Regression 1982q4-2018q3:
f = -18.52 + 0.028pop + 5.182trump
Adj.R2 = 0.18, N = 144, DW = 2.07, bold denotes significance at 10% msl using HAC robust standard errors.
Where f denotes mass shooting fatalities, pop is population in millions, trump is a dummy variable for Trump administration.
One can interpret this as follows: a Trump administration quarter is associated with 5.2 greater fatalities from mass shootings, or 20.8 on an annualized basis. (Over 1982q4-16q4, the average frequency per quarter is 4.876). Inclusion of a deterministic time trend yields a negative coefficient on population, and a trump coefficient (4.093) significant at 11% msl.
Figure 1: Cumulative mass shooting fatalities (dark red), non-fatal injured (pink), from 1982M08. Orange denotes 2017M01-2018M10, light orange 2016M11-2017M01. Source: Mother Jones, accessed 10/28/2018, and author’s calculations.
The corresponding regression for mass shooting wounded:
w = -13.84 + 0.022pop + 26.815trump
Adj.R2 = 0.13, N = 144, DW = 2.36, bold denotes significance at 10% msl using HAC robust standard errors.
Where w denotes mass shooting wounded. This can be interpreted as 26.8 greater mass shooting wounded per quarter of Trump administration, or about 107.2 on an annualized basis. (Over 1982q4-16q4, the average frequency per quarter is 4.672). The results are robust to the inclusion of a time trend.
Now consider a count variable, and a Poisson regression:
Figure 2: Mass shooting count, from 1982M08. Orange denotes 2017M01-2018M10, light orange 2016M11-2017M01. Source: Mother Jones, accessed 10/28/2018, and author’s calculations.
events = -5.98 + 0.006pop + 0.267trump
Adj.R2 = 0.36, N = 144, DW = 2.36, bold denotes significance at 10% msl.
Where events denotes mass shooting event count. This means there are 0.267 more events per quarter of the Trump administration, or slightly more than one more per year on an annualized basis. (Over 1982q4-16q4, the average frequency per quarter is 0.577). The results are robust to inclusion of a time trend.
Correlation is not the same as causation — but causality is a very hard thing to identify. Granger causality suggests a less rigorous approach. One interesting fact is that the Charlottesville riot, and the president’s remarks about “good people on both sides”, precedes an intensification of mass shootings.
Figure 3: Mass shooting fatalities (dark red), from 2000M01. Orange denotes 2017M01-2018M10, light orange 2016M11-2017M01. Dark orange line at 2017M08. Source: Mother Jones, accessed 10/28/2018, and author’s calculations.
Figure 4: Mass shooting count, from 2000M01. Orange denotes 2017M01-2018M10, light orange 2016M11-2017M01. Dark orange line at 2017M08. Source: Mother Jones, accessed 10/28/2018, and author’s calculations.
Update, 10/30 8PM Pacific: Several readers have evinced interest in what results would be obtained by trying a negative binomial count regression instead of Poisson. I did not test for overdispersion, but just estimated the regression, using QML. I obtain:
events = -5.74 + 0.006pop + 0.286trump
Adj.R2 = 0.36, N = 144, bold denotes significance at 10% msl.
Are you making a causal assertion? If not, what is your point? [Extra credit: prove that the “increase” in mass murders isn’t caused by leftwing nutjobs who lost it completely because Hillary isn’t president.]
A. Non You might want to read the narratives for each shooting in the source data. I didn’t find any that indicated the Trump spikes were due to “leftwing nutjobs.” You also might want to change your calendar. This is 2018, not 1968.
apparently the trump solution is to arm the churches and make a more robust death penalty. more guns and violence is the cure for current violence!
Yep but this should not be called the Trump Solution. Its name is the NRA Solution. Of course Trump is the puppet of the NRA.
Is that really statistically significant, given the U.S. population?
The variance of being hit by lightening may be greater.
PeakTrader Do you know what it means for something to be statistically significant? Did you ever take an econometrics course? Hint: coefficients “in bold denotes statistical significance at the 10% MSL.”
Maybe in Peaky’s mind, it is only “significant” when white Christians are killed.
I got a hoot out of PeakTrader’s comment about variance. It’s a “Poisson” regression, so anyone with the tiniest familiarity with econometrics should be able to describe the relationship between the mean the variance. Duh! If he wanted to comment intelligently about the count model, then he should ask if there was a “zero inflate” test, so maybe a NegBi regression would be appropriate. Although just looking at the data I don’t see any obvious “zero inflate” or overdispersion red flags. Or he might want to ask if there isn’t a two-stage “hurdle” problem; e.g., an unstable attacker must be both exposed to Trump’s nonsense and have a cache of weapons needed to carry out the assault. Or there might be a truncation problem, so only attacks above a certain threshold are counted. Those might be the kind of research questions Menzie would want to do if he planned on writing a refereed academic paper, but for blog purposes the simple Poisson regression presents a prima facie case that the Trump factor seems to be significant.
