Bachelor Thesis
Econometrie & Operationele Research
The connection between risk aversion and religion: The Netherlands and America compared
Joris Plaatsman 10375708
University of Amsterdam Supervisor: Z. Huang 27-06-2015
Abstract
This paper investigated the connection between risk aversion and religion. Previous research found this connection to be positive. It is explored how life satisfaction and subjective health status relate to this connection. The research was performed for both Dutch and American data, to see whether the explored relationships behave in the same way in different countries. It turned out that the relation between risk aversion and religion is positive in both countries, but only the Dutch data clearly showed that this relation probably goes via religious service attendance. The American data merely showed a relation between religion and risk aversion. In addition with some other differences found between the Netherlands and America, it is concluded that culture probably has a significant effect on the explored relationships. This led to the most important conclusion, which is that research in general should always take into account the effect culture has on relationships. Furthermore, life satisfaction and subjective health status had no relation with risk aversion.
Contents
1 Introduction ... 1
2 Theoretical framework ... 3
2.1 Measuring risk aversion ... 3
2.2 Previous literature ... 6 2.3 Reverse causality ... 9 3 Experiment design ... 11 4 Results ... 15 5 Analysis ... 22 6 Conclusion ... 25 References... 28 Appendix A ... I Appendix B ... II
1 Introduction
People have to make choices that involve risk every day. Especially in the business world, economic decisions that involve a severe amount of risk are commonplace. For example, Rijksoverheid (n.d.) states that banks took too much risk in the years prior to the financial crisis of 2008 when providing mortgages. The way someone behaves when faced with a risky choice is determined by his risk attitude. A term closely related to risk attitude, risk aversion, is thought of as the inverse of someone’s willingness to participate in risky choices (Wakker, 2010), and is a much investigated topic. The more risk aversion an individual possesses, the more likely he will try to avoid choices that involve risk (Wakker, 2010).
Research performed on risk aversion can be categorized into two categories. The first category contains the studies in which research is performed on the causal relation between personal characteristics (e.g. age, gender) and someone’s level of risk aversion. The second category contains the studies which focus on finding out what impact someone’s level of risk aversion can have on the decisions he makes.
The study performed by Fellner and Maciejovsky (2007) gives an example of a possible consequence of having a higher level of risk aversion and thus belongs in the second category as described above. They find that people who have more risk aversion, have a relatively lower market activity, meaning that more risk aversion results in a more cautious buying and selling behavior on the asset market.
While investigation on the impacts of a higher level of risk aversion still needs further attention (Fellner and Maceijovsky, 2007), the focus of this study is on the causal relationship between personal characteristics and risk aversion. Therefore, this study belongs in the first category of studies, as described before. An example of a study from this category is the study from Morin and Suarez (1983). They investigate the effect of age on risk aversion, and find that risk aversion increases with age. Another example is the study by Halek and Eisenhauer (2001), who find a positive relation between marriage and risk aversion.
A subject that has been investigated in many papers, is the influence of religion on risk aversion. Most studies conclude that these two are positively related (Noussair, Trautmann, Van de Kuilen, and Vellekoop, 2013; Sinha, Cnaan, and Gelles, 2007). Religion has also been associated with life satisfaction and subjective health status (Elliot and Hayward, 2009; Ferraro and Albrecht-Jensen, 1991). With life satisfaction, the global satisfaction with life is meant, and subjective health status is the way an individual values his own health. Subjective health status is assumed to have a negative relation with risk aversion (Bellante and Green, 2004). Life satisfaction and risk
aversion have not been associated very often, but the paper by Dohman et al. (2011) suggests a negative relation, while the paper by Proctor, Linley, and Maltby (2009) suggests a positive relation. Based on this, this paper assumes that there is reverse causality present between life satisfaction and risk aversion.
All over the world people try to insure themselves against risks. While insurance companies are not allowed to apply price discrimination between religious and non-religious people, they can use information about the relation between religion and risk aversion when forecasting the claim behavior of the policyholders. Therefore, this study focuses its attention on the relation between religion, life satisfaction, subjective health status, and risk aversion with the purpose to add to the existing knowledge about risk aversion and to further understand how these variables can influence choices.
As a robustness check, the research in this study is performed for both the Netherlands and America. To do this, data from the LISS (Longitudinal Internet Studies for the Social sciences) panel and the ALP (American Life Panel) are used.
Furthermore, this study assumes that people value choices according to a utility function from either the exponential or power family (Wakker, 2010). The parameter 𝜃, that is used to define these utility functions, measures an individual’s risk aversion (exponential utility) or the inverse of his risk aversion (power utility) (Wakker, 2010). In this study, both families of utility functions are reviewed. Lotteries are used to estimate the parameter 𝜃 for both the exponential and the power utility. In a lottery, you get a certain monetary amount with a certain chance (Wakker, 2010).
The purpose of the research is to find an answer to the following question: to what extent is the positive relation between religion and risk aversion influenced by life satisfaction and subjective health status? Three sub questions support this main
question. The first sub question is: what is the relation between life satisfaction and risk aversion? The second sub question is: what is the effect of religion on risk aversion when not controlling for life satisfaction and subjective health status? The third sub question is: are the results the same when performing the research for both Dutch and American data?
This paper is structured as follows. In section two, the theoretical framework is discussed, where the existing literature about the relations between religion, life
satisfaction, subjective health status, and risk aversion is reviewed, with the goal to form hypotheses. Furthermore, several measures for risk aversion are discussed. In section three, the research design is discussed. In section four, the research results are given, and the three sub questions mentioned above are examined. Section five analyzes the results, and section six concludes.
2 Theoretical framework
In this section, different measurements of risk aversion are reviewed. Furthermore, the existing literature and ideas about the relationships between religion, subjective health status, life satisfaction, and risk aversion are discussed with the goal to form hypotheses.
2.1 Measuring risk aversion
First, a way to measure risk aversion needs to be defined. An important difference between studies on risk aversion is the way they measure it, which usually depends on the view on risk. For example, when the main focus of a study is to explore the way people invest money, a measure of risk aversion would be the proportion of total wealth
held as risky assets (Jianakoplos and Bernasek, 1998; Wang and Hanny, 1997). Another measurement for risk aversion, when the main focus is financial decision making, is to look at the choices people make when faced with the decision between a lottery and a sure payoff (Holt and Laury, 2002; Noussair et al., 2013).
Noussair et al. (2013) use lotteries to measure risk aversion when investigating the relationship between religion and risk aversion. They asked respondents to choose between a lottery and a sure payoff five times, where the sure payoff differed each question. The writers use the number of instances in which a respondent chose the sure payoff as a measure of risk aversion.
Although the measure of Noussair et al. (2013) has the advantage that it is easy to measure the number of choices someone picks the sure payoff, there is a
disadvantage as well. The problem with this measure is that it treats the difference in risk aversion between people who choose the sure payoff three or four times the same as the difference in risk aversion between people who choose the sure payoff two or three times. In other words, Noussair et al. (2013) assume that every ‘step’ in the risk
aversion scale ranging from zero to five represents the same amount of risk aversion. It is an assumption that it is very useful in the sense that it simplifies the measure of risk aversion, but the writers do not prove that every ‘step’ indeed represents the same amount of risk aversion.
In contrast with the former methods of measuring risk aversion, a more mathematical measure is obtained by using a utility function 𝑈(𝑥). This way of measuring risk aversion does not use the same assumption that Noussair et al. (2013) use (see previous paragraph). Wakker (2010) defines a utility function as a function which usually has money as the input variable, and gives as output the utility an individual experiences from having (or receiving) that amount of money. A basic assumption of a utility function is that it is a strictly increasing (Wakker, 2010).
