I think the authors raised an important point. We often look at the average assuming cases and effect estimates are homogeneous, but more often they are not.
My thoughts about Jim's question on whether it is worth the while to publish cross-sectional evidence on parenting-alcohol relationship: it would come down to the assumptions. If it is safe to assume the relationship is a positive feedback loop, then yes. If it is a one-way positive or negative relationship, then yes. Would provide initial estimates for future longitudinal studies. If it is a two-way negative relationship or a mixture of positive and negative relationship, then probably no, because it becomes complicated very quickly depending on when kids are assessed in a cross-sectional survey.
But again, that is population average. Does it apply to everyone? Maybe not. In addition, there may be unobserved heterogeneity in the background. Parental drinking, peer affiliation, number of kids in the household, neighborhood environment...
Comments from Jim:
Chicken and egg problem: What if drug use causes parents to relax their supervision and monitoring?There is a quite rational decision in epidemiology to start
with relatively inexpensive and logistically feasible study designs, such as
case-control research, before moving on to the more expensive prospective
two-wave or longitudinal multi-wave designs that can throw more light on issues
such as uncertain temporal sequencing. (Here, some epidemiologists would turn
to the within-field jargon term of 'reverse causality,' but let me again warn
that communication across disciplinary boundaries is far more important than
any display of group membership as represented by use of within-field jargon
terms. 'Uncertainty about temporal sequencing' is apt to serve well in a public
forum or research article, whereas epidemiology 'bull sessions' might be
appropriate venues for the within-field jargon of 'reverse causation' and the
like.)
This new contribution is a very interesting longitudinal
multi-wave study that deserves attention because it has a capacity, within the
limits of its assumptions, to estimate the degree to which drinking might cause
parents to back off and relax their supervision and monitoring behaviors. The
assumptions, both conceptual and methodological, should be studied with care,
and after you consider the evidence, ask yourself whether it was a mistake for
journals to publish the early articles with no more than cross-sectional or
two-wave study data on the monitoring hypothesis.
If interesting comments are provoked, I will leave a later
comment to tell you what I think.
