Fit linear regression weighted by number of homicides. Homicides are rare, therefore variance of homicide rate estimates is primarily driven by number of homicides rather than by population. There is no circular reasoning here – we’re predicting homicide rate rather than absolute number of homicides.
Exclude countries with gross domestic product (at purchasing power parity) per capita lower than the median because we are primarily interested in developed countries.
Firearms are statistically significant and have negative coefficient in the regression.
Conclusion: in developed countries, more firearms means less homicide.