presenting Quantifying the Robustness of Inferences on April 6 (1pm-2:20pm) and April 11 (1pm-2:20pm) , new paper re: Oster's coefficient of proportionality
Ken Frank
I hope you are doing well.
On April 6 (1pm-2:20pm) and April 11 (1pm-2:20pm) I will be presenting on Quantifying the Robustness of Inferences on my zoom https://msu.zoom.us/j/783760435.
This is in the context of my course on use of multiple regression. So I open it up but priority goes to the students.
Note we also have a new paper. It’s still a little rough, but thought you might be interested.
Exact Calculation of Coefficient of Proportionality including Evaluation of Oster’s δ*, Corresponding Bounds, and Alternatives
Kenneth A. Frank
With
Shimeng Dai
Nicole Jess
Hung-Chang Lin
Qinyun Lin
Yuqing Liu
Sarah Maestrales
Ellen Searle
Jordan Tait
Abstract
Sensitivity analyses characterizing the hypothetical unobserved conditions that can alter statistical inferences are increasingly being applied in the social and health sciences. One of the most ascendant techniques is Oster’s (2019) coefficient of proportionality, which intuitively appeals to how strong selection on unobservables must be compared to selection on observables to reduce an estimated coefficient below a specified threshold. In this paper, we derive an exact expression for the coefficient of proportionality that satisfies Oster’s constraints by reducing the estimated effect to a specified threshold and a specified coefficient of determination (R2). The derivation is exact instead of an approximation, requires fewer assumptions than Oster’s, and the expression is more parsimonious informing theoretical understanding of the factors contributing to robust inferences. From a practical standpoint, the quantities required for the exact expression are conventionally reported in most published studies. Simulations demonstrate the improved performance of the exact expression relative to Oster’s approximation. Specifically, Oster’s coefficient can overstate the robustness of an inference, especially for strong designs in which the observed covariates account for a large portion of an unconditional estimated effect. An application shows that Oster’s expression overstates the robustness of the inference of an effect of low birth weight and preterm birth on IQ by an order of magnitude.
Ken