*Student number: 2938693, e-mail: woutermarsman@gmail.com
**Course code: EBM877A20, ***University of Groningen, Netherlands
Ex-ante moral hazard effects: the influence of the Dutch voluntary
deductible on lifestyle. Examining excessive smokers and drinkers
Abstract
This research evaluates the existence of ex-ante moral hazard effects around the uptake of the voluntary deductible in the Dutch healthcare system. By investigating individual decisions concerning the voluntary deductible, smoking, and alcohol use, the influence of the voluntary deductible on the likelihood of the latter two is estimated. To do so, this thesis makes use of data from the LISS panel for the years 2009-2019. The examined sample comprises 6.587 observations, dispersed over 1.171 individuals. The model employed is a univariate probit model that controls for individual socio-demographics, risk preferences and risk type. The coefficient that captures the
ex-ante moral hazard effect is insignificant. Opting for the voluntary deductible does not influence the probability
to be categorized as an excessive smoker/drinker or stopped excessive smoker/drinker for the average individual. The findings are robust to the exclusion of a risk aversion parameter that considerably increased the sample size. The Dutch healthcare system does not have to be changed to accommodate for the existence of ex-ante moral hazard effects.
Key words: ex-ante moral hazard, voluntary deductible, excessive lifestyle decisions JEL codes: I12, I13, I14, I18
Author: Wouter A. Marsman* Supervisor: Dr. L. Viluma
Date: January 08, 2021
1
1 Introduction
Insurance markets emerge when there is demand to mitigate the consequences of certain behavior. For insurance markets to last, various assumptions need to be satisfied. Even long-lasting insurance markets are imperfect as they are often hampered by the consequences of consumer behavior. Behavioral economics can explain these market imperfections. Moral hazard effects are one of the causes for imperfect insurance markets. Moral hazard – the loss-increasing behavior that arises under insurance – is an undisputed concept among economists that has been empirically proven in diverse situations. There are two types of moral hazard effects; ex-ante and ex-post, of which the literature on the former is relatively scarce and inconsistent. Ex-ante moral hazard effects therefore allow for interesting research.
This thesis will evaluate the existence of ex-ante moral hazard effects, i.e., the reduction in preventive effort as a consequence of insurance, in the Dutch healthcare market (Zweifel and Manning, 2000). In this setting, ex-ante moral hazard effects will be evaluated for individuals who adhere to excessive and unhealthy lifestyles concerning smoking and alcohol use.
From 2008 onwards, the Netherlands have introduced mandatory and voluntary deductibles in their healthcare system. Individuals have a mandatory deductible of 385 euros in 2020 – the so-called “own-risk” – and can opt for a maximum voluntary deductible of 500 euros, effectively increasing the overall deductible to a maximum of 885 euros. The deductibles were introduced to limit the overconsumption of healthcare, as individuals are more likely to opt for the most efficient healthcare when healthcare expenses are out-of-pocket. The Dutch healthcare market has been popular among researchers due to the availability of data and the design of the Dutch healthcare system, which lend it suited for research purposes. Ex-post moral hazard has already been empirically proven for the Dutch setting (Remmerswaal, Boone and Douven, 2019; Alessie, Angelini, Mierau and Viluma, 2020). Research on ex-ante moral hazard in the Dutch healthcare market is scarce but does exist. Zwemstra (2019)’ analysis of the Dutch healthcare market found significant ex-ante moral hazard effects by establishing an instrumental variable probit model, using an elaborate set of controls.
2
deductible influences lifestyle improvements will be answered. To answer these questions, this thesis makes use of data from Longitudinal Internet Studies for the Social sciences (LISS) panel for the years 2009-2019. The panel contains necessary information on individual lifestyle- and insurance decisions.
Evaluating reasons for lifestyle decisions regarding smoking and alcohol use is highly interesting, as the health consequences of both are stark. Excessive smoking and excessive use of alcohol are important contributors to global healthcare expenditures, and it is therefore essential that the prevalence of both is reduced. Governments already divert increasing attention towards the health of its population. The National Prevention Pact of the Netherlands (2019), which stipulates a route to healthier generation is a good example of an incentive aimed to reduce the prevalence of both smoking and excessive alcohol use.
The contribution of the analysis of the voluntary deductible in the Dutch healthcare system is twofold. Firstly, the univariate probit model employed in this thesis allows, at least for parts, for the isolation of ex-ante moral hazard effects by controlling for other factors that influence excessive lifestyle decisions. The remaining relationship between the voluntary deductible and excessive lifestyle decisions that captures the moral hazard effect is insignificant, indicating no robust presence of ex-ante moral hazard. In other words, opting for the voluntary deductible does not incentivize, or disincentivize excessive lifestyle decisions. The fortunate result is that the Dutch healthcare system is not flawed in this sense. The design of the healthcare system does not have to be changed to accommodate for the occurrence of
ex-ante moral hazard effects.
3
This paper proceeds as follows. Section 2 will discuss the relevant literature. Section 3 will briefly explain the regulatory framework of the Dutch healthcare system. Section 4 will provide a general description of the used data. Section 5 provides the theoretical framework, section 6 provides the empirical framework. Section 7 will describe the data and section 8 shows the results of the employed model, with a robustness check. Section 9 will address the limitations of the study and section 10 will provide policy recommendations based on the descriptive statistics and the results. Section 11 concludes.
2 Literature review
Health insurance markets have received ample attention among economists over the past decades, starting with the early works of Arrow (1963) and Pauly (1968). This section serves as an overview of the existing literature regarding the societal burden of smoking and excessive alcohol use and will provide an overview of the existing literature on health insurance, paying special attention to ex-ante moral hazard. Moreover, I will discuss the popular solution put in place in the Dutch healthcare system intended to limit moral hazard.
2.1 Societal burden of smoking and excessive alcohol use
4
these additional years which may exceed the quantity and quality of the care received by earlier diseased smokers.
2.2 Early works on health insurance
Arrow (1963), in his work, examines the competitiveness of the health insurance market. Comparing characteristics of a competitive market in optimal state with the market for health insurance, Arrow (1963) is able to blueprint an ideal insurance market. In an ideal insurance market two assumptions need to be satisfied. Firstly, individuals need to maximize expected utility. Utility is a derivate from income and the cost of medical care should be seen as a negative and random income shock. Secondly, individuals should be risk averse. That is, individuals do not gamble with their health. Individuals, when given the option between a certain income stream m and a probability distribution with mean m, opt for certainty. Optimally, insurance companies offer actuarially fair premiums, and the law of large numbers makes sure there is no welfare loss. However, the law of large numbers is not satisfied as there are only so many potential individuals in a health insurance market, meaning welfare losses occur. Arrow (1963) proposes that this welfare loss can be offset by offering premiums slightly higher than the actuarially fair level. Since individuals are risk averse, they should still favor a just actuarially unfair premium, as opposed to bearing the entire health risk themselves, as long as the premium is not too far off from the actuarially fair premium. Unfortunately, this ideal insurance market is unrealistic. Not all individuals are able to forecast their expected income correctly and attach the right price to medical care. Furthermore, not all individuals are risk averse to the same extent.
