This paper re-examines the appropriate procedures to employ when sampling unchosen alternatives when estimating multinomial (aka conditional) logistic regression models. I show that recent advice presented in Sociological Methodology (SM) deviates from the econometric literature. I use data from the Multi-City Study of Urban Inequality to demonstrate that the original econometric advice is sound, and that the recently presented advice in SM can lead to biased parameter estimates, especially when the sampling fraction is small. [Article]
Benjamin Fletcher Jarvis 