Hell JMP Community,
I am writing to see if anyone has any suggestions for my analysis comparing mortality distributions between fly populations. I tested 4 doses at 4 time points and I would like to see how the proportion of losses varried over time between populations to understand variance in response to exposure at different levels.
Presently I have the table set up as the percent of total loss at each time for each cage in a univariate format and have been running it as X=Time, Y/response=%loss, GROUP=Population, and then fit curve => Quadratic. I then have been using BY=Dose to compare the distributions between population at the different doses.
Does anyone have any better suggestions for another ways to analize this dataset.
I appreciate any input- Joe
Could you kindly provide a sample of your data and demonstrate what you are trying to do and achieve or share the script and the test data so the users in the community can reproduce and relate to what you are seeking ?
If I understand what you are trying accomplish, maybe Graph Builder could help? Place Time in the X drop zone. Place % loss in the Y drop zone. Place population in the Group drop zone. Choose the box and whisker plot from the graph type icons. The advantage of the box and whisker approach is it will show the middle and spread of each distribution. Not much there to show since it sounds like you only have 4 observations for each time/group combination...but it's a start.
You could use Survival Analysis. Rather than computing the % mortality, use the raw data showing the number that survived and the number that did not. It will be censored data since those that survived only survived at the last time observation (unless all died by the end of your measurements - which would call for a modification of the analysis - it would no longer be censored). Survival analysis is set up to analyze the time to death as a function of your other variables and the graphs it will display will look like what you are trying to graph (or what you would get from graph builder as suggested by others). Given how little data you have, I doubt you will get any meaninful inferential results, but the graphs and analysis should comport with the question you are asking.
Following on Dale's reply and the nature of your raw data, you could use a probit analysis. Set it up using Fit Model with the Generalized Linear Model personality, the binomial distribution, and the probit link function. Be sure to enter both data columns for the number that did not survive and the number tested in that order.
You can use any linear predictor you want, as is the case in any regression.