topic Re: Residual analysis DOE in Discussions

Residual analysis DOE

William29 — Mon, 10 Jun 2024 17:57:51 GMT

Dear community,

I would like to ask for a bit of support on the interpretation of raw residuals. I analyzed data using 3K factorial with 3 factors and three levels, total runs 27. I got the next raw residuals vs predicted values.

I am not pretty sure that the plot is ok, the variance seems quite equal from negative and positive a small pattern in the center I am not sure if there is a big issue. if someone can help me I will really appreciated.

Re: Residual analysis DOE

statman — Mon, 10 Jun 2024 19:32:54 GMT

I'm a bit confused. You say you ran a 3^3 factorial (27 treatments), but in the picture you attached, it mentions 3 blocks? What are the blocks? What was the model you used? What do the other diagnostic statistics look like? (e.g., R-Square-R-Square adjusted, RMSE, p-values). I assume you are trying to create a useful model. Does the model work?

Try plotting the residuals some different ways

1. vs. run order

2. Save residuals and look at the distribution.

3. Take a look at the leverage plots.

4. Residuals vs. predictor (that is each x)

5. JMP will also plot studentized residuals

It does look interesting. It looks like the first 3rd is quite what you'd expect. The second third looks strange, are these center points? Then the last third look like it trends. None of these may matter if the order is not meaningful.

Re: Residual analysis DOE

William29 — Mon, 10 Jun 2024 22:49:12 GMT

I appreciate the reply, regarding the questions:

-It is a 3^3 with 27 runs full factorial, it has 3 blocks since the 27 runs were divided into three blocks of 9 runs. I am using a software statistica, that generates these residual plots, in this case the residual vs the predicted. I have generated as well the normal plot of residuals and the residuals vs case number.

I was planning to do an sqrt root transformation of the response variable

Re: Residual analysis DOE

statman — Tue, 11 Jun 2024 13:21:04 GMT

You have a significant block effect. BTW, while we are often happy to answer any questions regarding the application of statistical method, this forum is intended for JMP users.