Radical
Statistics
BOOK REVIEW
Home Sweet Home? The impact of poor housing on health
Marsh, A., Gordon, D., Pantazis, C. and Heslop, P. Bristol: ThePolicy Press, 1999, (pp. 90+x), ppb. £16.99, ISBN: 1-86134-176-8
Review by Roy Carr-Hill
This is a monograph in the true sense of the word; it is the report of what can be learnt from (the analyses of) a single study - the National Child Development Study (NCDS) - about the impact of poor housing on health.
What They Did
The report includes a brief but comprehensive review of the associations between housing and health oriented to explaining the necessity for longitudinal data; hence the choice of the NCDS which has observations at birth, ages 7, 16, 23, and 33. In chapter 3, there is a careful discussion of the construction of the health measures in the study1 and a comparison of the measures used with population values around the same time (and there is a similarly careful discussion of the construction of a housing deprivation measure in the first part of Chapter 4).
They attempt in Chapters 4 and 5 to test the hypothesis that the impact of housing on health is cumulative by examining the odds ratios of contracting a respiratory or infectious disease according to lifecourse 'pathways' followed by the individual, whether defined in terms of 'overcrowding' or their constructed housing deprivation measures. Whilst, once again, it is exemplary in explaining precisely what they did, it is relatively disappointing (for the authors) because, even with such a large sample, several of the pathways end up with very few in the cells at age 33; and - they suggest (probably correctly) - because the meaning of 'deprivation' has changed substantially between 1958 and 1991.
These are all really preliminary analyses to the analysis in the final substantive chapter (Chapter 6), where they explore the extent to which lifecourse housing deprivation affects health after controlling for household income (and other confounders) using Cox's proportional hazard modelling and, although they acknowledge that there is a problem with the income variable (p57), they conclude that there is a substantial effect relative to other analysis in the field (see Table).
Table: Odds ratio for impact of housing deprivation over the lifecourse on health - CLICK HERE
Note: Taken from their Table 19
Commentary
Overall, relative to other analyses in the field, the approach taken by these authors is careful and comprehensive. However, to be purist, one might argue that:
- They could have made greater effort to devise a family/household income variable (just as they took a lot of trouble to devise the household deprivation variable), for example, by estimating an average permanent family income variable, during childhood (as others have done with NCDS) based on parental characteristics.
- They have not taken sufficient account of omitted variable bias, again as others have done, by including prior health status as a control. Indeed, whilst I would approve in principle of their exclusion of the social class variable, both on the grounds only that it is portmanteau and non-specific but also that there is no plausible direct link between social class and health (p.48), I did wonder whether the effect they identify would have survived the introduction of such a portmanteau control for earlier living standards.
Nonetheless, it is an interesting and important analysis. Moreover, I find that because the reporting of their exercise is so thorough, it highlights the disadvantages of using a longitudinal study. Apart from the obvious problem of attribution - which they document in Table 1 (p2) and demonstrate has little effect from their point of view in Appendix E2 - there are three other features of their analysis which derive from the longitudinality of the design but which generate problems for them.
- The relevance of their results for today when nearly all their sample had the four common infectious diseases as children (p12) - which could not be the case today.
- limitation of the health measures available to them (p14) because this was a general cohort study.
- changing meaning of overcrowding and housing deprivation (p39) (although it is unclear to me why they are so concerned to devise a single measure of housing deprivation at each stage).
As can be seen, from the page references to their text, they clearly acknowledge each of these difficulties and one can only wish other authors would do the same. However, it does make one wonder whether this is, in fact, the most suitable data set to use to test this hypothesis. They themselves, in particular, highlight the methodological importance of designing a measure of housing deprivation which is an appropriate and relevant measure of housing for current circumstances.
Despite these caveats, judged by the cannons of other analyses of the relationship between poverty and health, this analyses is exemplary and well worth ordering for your library.
NOTES
- They also have constructed what they call an 'innovative' index of severity.
- Although, I am rather surprised given the differences between the NCDS sample where the proportion overcrowded at age 33 is just over 3% whilst it is 5% in SARS (see Fig 2, p19) - the value for the 23 year olds in bizarre.
Roy Carr-Hill
Centre for Health Economics
University of York
Heslington
York
YO1 5DD
Tel: (01904) 430000
irss23@york.ac.uk
