Tuesday, May 09, 2006

A gap that must be closed

Shortly after Don Brash made his Orewa speech and National started pontificating about "need, not race", researchers for the Ministry of Health revealed that, where mortality rates were concerned, race is need - that Maori die younger than Pakeha, even after correcting for socioeconomic status. The research was part of a long-term study on ethnicity, socioeconomic status, age, and mortality conducted by the Ministry of Health and the Wellington School of Medicine, and the third part of it - Decades of Disparity III: ethnic and socio-economic inequalities in mortality, New Zealand 1981 – 1999 [PDF] - was released yesterday. The aim of this part was to analyze the interaction between ethnicity and socioeconomic status and try and determine the broad effects of each on mortality. And their conclusions are just as devastating for the National Party's dogmatic position as the last lot.

Firstly, the money quote: there is a clear ethnic inequality in mortality rates. And it increased significantly over the 80's and 90's:

The first report... demonstrated a growing disparity in life expectancy between Maori and non-Maori throughout the 1980's and early 1990's. After correcting for under-recording of Maori ethnicity on mortality records, we found that the ethnic disparity in life expectancy at birth increased from six to seven years in the early 1980s to eight to nine years at the end of the 20th century

The reason for this is that while life expectancies generally rose, those of Maori rose slower than those of Pakeha - and those of low-income male Maori stayed stagnant (and even dropped between 1986 and 1999).

About half of this ethnic difference, and between a third and half of the widening, can be attributed to socioeconomic factors. For various reasons, this is almost certainly an underestimate - but it certainly doesn't explain everything. There is a racial difference here which is not just explained by the fact that far too many Maori are poor.

The preferred right-wing explanation for this is "lifestyle" - Maori just choose to die younger, by living unhealthy lives. Unfortunately, this doesn't really cut it. The chief "lifestyle" culprit - smoking - is already strongly correlated with socioeconomic status, and when that is corrected for, it tends to drop out of the equation. A subsidiary study on the same data showed that smoking made only a modest contribution to ethnic inequalities in mortality rates, and other lifestyle causes - diet, alcohol, exercise - are likely to be even smaller.

Which leaves the explanations the right don't want to talk about: differential access to the health system. Racial discrimination, both in quality of care and in wider society. Increased psychological stress as a result of the above. These are real problems, with real effects, and tackling them will require the commitment of targeted funding - but it is not something we can afford not to do, unless we want to abandon our traditional commitment to equality entirely.


Look at the Pou case - a woman of Maori descent dies of cancer at the age of 52 after succumbing to the blandishments of the tobacco industry and the judiciary says its a matter of individual responsibility. The gaps are opening wider than ever, although I think it is class based, not race.

Posted by Anonymous : 5/09/2006 05:40:00 PM

It's hard to dispute the pattern, but I can't leap to the conclusion that differential access to health care is the cause. For starters, it's questionable whether all or most of a 8-9 difference in life expectancy could be attributed to health care. Second, I'm not convinced of any systematic barriers to health care in New Zealand, at least among citizens who can speak passable English, which would presumably capture almost all of the Maori population.

Diet + alcohol + exercise + health care = a possible explanation for a difference of this magnitude.

Posted by dc_red : 5/09/2006 07:56:00 PM

> that Maori die younger than Pakeha, even after correcting for socioeconomic status.

It is not rocket science - genetics play a role in life expectancy.

Also as a side note - the world isn't fair.

That doesn't mean you give up on improving maori's lot - but instead that that people with parents who lived to be 90 tend to live longer than people with parents who lived to be 60 is not in itself evidence that the latter group were abused.

Posted by Genius : 5/09/2006 08:05:00 PM

Of course I could tolerate a health care benefit for people with grandparents and great grandparents who died young which would probably be a VERY strong indicator that they "won't be on the super very long" (to use a euphamism).

Posted by Genius : 5/09/2006 08:08:00 PM

I also wonder whether access to health care can be the whole of the story. I don't know but I wouldn't be suprised, given the prison population statistics, if there is a higher chance of being murdered, badly injured leading to a decreased life span, or killed in a car accident for Maori.

