TCRM Paper Session 8 - Using Complex Statistics Correctly
About the Session
Discussant: Jeremy Yorgason
Presider: To be Announced
106-01: A Novel Use of Zero-Inflated Models: Combining Events and Intensity
Isaac Washburn, Clint Broadbent
106-02: “Fishy” Results? Six Formal and Informal Ways to “Catch” Them
Walter Schumm, Duane Crawford, Lorenza Lockett (presenter), Abdullah Alrashed, Asma M. Bin Ateeq
106-03: (Presented by Lorenza Lockett) P-Hacking? How About P-Slashing?
Walter Schumm, Duane Crawford, Lorenza Lockett (presenter)
Abstract(s)
106-02: “Fishy” Results? Six Formal and Informal Ways to “Catch” Them
SUMMARY
Sometimes social scientists fabricate their data; sometimes, the reports contain typographical errors. How can we detect such flaws and tell the difference among them? We review six ways to detect problematic data, using social science articles published since 2000, some of which have been found to have been total fabrications. Some mistakes can easily be corrected and are not much of an issue; however, in some cases, our approach may help detect fraudulent or fabricated data.
106-03: P-Hacking? How About P-Slashing?
SUMMARY
“p-hacking” is a term that labels attempts to find significant results when there may be none. We propose a term “p-slashing” which refers to avoiding findings of statistical significance. Several popular methods for this are described, using recent published articles in major journals. Potential remedies for discovering “p-slashing” are discussed.
106-01: A Novel Use of Zero-Inflated Models: Combining Events and Intensity
Zero-Inflated models are widely utilized to deal with an over-abundance of zeros in intensity data where the data has a natural zero. This paper will show how using Structural Equation Modeling (SEM) versions of Zero-Inflated models frees up the restriction of a natural zero. In particular, we will present an overview of common regression models for zero-inflated data (Poisson and negative binomial) and contrast those with their SEM counterparts (Poisson, negative binomial, and semicontinuous). Lastly, we will provide examples of novel outcomes and how they are handled in these various models using both real examples and simulated data.