Temporal Issues in Studying Marriage
Conceptualizing Cause and Effect
This essay was presented as part of a PAIR Project methodological symposium at the 1996 International Conference on Personal Relationships in Banff, Canada.
Social science research is fundamentally concerned with identifying cause-effect linkages. Causes can be conceived in terms of molecular elements or in terms of broader systems; effects can be observed immediately, or at later times. The study of marriage as an interpersonal system is implicitly anchored in what Cook and Campbell (1979) refer to as an "activity" theory of causation in that the overt actions of the partners, and psychological substrate that underpin such actions, are causally interdependent. Our focus here is on how the researcher conceptualizes "causes" and "effects." In this document, we lay forth several issues for researchers to consider as they attempt to conceptualize causal relationships.
Locating the cause on the molar-molecular continuum
Molar causal conditions are characterized in terms of large and often complex objects (Cook & Campbell, 1979). They are often linked causally to particular effects through smaller units thought of in terms of a finer time scale. Some researchers study causal conditions in terms of these smaller, micromediational units. Thus, for example, a researcher who conceptualizes causal conditions at a relatively molar level might propose that wives married to husbands who are high in negative affectivity (a personality trait) will become increasingly dissatisfied with their marriage. This researcher might argue that husbands who are high in negative affectivity are more likely (than other men) to be irritable and critical, and less likely to be affectionate, and that these behavioral tendencies might, over time, create conflict, ultimately decreasing their wives' satisfaction. It is important to recognize, as Cook and Campbell (1979) point out, that causal assertions are meaningful at the molar level even when the ultimate micromediational processes are not known (p. 32). It is equally important to keep in mind that the identification of micro causal processes is meaningful even when the extent to which they might reflect more molar causal conditions is not known.
Consider whether the proposed cause-effect relationship is psychological, interpersonal, contextual, or some combination of these types of elements
Intrapsychic causal linkages fit within the domain of individual psychology, unless the theory explicitly hypothesizes contextual or interpersonal micromediational processes linking the mental states to one another. A study that examines spouses' personality characteristics (e.g., neuroticism, chronic depression in connection with their own marital satisfaction includes variables on only one side of the interpersonal equation and thus fits within the domain of individual psychology. Similarly, research which examines linkages between personality and behavior exhibited in a particular relationship is psychological in nature, unless contextual or interpersonal factors are brought into the causal equation. Studies are interpersonal in nature if they examine interspousal causal processes (e.g., one spouse's behavior, the other's happiness) or characterize either causes or effects in dyadic terms. Studies that examine compatibility, defined in terms of combinations of spousal attributes fit this category, as do studies that examine influence patterns, conflict, division of labor, or marital stability. Contextual causes or effects are located outside the individual or the relationship. A person may be interested in the impact of a natural disaster on some aspect of marriage (such as conflict or stability). Such a study would identify as the cause something that is contextual, with its effect being interpersonal.
Direction of causality
Longitudinal studies have the advantage over concurrent studies in that they have the potential of allowing researchers to study processes over time and, thus, of identifying the onset of causal conditions and their consequences on events. The stability of one condition, coupled with a change in the other, provides evidence suggesting the first condition affects the second.
Temporal Shape of Causal Conditions
Causal conditions can take any of at least four temporal shapes (Kelly & McGrath, 1988, p. 129), as illustrated in the table and figure below:
Temporal Shape of the Effect
Researchers need to consider how the length of time from X to Y affects the resulting pattern of changes in Y, hence the conclusions that can be drawn about the meaning of the X-Y events (i.e., the causal process) of the study. Kelly and McGrath (1988) identify five temporal shapes of the functional relationship of the causal variable (X) and the outcome variable, Y, over time.
The figure to the right illustrates that the conclusion an investigator draws on the basis of a single subsequent measurement will depend both on: which interval is chosen for comparison; and the shape of the process being examined.
The five functional relationships of the putative cause and the effect are:
(1) A gradually increasing, linear process
In this case, the impact of the cause, X, on the outcome, Y, increases with time. The connection between a spouse's "negative affectivity" (a personality trait) on the partner's marital satisfaction might be hypothesized to become increasingly strong with time. Another example: Parenthood may have an increasingly strong impact on marital satisfaction, as a consequence of a variety of processes that reduce the pleasantness of marital interaction. Note that the likelihood of detecting the effect of the causal variable, X, would increase with time, assuming, of course, that countervailing causal forces do not occur, but the magnitude of the effect will depend on whether the final measurement is taken at O2, O3, or O4. For example, there will be a conclusion of a greater effect if the second measurement is taken at O4, rather than at O2.