I also don't think you can completely discount diet even though unhealthy diets leading to health problems may be a consequence of poverty rather than an intentional life style "choice".

Posted by Amanda : 5/09/2006 09:12:00 PM

Anon: I think it is class based, not race.

That's really not what the statistics say. Look at the graphs on p 18, which seperate mortality by both class and race. I think they tell their own story.

DCRed: Firstly, race accounts for half of that difference, not all of it. There is a clear socioeconomic factor (which they recommend tackling through full employment), and a clear ethnicity factor. As for exactly what is going on with the latter, I suggest reading pages 59 and 60 of the report, which cover other factors at play. It suggests that there is differential access - in the form of both differential utilisation, and of differential quality of care - particularly with respect to cancer and heart disease (which are two of the biggest killers). Europeans seem to get more and better care than Maori, even though the latter have greater specific need in at least one of the areas in question.

As for diet, here's a quote:

Beyond tobacco smoking, diet, alcohol consumption and possibly physical activity may all contribute to ethnic mortality inequalities through pathways that are both dependent on and independent of socioeconomic position. However, their contribution to the latter is likely to be even smaller than that of tobacco consumption.

Basically, almost all of that is already taken out by socioeconomic status, and the leftovers simply aren't strong enough to give the effects we're seeing here.

Posted by Idiot/Savant : 5/09/2006 11:34:00 PM

The next question would be that if access to health care explains mortality differences - why do Asians have better access to health care than Europeans (in basically every country in the world) despite being a very broard group and being significantly poorer and part of much smaller communities etc? why do southern people have less acess in basicaly every country in the world even where they are the vast majority?

Actually there might be something to that - it would imply what you need to do is train more Polynesian doctors, of course to do that you would have to lower the standards (as you would if you wanted to artificially increase the amount of any group). With the idea that doctors tend to not treat people of other races properly.

Posted by Genius : 5/10/2006 07:30:00 AM

I/S: could you be a bit more precise with your description of the (negligible) effects of diet, smoking,exercise and alcohol on the ethnic disparity? They could all have small effects considered individually, but have synergistically larger effects when they occur together in the same person. And even if they don't interact synergistically, it's not clear to me what the stats look like after adjusting for all of them individually .

Posted by stephen : 5/10/2006 08:30:00 AM

I have similar queries to genius, stephen and others above... the hand-waving about diet and other lifestyle issues in the report (and lots else) simply won't do.
I/S says:
"As for diet, here's a quote [s.g. -note that's *the* quote, the report has nothing more to say except for a little early on that simply expresses the report's prejudices which I'll come back to below]:
'Beyond tobacco smoking, diet, alcohol consumption and possibly physical activity may all contribute to ethnic mortality inequalities through pathways that are both dependent on and independent of socioeconomic position. However, their contribution to the latter is likely to be even smaller than that of tobacco consumption.'

Basically, almost all of that is already taken out by socioeconomic status, and the leftovers simply aren't strong enough to give the effects we're seeing here."

The authors of the report claim to know, with I/S's enthusiastic support, that, say, poor maori have essentially the same lifestyles as poor non-maori except for racial discimination of various sorts [the non-socioeconomic mediated differential access to healthcare the authors allow may be important largely boils down to a form of racial discrim.: "physicians may unconsciously activate racial stereotypes in patient consultations, resulting in discriminatory treatment or referral patterns." (59))]

I don't think they do know that, indeed I suspect it's false. The authors admit that there are smoking differences and that *they* may account for 10% of the controlled-for-socio-economics difference in mortality rates. But a few more contributing factors at the life-style level - particularly if ginned up with some strong interaction effects (as Stephen remarks above) could easily explain a hell of a lot more and socan't just be waved away. Put it another way - the authors are claiming to *know* that either there is no soc-ec status-controlled-for additional obesity in maori or that if there is then that must be a result of broadly racial discrmination. And that's nuts.

If you want to get a sense of what's wrong or at least very shallow with the report technically look at the theoretical framework model on p. 3 (and in Fig 1.).

