June 26th 2015
Z. Huang 2014-2015
Bachelor thesis 2nd semester
Econometrics
University of Amsterdam
First draft
Maurits van Vliet 10428690
What is the influence of civil status on risk attitude?
The differences in risk attitude between several civil statusesAbstract
In this paper I investigate the relationship between civil status and risk attitude of individuals using a Dutch sample, based on data obtained by CentERdata, an affiliation of Tilburg University. The differences between the risk attitude of men and women for several civil statuses is examined. Several regressions are performed, including logit and least squares regressions. From these regressions significant and remarkable results are obtained. The results show that divorced individuals are the most risk-seeking sub-group, compared to married, widowed or never married individuals. Next to this, the deviation in risk attitude is higher for men than for women when compared between sub-groups, i.e., divorced versus never married individuals. Finally, a high correlation between age and the probability of being divorced or widowed, when compared to never married individuals, is found.
Introduction………..1
2 Previous research……….….3
2.1 Results from previous research……….….3
2.2 Measurements of risk aversion………..…5
2.3 Discussion………..6
3 Research design………8
3.1 Research question ………..8
3.2 Data and measure of risk aversion……….8
3.3 Models...………10
4 Results………..11
4.1 Married versus non-married..………12
4.2 Married versus divorced………13
4.3 Married versus widowed………...15
4.4 Married versus never married………16
4.5 Divorced versus widowed………..16
4.6 Divorced versus never married………..……17
4.7 Widowed versus never married……….20
4.8 Logit regression……….20
4.9 Discussion ………..21
5 Conclusion………24
Reference……….26
Appendix……….27
The differences in risk attitude between several civil statuses
Individuals with a low risk tolerance may face difficulties in achieving their financial goals, due to a low return on their investment as a result of their investment choices. This is of importance since people need to build up sufficient capital from which they can retract an adequate retirement. While on the other hand some individuals have a very high-risk tolerance, which causes unnecessary financial losses. An extreme example here is people with a gamble addiction. The relation between characteristics of individuals and risk attitude can be used to predict someone’s risk profile. These predictions can be of importance to a lot of scientific fields. Risk attitudes influences not only financial decisions but also decisions of patients in healthcare and, study choices of students, etc. An adequate insight into the relation between marriage and the risk profile of married couples is of importance since it can help for example financial institutions to offer adequate financial products for married couples with which they can realize their retirement objectives.
Many researchers have investigated the demography of risk aversion and found that demographic characteristics and economic characteristics have a significant effect on one’s financial risk tolerance. Halek and Eisenhauer (2001) examine attitudes to risk across various segments of the population. They find significant effects of amongst other, migration, gender and employment on risk aversion. Next to Halek and Eisenhauer (2001) there are various researchers that have examined the effect of marital status on risk aversion.
For instance, Yao and Hanna (2005) suggest that people who have longer life expectancies will be more risk tolerant according to a rational economic model. Yao and Hanna (2005, p. 68), under the application of the rational economic theory, assume that married people have longer life expectancies and so will be more risk tolerant than singles of the same age. Hinz, McCarthy and Turner (1997) also suggest that married people have longer life expectancies, but they are not sure what the direction of this effect is. They state that marriage provides insurance through income pooling. This could cause that married couples would be willing to take greater risks than non-married couples. So there are some theories about the effect of marital status on risk aversion. The question is whether these theories hold in practice.
Few studies find the negative effect of marriage on risk aversion that is expected by the rational economic model. Gutter, Fox and Montalto (1999) state that white households headed by an unmarried householder are less likely to hold risky assets than otherwise similar
households headed by a married couple. While many other studies find a positive effect of marriage on risk aversion, which are discussed in the following section.
In the research of Yao and Hannah (2005) and Sunden and Surette (1998) significant effects of gender on risk aversion are found. Next to this effect they also identify interaction between marriage and gender. This interaction variable affects risk aversion. For this reason Sunden and Surette include gender and the interaction variable of marriage and gender as control variables.
This study investigates the effect that civil status has on someone’s risk aversion. The question is, however, whether civil status affects risk aversion or vice versa. Although many studies find a significant, positive effect of marriage on risk attitude, it could be the case that people with a lower risk profile are more likely to be married. Hence, there may consists reverse causality. This is further discussed in later sections. In this paper I make use of data of the LISS panel administered by CentERdata. It contains a large number of background information on a representative sample of Dutch population. The link between civil status and risk aversion is estimated by investigating the descriptive statistics and OLS- and logit-regressions are performed.
The next session discusses some previous studies on the effect of marriage on risk aversion. The results from these previous studies are compared. The results from other studies are used to determine the experiment design used in my thesis. Section 3 illustrates measurements of variables and models that are used. After that, in section 4, the results are analyzed and discussed. Finally, in the last section, a conclusion is drawn.
This section discusses previous research about the relationship between civil status and risk attitude. First the results found in these researches are discussed. After that the measurements of risk aversion used in previous research are discussed.
2.1 Results from previous research
The first section provides theories on the effect that marriage has on risk aversion. One of the studies that discusses the rational economic model is done by Yao and Hanna (2005).
According to the rational economic model people with longer life expectancies are more risk tolerant. Since it is assumed that married couples have longer life expectancies they should be less risk averse than other singles of the same age (Yao and Hanna (2005, p. 68)). The
question is whether these theories hold in practice. Yao and Hanna (2005) examine the effect of gender and marriage on risk aversion and the interaction of these two variables.
One hypothesis that Yao and Hanna (2005) deduce is that married respondents are more risk tolerant than unmarried respondents. Their hypothesis is supported by the rational economic model, which expects that people who have longer life expectancies are more risk tolerant. Since married people have longer life expectancies (Yao and Hanna (2005)), they should be more risk tolerant than non-married people. Their hypothesis is only proven for some of the comparisons. For males, 65.9 percent of married males were willing to take some risk versus only 61.9 percent of unmarried males that were willing to take such risk. On the other hand, married males were significantly less likely to take substantial risk than
unmarried males (4.5 percent vs. 7.4 percent). For females, more than half of the married females were willing to take high risk, and only 40.8 percent of unmarried females were willing to take the same amount of risk.
