Are the hypothesis tests used in time-on-market studies powerful enough?

Two approaches are widely used when estimating how long it takes to sell a property: one approach uses least squares and the other is based on a hazard model. Despite many years of study, the field does not seem to have reached a consensus on the “typical” effect of important variables. This paper uses numerical simulations and some simple reasoning to study the power of tests to reject hypotheses which are false. The numerical simulations offer three key insights. First, they quantify the effect of a difference in the hypothesis. Small differences are hard to detect while the power to reject a big difference varies with the characteristics of the data. Second, since the simulation can impose a true distribution, it is possible to study the effect of using a knowingly misspecified model with less statistical theory and more focus on practical implications. I show that analysis based on estimating a hazard model seems to be more robust. Third, the simulations are used to explore the conceptual difference between testing the hypothesis that an effect exists vs. tests aimed at measuring the magnitude of an effect. This difference is very important when the different approaches use different parameters which affect the estimates without being studied directly. I discuss the implications for practitioners and for researchers.