A measure that is often used to measure risk aversion is the Pratt-Arrow measure −𝑈𝑈′′′(𝑥)(𝑥), where 𝑈′ and 𝑈′′ are the first and second order derivative of 𝑈(𝑥), respectively (Wakker, 2010). The Pratt-Arrow measure is also called the absolute measure of risk
aversion (Wakker, 2010). Wakker (2010) proves that when person A possesses more
risk aversion than person B, person’s A Pratt-Arrow measure is bigger than that of person B. Another popular measure of risk aversion, related to the Pratt-Arrow measure, is the relative measure of risk aversion, defined as −𝑥𝑈𝑈′′′(𝑥)(𝑥) (Wakker, 2010).
Two popular families of utility functions are the power family and the exponential family, which are often used because of their convenient properties
(Wakker, 2010). The power family is defined on ℝ+ and with a parameter 𝜃 (for α > 0): for 𝜃 > 0, U(α)=αθ;
for 𝜃 = 0, U(α)=ln(α); for 𝜃 < 0, U(α)=-αθ.
It is characterized by having a constant relative risk aversion equal to 1 − 𝜃, which leads to the explanation why the power utility is often called the CRRA family (Wakker, 2010). The exponential family is defined as follows (on ℝ):
for 𝜃 > 0, U(α)=1-𝑒−𝜃𝛼; for 𝜃 = 0, U(α)=α; for 𝜃 < 0, U(α)=𝑒−𝜃𝛼-1.
The main quality of the exponential family is that it has a constant absolute risk aversion equal to 𝜃, which is why the exponential family is often called the CARA family (Wakker, 2010).
An example of a study in which a utility function from the power family is estimated, is the study by Dohmen et al. (2005). They use a survey about investment choices and some information about the individual’s wealth to estimate an interval for the individual’s parameter 𝜃.
The advantage of measuring risk aversion with a parameter 𝜃, is that this
parameter is uniquely defined for each individual (Wakker, 2010). However, it is a little more difficult to estimate this parameter, than to just observe the number of times an individual prefers a sure payoff to a lottery (like the measure from Noussair et al. (2013)).
2.2 Previous literature
In this section, the existing literature and ideas about the relationships between religion, subjective health status, life satisfaction, and risk aversion are discussed. Furthermore, hypotheses are formed. Figure 1 (see Appendix B) depicts the direction in which previous studies have estimated relationships. What stands out is that religion can influence risk aversion via multiple ways. Of course, it is likely that there are other factors playing a role, but the attention of this study is restricted to these variables. For the sake of clarity, the existing literature about the relationships between the variables in figure 1 is discussed in the order as depicted in the figure.
The influence of religion on decision making is a much investigated topic. Noussair et al. (2013) hypothesize that investigation of the relation between religion and risk aversion (relation one in figure 1) gains more insight in the way religion shapes the economy. Another similar research is the study from Sinha et al. (2007). To measure religion, both studies use the number of times a respondent attends religious worship services. In addition, Noussair et al. (2013) also measure the strength of religious beliefs, while Sinha et al. (2007) additionally measure the personal importance of religion and the participation in a youth group. Both studies find a positive relation between church membership or attendance and risk aversion. Noussair et al. (2013) find that this relation is in some cases stronger for Protestants than Catholics. However, they find no connection between the strength of religious beliefs and risk aversion, which leads the authors to the conclusion that the positive relation between religion and risk aversion is probably driven by the social aspects of church membership.
Another study which suggests a positive relation between religion and risk aversion, is the study by Dohman et al. (2005). Despite the fact that they do not make a distinction between different aspects of believing, they do make a distinction between several denominations and find a weaker effect for Catholics than for Protestants, similar to the results of Noussair et al. (2013).
The second relation in figure 1 is the relation between religion and life
measured in several ways, as will be discussed. Hadaway (1978) and Dorahy et al. (1998) both report a positive correlation between religion and life satisfaction.
However, Dorahy et al. (1998) find no significant correlation between life satisfaction and religion for women, only for men. An example of a study which dives deeper into the relation by regressing life satisfaction on religion, is the study from Elliot and Hayward (2009). They find that, although both effects are positive, the effect of the strength of personal faith is a little bigger than the effect of attending church services. In contrast, Lim and Putnam (2010) conclude in their study that the positive relation is likely driven by the attendance of religious services.
When looking at the measurement of life satisfaction, a clear division can be seen. Hadaway (1978), Elliot and Hayward (2009), and Lim and Putnam (2010) use one single question on global life satisfaction, while Dorahy et al. (1998) use the
Satisfaction With Life Scale (SWLS) (Diener, Emmons, Larsen, and Griffin, 1985). This scale measures an individual’s satisfaction with life as a whole by using five statements, where the individual states how much he agrees with them. It is considered as a good measure of life satisfaction, because the scores on the SWLS have a
significant relation with several personal characteristics (Diener et al., 1985).
Furthermore, instead of estimating satisfaction in specific domains, such as health or finances, individuals can judge their own life by using their own personal criteria (Pavot and Diener, 1993).
The next relation in figure 1 is the relation between religion and subjective health status (relation three). This relation has been the subject of a couple of studies. Subjective health status is the way an individual values his own health. For example, Feraro and Albrecht-Jensen (1991) report that higher levels of conservative religious affiliation (measured by questions about closeness to God and life after death) are negatively related with subjective health status. The writers find that more conservative religious people usually live in the lower classes of the social ladder, which leads them to speculate that this might be the reason for the lower reported subjective health status. However, they also find that that higher levels of religious practice, measured by attending church services and praying, leads to higher levels of subjective health status.
They conclude that religion may have positive as well as negative effects on subjective health status, although they report that the positive effect of religious practice on subjective health status is larger than the negative effect of conservative religious affiliation on subjective health status.
Another study about the link between religion and subjective health status indicates a different effect for black and white people (Musick, 1996). Musick (1996) reports that religious activities that are performed in private (e.g. praying) have a positive relation with subjective health status for black people, and that religious activities that are performed in public (e.g. church services) have a positive effect for both black and white people.
Relation four in figure 1 is the relation between subjective health status and life satisfaction. Palmore and Luikart (1972) find strong evidence for a positive relation between the two. After a regression of life satisfaction on subjective health status and several other variables, they find that subjective health status explains about two thirds of the total variation in life satisfaction they find in the groups analyzed. It is striking that the effect of an individual’s own judgement of his health on his life satisfaction is more than twice as big than the effect of a physician’s rating of his health (Palmore and Luikart, 1972). More recent studies which find similar positive results, are the studies from Melin, Fugl-Meyer, and Fugl-Meyer (2003), who perform their study in Sweden, and Gwozdz and Sousa-Poza (2010), who focus their research on the elderly.
The fifth relation in figure 1, is the relation between subjective health status and risk aversion. Dohman et al. (2011) report a positive effect of subjective health status on risk aversion, when focusing on taking risks in general. However, they find a small but significant, negative effect when looking at financial risk taking. Bellante and Green (2004) also find a negative effect of subjective health status on risk aversion, when looking at risk in a financial way. A difference between the two studies is that Bellante and Green (2004) study risk aversion among the elderly, while Dohman et al. (2011) study a wide variety of ages. Bellante and Green (2004) mention the fact that subjective health and age are not independent, and that the negative effect of subjective health status on risk aversion may be partially capturing the effect of age. However, both
Dohman et al. (2011) and Bellante and Green (2004) find a negative effect (when looking at financial risk taking), so the effect of subjective health status on risk aversion is assumed to be negative.
