Confidence Interval for Skewed data?

[#1]

Hello everyone,

I have a data distribution that is both negative and positive (it's the difference between two values) and it is not normally distributed by shapiro-wilk's test. Is there any way in JMP to calculate a confidence interval for skewed data?

Many thanks for your advice!

[#2]

Confidence interval for what estimate?

[#3 · Accepted solution]

As @Mark_Bailey suggested it would help to know you are trying to do, but two general methods include:

• Normalize the data first: transform it using formula columns or separate it into multiple sets, calculate confidence intervals, and then transform the results back.
• Use bootstrapping to develop confidence intervals

[#4]

I looked into bootstrapping (I have JMP pro) and I think this is the most appropriate. Thanks for all the suggestions!

[#5]

Hello, 

Thanks for the replies so far. I'm a novice at stats and just want to do basic stats for simple clinical studies. I'd also like to stick to JMP as this is the only stats software I have.

I'd like a confidence interval for the mean and because the values are differences between other values, I have negative and positive values.  I am interested especially if the interval crosses 0.

Thanks again!

[#6]

@ih has some good suggestions. The bootstrap estimate requires JMP Pro and you might have JMP instead.

The distribution model fitters in the Distribution platform provide confidence intervals. Here is an example using the sample Big Class data table and the variable weight. I launched the platform with weight, click the red triangle next to weight and selected Continuous Fit > Fit Lognormal because this distribution is skewed. (Another model might provide a better fit, but I am just illustrating the approach.) Notice that the table of estimates includes both the point estimate and interval estimate for the distribution model parameters.

Does this approach help?