Too early to evaluate effects of cannabis policy change?

Consider two time series: (1) the rate of suicide deaths, and (2) the unemployment rate.

Does the unemployment rate influence the suicide rate?

I maintain that research on this question remains non-definitive until we can specify the plausible distribution of  the analogue to an infective agent's incubation interval or a drug's induction interval.

One alternative to specifying a plausible distribution is to assert that any such epidemiological effect can be estimated almost immediately after a shift upward or downward -- i.e.  an almost immediate effect within a span of 1-48 months. In the unemployment-suicide context, this assertion ignores commonsense ideas about buffering against adversity that is connected to individual wealth (e.g., size of an employee's retirement savings account) or social capital (e.g., a Dutch uncle or aunt).

Another alternative is to try many different lag values, which converts the activity to a fishing expedition, with a false discovery penalty to be paid as a consequence of being undisciplined in these matters.

(A hidden fact is that the first assertion essentially can become a slippery slope in the direction of the fishing expedition, unless the research team finds evidence of an immediate effect. Why? One can look into the discussion section of a failed 'time series experiment' and see why. One limitation is almost guaranteed to show up there -- i.e., "It is possible that we did not specify the lag time correctly.")

To this point, the kind reader may see where I am going with this line of reasoning:

(1) Any attempt to specify an analysis approach for problems of this type must begin with Bayesian thinking about plausible distributions for the incubation or induction interval or its "lag time" analogue.

(2) We should be skeptical when we see published evidence based on an incautious assertion of a short lag time with no evidence of advance thinking about plausible alternatives.

So what are we to make about the series of articles being published on effects of changing cannabis policy environments in these United States of America?

* When they assert a short lag time, but consider no alternatives to the just-mentioned slippery slope?

* When they try out many different alternatives willy-nilly?

See Pacula & Sevigny, 2014

Then, for some technical details to help in advance specification of Bayesian distributions :

Pacula & Lundberg, 2014

Estimation of drug policy impact is not a simple matter.