Other research that examines the interaction between gender and marriage is done by Sunden and Surette (1998). They conclude that gender interacts with marital status and has an effect on household’s investment choices. Sunden and Surette (1998) investigate gender differences in the allocation of assets in retirement savings plans. Their sample consists of individuals working, covered by a defined contribution plan and under the age of 75.3. The results of Sunden and Surette (1998) demonstrate that investment decisions seem to be driven more by a combination of marital status and gender, rather than by gender alone. Sunden and Surette (1998) find that married men are less likely than single men to choose ‘mostly stocks’ and married women are more likely than single women to choose ‘mostly bonds’.
Estimates from models that omit the interaction variable of gender and marital status suggest that marital status has no effect on investment decisions (Sunden and Surette (1998),
p. 209). However, their results show clearly that marital status matters and interacts in important ways with gender.
Both Sunden and Surette (1998) and Yao and Hanna (2005) find that marriage
interacts with gender and this interaction variable affects financial decisions. The direction of the effects depends on gender and the amount of risk that is associated with the financial decision.
Hinz, McCarthy and Turner (1997) point out the predictions of the rational economic model like Yao and Hanna (2005). While Yao and Hanna (2005) find evidence that supports the predictions of the rational economic model, Hinz et al. (1997) find no evidence that supports these predictions. Hinz et al. (1997) compare married people to people who are divorced, never married or widowed. Their results show a significantly positive effect of marriage on risk aversion: married people are much less likely to invest either in stock or the fixed income effect.
Next to Hinz et al. (1997), Hartog, Ferrer-i-Carbonell and Jonker (2002) also find a positive relationship between marriage and risk aversion. The cost of breaking up the relationship increases when individuals get married, thus it makes sense to expect that married individuals are more risk averse, as they are the ones who will be more eager to reduce the risk of the partners running off (Hartog et al. (2002, p.16)).
Hinz et al. (1997) examine the differences between married men and unmarried women; they find that married men and unmarried women take similar investment risks. This result is in conflict with the results obtained by Sunden and Surette (1998) and Yao and Hanna (2005) who find that the effect of marriage on risk aversion also depends on gender.
Other researches examine the effect of marriage on risk aversion within sub groups of the population. Schooley and Worden (1996) use data from the 1989 Survey of Consumer Finances to examine household’s attitudes toward risk, a measure of their relative level of risk taking. While Gutter, Fox and Montalto (1999) investigate racial differences in investor decision making. Gutter et al. (1999) use data form the 1995 Survey of Consumer Finances.
An examination of marital status reveals that single respondents have significantly fewer risky assets per dollar of wealth than other households (Schooley and Worden, p. 92). The existence of increasing relative risk aversion would be an explanation since married couple households are more likely to have two incomes and thus a larger amount of human capital.
Gutter et al. (1999) find that white households are more likely to be married than black households (55.19% vs. 29.9%). The majority of black households is headed by single females (51%). For white households, they obtain a significant effect of marriage on risk attitude among white households; white households headed by an unmarried householder are less likely to hold risky assets than otherwise similar households headed by a married couple.
Subsequent research that finds a significant effect of marriage on risk aversion is the research of Halek and Eisenhauer (2001). Halek and Eisenhauer (2001) examine differences in risk attitudes across demographic groups based on survey responses. In their first model Halek and Eisenhauer (2001) include marriage and they obtain a significant coefficient of 29.28. Which means that marriage increases risk aversion by 29.28%. Halek and Eisenhauer (2001) argue the direction of the relationship between marriage and risk aversion. While it can be argued that marriage affects one’s risk aversion, it may also be argued that more risk-averse individuals choose to marry. This possible reverse causality could damage the internal validity of the model. Consequently, Halek and Eisenhauer (2001) included marriage in their first model but omitted marriage from their second model.
2.2 Measurements of risk aversion
In this section studies are discusses that have estimated a measurement for risk aversion. One possible method to measure risk aversion is by survey questions. Yao and Hanna (2005) use survey questions to measure one’s risk aversion. Respondents are asked whether they are willing to take substantial risk, above average, average or no risk to earn an amount of returns equivalent to the risk they take.
Halek and Eisenhauer (2001) use survey data on life insurance purchases to estimate the risk aversion parameter empirically for individual households. Schooley and Worden (1996) obtain a measure of individual risk tolerance from the actual household portfolio allocation.
Hartog et al. (2002) ask individuals to state the reservation price for a lottery ticket, after specifying the probability of winning a prize of particular magnitude. Using expected utility theory, it is then straightforward to deduce the Arrow-Pratt measure of risk aversion. A study that also uses lotteries to measure one’s risk aversion is the study by Noussair,
Trautmann, van de Kuilen and Vellekoop (2013). Noussair et al. (2013) investigate the influence of religion on risk aversion. Noussair et al. (2013) measure risk attitude in an experiment, by letting participants choose between a lottery and a safe amount of cash. The
amount of safe choices, choosing the safe amount of cash, indicates whether participants are risk averse, neutral or seeking. Some of the participants have real payouts, while others have hypothetical payouts. The problem with this measurement is that it can not be assumed that the difference in risk aversion between one safe choice and two safe choices is as large as the difference in risk aversion between three safe choices and four safe choices, and so on.
Wakker (2008) gives, in his book, two ways to measure risk aversion via utility functions; via the power utility function and via the exponential utility function. Power utility, often called constant relative risk aversion (CRRA):
For a>0:
1-θ >0: U(a)= a^ θ 1-θ =0: U(a)= ln(a) 1-θ <0: U(a)= - a^ θ
The smaller θ, the more U is concave. Concave utility is an indication of risk aversion, neutral utility correspondents to risk neutrality and a person with convex utility is risk seeking.
Exponential utility, often called constant absolute risk aversion (CARA): For θ >0: U(a)= 1 - e^(-θa)
θ =0: U(a)= a
θ <0: U(a)= e^ (-θa) - 1
θ is an index of concavity: θ >0 concave utility θ =0 linear utility θ <0 convex utility
When utility functions are known or estimated it is possible to measure risk aversion by . It may be difficult to measure utility functions as these functions may differ among individuals.
2.3 Discussion
It is clear that there are many studies that examine the effect of marriage on risk aversion. In the studies discussed, various measurements of risk aversion are used. The most used
measurements are computed from lotteries and survey questions. Lotteries are more reliable measures of risk aversion since it comes closer to reality than survey questions. Individuals can see themselves as more risk averse or seeking than they in reality are, which makes survey questions a less reliable method.
