The biggest piece of advice I could give is to take a course in microeconometrics/labour econometrics as a part of your course. If your course coordinator won’t let you, beg. If they still won’t let you, then go off-line for a week or two and properly digest Mostly Harmless Econometrics (or if your stats isn’t too good yet, Mastering Metrics). If you want to go and work in health analytics, then replace what I just wrote with the equivalent for research design.

Why learn microemet? Basically, many of the big questions in business are of the form “what will happen if we do x”. Predictive models that aren’t informed by causal reasoning do *terribly* at this question–they answer the question “what do we see happening to y when we see x”. Inferring what will happen to y when you fiddle with x is a difficult task when all your data come from a world in which you did not fiddle with x. Too often we come across people with great technical chops who aren’t even aware they’re making mistakes when answering these questions. Don’t be one of these people.

The second biggest piece of advice would be to not become too enamoured by the sexy end of data science (especially predictive algorithms), but *do spend the time learning this stuff in depth*. Often the simple stuff done well is far more useful to real-world decisionmaking.

Third: read very widely.