Commenting on the work of Arrow (1963), Pauly (1968) points out an additional pre-condition for the existence of an insurance market. Pauly (1968) states that due to heterogeneity in a population, it is optimal for some medical expenses to be non-insurable. Insurance can only exist for those goods or services for which the demand is almost perfectly inelastic with respect to the market price. Insurance markets often fail to sustain whenever this pre-condition is not nearly satisfied. Furthermore, the occurrence of the event should be random, and individuals should be risk averse towards the event in order for insurance markets to sustain.
5
taken by economists as the “golden standard” (Aron-Dine, Einav and Finkelstein, 2013). Later research has found that medical care price elasticities, under different circumstances, may even take magnitudes reaching as far as -2.3 (Kowalski, 2016). Healthcare demand does thus depend on price. Besides this, the occurrence of negative health shocks is often not completely random. Especially diseases that are tightly associated with lifestyle decisions regarding alcohol and smoking, are highly subject to the individuals own behavior. These and other issues have been the ground for the extensive literature on health insurance markets, of which I will discuss more in the subsequent sections.
2.3 Moral hazard
Another issue, inherently possessed by insurance markets, is called moral hazard. Generally, moral hazard is defined to describe the “loss-increasing behavior that arises under insurance” (Rowell and Connelly, 2012, p. 1051). Insurance markets are often born out of desire to cushion the outcome of certain behavior. Paradoxically enough, insurance itself may alter that specific behavior. Within the literature on moral hazard, it is necessary to make a clear distinction between the so-called ex-ante and ex-post moral hazard, of which the former enjoyed relatively little attention. The following sections will allow for a better understanding of the different concepts.
2.3.1 Ex-ante moral hazard
6
moral hazard, examining smoking and alcohol use often find no significant relationships. Rezayatmand, Groot, and Pavlova (2017)’ analysis of ex-ante moral hazard concerning smoking behavior in Europe, find no significant effect. In their attempt to isolate ex-ante moral hazard by means of an instrumental variable strategy, they find no conclusive evidence. That is, large part of the (positive) correlation between groups of excessive smokers and their degree of insurance coverage, can be explained by controlling for selection effects. De Preux (2011), when examining ex-ante moral hazard concerning alcohol use in the US Medicare program, also finds no statistically significant evidence for the ex-ante moral hazard case.
In line with the argument of Cook and Graham (1977), the presence of ex-ante moral hazard has been less undisputed in other insurance markets, such as the market for car insurance (Abbring, Chiappori and Zavadil, 2008; Cohen and Deheija, 2004). Cohen and Deheija (2004) find a significant 2 percent increase in car fatalities for each percentage point decrease in uninsured motorists using U.S. panel data. Intuitively, it is easier for individuals to attach a price to their motor vehicle than to their health, but this argument may be flawed. As Dave and Kaestner (2009) rightly note, the potential consequences of less self-protection because of car insurance, e.g., drunk driving, are quite severe and may result in a stark and direct negative health shock. Additionally, the consequences of lifestyle decisions on health, such as alcohol consumption and smoking, are often not immediate and take longer to accrue. By identical reasoning to Kenkel (2000), we could therefore expect higher levels of moral hazard in lifestyle decisions compared to car insurances as the utility losses of the latter event are far greater. However, empirical research is inconclusive on the presence of ex-ante moral hazard in health insurance markets. Identifying and separating ex-ante moral hazard from ex-post moral hazard has found to be challenging, although some have attempted.
2.3.2 Ex-post moral hazard
7
econometric issues often related to the observational data used in other studies. The results were conclusive, higher levels of cost-sharing declined the demand for healthcare. The argument that better insurance leads to increased medical expenditures does not necessarily hold. Einav and Finkelstein (2018) make a fair note. Insured individuals are more likely to visit the doctor, compared to uninsured individuals, as their expenses are covered. Visiting the doctor and receiving preventive care may avoid the more expensive care necessary related to postponing effective treatment. Insurance therefore does not necessarily lead to overconsumption of healthcare.
The RAND experiment is a good illustration for the ex-post moral hazard case. However, the experiment was U.S. specific and was conducted over three decades ago. The results may not necessarily hold for the Netherlands as cultural and institutional differences play a role. The subsequent section will review the evidence on moral hazard in the Dutch healthcare system.
2.3.3 Moral hazard in the Dutch healthcare system
Alessie et al. (2020) studied the effects of the voluntary deductible – an additional amount of cost-sharing on top of the mandatory deductible in return for a premium rebate – in the Dutch healthcare system. Using bivariate models, they found that opting for the voluntary deductible significantly decreases the probability that an individual visits a specialist, the GP, or is hospitalized (respectively by 31, 23 and 16 p.p.). Ergo, the voluntary deductible is an effective tool for reducing moral hazard. Empirically, it has been difficult to determine the moral hazard effect using observational data, as selection effects and heterogeneity between individuals may also provide an explanation for the increased demand in healthcare when better insured. By cleverly exploiting the design of the Dutch healthcare system, Remmerswaal et al. (2019) are able to employ a panel regression discontinuity design to disentangle the moral hazard from the selection effect. In line with Alessie et al. (2020), they find evidence that the voluntary deductible does modestly decrease total healthcare expenditures for the Netherlands, and this difference can be explained by the moral hazard effect.
8
stronger to deductible incentives and that behavioral economics play a key role in this difference. Augmenting on loss-aversion theory, they argue that individuals perceive a deductible as a loss and a no-claim refund as a foregone gain, and theory suggests that individuals respond stronger to the former (Kahneman and Tversky, 1991). Individuals perception of costs may thus cause moral hazard effects.
2.4 Selection effects
Differences in healthcare expenditures between individuals with and without the voluntary deductible can be (partly) decomposed into two effects: the moral hazard effect and the selection effect, of which the latter may dominate (Remmerswaal et al., 2019). Selection effects may occur when low-risk individuals – with low expected healthcare expenditures – opt out of additional insurance, leaving insurances companies to recoup their costs over a lower number of profitable individuals, putting upward pressure on the premium, resulting in so-called “adverse selection”. Evidence of Geoffard, Gardiol and Grandchamp (2006) for the Swiss health care market estimates that 76 percent of the correlation between healthcare expenditures and insurance coverage may be attributed to selection effects, whereas moral hazard effects only account for 24 percent.
9
suboptimal outcomes, and less asymmetry will lead to market wide Pareto-improvements (Rothschild and Stiglitz, 1970).
Educational level of individuals can also cause selection effects. According to van Winssen, van Kleef and van de Ven (2015), higher educated tend to be healthier compared to lower educated and higher educated individuals therefore profit the most from opting for the voluntary deductible. An additional argument for why higher educated individuals opt for the voluntary deductible in larger proportions is their level of financial literacy (Zwemstra, 2019). Individuals who enjoyed higher quality and quantities of education are able to make a more thorough analysis of the costs and benefits of the voluntary deductible and therefore make better use of private information.
