topic Re: Specifying levels of sensitivity in Statistical Procedures

Specifying levels of sensitivity

lbogar314 — Tue, 19 Jun 2018 18:33:09 GMT

For proc logistic, can you look at specific levels of sensitivity?

For example:

proc logistic data = dataset;

by group;

model y = x / ctable;

roc "x" x;

run;

Ctable gives the classification table but is there a way I can see everything that gives me except with specific levels of sensitivity (70, 80, and 90%)?

Re: Specifying levels of sensitivity

PGStats — Tue, 19 Jun 2018 18:43:38 GMT

model y = x / ctable pprob=(0.7 0.8 0.9);

Re: Specifying levels of sensitivity

lbogar314 — Tue, 19 Jun 2018 19:20:02 GMT

It was my understanding that pprob specifies cutpoints. Basically what I am trying to do is look at Sensitivity levels of 70%, 80%, and 90% as the cutpoints.

The exact wording of what I am working on requires me to "provide specificity for the following three levels of sensitivity: 70%, 80%, 90%"

Is there a way to do this?

Re: Specifying levels of sensitivity

StatDave — Wed, 20 Jun 2018 12:35:37 GMT

It isn't possible to specify the sensitivity values to report in CTABLE results. The best you can do is specify a fine list of PPROB= values and see which cutpoints correspond to the desired sensitivity values. If the range of cutpoints that you see without PPROB= is, say, .3 to .7 then you could specify something like PPROB=(.3 to .7 by .05). Once you find the cutpoints that bracket the desired sensitivity values, then you can rerun with even finer PPROB= list to get what you want.

Re: Specifying levels of sensitivity

lbogar314 — Wed, 20 Jun 2018 12:57:57 GMT

Ah, thank you! That's what I ended up doing but was hoping there was a better way because my data is stratified into several different groups and I need the stats for each so it is a lot to look through. Is there a different way to do it that I am not considering?

Thank you so much again!