The direction of the effect that marriage has on risk aversion differs among these studies. Gutter et al. (1999) and Schooley and Worden (1996) both find a negative effect of marriage on risk aversion within sub-groups of the population. On the other hand Sunden and
Surette (1998) and Yao and Hanna (2005) do not find a clear direction of the effect of marriage on risk aversion. However, they both find that gender interacts with marriage and this interaction variable has a significant effect on one’s risk aversion. There are also studies that obtain a positive relationship between marriage and risk aversion. Among these are the studies by Hartog et al. (2002) and Hinz et al. (1997).
The studies that are discussed in this section all investigate the influence on people characteristics on risk aversion. Except for Yao and Hanna (2005) all the discussed papers examine another characteristic than marriage as main topic. The results found by previous research are used to determine the model that I use in my research. This model is further discussed in the next section.
3 Experiment design
The results and models from previous research are used to develop a model for this thesis. In this section the data is discussed first. After that, the measure of risk aversion is mentioned and the computation is explained. Finally the model is presented.
The main question in my thesis is what the differences in risk aversion are between several civil statuses. Yao and Hanna (2005) and Sunden and Surette (1998) conclude that gender and the interaction variable of marriage and gender have a significant effect on risk aversion. Because of their findings gender is included as variable and risk aversion among men and women is studied. The difference between men’s risk aversion and women’s risk aversion caused by civil status is a sub question in my research and therefore discusses in the next section. Next to gender also age, net income and education are included as explanatory variables.
3.2 Data and measure of risk aversion
In this thesis the LISS data obtained by CentERdata, an affiliation of Tilburg University, is used. It contains information of approximately 9000 individuals. The main topic of this research is the effect of marriage on risk aversion. To test the effect of marriage on risk aversion a variable of risk aversion and marriage is needed. For the variable of marriage a dummy is used, this is easy to obtain as the LISS data contains information about marital status. For the variable of gender another dummy is created, this dummy equals one when a respondent is male and equals zero when a respondent is female.
For the last three variables, age, net income and education, category variables are used. Age differs from 14 year and younger to 65 years and older. The ages in-between 14 and 65 are divided into categories of ten years. Net income is split into categories of 500 euro’s. For education there are six increasing options, from primary school to university. Respondent’s that did not stated what their income of highest received education was were removed of the sample.
Since lotteries are more reliable than survey questions, this is used as measure of risk aversion. As pointed out in the section two, there are some problems with the measurement used by Noussair et al. (2013). The lottery Noussair et al. (2013) use is also used in this paper to compute a measure of risk aversion, but in another way.
Noussair et al. (2013) let each participant choose, in five trials, between a lottery that paid €65 or €5 with equal probabilities and a sure payoff that differed by trial. The expected value of the lottery is €35. The sure payoff varied from €20 to €40 in steps of €5. Noussair et al. (2013) counted the amount of safe choices (choosing the sure payoff instead of the
lottery). Noussair et al suggest that a respondent is risk averse when he or she makes two safe choices. A problem with this measurement is that the difference in risk aversion between two and three safe choices is not equal to the difference between four and five safe choices.
Furthermore, the basis on which the assumption is made whether a respondent is risk averse or not may be interpreted different by another person.
In my thesis the same trials with respondent’s choices are used. For every respondent a certainty equivalent for the lottery is computed. Suppose a respondent prefers the lottery above €X but €Y above the lottery, then there is an amount of cash in-between X and Y for which the respondent is indifferent between the sure payoff and the lottery. When this is the case it is assumed that the respondent is indifferent between the lottery and ½*X + ½*Y, his or her certainty equivalent, for X and Y in-between 20 and 40. Problems arise when a respondent has only safe choices or no safe choices because there is no switching point. In the case that the respondent makes only safe choices we assume that the respondent has his switching point in-between €15 and €20, so his certainty equivalent amounts €17.50. The same solution is used when the respondent has no safe choices, than his certainty equivalent is €42.50.
The certainty equivalents of respondents are used to obtain measurements of individual risk aversion via the power utility and exponential utility functions that are discussed in section two. The equation 1/2*U(5) +1/2*U(65) = U(CE) is solved for every respondent, with CE the certainty equivalent of the respondent and U the utility function. There is only one theta for which this equation is solved; this theta is the measure of risk aversion for the corresponding respondent. Using both the power utility and exponential utility function gives two measures of risk aversion for every respondent. These two measures are used in the regressions and the outcomes are compared and analyzed.
3.3 Models
Among the several civil statuses some regressions are conducted. Each civil status is
compared to the other three civil statuses. In these cases three regressions are performed. The first (1) regression includes only a dummy variable for civil status as explanatory variable and risk attitude as dependent variable, to examine the effect of marriage on risk aversion. The measurement of the variable for civil status explains the difference in risk aversion between the two statuses that are compared. In the second (2) regression gender is added as explanatory variable to control for the effect that gender has on risk aversion. The third (3) regression also includes age, net income and education as explanatory variables next to gender and civil status. Finally, the first (1) and third (3) regression are also performed among subgroups, containing only men and women; (4), (5), (6) and (7). By analyzing these regressions the influence of civil on a man’s risk aversion and woman’s risk aversion can be compared. By adding more explanatory variables a better measure of the effect of marriage on risk aversion can be derived. The models mentioned in this subsection are given below:
(1) θ = α + β1*Dcivil
(2) θ = α + β1*Dcivil + β2*Dgender
(3) θ = α + β1*Dcivil + β2*Dgender + β3*Age + β4*Netincome + B5*Education For male: Dgender = 1 and (4) θ = α + β1*Dcivil
(5) θ = α + β1*Dcivil + β3*Age + β4*Netincome + B5*Education
For female: Dgender = 0 and (6) θ = α + β1*Dcivil
(7) θ = α + β1*Dcivil + β3*Age + β4*Netincome + B5*Education
4 Results
Results from prior research are used to develop the model presented in section 3. Section 2 points out that there may be reverse causality. The problem is that individuals who get married, divorced, widowed or separated are possibly already more risk averse before their civil status changes. This reverse causality is further discussed in this section. This section discusses the descriptive statistics and describes the results that are obtained through regressions. First, multiple cases are evaluated. Subsequently, the results are discussed and the model is critically evaluated.
As there are multiple civil statuses, there are several cases that can be compared. In the questionnaire on internet respondents have five options: (a) married, (b) separated, (c) divorced, (d) widow or widower and (e) never married. Since there is only one respondent that has separated as civil status, separated status is not compared with other civil statuses. Several cases are investigated. For each case the seven regressions mentioned in section 3 are performed. The first regression includes only civil status as explanatory variable.
