Model Specification

Model Uncertainty and Sensitivity Analysis

Model uncertainty has multiple origins; it occurs when researchers do not know the correct set of right-hand-side variables, the correct functional form that relates the dependent to the independent variables, the correct structure of conditionality between exogenous variables (those whose values are independent of the states of other variables in the system under study), and/or the causal dependences between exogenous variables, to mention just a few model uncertainties. Logically, researchers need to start with identifying all exogenous factors that influence the variation of the dependent variable across observations. Both the inclusion of too few or too many regressors may cause bias and will render inferences potentially invalid. [Page 1596]Assume, as an example for the “too many” case, that researchers intend to estimate the effect of income on vote choice but include the vote intention into the model. Since vote intentions depend on structural factors such as income and education, the inclusion of the vote intention variable causes right-hand-side endogeneity (i.e., regressors may not be independent of the dependent variable or the error term; see below). If this endogeneity is not correctly modeled, it makes the correct interpretation of results impossible. Indeed, the vote intention variable will capture much of the influence of income and education on vote choice, but researchers cannot separate the two causal mechanisms unless they estimate a simultaneous equation model or unless they drop vote intention from the list of regressors, hoping that income and education fully determine vote intentions.

...