Response

摘要

The problem of how to describe, detect, and measure causal effects has an importance in economic and social policy analysis that transcends the specifics of our paper, and our discussants provide valuable perspectives on possible approaches. We thank them for their comments. In this response, we identify points in the discussion that we find particularly useful, and try to clarify some issues where there appears to be disagreement. The comment by Jerome Adda, Tarani Chandola. and Michael Marmot focuses on the association of health and wealth in panel data. They replicate our statistical analysis on two data sets, the Whitehall II panel in Great Britain and the Swedish Survey of Living Conditions ULF, and conclude that these replications give similar results. This is extremely valuable, providing a powerful cross-population/cross-institutions test for model invariancc. However, their general assessment that our model is transferable to these data sets may be overly generous; a detailed comparisonreveals some significant differences whose exploration would be a good starting point for further research. Their finding that mental diseases in the British and Swedish data also fail the test for no direct causality suggests strongly that the sources of this rejection are behavioral factors, rather than our proposal of a possible gradient in affordability of prcventative mental health services within Medicare. James Poterba points out that a detailed look at disease-specific therapies and Medicare reimbursement rules may permit a sharper test for a causal link from affordability of prevcntative care to health outcomes, and changes in medical insurance coverage over time or with recipient age may induce in variance failures unless they are accounted for explicitly.