Download presentation

Presentation is loading. Please wait.

Published byEden Breckenridge Modified about 1 year ago

1
M. Azhar Hussain, Mette Møller Jørgensen & Lars Peter Østerdal Refining Population Health Comparisons: A Multidimensional FOD Approach M. Azhar Hussain Associate professor, Department of Society and Globalisation Roskilde University

2
Agenda Motivation What is FOD Data Indicators Joint health distributions Results Conclusion

3
Motivation There is great focus on ranking different populations Ranking of countries/regions Learn from best practise Ranking of population sub-groups Where to focus health inititatives Many rankings in the litterature and official reports, but often consistency problems when multiple indicators Our question: How to determine if a population sub-group has better overall multidimensional health status than another

4
Problem I What is the problem? When there are multiple health indicators we are not able to rank all combinations of indicators We can say: best combination when outcomes good in all dimensions We can say: worst combination when outcomes bad in all dimensions

5
Problem II Undecided: All remaining combinations are not rankable among each other Nevertheless many analysts will also rank in these cases without hesitation But this is not possible without problematic assumptions regarding weighting of different health dimensions In contrast, the first order dominance (FOD) approach is applied here do not weight health dimensions, and thus there are no problematic assumptions on substitution or complementary Only assumption: better health is preferred to worse health

6
Methodology I Stochastic first order dominance is applied We look at two distributions of populations across all possible binary health outcomes 0 is a bad outcome 1 is a good outcome

7
Methodology II n health dimensions 0/1 (unhealthy/healthy) valued binary health indices i, i : shares of population A and B, respect., with health status i ij : probability mass transfer from health status i to health status j : source-destination of pairs ij that move probability mass from a health status i to a less preferred health status j - meaning i is at least as good as j in all n dimensions Under these conditions, population A dominates population B if and only if there exists ij with ij ≥ 0, ii = 0, such that i + ∑ ji ∈ ji − ∑ ij ∈ ij = i for all i

8
Methodology III So a lot of technique and inequalities that need be fulfilled in order to decide whether pop A first order dominates population B But basic idea is simple Can we bilaterally move probability mass from better to worse within A’s distribution such that we get B’s distribution? In that case people multidimensionally must be better off in A than in B

9
Simple nummerical example, n =1 In the simple trivial one-dimensional case we e.g. have n =1 Indicator: self-reported health Outcomes: 1=good, 0=not good Population A: (1,0) is a=(80%,20%) Population B: (1,0) is b=(70%,30%) No doubt about ranking: A better than B But note:a + (-10%,+10%) = b Slightly more complicated technique… …but this is the FOD way of detecting dominance

10
Simple nummerical example, n =2

11
Simple num- meri- cal exam- ple, n =3

12
Simple nummerical example, n >3 Although, the FOD criteria are very demanding we nevertheless see FOD in previous studies Arndt et al. (2012). Child poverty In this study FOD is applied to health indicators for the first time Also we stress test the FOD using many indicators

13
Data Sample: The National Health Interview Survey 2010 Source: The Danish National Institute for Public Health Representative of the Danish adult (16+) population and for the five regions of Denmark Sample size is 11,433 individuals The basic indicators applied are selected from an even broader set of indicators from which we have chosen 22 They describe individuals’ general health status We use a combination of indicators covering self-reported health, pain and discomfort, chronic diseases, own health efforts, and risk factors

14
Indicators 22 to start with All 22 are used to create ”aggregated” indicators n =1: no problems in any of 22 health indicators n ={4,7,10} Different types of indicators used to create varying aggregated indicators All created such that a person has a good outcome in a dimension if she/he do not have problems with respect to any health indicator in a dimension

15
Aggregation rules and prevalence of good health in each dimension. %

16
Joint distrib ution of indica tors I

17
Joint distribution of indicators II people aged years medium higher education: 35% have good health with respect to all four dimensions, and 2.2% have bad outcome in all four indicators vocational training: 30% have good health with respect to all four dimensions, and 2.5% have bad outcome in all four indicators Thus, the better educated group have (relatively) more people with the best outcome and fewer people with the worst outcome So, regarding the two most extreme cases, we see medium educated perform better than people with vocational training We even see that using a summary measure (bottom of table) taking all sixteen outcomes into account, medium education outperforms vocational training, when equal weights are assumed across the four dimensions But if we change the weights and give much higher weight to dimension iii then people with vocational training have better multidimensional health than people with medium training In other words, in this case the best performing of the two educational groups depends on the applied weighting scheme This is because in the intermediate indicator combinations we cannot in many instances unambiguously classify one outcome as being better than the other

18
Results: only age

19
25-44 years

20
45-64 years

21
Other than education

22
Fraction with ranking

23
Non ranking We observe a few dominances in the 7+ dimensional cases (7 or 10 dimensions) Reflects that in an analysis of multidimensional health we can actually often not rank groups unambiguously when there are a lot of indicators Lack of dominances is therefore not a methodological flaw in the FOD technique

24
Conclusion Of course, allways rankability in 1-dimensional case Often able to detect FOD in the 4-dimensional case Sometimes in the 7-dimensional case But only once in the 10-dimensional case => Difficult to conclude robustly about which group dominates another when more and more indicators are included, since with an increasing number of dimensions the prevalence of FOD decreases drastically Other data or other countries with larger inequality, maybe more FOD in 10-dimensions as well – need to analyze But in any case more indicators => fewer FODs

25
Thank You

Similar presentations

© 2017 SlidePlayer.com Inc.

All rights reserved.

Ads by Google