Thursday, July 9, 2009

"Guns are Bad News for Women?" Huh ... - Part III

In Part I and Part II, I have discussed the basis of mikeb hypothesis that "guns are bad news for women." I ended Part II with three questions yet to answer. Going back to the original table, mikeb and Hemenway posit that the data prove more guns equates to more death for women. I created an updated table based on rates in Part II. My numbers correlate well with Hemenway's. So why don't I agree with his conclusion?

First, the table contains two datapoints - high gun states and low gun states. It so happens that looking at the extremes of gun ownership, states with more guns have more homicides and suicides. So let's not stop there, let us break out the data and look at individual states, that will then give us a total of 17 data points. If mikeb's hypothesis is correct, then we should see an increasing trend among the high gun states (as % ownership goes up), and a decreasing trend among low gun states (as the % ownership goes down).

2006 Death Rate of Women (per 100,000) High Gun States

Firearm Ownership 47.7 50.7 51.7 55.3 55.3 55.3 55.4 56.6 57.7 57.8 59.7

Firearm homicide 0.84 0.00 2.45 1.88 1.24 2.61 2.38 0.25 0.42 0.92 0.77
Other homicide 1.26 1.58 1.35 1.88 0.69 1.94 1.41 2.02 1.69 1.23 0.39
Total 2.09 1.58 3.80 3.77 1.92 4.55 3.79 2.28 2.11 2.14 1.16

Firearm suicides 2.09 1.26 2.32 1.60 2.20 2.41 2.81 2.02 1.06 3.06 2.70
Other suicides 2.89 4.10 1.73 2.79 3.71 2.41 3.03 3.29 4.23 6.43 6.17
Total 4.98 5.36 4.06 4.40 5.91 4.81 5.85 5.31 5.29 9.50 8.86

Firearm accidents 0.09 0.00 0.30 0.14 0.14 0.27 0.54 0.00 0.00 0.31 0.39
Other accidents 40.03 30.56 36.50 33.84 32.71 40.72 45.36 41.73 41.65 25.42 37.77
Total 40.13 30.56 36.8 33.98 32.85 40.99 45.9 41.73 41.65 25.73 38.16

For high gun states, we don't see a correlation (trend) between gun ownership and homicides. Same for suicides and same for accidents. So lets look at low gun states:
2006 Death Rate of Women (per 100,000) Low Gun States

Firearm Ownership 3.8 8.7 12.3 12.6 12.8 16.7

Firearm homicide 2.27 0.47 0.72 0.27 0.91 0.84
Other homicide 2.59 1.42 1.02 0.87 1.09 1.00
Total 4.86 1.90 1.74 1.15 2.01 1.84

Firearm suicides 0.00 0.47 0.18 0.39 0.55 0.22
Other suicides 1.94 4.58 2.41 3.50 2.19 3.12
Total 1.94 5.05 2.59 3.89 2.74 3.35

Firearm accidents 0.00 0.00 0.07 0.00 0.00 0.00
Other accidents 25.58 23.06 20.35 24.92 28.64 24.88
Total 25.58 23.06 20.42 24.92 28.64 24.88

Again, same result, there is no correlation(trend). Even if we ignore DC, there is still no correlation. So the hypothesis fails when more than two data points are looked at (using the exact same set of data).

Comparing the two, the low gun states in general have lower homicide rates than the high gun states. Except for SD and ND which have lower rates than all of the low gun states. And MT is lower than all but 1 low gun state. Interestingly enough, SD, ND and MT are contiguous to each other. Which leads us to the first major problem with Hemenway's table: Simplified conclusions (such as mikeb's) only work if all other variables are equal. In this case they are clearly not.

Hemenway and mikb make the mistake of using only a 2 datapoint set to formulate a conclusion. With only 2 datapoints, no outliers or exceptions are identified. Once the same data to make the 2 datapoints is taken back to its original 17 datapoints, at least 4 outliers (the 3 mentioned above and DC) begin to emerge for homicides alone. Whenever one is examining a dataset, you cannot ignore the outliers. You may decide not to include them in your dataset (DC) but you must provide a valid reason for the expulsion. By crunching the numbers down to only 2 datapoints, Hemenway and mikeb cover up the datapoints that go contrary to their hypothesis.

The United States is not a homogenous society. I complain about California drivers and Boston drivers. I have never complained about Oklahoma or Texas drivers. Rhode Island and Utah have very homogeneous religious makeups. Most every other state does not. North Dakotans don't have problems with punch card ballots. Floridians do. The simple fact is, people (in general) act differently depending on where they live. Even within a state, people are different in different counties (like the Texans who can't stand people from Dallas). So to simplify an issue like homicide and guns in the whole nation and assume that gun ownership is the major factor contributing to it is dishonest. Looking at the datapoints I see three distinct regions that have very similar numbers: 1) the previously mentioned ND, SD, MT and I'll add in WY, make a contiguous region that has firearm homicide less than 0.8. 2) AL and MS are next to each other and have very similar homicide rates. 3) MA, CT, RI, and NJ are all right next to each other and while MA firearm homicide rate is about 1/3 the rate of the other three states, the non-firearm homicide rate is very similar. So, 10 of 17 of the data points fall rather neatly into a regional category. Perhaps then this has more of an influence than firearm ownership? I'll have to look at that.

Next let's compare suicides between the two tables. Unlike with homicides, there are no obvious firearm outliers. All of the low gun states have lower firearm suicides than the high gun states, so one could safely conclude that states that have access to more guns see more people use guns for suicide. But isn't this stating the obvious? What the real question is, is whether the overall suicide rate would decrease if the availability of firearms decreased. Looking at the non-firearm rates and overall rates we do see a few outliers: 1) HI has a overall suicide rate higher than 4 of the high gun states, and it has a higher non-firearm suicide rate than all but 2 of the high gun states; 2) AL is the only state of the 17 that has a non-firearm suicide rate lower than the firearm suicide rate (although their neigbor MS is equal); 3) DC has a suicide rate far lower than any other of the 17 states; 4) AK and WY (the highest gun states) have firearm suicides that are not much different than the other high gun states, but their non-firearm suicide rates are 50% more than the next highest state.

Since I saw a potential for regional influence on homicides, let me look at suicides as well. The four northeastern states have a nice cluster of rates that are similar. AL, AR, and MS have more variation in their firearm suicide rates, but the total suicide rates match up well. Finally, SD, ND, MT and ID this time have a similar situation (overall suicide match well, even though firearm suicide have a disparity).

Finally, accidents. The first thing that should jump out at you is that there are no firearm accidents in low gun states (except New Jersey). Even more fascinating is that there are no gun accidents in ND, SD and MT (hey that regional thing pops up again! - I may be on to something). AL and MS once again have similar firearm rates.

So to sum up what I have looked at, the numbers for the states that make up the high gun and low gun states do not mesh well with the hypothesis of more guns = more homicides, suicides, and accidents. There is no trend. In general one can conclude that there are more firearm homicides, suicides, and accidents in the states that have the most guns compared to the states that have the least guns. But this does not mean that guns are the cause (if that were the case we should be seeing a trend).

The other theme that kept popping up is that certain regions of the country have similar numbers. This is something I will have to explore more to answer the hypothesis.

So on we go to Part IV

No comments:

Post a Comment