This paper studies the effectiveness of the Queensland School Transport Safety Program (SafeST). The authors main general conclusion is that good statistical practice can improve the accuracy of forecasts of the effects of a countermeasure by an order of magnitude. Conversely, poor practice, in particular an inappropriate model specification coupled with a small data sample can produce imprecision. In particular, the authors suggest that unobserved demographics are generally best modelled by stochastic trends, as these allow for the demographic effect to wander away from a linear trend if this improves the fit. The authors show how the model can be estimated by GLS in these circumstances and demonstrate that efficiency gains can result. (Author/publisher) For the covering entry of this conference, please see ITRD abstract no. E210413.
Abstract