作者: Aris Spanos
DOI:
关键词:
摘要: Statistical model specification and validation raise crucial foundational problems whose pertinent resolution holds the key to learning from data by securing reliability of frequentist inference. The paper questions judiciousness several current practices, including theory-driven approach, Akaike-type selection procedures, arguing that they often lead unreliable inferences. This is primarily due fact goodness-of-fit/prediction measures other substantive pragmatic criteria are questionable value when estimated statistically misspecified. Foisting one’s favorite on yields models which both substantively misspecified, but one has no way delineate between two sources error apportion blame. argues statistical approach can address this Duhemian ambiguity distinguishing premises viewing empirical modeling in a piecemeal with view various issues more effectively. It also argued Hendry’s general specific procedures does much better job than primary because its underpinnings.