The sixth and last relation in figure 1 is the relation between life satisfaction and risk aversion. Not many studies have assessed this relation. Dohman et al. (2011) is one of the few studies that explores the relationship between life satisfaction and risk taking in a financial way. They find that more life satisfaction leads to less risk aversion. Another study, performed by Proctor, Linley, and Maltby (2009), finds the opposite result. They find that taking more risks in, for example, violence and sex (less risk aversion), leads to less life satisfaction. Of course, taking financial risks cannot be easily compared to taking risks in violence or sex, but they both find a different effect. Both studies approach the causal relationship from a different perspective, but neither of them argue why a relationship in a certain direction should be expected, or mention the
possible presence of reverse causality. The next section treats the possibility of the presence of reverse causality for this and the other relationships of figure 1.
2.3 Reverse causality
With the exception of the literature about relation one, none of the literature discussed so far argues whether there is reverse causality present or not in relationship one to six. This section discusses arguments for and against reverse causality. The first relation that is discussed is relation six, and after that relation one to five are discussed.
As the last paragraph of section 2.2 already mentioned, the literature about relation six does not say anything about reverse causality. However, it could be argued that someone who has more life satisfaction, takes more risky choices (so possesses less risk aversion), because he does not have the same view on potential losses as a person who possesses less life satisfaction has. More life satisfaction could lead the person to put a more heavy weight on positive outcomes of a risky choice than on negative outcomes. This negative relation is in line with Dohman et al. (2011). On the other
hand, more risk aversion might also lead to more life satisfaction, because someone with more risk aversion could take less risky choices and thus has not to deal with the potential consequences of risky decisions, which could have led to a lower life
satisfaction. This positive relation is in line with Proctor, Linley, and Maltby (2009). This reasoning results in the possible presence of reverse causality.
The paper of Noussair et al. (2013) provides an argument against the reverse causality of relation one. They state that the positive causal relation between religion and risk aversion could be there because of certain teachings of the church. Gambling, something that can be linked with taking financial risks, is something that the church does not encourage and could thus lead to more risk aversion (Noussair et al., 2013).
An argument against reverse causality in relation two and three of figure 1 is that someone who has more life satisfaction or values his own health better, is not logically more easily convinced of the presence of God. On the other hand, when being religious, you are convinced that you can trust in God and that He will help and heal you. These are arguments against reverse causality in relation two and three and for a positive causal relation, in line with Hadaway (1978) andFeraro and Albrecht-Jensen (1991).
Furthermore, for relation four, it is easy to see why subjective health status probably has a causal effect on life satisfaction and not the other way around. Life satisfaction measures someone’s satisfaction with his life as a whole. Individuals can judge their own life by using their own personal criteria, and subjective health status is one of the elements that an individual takes into account when valuing his life
satisfaction (Pavot and Diener, 1993).
The last relation to discuss is relation five. An argument for a causal effect of subjective health status on risk aversion, is that the way you feel about yourself (your health) can influence certain choices you make and thus influence your risk aversion. Someone who feels less healthy can be more cautious when making certain decisions (so a negative effect, in line with Bellante and Green (2004)). However, being more cautious in making decisions (more risk aversion), does not logically lead to feeling less healthy.
To summarize section 2.2 and 2.3, the existing literature about relation one to four in figure 1 suggests that these are positive, and relation five is negative. The
previous literature about relation six is unclear and gives a reason to believe that reverse causality plays a role. However, as discussed in the second paragraph of section 2.3, there are arguments for a negative causal effect of life satisfaction on risk aversion. The possibility of reverse causality explains why the research performed in section three and four is done by two-stage least squares (2SLS). The study by Bjørnskov, Dreher, and Fischer (2008) finds a positive relation between life satisfaction on one hand and marriage, employment, gender, and age on the other hand. These variables can be used as instruments for life satisfaction, among religion and subjective health status.
The following hypotheses can now be deducted. The first hypothesis is: the relation between life satisfaction and risk aversion is negative. The second hypothesis is: controlling for subjective health status has a positive effect on the relation between religion and risk aversion. The third hypothesis is: when additionally controlling for life satisfaction, the relation between religion and risk aversion increases even more and the relation between subjective health status and risk aversion increases as well.
3 Experiment design
This section describes the design of the research that is performed in section four. As mentioned in the introduction, both Dutch and American data is used in the research for a robustness check of the results. Firstly, this section describes the general information about the LISS (Longitudinal Internet Studies for the Social sciences) panel and the ALP (American Life Panel). Secondly, the surveys that are used are described. Thirdly, the variables that are used in the ultimate model are reviewed, and the four models that are estimated in section four are given.
As already mentioned, in this paper the data of the LISS panel is used, which is administered by CentERdata (Tilburg University, The Netherlands). The LISS panel is a representative sample of Dutch individuals who participate in monthly internet surveys.
The panel is based on a true probability sample of households drawn from the
population register. The second panel that is used in this paper is the ALP. The ALP is a representative database for America. The panel is probability-based and contains over 1000 members ages 18 and older who are regularly interviewed over the internet for research purposes.1
Both databases contain surveys about religion, where a distinction is made between denomination, attending religious services, importance of religion (only ALP), and strength of belief (only LISS), as well as information about several background variables (e.g. age, gender). Furthermore, both databases hold surveys about risk attitude (measured by lotteries), life satisfaction, and subjective health status. The sample that is used from the LISS panel consists of 1586 respondents, and the sample from the ALP consists of 1023 respondents.
A total of four regressions is estimated, where a variable is added each regression. The first two regressions are about the relation between religion and risk aversion, the third regression adds subjective health status, and the fourth regression adds life satisfaction. This last regression is performed using 2SLS, as was mentioned in the last part of section 2.3. All variables are described below.
Concerning the measurement of religion, several studies find a distinct relation between different denominations of religion and risk aversion (see section 2.2, relation one). This is why the research performed in this study also makes this distinction. The categories are: Protestant, Catholic, and other religion. Moreover, previous literature suggests a different effect of attending religious services and importance of belief in relation one, two, and three of figure 1 (see section 2.2). This distinction is also made in this research. Because the LISS panel holds no data about importance of belief, the variable strength of belief in God is used instead. Importance of belief is measured on a scale from one to three (where a higher score represents more importance), and strength of belief in God is measured on a scale from one to six (where a higher score represents
1 More information about both the LISS panel and the ALP can be found at: www.lissdata.nl and
more strength of belief). Attending religious services is measured with the following five dummy variables: “Never”, “One or more times a year”, “Two or three times a
month”, “Once a week”, and “More than once a week”. “Never” is left out of the
regressions, to avoid collinearity.
As for life satisfaction, this is measured by the SWLS (Diener et al., 1985), because of the advantages that are discussed in section 2.2 (relation two). The five statements of which the SWLS consists are: “In most ways my life is close to my ideal.”, “The conditions of my life are excellent.”, “I am satisfied with my life.”, “So far I have
gotten the important things I want in life.”, and “If I could live my life over, I would change almost nothing.” The respondents indicate their agreement with the five
statements on a scale from one to seven, where seven stands for ‘strongly agree’. The scores of the five questions are summed to obtain the score on the SWLS. Subjective health status is measured with one question (“How would you value your health?”), where answers are given on a scale from one to five (where one stands for ‘poor’, and five for ‘excellent’).
Concerning the measure of risk aversion, it is assumed that every individual values choices according to a unique utility function that is either from the power or the exponential family (see section 2.1). This study uses data from lottery questions, only to make an estimation of the parameter 𝜃, which is used to define these utility functions. In case of the power family, 1 − 𝜃 is used as the measure of risk aversion. For the
exponential family, 𝜃 is used. See Appendix A for an example of this estimation technique. Because the lottery data from the ALP is expressed in very large numbers, the parameter 𝜃 of the exponential family could not be estimated. Therefore, the ALP data focusses only on the power family.