Subsequently, in the second regression a dummy for gender is included. The third regression includes also category variables of age, net income and education. Finally, regressions among subgroups containing only men and women are performed. In order to investigate whether marriage and other civil statuses have a different effect on male’s risk aversion than female’s risk aversion, I look for differences in the influence of civil status on risk attitude between male and female. Finally, I discuss the results and explain the direction of the coefficients and the internal and external validity of the model that I use.
Using the descriptive statistics, presented in table 1, the average risk aversion within groups can be compared. The measure of risk aversion computed by the power utility function gives an average theta of 0.322 among the divorced respondents, 0.216 for the widowed, 0.179 for the married and 0.137 for the never married. An individual with a higher theta is more risk seeking than an individual with a lower theta. As a result it can be
concluded that divorced individuals are the most risk seeking, followed by widowed, married and non-married respectively. However, the question is whether this is caused by civil status or other factors. This is further investigated by regressions. In the sequel, unless indicated otherwise, I refer to the measure of risk attitude as the one computed by the power utility function, since this measure gives the most significant relations. This measure is also named theta, since that is the common parameter for risk attitude.
Descriptive statistics married separated divorced widow/widower never married
All 1372 1 202 122 619
female 669 0 91 35 278
male 703 1 111 87 341
Average power theta 0,178855685 1,17 0,322029703 0,215983607 0,137318255 Average expo theta 0,012069971 -0,056 -0,007153465 0,00445082 0,021807754
Table 1. Descriptive statistics: average theta’s computed by the power and exponential utility function and the number of males, females and all respondents per civil status.
4.1 Married versus non-married
From descriptive statistics it can be concluded that the average theta among non-married is 0.187. This theta is higher than the average theta among married individuals (0.179). I create a dummy, which equals one for married individuals and equals zero for non-married
individuals, this dummy is regressed on the measures of risk aversion. The regression of this dummy on the three measures of risk aversion gives insignificant coefficients. When more variables are added, the regression returns a more significant coefficient. In the second regression gender is included as variable. Although the coefficient of gender is significant, the dummy for marriage is not. In the last regression there are three category variables added. In this regression I find a significant coefficient for marriage on risk aversion, at the 10% rejection interval. For both measures of risk attitude a negative relation between marriage and risk attitude is found. This means that married individuals are more risk averse than non-married individuals.
Among men there is no significant difference in risk aversion found between married and non-married individuals. This is the case for the regression of only civil status on risk aversion as for the regression that includes all variables. However, the insignificant
coefficient has the same direction as the coefficient in the regression among all individuals. Although the p-values found in the regression among females are more significant than the ones found in the regression among males, the results are still not significant. The direction of the coefficient of the marriage dummy in the regression among women does not equal the direction of the dummy in the regression among all individuals for the power utility measure of risk aversion, while it does for the one computed by exponential utility.
4.2 Married versus divorced
A dummy that equals one when a respondent is married and equals zero when a respondent is divorced is used in this regression. The regression of civil status dummy on the power utility measure of risk aversion gives a very significant coefficient of -0.124. Hence, a married individual is 12.4 percent more risk averse than a divorced individual. Including gender, age, education and net income does not change the coefficient of the civil status dummy much. In this regression a coefficient of -0.123 is found. Controlled for the variables gender, age, education and net income there is still a significant effect of marriage on risk aversion found, when compared with divorced individuals. The direction of the coefficient is the same in the regressions on risk attitude measured by the exponential utility function.
When controlled for all variables, a significant coefficient, at the 10 percent rejection interval, is found in the regression among all men. For the power utility function measure of risk attitude, a decrease of 13.8 percent in risk attitude is found for married individuals compared with divorced ones. This means that married men are 13.8 percent more risk averse than divorced individuals.
For women, when controlled for age, education and net income we find a less significant coefficient than in the regression that only includes marriage as variable. The coefficient in the first regression, with only marriage as explanatory variable, is significant at the 5 percent rejection interval. While the coefficient of marriage in the regression including age, education and net income is significant at the 10 percent rejection interval. However, controlled for age, education and net income, a decrease of 11.3 percent in risk attitude is found for a married woman compared with a divorced woman. Thus, a married woman is 11.3 percent more risk averse than a divorced woman. When the exponential utility function measure is used, and controlled for age, education and net income, there is a decrease of 0.55 percent in risk attitude found for a married woman, compared with a divorced woman.
Overall, there is more difference in risk aversion between married individuals and divorced individuals for men than for women. This explains the coefficient for marriage among all individuals, which lies in-between the one for men and the one for women.
Table 2. The regression in the case married versus divorced, including all respondents. Theta computed by power utility function.
Table 3. The regression in the case married versus divorced, including only male respondents. Theta computed by power utility function.
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.005 0.020 0.022 Observations 1487 1487 1487 (11.28) (9.38) (3.54) Constant 0.486*** 0.421*** 0.341*** (-0.41) education -0.00478 (1.73) net income 0 .0184 (0.69) age 0.00902 (4.80) (2.74) male 0.148*** 0.105** (-2.68) (-2.86) (-2.67) married -0.124** -0.131** -0.123** Model 1 Model 2 Model 3 Married vs divorced(Power) * p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.005 0.011 Observations 716 716 (8.36) (3.58) Constant 0.570*** 0.548*** (0.21) education 0.00366 (1.68) net income 0.0269 (-0.93) age -0.0190 (-1.82) (-1.89) married -0.132 -0.138 Model 1 Model 2 Married vs divorced(Power, only male)
Table 4. The regression in the case married versus divorced, including only female respondents. Theta computed by power utility function.
Table 5. The regression in the case married versus divorced, including only female respondents. Theta computed by exponential utility function.
