Hi All,
I have some highly skewed count data and have fitted various distributions.
I am interested in a statistical goodness of fit test - any thoughts?
I have tried the Pearson Chi^2 GOF test however it requries expected cell
counts > 5 for the Chi^2 approx to work well and due to the highly skewed
nature of my data I have a lot of cells with expected counts < 5. I have
tried combining these cells but in some cases I end up with only a few cells
left, maybe less than half of what I started with. The other question is how
do I combine the cells do I start at the tail and work inwards or the other
way round.
I can look at qq plots to see graphically if my model is a good fit, or
indeed compare likelihoods between models, but I was wondering if there was
a way to simply compare the observed and expected values to see if (from a
stats view point) the model is a "good" fit.
Just for clarification here is some example data
#widgets Frequency
0 5521
1 459
2 41
3 6
4 3
6 1
7 1
19 1
Best regards,
Richard
|