BookmarkSubscribeRSS Feed
Choose Language Hide Translation Bar
aj_p
New Contributor

GLM

I am new to JMP, but very interested to use it for my data analysis. I have a data set: 8 treatment groups, each group have 12 to 14 flies, each fly was provided with 4 different fruits together (4 choice)  for oviposition. Oviposition counts on each fruit out of total number of oviposition per fly is the final data. I am studying the effect of host fruits and treatments on the oviposition, and how treatments alter host preference?. As it is a count data I am not happy with ANOVA. I decided to use GLM. Should I use nest at any point here?

0 Kudos
1 ACCEPTED SOLUTION

Accepted Solutions

Re: GLM

It sounds like you crossed 8 treatments with 4 fruits. (Are the treatments from one factor with 8 levels, two factors, or three factors?) I assume that your choice for a GLM model is Poisson distribution with the canonical log link function, right? Why do you think that treatment and fruit are nested?

Learn it once, use it forever!
3 REPLIES 3

Re: GLM

It sounds like you crossed 8 treatments with 4 fruits. (Are the treatments from one factor with 8 levels, two factors, or three factors?) I assume that your choice for a GLM model is Poisson distribution with the canonical log link function, right? Why do you think that treatment and fruit are nested?

Learn it once, use it forever!
aj_p
New Contributor

Re: GLM

Thanks lot for the response. In my case the treatments are different
bacterial treatments but no levels inside.
I have arranged the data in 3 columns C1: Treatment, C2: Fruit, C3:
Oviposition count. Count in every four rows are from the same fly of a
particular treatment. Moreover, I want to include the total count of each
fly also into the analysis. So I get confused with nesting. Sorry. Please
suggest me a better way to structure my data for the GLM.
0 Kudos

Re: GLM

I don't see any need to structure the data differently. You can enter the counts in the Y analysis role and then enter Treatment, Fruit, and Treatment * Fruit (interaction effect) in the terms list.

 

At the same time, I don't see the need to model the total. This would mean summarizing (e.g., 'rolling up') the counts of each fruit. There is nothing wrong with doing that operation but what does it gain the analysis?

 

One consideration for another design would be to include more flies. Possibly replicate treatment (Treatment / Fruit combination with new fly) or simply introduce more than one fly to each combination. (I guess that multiple flies raises other issues. Simultaneous introduction of multiple flies might modify their behavior. Sequential introduction might be subjected to a carry-over effect from previous fly.)

Learn it once, use it forever!
0 Kudos