When I regress 86 pairs of lab duplicates, I get a better fit by hard-coding
0 intercept than letting XL calculate it. This seems wrong: the least squares
regression should be the best fit of the data. I compared LINEST and the
Regression tool in the Data Analysis Tool Pak and they yield the same answer.
I suspected that XL adds a (0,0) to the data set because the total df in the
ANOVA output is one larger for the fixed intercept, but testing that with RSQ
yielded a different value. The stats (below) all favor the fixed intercept;
even the calculated slope is closer to 1 and the confidence limits are
tighter:

Stat b=calc b=0
Intercept -0.02048 0
Rsq 0.9929 0.9991
std Err 0.2318 0.2306
Slope 1.0045 1.0020
std Err 0.0093 0.0033
L 95.0% 0.9861 0.9955
U 95.0% 1.0230 1.0085

Thanks in advance for any input.