Hi everyone, I ran a moderated moderation analysis using PROCESS model 3 in R with the variables X (predictor), Y (outcome), W (moderator 1), and Z (DV). I found significant interaction terms between X and W. However, when I used model 1 to explore the interaction terms between X and W, the results were not significant. Is there another model that could help me support my hypothesis? Also, why did the same predictor and moderator yield different results in model 3 compared to model 1?
Are you now selecting a model that best 'proves your hypothesis'? I understand that this what you are taught to do. I am however worried that you are now data-dredging and this is acceptable practice (https://en.wikipedia.org/wiki/Forking_paths_problem or The Garden of Forking Paths: Why Multiple Comparisons Can Be a Problem, Even When There Is No ‘Fishing Expedition’ or ‘p-Hacking’ and the Research Hypothesis Was Posited Ahead of Time). Using such methods all hypothesis will be 'proven', but no argument that is placed will be internally valid. As such all 'scientific' articles 'proof' their hypothesis. However, nothing of what is written in the before sentence is semantically correct. Proofs are something for the deductive sciences and frequentsist methods cannot support hypothesis (https://en.m.wikipedia.org/wiki/Misuse_of_p-values). This might come over as harsh, and it is, but somehow these issues are written about in the tons, but nobody really dares to say something. If you say something it might harm my your career or you seem to be nitpicking or overly critical. But now, none of our results are practically meaningful, because we only select what we want to see. Then to stay in 'science' you have to perform the same because thats how you get 'positive' results.
- Badges
- Science topic
- Similar topics
- Statistics
- Survey