James P. Scanlan, Attorney at Law

There is a considerable body of literature on so-called reporting heterogeneity in self-assessed health – i.e., differences between the ways that persons in different socioeconomic groups with the same objective health will categorize their health status – and the implications of such heterogeneity with regard to the value of differences in self-rated health as an indicator of health inequalities.  Eventually on this page, and in conjunction with comments on the Huisman et al,[1] Dowd and Zajacova,[2] and Sing-Manoux et al.[3], Delpierre et al.,[4], and Burström and Fredlund,[5], I shall address the way that the literature is flawed for failure to recognize the implications of the pattern whereby the rarer an outcome the greater tends be the relative difference in experiencing and the smaller tends to be the relative difference in avoiding it – and its corollary whereby a factor that increases the likelihood of experiencing an outcome will tend to increase it proportionately more in the group with the lower base rate, while decreasing the opposite outcome proportionately more in the other group – as discussed, among other places, in references 6-9 below the Subgroups Effects sub-page of the Scanlan’s Rule page of jpscanlan.com.  This page merely sketches the topic.[i]

The patterns whereby the rarer an outcome the greater tends to be the relative difference in experiencing it and the smaller tends to be the relative difference in avoiding it can be illustrated with the data from Table 1 of the article by Huisman et al.  The table provides, or allows one to derive, the rates at which groups with different levels of education experience the following adverse outcomes: (a) having self-rated health less than very good (the four lowest categories); (b) having bad health (the lowest category); and (c) dying.

Table A below sets out these rates along with the relative differences in experiencing (AdvRR= adverse rate ratio) or avoiding these outcomes (FavRR = favorable rate ratio).  The table shows that the relative difference between the risk at which the highest and lowest education levels experience each outcome increases as the outcome becomes less common – with the increasing order being (1) self-rated health less than very good (risk ratio = 1.19), (2) mortality (risk ratio = 2.67); (3) bad self-rated health (risk ratio = 3.71).  On the other hand, the relative difference between the rates of avoiding these outcomes increases as the outcome becomes less common with the order being: (1a) avoiding bad health (risk ratio = 1.02); (2a) avoiding mortality (risk ratio = 1.16); (3a) avoiding health less than very good (risk ratio = 2.49). 

Thus, based on the relative differences in adverse outcomes, one might say that the disparity in mortality is greater than the disparity in health less than very good, and that the disparity in bad self-rated health is greater than the disparity in mortality.  So whether the disparity in the subjective indicator of health is greater or smaller than the disparity in the objective indicator (mortality) depends on where one dichotomizes self-rated health. 

On the other hand, based on the relative difference in avoiding these outcomes, one would say that the disparity in avoiding mortality is smaller than the disparity in avoiding self-rated less than very good health and that the disparity in avoiding bad health is smaller than the disparity in mortality.  Thus, focusing on disparities in favorable outcomes, one would again say that whether the disparity in the subjective indicator of health is larger or smaller than the disparity in the objective indictor of health turns on where one dichotomizes self-rated health – but with opposite patterns from those based on relative differences in adverse outcomes.

Table B below is based on the information concerning differences in the way clinical high blood pressure affects the self-rated health of women with the highest and lowest education found in Table 4 of Delpierre et al.  The penultimate column shows risk ratio for poor self-rated health for women in the two educational groups with the condition compared with those without the condition.  The column shows that having the condition increases the likelihood of poor self-rated health more among high education women than low education women, a pattern that led Delpierre et al. to conclude that women with high education are readier to let health conditions cause them to lower their self-rated health.  But that pattern is to be expected commonly to occur simply because the baseline poor self-rated health rate is lower among women with high education. 

The final column shows the risk ratio for avoiding poor health (i.e., experiencing health better than poor) for women without the condition compared with women with the condition for the two education groups.  The risk ratio is smaller for women with higher education.  So it could be said that having the condition reduces health better than poor less among high education women than low education women.  Thus, examining the opposite outcome leads to conclusions that are the opposite of those drawn by the authors. 

Note:  The pattern found by Singh-Manoux et al. (not shown) differs from that commonly observed in that poor self-rated health increases mortality more among the lower SES groups and is contrary to the statistical pattern described above.  This departure from the standard pattern warrants examination.  It could suggest that self-rated health in fact operates differently in this cohort.  It should be borne in mind that the same statistical tendencies observed in the other studies presumably exist in the GAZEL cohort, but that in the GAZEL study other forces were simply strong enough to overcome the statistical forces.  Given that the GAZEL cohort was relatively young, with very low mortality generally, there also exists the possibility that irreducible minimums played a role, as discussed in the Irreducible Minimums sub-page of the Measuring Health Disparities page of jpscanlan.com.

I am prompted to add the following point by a 2011 study by Janavek et al.[10] on differences in self-reported health between Roma and non-Roma in Serbia.  My being so prompted is more related to the subject of self-reported health than reporting heterogeneity, though aspects of the matter can be cast in those terms. 

In any case, as reflected in the Life Tables Illustrations sub-page of the Scanlan’s Rule page and several tables in the Mortality and Survival page, data on adverse or favorable outcome rates of advantaged and disadvantaged groups broken down by age groups tend to be particularly reflective of the standard patterns of correlations between prevalence of an outcome and relative differences in experiencing it and avoiding it because adverse outcome rates tend generally to differ greatly among age groups, at least when one compares, for example, a 70-79 year old group and the and 40-49 year old group.  Thus, the prevalence-related forces tend to be evident regardless of any meaningful differences between the comparative situations of the groups among 40-49 year olds and 70-79 year olds.

