Hi JMP users, I've tried to figure this out but the statistics guide is vague enough that I can't quite get to the answer that I need to make this work.
I'm trying to get JMP to work for computing LD50 values based on observed mortality proportions.
The documentation and routines in JMP seems to rely on a logistic regression and imply that inverse prediction at 0.5 will yield an LD50. The problem is that this approach relies on binomial (live/dead) data for each individual in the population at each [log] dose level...I don't have individual animal data, just the number dead from the number challenged at each dose.
So, for example, I perform a log-dilution series on a drug and administer the drug at each dose level to each of ten animals. At level one, none survive, 2->none, 3->75% die, 4->35%, 5, 6, 7, all animals survive.
To use the logistic regression platform, one would enter a 1 or 0 (or "live"/"dead") for a row for each animal, repeating this for each dose level (70 lines in the table), and this would work fine, but I don't know which animal is which.
What I'm looking for is a way to enter 1, 1, 0.75, 0.35, 0, 0, 0 as a response to log dilution levels [negative] 1,2,3,4,5,6,7.
You will need to know the actual number that died or survived at each level rather than just the proportion. Then have a binary outcome variable, with 1's and 0's or "Dead" and "alive" and add an extra column with the count. Place the count column in the FREQ field of the logistic dialog. Then inverse prediction will get you the LD50 (or whatever LD you require) with confidence intervals.
A lot of bioassays use probit rather than Logit. JMP can do Probit using Generalized Linear Model under Fit Least Squares, just select Binomial Distribution and Probit Link. This platform doesn't have the reverse prediction option (Maybe next version, please ?) But you can use the Prediction profiler with confidence bands to read off the inverse prediction and approximate confidence limits. Just slide the vertical red dotted line until the vertical line hits 0.5 (or whatever LD you're looking for). Then you can read off the log dilution and confidence interval off the graph. This CI is probably not the same as the fiducial limits that another package would produce, and I'm not sure it matches the CI that SAS Proc Probit gives you. IF you have SAS you could submit statements to SAS and see what you get.