Health and Consumption Patterns
M.J.W. van Megen
University of Groningen
Master thesis Econometrics
Health and Consumption Patterns
By M.J.W. van Megen
Abstract
In this paper, the relation between consumption and health is closely examined. A unique Dutch dataset providing detailed information on consumption is used to uncover changing patterns as a result of deteriorating health. Turbulent years in the organisation of the Dutch health care system underline the importance and relevance of studying these effects. The lack of literature in this field calls for an exploratory study of what effects may be expected, particularly useful for future hypothesis testing. The main findings in this paper include a drop in total non-durable spending following a negative health shock of up to 300 Euros per month; clearly identified substitute and complement goods and services to bad health; implications of recent reforms on the allocation of budget.
1 Introduction
This paper aims at providing insight in consumption (or rather expenditure) and health patterns in the Dutch elderly population. Moreover, it takes a first look at the changes brought about by recent reforms in the Dutch health care system using a before-after strategy. In particular, interest lies in the economical concept of marginal utility of consumption and its health state dependence. We investigate both for aggregated non-durable consumption as well as several consumption categories that our unique dataset provides us with the relation with health. Research on the relation between health and the spending categories is done with the well-known Almost Ideal specification of the demand system developed by Deaton and Muellbauer (1980), allowing us to observe shifts of allocation within the budget resulting from health shocks. Notable results of this paper reveal clear complements and substitutes
to poor health. Moreover, we find that health-related consumption dynamics are more
explicitly present at ages before pension. We further find that increased medical spending in 2015 relative to the years 2009-2012 is compensated by decreased spending in goods that are complement to bad health. As these findings indicate, this paper is exploratory of nature. The value of this study lies in the poor investigated effects of health on consumption and the relevance thereof at this time, addressed in the remainder of this section.
resulting in, among other things, less compensation for physiotherapy and seeing the psy-chologist for mental health problems. Then in 2013, mandatory deductibles were increased significantly, from 220 Euros per month in 2012 to a monthly 350 Euros in 2013. In that same year, walking aids were no longer covered in the basic health care package. Finally, since 2013, the Dutch government aims at more extramural long-term health care. This goal to have elderly live longer in their own homes is especially reinforced by the decision to carry over responsibility of long-term care to municipalities as of 2015. Among other things, accompaniment, daytime activities, and informal care are no longer generally financed, but may now differ from one municipality to another.
The question that arises is whether the elderly part of the population is able to handle the implications of these recent reforms when confronted with a decrease in health. In particular, a result of the reforms is an increase in out-of-pocket medical expenditures. Whether this affects non-medical spending patterns depends in part on shifting preferences associated with a change in health. Assuming that there exists such a shift, Figure 1 will help to
further clarify the research done in this paper. Following the figure, we expect that a
change in health affects both the budget (through necessary medical expenditures) and the consumption preferences of people. Consequently, there is less choice in consumption, but at the same time preferred consumption choices change. Affected by recent reforms is the budget through higher necessary medical expenditures, while we assume preferences to be unaffected. Health effects on consumption patterns before recent reforms, say in the period 2009-2012, will be considered revealed preferences, while changes in these patterns after the reforms will be considered consequences thereof.
Figure 1: The consequences of changing health to well-being Health
Budget
Preferences
Well-being
The claim of higher medical expenditures in 2015 for the elderly population is supported by the data. In Figure 2, medical spending, conditional on people having medical expendi-tures, is plotted for two age groups (65-74, 75-85) in the four years for which we have the data. Note that people were asked to report the amount they monthly spend on ‘medical care and health costs NOT covered by insurance’. (More on the consumption questions and data in Section 3.) The figure clearly shows that for both age groups the mean, median, and IQR of (conditional) medical spending have significantly increased in 2015, while staying relatively constant in the years before.
Figure 2: Medical spending (real, 2010) of two age groups in four different years 0 50 100 150 Medische uitgaven 2009 2010 2012 2015 Jaar 65-74 0 50 100 150 Medische uitgaven 2009 2010 2012 2015 Jaar 75-85 IQR Mean Median
trips and holidays. He may now be unable to drive his car and thus saves money on trans-port. But if these consequential cuts do not provide enough to pay the medical bills, what will he save on next? Possibly, an undesirable effect would be if the elderly population starts saving on food or even heating, for example. Ambiguity lies of course in all other possible examples. Possibly, this person’s mobility may cause him to spend more on holidays through comfortable travelling. Instead of the car, he may call a cab raising costs on transport. In general though, we may expect to find some logical shifts in the spending categories that might point us in the direction of possibly undesirable effects. Concretely, how do consump-tion patterns change with health? What spending categories are complement and which are substitute to deteriorating health? Do we see notable and significant differences after recent reforms? If so, are these differences possibly alarming? These are the questions to which we seek an answer in this paper.
