When I use Prism 4.00 or 4.0a to fit data entered as mean, SEM (or SD) and N, the standard errors of the parameters seem too low, and the confidence intervals too narrow.

Most people enter actual replicate values into Prism, and it fits these data without problem. But Prism also lets you enter data as mean, SD (or SEM) and N. Fitting these data is trickier.

Prism 3 fit the mean values, ignoring the SD (or SEM) and N. This works pretty well, but can give invalid results if the sample size N varies a lot from point to point. You'd want to give the points with higher N more weight, but Prism 3 didn't do this.

Prism 4.0 and 4.0b attempt to solve this problem by giving you (in the weighting tab) the choice to weight by N. This works properly to give you the most accurate best-fit values of the parameters. But it works by simply entering each mean into the fitting program N times. This results in a sum-of-squares that is lower than it would have been had you entered raw replicates. Accordingly, the SE of the parameters are too low and the confidence intervals too narrow.

Prism 4.01 and 4.0b fix this problem by taking into account the SD (or SEM) you entered, as well as N. As a result, you'll get identical results when you fit data entered as mean, SD (or SEM) and N as you do when you enter the data as raw replicates.