(2) An all-at-once change that is then maintained over time (persistent)
In this case, the impact of cause, X, on the outcome, Y, occurs immediately and persists over time. This kind of effect might occur, for example, when a husband finds his partner in bed (and obviously not merely "sleeping") with a friend. Such an event could have an immediate and long term effect on trust. Note that the effect would be detected regardless of when the final measurement is taken.
(3) An all-at-once change that is not maintained over time (not persistent)
In this third case, the impact of the cause, X, on the outcome, Y, is all-at-once, but is not maintained over time. Parenthood might create an immediate, but short-term, effect on the parent's level of stress. Also, after a couple moves to a new community, they may feel particularly lonely; later, as they become integrated into the social fabric of the community, their loneliness may dissipate. Note that the investigator would conclude that X had a large effect if the measurement were taken at O2, a moderate effect if the measurement were taken at O3, and no effect if the measurement were taken at O4. Unless at least two measures were taken after the treatment, the investigator would not know that the effect deteriorated with time.
(4) A delayed effect
Sometimes the effect of a cause, X, on the effect, Y, might be delayed. Consequently, these effects are sometimes called "sleeper effects," or "viral effects." For instance, the impact of spouses' contrary ideas concerning the appropriateness of various ways of disciplining children on marital satisfaction may be slow to surface. Also, the impact of a husband's level of narcissism on how much he takes his wife's desires into account in dealing with conflict-of-interest situations may not surface when the couples are newlyweds, but become manifest later in marriage. Note however that even if a measurement was taken at O4, the investigator would still not know that the effect was delayed unless at least two measurements were made after the treatment - either O2 and O4, or O3 and O4.
(5) A cyclical effect
A cyclical effect occurs when a cause, X, creates changing effects on Y over time. For instance, in a marriage in which the husband is an avid hunter and the wife is uncomfortable with the idea of stalking and killing animals, the effects of their differences may be cyclical. Such a couple might experience a seasonal downturn in marital satisfaction as a result of the conflict over his going on hunting trips. Another example of a cyclical effect might be reflected in the marital satisfaction of couples in a "commuter marriage." Note that with a cyclical process, a single measurement at O2, when the cycle is at its peak, would indicate that the treatment exerted a large, positive effect on the process being measured. Measurement at O4, however, when the cycle is at its ebb, would indicate that the treatment exerted a large, negative effect on the process. Further, measurement at O3, when the cycle is midway between its peak and its ebb, would indicate that the treatment had no effect. Indeed, the investigator would probably not be aware of the cyclic nature of the process being observed unless all three measurements, O2, O3, and O4, were made, plus one additional one.
Clearly, it is important to specify, in both theoretical formulation and empirical practices, both the time interval between the cause and the effect necessary for the causal processes to unfold, and the temporal shape, or time-course, of the process under observation. Furthermore, a single measurement of the effect is usually insufficient to describe these causal processes.
Temporal Hierarchy of Effects
If we assume that behavioral change in humans is a process that takes time, then a single measurement of the behavior is inadequate to describe that process. And the interpretations arrived at on the basis of that single measurement will be a function of the point in time chosen (usually arbitrarily) to measure the effect.
There is a hierarchy of levels of information about potential time related effects, illustrated in the figure to the right, and these are tied to different degrees of temporal fine-grainedness of observations. To explore relations of any shape more complex than the linear case - or even to test the assumption of a linear relation - requires that we observe Y at multiple times. At least three waves of observations are needed to assess linearity; four or more observation waves are needed to explore cycles.
Temporal level 1: Static designs
Measurement of some behavior on a single occasion. Such a data set potentially contains useful descriptive information about that occasion, but it contains no time related information at all.
Temporal level 2: Change in level designs
Measurement of some behavior on two occasions, which can tell us about the change in effect from the first to second occasion (more, less, or stable).