The remainder of this paper is organised as follows. We examine related literature in Section 2. In Section 3, we closely examine the consumption and health data at hand. Section 4 presents the model we use to estimate the parameters of interest. Results of the estimation are presented and discussed in Section 5. Finally, we conclude, evaluate our methods, and give directions for future research in the last section.
2 Link to the literature
of deteriorating health. Willingness to pay is therefore a clear indication of the health state dependence of the utility function. A second suggested approach is to make use of utility proxies such as subjective well-being. One may then infer how this proxy changes with health and how this varies for households with different consumption levels. This approach is implemented by Finkelstein, Luttmer, and Notowidigdo (2012) using permanent income as a proxy for consumption and they find a substantial negative effect on the marginal utility of consumption in case of deteriorating health. A third approach is to examine consumption profiles and changes associated with health directly. We may expect individuals to consume more when marginal utility is high and less when it is low. Observationally equal individuals of which one experiences an unexpected health shock should allow for inference about the health state dependence through changing consumption patterns. Lillard and Weiss (1997) make use of panel data on consumption, inferred from income flows and asset changes, and find a substantial positive effect on marginal utility of consumption following this approach. This approach may, however, be sensitive to what is assumed about the bequest motive. For example, a decline in consumption following an unexpected negative health shock may be evidence for decreasing marginal utility of consumption in poor health, but it may also be contributed to a strategic bequest motive, i.e., raising bequests to induce informal care from close ones.
In this paper, we examine patterns both in specific consumption categories and in total consumption. Health state dependence of the marginal utility of these categories will im-plicitly be subject of study. Making use of our panel data on consumption and not concern ourselves with the issues of utility proxies, we move along the lines of the last mentioned approach.
Furthermore, we base our approach largely on Banks, Blundell, Levell, and Smith (2015). This recent paper of Banks et al. (2015) gets important insights regarding the relation between consumption and medical expenditure. Comparing non-durable expenditures in the US and the UK between 1988 and 2009, they address the huge difference in the consumption profiles of households between these two countries. In fact, they note an average rate of decline in spending for non-durable goods between the age of 45 and 75 of 1% for the US and over 3% for the UK. Closely examining all factors that could possibly explain the difference, they find that about a quarter of the gap can be attributed to medical expenditures, which are considerably higher in the US. The rest of the gap is filled by the uncertainty that households have about future medical expenditures. US households are inclined to save much more, hence consume less, in their earlier years for precautionary reasons. Consequently, at older ages, they have much more left to spend as most of this saving turns out superfluous. Roughly speaking, one could say the Dutch system is moving away from the UK system and closer to that of the US. Policy makers should therefore expect to see more precautionary saving in the future.
3 Data
updated background variables of all members of the household. Moreover, panel members complete online questionnaires of which we consider two for our research purposes: one on health, yearly available from 2007 until 2013 and in 2015, and one on time use and con-sumption which has been issued in 2009, 2010, 2012 and 2015. All members of participating
households aged 16 or older were asked to fill in the questionnaires.1
The years we select are the same as in which we have data on consumption, 2009, 2010, 2012 and 2015. We limit our focus to single households (i.e., households without cohabitant partners) aged between 55 and 85. The reason for limiting analysis to singles is that analysis of households with spouses comes with various practical considerations. For example, we would require a measure for household health, not trivially available. In addition, differences in reporting of household consumption should then be handled. We therefore decide to limit focus on single households. Furthermore, a conservative method of defining outliers was used. Outliers were found in response to questions on income as well as consumption. We end up with a sample containing 1781 observations (approximately equally) divided over the four years.
Regarding consumption of the households, the LISS data provides very detailed infor-mation on expenditures, uniquely useful for our purposes. The concept is based on survey questions on personal and household non-durable spending per month. This kind of sur-vey methods is more noisy than, for example, diary based measures. Nonetheless, it con-tains a useful signal on consumption behaviour (Browning et al., 2003). Specifically, the questionnaire asks for two types of consumption: non-durable goods and services that are non-assignable within the household and non-durables that are assignable to the members of the household. A part of the questions and the consumption categories are listed in Table 1. These will generally be of interest to us and the names of the categories will appear as variables, albeit abbreviated, throughout the paper.
