Shots Fired at University of Texas Campus; Gunman Dead

[Alyrium Denryle]

Can you break up the variables and figure out which ones are responsible for how much of the variance?

I can, but do not have access to my good software package right now (I only have SPSS right now, not Systat, and I am still on SPSS's learning curve). I can get it done easily enough tomorrow though.

The other issue is co-linearity in my variables which I cannot easily test for in SPSS, and I would like to include other measures for the permissiveness of gun laws. Brady Scorecards as a continuous variable, a few more categoricals like the necessity to get a permit to purchase guns etc. The above was a "quick" analysis. I use quotes because I had to hand-enter all the data.

[Simon_Jester]

If by normally distributed you mean a Gaussian curve, I'm pretty sure the distribution of GINI coefficients and population figures for the states don't fit a Gaussian.

I ran, as part of the ANOVA, a Kolmogorov-Smirnov Test on both variables. GINI was normaly distributed. Population was not, so I log-transformed it and re-tested. It came out nice and normal. Subsequent runs on the log-transformed data did not appreciably change the outcome, just change the fit slightly.

Ah. I wouldn't have expected GINI to be normally distributed. My math-intuition isn't surprised that log(population) was, but I had no right to expect that either. Cool.

I can, but do not have access to my good software package right now (I only have SPSS right now, not Systat, and I am still on SPSS's learning curve). I can get it done easily enough tomorrow though.

The other issue is co-linearity in my variables which I cannot easily test for in SPSS, and I would like to include other measures for the permissiveness of gun laws. Brady Scorecards as a continuous variable, a few more categoricals like the necessity to get a permit to purchase guns etc. The above was a "quick" analysis. I use quotes because I had to hand-enter all the data.

Wow.

Well, if you do you do and if you don't you don't; I was just wondering.

Thanks.

[His Divine Shadow]

I collected data on the annual gun deaths per hundred thousand, and performed an Analysis of Covariance using Concealed Carry laws as a proxy for the general permissiveness of a state's gun laws. Specifically I used unmodified Shall Issue laws vs May issue or Right Denied. I used population and GINI as covariates.

The analysis explained 31.5% of the variance in gun laws, and all terms were significant. Population size was significant at .035, GINI was significant at <.001 and permissiveness of gun laws was significant at .008

I wish I actually understood what you just said, because I think it's significant information. While it may require you to use more words of fewer syllables, and probably also fewer numbers, could you perhaps explain that in American dialect for the non-statistically educated?

Basically, the issue of gun violence is a multi-faceted problem. Comparing one state to another is absolutely useless, what actually needs to be done is to compare the result policies such as gun control while statistically controlling for other variables. What an Analysis of Covariance does is exactly that. It looks at the effects of a categorical variable (namely the CCW laws) on the outcome, firearm deaths. It also controls for the effect of a normally distributed continuous variable such as population or GINI (an index of income inequality) on your outcome variable.

In this case, the combination of these factors explained 31% (though a subsequent analysis I just ran on log transformed population explained 36%) of the variation in firearm related deaths. I can also run this same regression for armed robbery and forcible rape.

Statistical significance is the probability that you have rejected the null hypothesis(that your independent variables will have no effect) in error. The standard convention is .05. A five percent chance that your independent variables have no effect. Anything lower than that and you have increasing certainty that your independent variables are actually doing something to your dependent variable.

What my analysis indicates is that population, income distribution, and gun laws all impact the rate of firearm deaths, in this case, more permissive gun laws lead to more firearm related deaths. Also, uneven income distribution and high population also lead to more gun related deaths.

Measuring CCW laws in correlation to firearms deaths (also how does it affect the total murder rate, rather than just firearms deaths?) strikes me as an odd method. Given how most homicides and gun crimes are not comitted by CCW holders. It seems to me that yes, you can indicate a correlation like this but not a causation. I mean your stats could indicate that increased population and higher income equality and higher firearms deaths wil lead to more people getting a CCW instead of CCW holders causing it.

Seems to me it would be better to compare gun ownership ratios (if such exists) to gun murders instead of CCW laws, while accounting for other variances, it would then also be good to account for slums and such areas as they can skew an state or city's numbers. I wonder though when such a situation is found, does it then indicate social problems causing firearms death rather than firearms laws?

[His Divine Shadow]

Ninja edit: Would it also not be usefull to compare states homicide rates before and after CCW laws where liberalized (this is a trend across most of the USA) and see how firearms death where affected by that while compenstating for the same covariances as you have done?

[Alyrium Denryle]

Measuring CCW laws in correlation to firearms deaths (also how does it affect the total murder rate, rather than just firearms deaths?) strikes me as an odd method. Given how most homicides and gun crimes are not comitted by CCW holders.

They are not, however CCW laws serve as an easily accessible proxy for the general permissiveness of gun laws, and easy access to firearms. Tomorrow, when I have some spare time, I will add more values relating to gun laws into the equation. I just had to enter every datapoint I used by hand

Seems to me it would be better to compare gun ownership ratios (if such exists) to gun murders instead of CCW laws, while accounting for other variances, it would then also be good to account for slums and such areas as they can skew an state or city's numbers.

I tried to find guns per capita by state but was unable to find the relevant data. Otherwise I would have used it. As for slums, that is what I used GINI for. Income distribution indices in a nation with a high per capita GDP accounts for the relative frequency of slums/inner city regions. To try to get all of that data for each state would be Insane. With a capital I. I would not do that unless I wanted to study the Public Policy and Demographics lit and actually publish something. That is if I could find the data at all.

Ninja edit: Would it also not be usefull to compare states homicide rates before and after CCW laws where liberalized (this is a trend across most of the USA) and see how firearms death where affected by that while compenstating for the same covariances as you have done?

Within-subject designs are very very messy to perform. They have this odd tendency to violate just about every parametric assumption.

[His Divine Shadow]

GINI is an indicator but it seems like it would introduce a high margin of error. In your stats GINI seems to account for very little (<0.001), while if one looked closer upon a state one might find that poor areas are far more of a factor in gun crime than GINI would indicate. So I'm not sure how reliable the results are going to be in the end, the significance for guns seemed pretty low as well as it stands, below the margin of error perhaps?

[Alyrium Denryle]

GINI is an indicator but it seems like it would introduce a high margin of error. In your stats GINI seems to account for very little (<0.01), while if one looked closer upon a state one might find that poor areas are far more of a factor in gun crime than GINI would indicate. So I'm not sure how reliable the results are going to be in the end, the significance for guns seemed pretty low as well as it stands, below the margin of error perhaps?

No no. That is not what significance is. That is the probability that there is no effect Less than a percent in the case of .01

My stats explained 36% of the variation in gun deaths (after I log-transformed population). That will increase when I add more gun laws and population parameters.

[His Divine Shadow]

Right, I obviously don't have a background in statistics here.

[His Divine Shadow]

This helped me understand, I think
http://onlinestatbook.com/chapter9/significance.html