Level: Power
William Q. Meeker, Professor of Statistics and Distinguished Professor, Iowa State University
Chris Gotwalt, JMP Director of Statistical Research and Development, SAS
In this session we will show several examples of how JMP has made it easy to take advantage of simulation-based statistical methods. We will begin with reliability applications and show how to obtain more refined inferences when there are very few observations using Bayesian techniques in the Life Distribution Platform and in the Fit Life By X platform. We will then introduce the One-Click Bootstrap in JMP Pro with several use examples, with special emphasis on the fractional-random-weight bootstrap.
The classic bootstrap, based on resampling, has for decades been widely used for computing trustworthy confidence intervals for applications where no exact method is available and when sample sizes are not large enough to rely on easy-to-compute large-sample approximate methods.
There are, however, many applications where the resampling bootstrap method cannot be used. These include situations where the data are heavily censored; logistic regression when the “success” response is a rare event or where there is limited mixing of successes and failures across the explanatory variable(s); and designed experiments where the number of parameters is close to the number of observations. What these three situations have in common is that there it may be impossible in a substantial proportion of the resamples to estimate all the parameters in the model. The fractional-random-weight bootstrap method can be used to avoid these problems and provide trustworthy confidence intervals.
Next we will demonstrate how to do power calculations using the Simulate Feature in JMP Pro. This makes it straightforward to do power calculations for many analyses throughout JMP, particularly designed experiments where the outcome is binary. Finally, we will show how to obtain confidence intervals on neural network predictions using the Save Bagged Estimates option in the Profiler.