History as the Mediator of Meta-Analysis

In regression class I finally understood the difference between mediation & moderation.  A significant break through in the way I thought of variables influencing one another.  The most interesting thought was the relationship between time & meta-analyses. 

During a meta analysis we collect as many studies as possible which experimented on or evaluated  a particular variable.  We then compile these studies, do some rather interesting statistics and postulate a conclusion about the influence of this  particular variable on a dependant variable.  An example would be taking all the studies on Self-Esteem and exploring its influence on the personality trait extroversion.  However, in my work with the psycho-market model, slopes & relationships gain & lose significance over time.  A relationship may be significant one year and non-significant the next.  Time has mediated (or something else) the significance of this relationship.  In running experiments we try to control for "the something else," but we often never control for time.  I propose that the fluctuation of effect sizes & significant/non-significant relationships are mediated by not only standard experimental issues, but time.  I would go as far as saying that time would account for the majority of variance above & beyond standard experimental issues related to the fluctuation of meta analysis data.

Several time & other related errors do occur that need to be addressed.  When conducting a psychological experiment we often forget it was done at a certain period in time, with a particular zeitgeist.  Therefore when conducting meta-analyses we may be combining significant & non-significant time sensitive data which may eliminate any true relationship or produce a type 1 error.  In addition, every experimenter knows of the bias journals have towards significant results & meta-analyses typically only use these results, collecting very few datas from non-significant sources.  Lastly, it's erroneous to think that the knowledge an individual holds about Self-Esteem or Organizational Justice was the same when the study was first conducted to when the meta-analysis was calculated.  Some erudite professors may rebuttal with "This is why we operationalize constructs."  My response to that would be look at the literature on survey responses & actual occurrences of these events. I think one would be hard pressed to find someone who hadn't incorporated their own version of what the variable means.

In short, I would attest that the fluctuation across experiments, evaluating the same variable, is most attributable to time above & beyond standard experimental errors.  The question becomes how can we control for time?  I have outlined some suggestions.

1) Use a linear variable to depict the change of time.

2) Create a zeitgeist construct to be included in the analysis

3) Understanding this may be one variable that is uncontrollable & accepting it's placement within the experimental meta-analytic frame work.  Therefore, not accepting meta-analyses as the end all tell all, as they often are.