Profiler Confidence Intervals with One Data Point

[#1]

I have a question about the confidence intervals in the fit model > profiler platform: How does JMP generate confidence intervals when there is only one datapoint in a given category? I was under the impression you needed a standard error (which requires some measure of variability) to calculate it, but that shouldn’t exist with only a single datapoint. Here’s an example of what I’m referring to:

I excluded the labels, but the far right condition (with the largest confidence interval) only has a single datapoint feeding into it. Are there assumptions JMP makes to calculate that or maybe I’m missing something fundamentally about the need for multiple datapoints to estimate the standard error for parameter estimates (?).

Thanks!

[#2]

Hi,

The confidence interval is based on estimation of the error for the whole model. The error can be estimated as long as you have degrees of freedom that are not required for model parameter estimation. Basically, as long as you have more rows of data than parameters in the model, JMP should be able to estimate error and confidence intervals.

This is not an independent estimation of error for each unique combination of conditions based on repeated measures for each treatment. That is not how multiple linear regression models work.

For a proper understanding of confidence intervals and how to interpret them I would recommend you get into the mindset of statistical thinking. And there is a really good free online course for that!

I hope that helps,

Phil

[#3]

Hi Phil,

That's very helpful, thank you! I've heard about the STIPS program before, so I'll have to make some time to work through it.

Cheers,

Tim