Are you still referring to the anomaly plot for January? I'm not sure exactly what you mean, but any kind of averaging process over time reduces the influence of variability within that time interval. (palmer always reminds me about this for ice-core data, because the CO2 in ice cores is essentially an averaged sample over whatever time interval is represented by the sampled ice core layer.)
I still think we must use the raw data before we make all the corrections to better fit the modeling process.
Taking that statement at its basic level, I can't imagine any environmental observational data set that won't have an occasional outlier that would get tossed by a simple Q-test or similar. The data has to be QA/QC'ed to some extent. How much and how its done -- that is a process that is, and should be, constantly scrutinized.
Outlier is a term that should be reserved for obviously wrong data - for example, if St. Louis had an average temp of 54F on one thermometer for one day and another thermometer one mile away had 40F degrees, both are suspect; while if one out of ten in a close radius differ where only one was off by more than one degree the spurious one becomes obvious.
Move out beyond ten miles and narrow your time allowed to ten minutes and that one thermometer becomes a predictor.
I rechecked St. Louis for February, 2008 and it was warmer than 2007 by 1C, but not 3C.