Dear Allstat,
I have a number of curves, which represent the result of a stochastic
search for the best model in a model selection context. Briefly, the x
axis represents a parameter, in this case, the quantity of data. For
different values of this parameter, some data is simulated from a model, a
stochastic search is done for the model that the data was simulated from,
and the result of the search is returned. The y axis represents a
statistic which summarizes the difference between the true model and the
estimated model.
The plot shows that for increasing values of this parameter, the parameter
goes to zero.
My reason for wanting to fit the curves is so that I can produce a table
of comparative results for different curves, showing that some curves
decrease more slowly, and some curves decrease more quickly.
I don't have any theoretical model for what these curves should look like,
but I assume that if I want to do such a comparative table, I would need
to use a parametric model, where one parameter reflects how quickly the
curve decreases. If this is incorrect, please let me know. I'm not very
familiar with the curve fitting options available.
In any case, does anyone have suggestions about how one could do curve
fittings in such a scenario? Any suggestions gratefully appreciated.
Sincerely, Faheem Mitha.
|