We examine the data more closely in the remainder of this section, starting with data on health. We choose to use the (subjective) self-reported health measure derived from the question ‘How would you describe your health, generally speaking?’. Respondents answer on a 5-point scale ranging from poor to excellent with moderate, good, and very good in between. Response density to this question for three different age groups, namely ages 55-64, 65-74, and 75-85, is captured in Figure 3. Expected differences between age groups are found; older people report relatively worse health and differences are larger for the oldest group. Notably, people rarely report ‘poor’ or ‘excellent’ health. Moreover, qualitatively, ‘good’, ‘very good’ and ‘excellent’ are contrasted by ‘poor’ and ‘moderate’, leading to the decision to take them together and divide them into good and bad health, respectively.
Next, we closely examine consumption data. First, note that most people either pay mortgage or they pay rent, so in order to have a category ‘housing costs’, we would need, for example, a measure for imputed rent. Imputed rent is the rent-equivalent that home owners would pay for a house with similar characteristics to the one they own. Accompa-nied with computing such a measure are mortgage interests, mortgage redemption, value of the house, tax, tax relief, etc. We unfortunately lack the proper data to come up with
Table 1: List of consumption categories elicited in questionnaire
‘Can you indicate for each type of expenditure how many euros your household spends on this on average, per month?’
Mortgage Rent
General utilities
Transport and means of transport Insurances
Children’s daycare
Alimony and financial support for children not (or no longer) living at home Debts and loans
Expenditure on cleaning the house or maintaining the garden Eating at home
Other household expenditures
‘For each type of expenditure, please indicate how many euros you spend on this personally per month, on average.’*
Food and drinks outside the house Cigarettes and other tobacco products Clothing
Personal care products and services
Medical care and health costs NOT covered by insurance Leisure time expenditures
(Further)schooling Donations and gifts Other
Figure 3: Self-reported health measure response in three age groups 0 .2 .4 .6 Fraction poor moderate good
very good excellent Self-reported health
55-64 65-74 75-85
Health measure by age group
this measure. With regard to Table 1, mortgage and rent were therefore not taken into ac-count. Furthermore, the household categories children’s daycare, alimony, debts, and ‘other’ were also excluded as their magnitude is negligible in our sample. The personal expenditure categories were aggregated across categories in all years so to be most similar to the mea-sure we have in 2015, i.e., with the exception of the medical expenditure category which is separately available in all four years. We end up with seven spending categories for which we provide summary statistics in Table 2. For reference, we also include total non-durable spending (sometimes abbreviated by ndbs), defined as the sum of the seven categories, and net income. Finally, we include the mean percentage of total non-durable spending and of net income.
Table 2: Summary statistics of consumption data reported in real (2010) Euros per month
Consumption category Mean SD p25 p50 p75 % of ndbs % of income
Utilities 188 90 132 178 229 19.9 12.4 Transport 93 101 28 64 122 8.6 5.4 Insurances 180 127 120 165 213 18.2 11.6 Maintenance2 40 63 0 18 51 3.5 2.2 Food at home 227 140 138 200 300 22.6 14.3 Personal 278 378 92 185 349 23.4 16.4 Medical 46 149 0 20 45 3.9 2.9 Total ndbs 1051 573 700 945 1262 - -Net income 1738 796 1160 1590 2100 -
Table 2 reveals that the biggest spending categories are on utilities, insurances, food at home, and the aggregated personal spending. Furthermore, categories transport, mainte-nance, personal consumption, and medical expenditure show a high standard deviation and a mean well above the median, indicating small parts of the sample having high expenses on these categories. Finally, notice that for both the consumption category maintenance as well as medical, there is at least 25% (but less than 50%) of the population that indicates not having any monthly expenses on that category. The observations done here will be kept in mind when interpreting the results.
4 Model
The theoretical model we employ is the well-known Almost Ideal Demand System (AIDS) developed by Deaton and Muellbauer (1980). This system allows us to examine in detail how consumption varies with demographics, and in particular health. Specifically, the dependent variables in this system are the consumption categories mentioned earlier to be subject of research. By design, the system shows how increased demand for one group of goods is financed by decreased demand for another group. Hence, differences in budget allocations to the specific consumption categories as a result of a change in health allows us to identify which are complement and which are substitute to bad health. Moreover, it allows us to examine and compare these health-related shifts in the before-reforms period (roughly 2009-2012) and the after-reform period (represented by the data available in 2015). We write the model as: wkht = βk0 + β 1 kzht+ βk2ln xht Pt + βk3 lnxht Pt 2 ,
where wkht is the budget share of household h at time t for good/service k, zht includes
all to-be-specified demographic variables of household h at time t, xht is total non-durable
spending on all goods/services included in the demand system which is deflated by Pt which
is the consumer price index at time t. This model is almost equal to the earlier mentioned AIDS model under the assumption that relative prices have been constant over time. It then still differs slightly through the addition of a quadratic term on total expenditure (though falls short of the fully integrable QUAIDS model (Banks, Blundell, and Lewbel, 1997)). The model is finalized by the inclusion of the following parameter restrictions ensuring that the budget shares sum to one:
K X k=1 βk0 = 1, K X k=1 βki = 0, ∀i ∈ {1, 2, 3}.
