Has anybody tried to capture the variability between two pairs of instruments/users (you have a big experiment and two MultispeQ, each run by a different user). How would you measure those fixed and/or random effects and include into your model/stats?
Specially when measuring so many parameters, and many are not easy to reproduce. e.g. for SPAD it can be quite straightforward to cross-calibrate between two devices, and even against an external reference... but for NPQt or oxidized PS1 (for example)?
Any experience or idea?
We measured corn leaf parameters at multiple development stages in 2018 in a large-scale, replicated, field trial using two devices. Each operator maintained the same order with respect to plot row number, plants within row, leaf number (MRM or ear leaf ±1), replicate measures per leaf, throughout. Thus far, I have not detected any appreciable difference in error variance with respect to operator when included as an R-side or G-side effect in a generalized linear mixed model. It is, however, important that each operator follow the same procedure, i.e. orientation of sensor, clamping position on the leaf blade, otherwise you are estimating operator bias, not that of the sensor itself. Moreover, I have been including PAR as a covariate when testing treatment effects on light reaction parameters due to the inherent variability of incident PAR under field conditions, particularly while working in a mature corn canopy. In trials with blocking structure, i.e. RCB, I would proceed measuring by block in order to partition "time of day" effects, if present.