Here's the attempt at p. 3 to justify the crucial step that the theoretical framework model represents (all the regressions they run - poisson and otherwise will largely just bounce around the model as drawn):
[Note I add my own emphases.]

"A wide variety of factors *may appear* to mediate between ethnicity or socioeconomic position and health outcomes. These factors include health-related and lifestyle behaviours, family stress, work and social environments, religious beliefs, personality and other psychological factors. However, these factors *may be considered proximal or surface causes*, because in the causal chain they are close to the health outcome. As the framework illustrates, *such surface causes are not independent factors but have themselves been shaped by underlying social forces (distal or basic causes)*. Paying attention to the distal as well as the proximal causes can guard both against
blaming the victims for their behaviour, and against analysing data in ways that reinforce deficit thinking. Also, interventions aimed at surface causes will be less effective if they do not taken account of the social contexts in which vulnerable populations live."

This is weird....variables that "appear to mediate" are originally cited as a possible source of objection to the study then that mediateness/closeness to health outcomes is cited as a reason to dismiss them as strictly dependent variables.... And "blaming the victim" is suggested as a worry if you do otherwise. Uh-oh.

I urge everyone to study the theoretical framework model on p. 3. Here's the problem with their attempt to justify it: the "appear to mediate" objection and its dismissal is a misdescription justified only by the model that's suppposedly at issue! A more neutral description of the real possible objection is not that there are mediating variables or that there are variables that aren't at all shaped by (are independent of) the social status level factors - maybe there are no such variable - but rather that there are *other* variables including those labelled as surface causes in Fig 1. whose values *aren't* functions *just of* the social status variables. But there's *no* reason to think that the "surface cause" layer variables (and other lifestyle/culture variables ) are strictly causally posterior to the so-called "social status" layer, or that they are screened off from all the basic causes by the social status layer variables.

By embracing the theoretical framwork model that they do the authors of the report in effect *force* lots of life style variables to impact on the data only through social-ec values and ethnicity values so that when you control for soc-ec status remaining differences are ethnic ones. Sorry. Unless I'm missing something pretty obvious, that's a classic social science mistake.

I'm prepared to believe that the marked causal links in Fig 1 exist (at least under some interpretations) but there should at least be direct arrows from the basic causes to the surface causes too, and there are many other links I'd want to see allowed in at the beginning theoretical framework stage (you can then let the data drive the causal influence value to zero, but if you just assume the possibility out of existence to begin with you're simply guaranteed to draw spurious conclusions). For example, it just *is* weird to have no direct causal arrow from "biology" at the basic cause level to items at the "biological process" level. Why should all biology/genetic influence on biological processes (and thence on mortality) be completely via biological/genetic influence on social statuses? They shouldn't be - Genius is right. The report/study therefore seems to exclude this whole dimension of the truth by fiat.

Similarly for all sorts of other culturally specific differences that might be very hard to quantify or even understand. For example, Maori families and groups could be more macho or be more structured by defense-of-honor-style interactions than non-Maori families and groups in NZ. Who knows? I suspect that stuff like that's important and is a chunk of what's coming out in the data after you control for soc-ec status etc.. The authors claim to know that that's not so and indeed that nothing else of importance other than racial discimination of various sorts is going on... I say, believe them and be a social science victim! (I in fact doubt very much whether all of the health-relevant ways people live that are culturally-specific are even understood well-enough to be *named* let alone measured precisely to be judged as having the same values...)

The Discussion (pp. 53) leans a *lot* on the inadequate theoretical framework model to introduce racial discrimination and its stresses as an important causal factor for explaining the soc-ec-controlled-for differences.... That's a hard thing to observe, to confirm, and even to get clear about conceptually. There must *be* influences from such factors in my view since racism is certainly real, but one would need to be very careful to have factored everything else out of consideration to be able to seriously claim to actually *find* it in the data. I therefore expected violence and crime and the stresses that come from dealing with *that* to somehow be allowed for in an ethnicity and mortality report, i.e., given basic violence and crime stats that are known to be ethnically polarized after you control for socio-economic status... If you don't control for that then you probably can't hope to observe or measure "the personal experience of racial discrimination [which] has been shown to elicit strong physiological stress responses" (59) in the strongly entangled data that the real world presents.
I was therefore disappointed and a little shocked to not see any of *that* stuff discussed let alone nicely factored out in the report. It doesn't appear to be discussed in suppporting materials either (unless I missed something). Once again, a whole fairly obvious side of culture appears to just drop out, excluded by fiat as a causal influence in the light of the theoretical framework model.

