Selection Bias Is A Fact Of Life, Not An Excuse For Rejecting Internet Surveys

Pushes back on the reflexive 'selection bias!' dismissal of amateur online surveys (Aella's, his own). The key distinction: selection bias is FATAL for polls/censuses (estimating how common something is -- any non-representative sample breaks the prevalence estimate) but only SOMETIMES a problem for correlations, which often generalize out-of-sample if the underlying mechanism does. He notes 'real' studies are also selected (Psych 101 undergrads, $10-flyer respondents, non-scientist-murdering tribes), so the right response to a correlation isn't reflexive dismissal but thinking hard about the mechanism and which out-of-sample populations it should or shouldn't generalize to (the banana-IQ and income-vs-obesity-across-scales examples).

Quality 70 · Strong. Strong. A concise, genuinely clarifying methodological correction (the poll-vs-correlation asymmetry in how much selection bias matters) that fixes a common error. Bounded and short, so mid-Strong.

Claude’s paradigm shift 48 · Moderate. Moderate. The poll-vs-correlation distinction for selection bias is a fresh, useful framing of standard methodology.

Real-world impact 2 · Minor. Minor. Statistical-literacy discourse (defending amateur survey work); no material reach. Within-niche.