From the epistemological point of view, criticism of statistical significance here is based on questioning a criterion that is stipulated, i.e., defined arbitrarily.
Doing so gives formalization and modeling a greater importance than advancing understanding. That is unscientific.
A pragmatic approach is more appropriate than a strictly formal one, especially as an institutional norm.
Formal rigor is a necessary condition but not a sufficient one. Don't lose the forest for the trees.
The quest for knowledge is the quest for a consensual world view based on reasoning and evidence. This is based on many inputs and their consilience within the framework. This is an ongoing enterprise and science is not the only contributor to it. But science based on naturalism is a very significant contributor, tethering knowledge to logical pedigree and empirical warrant.
Statistical Modeling, Causal Inference, and Social Science
Alan Sokal’s comments on “Abandon Statistical Significance”
Andrew Gelman | Professor of Statistics and Political Science and Director of the Applied Statistics Center, Columbia University
Also
true economics
Ambiguity Fun: Perceptions of Rationality?Constantin Gurdgiev | chairman of the Ireland-Russia Business Association, contributor and former editor of Business & Finance Magazine, and lecturer in Finance with Trinity College, Dublin
4 comments:
In his Oct 2 blog Gelman notes "p-values should be interpreted as graded measures of the strength of evidence ... dichotomous threshold thinking must give way to non-automated informed judgment."
If Gelman wants to phrase his position that way, then I can agree with him. In other words, don't do away with the statistical significance test, but report the result as a probability rather than "yes or no." Avoid dogmatic thinking. That was the gist of my original comment on the subject.
That is far different from Syll's proposal to eliminate statistical significance testing altogether.
The decision to publish is different than real-life decision making. There is something to be said for sharing all studies, even if the study results seem inconclusive at the time, because the information may be useful to others. Sometimes the experimenter is focusing on the wrong thing, then later the old data is viewed through a new frame that gives it meaning.
In real life, say you are considering a drug for approval, then the stakes may be higher and a more cautious view of the study may be justified.
The overall approach to social science stats is flawed https://papers.ssrn.com/sol3/papers.cfm?abstract_id=1775002 Why Inferential Statistics are Inappropriate for Development Studies and How the Same Data Can be Better Used
Thanks, Clint. I just downloaded the paper and read the beginning, and it looks good from the epistemological perspective. I'll at the whole thing later when I have time.
Post a Comment