Great series - this is the kind of in-depth review I pay for.
I decided to do a little statistical investigation of these loft, fill, price, and surface temperature data to see whether loft or fill was a better predictor of insulating capability. I compared linear models with loft, fill volume (fill weight x fill power), and price as predictors. The results were quite interesting.
Comparing models with only a single predictor each, the best model (as determined by AIC) contained loft and had an r2 value of 0.30, meaning that 30% of the variation in surface temperature could be explained by loft. Models containing only fill volume or price had r2 values of .22 and .21.
Interestingly, price added more information to a model already containing loft than did fill volume. Comparing models containing multiple predictors, the best model (as determined by AIC) contained loft and price and explained 45% of the variation in surface temperature (r2 = .45). A model with fill volume and loft explained 38%; fill volume and price 32%.
A model with all three predictors has an r2 value of 0.48, but r2 will always increase as additional predictors are included even if those predictors are random numbers so you need a way to determine whether additional predictors are statistically significant: I used AIC values.
So for this fairly small sample of 19 jackets, loft appears to be more important than fill weight/volume. This fits with what we're normally told but not with Richard Nisley's experimental evidence that down can be compressed to 40% of its maximum volume without losing thermal efficiency. I don't doubt Richard's experiments, this is an observational study and there could be lots of confounding factors. Perhaps though the down in some of these jackets (those with high fill and low loft) is compressed to less than 40% of its maximum volume and so the amount of loft allowed by the design does become important. This may not normally be the case for sleeping bags or bulkier winter down jackets.
It's interesting that price adds more information to a model already containing loft than does down fill volume. Perhaps the price charged for a jacket is a better measure of the true amount and quality of down it contains than are the specifications listed in catalogues.
Linear model results table.
Predictors df AIC dAIC r2
lm1 2 101.49 7.31 0.00
lm2 FillVol 3 98.80 4.62 0.22
lm3 Loft 3 96.73 2.56 0.30
lm4 Price 3 98.90 4.73 0.21
lm5 FillVol, Price 4 98.04 3.86 0.32
lm6 Loft, Price 4 94.17 0.00 0.45
lm7 FillVol, Loft 4 96.32 2.14 0.38
lm8 FillVol, Loft, Price 5 95.03 0.86 0.48
19 Jackets with loft, fill weight and price information (Salomon excluded as fill weight wrong).
FillVol, Loft and Price are correlated with each other but no so much as to preclude having them in the same model.