4.3 Married versus widowed
A dummy that equals one when a respondent is married and equals zero when a respondent is widowed is used in this regression. The regression of this dummy and gender on risk attitude gives insignificant coefficients of marriage, while the dummy for gender is significant. When the variables for age, education and net income are included the coefficient of the dummy for
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.006 0.013 Observations 771 771 (7.78) (2.05) Constant 0.420*** 0.269* (-0.73) education -0.0118 (0.81) net income 0.0115 (1.66) age 0.0282 (-2.24) (-1.89) married -0.130* -0.113 Model 1 Model 2 Married vs divorced(Power, only female)
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.006 0.011 Observations 771 771 (12.76) (6.33) Constant 0.0308*** 0.0372*** (0.69) education 0.000496 (-1.09) net income -0.000692 (-1.43) age -0.00109 (2.11) (1.72) married 0.00550* 0.00459 Model 1 Model 2 Married vs divorced(Expo, only female)
marriage even gets more insignificant. Among the subgroups including only women or men there are also found highly insignificant coefficients. However, for all three groups (all individuals, only men and only women) the same direction of marriage on risk aversion is found. The coefficient in all regressions is negative, which indicates that married individuals are more risk averse than widowed individuals. But since the p-values are very large, it is doubtful whether this direction is correct.
4.4 Married versus never married
A dummy that equals one when a respondent is married and equals zero when a respondent is never married is used in this regression. First the relation between marriage and risk attitude compared with individuals who never were married is measured. For all individuals the same problem as in 4.3 occurs; including age, education and net income gives a more insignificant coefficient for civil status. The coefficient for civil status is already not significant when only marriage is regressed on risk attitude. Furthermore, the p-value even increases from 0.153 to 0.667 when the other variables are included.
This increase in p-value does not occur when variables are added in the regression among only men. However, the decrease in p-value is small and they are still far from significant. For women the values are much lower, but still not significant. The lowest p-value found is 0.2. But like the regression among men, there is no increase in p-p-values when more variables are added.
4.5 Divorced versus widowed
A dummy that equals one when a respondent is divorced and equals zero when a respondent is widowed is used in this regression. For all three subgroups there are found insignificant results. Notably, in the case of divorced compared with widowed, I find more significant results in the subgroup containing only males than in the subgroups which contains only female respondents. While in all other civil states comparisons there were obtained more significant results for female than for male, probably caused by the fact that the sample contains more women than men (1180 vs. 1018). The direction of the coefficient is the same for all three subgroups; a positive relation between being divorced, compared with widowed respondents, and someone’s risk attitude is found. However, since the results are insignificant the direction and magnitude of the coefficients is doubtful.
4.6 Divorced versus never married
A dummy that equals one when a respondent is divorced and equals zero when a respondent is never married is created in this regression. In the regression including only civil status and the regression including civil status and gender gives a significant relation between being divorced, compared with being never married, and risk attitude. However, when the other three variables, age, education and net income, are included, the relation between civil status and risk attitude becomes insignificant, while for net income there is a significant relationship found. The p-value of civil status increases from 0.001 to 0.119. The coefficients of civil status and net income have the same direction. The coefficient of civil status falls from 0.150 to 0.093, while the coefficient of net income equals 0.033.
In the subgroup with only females there arises a similar problem. When only civil status is regressed on risk attitude, a significant coefficient of 0.154 is obtained. When age, education and net income are added, the p-value rises from 0.004 to 0.843. In this last regression age is significant at the 5 percent rejection interval, with a positive coefficient of 0.0594.
What stands out is that this drop in significance does not occur among men. When first civil status is regressed on risk attitude, a coefficient of 0.150 is found, significant at the 5 percent rejection interval. After age, education and net income are included the coefficient of civil status increases to 0.171, significant at the 10 percent rejection interval. Although a less significant relation is found compared with the first regression, it is a much smaller drop in significance than among women. Interpreting the coefficient; divorced men are 17.1 percent more risk seeking than men that were never married.
For all three subgroups there is a high correlation (0.60) between age and being divorced, compared to never married individuals. This explains the insignificant results among women and all individuals. However, it is remarkable that in the subgroup containing only men significant results are obtained. This can be caused by the differences in other correlation.
Table 6. The regression in the case divorced versus never married, including all respondents. Theta computed by power utility function.
Table 7. The regression in the case divorced versus never married, including only male respondents. Theta computed by power utility function.
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.012 0.030 0.041 Observations 785 785 785 (14.15) (8.87) (3.09) Constant 0.336*** 0.266*** 0.229** (-1.43) education -0.0225 (2.21) net income 0 .0328* (0.67) age 0.0121 (3.73) (3.18) male 0.153*** 0.132** (3.14) (3.22) (1.56) divorced 0.150** 0.152** 0.0932 Model 1 Model 2 Model 3 Divorced vs never married(Power)
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.012 0.038 Observations 353 353 (11.55) (4.62) Constant 0.420*** 0.512*** (-1.52) education -0.0356 (3.08) net income 0.0653** (-1.65) age -0.0465 (2.03) (1.87) divorced 0.150* 0.171 Model 1 Model 2 Divorced vs never married(Power, only male)
Table 8. The regression in the case divorced versus never married, including only female respondents. Theta computed by power utility function.
Table 9. Correlation table in the case of divorced versus never married, including male and female.
Table 10. Correlation table in the case of divorced versus never married, including only males.
Table 11. Correlation table in the case of divorced versus never married, including only females.
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.015 0.033 0.034 Observations 432 432 432 (8.64) (1.12) (1.19) Constant 0.266*** 0.0803 0.110 (-0.48) education -0.0103 (0.05) net income 0.0 00955 (2.85) (2.53) age 0.0576** 0.0594* (2.52) (0.33) (0.20) divorced 0.154* 0.0251 0.0156 Model 1 Model 2 Model 3 Divorced vs never married(Power, only female)
oplcat -0.0518 -0.0047 0.0081 0.0028 0.1041 0.5076 1.0000 nettocat 0.1899 0.1250 -0.1190 0.1521 0.5003 1.0000 lftdcat 0.5972 0.1265 -0.1137 0.0304 1.0000 gender -0.0133 0.1300 -0.1425 1.0000 Expo -0.0948 -0.9895 1.0000 Power 0.1116 1.0000 div 1.0000 div Power Expo gender lftdcat nettocat oplcat
oplcat 0.0588 0.0065 -0.0087 . 0.1409 0.4901 1.0000 nettocat 0.2920 0.1452 -0.1396 . 0.5523 1.0000 lftdcat 0.5986 0.0585 -0.0410 . 1.0000 gender . . . . Expo -0.0826 -0.9881 1.0000 Power 0.1075 1.0000 div 1.0000 div Power Expo gender lftdcat nettocat oplcat
oplcat -0.1428 -0.0156 0.0241 . 0.0736 0.5385 1.0000 nettocat 0.1030 0.0685 -0.0581 . 0.4570 1.0000 lftdcat 0.5973 0.1805 -0.1713 . 1.0000 gender . . . . Expo -0.1110 -0.9906 1.0000 Power 0.1204 1.0000 div 1.0000 div Power Expo gender lftdcat nettocat oplcat
4.7 Widowed versus never married
A dummy that equals one when a respondent is widowed and equals zero when a respondent is never married is used to use in this regression. An equally problem arises as in 4.5, the p-values drop when more variables are added. This occurs especially for women, while for men there seems to be no effect, of adding more variables, on both the coefficient as on the p-value. As presented in table 12 there is a high correlation (0.69) between being widowed, compared to being never married, and age. This can cause the drop in significance as in the first regression there is an indirect effect of age on risk attitude via civil status.
