topic Re: Why are the abscissa leverage values shifted in the Fit Model platform? in Discussions

Why are the abscissa leverage values shifted in the Fit Model platform?

pcarroll1 — Fri, 09 Jun 2023 00:35:54 GMT

There is an old discussion where the question above was posed. The answer was to view this.

https://itunes.apple.com/podcast/applied-statistical-essentials/id20769613

However, when I clicked on the podcast address below, the screen showed connecting for a long time and then I gave up. Am I too impatient or is that podcast no longer available? If the latter, is there another way I could understand this?

Re: Why are the abscissa leverage values shifted in the Fit Model platform?

Dan_Obermiller — Fri, 02 Jul 2021 15:54:18 GMT

Maybe knowing how a leverage plot is constructed will help?

Consider a multiple linear regression model with Y as the dependent variable and X1, X2, and X3 as the independent variables. To create a leverage plot for one of the independent variables (for example, X1), do the following actions.
• Regress Y on X2 and X3. These residuals are the vertical axis of the leverage plot.
• Regress X1 on X2 and X3. These residuals are the horizontal axis of the leverage plot.

This comes from the ANOVA and Regression course offered by SAS.

Re: Why are the abscissa leverage values shifted in the Fit Model platform?

Mark_Bailey — Fri, 02 Jul 2021 16:18:07 GMT

@Dan_Obermiller boiled it down to the essential fact: it is a scatter plot of two sets of residuals, each one from a different model. It is not a plot of the original variables.

You might find this page about Leverage Plots in the JMP Help enlightening if you want more detail.

Re: Why are the abscissa leverage values shifted in the Fit Model platform?

pcarroll1 — Tue, 06 Jul 2021 11:48:28 GMT

Dan,

Very good. That simple answer is what I needed to understand it. In the past (20+ yrs ago) I used to work in Splus (R) and wrote code to make my own partial residual plots. However, I plotted them against the actual X values. I never considered the second point that you made, but I guess that's done to attempt to account for cross correlation among the X factors.

Thanks,

Pat