While we estimated Eqs. 1 and 2 using linear regression, as a sensitivity check (available upon request) we also estimated these models using an ordered probit model. All results were qualitatively similar whether estimated using an ordered probit or linear regression. Because of concern that collinearity among the consumption variables might contribute to large standard errors, we calculated the proportion of the total variation in each consumption variable that can be explained by the other explanatory variables, . These values range between 0.11 (for health care) to 0.58 (for leisure spending). Larger values of can result in relatively high standard errors (Greene 2003).Footnote3