Data on rates of poor health of Roma and non-Roma from Figure 1 of the Janavek study by age group are presented in Table C.  The table shows the common pattern in which the relative difference in rates of avoiding poor health are much smaller among the young (where such avoidance is more common).  While one observes the common pattern of larger relative differences in the adverse outcomes (here, poor health) within the 31-50 year group than in the 50 and up group, the relative difference in the adverse outcome is not largest in the 18-30 group.  But rates of poor health in that group are so low that it is hard to attribute significance to this departure from the standard pattern.  Also irreducible minimums may play absolute differences may play a role.  The final column provides the EES figures discussed in the Solutions sub-page of the Measuring Health Disparities page, which figures provide the best indicator or the actual difference in self-rated health among Roma and non-Roma in each age group.

Returning to the reporting heterogeneity issue, the information for the 31-50 group and the 50 and up group does present the situation where one might say that growing older causes the non-Roma to increase the rates at which they regard themselves as in poor health more than it does for the Roma.  But one could just as well say that growing older causes the Roma to reduce their rates of reporting that their health is not poor more than growing older does for the non-Roma.

[to be continued]

Addendum Regarding Reverse Regression

(May 9, 2010)

The following is a tentative treatment of a matter that I am unlikely for some time to have the opportunity to think through carefully enough to treat in detail, but which is a matter of sufficient potential importance to warrant noting at this time.  

The unpublished article "Direct Regression and Reverse Regression: Contrasting Ways of Looking at the Wrong Information" (1992) addresses the pattern where the group with weaker qualifications will tend on average to be in a worse position relative to qualifications than the group with greater qualifications (direct regression) but that the group with weaker qualifications will tend on average to have weaker qualifications relative to position than the group with the greater qualifications (reverse regression).  Thus, from one perspective the group with weaker qualifications would appear to be disfavored while from the other perspective it would appear to be favored.  The subject is related to the matters addressed in the published "Illusions of Job Segregation," Public Interest 93 (Fall 1988), 54-69.

I am inclined to think that some of the same forces underlie patterns observed in Delpierre et al. in the following way:  Having clinical high glycosylated hemoglobin (HGH) seems to increase poor self-rated health (PSRH) more among highly educated women than low-educated women, while having PSRH seems to increase HGH more in low-educated women than highly-educated women.  Thus, from one perspective, the relationship between HGH and PSRH would seem stronger among highly-educated women and from the other perspective, the relationship would seem stronger among low-educated women.

But I need first to think through whether the description just presented is even correct.  I also need to reread the 1992 unpublished article and (1) make sure I still agree with it; (2) make sure that the pattern just-described reflects the same phenomenon described in the article; (3) determine whether such phenomenon is at all different from that illustrated in Table 2 above; and (4) determine whether the same statistical forces are driving the various phenomena.  See discussion the relationship of points illustrated in “Illusions of Job Segregation” to the patterns described in "The Perils of Provocative Statistics," Public Interest 102 (Winter, 1991), 3-14 (which, in the main, are the pattern whereby the rarer an outcome the greater tends to be the relative difference in experiencing it and the smaller tends to be the relative difference in avoiding it) in note 2 of the latter item.   

1. Huisman M, van Lenthe F, Mackenbach JP. The predictive ability of self assessed health for mortality in different educational groups. Int J Epidemiol 2007;36: 1207–1213.

2. Dowd JB, Zajacova A.  Does the predictive power of self-rated health for subsequent mortality risk vary by socioeconomic status in the US?  Int J Epidemiol 2007;36:1214-1221.

3. Singh-Manoux A, Dugravot A, Shipley M, et al. The association between self-rated health and mortality in different socioeconomic groups in the GAZEL cohort study. Int J Epidemiol. 2007;36:1222–1228.

4.  Delpierre C, Lauwers-Cances V, Datta GD., et al.  Impact of social position on the effect of cardiovascular risk factors on self-rated health.  Am J Public Health 2009;99:1278-1284.

5.  Burström B, Fredlund P.  Self rated health:  Is it as good a predictor of subsequent mortality among adults in lower as well as in higher social classes? J Epidemiol Community Health 2001:836-840.

6. Scanlan JP. Race and mortality.  Society 2000;37(2):19-35 (reprinted in Current 2000 (Feb)):

 http://www.jpscanlan.com/images/Race_and_Mortality.pdf

7. Scanlan JP.  Divining difference. Chance 1994;7(4):38-9,48: http://jpscanlan.com/images/Divining_Difference.pdf

8. Scanlan JP. Interpreting Differential Effects in Light of Fundamental Statistical Tendencies, presented at 2009 Joint Statistical Meetings of the American Statistical Association, International Biometric Society, Institute for Mathematical Statistics, and Canadian Statistical Society, Washington, DC, Aug. 1-6, 2009:

Oral Presentation: http://www.jpscanlan.com/images/JSM_2009_ORAL.pdf;

PowerPoint presentation : http://www.jpscanlan.com/images/Scanlan_JSM_2009.ppt

9.  Scanlan JP. Understanding Variations in Group Differences That are the Results of Variation in the Prevalence of an Outcome,  presented at the American Public Health Association 134th Annual Meeting & Exposition, 2006, Boston, MA, Nov. 4-8, 2006: Oral presentation: http://www.jpscanlan.com/images/APHA_Oral_Presentation.pdf; PowerPoint presentation: http://www.jpscanlan.com/images/APHA_Presentation.ppt

10.  Janevic T, Jankovic J, Bradley E.  Socioeconomic, gender, and inequalities in self-rated health between Roma and non-Roma in Serbia. Int J Pub Health 2011DOI10/1007/s00038-011-0277-1.