My last post inspired a number of requests for a multiple regression analysis with potential suppressor variables included. The table below shows standardised coefficients from OLS models of the state murder rate. In addition to population density (which was requested twice), Southern state, fraction in poverty, and fraction black were selected as covariates, due to their well-documented associations with the frequency of homicide in the US. These variables were all taken from the US Census Bureau.
The estimate for gun ownership is quite unstable across the different models, ranging from marginally significant and positive to significant and negative. In the final model, which explains just under 70% of the variance in state murder rate, gun ownership has a negative estimate; both fraction in poverty and fraction black have positive estimates. If anyone wants to play around with the data, I'd be happy to share them. Or else they can be retrieved through the various links provided.