To keep healthcare systems equitable, countries like the Netherlands, Germany and Switzerland have made health insurance mandatory. Pooling low-risk and high-risk individuals together, dampens the upward pressure on premiums caused by adverse selection effects, ensuring health insurance is still financially feasible for those who have been less lucky in the “lottery of life”. However, self-selection effects do not necessarily put upward pressure on premiums. Downwards pressure may be excelled in the case of “propitious selection” (Hemenway, 1992). Propitious selection – known in later literature as “advantageous selection” – effects occur when risk averse individuals who take more precautionary activities to self-protect, take out additional insurance because of their risk averse nature. These individuals are highly profitable for an insurer as the associated probability of a claim of such individuals is particularly low and they can therefore excel a downward pressure on the premium.
2.5 Deductibles
Insurers and governments have come up with several solutions intended to limit moral hazard such as deductibles, coinsurance, and no-claim refunds. Since the Dutch healthcare system makes use of deductibles, this section will solely discuss this solution.
10
all future claims, individuals demand significantly more healthcare (Hayen et al., 2018). A short-sighted solution would be to increase the deductible amount, but this would disproportionately foist the chronically unhealthy with larger medical bills. The Dutch healthcare system, from 2008 onwards, has made use of a mandatory deductible and an additional voluntary deductible. The benefit of the additional amount of cost-sharing when opting for the voluntary deductible, is the received premium rebate. The received premium rebate when individuals opt for the highest voluntary deductible can save roughly 300 euros in yearly insurance expenses.
3 Regulatory framework of the Dutch healthcare system
This section will discuss the unique regulatory framework of the Dutch healthcare system. From the age of 18, all Dutch inhabitants need to take out mandatory health insurance. The coverage of the mandatory health insurance is pre-determined by the government, but the packages are offered on the private market where individuals can pick their own health insurer. Since the insurance package is pre-determined, insurers can only compete on price and service. Premiums are community rated, i.e., the same premiums are charged to everyone, regardless of differences in health status. Risk-rating, i.e., charging premiums based on the individuals perceived risk type, is forbidden. By law, insurers are obliged to accept every individual. The general practitioner serves as a gatekeeper, and with referral of the practitioner the pre-determined package covers a variety of basic healthcare needs. Since an individual cannot be denied insurance, a risk-fund was set up to compensate insurers if necessary. The risk-fund spreads the risk that a specific insurance company only has accepted “high-risk individuals” – putting pressure on the insurer’ profitability – by redistributing profits of other insurers. On top of the mandatory insurance, individuals could opt for supplementary health insurance, covering a wider range of medical needs such as eye contacts, dental care, and hearing aid.
cost-11
sharing, individuals were compensated with a premium rebate. Low-risk individuals with low expected medical costs benefit the most from the premium rebate. Paradoxically, evidence from van Winssen, van Kleef and van de Ven (2016) has shown that only 11 percent of the Dutch population opted for a voluntary deductible in 2014, whereas almost 50 percent would have benefitted financially from the premium rebate. Van Winssen et al. (2016) argue that behavioral economics explain a vast amount of the observed discrepancy.
4 Data
For the purpose of my research, I will make use of data from the LISS (Longitudinal Internet studies for Social Sciences) panel. The panel comprises 5.000 households, and roughly 7.500 individuals, which are drawn from a true probability sample based on a population register of Statistics Netherlands. The panel members are paid for each completed questionnaire, and the data concerning individual health status, health insurance and lifestyle decisions is administered on annual basis. I pool together the longitudinal waves from 2009 up and until 2019, with the exception of 2014, when no health-related data are available. No use is made of data before 2009, as the voluntary deductible was only introduced in 2008 and the questionnaire of that year did not contain identical questions regarding the voluntary deductible as the subsequent years. I exclude all individuals of age 17 or younger since individuals are only obliged to be individually insured from the age of 18. After a listwise deletion of observations with missing data on the key variables, I am left with a sample that comprises 6.587 observations dispersed unevenly over 1.171 individuals.
5 Theoretical framework
5.1 Hypotheses
12
decisions. Ex-ante moral hazard effects would be less likely to occur if individuals are fully aware of the (longer-term) consequences of smoking and excessive alcohol usage and account for these consequences accordingly. The wide range of smoking and alcohol cessation incentives already suggest that individuals do not have full awareness yet. Empirical evidence on smoking behavior also suggests that considerable terrain could be gained regarding the awareness of smoking consequences among smokers (Steptoe et al., 2002, Milcarz, Polanska, Bak-Romaniszyn and Kaleta; 2018).
It will be interesting to see if ex-ante moral hazard effects can be identified from real-world data. Endogeneity within the uptake of the voluntary deductible complicates the analysis. To accommodate for endogeneity within the voluntary deductible uptake, I control for variables that influence the decision of the uptake and are related to the lifestyle behavior of the individual. In this setting, this means I control for variables that cause possible self-selection effects such as variables addressing risk type and risk preferences, much in line with Alessie et al. (2020). Individuals who expect high costs as a result of their lifestyle, self-select themselves into the for them most profitable insurance coverage, and therefore demand fuller coverage. Individuals make use of the “inside information” regarding their risk type and risk preferences when deciding on insurance. Controlling for selection effects around the uptake of the voluntary deductible isolates other relationships of the voluntary deductible. The sign and significance of the coefficient that remains after controlling for selection effects is of interest, as it captures the
ex-ante moral hazard effect. I test the following hypothesis:
1. The individual decision to uptake the voluntary deductible influences lifestyle decisions
The lifestyle decisions I test for relate to excessive smoking behavior and excessive alcohol use. The relationship is expected to be negative, as opting for the voluntary deductible, i.e., less coverage, is likely to result in more self-protection activities, and thus the decreased likelihood that an individual is categorized as an excessive smoker or drinker. In this case, self-protection activities are illustrated by individuals who adhere to healthier lifestyles. Additionally, I examine whether the uptake of the voluntary deductible influences individuals’ decisions to improve their excessive lifestyle. This results in the following hypothesis:
13
The following section provides more details on when an individual improves its lifestyle. In contrast to hypothesis 1, the relationship between the voluntary deductible uptake and lifestyle decision improvements is expected to be positive.
5.2 Variables of interest
To test my hypothesis, I will construct four response variables and one explanatory variable. Due to questionnaire format of the data, it is necessary to transform variables accordingly. The dataset allows the data to be restructured to useful binary variables.
Firstly, it is necessary to construct a variable indicating whether an individual has opted for the voluntary deductible. Individuals either opted for no voluntary deductible, or a positive voluntary deductible of 100, 200, 300, 400 or 500 euros. I will construct a binary variable that either indicates whether an individual has opted for no voluntary deductible or any amount of positive voluntary deductible. I do this because, the voluntary deductible – when introduced in 2008 – has gained considerable popularity, but only slowly over time. According to Vektis (2020), 13,3 percent of eligible Dutch population opted for a voluntary deductible in 2020, more than double the number of individuals who opted for the voluntary deductible 10 years ago. Separating the sample size based on the specific amount of voluntary deductible, would overcomplicate the analysis and considerably decrease the power of the estimated effects.
