Improve linear model

[#1]

Hello,

Attached is a Residual by Predicted Plot for some model built on historical data from some process.

When you see such a graph for residuals, what would you recommend to change in the model in order to solve the slope and make residuals be randomly distributed around 0?

Thanks in advance!

[#2]

Can you share the data and model specified that resulted in the residual plot you shared? And some context on the practical problem/issue you are trying to resolve? Lacking these three things it's hard to recommend a next step that would achieve your goal.

[#3]

Thank you for you reply

Unfortunately, the data are sensitive and cannot be shared :(

i will try to optimize with all the feedbacks i've got

[#4]

Understood about the '...sensitive...' nature of your data. Not uncommon. But did you know within the JMP ecosystem there is a pathway to anonymize data that is easy to use and works for many users? It's described here in the JMP documentation: Anonymize Data in JMP 

[#5]

Hi @LambdaCapybara4 : As @P_Bartell  indicated, it is hard to diagnose with the limited information given. That said,  my guess would be that your model is missing a predictor.

[#6 · Accepted solution]

I retract the idea of no intercept. I also think it is a strong linear effect missing from the model.

[#7 · Accepted solution]

Just reviewed my Fit Model dialogue window and found out "No Intercept" was checked!

Unchecking it resolved the problem, attached is the resolved Residuals Plot.

Can I ask you how did you know that from this plot? Besides tons of experience, that i guess you have...

And, if you see abnormalities in the attached improved plot, I'd also be glad to know.

Thanks a lot!

.

[#8]

@LambdaCapybara4 : Good discussion about this here.

https://stats.stackexchange.com/questions/200103/what-can-be-inferred-from-this-residual-plot