Researchers often use partial observability models to separate effects of a single variable across multiple outcomes. This paper argues these models are unreliable when the explanatory variable affects all outcome variables.
In their analysis, authors show that small errors in model assumptions—specifically functional form specifications—even large-sample bias. They demonstrate this using Monte Carlo simulations, revealing substantial estimation problems under partial observability conditions.
Contrastingly, they find that identical misspecifications produce minimal to no bias when outcomes are fully observable.
| Unreliable Inferences About Unobserved Processes: A Critique of Partial Observability Models was authored by Carlisle Rainey and Robert A. Jackson. It was published by Cambridge in PSR&M in 2018. |
