Author s response to reviews Title: Material deprivation and health: A longitudinal study Authors: Anne Grete Tøge (anne-grete.toge@hioa.no) Ruth Bell (r.bell@ucl.ac.uk) Version: 1 Date: 04 May 2016 Author s response to reviews: Dear reviewers, We are grateful for your comments, and believe they have improved both the analysis and manuscript. Our response is attached point-by-point below. All page and line numbers refer to the revised version with track changes: Reviewer #1: I really like to thank the author for this manuscript. It was a pleasure reading it as it totally fits to my research interests. Therefore, my comments on the theoretical foundation of your manuscript are a bit longer. I think your manuscript has very interesting results and a fine discussion. However, in the introduction I found some inconsistencies that you need to address. Abstract 1. I think it is a bit misleading to speak about "Longitudinal data on indicators of material deprivation" if you use only one indicator of the living standard questionnaire.
We have deleted indicators of. Line 32, page 2. Background 4. The idea of your paper is very good, but reading only a few lines of it makes me a bit confused. I know from my own experiences in this field that there are a lot of terms referring to material deprivation (relative & absolute deprivation, social & material deprivation, material hardship, relative & absolute poverty, living standard). Material status was new for me and will be added to my list of available terms. However, I think it is really important to stick to one term. We have now used material deprivation instead of material status/ material hardship. Lines 30 and 32, page 2; Lines 64, 67 and 70, page 4; Lines 103, 105, 107 and 116, page 6; Lines 146-147, page 8; line 204, page 10; line 223, page 11; line 250, page 13; Lines 285 and 287, page 14; Lines 304, 308 and 310, page 15; Line 332, page 16; Line 357, page 17; I also would not address the theoretical background of social status and the social gradient in health as it is still not clear how the socioeconomic status is related to the concept of relative/material deprivation. On the other hand, you might argue that income is related to material assets such as affording a car, which has an influence on health. In this really simple case, it is not relevant to discuss the whole story about poverty and deprivation. I hope you understand what I mean. We have changed the background section according to this comment. Lines 55-62, page 4.
Measuring poverty 5. Don't ou think that material deprivation and material hardship are two different concepts. Relative deprivation refers to the concept of Runciman while material hardship to the concept of absolute poverty. We agree, and do now use material deprivation as the main concept (see response above). We have also added this as a limitation of our study, as using cannot afford a car cannot separate between the effects of relative deprivation and material hardship on the limitations among longstanding ill. Lines 337-341, page 17. 6. I do not see what lines 35-43 to your argumentation. We assume that the reviewer refer to page 4 (in the previous draft), and have removed the removed the paragraph. Lines 93-98, pages 5-6. 7. This section was a bit confusing for me. You start with a really good definition of "relative" poverty. Why did you talk about relative and absolute standards in the next section, when you use a relative definition of poverty? We are glad to be made aware of this point, and have now deleted this heading and the first sentence (Lines 117-118, page 6), and made changes in lines 123-129, on page 7. As it is very difficult to distinguish the effect of absolute and relative poverty using cannot afford a car as a proxy, we have moderated the definition of material deprivation. Lines 337-341, page 17. You might change the title to material deprivation and health.
We have changed the title to Material deprivation and health: A longitudinal study However, this definition is based on the living standard concept of Townsend (you might read one of our publication on this from Pförtner et al. 2011) which has been implemented in the EU- SILC. The EU-SILC claims to measure a specific level of necessities which allows to participate in society. If an individual could not afford a specific set of items, than relative deprivation occurs. I'm aware that Guio split the index, but from a theoretical and conceptual point of view, these items are made to indicate societal living standards and relative deprivation (Therefore, it really makes me wonder, why these authors talks about material deprivation, also I'm aware of the aim). I missed this discussion in your text. We have incorporated this comment into the manuscript. Lines 123-129, page 7; Lines 337-343, page 17. You talk too much about the specifics in the EU SILC, which is not necessary as this is a theoretical and not the method section. Specifics on data and analysis are now removed. Lines 139-145, pages 7-8. Also, I would strictly separate the argumentation about poverty and the effects of poverty. The most tricky problem might be that you use only one living standard item; the car. It is tricky as the affordability of a car only indicates the affordability of a car and not a latent construct of deprivation and living standards. It is totally fine to do that, but by doing this you are moving away from this complex theory of multidimensional poverty to a concept of absolute material hardship.
