Code and Data for the Social Sciences: A Practitioner's Guide

ion Rules (A) Abstract to eliminate redundancy. (B) Abstract to improve clarity. (C) Otherwise, don’t abstract. We are concerned about spatial correlation in potato chip consumption. We want to test whether per capita potato chip consumption in a county is correlated with the average per capita potato chip consumption among other counties in the same state. First we must define the “leave-out” mean of per capita consumption for each county: egen total_pc_potato = total(pc_potato), by(state) egen total_obs = count(pc_potato), by(state) gen leaveout_state_pc_potato = (total_pc_potato pc_potato) / (total_obs 1) We can now test whether pc_potato is correlated with leaveout_state_pc_potato. If so, we may need to adjust how we compute the standard errors in our model. We perform our analysis and are comforted to find little evidence of spatial correlation. But what if we are using the wrong level of aggregation? Maybe spatial correlation will show up at the level of the metropolitan area. Let’s copy and paste the code above and then adapt it to use metropolitan area instead of state as the level of aggregation: