What I should have expounded on is the observed variation coupled with the temperature dependence in the fit coefficients (noted by others as well), and the comparative sizes of the terms used to compute COP in Table 4. One can see that the COP has 3 parts: DeltaQ_heater, Q_k, and Q_pulse, which are all of the same approximate size. Thus the calculation of all is relevant, but the data supplied for the m’s and b’s for the Q_k calculation suggest a very wide spread, which means a large % error, which means it is unlikely the listed COPs are statistically different. But as I noted, with no information on the standard error of the slope and intercept, or statements on the errors of the computed points, I can’t go any further than simple cautionary speculation at this point. BTW, these numbers are standard output for most linear regression routines. Microsoft Excel 'trend analysis' however doesn't show them. You have to use a canned function in a cell to produce tabular output that will show this.
Q_k is an intermediate result; if you insert its formula into the COP formula, then the COP formula simplifies to:
COP=DeltaQ_heater/Q_pulse + m + b /Q_pulse
If we take the most extreme values for b and Q_pulse, we have b /Q_pulse=0.1/3.75<0.03.
So I'd submit b is not significant when assessing error.
Now, m is directly added to the COP calculation and varies between 0.41 and 0.57. So that's a +/- 0.08 range when we're working with a COP of 1.2. Are we really gonna see a confidence interval of +/- 0.2 for m? I tend to doubt that. And then we have to deal with the 1.4 and 1.6 COPs. We would have to assume no energy was dissipated (m=0) for the highest COPs to be reduced to 1.
That leaves us with the question of whether using a linear model is valid. I agree we would need to see their modelization data.
I do question even the need to have a linear model for power dissipated vs pulse power, when they could just use actual measured values (or at least they could do both).
Finally the last term DeltaQ_heater/Q_pulse ; those power values are given with 3 significant digits. Unless they totally messed up, are we going to see an error higher than 0.01?
I'm kinda rambling, but somehow I doubt that the error discussion will be what brings down the COP (if anything does). We'll see!