Lines 341-343, page 17. 8. I'm really not aware about the differences in longstanding illness and limiting longstanding illness. Are there any differences (empirically and theoretically)? We have now included a section on longstanding and limiting longstanding illness in the start of the paper. Lines 55-62, page 4. Relative vs absolute standards 9. Line 49, page 5: a "that" too much? This is now removed 10. Line 10, page 10: "understanding of relative???," This is now specified (outcomes as relative to other respondents/groups or oneself at another point in time). Lines 130-131, page 7. 11. As said above, you might change the title. We have changed the title to Material deprivation and health: A longitudinal study
12. Line 44-47, page 5: I would disagree, absolute lack of material resources can also contribute to psychosocial pathways, for example by shame, economic worries and anxiety of being not able to afford the next day. We agree with the reviewer, and that was what we meant, however, we recognise our phrasing easily could be misinterpreted. Lines 123-129, page 7. 13. Line 52, page 5: "Low material standards may also" Why also? The idea of relative deprivation (I don't know whether it is helpful to talk about standards in this case) and health is that it is based on social comparisons. It is an implicit assumption of the theory of relative deprivation (see Pförtner et al. 2011; Imlach Gunasekara et al. 2013; Pförtner and Elgar 2016;), which has its associations to arguments of Wilkinson. We have rephrased this section according to this comment. Lines 123-129, page 7. 14. Line 10-23, page 6: In two publications, we argue that material deprivation indicates a long period of poverty and deprivation as it is much stronger associated with health than income and other indicators of the socioeconomic status. This could be added here. We have added this in lines 136-138, page 7. Data and methods Explanatory variable
15. I really liked the argument using a car as a proxy for material status/deprivation/hardship, but do you have any empirical proof for your argumentation? Whether people can afford a car is shown to have a clear income gradient. Lines 181-183, page 9. 16. Line 30-35, page 8: This sentence needs to be edited. I do not really understand it. We have rephrased this section. Lines 184-189, page 10. 17. "All observations after the transition are coded 1 in order to avoid transition back to affording a car again contributing to the estimates". I do not really understand this sentence. If I understood you correctly, 0 = have a car/have no car because of other reasons; 1 = cannot afford a car. Thus, you might use a common dummy variable and - as you are using fixed effect regression - you are implicitly looking only at transitions from 0 (t-1) to 1 (t). We agree, and have no simplified the coding according to this comment. Lines 184-189, page 10. 18. I don' think it is necessary to introduce a second variable on transition as it is too strongly linked to the other transition dummy. We agree and have removed this dummy. Lines 191-193, page 10. Discussion
19. Line 18-19, page12: I wouldn't argue that strict and rather say that it contribute to a restriction. We have rephrased this section according to this comment. Line 286, page 14. 20. The discussion is very fine written, but does not correspond with the introduction at all. Also, it would be nice to read any suggestions for prevention. What should be done? Response: We are very grateful for the detailed suggestions for revisions of the background part of the paper, and believe these revisions have made the discussion more compatible. Unfortunately, our study does not provide any evidence for possible effective prevention, but suggests that further research should investigate effects of policies and schemes aimed at reducing the negative health effects of poverty and material deprivation. Lines 364-366, page 18. Literature: Pförtner, T-K, Elgar, FJ. Widening inequalities in self-rated health by material deprivation? A trend analysis between 2001 and 2011 in Germany. J Epidemiol Community Health, 2016; 70: 82-89. Pförtner, T-K, Andreß, H-J, Janßen, C. Income or living standard and health in Germany: different ways of measurement of relative poverty with regard to self-rated health. Int J Public Health; 56: 373-384. Imlach Gunasekara F, Carter KN, Crampton P, et al. Income and individual deprivation as predictors of health over time. Int J Public Health 2013; 58: 501-511.
Reviewer #2: This is an important manuscript that investigates association between change in material status and illness. However, the analysis in the paper is flawed. The paper is not suitable for publication in its current form. Specific comments * The authors should account for the longitudinal nature of their data. Repeated data per patient over time are expected to be correlated and failure to account for such correlation may be prone to false conclusion. It is also important to account for country effects due to the different social and welfare systems. The authors should note that cross-national variation will affect both the estimate and its associated standard error when dealing with a binary outcome. If helpful, the authors may want to look at materials on generalised estimating equations and generalised linear mixed effect models (GLMM). To make robust conclusion about the association between change in material status and illness, the clustering (country) and the longitudinal (repeated over time) nature of the data must be accounted for in the analysis. We are wondering whether this could be a misunderstanding based on our insufficient description of the applied estimator. All models in this paper are longitudinal and estimated with the xtlogit command in Stata. To control for differences between individuals and countries, we use the fixed effects model option (, fe ). This estimator utilises the variation within individuals over time, but not the cross-sectional variation, and it therefore indicates the change in risk of LLSI associated with moving into material deprivation. The description of this estimator is now improved. Lines 204-216, pages 10-11. We have also estimated binominal generalized linear mixed-effects models (GLMM), and show them in appendix (Lines 232-236, page 11). In contrast to the longitudinal fixed effects model, the GLMM does not control for time-invariant factors at the individual and country level. This implies that selection, i.e. people with LLSI moving into material deprivation, is impossible to separate from causation, estimated as the change in LLSI associated with moving into material deprivation. This challenge could possi ly e handled using a hy rid approach (also known as the Mundlak or within- etween approach), i.e. split the explanatory varia le into individual mean (x i) and time-specific deviation from individual mean (xij - x i) (Allison, 2009; Bell & Jones, 2014; Debarsy, 2012; Dieleman et al., 2014; Greene, 2012, Sjölander, Lichtenstein,
Larsson, & Pawitan, 2013; Snijders & Bosker, 2012). Nevertheless, in this case, that would be a very complicated way of running a longitudinal individual fixed effects model, and we have therefore considered it most appropriate to retain the simplest way of estimating the individual fixed effects model. * On page 9 line 17, the authors should justify why they divided age squared by 1000, but not age. Both age and age squared are part of a single continuous function. age squared is no longer divided by 1000 (Line 222, page 11). * On page 9 line 37, the authors should clarify the error term (the remaining unexplained variation) in their model formulation. It is unusual for Bernoulli and binomial distributions on which logistic regression is based to have error term. We agree, and have adjusted the equation according to Allison (2009). Lines 210-211, page 10. * In Table 1, the authors should summarise binary data using p(n) instead of mean(sd). We have changed the table according to this comment. * To facilitate understanding of the results, the authors should also present a frequency table (including proportion or percentage) showing the possible combination of "cannot afford a car" across the three years. Other form of presentation may also be considered.
We are now presenting the suggested frequency table (Table 2 and lines 246-248, page 13) * The authors may want to present OR (95% CI) instead of OR (SE) in the abstract. We have changed the abstract according to this comment. Lines 43-48, page 2.