Since my own analysis didn't show a robust HIE, I looked for bias in the dset1 data.
I thought it might be due to urban records being longer than rural ones and therefore dominating the data set. This did not prove out. The average length of rural and urban records is about the same.
I thought it might be inaccurate rural, suburban, and urban designations. But a review of 300 rural sites selected at random did not reveal a single obvious misclassification.
No matter how I parsed and combed through this data I was getting no significant difference in rural and urban stations. The best I've done is to see a 1.8 degree difference in the aggregate analysis that I've posted previously, and that does not necessarily represent a trend.
So, what are the possible explanations?
Here is my list of guesses:
1) The database is contaminated with thermometers that are sitting on asphalt or next to air conditioner compressors that are purported to be in rural locations.
2) The population density in the areas of the USA where most of the thermometers are located (the Northeast) has been flat or declining since 1950. Therefore, a grouping of U, R, or S or brightness in one cross-section in the database does not adequately represent the situation.
I don't have lot of enthusiasm for working through this possibility. It seems to me that even if the U, R, and S designations in the GISS data had errors that there would still be some differences emerging.
For what it's worth, here are the results of comparing locations that are designated rural and are not designated as airport locations versus all the rest of the locations using the random effects statistical algorithm described in previous posts:
These are not the results I hoped for. But hiding them would make me no better than a climate scientist.