Continuous Predictors of Pretest-Posttest Change: Highlighting the Impact of the Regression Artifact
| Autor: | Chantal A. Arpin-Cribbie, Robert A. Cribbie, Linda Farmus |
|---|---|
| Jazyk: | angličtina |
| Rok vydání: | 2019 |
| Předmět: |
Statistics and Probability
education Artifact (software development) 01 natural sciences 010305 fluids & plasmas pre-post pre-post analysis 0103 physical sciences Covariate 010306 general physics Baseline (configuration management) Lord's paradox Applied Mathematics Psychological research lcsh:T57-57.97 Regression analysis Outcome (probability) Regression continuous predictors lcsh:Applied mathematics. Quantitative methods change analysis sense organs regression artifacts lcsh:Probabilities. Mathematical statistics Psychology lcsh:QA273-280 Clinical psychology Type I and type II errors |
| Zdroj: | Frontiers in Applied Mathematics and Statistics, Vol 4 (2019) |
| ISSN: | 2297-4687 |
| DOI: | 10.3389/fams.2018.00064/full |
| Popis: | Researchers are often interested in exploring predictors of change, and commonly use a regression based model or a gain score analysis to compare degree of change across groups. Methodologists have cautioned against the use of the regression based model when there are non-random group differences at baseline because this model inappropriately corrects for baseline differences. Less research has addressed the issues that arise when exploring continuous predictors of change (e.g., a regression model with posttest as the outcome and pretest as a covariate). If continuous predictors of change correlate with pretest scores, the modeled relationship between predictors and change may be an artifact. This two-part study explored the statistical artifact, or overestimation of effect estimates, that may arise when continuous predictors of change are included in pretest-posttest regression based models. Study 1 revealed that the problematic regression based model that covaries out pretest scores is currently being applied in psychology literature more often than models that do not covary pretest scores, and that the conditions leading to the artifact (i.e., spurious effects) were met in a significant amount of studies reviewed. Study 2 demonstrated that the artifact arises in conditions common within psychological research, and directly impacts Type I error rates. Recommendations are provided regarding which regression based models are appropriate when pretest scores are correlated with continuous predictors. |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|