As regards the control variables, the dummy variable female is added, because the relation between religion and life satisfaction seems to be influenced by gender (see section 2.2, relation two). Furthermore, age is added as a control variable, because age seems to have a relation with subjective health status (see section 2.2, relation five) and with risk aversion (see the introduction). Moreover, the dummy variables married,
divorced, and widow are added, because they have a relation with risk aversion (see
introduction) and with life satisfaction (see the end of section 2.3).
As already mentioned at the end of section two, the effect of life satisfaction on risk aversion has to be measured with 2SLS, due to the probable existence of reverse causality. All of the previously described variables are used as instruments (except for risk aversion), in addition with the dummy variables paid job, student, retired, and other
job. These last variables are related with life satisfaction (see the end of section 2.3).
Unfortunately, the ALP data does not contain information about whether someone is a student or not, so the variable student is not used as an instrument in the regressions using the ALP data.
To find the answers to the main and sub questions, a total of four regressions is estimated. The first model is:
𝑟𝑖𝑠𝑘 𝑎𝑣𝑒𝑟𝑠𝑖𝑜𝑛 = 𝛽1 × 𝑃𝑟𝑜𝑡𝑒𝑠𝑡𝑎𝑛𝑡 + 𝛽2 × 𝐶𝑎𝑡ℎ𝑜𝑙𝑖𝑐 + 𝛽3 ×
𝑂𝑡ℎ𝑒𝑟 𝑟𝑒𝑙𝑖𝑔𝑖𝑜𝑛 + 𝛽4 × 𝑀𝑎𝑟𝑟𝑖𝑒𝑑 + 𝛽5 × 𝐷𝑖𝑣𝑜𝑟𝑐𝑒𝑑 + 𝛽6 × 𝑊𝑖𝑑𝑜𝑤 + 𝛽7 × 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛽8× 𝐴𝑔𝑒 + 𝜀
The second model that is estimated, adds religious service attendance and importance of religion to the first regression:
𝑟𝑖𝑠𝑘 𝑎𝑣𝑒𝑟𝑠𝑖𝑜𝑛 = 𝛽1× 𝑂𝑛𝑒 𝑜𝑟 𝑚𝑜𝑟𝑒 𝑡𝑖𝑚𝑒𝑠 𝑎 𝑦𝑒𝑎𝑟 + 𝛽2 × 𝑇𝑤𝑜 𝑜𝑟 𝑡ℎ𝑟𝑒𝑒 𝑡𝑖𝑚𝑒𝑠 𝑎 𝑚𝑜𝑛𝑡ℎ + 𝛽3 × 𝑂𝑛𝑐𝑒 𝑎 𝑤𝑒𝑒𝑘 + 𝛽4 ×
𝑀𝑜𝑟𝑒 𝑡ℎ𝑎𝑛 𝑜𝑛𝑐𝑒 𝑎 𝑤𝑒𝑒𝑘 + 𝛽5 × 𝑃𝑟𝑜𝑡𝑒𝑠𝑡𝑎𝑛𝑡 + 𝛽6 × 𝐶𝑎𝑡ℎ𝑜𝑙𝑖𝑐 + 𝛽7 × 𝑂𝑡ℎ𝑒𝑟 𝑟𝑒𝑙𝑖𝑔𝑖𝑜𝑛 + 𝛽8 × 𝐼𝑚𝑝𝑜𝑟𝑡𝑎𝑛𝑐𝑒 𝑟𝑒𝑙𝑖𝑔𝑖𝑜𝑛 + 𝛽9 × 𝑀𝑎𝑟𝑟𝑖𝑒𝑑 + 𝛽10 × 𝐷𝑖𝑣𝑜𝑟𝑐𝑒𝑑 + 𝛽11 × 𝑊𝑖𝑑𝑜𝑤 + 𝛽12 × 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛽13× 𝐴𝑔𝑒 + 𝜀
𝑟𝑖𝑠𝑘 𝑎𝑣𝑒𝑟𝑠𝑖𝑜𝑛 = 𝛽1× 𝑂𝑛𝑒 𝑜𝑟 𝑚𝑜𝑟𝑒 𝑡𝑖𝑚𝑒𝑠 𝑎 𝑦𝑒𝑎𝑟 + 𝛽2 × 𝑇𝑤𝑜 𝑜𝑟 𝑡ℎ𝑟𝑒𝑒 𝑡𝑖𝑚𝑒𝑠 𝑎 𝑚𝑜𝑛𝑡ℎ + 𝛽3 × 𝑂𝑛𝑐𝑒 𝑎 𝑤𝑒𝑒𝑘 + 𝛽4 × 𝑀𝑜𝑟𝑒 𝑡ℎ𝑎𝑛 𝑜𝑛𝑐𝑒 𝑎 𝑤𝑒𝑒𝑘 + 𝛽5 × 𝑃𝑟𝑜𝑡𝑒𝑠𝑡𝑎𝑛𝑡 + 𝛽6 × 𝐶𝑎𝑡ℎ𝑜𝑙𝑖𝑐 + 𝛽7 × 𝑂𝑡ℎ𝑒𝑟 𝑟𝑒𝑙𝑖𝑔𝑖𝑜𝑛 + 𝛽8 × 𝐼𝑚𝑝𝑜𝑟𝑡𝑎𝑛𝑐𝑒 𝑟𝑒𝑙𝑖𝑔𝑖𝑜𝑛 + 𝛽9 × 𝑆𝑢𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 ℎ𝑒𝑎𝑙𝑡ℎ 𝑠𝑡𝑎𝑡𝑢𝑠 + 𝛽10 × 𝑀𝑎𝑟𝑟𝑖𝑒𝑑 + 𝛽11 × 𝐷𝑖𝑣𝑜𝑟𝑐𝑒𝑑 + 𝛽12 × 𝑊𝑖𝑑𝑜𝑤 + 𝛽13 × 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛽14× 𝐴𝑔𝑒 + 𝜀
The final model that is estimated, adds life satisfaction and is estimated using 2SLS:
𝑟𝑖𝑠𝑘 𝑎𝑣𝑒𝑟𝑠𝑖𝑜𝑛 = 𝛽1× 𝑂𝑛𝑒 𝑜𝑟 𝑚𝑜𝑟𝑒 𝑡𝑖𝑚𝑒𝑠 𝑎 𝑦𝑒𝑎𝑟 + 𝛽2 × 𝑇𝑤𝑜 𝑜𝑟 𝑡ℎ𝑟𝑒𝑒 𝑡𝑖𝑚𝑒𝑠 𝑎 𝑚𝑜𝑛𝑡ℎ + 𝛽3 × 𝑂𝑛𝑐𝑒 𝑎 𝑤𝑒𝑒𝑘 + 𝛽4 × 𝑀𝑜𝑟𝑒 𝑡ℎ𝑎𝑛 𝑜𝑛𝑐𝑒 𝑎 𝑤𝑒𝑒𝑘 + 𝛽5 × 𝑃𝑟𝑜𝑡𝑒𝑠𝑡𝑎𝑛𝑡 + 𝛽6 × 𝐶𝑎𝑡ℎ𝑜𝑙𝑖𝑐 + 𝛽7 × 𝑂𝑡ℎ𝑒𝑟 𝑟𝑒𝑙𝑖𝑔𝑖𝑜𝑛 + 𝛽8 × 𝐼𝑚𝑝𝑜𝑟𝑡𝑎𝑛𝑐𝑒 𝑟𝑒𝑙𝑖𝑔𝑖𝑜𝑛 + 𝛽9 × 𝑆𝑢𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 ℎ𝑒𝑎𝑙𝑡ℎ 𝑠𝑡𝑎𝑡𝑢𝑠 + 𝛽10× 𝑆𝑊𝐿𝑆 + 𝛽11 × 𝑀𝑎𝑟𝑟𝑖𝑒𝑑 + 𝛽12 × 𝐷𝑖𝑣𝑜𝑟𝑐𝑒𝑑 + 𝛽13 × 𝑊𝑖𝑑𝑜𝑤 + 𝛽14 × 𝐹𝑒𝑚𝑎𝑙𝑒 + 𝛽15× 𝐴𝑔𝑒 + 𝜀
The first three regressions are estimated using OLS, and the last regression is estimated using 2SLS (to account for reverse causality). To estimate SWLS, a similar regression is performed, where SWLS is regressed on religious service attendance, denomination, importance of religion, subjective health status, marriage, employment, gender, and age. The next section discusses the results of these four regressions.
