Do poor people have a stronger relationship between income and mortality than the rich? Implications of panel data for health-health analysis