Sometimes you take a look at data from an odd direction and something about it doesn’t look right, but at the same time you can’t say what. Maybe it really is just a random bit of spaghetti on the wall and it doesn’t mean much. Maybe it’s a ‘smoking gun’ for human intervention that isn’t so rational. Like folks fudging numbers tend to pick “7”. They shy away from 1 and 9. There’s a bunch of those kinds of “Human Factors” things in forensics. Places where the human “call random” isn’t quite random or places where a human does make things random and they ought to have a pattern…
So I semi-randomly picked Mexico to look at the temperature data of v1 vs v3 by months. Why Mexico? Well, I like Mexico. It’s generally got a nice water to land ratio. The equipment is pretty good, but not frequently changed for “the newest stuff” and it hasn’t had a lot of wars causing loads of data dropouts. There are also interesting countries “nearby” in all those Central American analogs and all the Caribbean islands. So it’s possible to sort of “poke around” near it and see if other places have similar trends.
And I made this graph…
Stare at is as I might, something is disturbing about it, but I’m not able to just say “THAT! THAT’S The Culprit!”.
Maybe it’s the way different months wander off on their own… If the changes were some regular adjustment (like for TOBS or for change of method) it ought to hit all months more or less similarly. It also ought to be that they would show up as discontinuity in all the lines, then pass.
But this looks different. Part of it will be added / removed thermometer artifacts. (So add a thermometer in a cool steady ocean coastal town, summers will drop and winters will rise). But some of it is just inexplicable.
Each line is the ‘cumulative anomaly’ between v1 and v3 in that monthly series. One line is the average of the data, and it lets you see how the data ‘averages out’ or averages together. The others are the running total anomaly in that month.
So look it over. Click on it to get a big one. The wide magenta line is the cumulative change of averaging all those other lines together.
It looks like the major dropping lines are April, May, June, Sept, November, December. Not a lot in common there. March, July, and August all rise over the series. Other than being interleaved with the falling months, not a lot of pattern.
If this is the result of semi-random “corrections” I’d expect a smoother distribution.
I donno… take a look at it. Maybe you will notice something.
In a while I might add some “nearby” other countries to see if anything pops then….