ZELNER, Bennet A.
Using simulations to interpret results from logit, probit, and other nonlinear models - Bognor Regis : Wiley-Blackwell, December 2009
In a recent issue of this journal, Glenn Hoetker proposes that researchers improve the interpretation and presentation of logit and probit results by reporting the marginal effects of key independent variables at theoretically interesting or empirically relevant values of the other independent variables in the model, and also by presenting results graphically (Hoetker, 2007: 335, 337). In this research note, I suggest an alternative approach for achieving this objective: reporting differences in predicted probabilities associated with discrete changes in key independent variable values. This intuitive approach to interpretation is especially useful when the theoretically interesting or empirically relevant changes in independent variables values are not very small, and also for models that contain interaction terms (or higher-order terms such as quadratics). Although the graphical presentations recommended by Hoetker implicitly embody this approach, they typically fail to include appropriate measures of statistical significance, and may therefore lead to erroneous conclusions. In order to calculate such measures, I recommend and demonstrate an intuitive simulation-based approach to statistical interpretation, developed by King et al. (2000), that has gained widespread adherence in the field of political science. Throughout the article, I provide a running example based on research that has previously appeared in the Strategic Management Journal.
Using simulations to interpret results from logit, probit, and other nonlinear models - Bognor Regis : Wiley-Blackwell, December 2009
In a recent issue of this journal, Glenn Hoetker proposes that researchers improve the interpretation and presentation of logit and probit results by reporting the marginal effects of key independent variables at theoretically interesting or empirically relevant values of the other independent variables in the model, and also by presenting results graphically (Hoetker, 2007: 335, 337). In this research note, I suggest an alternative approach for achieving this objective: reporting differences in predicted probabilities associated with discrete changes in key independent variable values. This intuitive approach to interpretation is especially useful when the theoretically interesting or empirically relevant changes in independent variables values are not very small, and also for models that contain interaction terms (or higher-order terms such as quadratics). Although the graphical presentations recommended by Hoetker implicitly embody this approach, they typically fail to include appropriate measures of statistical significance, and may therefore lead to erroneous conclusions. In order to calculate such measures, I recommend and demonstrate an intuitive simulation-based approach to statistical interpretation, developed by King et al. (2000), that has gained widespread adherence in the field of political science. Throughout the article, I provide a running example based on research that has previously appeared in the Strategic Management Journal.