For the variables that tell us something about the lifestyle decisions of the individual, I will also construct binary variables. As for smoking, I will construct a binary variable that indicates 1 if an individual smokes cigarettes, pipe or cigars and smokes at least 20 of one of these on average per day. The threshold is determined by making use of the definition of the CBS (2017), who defined heavy smokers as individuals who smoke more than 20 cigarettes per day. I slightly adjust this definition by including individuals who smoked the exact amount of 20 cigarettes on average per day, since 20 is the average number of cigarettes in the most common package size. Excluding these individuals would mean 283 observations would not be considered heavy smokers, as they sit just on the threshold.
14
Constructing a binary variable for excessive alcohol usage of individuals is more cumbersome. Individuals who consume small volumes of alcoholic drinks on non-frequent basis are not harming their health to the same extent as more excessive users of alcohol. It is the combination of volume and frequency that defines the associated health risk (WHO, 2021). Non-frequent binge drinkers have lower associated health risks compared to frequent binge drinkers and frequent small volume drinkers have higher health risks compared to non-frequent small volume drinkers. However, it is difficult to determine whether non-frequent binge drinkers have higher health risks compared to frequent small volume drinkers. Definitions of harmful alcohol usage often used in related studies define harmful alcohol usage as a certain volume per day, covering both the volume and frequency of the alcohol consumed (Patra, Taylor and Rehm, 2009; Ezzati, Lopez, Rodgers and Murray, 2004). Unfortunately, such a variable is difficult to extract from the LISS data set. To construct a binary variable, I follow the definition of Zwemstra (2019). She constructs a variable for excessive alcohol usage using the LISS data set, that indicates 1 whenever an individual has drunk alcohol on 5 or more days per week for the last 12 months and on the day most alcohol is drunk the last week, consumed 5 or more glasses. This allows me to separate frequent small volume drinkers from frequent binge drinkers and helps to give a more accurate representation of the excessive alcohol users in the data. The argument for the 5 or more days per week frequency stems from a report by the UK parliament, that advises at least two alcohol-free days a week, an advice that should help combat the negative societal- and health consequences of excessive alcohol use (NHS, 2012). The argument for 5 drinks or more on the day an individual drank the most alcohol in the last week stems from the “4-5 rule” as introduced by Wechsler, Dowdall, Davenport and Castillo (1995). They first defined binge drinking as 4 glasses or more for females and 5 glasses or more for males.
Similar to the lifestyle improvement variable for smoking behavior, I will include a binary variable indicating 1 for those individuals who have stopped their excessive alcohol use compared to last year according to my definition.
5.3 Control variables
15
subjective control variable comprises a binary variable indicating 1 whenever an individual has assessed its own health status as good, very good or excellent. The arguably more objective control variable comprises a binary variable indicating 1 whenever an individual has reported no long-standing disease, affliction, handicap or suffers from the consequences of an accident.
16
such a way, that it makes no sense to switch from safe to risky as the certain payoff increases, as this would indicate inconsistent preferences. Additionally, it makes no sense for individuals to switch more than once from risky to safe as the payoff increases. I control for inconsistencies by excluding all individuals with inconsistent preferences. I include a binary variable indicating 1 for risk averse individuals. Individuals are risk averse if they prefer 20, 25 or 30 euros with certainty over the expectational payoff of 35 when opting for the lottery. In other words, risk averse individuals opt for more than two safe choices. If an individual is classified as risk averse in 2009, the individual is modelled to not change risk preferences in later years. This is the unfortunate consequence of limited availability on risk preferences of the participants in subsequent years.
6 Empirical framework
To establish if ex-ante moral hazard is present, I establish a univariate probit model. In the model, smoking and excessive alcohol use are dependent variables, explained by whether an individual has opted for a positive voluntary deductible and a vector of controls. The controls include variables regarding risk type, risk preferences, and socioeconomic status, where risk type is assessed by both a subjective and an objective measure. Socioeconomic status controls are lagged and include gender, logarithmic household income, age and squared age and educational level. The controls help to account largely for possible selection effects, i.e., endogeneity within the decision to uptake the voluntary deductible. The standard errors are clustered at individual level. This is necessary to obtain robust estimates, as multiple observations of the same individual are unlikely to be independent. An auxiliary regression on the uptake of the voluntary deductible showed autocorrelation. The uptake of the voluntary deductible is largely time invariant, yielding a fixed-effects model unsuitable for correct specification as fixed-effects models remove time-invariant variation.
In general, univariate probit models are suited in regression analyses with binary dependent variables, as ordinary least squares regressions often result in highly non-normal error term distributions. Equation (1) shows the univariate probit model, where the error term is normally distributed according to 𝜀𝑖𝑡 ~𝑁(0, σ2)
(1) 𝐿𝐼𝑖𝑡 = {1 𝑖𝑓 𝛽 𝐼𝑣
𝑖𝑡+ 𝜶′𝒊𝒕𝜿𝑰+ 𝜀𝑖𝑡 > 0
17
𝐿𝐼𝑖𝑡 indicates whether an individual smokes or uses alcohol according to the definitions in year
t and thus is specified the following way.
𝐿𝐼𝑖𝑡 = 1 if an individual is regarded an excessive smoker/drinker 𝐿𝐼𝑖𝑡 = 0 if an individual is not regarded an excessive smoker/drinker
I = s or a, indicating a smoking or drinking individual. 𝑣𝑖𝑡 indicates whether an individual has opted for a positive voluntary deductible in year t, and 𝛽𝐼 measures the ex-ante moral hazard effect. 𝜶′𝒊𝒕𝜿𝑰 is a vector of controls, where 𝜶′𝒊𝒕 is a 1 by 18 vector of coefficients and 𝜿𝑰 is an
18 by 1 vector of control variables.
To establish whether individual decisions to opt for the voluntary deductible influences lifestyle improvements the model specified in (1) is extended. I now also allows for ss or sa indicating whether an individual has stopped smoking or drinking in year t.
7 Descriptive statistics
Table 1 gives the description and proportions of the most important variables in the analysis. The table provides the proportions of observations with certain characteristics for the full sample and the sub-samples, where individuals either opted for no voluntary deductible or a positive voluntary deductible. This section serves to highlight the most important and striking statistics. Important to note is that table 1 is merely concerned with observations at the individual level and thus no inference can be made from this table about the number of individuals with specific characteristics.