4 Results
In this section, the descriptive statistics of the variables that are used in the regressions are discussed. Then, the results from the four regressions that were discussed in the last section are given. In particular, the results from the LISS panel and ALP are compared to see whether the results are robust and can be applied to other countries. Furthermore,
the three sub questions, which were introduced in the introduction are reviewed to ultimately find an answer to the main question.
First, the descriptive statistics of the variables that are used are given in figure 2 in Appendix B. The variable, number of observations, and mean of the Dutch data are given in column one, two, and three, respectively. Column four and five contain the number of observations and the mean for the American data. Some variables also display the standard deviation in brackets. The first interesting thing to note is that 81.2% of the American sample is affiliated with a religion, against only 42.4% of the Dutch sample. This is connected to the fact that more than half of the Dutch sample (53.5%) never attends a religious service. The American data shows a more equally distribution when looking at the attendance of religious services.
Furthermore, the average age of the respondents in the America sample is almost ten years higher than in the Dutch sample, and the American data contains somewhat more females than the Dutch data. About ten percent more people are divorced in the American data, and about ten percent more people have a paid job. Lastly,
approximately seven percent less people fall in the category “Other job” in the American data.
The most important statistic for this paper, risk aversion, is summarized in figure 3 in Appendix B. As noted in section three, the lottery data from the LISS panel makes a distinction between hypothetical and real stakes. This distinction is not made in the American data. The measure of risk aversion for the exponential utility is 𝜃, and for the power utility it is 1 − 𝜃, where a higher 𝜃 (exponential) or 1 − 𝜃 (power) stands for more risk aversion.
Something that stands out, is that the average value of risk aversion of the real stakes is lower than that of the hypothetical stakes for both the exponential and power family. In addition, the average value of 1 − 𝜃 in the American data is much higher than in the Dutch data. The reason for this, is that the American lottery data is expressed in much higher numbers. These higher estimates of 1 − 𝜃 result in other magnitudes of the coefficients in the following regressions. However, despite the fact that the coefficients
can thus not be easily compared between the Dutch and American data, this difference in the height of 𝜃 has no influence on the significance of coefficients, so it is possible to compare the outcomes between the Netherlands and America.
A total of four regressions are performed, of which the last one is performed using 2SLS (see section three). Figure 4 on page 18 contains the most important results for regressions one to four, where the coefficients of denomination, religious service attendance, importance of religion, life satisfaction and subjective health status are displayed. Figure 5 in Appendix B displays the results of regression one and two more extensively, where the coefficients of marriage and the 𝑅2 are also displayed. All regressions are performed for both the exponential and the power measure of risk aversion, and for both the Netherlands and America. The coefficients of both “age” and “gender” are not reported as are the coefficients of religious service attendance less than once a week.
Because this paper uses two datasets, the number of variables employed in the regressions is less than it would have been if only one dataset was used. This is because the way certain variables are measured is not the same in the two datasets, and can thus not easily be compared. Furthermore, risk aversion is measured using only five
questions, which can unlikely capture all aspects of someone’s risk aversion. Moreover, because several datasets have been merged, the data of a few thousand respondents has been deleted in order to have a complete dataset for every respondent. All these
limitations to the research could lead to a bias in the significance of the coefficients estimated. As a compromise, the significance level that is used in this paper is 10% instead of 5% (as is usually the case).
First, the results of regression one and two, obtained with the LISS data panel are discussed. The pattern of results is approximately the same, when comparing the power utility with the exponential utility. The first regression measures religion with the three denominations this study employs (see section three). It is notable that, for both the power and exponential utility, only the denomination “Protestant” is significantly, positively related with risk aversion (b = .078, p = .063 (power); b = .003, p = .069
(exponential)). When additionally controlling for attendance at religious services (regression two), the coefficient for “Protestant” is no longer significant (b = .051, p = .360 (power); b = .002, p = .374 (exponential)), but the coefficient of “Catholic” for the power utility suddenly is, although only just on the 10% level (b = .075, p = .094). Furthermore, the coefficient for attendance “More than once a week” is significant and positive (b = .181, p = .052 (power); b = .008, p =.047 (exponential)), and “Strength of belief” is not significantly related with risk aversion. The coefficient for “Divorced” is negative and significant for all regressions, with all p-values below 0.05, as opposed to insignificant results for “Married” and “Widow” (see figure 5, Appendix B).
In contrast with the results from the Dutch data, the results using the American data are slightly different. In the first regression, which measures religion with several denominations, the coefficients of “Protestant” and “Catholic” are not significantly related with risk aversion, with p-values of 0.600 and 0.407, respectively. However, the coefficient of “Other religion” does show a very significant relation with risk aversion (b = 1.420, p = .007). This coefficient stays significant when controlling for attendance of religious services. Moreover, just like “Strength of belief”, “Importance of religion” is not significantly related with risk aversion. Lastly, all variables related to marriage are positively related with risk aversion in both regressions, where “Married” and “Divorced” are both only significant at 10% level, while “Widow” is significant at 5% level (see figure 5, Appendix B).
The results based on the ALP data are not the same as the results based on the LISS data. In the Netherlands, being Protestant or Catholic, and attending religious services seems to have a positive relationship with risk aversion, but this relation is not visible in the American data. Also, being divorced seems to have a negative effect on risk aversion in the Netherlands, while this effect seems to be less significant, but positive in America (see figure 5, Appendix B). In addition, being married or being a widow also has a significant positive relationship with risk aversion in America (although only at 10% level), but not in the Netherlands. This is an interesting pattern, but before discussing these differences a bit more, the results of regression three and four are given (see section three).
Just as before, figure 4 on page 18 contains the most important results, and figure 6 in Appendix B contains the results of regression three and four more extensively. As before, the results of the Dutch data are discussed first. Again, the results from the power utility and exponential utility are comparable. When adding subjective health status to regression two, the coefficient of “Catholic” stays significant in the power regression (again, only at 10% level: b = .077, p = .094). This coefficient was not significant in regression two for the exponential utility, and stays insignificant in regression three. Both the power and exponential regressions report an insignificant effect of subjective health status on risk aversion (p = .710 (power); p = .698
(exponential)). Regression four, performed using 2SLS, shows an insignificant effect of life satisfaction on risk aversion, for both the exponential and power utility. In this regression, the coefficient of “Catholic” is no longer significant (p = .101 (power)), but the coefficient for “More than once a week” stays significant for both the exponential and power utility (b = .190, p = .042 (power); b = .009, p = .039 (exponential)).
