PDA

View Full Version : Recruitment Discrimination

agulland
01-28-2011, 09:21 AM
Hi,

I did stats years ago and am now quite rusty. At work we're looking at a straight forward problem: how to demonstrate we have no discrimination in our recruitment process.

We have data,

Recruitment Step Number Male Number Female
2. Interview 18 12
3. Job Offer 3 2

In the above data I'm assuming the "CV Received" data is our control group as we can't affect this. And so we expect same ratio to go through to interview and then job offer. How do we show that it is not likely that we're discriminating?

thanks for any help

AL

spunky
01-31-2011, 05:12 AM
you would be surprised that showing similarity instead of differences through the regular statistical machinery is NOT a straightforward problem at all. the logic of statistical inference is always discomfirmatory and you can never provide evidence in favour of the null hypothesis of similarity between groups. you can only reject it.

there are only two solutions to this problem that i know of... one which i like and one which i dont.

the one i like:
take a look at jacob cohen (1990) "things that i have learned so far" (just google it, it should pop up) and how he uses the power of a statistical test and type-2 error rates to imply no difference between samples.

the one i dont like:
take a look at what's going on in the biostatistical community with "equivalence testing". they essentially use this to show that there is no difference between treatments (like new VS old drugs) trying to invert the logic of confidence intervals. i say i dont like it because they kind of introduce some arbitrary criteria (what they call "clinical significance") and i have some serious criticism about how they define their interval of "non significance" based on the likelihood ratio test. however, medical researchers use that thing like crazy and nobody seems to care so you might as well jump in the bandwagon.

if you (or anyone out there) knows how to show statistical similarity through hypothesis testing do let me know because i've always been intrigued with that.