Hi,
I have a multinomial logistic regression of 3 classes: 0, 1, and 2 and have associated individual models (via Fit Y by X) for N parameters.
I am now looking into a sensitivity like study in which each parameter is assumed at certain levels (e.g. at min, max etc..). I have probabilities computed at each such parameter level. I like to rank order probability change in each class (0, 1 and 2) due to each of these parameter actions which is also easy.
Obviously each Prob is computed from Lin functions (in this case both functions are linear) - each line w/ specific p value on slopes and intercepts. What is the recommended and/or fastest way to compute a sort of confidence level of probability on the selected parameter ?levels ?
At this point, I have rather scalar data from my initial population - scalar data for each parameter w/ associated formulas of Lin[Y] and Prob[Z] w/ population statistics... I think I should have applied parameter changes to my original population data and then look into confidence interval from distributions of probabilities (as suspect this will be your answer) .. I was hope there may have been some other way -- perhaps an explicit formula - through a script.
Thx in advance for your help.
