OK well, reporting back to add a couple more meter drift data points...
Below graphs of:
(1) straight PPM drift and internal temperature for two meters (I think the room temperature data on the graph is bogus btw).
- which give these regression results (units are CAL 72 PPM/day and /C):
Meter 1 Coefficients Standard Error
Date/Time -0.086 0.000
Internal Temperature -0.230 0.003
Meter 2
Date/Time -0.188 0.001
Internal Temperature 0.520 0.005
Curiously one meter is +ve temp coefficient and one -ve (these are both properly old meters). Also, one reports internal temperatures 40-44C and one 35-39C (I cleaned both filters). Worth noting I guess is that the temperature coefficient is substantially bigger than the time one - so we might expect a bit of effort needed to get the time one out from under the temperature-based 'noise'.
(2) PPM/day over time calculated in the SN18 sense (i.e. ignoring temperature which moved about -1.5C over the 20-day period). This gives average 'drifts' of about +0.03 and -0.2.
(3) 'Temperature corrected' drifts using the regression results in (1) to correct the CAL 72 gain numbers. Fitting a line to the 'adjusted CAL 72' numbers gives drifts of -0.08 and -0.19. That is, roughly the same as the regression numbers - as it should. Curiously, there still seems to be a temperature variation observable and I believe I've done my calculations right. Maybe some delay or non-linear effects?
Alan