This study examines the estimation of linear regression models for length of stay using on-site versus border-point samples. We show that Ordinary Least Squares (OLS) based on on-site samples is inconsistent, whereas OLS estimators from border-point samples are consistent. However, consistent estimates can be recovered from on-site samples by applying Weighted Least Squares. We also demonstrate that the inconsistency of OLS affects survival models that admit an Accelerated Failure-Time (AFT) specification. These findings challenge the validity of empirical studies relying on on-site samples and linear regression or survival models that admit an AFT formulation.
Andreu Sansó (2026) studied this question.