RE: Re: Income Inequality and Health

["Brown, Martin - ARP (NCI)" <mbrown@xxxxxxxxxxxx>]

I'll try to make the issues clear as mud...

I was asking for opinions about the worth in this particular case, not
asserting that they are not worth the effort.  The most commonly noted
weakness of ecological regressions has to do with measurement error.  More
example, say  we are trying to establish a relationship between health
status and income and we have individual data on stage-at-diagnosis for
cancer (how early or late the cancer is diagnosed) and census tract level
data on average income.  There is measurement error in the latter as a proxy
for individual income.  The most common criticism is that this result in
lower power to detect a relationship when one actually exists, but some
statistical purists say the bias can actually go in either direction.  The
discussion starts to get murky in the case of the Wilkinson hypothesis
because this is a hypothesis that is inherently ecological, i.e., the
relationship between some measure of average health status and some measure
on social structure.  The problem is that this hypothesized ecological
relationship is confounded by the fairly well established relationship
between individual health status and individual income (or other measures of
individual social status).  To deal with this problem Michael Wolfson
simulates the expected effect of the individual level relationship on the
ecological level and shows that there is still a residual effect at the
ecological level that cannot be explained by the individual level
relationship (this is for income/health, income distribution/average health
for U.S. SMSAs).  Deaton, who have been very critical of the Wilkinson
hypothesis accepts the Wolfson analysis but then says that percent black
performs better in the ecological relationship than measures of income
distribution.  But I find percent black not to be nearly as conceptually
compelling as income distribution as an ecological variable.  E.g. what kind
of causal mechanism stories go with these measures???

-----Original Message-----
From: Bill Burgess [mailto:burgess@xxxxxxxxxxx]
Sent: Monday, August 27, 2001 3:07 PM
To: pen-l@xxxxxxxxxxxxxxxxxxx
>Subject: [PEN-L:16413] Re: Income Inequality and Health


Martin Brown wrote that ecological regressions (like average health against
average income plus income inequality) are not worth the effort.  Could you
expand a bit on why? I think regession assumptions like linearity,
independendence of variables and unidirection of causality are big problems
(on top of many issues regarding measuring health), but is this what you
have in mind? If so, can you cite a non-econometric-technical summary of
these problems, especially as they apply to health?

Bill Burgess


At 03:12 PM 24/08/01 -0400, you wrote:
>I'll try to respond to this when I have more time to do it right.  But
there
>is something else I wanted to bring up from the International Health
>Economics Association meeting.
>
>There were several plenary and regular sessions focusing on the "Wilkinson
>Hypothesis".  That is to say the theory that there is a relationship
between
>macroeconomic measures of income inequality and average health status.
This
>relationship is above and beyond that expected by the "Prescott Curve,"
that
>says there is a strong relationship between the level of individual income
>and individual health.  To make a long story short, the consensus at the
>meeting both from those who had been advocates and detractors of the W
>hypothesis in the past is that current data and/or sophisticated analysis
>does not support the hypothesis for most situations examined - e.g. OECD
>countries, within UK, within Canada, within Australia.  The remaining, very
>important case, is within the U.S.  Some cross-sectional analyses of SMSA
>data within the U.S. - notably by Michael Wolfson of Statistics Canada -
>strongly support the hypothesis.  The counter-argument, put forward by
Angus
>Deaton - an econometrician/development economist - is that when one enters
>percent black population into the regression for the U.S. the coefficients
>on the inequality measures drop out.  This only happens if one looks
>separately at health status (e.g. mortality) for blacks and whites
>separately.  And, note, white mortality is inversely related to percent
>black population.
>
>There was some discussion to the effect that macro measures of social
>structure still matter but that things like Gini coefficients of measured
>income were never very good measures.  Some discussion about dysfunction
>urban structures in the U.S. being the real issue, etc....but apart from
>this what should we make of this debate??
>
>1]  All attempts a these kinds of ecological regression are not worth the
>effort.
>2]  There is rationale for Deaton to substitute percent black for income
>inequality.
>3]  Percent black is a proxy measure for something that really is important
>- but what is it??
>
>I will say this for health economics.  1] Would the questions of inequality
>ever dominate a meeting of AEA?   2] Would everybody at an AEA meeting,
even
>those on the political right end of the debate, concede the importance of
>the Prescott curve, say that economists have ignored this for far too long
>and that we need to learn a lot more about the specific mechanisms behind
>this statistical relationship and intervene with social programs to address
>it?  3] Acknowledge that the Prescott Curve, alone, tells us that total
>social welfare would/should be improved by transfering social resources
>toward lower end of the income distribution (because 99% of health
>economists have pretty much accepted the proposition that a additional unit
>of health is/ought to be worth at least as much to a poor person as a rich
>person).
>
>On the down side, this debate has received the least visibility in the one
>country where the evidence suggests that both the Prescott Curve and
>(perhaps) the Wilkinson effects are the strongest - the US.