18
Table 1: Variable definitions and their occurring proportions
Sample
Full Vd = 0 Vd > 0
Variables Description Mean St. dev Mean St. dev Mean St. dev
Lifestyle decision
Excessive smoker 1 if smoking 19+ cigarettes/cigars/pipes
per day, 0 otherwise 0.068 0.251 0.065 0.247 0.081 0.272 Excessive drinker 1 if drinking on 5 or
more days a weeks & drinking 5 or more drinks on the day most alcohol is drunk, 0 otherwise 0.043 0.202 0.042 0.202 0.043 0.202 Stopped excessive smoker 1 if categorized as excessive smoker in year t but not in year
t+1 0.023 0.149 0.022 0.148 0.024 0.154 Stopped excessive drinker 1 if categorized as excessive drinker in year t but not in year
t+1
0.022 0.148 0.022 0.148 0.022 0.148
Voluntary deductible 1 if voluntary deductible
is positive, 0 otherwise 0.156 0.363 Sociodemographic
controls (lagged)
Male 1 if male, 0 if female 0.554 0.497 0.547 0.498 0.596 0.491
Age Age in number of years 57.567 14.521 58.334 14.376 53.425 14.604
Low education 1 if highest completed education is primary or no education is completed, 0 otherwise
0.052 0.223 0.054 0.225 0.046 0.209
Middle education 1 if highest completed education is: secondary or intermediate vocational, 0 otherwise
0.608 0.488 0.627 0.484 0.506 0.500
High education 1 if highest completed education is: higher vocational or university, 0 otherwise
0.340 0.474 0.320 0.466 0.449 0.498
Income Net household income
in euros 3,017.467 6,169.572 3,025.809 6,691.962 2,972.464 1,349.448
Risk type
Good objective health 1 if suffers from no long-standing disease, affliction, handicap, or the consequences of an accident, 0 otherwise 0.617 0.486 0.595 0.491 0.730 0.444 Good subjective health 1 if self-assessed health status is: good, very good or excellent, 0 otherwise
0.799 0.401 0.787 0.409 0.861 0.346
Risk preference
Risk aversion 1 if opted for three or more consistent safe lottery choices, 0 otherwise
0.459 0.498 0.471 0.499 0.397 0.490
Number of
observations 6587 5557 1030
19
Higher proportions of excessive smokers in the sub-sample that opted for the voluntary deductible is an indication of advantageous selection. Individuals who opted for the voluntary deductible may be more risk-seeking and therefore opt for lifestyles with higher health risks. The full and sub-sample proportions of excessive drinkers are almost identical, meaning that within the full sample there is no large discrepancy between the proportions of excessive drinkers that opted for no voluntary deductible or a positive voluntary deductible. The full and sub-sample proportions of stopped excessive drinkers and stopped excessive smokers are identical. This indicates that opting for a voluntary deductible or not does not lead to different incentives to stop these excessive lifestyles. Unfortunately, part of this result can also be explained by the relatively low number of stopped excessive smokers and drinkers in the sample. The proportion of observations who opted for a positive voluntary deductible is 15,6 percent.
Observations who opted for a positive voluntary deductible have relatively lower net household income compared to observations without the voluntary deductible. Intuitively, wealthier individuals are less prone to liquidity risks as out-of-pocket expenses make up a smaller proportion of their overall wealth. This makes it more attractive to opt for less coverage in return for a premium rebate as liquidity problems as a result of out-of-pocket expenses pose a relatively lower risk. This is not observed. An argument for the observed descriptive is that individuals with lower liquidity, due to their lower financial capabilities, could also be the ones who need the premium rebate the most, as this leaves more finances for other essential services and commodities that could otherwise not be bought.
20
Figure 1 serves to depict the yearly proportions of excessive smokers and drinkers in the full sample and the sub-samples who opted for no voluntary deductible or a positive voluntary deductible. The proportions are calculated as the total number of individuals who adhered to either one of these lifestyles relative to the total number of observations of that year. For the sub-samples, the proportions are calculated as the total number of individuals who adhered to either one of these lifestyles relative to the total number of individuals who opted for no or a positive voluntary deductible. Because of this construction, it is impossible for both sub-samples lines to lie above or below the full sample line within one lifestyle category, i.e., the sub-sample lines work in opposite directions relative to the full sample line.
Strikingly, the graph shows that in 2009 excessive lifestyles are more common in the sub-samples who opted for a positive voluntary deductible. The overall trend of individuals who adhere to excessive lifestyles is decreasing. However, in 2019 roughly 1 in 25 individuals were considered an excessive smoker or drinker within the sample, meaning there is still terrain to be gained.
Figure 1: Yearly proportions of excessive smokers and drinkers
21
8 Results
Table 2 shows the results of the univariate probit model. The control for high education and the year dummy for 2019 are omitted due to perfect collinearity. In all four specified models, the hypothesis that opting for the voluntary deductible influences lifestyle decisions or lifestyle improvements can be rejected, i.e., the coefficient of the voluntary deductible is not significant in any of the specified models. In other words, in model 1 and 2 the decision to uptake a positive voluntary deductible in year t does not influence the probability of being categorized as an excessive smoker or drinker that year. For excessive smokers, the descriptive statistics showed that the sub-sample who opted for a positive voluntary deductible contained relatively more excessive smokers, suggesting other variables play a role in as to why individuals adhere to excessive lifestyles. Omitted variables not captured within the model possibly provide a better explanation for varying proportions between sub-samples.
Model 3 and 4 show that the decision to uptake the voluntary deductible does not influence the probability of a change to a healthier lifestyle, i.e., opting for the voluntary deductible does not influence the decision to stop excessive smoking or drinking.
