David Weakliem is a sociology professor at the University of Connecticut. (screenshot/University of Connecticut)
The probability value (p-value) is justifiable as a standard for statistical significance, according to David Weakliem, University of Connecticut sociology professor, in his presentation “The Magic Number: What Should be the Standard of Statistical Significance” on Friday.
A p-value measures the probability of obtaining the trial observed if the trial was up to chance, according to the University of Chicago Department of Statistics.
The larger the p-value, the less confidence a statistician has in a trial being statistically significant.
Weakliem, the author of “Hypothesis Testing and Model Selection in the Social Sciences,” began by talking about the history regarding a universal standard for statistical significance.
The standard p-value, Weakliem said, used to be .003, because a smaller p-value results in greater confidence in interpreting data as statistically significant. Now, with a value of .05, critics are pushing to decrease the value to .005 to ensure more statistical procision.
“Does p-value equals .05 represent ‘strong evidence’ against the null hypothesis, or ‘pretty strong evidence’ against the null hypothesis, or maybe just ‘some evidence?’ Is there a point you say there is no evidence against a null hypothesis?” Weakliem said. “That’s why we got the competing standards. Its arbitrary.”
Weakliem brought up a New York Times article titled “Science’s Inference Problem: When Data Doesn’t Mean What We Think It Does.”
The article is about scientific research experiments published in journals that are deemed “statistically significant” and, after being tested again, show evidence against significance, according to the article.
“Critics point towards a standard of statistical significance that is too easy to meet,” Weakliem said.
The p-value is relevant to students working with research and statistics, said Brittney Hernandez, a measurement and evaluation assessment (MEA) graduate student who attended the lecture.
“(We) looked at the core theme of this critique of a p-value of .05 as being a standard for statistical significance. What this talk was honing in on is whether or not this is a valid standard,” Hernandez said.
Weakliem said there is merit to a standard p-value of .05; however, the data observed must be interpreted on a case to case basis.
“(P-value) .05 is kind of regarded as the big one,” Weakliem said. “But, it would be more valid to think of things in terms of a sliding scale, that is, you have a range of pretty strong evidence and range of strong evidence. It’s a matter of degree.”
The most influential idea of Weakliem’s argument, Hernandez said, was the point that researchers are obliged to clearly and thoroughly interpret their contrived data.
“The biggest thing to focus on is the actionable part,” Hernandez said. “I think it’s the responsibility of researchers to clearly communicate these discrepancies so we say that p-values are the standard and if you get that you’re good in terms of significance, yeah it’s great and all, but in order to be clear to communicate what that means, we have to be a little bit more precise in how we are defining. The biggest thing is clarifying between the standard and the fact it’s not the absolute best evidence for every circumstance.”
Lillian Whittaker is a campus correspondent for The Daily Campus. She can be reached via email at lillian.whittaker@uconn.edu.