Money and lies: proxy respondents and the mismeasurement of income in surveys