Table 12. Correlation table in the case of widowed versus never married, including male and female.
4.8 Logit regression
From the previous subsections it is obvious that divorced are the most risk-seeking subgroup. Since this subgroup gives the most significant results a logit regression is performed to investigate what increases the probability to be divorced. In this regression, see table 13, risk attitude, gender, age, net income and education are used as explanatory variables. Table 13 contains odds ratio’s. A strong relationship between risk attitude and being divorced is found, with a p-value of 0.008. An increase of 1 unit in theta increases the possibility of being divorced with 37.2 percent. Next to this result, age and net income also have a positive relationship with the possibility of being divorced. It is found that women are more likely to be divorced than man. This logit regression indicates a positive relationship between risk attitude and being divorced, compared to the other civil statuses. This is in line with the predictions from the descriptive statistics.
oplcat -0.2372 -0.0498 0.0439 0.0132 -0.0257 0.4582 1.0000 nettocat 0.2196 0.0543 -0.0574 0.0726 0.5223 1.0000 lftdcat 0.6910 0.0852 -0.0746 -0.0639 1.0000 gender -0.1267 0.1272 -0.1404 1.0000 Expo -0.0275 -0.9892 1.0000 Power 0.0376 1.0000 wid 1.0000 wid Power Expo gender lftdcat nettocat oplcat
Table 13. Logit regression, in the case of divorced versus individuals that are not married, including all respondents. Civil status is the dependent variable. Theta computed by power utility function.
4.9 Discussion
The regressions that are discussed in the subsections 4.2 to 4.8 confirm the descriptive statistics mentioned in subsection 4.1. In three subsamples that contained divorced
individuals, divorced individuals seem to be the most risk-seeking subgroup. However, the question is whether the status of being divorced influences someone’s risk attitude. Another possibility is that divorced individuals are more risk seeking due to other characteristics, and that this is the reason why they are divorced.
When looking at the logit regression, as discussed in subsection 4.7, a positive relationship is found between risk attitude and the probability to be divorced. It makes sense that more risk-seeking individuals are more likely to be divorced as the decision to get divorced may be viewed as difficult and involving high risks. As aconsequence, more risk-seeking individuals make the choice to get divorced while more risk-averse individuals stay married, assuming that both individuals are equally (un)happy about their marriage. Hence, it is clear that risk attitude plays a role in the civil statuses of individuals.
It seems to be that men’s risk attitude reacts stronger to being divorced than women’s risk attitude. This result indicates an interaction between gender and civil status. This result is in line with the results found by Yao and Hanna (2005) and Sunden and Surette (1998), which says that there is an interaction between gender and marriage. It can be the case that
* p<0.05, ** p<0.01, *** p<0.001
Exponentiated coefficients; z statistics in parentheses
R-squared Observations 2198 2198 2198 (-0.30) education 0.983 (2.06) net income 1.100* (4.59) age 1.293*** (-1.09) (-2.24) male 0.846 0.681* (2.91) (3.03) (2.65) Power 1.411** 1.435** 1.372** div Model 1 Model 2 Model 3 Logit Divorced vs rest(Power )
men react different to being divorced. Taking more risk may for example be a way of dealing with the process of separation.
As presented in the tables, the models used in this section have a very low R-squared. Including more variables in the model drives up the R-squared. This may be a solution to increase the credibility of the coefficients estimated by the model.
In the subsections 4.5 and 4.6 remarkable results are found. Both widowed and divorced individuals are compared to individuals who were never married. When age is included as explanatory variable the coefficients of civil status become insignificant. This may not be very surprising as there is a high correlation between on the one hand being divorced or widowed and on the other hand the age of the individual. Regarding widowed individuals, as the probability of dying increases with age, the probability of becoming widowed increases too. Since divorced individuals were married before they will be on average older than never married individuals, as never married individuals will be mostly young respondents. This causes that the coefficient of civil status in the first regression contains an indirect effect of age on risk attitude.
The model used in this research faces some lack of internal and external validity. They will be discussed and some possible solutions are mentioned.
The research shows quite some insignificant results. This is especially true for the cases where widowed are compared with other civil statuses. One reason may be that there are to few individuals widowed, as there are only 122 widows or widowers. A solution for this lack of internal validity may be to take a larger sample of the population. Hence, the sample will contain more widowed individuals.
Another possible cause of the insignificant results found for widowed individuals is a high correlation between age and being widowed, as the probability to die increases with age. This can cause that the effect of being widowed is incorporated as an indirect effect in the coefficient of age. A solution may be to look at subgroups of age, as the difference in age among the respondents decreases.
Another lack of internal validity that is faced in these regressions occurs at the measurement of risk attitude. The amount of safe choices is used to compute a measure of risk aversion. Since there are only five trials, people have either none, one, two, three, four or five safe choices. This causes that there is little variation in risk attitude. The little variation in risk attitude causes that less significant estimation of coefficients are found. A possible
solution is to increase the amount of trials to obtain more variation in risk attitude among the respondents. Another solution is to use a different way to measure someone’s risk attitude.
The third lack of internal validity in this model is the possible reverse causality. It is hard to say whether civil status influences risk attitude or that it is the other way around. A solution may be to use instruments to measure the influence of civil status on risk attitude. The final lack of internal validity is the question whether civil status influences risk attitude or that there is an indirect effect of people characteristics via civil status. It may be that someone’s characteristic determines whether someone gets married or divorced and also determines this individual’s risk attitude. To increase the internal validity more explanatory variables could be added since the indirect effect of these explanatory variables will
disappear.
Finally, there is also a possible lack of external validity. As the data only contains Dutch people, the results may not be useful for other countries. Perhaps Dutch individuals have another average risk profile and therefore are not comparable with other populations. Using data that contains people from all over the world might increase the external validity.