8.1 Marginal effects for the average individual
22 Table 2: Results of the univariate probit model
(1) (2) (3) (4)
Variables
Excessive smoker
Excessive
drinker Stopped excessive smoker Stopped excessive drinker
Voluntary deductible 0.0554 -0.0539 0.0448 -0.0440 (0.1145) (0.1088) (0.1022) (0.0947) Male 0.1374 0.5983*** 0.0496 0.2697*** (0.1121) (0.1056) (0.0798) (0.0874) Age 0.0065** 0.0030 -0.0052 0.0097** (0.0029) (0.0033) (0.0043) (0.0048) Squared age -0.0002*** -0.0001 -0.0001** -0.0001** (0.0000) (0.0000) (0.0000) (0.0000) Low education 0.0490 0.1443 0.3799** -0.1118 (0.1857) (0.2165) (0.1522) (0.1903) Middle education -0.0595 -0.0151 0.2481** -0.0002 (0.1163) (0.0935) (0.0994) (0.0846)
High education (omitted)
Income -0.4801*** 0.0953 -0.3649*** 0.1157
(0.0889) (0.0931) (0.0759) (0.0828)
Good objective health 0.1190 -0.0103 -0.0429 0.0056
(0.1073) (0.0970) (0.0872) (0.0889)
Good subjective health -0.2402*** -0.1341 -0.2147** -0.0133
(0.0929) (0.1069) (0.0938) (0.1101) Risk aversion 0.1164 0.0662 0.1761** 0.0749 (0.1084) (0.1066) (0.0790) (0.0775) Year 2009 0.2742** 0.3262** -0.0013 0.0065 (0.1250) (0.1388) (0.1765) (0.1873) Year 2010 0.2705** 0.2397* -0.0595 0.3273** (0.1168) (0.1399) (0.1726) (0.1431) Year 2011 0.2536** 0.1725 0.0260 0.1706 (0.1133) (0.1224) (0.1705) (0.1638) Year 2012 0.2929*** 0.2012 -0.1322 0.1304 (0.1099) (0.1305) (0.1796) (0.1727) Year 2013 0.1862* 0.2396* 0.0307 0.1495 (0.1065) (0.1241) (0.1569) (0.1588) Year 2015 0.1842* 0.1579 -0.0261 0.1064 (0.1100) (0.1151) (0.1721) (0.1829) Year 2016 0.2017** -0.0087 -0.0131 0.1419 (0.0996) (0.1228) (0.1874) (0.1816) Year 2017 0.1502* 0.0595 0.0714 -0.0227 (0.0834) (0.1295) (0.1857) (0.1706) Year 2018 0.0238 -0.0868 -0.0492 0.0621 (0.0952) (0.1386) (0.1992) (0.1922) Year 2019 (omitted) Constant 2.3572*** -2.9013*** 1.3313** -3.3811*** (0.7759) (0.8145) (0.6736) (0.7029) Observations 6,587 6,587 6,587 6,587
Cluster-robust standard errors in parentheses, coefficients of high education and the year 2019 are omitted due to perfect collinearity
23
Table 3: Marginal effects of the univariate probit model for the average individual
(1) (2) (3) (4) Predicted probabilities Variables Excessive smoker Excessive
drinker Stopped excessive smoker Stopped excessive drinker
Voluntary deductible 0.0062 -0.0042 0.0019 -0.0022 (0.0127) (0.0084) (0.0043) (0.0047) Male 0.0153 0.0462*** 0.0021 0.0132*** (0.0126) (0.0090) (0.0034) (0.0041) Age 0.0007** 0.0002 -0.0002 0.0005** (0.0003) (0.0003) (0.0002) (0.0002) Squared age -0.0000*** -0.0000 -0.0000** -0.0000*** (0.0000) (0.0000) (0.0000) (0.0000) Low education 0.0055 0.0112 0.0160** -0.0055 (0.0207) (0.0168) (0.0062) (0.0093) Middle education -0.0066 -0.0012 0.0105** -0.0000 (0.0130) (0.0072) (0.0041) (0.0042)
High education (omitted)
Income -0.0536*** 0.0074 -0.0154*** 0.0057
(0.0107) (0.0072) (0.0032) (0.0041)
Good objective health 0.0133 -0.0008 -0.0018 0.0003
(0.0120) (0.0075) (0.0036) (0.0044)
Good subjective health -0.0268*** -0.0104 -0.0090** -0.0007
(0.0103) (0.0083) (0.0040) (0.0054) Risk aversion 0.0130 0.0051 0.0074** 0.0037 (0.0121) (0.0083) (0.0034) (0.0038) Year 2009 0.0306** 0.0252** -0.0001 0.0003 (0.0137) (0.0108) (0.0074) (0.0092) Year 2010 0.0302** 0.0185* -0.0025 0.0161** (0.0128) (0.0108) (0.0073) (0.0069) Year 2011 0.0283** 0.0133 0.0011 0.0084 (0.0125) (0.0095) (0.0072) (0.0080) Year 2012 0.0327*** 0.0156 -0.0056 0.0064 (0.0122) (0.0101) (0.0076) (0.0084) Year 2013 0.0208* 0.0185* 0.0013 0.0073 (0.0118) (0.0097) (0.0066) (0.0077) Year 2015 0.0206* 0.0122 -0.0011 0.0052 (0.0123) (0.0089) (0.0073) (0.0089) Year 2016 0.0225** -0.0007 -0.0006 0.0070 (0.0112) (0.0095) (0.0079) (0.0088) Year 2017 0.0168* 0.0046 0.0030 -0.0011 (0.0093) (0.0100) (0.0079) (0.0084) Year 2018 0.0027 -0.0067 -0.0021 0.0030 (0.0106) (0.0107) (0.0084) (0.0094) Year 2019 (omitted) Observations 6,587 6,587 6,587 6,587
Cluster-robust standard errors in parentheses, coefficients of high education and the year 2019 are omitted due to perfect collinearity
24
8.1.1 Excessive smokers
Opting for the voluntary deductible does not significantly influence the probability of being categorized as an excessive smoker for the average individual. A 1 percent increase in income leads roughly to a 5.36 percent decreased probability of being categorized as an excessive smoker. This is in line with empirically examined correlations between socio-economic status and smoking behavior (Hitchman et al., 2014). Individuals who assessed their health as good or beyond have a decreased probability of being categorized as excessive smokers of 2.68 percent.
8.1.2 Excessive drinkers
Opting for the voluntary deductible does not significantly influence the probability of being categorized as an excessive drinker for the average individual. Males have a 4.62 percent higher probability to be categorized as excessive drinkers. This is in line with evidence indicating binge drinking is more common among men then among women (Wilsnack R.W., Wilsnack S.C., Kristjanson, Vogeltanz‐Holm and Gmel, 2009)
8.1.3 Stopped excessive smokers
Opting for the voluntary deductible does not significantly influence the probability of being categorized as a stopped excessive smoker for the average individual. A 1 percent increase in income roughly decreases the probability of being categorized as a stopped excessive smoker by 1.54 percent. Individuals who assessed their health as good or beyond have a decreased probability of 0.09 percent of being categorized as a stopped excessive smoker. Risk averse individuals have 0.74 percent increased probability of being categorized as stopped excessive smokers.
8.1.4 Stopped excessive drinkers
25
8.2 Robustness check
Part of the literature on moral hazard already led to the anticipation of possible insignificant ex-ante moral hazard effects (Kenkel, 2000; Einav and Finkelstein, 2018). Therefore, no indication of ex-ante moral hazard effects is a true possibility. Nonetheless, other explanations for insignificant effects should be precluded. Smaller sample sizes are often more hampered by insignificant results. The inclusion of the risk aversion parameter controlling for risk preferences significantly decreased the size of the sample. Therefore, table 4 in the appendix shows the results of the univariate probit model with the exclusion of the risk aversion parameter.
Exclusion of the risk aversion parameter leads to an increase in observations from 6.587 to 21.520, where the 21.520 observations are dispersed over 4.908 individuals. Again, the standard errors are clustered at individual level. Multiple controls show higher levels of significance compared to the model where the risk aversion parameter is included. Despite the larger sample size, the voluntary deductible has no significant effect on an individual probability to be categorized as an excessive smoker, drinker or an individual who has stopped excessive smoking or drinking.
9 Limitations
This section serves to discuss some of the most important limitations of my research. Firstly, this thesis makes use of several definitions which should carefully be reviewed. To determine the relationship between the voluntary deductible and unhealthy lifestyle decisions, it was necessary to define excessive smokers and drinkers.
