Skip to main content

Cheat Sheet: Three-Step EITM Framework

Step 1: Identify a theoretical concept of human behavior of your interest and relate it to a statistical concept.

  • Given that human beings are the agents of action, theoretical concept should reflect overarching social and behavioral processes.  Examples include (but are not limited to): decision making; bargaining; expectations; learning; and social interaction.  (i.e., what is the nature of behavior in which you are interested?)

  • It is also important to relate your theoretical concept to a statistical concept that can allow you to test your theory empirically - make sure that whatever you measure properly reflects your theory.  Examples of statistical concepts include (but are not limited to): persistence; measurement error; nominal choice; and simultaneity.  (i.e., what is the statistical characteristic of behavior in which you are interested?)

Step 2. Develop behavioral (formal) and statistical analogues [see Granato et al 2010].

  • We use "analogues" to express theoretical and statistical concepts.  An analogue represents a concept by variable - and measurable - quantities.
    Examples of theoretical analogues for concept such as decision making, expectations, and learning include (but are not limited to): utility maximization; conditional expectations; and adaptive and Bayesian learning. (i.e., how do you use the formal model to show the nature of behavior in which you are interested?)

  • Examples of statistical analogues for the statistical concepts of persistence, measurement error, nominal choice, and simultaneity include (respectively): autoregressive estimation; error-in-variables regression; discrete choice modeling; and multi-stage estimation (e.g., two-stage least squares).  (i.e., how do you use the statistical model to show the statistical characteristic of behavior in which you are interested?)

Step 3. Unite your theoretical and statistical analogues in testable theory.

  • Here, we are trying to relate the formal and applied statistical analogues for purposes of creating linkages between model and test. Your behavioral theory (often expressed in one or more equations) should clearly show what the specific parameters of interest are, and how you have arrived at them. This clarity in building your model is crucial because your subsequent empirical test can then clearly show whether your theory is supported by evidence or not; if it is not, you can retrace your steps and correct your theory. (i.e., what are your hypotheses? And how do you test your hypotheses?)

Example: Riker, William H., and Peter C. Ordeshook.  1968.  "A Theory of the Calculus of Voting." American Political Science Review 62(1): 25-42.

  • Research question: When do voters turn out to vote?

  • Contribution: Unlike previous authors, Riker and Ordeshook argue that reward of voting must be calculated in such a way that it is positive for voters and zero or negative for non-voters. The authors relax the assumption that the gain from the success of one’s favored candidate is always positive and that the cost of voting is always negative. Those who turn out do not necessarily have the same PDF as those who do not turn out; rather, it is a function of certainty.

  • Theory fitted into the three-step EITM framework:

    • EITM Step 1: Intuition: voters participate in voting to maximize their utility

      • Theoretical concept: Decision theory

      • Statistical concept: Nominal choice

    • EITM Step 2:

      • Behavioral analogue: Utility maximization

      • Statistical analogue: Discrete choice modeling (participate/not participate)

    • EITM Step 3: Based on the assumptions and the previous steps, the authors offer a theory about how voters behave in particular situations – particularly, how they calculate P (probability of casting a deciding vote) and B (gain from success of one’s favored candidate). From the theory they derive three testable hypotheses (on p. 35).

  • Note on the test of the theory: the theory is a micro-level theory based on an individual behavior, but due to the lack of information on the micro-level, the subsequent test of this theory is conducted on the aggregate level. Thus, their theory was ahead of data, which is not unusual in theoretical modeling.