Descriptive, Predictive, and Causal Research

A few years ago, I wrote an open letter to the NASSM community. Here is what I wrote:

Dear Sport Management Colleagues,

In this email, I would defend the value of descriptive studies to sport management research, and differentiate explanatory and predictive research.

I would argue even purely descriptive research can be very much needed and useful. It becomes a cliché that good descriptive research provokes why questions and good descriptive research builds the foundation for further causal studies. However, because the favors given to causal studies and theory testing, we often take on a causal study too hastily (I consider SEM as a causal technique). Without a solid description of phenomenon, causal studies, even informed by strong theory, can be on shaky ground. All theories can be wrong, and wrong theories can be supported by the means of poor science (such as problematic data generation process, highly flexible modelling methodology, p-hiking, and over-fitting). For instance, all theories can be potentially supported with SEM when there exist common method variances.

Instead, I would argue sport management field is in great need of good descriptive studies for further growth. I recently ask a question relating the two fundamental pillars of sport management - spectatorship and participation. What is the relationship between them? Whereas many studies examined the causal effect of spectatorship on participation, the covariation of these two behaviors are not fully established. And my work is readily rejected based on the theory that descriptive research has no value. In research methodology class, we teach students that the three conditions of establishing causality are temporal order, covariation of the cause and the consequence, and no alternative explanations. Sadly, many presumably causal studies simply ignore those conditions.

Explanatory research and predictive research are different. Predictive research is relational in nature. Theory is not a necessary condition for prediction. More often than not, prediction does not need theory. My four-year old can predict rain is coming when he sees dark clouds; my two-year old can predict train is coming when he hears whistle; the size of a person's right foot can be well predicted by the size of her left foot; the Pavlovian dog would predict food is coming when the bell rings. Accurate prediction can be made without taking recourse to a theory. Prediction only requires the learning of covariation. Prediction does not require an understanding of what cause the covariation. Yet, prediction is not of less value because it does not have a theoretical framework. Rather, prediction is of central interest to many sports management practitioners. Team managers want to predict the potential of a player based on her past performance; merchandise managers want to predict what products would be in vogue; sport entrepreneurs want to predict when their new products would take off; racing pundits and tipsters want to predict the winner of a race. Yet all those scholars could care less.

Will theory facilitate accurate prediction? First, good theory facilitates the learning of covariation as theory can direct observations to the relevant variables; second, theory allows reliable (defined as consistent, not necessarily accurate) prediction to minimize the impact of haphazard or wishful thinking. Third, a good theory is inherently very powerful in prediction.

Whereas, prediction is an attribute of a good theory, not all theories are predicative. Theories involving all unobservable are poor in prediction. Sports identification is not a good predictor of sport consumption because identification is unobservable. And if identification is measured based on consumption, then it becomes tautological to claim that identification predicts consumption. A theory that can explain everything is poor in prediction. A good theory must have its boundary conditions to allow accurate predictions. In theory testing studies, unfortunately, those boundary conditions often got neglected, and a theory was proven correct when it was not supposed to be correct. A great example of such is the theory of reasoned action. This theory is only supposed to explain a certain type of behaviors, namely reasoned action. However, we found that this theory has been used to predict all different kinds of behaviors, reasoned or not.

Theoretical testing becomes a trivial scholarly endeavor because of the ignorance of the boundary conditions, compounded with extremely powerful statistical methods. Alas, now all is good, all theories are supported! It is a pity, based on my own experience, that the reviewers often favor powerful statistical methods. My own work has often been criticized for using conservative statistical techniques. I would sometime use multivariate regression instead of SEM; I would use a simple correlation analysis instead of other more complicated zero-inflated models. If a study can find a significant relationship by using a more conservative method, the relationship must be detected by a more powerful technique.

In summary, I argued:

1. Descriptive study is very useful and needed to sport management field;

2. Predictive study is relational in nature which does not require theory;

3. Good theory can help accurate prediction;

4. Theory testing is trivial when boundary conditions are neglected and extremely powerful techniques are used.

5. Conservative technique should not be disregarded based on the belief/perception that they are too simple to be scientific.