In this video, a Google hiring committee were given their own anonymized hiring feedback and wouldn't even hire themselves.

To complicate this: 1. The range in interview performance of those who get hired is really quite narrow. So you're looking at essentially the difference between a 3.1 and a 3.3. You'll need a lot of data to try to find a correlation there.

2. The interview score is only a partial measure of interview performance. If you and I experience an identical interview, I might call that a 3.1 and you might call that a 3.3. The hiring committee will hopefully understand that a 3.1 from me and a 3.3 from you are equally good (because they see our past scores). This makes the data on interview performance harder to correlate, and you'll need even more data. (Or you'll need to properly take this into account in the study, by looking at a score relative to that interviewer's past scores, rather than the score itself.)

3. The above point, repeated for job performance.

4. When a candidate with a lower score gets hired, they're more likely to have something that offset these lower scores. The data set of people who got hired is {people who interviewed well and had good or bad resumes} + {people who interviewed just okay but had great resumes}. When you correlate interview performance of those who got hired, the lower scores in there are skewed towards people with great resumes (or referrals or something else). That factor could totally eliminate any correlation between interview scores and job performance.