topic Re: "Blue Line" Significance Test in Effect Estimates in Fit Model in Discussions

"Blue Line" Significance Test in Effect Estimates in Fit Model

ar2 — Wed, 16 Nov 2016 21:02:09 GMT

Can anyone explain why blue line in Effect Estimates in Fit Model which (implicitly) signifies significance is set at P = 0.01 rather than P= 0.05. Is this alterable in preferences?

Thanks for any advice/guidance/Logic

Re: "Blue Line" Significance Test in Effect Estimates in Fit Model

Jeff_Perkinson — Tue, 22 Nov 2016 14:17:42 GMT

I'm not sure which "Blue Line" you're talking about. Do you mean the one in the Sorted Parameter Estimates report?

I believe those vertical blue lines are set at the .05 significance level.

If that's not the part of the report you're asking about can you post a picture of the lines you're interested in?

Re: "Blue Line" Significance Test in Effect Estimates in Fit Model

ar2 — Tue, 22 Nov 2016 14:38:47 GMT

Jeff - it is the blue line in the Effect Summary report in Fit Model which is set at logworth value equivalent to P= 0.01 (I believe)

Re: "Blue Line" Significance Test in Effect Estimates in Fit Model

David_Burnham — Tue, 22 Nov 2016 17:03:00 GMT

This is just my personal interpretation.

The table is showing p-values but also LogWorth values. Personally I would use these in difference scenarios:-

If I'm doing traditional statistical modelling (lets say modelling DOE data) then I would be interested in p-values. Then I would expect the blue line to be to 0.05 threshold. I would also want the table to use p-value formatting (i.e. colour coded p-values) - something it doesn't do.

If I'm dealing with much larger volumes of data and doing predictive modelling then I kind of expect everything to be statistically significant, so LogWorth is more helpful to me in terms of ranking variables. Then a LogWorth threshold is useful to me, maybe 2 or 3 or whatever helps me apply some form of Pareto principle. In this context, the blue line make sense, giving me a threshold of 2 for LogWorth.

So this is what I would like to see:

An option to format the effect summary either in terms of p-value or LogWorth. In p-value format, I want to see the p-value formatting and a thresold line of 0.05. In LogWorth format I don't really want to see p-values. Just show me LogWorth values and the threshold line. Happy to have it at 2, but give me the choice to change the value.

Re: "Blue Line" Significance Test in Effect Estimates in Fit Model

ar2 — Tue, 22 Nov 2016 17:23:42 GMT

David - really helpful insight. Thank you. And I agree with your proposed format chnage suggestions!