I'll be using only the RAW datasets, the recorded thermometer readings, with no adjustments. Here they are:
As you can see, there is quite a large variability at the beginning of the record, enough to throw off the slopes of the linear fit. I'm uncertain the best way to handle it. My first impression is to delete the record before 1890, after which the variability diminishes. Tavg and Tmin are certainly being influenced by the pre-1890 data. Deleting that data gives us a good 130-year temperature timeline, minimally enough to see what's happening to the climate.
That's what I'll do before assessing the trends in the US temperature record.