It's very difficult to correlate interview performance and job performance. No one is willing to hire at random to really get the data. Companies only look at who got hired.
I've seen some companies decide "well, whiteboard interviews are flawed" and just scrap the whole thing and replace it with something else... that's much more flawed. They're really not being honest with themselves about what the flaws exactly were in whiteboard interviews and if those are fixable, or what the flaws are in their new process.
There's a lot that can be done to make whiteboard coding/algorithms interviews less flawed.
- Be clear with candidates and interviewers about expectations. What knowledge is okay to assume? What knowledge is not?
- Clear guidelines on what makes a good question. A lot of companies ask very simple or well-known questions but think they're emulating what Google does. They are not.
- Good interview training. Most companies don't have interview training, beyond maybe basic legal stuff. Even the companies that do have training typically have crappy training.
- Interviewers should candidates with certain things that might not be obvious. For example, writing examples of input and output on the whiteboard is helpful. Don't rush into coding.
There is A LOT that can be done to make whiteboard interviews more effective.
You can also replace them with some other things, but understand that those will come with their own flaws. Scrapping whiteboard interviews is not going to deliver you some perfect system.