It is important to note that we use a substantially simplified version of the AIDS/QUAIDS model. The exact implications are hard to pin down, but not considered harmful for the purpose of this study which is exploratory of nature.
5 Results
The main topic in this paper is the relation between consumption and health. Our unique dataset allows us to closely examine shifts within the budget from one category to another. In this section, we will fully exploit this feature and gain interesting insights in the within budget dynamics. To start off with, however, we relate back to the earlier literature and examine what happens to total consumption in case of a health shock, possibly allowing us to make inferences about the health state dependence of the marginal utility of consumption.
5.1 Health state dependence of total consumption
We have run fixed effects regressions to see how a health shock (i.e., the transition from the good health state to the bad or vice versa) affects total non-durable spending. Moreover, to put these findings in perspective, we have done the same for net income. In the regressions, we control for cohort effects and household composition of which the effects are allowed to vary with age. Health effects are also allowed to differ with age, allowing us to construct the consumption-age profile for different health states and the income-age
profile as shown in Figure 4.3 The interpretation of the figures is that a health shock at a
given age causes a vertical change along the y-axis towards the line that represents the new health state. Overlapping confidence intervals of the two health states around the mean non-durable spending or income at a certain age means that there is no statistically significant difference between the two health states.
In Figure 4, in the left panel, we observe decreasing non-durable spending with age in good health until the age of 70, after which it appears to be at a constant level or even slightly increasing level (with large confidence bands). We see a similar pattern in case of bad health, but there are statistically significant differences between the two health states for singles until around age 70. The difference between means is approximately 200-300 Euros per month. To draw conclusions from these observations, we first look at the right panel in Figure 4. We see that for healthy singles, net income decreases only slightly, most likely as more and more people stop working (but with a high replacement rate). A transition in health state has remarkable effects on net income of singles. For ages between 45 and 60, we observe a drop in net income of approximately 600 Euros per month. This gap becomes smaller until there is no statistically significant difference at around the age of 70. Hence, the drop in consumption due to deteriorating health may on the one hand indicate a decrease in marginal utility of consumption, but on the other hand it may be necessary cutbacks due to a more restricting budget constraint. We are therefore unable to identify the sign of the possible health state dependence of marginal utility of consumption.
5.2 Effects of health on disaggregated consumption
The model as defined in Section 4 was estimated through three-stage least squares; a com-bination of IV regression and Seemingly Unrelated Regression, where the total non-durable
spending variable xht is taken endogenous and is instrumented by net income. Furthermore,
3Note that, although the focus of our analysis is on households of age 55-85, we have broadened this range
Figure 4: Age profiles of total ndbs (left) and of net income (right), both in real terms (2010) for the period 2009-2012
200 600 1000 1400 1800 Total ndbs 40 50 60 70 80 90 Age Bad health 95% CI Good health 95% CI 200 600 1000 1400 1800 Net income 40 50 60 70 80 90 Age Bad health 95% CI Good health 95% CI
the vector of demographic variables zht includes health, three age group dummies, the
num-ber of living-at-home children, education dummies, a dummy for whether household head still has a paid job, a dummy for home ownership, and time dummies. In addition, as we expect more rigorous shifts in budget shares after recent reforms, we interact health with a 2015-dummy and present estimates on 2009-2012 health effects only in Figure 5. The figure shows the mean shift in the budget share of the respective consumption category together with its 95% confidence interval.
The interpretation in Figure 5 is as follows. For the consumption category utilities, for example, we observe that the transition from good to bad health (the so-called bad health shock) results in a, ceteris paribus, 1% increase in the allocation of the budget to this category, statistically significant at the 5% level. Hence, if the mean budget share of utilities is, say, 18% for healthy singles in our dataset, then it is 19% for singles in bad health. A similar effect is found for the category insurances, which notably, includes health insurance. Moreover, for categories transport and food at home we find opposite effects. Singles tend to allocate less budget to these categories in case of deteriorating health. The same holds true for the personal consumption category, but the effect turns out even stronger; a 2% decrease in budget allocation. Finally, the two smallest categories (with reference to Table 2), including medical expenditure and maintenance in and around the house, are observed to be complement goods to bad health. 1% increases in allocation are found to these categories, which can be considered an economically significant increase relative to their levels in good health. In conclusion, complement to bad health, we find the categories utilities, insurances, medical expenditure, and maintenance in and around the house. Substitute to bad health we consider categories transport, food at home, and personal consumption.
