>>Hmmm... an interesting approach, but I'm not sure they are mathematically equivalent.
>
>Not having used regressions in practice, I am not sure either. But in general, lots of things can be done converting back and forth, similar to the way I explained.
>
>Perhaps you can do a test in Excel, where, I believe, all required functions are available, and see whether you get identical results.

Well, although it was a good idea it's easy to show your original idea wouldn't work in this case. The data sets contain some negative y-values for which you can't get a logarithm.

It turns out the client really wants to duplicate the Excel logarithmic "trendline" function. In VFP parlance it calculates, using a least-squares method, the best-fit function

y = (m * LOG(x)) + b

I dug into my old university calculus texts and derived the equations for multiplier m and constant b. It turns out they are very similar to those for the least-squares fit for a line but it took quite a bit of work to convince myself it was correct!

In Excel it is possible, once you have put a logarithmic trendline on a chart, to display the equation Excel has calculated. It was gratifying to see that my solution gives exactly the same answers as Excel on the same test data sets.