Stalin and Summary Statistics

A statistical-literacy essay (follow-up to 'Beware Summary Statistics'). Takes two real studies spun in opposite directions — Noah Smith downplaying IQ→state-GDP (14% variance, 'Something Else matters') and the Atlantic hyping a Swedish adoption study as proof the 'birth lottery' rules wealth — and shows the same finding looks boring as % variance, suggestive as a correlation (r), and overwhelming as a scatter-plot-with-line. Drives home variance = r², notes almost nothing in social science exceeds r≈0.4, and uses the coin-flip explanation to reconcile the 5%-vs-23% intuitions. The 'Stalin' hook: the people who report a study (by choosing which representation to show) decide everything, and can hand you negative knowledge.

Quality 72 · Strong. Strong, upper. A lucid, memorable, genuinely useful public explanation of correlation vs variance-explained vs the raw scatter, plus the sharp meta-point that framing choice hijacks intuitions. Held below Excellent because the core statistics are textbook (variance = r²) and the piece partly reprises his own earlier essay; the value is the clear exposition + the media-framing angle, not a new idea.

Claude’s paradigm shift 50 · Moderate. Moderate (50). The packaging — same data, three presentations, reporters as the real deciders — was a fresh, non-obvious public-facing synthesis at publication, but builds on well-known statistics and his own prior post.

Real-world impact 2 · Minor. Minor (2). Improves statistical literacy among readers and the 'r vs variance / r≈0.4 ceiling' points get cited within the discourse, but no material-world effect.