Help with probit analysisHi All,
A customer in genetics sent me some anonymized data, they are interested in doing probit analysis.
It seems that they need to first figure out the distribution and somehow use that information to do a probit analysis on the data.
I am struggling a little in what to do to get started, can anyone help ?
Here is the email:
"Here is an anonymized table of results for LoD processing:
Concentrations tested
|
Positive replicates (N)
|
Total replicates (N)
|
5
|
21
|
21
|
1.667
|
20
|
21
|
0.556
|
12
|
21
|
0.185
|
2
|
21
|
0.0062
|
4
|
21
|
For the LoD determination, we would like to know which prediction model we should use when performing the probit analysis. Before using JMP, we were using Minitab which was offering the Weibull, LogNormal and the LogLogistic prediction models and we haven’t been able to find these on JMP. "
Thanks.
Elodie
Re: Help with probit analysisHi Elodie,
Fit Curve or Fit GLMM will do some of what they are after - a mixture of Probit, Logit and Weibull growth models (see scripts in attached table). I'm wondering if the other two options are achieved with a log transformation of the concentration variable then applying the Probit/Logistic models?
Malcolm
Re: Help with probit analysisYou can use GLM in Fit Model to fit a probit with binomial errors.
Re: Help with probit analysisNote: GLMM will have the probit link for the binomial distribution (which is presumably what they have when they have N positive out of N replicates) in JMPPro 18. The lognormal distribution has also been added to GLMM in v18.
But as it appears they do not have any random effect features requiring a mixed model (from that data table sample), @jiancao 's pointer to GLM would likely work and is available now.
Re: Help with probit analysisThank you very much @malcolm_moore1 , @eclaassen and @jiancao, it was really helpful.
