This is so refreshing. The Journal of Basic and Applied Social Psychology has banned p-values.
And confidence intervals.
Their reasons are extremely good reasons, and well-articulated.
I forgot to stress this point explicitly in the recent essay on the Lewandowsky fraud, but descriptive statistics always trump inferential.
A p-value has no inherent substantive meaning, nor does the underlying statistic (this is especially true of a linear correlation on scale items where the participants' actual placement on the items is undisclosed, as it was in the LOG12 paper, a situation that was apparently fully satisfying to Eric Eich, Psychological Science and APS.)
The way we use inferential statistics is often barbaric, though rarely fraudulent. This is a great first step. I think Trafimow and Marks' explanation will be informative to a lot of readers and scientists. I think a lot of us have the wrong idea of what a confidence interval represents.
This will be interesting long-term for its impact on our use of terms like "effect". A low p-value doesn't mean there's an effect. An inferential statistic doesn't mean there's an effect. Other data, including descriptive data, is needed to know if we have an effect. We're especially vulnerable to false or overstated effects when we use student samples because such samples artificially reduce noise (student samples are extremely homogeneous on various dimensions, dimensions that will contribute noise or unexplained variance in community samples.)
Leave a Reply.
José L. Duarte
Social Psychology, Scientific Validity, and Research Methods.