5.3 Effects of health on disaggregated consumption for different age groups
Figure 5: Bad health shock effects for singles between 2009 and 2012
6, which has similar interpretation as Figure 5, but displays health effects per consumption category for each of the three age groups. We are interested in this segregated age-health effect because we may expect notable differences between age groups.
groups, all statistically significant and an approximate 1% increase in budget allocation in case of deteriorating health. Finally, personal consumption is cut back on by singles in the first two age groups, while there is no significant (neither statistical nor economical) difference for singles in the oldest age group experiencing a bad health shock.
Figure 6: Bad health shock effects for singles between 2009 and 2012 for different age groups
5.4 Health effects before and after reforms
1% in 2015 relative to the other period. Moreover, this increased allocation results in de-creasing expenses to consumption groups that are complement to bad health. The difference between mean health effects for substitute groups in both periods is much smaller. Hence, we carefully conclude that the additional medical expenses that are a result of reforms in the years 2013-2015 are financed by cutbacks on goods complement to bad health.
Figure 7: Shifts in budget shares of singles for aggregated groups in and before 2015
6 Conclusion
In this paper, we have explored how health affects consumption. In particular, we have examined single households ranging from age 55 to 85. Using detailed data on consumption patterns of these households and a question on general health, we have gained insights in the health state dependence of the total consumption level as well as of levels of specific consumption categories. Moreover, we have uncovered these health effects for different age groups.
Furthermore, we find shifts in the budget allocation to be (close to) statistical signifi-cance at the 5% level for all seven considered consumption categories, i.e., utilities, insur-ances, transport, medical, food at home, maintenance, and personal in the years 2009-2012. Complement to bad health, all with approximately a 1% increase in budget allocation to the respective category in the event of a bad health shock, were utilities, insurances, medical, and maintenance. The other three categories were found to be substitutes to bad health with a decreasing budget allocation of between 1 and 2%. The magnitude of these changes relative to the levels in good health are found to be big for the small categories medical and maintenance; moderate for the category transport; and small for the other four larger categories.
Considering these same effects for three different age groups (55-64, 65-74, and 75-85), we find most statistically (and economically) significant effects for the youngest group. This ap-pears to be in line with earlier findings where we looked at total consumption. Furthermore, patterns of complements and substitutes hold for the better part for all three age groups. Differences between age groups could mostly be logically explained.
We, furthermore, took a first look at effects of the recent (2013-2015) reforms. The expected result of these reforms were an increased budget allocation to (necessary) medical spending in case of deteriorating health. Indeed we find, at least qualitatively, a higher increase in medical spending in 2015 relative to 2009-2012 following a bad health shock. The source of financing this increase appears to be the group that constitutes as complements (aside from the medical category). The interpretation is that, since recent reforms, single households save money on goods for which their marginal utility has increased following a negative health shock in order to handle the increase in necessary medical expenditure.
The contribution of this paper is better insights into the effects of health on consumption patterns. Existing literature on health-state dependence of the marginal utility of (total) consumption comes closest to what is investigated here, but is unable to predict policy implications regarding the health care system. A closer look into more specific types of consumption is needed to be able to evaluate implications for well-being. Recent turbulence in the Dutch health care system requires such evaluation. The lack of an obvious control group restrains from clean deduction of policy implications, but evidence supporting expected changes in medical spending indicates that a useful signal was found with the before-after strategy.
References
Banks, J., R. Blundell, P. Levell, and J.P. Smith (2015). Life-cycle consumption patterns at older ages in the us and the uk: can medical expenditures explain the difference? IFS Working Paper W15/12.
Banks, James, Richard Blundell, and Arthur Lewbel (1997). Quadratic engel curves and consumer demand. Review of Economics and Statistics 79, 527–539.
Browning, M., T.F. Crossley, and G. Weber (2003). Asking consumption questions in general purpose surveys. The Economic Journal 113, 540–567.
Deaton, Angus and John Muellbauer (1980). An almost ideal demand system. American Economic Review 70 (3), 312–26.
Finkelstein, A., E.F.P. Luttmer, and M.J. Notowidigdo (2009). Approaches to estimating the health state dependence of the utility function. American Economic Review 99, 116–121. Finkelstein, A., E.F.P. Luttmer, and M.J. Notowidigdo (2012). What good is wealth without health? the effect of health on the marginal utility of consumption. Journal of the European Economic Association 11, 221–258.