These results are partly in line with the theory and the three hypotheses. Based on the literature, a significant relation between subjective health status, life satisfaction and risk aversion would be expected, but the present results do not support this.
Although the estimated effect of life satisfaction is negative, it is not significant. This means that the Dutch data does not support the first hypothesis (see the end of section two).
Concerning the second hypothesis, it could be questioned whether it is supported by the Dutch data. The coefficient of “Catholic” increases from 0.75 to 0.77, when adding subjective health status to the regression (power utility). Although it indeed increases, it is a very small increase. Furthermore, these coefficients are only significant at 10% level, which leads to the conclusion that hypothesis two is also not supported by the Dutch data.
On the other hand, the third hypothesis of this paper is supported by the results of the Dutch data. The coefficient of “More than once a week” does not increase from regression two to three (for both the power and exponential utility), but it does increase from 0.181 to 0.190 (power utility) and from 0.008 to 0.009 (exponential utility) when
adding life satisfaction to the regression. However, the third hypothesis also states that the coefficient of subjective health status increases when adding life satisfaction to the regression. Although it does increase, it stays insignificant, so this part of hypothesis three is not supported by the Dutch data.
The results of the ALP data are again comparable, although there are slight differences. Adding subjective health status to the regression, decreases the significant coefficient of “Other religion” from 1.482 to 1.476 (p = .010), but adding life
satisfaction to the regression, increases it to 1.489 (p = .008). This means that
hypothesis two is not supported by the American data, but hypothesis three is. However, although the coefficient of subjective health status does increase when adding life satisfaction to the regression, it stays insignificant, meaning this part of hypothesis three is not supported by the ALP data. Furthermore, both subjective health status and life satisfaction are not significant, which is not in line with hypothesis one, as already explained for the Dutch data.
Concluding, both the Dutch and American data do not support hypothesis one, which states that the relation between life satisfaction and risk aversion is negative. The second hypothesis, which states that adding subjective health status to the regression, increases the coefficients of the variables measuring religion, is also not supported by the Dutch and American data. The third hypothesis is supported by both data sets, although the coefficient of subjective health status stays insignificant. Furthermore, when taking a look at the three sub questions that were introduced at the end of the introduction, it turns out that they can all be answered at this stage. The supposed relationship between life satisfaction and risk aversion, covered by the first sub question, is not significant in the analyzed datasets. The answer to the second sub question is in line with the theory: the relationship between religion and risk aversion is found positive. Finally, the third sub question can be answered negative. The results clearly show some differences between the Netherlands and America. The next section discusses these differences and further analyzes the results.
5 Analysis
This section continues where the last section left off. The differences between the Netherlands and America are reviewed and the results from the previous section are analyzed a bit more. In addition, the limitations of this research are discussed, and a recommendation is made for future research.
A similarity between the Dutch and American data is that both datasets reveal a positive relation between religion and risk aversion. The LISS data suggests that the positive effect of being religious on risk aversion goes via religious service attendance, and not via the strength of your belief in God. Possibly, certain teachings from the church influence someone’s view on risk. Another explanation could be that the social aspects of going to church influence risk aversion. These results are in line with the paper of Noussair et al. (2013). They also find a positive relation between attending religious services and risk aversion, and no relation between personal aspects of believing (strength of belief, prayer etc.) and risk aversion. However, Noussair et al. (2013) also find a significant positive relation between “Protestant” and risk aversion (when controlling for attendance), which the present results do not support.
In contrast, the American results concerning religion, are not in line with the paper from Noussair et al. (2013), who find no relation between “Other religion” and risk aversion. Even when controlling for religious service attendance in regression two, the positive coefficient of “Other religion” stays significant. Unfortunately, it is not possible to split “Other religion” into specific religions, as the ALP data does not contain this information. Apparently, the social aspects of believing do not influence risk aversion in America. Perhaps that for the religions in “Other religion”, a more personal aspect of believing, such as a strong personal relation with God, is important when shaping someone’s risk aversion. On the other hand, the variable “Importance of religion”, which is an indicator of how you value your relation with God, is not
significantly correlated with risk aversion, which does not support the theory that has just been put forward.
Another possible explanation for the different results in the Netherlands and America is that cultural differences play a role. These cultural differences can have two places of origin, of which only the second one is relevant. The first source of cultural differences comes from the religions in the category “Other religion”. It is possible that certain religions in this category find their foundation in eastern countries. These cultures differ very much from the western culture that dominates the Netherlands and America. However, if this cultural difference really plays a role, the coefficient of “Other religions” in the LISS data should also be significant, which is not the case. In other words, this cultural difference cannot explain the difference we find between the Netherlands and America.
Secondly, there exist cultural differences between the Netherlands and America. Certain traditions and customs in the Dutch and American culture can influence the way religion shapes risk aversion. Some studies report differences in risk perception between cultures. For example, Kapteyn and Teppa (2002) find that Dutch people are slightly less risk averse than American people. Zinkhan and Karande (1991) find that when it comes to financial decision making, American people are more risk averse than Spanish people. It is not ruled out that cultural differences explain the differences found between the Netherlands and America. For example, it is more natural to pay with a credit card in America than it is in the Netherlands. This way of paying brings more financial risk with it than paying with cash, as you create a debt.
Another interesting pattern where culture can play a role, are the different relations between marriage and risk aversion in the Netherlands and America. Being divorced seems to have a negative relation with risk aversion in the Netherlands (at 5% level), but a positive relation in America (only at 10% level). It is very likely that these differences are also present due to cultural differences. Perhaps that Dutch people who divorce, are less risk averse in the first place, causing the negative relation. Concerning America, maybe the people who divorce become more risk averse, because they want to avoid making the same mistake again of marrying someone of whom they divorce later. However, the purpose of the present study is not to explain the differences between the Netherlands and America concerning the explored relations. A suggestion for future
research is to explore the causes of the differences in risk perception between countries and in particular how traditions and customs of cultures can influence the relation between religion, marriage and risk aversion.
A similarity between the LISS data and the ALP data is that both subjective health status and life satisfaction do not have a relation with risk aversion. This result was not hypothesized, because it is not in line with the theory. There are several possible explanations for this. First, it could be that the measure of life satisfaction (SWLS) does not capture someone’s life satisfaction completely, despite the fact that it is generally found a good measure (Diener et al., 1985). To check this possibility, regression four is repeated for the American data, using another measure for life satisfaction. This measure exists of just one question, instead of five. The question is: “Taking all things together, how satisfied are you with your life as a whole these days?” On a scale from one to five, this question is answered. The left part of figure 7 in
Appendix B shows the results, obtained again using 2SLS, as in regression four in figure 5. Still, life satisfaction is not significantly related to risk aversion, so this gives more proof that life satisfaction is possibly unrelated to risk aversion.
Another explanation could be that the instruments used, do not explain life satisfaction enough. The right part of figure 7 contains another regression, where monthly income is added as an additional instrument for life satisfaction. This regression is performed for only the LISS and power utility data. The regression employs five income levels, of which the first one is left out, to omit collinearity. See the information under figure 7 to see which income levels have been used. The
coefficient of SWLS stays insignificant, which adds to the thought that life satisfaction and risk aversion are unrelated. Perhaps that the use of other or more instruments shows a different pattern. Concerning the insignificance of subjective health status, it is
perhaps too simplistic to measure it with just one question. Using more questions, like in the SWLS, could reveal a different pattern. Both these suggestions can be something for future research.
In addition to the limitations of this research that have just been discussed, are the limitations which were already mentioned in section four (see the one but last
paragraph of page 17). The fact that risk aversion is measured using only 5 questions is the most important limitation of this research. Perhaps that a more advanced measure would have given different results.
