Michael R Middleton wrote:
....

Second, as I recall, depending on the version of Excel, there may or may not
be an adjustment for the number of observations (degrees of freedom), but
maybe that was only for COVAR.
....


I do not recall that. As best I recall, CORREL has used a numerically
stable 2-pass algorithm in all versions, where RSQ and PEARSON only
switched to that in 2003. Prior to 2003, the R-squared value from
LINEST was wrong if no intercept was fit.

Jerry