At the top U.S. universities in psychology, there are 17 Democrats for every 1 Republican.

Some argue that this kind of political imbalance is not a cause for concern.

Others believe that it poses an existential threat to the field.

🧵 A dialectical thread 🧵 𝗧𝗛𝗘𝗦𝗜𝗦

Two studies suggest that the political orientation of researchers or their studies has little effect on research outcomes.

@dom_muller @mjbsp @FGabarrot @cedricbatailler I get your point & that of @AntalHaans/@seriousstats. FYI, we intend to adjust the wording of the piece to be more precise & add a subsection on interaction ES. That being said, I don't think there are any mathematical flaws in the preprint or any confound in the sims. 1/5 @dom_muller @mjbsp @FGabarrot @cedricbatailler @AntalHaans @seriousstats First, I agree that:
1) power calculation is the same for main effects & interactions (eq. 1) and
2) ES calculation is essentially the same for main effects & interactions (eqs. 2 & 3, respectively).
(note that the overall interaction ES is displayed by the web app) 2/5

Power analysis for #interactions can be tough!

📢 Our new preprint features:
𝟭 An intuitive taxonomy of 12 types of interaction
...with the 𝘕s to reach power = .80/.90
𝟮 A 😭 meta-study
𝟯 Simulations testing 3 ways to ↗️ power
𝟰 A cool web app!

🧵

osf.io/xhe3u/ 𝟭𝗮 As we know from popular blogs/papers, power analyses differ b/w main effects & interactions because:

👉a main effect corresponds to a difference b/w means

👉a two-way interaction corresponds to a difference b/w mean subdifferences

(using simple b/w-Ss designs as examples)

The Spirit Level has been cited ≈10K times (≈700 times in 2018).

The book is straightforward: It uses cross-sectional data to show negative effects of #IncomeInequality on health.

The problem: It does NOT hold up to scrutiny.

🍒 Thread 🍒 #1 🍒-picking.

In the Spirit Level, some countries are excluded from the analysis without justification. When including these countries and using the latest estimates available, the core findings of the book disappear. [2/5]