Concluding, the most important result that the data gives, is that the discovered relations between religion, risk aversion, subjective health status and life satisfaction cannot be easily applied to other countries. Possible reasons for this are the cultural differences that exist between countries. The influence of culture in these relations should thus not be underestimated. Furthermore, the relation between religion and risk aversion increases when controlling for subjective health status and life satisfaction, which is in line with the literature. Lastly, both subjective health status and life satisfaction seem to have no relation with risk aversion.
6 Conclusion
This section summarizes the most important aspects of the paper and tries to explicitly answer the main question. The research in this paper was about the influence of life satisfaction and subjective health status on the relation between religion and risk aversion. Knowing more about the behavior of this relationship could help insurance companies in making forecasts. While these companies are not allowed to use price discrimination between religious and non-religious people, forecasting the future claim behavior of policyholders can be improved when incorporating the relation between religion and risk aversion.
The main question of this paper was: to what extent is the positive relation between religion and risk aversion influenced by life satisfaction and subjective health status? As this question already suggests, the relation between religion and risk aversion is found to be positive in many studies. The relations between religion and life
satisfaction, religion and subjective health status, and subjective health status and life satisfaction are also found positive in several studies. Previous studies suggest that the relation between subjective health status and risk aversion is negative, and the relation
between life satisfaction and risk aversion is unclear. Because there are arguments for why this last relation goes in both directions, the effect of life satisfaction on risk aversion was estimated using 2SLS.
The answer to the above main question was found by estimating four
regressions, using two different datasets. One from the Netherlands (LISS), and one from America (ALP). The first two regressions focused on the relationship between religion and risk aversion, the third regression added subjective health status, and the fourth regression added life satisfaction.
It was found, in line with the theory, that religion and risk aversion are positively related. The Dutch data showed that this positive relation is likely driven by the fact if someone attends religious services (which could be related to certain teachings in the church), while the American data did not clearly show what caused the positive relation. The latter merely showed a relationship between “Other religion” and risk aversion. This paper did not find out what caused this relation, and why this same relation is not found in the Dutch data, but it was speculated that is has something to do with the cultural differences between the Netherlands and America. It is possible that certain traditions and customs of both countries influence the way people deal with risky choices. However, it was not the purpose of this paper to fully uncover the consequences of the cultural differences.
Furthermore, both datasets showed no significant relation between subjective health status and life satisfaction on one hand, and risk aversion on the other. This finding is not in line with the theory. Using an extra instrument for life satisfaction did not change this result. Perhaps that using a different measure for subjective health status or using more instruments for life satisfaction would have given a different result.
Adding life satisfaction and subjective health status to the regression of risk aversion on religion, increased (in both the Dutch and American data) the relation between religion and risk aversion. Despite the fact that it is hard to draw conclusions based on the magnitude of the relation between risk aversion and religion, it appears that these two elements are stronger related to each other when controlling for life satisfaction and subjective health status. The answer to the main question thus is that the
relation between religion and risk aversion is positively influenced by life satisfaction and subjective health status.
The last notable result from this paper was the fact that marriage and risk aversion seem to have a different relation in the Netherlands and America. Being divorced correlated negatively with risk aversion in the Netherlands, but positively in America. Again, it is possible that cultural differences affect the way people deal with a divorce, and changes the way this correlates with risk aversion.
Both the relations between religion and risk aversion, and marriage and risk aversion, differ between the Netherlands and America. Perhaps, the fact that the Dutch and American data find different significant relations is the most important result of this paper. It appears that drawing conclusions based on data from just one country, does not give enough proof for the existence of certain relations in other countries. The effect of culture on the way things are related to each other and people behave is something that should thus not be underestimated. In other words, a serious look should be taken at the external validity of studies when drawing conclusions.
Future research in general, and of course in particular about the relations that were researched in this paper, should always take the effect of culture into account and preferably be performed in more countries to give a good foundation upon which conclusions can be drawn. As regards the relations examined in this paper, future research could extent further and in more detail on the way culture affect people’s view on risk, and how this is related to religion.
References
Bellante, D., & Green, C. A. (2004). Relative risk aversion among the elderly. Review of Financial Economics, 13(3), 269-281.
Bjørnskov, C., Dreher, A., & Fischer, J. A. (2008). Cross-country determinants of life satisfaction: Exploring different determinants across groups in society. Social Choice and Welfare, 30(1), 119-173.
Diener, E. D., Emmons, R. A., Larsen, R. J., & Griffin, S. (1985). The satisfaction with life scale. Journal of personality assessment, 49(1), 71-75.
Dohmen, T. J., Falk, A., Huffman, D., Sunde, U., Schupp, J., & Wagner, G. G. (2005). Individual risk attitudes: New evidence from a large, representative,
experimentally-validated survey (Discussion Paper No. 1730). Retrieved from Social Science Research Network website:
http://papers.ssrn.com/sol3/papers.cfm?abstract_id=807408
Dohmen, T. J., Falk, A., Huffman, D., Sunde, U., Schupp, J., & Wagner, G. G. (2011). Individual risk attitudes: Measurement, determinants, and behavioral
consequences. Journal of the European Economic Association, 9(3), 522-550. Dorahy, M. J., Lewis, C. A., Schumaker, J. F., Akuamoah-Boateng, R., Duze, M. C., &
Sibiya, T. E. (1998). A cross-cultural analysis of religion and life satisfaction. Mental Health, Religion & Culture, 1(1), 37-43.
Elliott, M., & Hayward, R. D. (2009). Religion and Life Satisfaction Worldwide: The Role of Government Regulation*. Sociology of Religion, 70(3), 285-310. Ferraro, K. F., & Albrecht-Jensen, C. M. (1991). Does religion influence adult
health? Journal for the scientific study of religion, 30(2), 193-202.
Gwozdz, W., & Sousa-Poza, A. (2010). Ageing, health and life satisfaction of the oldest old: An analysis for Germany. Social Indicators Research, 97(3), 397-417. Hadaway, C. K. (1978). Life satisfaction and religion: A reanalysis. Social
Forces, 57(2), 636-643.
Halek, M., & Eisenhauer, J. G. (2001). Demography of risk aversion. Journal of Risk and Insurance, 68(1), 1-24.
Hilary, G., & Hui, K. W. (2009). Does religion matter in corporate decision making in America? Journal of Financial Economics, 93(3), 455–473.
Holt, C. A., & Laury, S. K. (2002). Risk aversion and incentive effects. American economic review, 92(5), 1644-1655.
Jianakoplos, N. A., & Bernasek, A. (1998). Are women more risk averse? Economic inquiry, 36(4), 620-630.
Kapteyn, A., & Teppa, F. (2002). Subjective measures of risk aversion and portfolio choice. Retrieved June 3, 2015, from
http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.320.8575&rep=rep1& type=pdf
Lim, C., & Putnam, R. D. (2010). Religion, social networks, and life satisfaction. American Sociological Review, 75(6), 914-933.
Melin, R., Fugl-Meyer, K. S., & Fugl-Meyer, A. R. (2003). Life satisfaction in 18-to 64-year-old Swedes: In relation to education, employment situation, health and physical activity. Journal of rehabilitation medicine, 35(2), 84-90.
Morin, R. A., & Suarez, A. F. (1983). Risk aversion revisited. The Journal of Finance, 38(4), 1201-1216.
Musick, M. A. (1996). Religion and subjective health among black and white elders. Journal of Health and Social Behavior, 37(3), 221-237.
Noussair, C. N., Trautmann, S. T., Van de Kuilen, G., & Vellekoop, N. (2013). Risk aversion and religion. Journal of Risk and Uncertainty, 47(2), 165-183. Palmore, E., & Luikart, C. (1972). Health and social factors related to life
satisfaction. Journal of Health and Social Behavior, 13(1), 68-80.