Regarding to the validity of the model it may be very helpful to use instruments for civil status and risk attitude. When exogenous instruments are used, reverse causality will evade. Furthermore, there may be no correlation between this instrument and age or other variables. So using exogenous instruments probably increases the credibility of the results.
It is evident that there are some shortcomingsin this model. Hopefully, subsequent research finds solutions for this lack of internal and external validities. Maybe this research points subsequent research in the right direction.
5 Conclusion
Financial institutions want to offer financial products which customers are willing to purchase. An insight in the relation between civil status and risk attitude can help these institutions to determine which products to offer. For individuals, (their) risk attitude plays a role in reaching their financial goals. An insight in their risk attitude can be of importance to determine what to invest in,
The main question in this research is what the influence of civil status is on risk attitude. To answer this question I have performed multiple regressions. A lot of this regression gave insignificant results. However, I did find significant results for some cases that were evaluated. As a result it can be concluded that there are differences in risk attitude between several civil statuses.
At first, the plan was to compare married individuals to widowed, as being widowed is a more random event than being divorced. Since being divorced gave the most significant results, while being widowed did not, this group was mostly discussed and used in the logit regression. Several reasons why being widowed did not gave significant results were discussed in the previous section. The few widowed individuals may be a possible explanation, or the high correlation with age, which causes an indirect effect. Possible solutions for this problem are to increase the sample or to look only at subsamples of age.
Although there were no significant results found for widowed individuals, they were found for divorced individuals. Divorced individuals are the most risk-seeking group.
Stronger results are found for men than for women. This indicates an interaction between age and civil status. Next to the interaction between age and civil status there is also found a high correlation between on the one hand age and on the other hand being divorced or widowed. This correlation caused that the coefficients of civil status became less significant when age was included as explanatory variable. The logit regression obtained a positive relationship between risk attitude and being divorced, meaning that more risk-seeking individuals are more likely to be divorced.
The interaction between gender and civil status supports previous research done by Sunden and Surette (1998) and Yoa and Hanna (2005). However, where previous research (Hinz et al. (1997), Hartog et al. (2002), Sunden and Surette (1998) and Yao and Hanna (2005)) finds a significant relationship between marriage and risk attitude, in this research there was find no significant relationship between marriage and risk attitude.
The computations that were performed to obtain measurements of risk attitude did take quite some time. Individuals that switch more than once between safe choices or the lottery had to be removed from the sample. Since these individuals have two switching points they do not have a unique theta. At first I did not notice that these individuals had to be removed, which caused a lot of insignificant results. Although the problem of multiple
switching points was eventually fixed, there are still quite a lot of insignificant results. This is partly due to the measure of risk attitude that was used, as there is to little variation in the possible thetas.
The regressions that were performed in this research had a low R-squared. This could be due to the low number of explanatory variables. A model including more explanatory variables would probably give a higher R-squared.During this thesis some lack of internal and external validities were faced. Regarding the lack of internal validities it is often hard to say whether risk attitude influences civil status or the other way around. Possible solution is to use instruments. This may be a solution for researchers that are investigating the same topic.
References
Gutter, M., Fox, J. & Montalto, P. (1999). Racial differences in investor decision making. Financial Services Review, 8(3), 149-162.
Halek, M. & Eisenhauer, J. (2001). Demography of risk aversion. The Journal of Risk and Insurance, 68(1), 1-24.
Hartog, J., Ferrer-i-Carbonell, A. & Jonker, N. (2002). Linking measured risk aversion to individual characteristics. Kyklos, 55(1), 3-26.
Hinz, R., McCarthy D. & Turner, J (1997) Are women conservative investors? Gender differences in participant-directed pension investments. M.S. Gordon, O.S. Mitchell, M.M. Twinney (Eds.), Positioning pensions for the twenty-first century, University of Pennsylvania Press, Philadelphia (1997), 91–103.
Noussair, C. N., Trautmann, S. T., Kuilen, G. v., & Vellekoop, N. (2013). Risk aversion and religion. Journal of Risk and Uncertainty, 47(2) , 165-183.
Schooley, D. & Worden, D. (1996). Risk aversion measures: comparing attitudes and asset allocation. Financial Services Review, 5(2), 87-99.
Sunden, A. E. & Surette, B. J. (1998). Gender differences in the allocation of assets in retirement savings plans. The American Economic Review, 88(2), 207-211.
Wakker, P. (2008). Prospect theory for risk and ambiguity. Cambridge, UK: Cambridge University Press.
Yao, R. & Hanna, S. (2005). The effect of gender and marital status on financial risk tolerance. Journal of personal finance, 4(1), 66-85.
Appendix I
Table 1. The regression in the case married versus the rest, including all respondents. Theta computed by power utility function.
Appendix II
Table 2. The regression in the case married versus the rest, including all respondents. Theta computed by exponential utility function.
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.000 0.016 0.022 Observations 2198 2198 2198 (19.27) (14.17) (4.51) Constant 0.376*** 0.314*** 0.236*** (-1.17) education -0.0109 (1.95) net income 0 .0159 (2.10) age 0.0187* (5.87) (4.37) male 0.147*** 0.121*** (-0.56) (-0.91) (-1.74) married -0.0143 -0.0231 -0.0477 Model 1 Model 2 Model 3 Married vs rest(Power) * p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.000 0.017 0.023 Observations 2198 2198 2198 (36.66) (35.37) (16.18) Constant 0.0321*** 0.0350*** 0.0379*** (1.15) education 0.000482 (-2.28) net income -0.0 00834* (-1.49) age -0.000597 (-6.20) (-4.54) male -0.00693*** -0.00561*** (0.74) (1.12) (1.70) married 0.000846 0.00126 0.00209 Model 1 Model 2 Model 3 Married vs rest(Expo)
Appendix III
Table 3. The regression in the case married versus divorced, including all respondents. Theta computed by exponential utility function.
Appendix IV
Table 4. The regression in the case married versus widowed, including all respondents. Theta computed by power utility function.
