### Discussion Paper

No. 2018-25 |
March 12, 2018

Can heterogeneity in reporting behavior explain the gender gap in self-assessed health status?

## Abstract

This paper explains the gender differences in self-assessed health status by providing a theoretical identification mechanism through a dynamic structural model which allows for heterogeneity in discount factors of individuals. Theoretical predictions are empirically tested and estimation results support the structural model implications. The authors conclude that accounting for heterogeneity in individual discount factors explains a significant portion of the gender gap in self-assessed health status.

## Comments and Questions

This is an interesting and thought provoking research question especially in light of the fact that HRQol are often used to determine cost-effectiveness for drugs that results in reimbursement decisions.

A few questions: In the discussion piece, the authors refer to the alternative Fuchs (1982) approach in explaining the difference ...[more]

... – is there any more recent literature that supports this different pathway?

In the results, the results are not clearly explained – e.g., interpretation of the cut-off table. Also, since Models 2 and 3 do not seem to show statistical significance, how do the authors interpret these findings on the interaction terms?

Also, how do the authors explain the coefficients magnitude as age decrease and education increase?

It would also be useful to get a more detailed analysis of the survey responders and the choice of the smoking variable.

Thank you for your comment. Below are our responses to your comments raised:

1)Of course there are more recent work in the literature on the same line, as some of them are referenced below. Here the emphasis is on the fact that Fuchs was the pioneer of the alternative approach ...[more]

... which in fact does not contradict with the association of education and health outcomes. It would not be incorrect to say that with Fuchs, the tradition of seeing the education and health outcomes in a “causal framework” started. In this respect, contrary to the previous research which emphasize the effect of education on health outcomes, this approach sees the higher level of health status as a result of time preference rather than having higher educational attainment. In other words, in this time preference approach, an individual spending a greater proportion of her time for education will also allocate more resources to health.

References:

Fuchs, V.R., 2004. Reflections on the socio-economic correlates of health. Journal of Health Economics 23 (4), 653–661.

Tenn S, Herman D, Wendling B (2010) The role of education in the production of health: An empirical analysis of smoking behavior. Journal of Health Economics 29(3):404–417.

Brown, H., and O. Biosca. "Exploring the relationship between time preference, body fatness, and educational attainment." Social science & medicine 158 (2016): 75-85.

2)Interpretation of the cut-off table is not explicitly made since as standard in this sort of estimation with ordered choices, the 4 cut-off values correspond to the points on the real line where normal distribution is separated into 5 regions. Each SAH outcome is the corresponding region under the normal curve. This is a standard identification assumption with a structure presented with equations 12 and 13.

We could not exactly understand what is meant by Model 2 and Model 3. Just assuming it is referred to Model 1 and Model 2, this time we could not understand what is meant by these models do not show statistical significance. The Model 1 and Model 2 are empirical counterparts of the theoretical estimation equation derived from the lifecycle model. This is equation 12 in the paper. However, we used dummy variables (instead of continuous variables used in the theoretical derivation) for some of the major variables such as education, age and income levels. As explained in Section 4.3, the estimation requires proxy z variables constructed from the demographic variables and the measure for the discount rates. Model 1 and Model 2 are the two empirical specifications proxying equation 12 where Model 1 is constructed with less variables by mainly ignoring the interaction terms that would arise while constructing the proxy z’s other than the plain pairwise interactions. However, Model 2 covers all the interactions that would arise when constructing z proxies. Therefore, neither the choice of the interaction terms nor the insignificance of them are of major concern, since the ultimate aim is to find support for the life-cycle health hypothesis. However still we could report and discuss the effects of interactions terms.

3)Reference age group is the ages between 25 and 34. Therefore the rest of the age coefficients should be interpreted with reference to this group. What we observe is as age decreases, individuals report better health outcomes on average as the negative coefficients on the age groups indicate. The incidence of worse SAH reporting with respect to the reference age group of 25-34 increases as age increases which we can see from the larger (in negative terms) coefficients of higher age groups. Similarly, for the education, no education is the reference group and is omitted. All the other education levels should be interpreted with reference to this group. In this respect we see individuals report better SAH outcomes as their education increases with respect to reference group of no education. For instance, for tertiary education, the effect is 0.417 whereas for high school it is 0.189 with respect to no education group. Education and age are reported to have the similar effects in the literature. Namely people face more health problems as they age and more educated people for various reasons take better care of their health.

4)A more detailed analysis of the survey respondents and the choice of the smoking variable will be provided in the revised version of the paper.

This paper asks and argues a really important question. Gender differences in individual behaviour and health are widely observed and the paper examine this relationship.

I think the paper adds important insights to the literature and well described. Further the method is easily applicable to different settings and ...[more]

... hence there is a potential for deep impact future studies.

Just one question: I understand the reasoning behind the use of smoking behaviour as a factor to identfy the discount rates. Is there other potential factors that can be used and do you think changing the factor would drastically change the results?

Thank you for your comments. Even though there were not many alternatives, as you have mentioned we think that smoking behavior is a good proxy to identify the discount rates. Therefore, we do not think that any other choice of factor will drastically change the results but it can be ...[more]

... a good practice for future research if and when data is available.

see attached file

Thank you for your comments. See attached file

see attached file

See attached file.

See attached file.

see attached file

Report of “Can heterogeneity in reporting behavior explain the gender gap in self-assessed health status”.

Women generally report to be in worse health than their male counterparts, despite the fact that women live longer than men. This paper argues that differences in reporting behavior may explain this gender gap. ...[more]

... To this end they specify and a health production model and show that within this model gender differences can be explained by differences in discount rates. The author(s) subsequently proxy the discount factors by a set of socio-economic variables and exploit non-linearities to identify the effect of the discount rate on health.

I value the approach taken by the author(s) to explicitly use economic theory to approach problems in applied health economics. However, I also see some major problems with the current paper.

The most important problem that I have with this approach is that ultimately the authors estimate a health production model, with (proxies for ) discount factors included as regressors and as such the model explains differences in SAH that can be true health differences (let’s label these as HT) as well as reporting differences (let’s label these as HR). Indeed, in theory, the discount factor may explain health investment behavior and influences life style choices such as smoking and subsequently health, but this is the health production part (HT). The observed health in the empirical model (SAH) is the joint effect of HT and HR and right now the authors do not disentangle these two components.

Furthermore, indeed, it is difficult to include the discount factor in the empirical model if one does not observe it in the survey. Smoking is a derived outcome, but is also a poor proxy of the underlying discount factor as other factors like risk attitude and contextual factors may explain smoking outcomes. Of course the other individual characteristics can be included to make the proxy better, but ultimately there is much noise there. So, the authors have to convince the reader that this is not a problem in the current context. By the way, in the proxy income plays an important role, but isn’t this an important endogenous variable? After all, in the context of a health/human capital production model, the discount factor influences investment decisions in the labor market as well and therefore also income (rather than the other way around).

See attached file.