Pavot, W., & Diener, E. (1993). Review of the satisfaction with life scale. Psychological assessment, 5(2), 164-172.
Proctor, C. L., Linley, P. A., & Maltby, J. (2009). Youth life satisfaction: A review of the literature. Journal of Happiness Studies, 10(5), 583-630.
Rijksoverheid (n.d.). Ontstaan kredietcrisis. Retrieved May 8, 2015, from
Sinha, J. W., Cnaan, R. A., & Gelles, R. J. (2007). Adolescent risk behaviors and
religion: Findings from a national study. Journal of adolescence, 30(2), 231-249. Wakker, P. (2010). Prospect theory: For risk and ambiguity. Cambridge: Cambridge
University Press.
Wang, H., & Hanna, S. D. (1997). Does risk tolerance decrease with age? Financial Counseling and Planning, 8(2), 27-32.
Zinkhan, G. M., & Karande, K. W. (1991). Cultural and gender differences in risk-taking behavior among American and Spanish decision makers. The Journal of Social Psychology, 131(5), 741-742.
Appendix A
This appendix gives an example of how to estimate someone’s parameter 𝜃, if it is assumed that this individual values choices according to a utility function from the power family. The power family, for 𝜃 > 0, is defined as 𝑈(𝑥) = 𝑥𝜃. Suppose that someone who is given the choice between receiving €5 or €65, both with a change of 50%, and receiving a sure payoff of €30, prefers the lottery. This means that for this person 0.5 × 𝑈(5) + 0.5 × 𝑈(65) ≥ 𝑈(30). Now suppose that he would prefer a sure payoff of €35 above the lottery. This would lead to the conclusion that 0.5 × 𝑈(5) + 0.5 × 𝑈(65) ≤ 𝑈(35). One of the basic assumptions of utility functions is that they are strictly increasing (Wakker, 2010). This means that, since 0.5 × 𝑈(5) + 0.5 × 𝑈(65) is just a number, there is exactly one unique value of 𝑎 where the following applies: 0.5 × 𝑈(5) + 0.5 × 𝑈(65) = 𝑈(𝑎), where 𝑎 must be a value between 30 and 35. This value is called the certainty equivalent. Because it is impossible to know the value of the certainty equivalent exactly with only the former information, it is supposed that it is exactly in the middle of 30 and 35. All that remains to do is calculating the unique value of 𝜃, by solving the equation 0.5 × 5θ+ 0.5 × 65θ= 32.5θ.
When assuming a utility function from the exponential family, the procedure is similar. For the power family, the value of 1 − 𝜃 is used as measure of risk aversion (relative risk aversion), and for the exponential family, the value of 𝜃 is used as a measure of risk aversion (absolute risk aversion).
Appendix B
LISS ALP
Variable # obs. Mean # obs. Mean
Denomination & Attendance
Religious 1586 42.4% 1023 81.2%
Protestant 1586 15.6% 1023 36.9%
Catholic 1586 21.4% 1023 19.9%
Other 1586 5.3% 1023 24.4%
Never 1586 53.5% 1023 30.3%
One or more times a year 1586 28.8% 1023 27.8%
Two or three times a month 1586 7.0% 1023 10.6%
Once a week 1586 7.5% 1023 18.9%
More than once a week 1586 3.6% 1023 12.5%
Strength belief 1586 3.5 (1.83) - -
Importance of belief - - 1023 2.7 (0.44)
Other variables
Age 1586 49.7 (16.25) 1023 58.0 (7.80)
Female 1586 53.7% 1023 58.8%
Subjective health status 1586 3.1 (0.74) 1023 3.4 (0.92)
SWLS 1586 25.3 (5.54) 1023 23.3 (8.00) Married 1586 65.0% 1023 68.4% Divorced 1586 8.5% 1023 17.5% Widow 1586 3.5% 1023 4.1% Paid job 1586 52.7% 1023 61.6% Student 1586 8.3% - - Retired 1586 18.6% 1023 22.7% Other work 1586 14.1% 1023 7.3% 2. Summary statistics
LISS ALP
Risk aversion # obs. Mean St. dev. # obs. Mean St. dev.
Exponential (𝜃) All 1586 0.033 0.026 - - - Real 600 0.029 0.027 - - - Hypothetical 986 0.035 0.026 - - - Power (1-𝜃) All 1586 0.635 0.585 1023 6.738 5.561 Real 600 0.563 0.606 - - - Hypothetical 986 0.679 0.567 - - -
Dependent variable: θ (exponential), 1 − θ (power).
Significance at 10%, 5%, and 1% level is indicated with *, **, and ***, respectively.
LISS ALP Variable Power (1) (2) Exponential (1) (2) Power (1) (2) Married -.004 (.044) -.002 (.044) -.0001 (.002) .0000 (.002) 1.126* (.576) 1.091* (.580) Divorced -.156** (.065) -.145** (.065) -.007** (.003) -.006** (.003) 1.205* (.676) 1.223* (.677) Widow .050 (.093) .065 (.093) .002 (.004) .002 (.004) 2.304** (1.000) 2.258** (1.002) Protestant .078* (.042) .051 (.055) .003* (.002) .002 (.002) .256 (.488) .354 (.536) Catholic .060 (.038) .075* (.046) .002 (.002) .003 (.002) .394 (.545) .572 (.590) Other religion .061 (.066) .004 (.078) .002 (.003) .0000 (.004) 1.420*** (.524) 1.482*** (.541) Once a week - .080 (.072) - .004 (.003) - -.301 (.546) More than once
a week - .181* (.093) - .008** (.004) - .476 (.616) Strength of belief - -.003 (.011) - -.0002 (.0005) - - Importance of religion - - - .069 (.413) 𝑅2 0.0360 0.0404 0.0352 0.0394 0.0638 0.0690
Dependent variable: θ (exponential), 1-θ (power).
Significance at 10%, 5%, and 1% level is indicated with *, **, and ***, respectively.
LISS ALP Variable Power (3) (4) Exponential (3) (4) Power (3) (4) Married -.002 (.044) .008 (.046) -.000 (.002) .0004 (.002) 1.101* (.581) 1.265* (.670) Divorced -.147** (.065) -.169** (.071) -.006** (.003) -.007** (.003) 1.223* (.678) 1.197* (.674) Widow .065 (.093) .060 (.093) .002 (.004) .002 (.004) 2.262** (1.003) 2.362** (1.015) Protestant .050 (.055) .048 (.055) .002 (.002) .002 (.002) .359 (.536) .412 (.543) Catholic .077* (.046) .075 (.046) .003 (.002) .003 (.002) .579 (.591) .656 (.593) Other religion .003 (.078) -.012 (.081) .0000 (.004) -.001 (.004) 1.476*** (.568) 1.489*** (.564) Once a week .081 (.072) .090 (.073) .004 (.003) .004 (.003) -.297 (547) -.169 (.605) More than once a week .181* (.093) .190** (.094) .008** (.004) .009** (.004) .459 (.619) .507 (.621) Strength belief -.003 (.011) -.003 (.011) -.0002 (.0005) -.0002 (.0004) - - Importance of religion - - - - .070 (.414) .144 (.439) Subjective health status -.008 (.020) .016 (.038) -.0004 (.0009) .0006 (.0005) -.047 (.188) .155 (.462) SWLS - -.009 (.013) - -.0004 (.0006) - .061 (.128) 𝑅2 0.405 0.0391 0.0395 0.0381 0.0690 0.0724