It'd take weeks (that I don't have) to work through all the supporting materials properly, and all my remarks here are highly provisional but (1) There's no data-driven discovery of causal structure going on in the report. (2) It assumes a very elaborate and not-especially-plausible causal structure to begin with. (3) Alternative hypotheses inconsistent with the theoretical framework model are barely entertained let alone disproved/disconfirmed. Those are all frankly terrible signs.

The report does confirm that controlling for soc-ec statuses leaves mortality gaps between ethnic groups. That's important and useful to know (although not especially surprising at least to me).

But the report (i) spuriously in my view forces all the unexplained stuff into the "There's Something about Maori" box and then (ii) equally preposterously in my view suggests that soc-ec similar maori and non-maori are essentially the same so that the downstream differences between members of those "boxes" must all be external, i.e., about how those boxes are treated by the wider society. I mean, bloody hell!

In sum every alarm bell I've developed over the years for assessing social science research is ringing. Several compensating errors appear to being made simaultaneously, and (i) in particular kind of traps critics - one might want to say, for example, that macho/honor-code culture kills and if maori (or certain iwi or iwi-less maori etc) have a lot of that then look out, but it comes out as saying in the first instance "Something about Maori" which can sound nasty. It's incredibly irritating. Ultra caveat lector.....unless I'm very much mistaken... :)

Thanks to i/s for linking to the report. What an eye-opener!

Posted by Anonymous : 5/10/2006 03:59:00 PM

The funny thing is, stephen glaister, that we may be on opposite sides in terms of our assessment of the effects of racism and oppression. My strong intuition is that these are powerful, and this report is fuel for that. But I absolutely think that one especially needs very very solid evidence when the conclusion is tempting.

Posted by stephen : 5/11/2006 08:44:00 AM

Stephen: I'm definitely prepared to believe that racial discrim. of various sorts is a powerful factor too (although I'd be surprised if it was), but I seriously don't think this report provides anything like the sort of evidence one needs to draw that conclusion. Unless you're an expert you won't see that there's just *so* much room for stuff to go wrong. Here's a simple data set/example I've used with students occasionally to illustrate a rather different worrying point from those I made above. Suppose there are 5 levels of performance/qualification that workers in an organization exhibit: 21, 11, 6, 1, and 0 (e.g. #'s of papers published in the last 5 years in an academic context). Now let the workforce be divided into two groups: the M's and the non-M's (e.g., men and women). It's then *easy* to flesh out the case so that at each qualification level the average M earns more than the average non-M, and yet it's also true that, at every pay level, the average M is more qualified than the average non-M at that level. The former regression suggests discrimination against non-M's whereas the latter suggests the reverse! Most people have *no idea* that this sort of conflict is even possible.

Here's a data set to get you the strange result:
21: M50 (i.e. there's a 21-level qualified M earning 50K)
11: M60,60,40,40; non-M48 (i.e., at the 11-level there are 2M's earning 60K, 2 earning 40K, and a non-M earning 48K)
6: M50,30; non-M38
1: M40,20; non-M28K
0: M30,20,20,20

Non-M's earn 2K less than the average simlarly qualified M. But at every pay rate M's are more qualified on average than non-M's (you may have to graph and interpolate or do a regression estimate line to convince yourself of the latter... or you can just trust me!)

I haven't had time yet to go through and test the (horribly named!) "Decades of Disparity" report along these sorts of lines yet. I couldn't see, right off the bat, how to turn their highly aggregated mortality data around....

Posted by Anonymous : 5/12/2006 01:21:00 AM