Temporal level 3: Monotonic change designs
Measurement of some behavior from three points in time. Designs at temporal Level 3 can show five different patterns of outcome:
Temporal level 4: Nonmonotonic change designs
Measurement from four points in time which can yield six different patterns of means. Five of the patterns (those just described above) permit us either to have greater confidence in, or to disconfirm (depending on the lie of the final point), the kinds of patterns already noted for Level 3. The sixth is a potential pattern that can be seen only at this level and above:
Temporal level 5: Cycle detection designs
It is only at temporal level five that one can detect a cycle with some confidence, and then only if the cycle is no longer than two observation intervals. The more fine grained the set of observation intervals and data analysis intervals, the more readily cycles can be detected if they are present.
Considerations for Designing Longitudinal Research
The variety of ways a proposed cause, X, may affect an outcome variable, Y, makes it important to carefully scrutinize longitudinal research to make sure that the duration of the study and the interval between occasions of measurement allow the temporal shape of both the putative causal condition and its effect to be revealed.
Unfortunately, very few studies include more than two times of measurement. Furthermore, the duration of the study and the intervals between occasions of measurement are often determined as they were in the PAIR Project by practical matters, or matters of convenience to the investigator, rather than on a conceptualization of the causal processes involved. More often than not, the duration of the interval is left unspecified both in our theoretical formulations and in our interpretations of concrete findings. The general failure of researchers to attend to these matters makes it likely that the durations and intervals used in some studies will be inappropriate, threatening the internal validity of the researcher's conclusions about cause and effect. Experts in longitudinal research have recently urged researchers to think through the causal processes, to use multiple occasions of measurement, and to give careful considerations to the length of time necessary for the causal process to operate (Kelly & McGrath, 1988; Nesselroade, 1991). A strong theoretical basis for predicting long-term effects, one that can be disproved or supported by empirical evidence, helps confirm the validity of such effects.
Bollen, K., & Lennox, R. (1991). Conventional wisdom on measurement: A structural equation perspective. Psychological Bulletin, 110(2), 305-314.
Cook, T. D., & Campbell, D. T. (1979). Quasi-experimentation: Design and analysis issues for field settings. Chicago: Rand McNally.
Huston, T. L. (1994, February). Temporality and causation: A template for analyzing longitudinal research on marriage. Unpublished manuscript.
Kelly, J. R., & McGrath, J. E. (1988). On time and method. Newbury Park, CA: Sage.
Nesselroade, J. R. (1991). Interindividual differences in intraindividual change. In J. L. Horn & L. M. Collins, (Eds.), Best methods for the analysis of change (pp. 92-195). Washington, D. C.: American Psychological Association.
Appendix: On Causes and Effects (From Cook & Campbell, 1979)
1. Causal assertions are meaningful at the molar level even when the ultimate micromediation is not known. Thus, for example, it is useful to know that incompatibility causes marital dissatisfaction even when the micromediational processes are not known.
2. Molar causal laws, because they are contingent on many other conditions and causal laws, are fallible and hence probabilistic. Many causal explanations are molar and contingently causal rather than ultimately micromediational and inevitable. Where any of a number of contingencies are not operating as assumed, a cause may not operate as it otherwise would. Given the difficulties in conceptualizing and testing all the relevant contingent conditions, many genuine effects will appear to occur sporadically. The evidence supporting molar causal laws will usually be probabilistic; it is probably the case that the more molar the causal assertion and the longer and more unspecified the assumed micromediational chain, the more fallible the causal law and the more probabilistic is the supporting evidence.
3. The effects in molar causal laws can be the result of multiple causes. These effects may be additive and/or interactive.
4. While it is easiest for molar causal laws to be detected in closed systems with controlled conditions, field research involves mostly open systems. The more open the system, the more apt there are to be multiple and contingent causes, and hence the more fallible will be causal inferences.
5. Dependable intermediate mediational units are involved in most strong molar laws.
6. Effects follow causes in time, even though they may be instantaneous at the level of ultimate micromediation. The aim is to discover points in the causal chain where changes in one variable lead to changes in another. Delayed causal connections imply real storage processes. Such storage processes lead us to accept the idea of delayed causation while not ruling out that within some logical framework, ultimate causes should have instantaneous effects.
7. Some causal laws can be reversed, with cause and effect interchangeable.
8. The paradigmatic assertion in causal relationships is that the manipulation of a cause will result in the manipulation of an effect.
The PAIR Project at the University of Texas at Austin
Principal Investigator, Ted L. Huston
Page last modified: 2 August 2002