2slugbaits, what I mean by not being statistically significant is mass shootings are very infrequent compared to the population, not that they haven’t increased over time and with population growth, which are statistically significant.
And, it looks like U.S. deaths by lightning ranged from 15 to 40 a year between 2008 and 2018.
Couldn’t find the other link with lightening deaths. Here’s another one:
http://www.lightningsafetycouncil.org/LSC-LightningFatalities.html
We had the lowest deaths by lightning in Trump year 2017.
PeakTrader You’re confused about what it means for something to be “statistically significant.” It doesn’t “infrequent”. It means that the true value of the coefficient is unlikely to be zero. And the regression equations that Menzie showed did not include a time trend, they included a population parameter and a “Trump” effect parameter. It’s the “Trump” effect that stands out.
2slugbaits, an increase in population causes more mass fatalities and you can see from the charts they’ve increased over time.
Yes, statistically it shows a positive relationship between “Trump” and mass fatalities with a P-value at or less than 0.10 rejecting the null hypothesis.
However, it’s basically insignificant or negligible that statistically you will be a mass shooting victim, because it’s so infrequent.
PeakTrader: You keep on using that word “insignificant”. I do not think it means what you think it means.
Menzie Chinn,
“Insignificance (noun): the quality of being too small or unimportant to be worth consideration.”
I’m not worried about being a mass shooting victim.
“Adj.R2 = 0.36, N = 144, DW = 2.36, bold denotes significance at 10% msl.”
I bet you have no clue what any of this means. Look Peaky – you suck at economics. Please do not comment on statistics as you are even worse at reading evidence.
“trump is a dummy variable” so very true.
If you ran “Hillary Looses” as the independent variable, you would get the exact same result. Maybe call it the “Hillary Effect.”
sammy: Do you mean “Hillary unleashed” by “looses”. Or do you mean “loses” as “does not win”?
Either one’
Sammy, I like!
You would as it was stoooopid beyond belief. Yes Hillary loses in 2016 = Trump wins. I guess that is why it is called a Dummy Variable as our usual Dummies have forgotten that the operative event is who is in power. And of course Trump is using the White House to divide us and inflame people to do these awful deeds.
But knowing you – you like that too.
Stop the presses! Sammy says Mrs. Clinton won the 2016 election. Well she did get more votes!
Ceaser Sayoc was a horrible speller which was part of how the FBI caught him so fast. It seems Sayoc learned to spell from Sammy!
Promises made? Promises broken.
“In his inaugural address, Donald Trump declared, “This American carnage stops right here and stops right now.” He knew it would not. We know it did not.
“I’ll be able to make sure that when you walk down the street in your inner city, or wherever you are, you’re not going to be shot,” he declared during the campaign. “Your child isn’t going to be shot.”
Off-topic I wanted to put this up in comments. I know it’s an “old” story. But I have never seen ProPublica’s breakdown on it, and feel it is very important to never forget. Kicking elderly people out of their home is no small moral crime. It’s viciously egregious. I could go into more detail about how Mnuchin doing this makes it even worse. But I’m afraid the comment will then be filtered, so we’ll leave it here.
https://www.propublica.org/article/trump-treasury-pick-excelled-at-kicking-elderly-out-of-their-homes
Counting military?
Trump’s attempt to shut down the investigation of collusion with Putin just reached a new low. Falsely accusing Mueller of sexual assault? Seriously?
https://lawandcrime.com/high-profile/looks-like-someone-may-be-planning-false-sexual-assault-accusations-against-robert-mueller/?fbclid=IwAR3rXjm3Bf1HWicu7tQq1vdVUD72rrqt-2y_I19LD9JCMb-AFi1IyBkZbJg
Professor Chinn,
Using the listed data source, I’d like to compare my data regarding, fatalities, injured and total.
From 1982Q4 to 2018Q3, I show 841 fatalities, 1,296 injuries and 2,137 total victims. I used Trump 2016q1 to 2018Q3, and POP=POPTHM/1000
Using this data, my attempt to tie-into your model, shows:
F = -15.5 + .072 POP + 14.04 Trump. Adj R^2 = 0 .19, N=144, DW=2.17.
AS. Why would you begin the Trump era at 2016q? I assume you meant 2017q and it is just a typo.
Correct, a brain-malfunction, should be 2017q1.