean in later years. The result would be an entirely spurious appearance that Connecticut was doing better due to the new law.

    I also always find it suspicious when people look at “gun homicide”, rather than overall homicide, which is of course the outcome of actual interest.

    • “I also always find it suspicious when people look at “gun homicide”, rather than overall homicide, which is of course the outcome of actual interest.”

      I, for one, would certainly rather be shot than stabbed or bludgeoned to death.

    • I think what we want is a proper timeseries model where we estimate a poisson rate as a function of time rather than try to look at the raw data. if we do that, I’m quite sure we will have too little information to do a good job of estimating this function, and we certainly won’t see a 40% drop in the rate. Furthermore, as you say we should really match against all the regional states, all of New England probably, so as to avoid the regression to the mean issue you point out.

      Finally, it’s not surprising at all that people look at “gun homicide” because this is a hot-button topic. I suspect many people would be happy to eliminate legal gun ownership even if it meant an increase in overall homicide. In the wake of the Heller case there have been a large number of cases liberalizing local gun control laws with essentially no real effect on local crime rates certainly nothing like a 2.5x increase (1/0.4).

      • Finally, it’s not surprising at all that people look at “gun homicide” because this is a hot-button topic.

        And yet most people fail to realize that from a public health policy perspective, the most gains that could be reaped by restricting gun ownership would come from reducing the number of gun suicides, which account for 60% of gun-related deaths in the US. (There are good reasons to think that the availability of a quick and easy death technology is a cause of suicide, in the sense that suicide would just not be attempted in its absence.)

        • We definitely have to distinguish between making more restrictive gun laws, and actually reducing the number of guns in the hands of people likely to commit crimes with them. I think that’s the biggest issue in the US. Gun laws are very effective at removing guns from the hands of very low risk people, people who will never commit even a serious misdemeanor much less a felony assault or murder. The liberalizations that have come after Heller largely reduce restrictions on that large non-violent group, and hence have had no major increases in gun violence associated.

          The topic is so terribly charged that it’s hard to talk seriously about things like simultaneously liberalizing gun laws to make it easier for low-risk people to use them for defense, which is an important topic for pro-gun people, and cracking down on enforcement of the remaining laws so that high risk people don’t have such easy access which should be an important topic for everyone.

          Mental health is a whole second topic where we as a society aren’t having useful productive policy discussions, and the intersection of mental health and guns is a big important issue, I totally agree with your point there, while still disagreeing that somehow the solution is a general “restricting gun ownership”.

      • 40% seems like a very large claim, but then again http://www.disastercenter.com/crime/ctcrime.htm raw data for murder (not the same as all homicides) do show a sizable change. They say in the abstract that they use synthetic control.

        There is some thought that the displacement away from guns leads to a lowering of homicide total because the lethality of knives and fists is so much lower. Among other things you have to be in close physical proximity for a knife or fist to be effective but guns work at a distance. Also, if you did look at the 1995 data you would see that there is a distortion for Oklahoma caused by the Oklahoma City bombing which was treated as a set of homicides (unlike the World Trade Center attacks which were treated as terrorism).
        The real issue is that homicides in general have been in decline both in the region and nationally as has other kinds of crime, but it is very to know why. There is a whole little subindustry of people who argue about why crime in New York has declined so sharply.

        • The real issue is that homicides in general have been in decline both in the region and nationally as has other kinds of crime, but it is very [hard] to know why.

          Is there any reason to doubt that lead exposure (or lack of same) is responsible for the big swings in crime rate over the course of the 20th century? (Genuine question, not rhetorical.)

        • I think people are always justifiably dubious with simple explanations especially when the variable is also highly correlated with other kinds of disadvantage. I don’t doubt that it has a role at the individual level that can lead to aggregate change because of how individual exposure is geographically and socially concentrated.

          http://www.samefacts.com/wp-content/uploads/2014/04/Crime-Justice-2002-after-epidemic.pdf starting around the bottom of p. 23 has a discussion of some of the issues.

        • Corey – I am going to agree with Elin here.

          It is sort of a consistency check in childhood exposure and cognitive development studies that if blood lead levels are available one should see clear effects (but I would always work through the background science with a biologist.)

          That this could be worked up into evidence of effects on specific adult behaviors removing confounding from other cohort effects would be very very surprising. Maybe not impossible given there many be some very good long term birth cohort studies with good individual level data.

      • Daniel,
        Based on the abstract and this news account http://www.washingtonpost.com/blogs/govbeat/wp/2015/06/12/gun-killings-fell-by-40-percent-after-connecticut-passed-this-law/ I don’t think they are looking “at raw data” but anyway that idea of a Poisson time series with comparison data is basically what was done for the Boston Gun Project/Operation Ceasefire. It definitely seems as though gun violence could be very sensitive to interventions, but the real question is whether there is any long term staying power.

        • Looking at that “synthetic connecticut” graph (figure 1 in your linked story) it also shows all control states equally weighted, which looks a lot more like a scaled version of real connecticut. So it seems like their result is highly dependent on the weighting they use to generate their synthetic control. That definitely seems problematic, especially because they’re using a point-estimate for the control weights. How robust is that control? we don’t see anything obvious in the graphs at least.

    • I think they looked separately at gun and non-gun homicides. The idea was that the law should only affect gun homicides, so if there seems to be an effect on non-gun homicides (as it happens, there wasn’t) there might be something wrong with the model.

  2. Most of the research on gun reduction finds no substitution effect or a very small one. The person who winds up killing someone else does not, in most cases, set out that day intent on killing that person. The motive arises in the course of the interaction — an argument, a fight, a robbery — and the motive is usually not so much to kill but to get the other person to stop doing what he or she is doing. So the question is not whether you’d prefer to die from a bullet or a bludgeon. It’s whether you’d prefer to get in a fight with someone armed with a club or someone armed with a gun. So if gun laws do reduce homicide, it’s probably these more spontaneous killings. The people who do have murder in mind from the very start can probably find ways to get a gun regardless of the laws.

  3. “In case you were wondering what “Bruno” Lacour…”

    I don’t get it, is this is a joke? Are you trying to make a parallel between Micheal Lacour and Bruno Latour? Why?

    • Bees:

      From Wikipedia: “Bruno Latour has undertaken an ambitious analysis and reinterpretation of modernity, and has challenged fundamental concepts such as the distinction between modern and pre-modern, nature and society, human and non-human.” Seems like a pretty good description of what LaCour’s work too, no?

      • I thought it might be a Bruno Frey reference. LaCour did get a lot of mileage out of the same project (a job, $190K in fake grants, a reputation as a rising star, a fake award to decorate his CV …).

  4. Why wait 20 years when you can just fast forward a few months:

    Michael Jules LaCour, Data Scientist & Visualization Engineer

    “I am a data scientist with five years experience in applied statistics, media research, and predictive modeling. M.S. and Ph.D. training in statistics and political science. I founded, Beautiful Data Inc., a data science consulting company, specializing in data visualization & presentation. My graphs have been featured in The Economist, Washington Monthly, and The Washington Post.”

    Why anyone would hire a known liar and cheat is beyond me. But then again, I’m fairly certain that LaCour would face no ethical dilemma providing his clients with the answers they want to hear, regardless of what the data actually say. So, I guess getting caught cheating badly has been a net win for him!

    • Anon:

      I’m not myself planning to hire Lacour but I think you’re being a little harsh. Sure, there’s a downside to hiring someone whose work you can never trust. But Lacour has good skills. If you hire him, I’d just suggest requiring that he make all his code and data open. This gives him a disincentive to cheat, if all his work can be checked. It may well be that, now that the pressure is off, he can do good, honest work. The graph on his website is pretty impressive.

  5. Explaining a variable with a constant is the favored method of frauds and extremists.

    Pity (very slightly) poor Lacour, whose one insight was to add a bit of noise to someone else’s data. It sheds a bit of light on what the rest of us have long suspected about the quality of oversight by UCLA faculty, and about the faculty themselves. Not a great program.

    I do hate to say it, but Prof. Greene saw some stars in his eyes as well.

    If people want (as seems appropriate) to inundate the legit channels with the tale of the miscreant Lacour, IMHO Greene should figure prominently.

    • +1 about Greene getting off too easily.

      Partly, this was due to a very savvy PR move by Greene who very early in the scandal vigorously portrayed himself as the well-meaning, trusting, innocent dupe.

      The good-guy-bad-guy contrast narrative was appealing and most of the media & bloggers lapped it up gladly.

Leave a Reply to Elin Cancel reply

Your email address will not be published. Required fields are marked *