We can observe that the idea is to elaborate a meaning analysis that creates theoretical constructs rather than measurable variables. Typically, these constructs represent an explanation of the phenomenon analyzed via a diagram that shows the causal relationships between them. Because constructs are abstractions, researchers need to establish connections between these and reality. The link between constructs and observable measurements is called operationalization. Once the operationalization is done, measuring the variables of interest is possible to generate the data needed to evaluate the research hypothesis in a statistical model. In summary, explanatory modeling refers to “applying statistical models to data for testing causal hypotheses about theoretical constructs.”

Predictive modeling is defined as “the process of applying a statistical model or data mining algorithm to data to predict new or future observations.” The idea is to predict an output value given the input values. This is clear for time series analysis when we try to forecast values in the future \( (t+k) \) based on previous observations.

The concept of prediction in social sciences has been historically ignored because scientists’ efforts have concentrated on finding causal explanations for social phenomena. But recently, scholars like Hofman, Sharma, and Watts (2017) have argued that the new tools provided by the current computational nature of social science should help consider a more extended use of predictions in the field. For the proper incorporation of prediction, the authors highlight three critical problems that must be addressed. First, it is necessary to define standard measures for prediction. These measures will improve the comparability of results by establishing consensus on the problems to be solved. With a set of specific problems defined, a “common task framework” should help evaluate claims about prediction. In the absence of common tasks and data, researchers must clearly differentiate between exploratory and confirmatory analysis. Second, theoretical limits to prediction must be better characterized; this way, it is possible to set expectations of what can be predicted or explained. It has been shown that there are events intrinsically impossible to predict in a meaningful sense, meaning that the predictive accuracy is subject to a limit, and those limits need to be comprehended. Third, predictive accuracy and interpretability must be recognized as a complement to evaluate explanations. This is based on noticing that simple models do not necessarily generalize better than complex models, there is increasing evidence that the trade-off between accuracy and interpretability may be less severe than we thought, and the concept of understanding can be used to refer to the feeling of having made sense of something (interpret) or having accounted for observing regularities (predict).

Descriptive modeling is the third type of modeling. Its objective is to summarize and represent the data compactly. The difference with explanatory modeling is that the underlying causal implications are absent or incorporated less formally in these models. On the other hand, compared with predictive modeling, descriptive modeling does not aim to generate a prediction of future values. A typical example of this situation would be when we stumble upon scientific papers whose main objective is to fit the best regression model to capture the association between dependent and independent variables rather than produce causal inferences or predictions. I want to add that even though there is a group of well-defined statistical techniques to make causal inferences, we can find an important number of social science papers that, without claiming causation, show association based on theoretical grounds (explanation). The common phrase “correlation is not causation” clarifies that path, regression, or structural equation models (SEM) do not necessarily allow us to compute the causal effect between variables in the statistical sense. Still, they do present an explanatory association within the model. No standard rule exists to determine whether a social science paper develops an explanatory or descriptive model. Still, in general, the presence of path analysis or SEM diagrams indicates the existence of relational (sometimes causal) implications derived from theoretical reasoning.

As I mentioned at the beginning of this post, I will explain a set of methods to serve as an introduction for researchers interested in learning analytical tools that can be used to develop explanations and predictions in social sciences. The methods I collected originated in different areas, such as causal inference, time series analysis, and information theory. In Table 1, I present a summary of the essential elements of each method.