Using Proc mixed (or other regression) for inverse distance weighting using latitude and longitude

[Talat]

Hi,

I have household pollution data. I want to build a prediction model based on household variables to predict household PM2.5( particulate matter). However, houses close by are more likely to have similar exposure compared to houses further away. So, I'd like to adjust the prediction model for the inverse of the physical distance. I have latitude-longitude data for each household.  Can inverse distance weighting be done in Proc mixed?  

[PGStats]

First, translate your lat-long coordinates into X-Y coordinates, such as UTMX and UTMY, so that both are expressed in the same units. Then ask proc mixed to fit a spatial covariance structure to your data with

random intercept / type=sp(pow)(utmx utmy);

also consider type=sp(sph)(utmx utmy) or type=sp(exp)(utmx utmy), (the choice will probably not matter much).

hth