It’s time to play “Fisk the Bogus Research” again. Today’s offering:
Cooling-off period for handgun purchases save lives, research shows
Malhotra, Michael Luca and Christopher Poliquin authored a study which found that handgun waiting periods reduces gun homicides by 17% and gun suicides by 10%.
[…]
Expanding the waiting period policy to all other US states would prevent an additional 910 gun homicides per year without imposing any restrictions on who can own a gun.
Really? Let’s look at the study.
Waiting period laws that delay the purchase of firearms by a few days reduce gun homicides by roughly 17%
A bold claim. Does their methodology really support it?
the model for gun homicides omits three state-years, and the model for non-gun homicides omits two state years because the death count was zero
We’re done. I’m not not going to do a full fisk on a study where they admit they tossed valid data that didn’t support their conclusions. You don’t omit data because the number is zero. You count the zero. Deliberately omitting zero when computing an “average” will yield a higher average.
1+1+1+1=4. 4/4=1.
0+1+1+1=3 3/4=0.75.
Oh, heck. Let’s look at some other stuff. For instance, they found that firearm waiting periods reduced nonfirearm homicides and suicides (Insignificant, they say, but that they found a definite effect at all should have prompted them to examine it for cause). How, pray tell?
What they also didn’t address is how waiting periods could have such a dramatic effect when roughly 93% of firearms used in murders are obtained through channels which bypass background checks and waiting periods: theft, black market, straw purchases, friends & family, and even “found it at the scene of my crime” (seriously; that’s a fairly common answer in the inmate surveys).
According to the FBI, 10,982 people were murder with firearms in 2017 The CDC says 14,542. If only 7% percent of those murders were committed with firearms subject to checks or waiting periods, we have a possible 769 to 1018 victims who might have benefited from waiting periods. But only if they were killed with firearms obtained no more than three days prior to the murder. But we’ll work with it. Let’s pretend we could have saved 17% as our researchers claim.
That gives us a range of 131 to 173 lives which might’ve been saved. Not the 910 claimed. They’re only off by a factor of 5.26 to 6.95. Truly though, it’s worse than that. 16 states (and DC) have waiting periods. I’d have to sort murders by state to find the fewer who might have benefited.
Hint to researchers using models: If your model doesn’t match reality, change the model, not reality.
[Permission to republish this article is granted so long as it is not edited and the author and The Zelman Partisans are credited.]
(More Tip Jar Options) |
Hint to researchers using models: If your model doesn’t match reality, change the model, not reality.
Working in IT like I do, I have to tell the database people something like this, all the time.
In essence, the data is what it is. If your search query isn’t getting the results you want, you don’t need to change the data; you need to change your search query.
It’s the same with “research”. The data is what it is. If you want to change it, you need to document and justify exactly how you changed it and why. Yes, you’ll have two datasets at the end (one original and unaltered, and one altered for the purposes of the experiment). Other researchers reading your work get to check your reasoning and decide for themselves whether your justifications were valid. If not, your experiment and conclusion are worthless. If so, then having both datasets means the experiment should be repeatable (and repeatability is a hallmark trait of a valid theory and test model).
That’s how science works.
But the bottom line of all that is, if the test model doesn’t match the data, the problem isn’t the data. Fudging the data (without a DAMN good reason!) to fit a model is just another way to lie, and lies have no place in science except as targets to debunk.