Current Issue

This Article From Issue

March-April 2013

Volume 101, Number 2
Page 85

DOI: 10.1511/2013.101.85

To the Editors:

In Megan Higgs’s Macroscope Do We Really Need the S-word?” in the January–February issue, I heartily endorse her critique of the term significance, which is misused all too frequently in the social sciences. Hypothesis testing and statistical significance are very useful in cases where data collection is expensive and random sampling is used to collect data. However, nowadays we often have oceans of data to wade through. Some years ago, I reviewed a report on recidivism for a federal agency. The authors had a large body of data, tens of thousands of cases, and found that every comparison they made was statistically significant. So they decided to take a 10-percent sample of the data, and found that many of the comparisons they made were no longer significant. Consider: to get “significant” findings they threw away 90 percent of the data!

My colleague Alfred Blumstein at Carnegie Mellon University has suggested “discernible” as a replacement for “significant,” because it indicates that two numbers are far enough apart that they are unlikely to have arisen from the same distribution.

Michael D. Maltz
Criminal Justice and Information & Decision Sciences, Emeritus
University of Illinois at Chicago
Evanston, IL

To the Editors:

Megan Higgs’s argument to stop using the term significant for reporting results of statistical tests is well received. To add to the argument, after attaining a significant result with p-value of 0.05, the probability that an exact replication of equal size from the same population will yield a significant result is only about .50 (see page 104 in my book Empirical Direction in Design and Analysis for more information). Furthermore, the meaning of results depends entirely on extra-statistical, empirical inference. I completely concur with Higgs’s emphasis on explicating the scientific meaning of the results.

But this argument against “significant” may not have much effect upon the mass of common usage. In my book mentioned above, I suggested the term statsig to avoid common language meanings of “significant.”

Norman Henry Anderson
Distinguished Professor Emeritus
University of California, San Diego