Подтвердите e-mail

Для публикаций, комментариев, реакций и сообщений подтвердите адрес.

←

Публикация

Dan Quintana

Selecting an effect size for power analysis is hard. Many researchers fall back on Cohen's thresholds, but they have no empirical basis and vary wildly by field. Our new paper offers a better option: field-specific effect size distributions built from meta-analytic data doi.org/10.3758/s134...

Обсуждение

3 прямых ответов · 8 сообщений

Stefano Coretta

Congrats! 🙌 Looks interesting! Now I want to see how people will manage to misuse this… 🤣 I guess it’s inevitable (see how some people are still doing the Null Ritual, but Bayesian).

Dan Quintana

Cohen never intended d = 0.2/0.5/0.8 to be applied across *every* field (in fact, he later regretted suggesting these thresholds). He proposed them as a last resort when no prior estimates were available. But they've become the default for interpreting results and powering new studies anyway

Ответ для Dan Quintana
Dan Quintana

The problem with this approach is that effect size distributions vary a lot between fields. A "medium" effect in social psych isn't the same as a "medium" effect in heart rate variability research, for instance. One-size-fits-all benchmarks can leave studies underpowered or overstate results.

Ответ для Dan Quintana
Dan Quintana

The 'ESDist' R package takes data you can easily pull from existing meta-analyses (effect size + SE) and gives you field-specific small/medium/large benchmarks, distribution plots, ranges of detectable effects for sensitivity power analyses, and 95% CIs around the benchmarks

Ответ для Dan Quintana
Dan Quintana

It also handles publication bias. Since published effects tend to be inflated, raw ESDs can mislead. ESDist can produce adjusted benchmarks alongside unadjusted ones, including an "iceberg plot" that shows both distributions together.

Ответ для Dan Quintana
Dan Quintana

We see two main use cases: justifying the effect size parameter in power analysis, and interpreting results in context. Instead of "we found a medium effect," with the package you can say "our effect was larger than 70% of effects in this literature." This is more precise than using benchmarks.

Ответ для Dan Quintana
Dan Quintana

Find the package, tutorial code, and example data here: github.com/berntgl/ESDist A Jamovi point-and-click ESDist module is also on the way. Congrats to Bernt Glaser, who led this work, and the rest of the team on this paper!

Roger Giner-Sorolla

👀