26
Another limitation of this study is the representability of the examined sample. Potential policy recommendations should be based on the empirical evidence of samples that provide a true representation of the whole population. The LISS panel itself is based on a true probability sample of households but missing data could bias the eventual examined sample. Listwise removal of observations with missing data can bias the estimates if the data is not missing at random. This can result in estimates that are not a good representation of the effects in a whole population. Additionally, the representability of some of the key parameters should be reviewed. This is done in the following sections.
The most recent numbers on the voluntary deductible show that 13.3 percent of the Dutch population has a deductible exceeding the mandatory 385 euros (Vektis, 2020), i.e., opted for a positive voluntary deductible. Compared to the 15.6 percent of individual observations who opted for the voluntary deductible in this sample, this is deemed an accurate representation. The male/female representation within the sample also provides an accurate representation of the Dutch population.
The risk aversion parameter possibly provides a less accurate representation of the entire population. Individuals are deemed risk averse if they prefer a certain payoff of 30 euros over an equiprobable payoff of 5 or 65 euros, i.e., an expectational payoff of 35 euros. This results in 46 percent of risk averse observations and 54 percent of the observations being classified as risk-neutral or risk-seeking. This contradicts a more elaborate research by Kapteyn and Teppa (2011), who classified roughly 74 percent of the Dutch population as risk averse. Unfortunately, the LISS panel data does not directly allow for better measures of risk aversion. As discussed, one pitfall of the inclusion of the risk aversion parameter is the decreased sample size. Another caveat of the risk parameter is a that part of the participants in the Risk Attitudes study were given lottery payoffs multiplied by a dimension K, where K equaled 150. The factor K multiplication of the payoffs may result in an upward bias. Individuals are less likely to gamble whenever the certain payoff already results in a stark increase in utility.
10 Policy recommendations
27
voluntary deductible have additional incentives to live healthier or unhealthier. This is useful information as significant ex-ante moral hazard effects would indicate a flawed healthcare system. Individuals who would not have opted for the voluntary deductible would have a disincentive to live healthier, leading to a further deterioration of their health and higher societal costs. A benevolent government that cares for the health of its population and puts capital and effort into initiatives aimed to promote healthier lifestyles should avert this outcome. Fortunately, there is no evidence for this case.
What is observed, is the relatively low proportion of individuals with a voluntary deductible. As the earlier mentioned research by van Winssen et al. (2015) mentioned, roughly 50 percent of the population would benefit from a positive voluntary deductible. The proportion of observations who opted for a voluntary deductible within the sample is only 15.6 percent. From the financial perspective of an individual, it would be worth evaluating the voluntary deductible option. At this moment in time, low-risk individuals transfer unnecessary high premiums to their insurers. Over time, as the popularity of the voluntary deductible keeps steadily increasing, individuals could be left with more money at the end of the year. Low-risk individuals can save roughly 300 euros in annual premiums by opting for the maximum voluntary deductible.
28
deductible. This could help decrease inequities between healthy and unhealthy individuals and partly tackle the ongoing debate about the voluntary deductible.
11 Conclusion
This thesis examined the relationship between opting for the voluntary deductible and lifestyle decisions concerning smoking and alcohol use. If the uptake of the voluntary deductible were to have a statistically significant effect on excessive lifestyle behavior, this would be an indication for ex-ante moral hazard, under the proper model specifications. The univariate probit model – controlling for socio-economic characteristics, risk type and risk-preferences – found no conclusive indication for the presence of ex-ante moral hazard within the examined sample. The results were robust to the in- and exclusion of a risk parameter, which considerably changed the sample size. In the first two specifications of the model, the uptake of the voluntary deductible has no significant effect on the probability to be categorized as an excessive smoker or drinker. The other two model specifications – which model whether the uptake of the voluntary deductible has a significant effect on lifestyle improvements – also found no significant effects.
29 References
Abbring, J. H., Chiappori, P. A., & Zavadil, T. (2008). Better safe than sorry? Ex-ante and
ex-post moral hazard in dynamic insurance data.
Akerlof, G. A. (1978). The market for “lemons”: Quality uncertainty and the market mechanism. In Uncertainty in economics (pp. 235-251). Academic Press.
Alessie, R. J., Angelini, V., Mierau, J. O., & Viluma, L. (2020). Moral hazard and selection for voluntary deductibles. Health Economics, 29(10), 1251-1269.
Aron-Dine, A., Einav, L., & Finkelstein, A. (2013). The RAND health insurance experiment, three decades later. Journal of Economic Perspectives, 27(1), 197-222.
Arrow, K. J. (1963). Uncertainty and the Welfare Economics of Medical Care. The American
economic review, (53), 941-73.
Barendregt, J. J., Bonneux, L., & van der Maas, P. J. (1997). The health care costs of smoking. New England Journal of Medicine, 337(15), 1052-1057.
Centraal Bureau voor de Statistiek. (2017, September 15). Zware rokers leven gemiddeld 13 jaar korter. Retrieved January 06, 2021, from https://www.cbs.nl/nl-nl/nieuws/2017/37/zware-rokers-leven-gemiddeld-13-jaar-korter
Cohen, A., & Dehejia, R. (2004). The effect of automobile insurance and accident liability laws on traffic fatalities. The journal of law and economics, 47(2), 357-393.
Cook, P. J., & Graham, D. A. (1977). The demand for insurance and protection: The case of irreplaceable commodities. In Foundations of Insurance Economics (pp. 206-219). Springer, Dordrecht.
Dave, D., & Kaestner, R. (2009). Health insurance and ex-ante moral hazard: evidence from Medicare. International journal of health care finance and economics, 9(4), 367.
De Preux, L. B. (2011). Anticipatory ex-ante moral hazard and the effect of Medicare on prevention. Health economics, 20(9), 1056-1072.
Ehrlich, I., & Becker, G. S. (1972). Market insurance, insurance, and self-protection. Journal of political Economy, 80(4), 623-648.
30
Ezzati, M., Lopez, A. D., Rodgers, A. A., & Murray, C. J. (2004). Comparative quantification of health risks: global and regional burden of disease attributable to selected major risk factors. World Health Organization.
Geoffard, P. Y., Gardiol, L., & Grandchamp, C. (2006). Separating selection and incentive effects: an econometric study of Swiss health insurance claims data.
Hayen, A., Klein, T. J., & Salm, M. (2018). Does the framing of patient cost-sharing incentives matter? The effects of deductibles vs. no-claim refunds.
Hemenway, D. (1992). Propitious selection in insurance. Journal of Risk and
Uncertainty, 5(3), 247-251.
Hitchman, S. C., Fong, G. T., Zanna, M. P., Thrasher, J. F., Chung-Hall, J., & Siahpush, M. (2014). Socioeconomic status and smokers’ number of smoking friends: findings from the International Tobacco Control (ITC) Four Country Survey. Drug and alcohol
dependence, 143, 158-166.
Jellinek. (2020, August 25). Hoeveel mensen overlijden er door alcohol, tabak en andere drugs? Retrieved December 11, 2020, from https://www.jellinek.nl/vraag-antwoord/hoeveel-mensen-overlijden-er-door-alcohol-tabak-en-drugs/
Kapteyn, A., & Teppa, F. (2011). Subjective measures of risk aversion, fixed costs, and portfolio choice. Journal of Economic Psychology, 32(4), 564-580.
