| > "Using average income, a clearly bad statistic, can't be justified by saying you could not find a good statistic." As lliamander already mentioned, I addressed this on my blog. Several quick points: 1) International comparisons of health spending overwhelmingly relate directly to mean health expenditures (e.g., per capita, % of GDP, etc). Even if you know nothing of this area it seems most reasonable to compare mean health spending to mean material living conditions (disposable income or consumption)! 2) Individual income and health expenditures on behalf of individuals (i.e., not just OOP) are effectively uncorrelated within countries in the developed world (including the United States). There are clearly large national level effects independent on one's place in the national income distribution, which are exceptionally well correlated with mean income (r^2>0.9), and we shouldn't expect to income of the median person to be terribly informative. 3) We know healthcare to be heavily socialized in the United States and throughout the developed world. Mean income is a much better indicator of overall financial wherewithal than median income. The median just doesn't make much sense from a theoretical point of view. Moreover, estimates of the income distribution and the medians are surely subject to greater measurement error and issues of comparability between countries, so there are practical issues with this as well. 4) More practically speaking, I've run many different regressions using medians, various indicators of income inequality, etc and found little to nothing to suggest the income distribution is an important independent predictor. The r-squared and various goodness of fit indicators suggest these metrics perform less well than the National Accounts-based means I have used for disposable income and consumption. In multiple regression on these necessarily smaller samples (which are easy to over-fit) the coefficient/effect size of these distribution-related indicators are rarely significant, don't significantly improve model fit, and would almost certainly be rejected in lasso or the equivalent. https://randomcriticalanalysis.com/2018/01/20/on-the-relevan... 5) With other aggregate consumption/expenditure statistics, such as food, housing, transport, entertainment, etc the fit is also (unsurprisingly) much better with my statistics than cross-sectional comparisons of median income, so I'm not sure why you would expect one of the most socialized and most elastic categories of expenditure at a national level (~1.4 as a function of GDP) to be any more favorable for the distributional perspective. https://rpubs.com/random_critical_analysis/oecd_2014_consump... |