Continuous fitting of a distribution with decay

Question

Hello

I am looking to do a continuous distribution fitting for a customized combination of a exponential decay and a log normal or weibull or gamma. Basically a  non-normal right skewed distribution with a exponential decay. I checked the distribution platform and the mixtures option but could not find a suitable way.

Any help much apprecited.

Thanks

theseventhhill · Answer

I think I am using my data in a non-conventional sense. Basically I am not looking at product reliability or failure times but looking at daily driving distances over a year to see what distribution/combinations will have better prediction. My hunch is that one distribution might do well for shorter distances and some other distribution will do better on relatively longer distances.

So far I did all the possible continuous fit, tinkered a little bit with the conditional mixtures but holding off now until I am confident that a mixture approach would fit better than a single continuous fit

theseventhhill · Answer

Thanks for the pointers. will check out the help manual

gzmorgan0 · Answer

Here are a few more comments and ideas:

* Do you think your data is a conditional mixture, sometimes one distribution is active and other times both are active. I saw this type of distribution for defects, particulate on product. Most of the time a random defects of a particular distribution but a portion of time another mode of defects were added. If that is the case, your predicted model might look something like Lognormal(p1, p2) +(Random Uniform()>p3)*p4*Exp(), where p1-p4 are parameters to be estimated. These will need starting values. I found that Distribution>Continuous Fit> Normal Mixture 2 will give a good starting point for p3 the proportion of the baseline distribution.
* JMP Pro Life Distribution fits Time to fail data, but pretend your data is time. It has a Fit Mixture for the distributions that you expect.

Attached is a table with random simulations there are two mixtures, results will vary since the data has formulas. The attached scripts for Life Distribution might not work, depending upon your version of JMP. The Life Distribution for the mix of Weilbull and Exponential for my simulations.

Random.jmp