I applied online. The process took 2 weeks. I interviewed at Google (Mountain View, CA) in Jun 2009
Interview
Submitted resume online. Sailed through a 1-hour technical phone interview. The question involved bayesian analysis of drug testing (determining false positive rate, etc). It's a classic and it hit my sweet spot. Easy.
Interview questions [2]
Question 1
The on-site interviews were not so easy. One thing to note: I have 20 years of experience and a strong track record of getting results. The interviewers had zero interest in what I've done before. Instead it was white boarding all the way. Felt more like an old-school math class -- go to the board and solve this problem -- than a job interview. Toughest question? Honestly, this was 4 years ago and I don't recall precisely. I think it was how to determine mean standard deviation of two blended datasets for which you know their means and standard deviations (and maybe Ns) but for which you've lost the original data. Turns out it's a classic you'll probably encounter in first year statistics, but I hadn't thought about it in about 25 years and so stumbled.
I applied through an employee referral. I interviewed at Google in Jan 2013
Interview
Get contacted by a recruiter to answer 13 questions on a questionnaires. Then a phone interview was scheduled. The identity of the interviewer was not told and he didn't introduce himself much unless you asked.
I got contacted by recruiter three months after submitting a resume and arranged a phone interview. The interviewer was nice and led me to answers. Most questions are about basic statistical concepts, especially the OLS assumptions and remedies for violation, hypothesis testing. He didn't ask anything about coding. The second interview was about open questions, for which she described a scenario and asked me about solutions. The part was hard for me. I had no idea what she was looking for, for some of the questions.
Interview questions [1]
Question 1
If each of the two coefficient estimates in a regression model is statistically significant, do you expect the test of both together is still significant?