Skip to main content

Hover over "social costs of gun ownership" at the top of the page to show the site menu


Regression findings in Cook and Ludwig's 2006 analysis on "the social costs of gun ownership" are completely spurious due to ignoring the ratio fallacy (see Kronmal 1993).

This website will guide you through obtaining the data, confirming Cook and Ludwig's (spurious) findings and finally taking the ratio fallacy out of the model and arriving at a null result.


2013-04-29: Added a section on state level data

2013-04-15: Working paper (subject to slight future changes) available, easier to read than this site if you are only interested in the findings

2013-04-08: Added a section on using Stata for extracting mortality data with code.

2013-04-05: Corrected slight undercount of E95 and E96 in 1997. All results updated: No significant change.

Guns & Crime in Economics (must read)

It is highly disputed if increased civilian ownership of firearms comes with an increased or lower rate of this-or-that type of crime.

John Lott wrote the book "More Guns, Less Crime" (1st ed. in 1998, 3rd ed. in 2010) where he found negative association between gun ownership and certain crimes.

Other scientists like Mark Duggan ("More Guns, More Crime"), Jens Ludwig and Philip Cook found contradicting evidence. Their analysis culminates in Cook's and Ludwig's 2006 article "The social costs of gun ownership", with a preceding NBER working paper of the same title. The latter lists the exact data sources used.

In their analysis they assume guns to somehow travel from the hands of law abiding to criminal individuals who then abuse them to commit more homicides. I think this mechanism isn't needed. One could simply assume, the general increased availability of guns makes on average all people more likely to commit certain crimes. This would be a much simpler explanation for increased crime coming with an increased amount of firearms. So let us generally agree there may be some mechanism to increase homicide rates through increased gun ownership, it does not matter if it is their complicated or my simple mechanism or both.

Surely then one should be able to find statistical association between the numbers measuring gun ownership and homicides. This is what their analysis is about. They use panel data for 200 U.S. counties (the counties with the highest population in 1990) from 1979 to 1999. Gun ownership cannot be directly measured, therefore they use a proxy variable of

FSS = "Suicides by firearm" / "Suicides"

This is pretty plausible. Imagine everyone who wants to commit suicide having a gun at hands: Surely more people will kill themselves with guns compared to a situation where guns are unobtainable. It does not matter for this measure if the suicide decision depends on the gun availability, i.e. if the gun availability increases the overall rate of suicides.

Out comes an elasticity of the homicide rate with respect to gun prevalence in the region of 0.1 to 0.3, meaning (a bit simplified):

Increase gun prevalence by 1% then the homicide rate increases by somewhere between 0.1% and 0.3%.

This of course is a hurtful effect on society. They then go on to calculate the approximate social costs from that relationship. Their calculation is arguable as they ignore all benefits like jobs created from gun ownership. But nevertheless it is an interesting study.

I chose to reanalyze the data. I had to obtain the data from scratch. That is a lot of work and there are many possible sources for mistakes. Here I documented how I extracted the data so you can follow what I have done and (hopefully not) spot mistakes.

So what to do?

If you are only interested in my results, you may start reading with section "Regression Analysis, 2. Replication of results" and go from there.

The Menu on the top will guide you through obtaining and analyzing the data. Should you plan on obtaining and extracting the data yourself you will need a lot of time and preferably an empty external hard disk. I suggest you start with the overview.


A quick overview: How to proceed to see yourself how Cook and Ludwig's findings are some sort of nonsense regression.

Data Sets for Analysis

For those who trust my data extraction, I will provide the data sets for analysis in various formats. They are great teaching examples as well.

Obtaining the Data yourself

Obtaining the data is a hassle. Cook and Ludwig used various data sources described in detail in their 2004 working paper. Unfortunately they do not care to share their data set. So I went as close to the raw data as I could to extract the numbers…

My R code

Opposed to CL I publish all my code for individual use and critical review. This page gives a short description of the code files. You may download all files from here.