Income-based policy preferences are estimated at the 10th, 50th, and 90th percentiles.

Evidence

For each case, Gilens used the original survey data to assess responses by income level. In order to cope with varying income categories across surveys, he employed a quadratic logistic regression technique to estimate the opinions of respondents at the tenth income percentile (quite poor), the fiftieth percentile (median), and the ninetieth percentile (fairly affluent). — Martin Gilens and Benjamin I. Page, Testing Theories of American Politics, 51

Categorization

• QUESTION: Why quadratic logistic regression specifically - and why ‘categorical rather than conditional phrasing’ as a case criterion?
• XREF: connects to: Affluent-proxy preferences understate the true gap between elites and ordinary citizens. — Martin Gilens and Benjamin I. Page, Testing Theories of American Politics:…, 105

Related

• Top-10-percent income earners’ policy preferences correlate at r=.91 with the top 2 percent’s. (similarity 0.83)
• Income-percentile measures cannot isolate wealthiest elites from prosperous ninetieth-percentile citizens. (similarity 0.80)
• Affluent-proxy preferences understate the true gap between elites and ordinary citizens. (similarity 0.78)