Professor Long introduced me to Agresti & Finlay's (1997) non-exhaustive list of necessary (but not sufficient) conditions to establish causation between two variables, X and Y:
1) a theoretical or common sense linkage,
2) empirical association (correlation),
3) elimination of common causes: some other variable must be ruled out as a cause of the correlation,
4) responsiveness: altering X leads to an alternation in Y,
5) asymmetry: X must cause Y, and not vice vera.
" Non-experimental studies can only address #1 and #2, and partially address #3. The most common method of addressing #3 is statistical control. Experimental studies that use random assignment can address all the conditions. However, the list of conditions is not exhaustive and some would not admit causation even if all five of these conditions were met. In addition, random assignment works in theory but may not work in practice. A check of random assignment consists of comparing randomly assigned groups before the treatment is introduced to see if they are equal on a number of extraneous variables" (Supplemental Notes, Statistical Methods II: Regression and the General Linear Model, Fall 2002).
Ted Huston's PAIR Project (University of Texas) has an excellent discussion about "Temporal Issues in Studying Marriage: Conceptualizing Cause and Effect". He discusses such issues as whether the proposed cause-effect relationship is psychological, interpersonal, contextual, or something else ... and he also talks about direction of causality and the temporal shape of causal conditions. It is WELL worth reading as we think about issues of research design in family and relationship science. Enjoy!