topic Re: Proc Freq and Variable Worth in Statistical Procedures

Proc Freq and Variable Worth

DmytroYermak — Wed, 08 Mar 2023 07:36:26 GMT

Hi all,

Proc Freq is being used in assessing variable importance. For this purpose the results of the chi-square test for ordering the predictor variables by importance is being used as the variable worth:
Worth = -2log(p).

Could you please explain how to manage the cases where p is not a number but rather an inequality like here:

If i'm wrong with the interpretation please avise how the variable worth should be calculated.

Thank you!

Re: Proc Freq and Variable Worth

sbxkoenk — Wed, 08 Mar 2023 08:10:55 GMT

Hello,

The p-value being displayed as <0.001 ... ... that's only a format being used (pvalue6.4).

Use ODS output to send that output object to a table.
In the table you can then see the real internal value.

Formats for p-values and odds ratios in SAS
By Rick Wicklin on The DO Loop August 15, 2016
https://blogs.sas.com/content/iml/2016/08/15/formats-p-values-odds-ratios-sas.html

Cheers,

Koen

Re: Proc Freq and Variable Worth

DmytroYermak — Wed, 08 Mar 2023 08:49:02 GMT

Thank you, Koen!

Re: Proc Freq and Variable Worth

DmytroYermak — Wed, 08 Mar 2023 08:59:13 GMT

P.S. Just one comment after some manual calculations: Log in Worth = -2log(p) is the logarithm on base 10 ('LOG10()').

Re: Proc Freq and Variable Worth

sbxkoenk — Wed, 08 Mar 2023 09:21:05 GMT

Hey @DmytroYermak ,

@DmytroYermak wrote:
P.S. Just one comment after some manual calculations: Log in Worth = -2log(p) is the logarithm on base 10 ('LOG10()').

Is that a question or a statement??

LOG-function in SAS has base number e.
LOG10-function in SAS has base number 10.
LOG2 function in SAS has base number 2.

Or do you simply mean that for the worth to be correct LOG10 is needed?

Cheers,

Koen

Re: Proc Freq and Variable Worth

StatDave — Wed, 08 Mar 2023 16:26:09 GMT

I think this is usually computed as -log10(p-value) as is used in the Manhattan plot in PROC MULTTEST to make the most significant p-values easier to see in a large set of p-values. See the example titled "Adaptive Adjustments and ODS Graphics" and the PLOTS= option in the MULTTEST documentation.