again. More bad averaging and USHCN. This time, it arises following a very good post by Zeke Hausfather on USHCN adjustments. He showed this plot of the effect of infilling. It isn't much.
A blogger and commenter there, sunshinehours1, said, no, that's misleading information. And he shows how the average of estimated final less rises faster than the average of non-estimated.
It has been reblogged by Paul Homewood, and looks like it is getting around. But it's the same bungled methodology that Steven Goddard used. The stations in that average, plotted over the years, change substantially from year to year. They could be just be an increasing number of warmer stations. Since climate differences are large, it doesn't need a big imbalance to show up.
So the same refutation will work here. Simply work out the difference using just the climatology of the stations. No use of estimation, or indeed annual data. And you get the same result. It isn't telling you anything about the effect of estimation. It is just telling you about the changing nature of the stations being estimated.
Sunshinehours (SH) prefers to work with Max data, and do individual months. He shows this plot:
It's bigger at the source. Red is average estimated, Blue is not estimated, and green is combined. My equivalent plot is below - I've omitted combined, since it is very close to "not estimated":
It looks the same, and the estimated sure seems to have a higher slope. So I'll plot the difference between them:
Yes, rising and positive. That's his result. But now I'll plot the same difference calculated with just climatology. To get that, I just average each station, for December, over the whole range of years, and substitute this constant value for the year-by-year data. So there is no issue of estimation. If the composition of the estimated wasn't changing, the result would be absolutely flat. But it isn't:
Instead it tracks the data-based difference very closely, with similar trend.
The difference between average estimated and average non-estimated, doesn't reflect estimation. It just reflects changes in the kind of stations that were being estimated. For some reason, they were more likely to be warmer. I don't know why, but they were. It's just the wrong way to do it.
My R code is here.