Kenkel, D. S. (2000). Prevention. Handbook of health economics. AJ Culyer and JP
Newhouse, editors.
Kowalski, A. (2016). Censored quantile instrumental variable estimates of the price elasticity of expenditure on medical care. Journal of Business & Economic Statistics, 34(1), 107-117. Milcarz, M., Polanska, K., Bak-Romaniszyn, L., & Kaleta, D. (2018). Tobacco health risk awareness among socially disadvantaged people—a crucial tool for smoking
cessation. International journal of environmental research and public health, 15(10), 2244. National Health Service (NHS). (2012, January 09). Have two alcohol-free days a week, say MPs. Retrieved January 06, 2021, from
https://www.nhs.uk/news/lifestyle-and-exercise/avoid-alcohol-2-days-per-week/
31
Patra, J., Taylor, B., & Rehm, J. (2009). Deaths associated with high-volume drinking of alcohol among adults in Canada in 2002: A need for primary care
intervention?. Contemporary Drug Problems, 36(1-2), 283-301.
Pauly, M. V. (1968). The economics of moral hazard: comment. The american economic
review, 58(3), 531-537.
Rehm, J. (2011). The risks associated with alcohol use and alcoholism. Alcohol Research &
Health, 34(2), 135.
Remmerswaal, M., Boone, J., & Douven, R. (2019). Selection and moral hazard effects in healthcare (No. 393. rdf). CPB Netherlands Bureau for Economic Policy Analysis.
Rezayatmand, R., Groot, W., & Pavlova, M. (2017). Smoking behaviour and health care costs coverage: a European cross-country comparison. International journal of health economics
and management, 17(4), 453-471.
Romp, M., Nijhof, E., & 't Hoen, E. (2020). Verzekerden in beeld 2020 (Rep.). Vektis. Rothschild, M., & Stiglitz, J. E. (1970). Increasing risk: I. A definition. Journal of Economic
theory, 2(3), 225-243.
Rowell, D., & Connelly, L. B. (2012). A history of the term “moral hazard”. Journal of Risk
and Insurance, 79(4), 1051-1075.
Shavers, V. L. (2007). Measurement of socioeconomic status in health disparities research. Journal of the national medical association, 99(9), 1013.
Steptoe, A., Wardle, J., Cui, W., Baban, A., Glass, K., Karl Pelzer, ... & Vinck, J. (2002). An international comparison of tobacco smoking, beliefs and risk awareness in university
students from 23 countries. Addiction, 97(12), 1561-1571.
Tversky, A., & Kahneman, D. (1991). Loss aversion in riskless choice: A reference-dependent model. The quarterly journal of economics, 106(4), 1039-1061.
Van Winssen, K. P. M., Van Kleef, R. C., & Van de Ven, W. P. M. M. (2015). How profitable is a voluntary deductible in health insurance for the consumer?. Health
32
Van Winssen, K. P. M., Van Kleef, R. C., & Van de Ven, W. P. M. M. (2016). Potential determinants of deductible uptake in health insurance: How to increase uptake in The Netherlands?. The European Journal of Health Economics, 17(9), 1059-1072.
Wechsler, H., Dowdall, G. W., Davenport, A., & Castillo, S. (1995). Correlates of college student binge drinking. American journal of public health, 85(7), 921-926.
Wilsnack, R. W., Wilsnack, S. C., Kristjanson, A. F., Vogeltanz‐Holm, N. D., & Gmel, G. (2009). Gender and alcohol consumption: patterns from the multinational GENACIS project. Addiction, 104(9), 1487-1500.
World Health Organization (WHO). (2021) Alcohol, risks. Retrieved January 06, 2021, from https://www.who.int/health-topics/alcohol#tab=tab_2
Zweifel, P., & Manning, W. G. (2000). Moral hazard and consumer incentives in health care. In Handbook of health economics (Vol. 1, pp. 409-459). Elsevier.
33 Appendix
Table 4: Results of the univariate probit model with the risk-aversion parameter excluded
(5) (6) (7) (8) Variables Excessive smoker Excessive drinker Stopped excessive smoker Stopped excessive drinker Voluntary deductible 0.0208 -0.0261 0.0162 -0.0728 (0.0568) (0.0573) (0.0543) (0.0533) Male 0.1396** 0.5628*** -0.0052 0.3287*** (0.0569) (0.0584) (0.0445) (0.0456) Age 0.0030** 0.0018 -0.0034* 0.0011 (0.0014) (0.0016) (0.0020) (0.0021) Squared age -0.0001*** -0.0001*** -0.0000* -0.0000** (0.0000) (0.0000) (0.0000) (0.0000) Low education 0.2421** -0.1905 0.3097*** -0.2559** (0.1038) (0.1169) (0.0902) (0.1080) Middle education 0.0559 -0.0693 0.1968*** -0.0448 (0.0599) (0.0531) (0.0549) (0.0469) High education (omitted) Income -0.2786*** 0.0976** -0.2780*** 0.1050** (0.0495) (0.0492) (0.0394) (0.0455)
Good objective health 0.0400 -0.0534 0.0315 0.0071
(0.0579) (0.0566) (0.0493) (0.0482)
Good subjective health -0.2344*** -0.0272 -0.2359*** -0.0038
(0.0534) (0.0621) (0.0537) (0.0592) Year 2009 0.3432*** 0.2835*** 0.1184 0.0702 (0.0640) (0.0720) (0.0981) (0.0963) Year 2010 0.3740*** 0.2523*** 0.0265 0.2665*** (0.0612) (0.0701) (0.0960) (0.0844) Year 2011 0.2872*** 0.1508** 0.1682* 0.1601* (0.0601) (0.0678) (0.0910) (0.0888) Year 2012 0.2751*** 0.1648** 0.1064 0.1128 (0.0582) (0.0672) (0.0911) (0.0882) Year 2013 0.2021*** 0.1226* 0.1180 0.2188*** (0.0574) (0.0686) (0.0899) (0.0840) Year 2015 0.1719*** 0.1474** 0.1338 0.0154 (0.0578) (0.0654) (0.0918) (0.0974) Year 2016 0.1985*** 0.1050 -0.0095 0.1331 (0.0559) (0.0672) (0.1037) (0.0867) Year 2017 0.0605 0.1090 0.1020 0.1477* (0.0543) (0.0671) (0.0966) (0.0885) Year 2018 0.0423 -0.0278 0.1306 0.1260 (0.0481) (0.0640) (0.0946) (0.0904) Year 2019 (omitted) Constant 0.7333* -2.7982*** 0.3423 -3.0176*** (0.4139) (0.4175) (0.3369) (0.3895) Observations 21,520 21,520 21,520 21,520
Cluster-robust standard errors in parentheses, coefficients of high education and the year 2019 are omitted due to perfect collinearity
