I don't remember the exact details anymore, but more or less they claim to address causation by 1) starting with a non-schizophrenic population, 2) segmenting by marijuana use, and 3) evaluating schizophrenic status N years later. The theory is that by limiting 1) to non-schizophrenics, they've factored out the possibility that schizophrenia drives drug use rather than the other way around.
Of course, most schizophrenics are not actually diagnosed until waaaay past the point where they are showing symptoms, and if you've ever known a schizophrenic who was diagnosed in college, for instance, you know that that diagnosis usually comes at the end of a long chain of subtle and not so subtle mental issues, shrink appointments, and life problems that start much earlier but don't present as full blown schizophrenia. So the whole supposed filter in 1) is complete trash, and they're probably including a lot of people who are already symptomatic, just not to a clinically diagnosable degree.