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.004 0.020 0.023 Observations 1487 1487 1487 (14.52) (15.45) (7.87) Constant 0.0279*** 0.0308*** 0.0337*** (0.38) education 0.000200 (-2.08) net income -0.0 00979* (-0.33) age -0.000190 (-4.96) (-2.69) male -0.00682*** -0.00461** (2.47) (2.65) (2.44) married 0.00507* 0.00541** 0.00501* Model 1 Model 2 Model 3 Married vs divorced(Expo) * p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.000 0.015 0.016 Observations 1412 1412 1412 (7.23) (6.41) (2.65) Constant 0.392*** 0.350*** 0.296** (-0.69) education -0.00817 (0.69) net income 0. 00719 (0.59) age 0.00789 (4.66) (3.37) male 0.147*** 0.131*** (-0.53) (-1.05) (-0.59) married -0.0298 -0.0595 -0.0357 Model 1 Model 2 Model 3 Married vs widowed(Power)
Appendix V
Table 5. The regression in the case married versus widowed, including all respondents. Theta computed by exponential utility function.
Appendix VI
Table 6. The regression in the case married versus never married, including all respondents. Theta computed by power utility function.
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.000 0.016 0.017 Observations 1412 1412 1412 (13.11) (13.82) (7.14) Constant 0.0317*** 0.0337*** 0.0356*** (0.58) education 0.000309 (-1.17) net income -0.0 00549 (-0.24) age -0.000141 (-4.80) (-3.20) male -0.00675*** -0.00557** (0.47) (1.01) (0.58) married 0.00119 0.00255 0.00155 Model 1 Model 2 Model 3 Married vs widowed(Expo) * p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.000 0.017 0.020 Observations 1884 1884 1884 (14.21) (10.18) (4.14) Constant 0.336*** 0.268*** 0.223*** (-0.64) education -0.00636 (1.61) net income 0 .0138 (1.23) age 0.0131 (5.71) (4.16) male 0.150*** 0.123*** (0.92) (0.74) (-0.42) married 0.0262 0.0210 -0.0149 Model 1 Model 2 Model 3 Married vs never married(Power)
Appendix VII
Table 7. The regression in the case divorced versus widowed, including all respondents. Theta computed by power utility function.
Appendix VIII
Table 8. The regression in the case divorced versus widowed, including all respondents. Theta computed by exponential utility function.
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.005 0.016 0.024 Observations 313 313 313 (6.55) (5.53) (0.96) Constant 0.392*** 0.351*** 0.282 (-1.52) education -0.0429 (0.91) net income 0 .0246 (0.33) age 0.0131 (1.87) (1.70) male 0.143 0.139 (1.24) (0.94) (1.42) divorced 0.0941 0.0722 0.130 Model 1 Model 2 Model 3 Divorced vs widowed(Power) * p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.004 0.016 0.026 Observations 313 313 313 (12.19) (12.13) (2.82) Constant 0.0317*** 0.0335*** 0.0360** (1.66) education 0.00204 (-1.13) net income -0. 00134 (-0.24) age -0.000415 (-1.87) (-1.64) male -0.00623 -0.00586 (-1.17) (-0.88) (-1.39) divorced -0.00389 -0.00294 -0.00554 Model 1 Model 2 Model 3 Divorced vs widowed(Expo)
Appendix IX
Table 9. The regression in the case widowed versus never married, including all respondents. Theta computed by power utility function.
Appendix X
Table 10. The regression in the case widowed versus never married, including all respondents. Theta computed by exponential utility function.
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.001 0.019 0.029 Observations 710 710 710 (14.69) (9.02) (3.12) Constant 0.336*** 0.268*** 0.236** (-1.64) education -0.0265 (0.67) net income 0. 00995 (1.68) age 0.0318 (3.57) (3.35) male 0.151*** 0.142*** (1.00) (1.46) (-0.78) widowed 0.0560 0.0813 -0.0633 Model 1 Model 2 Model 3 Widowed vs never married(Power)
* p<0.05, ** p<0.01, *** p<0.001 t statistics in parentheses R-squared 0.001 0.022 0.031 Observations 710 710 710 (32.19) (27.36) (11.07) Constant 0.0336*** 0.0370*** 0.0382*** (1.62) education 0.00119 (-0.85) net income -0.0 00574 (-1.45) age -0.00125 (-3.90) (-3.66) male -0.00750*** -0.00709*** (-0.73) (-1.23) (0.80) widowed -0.00186 -0.00312 0.00294 Model 1 Model 2 Model 3 Widowed vs never married(Expo)
Appendix XI
Table 11. The logit regression in the case divorced versus rest, including all respondents. Theta computed by exponential utility function. Table contains odds ratio’s.
Appendix XII
Table 12. The logit regression in the case divorced versus rest, including only males. Theta computed by power utility function. Table contains odds ratio’s.
* p<0.05, ** p<0.01, *** p<0.001
Exponentiated coefficients; z statistics in parentheses
R-squared Observations 2198 2198 2198 (-0.31) education 0.983 (2.04) net income 1.099* (4.63) age 1.295*** (-1.08) (-2.22) male 0.848 0.683* (-2.61) (-2.73) (-2.40) Expo 0.000863** 0.000579** 0.00142* div Model 1 Model 2 Model 3 Logit Divorced vs rest(Expo)
* p<0.05, ** p<0.01, *** p<0.001
Exponentiated coefficients; z statistics in parentheses R-squared Observations 1018 1018 (0.72) education 1.061 (-0.40) net income 0.971 (3.22) age 1.318** (1.94) (1.97) Power 1.402 1.406* div Model 1 Model 2 Logit Divorced vs rest(Power, only male )
Appendix XIII
Table 13. The logit regression in the case divorced versus rest, including only males. Theta computed by exponential utility function. Table contains odds ratio’s.
Appendix XIV
Table 14. The logit regression in the case divorced versus rest, including only females. Theta computed by power utility function. Table contains odds ratio’s.
* p<0.05, ** p<0.01, *** p<0.001
Exponentiated coefficients; z statistics in parentheses R-squared Observations 1018 1018 (0.71) education 1.060 (-0.41) net income 0.970 (3.25) age 1.321** (-1.68) (-1.76) Expo 0.00122 0.000882 div Model 1 Model 2 Logit Divorced vs rest(Expo, only male )
* p<0.05, ** p<0.01, *** p<0.001
Exponentiated coefficients; z statistics in parentheses R-squared Observations 1180 1180 (-1.01) education 0.923 (3.15) net income 1.217** (3.34) age 1.284*** (2.34) (1.81) Power 1.464* 1.350 div Model 1 Model 2 Logit Divorced vs rest(Power, only female )
Appendix XV
Table 15. The logit regression in the case divorced versus rest, including only females. Theta computed by exponential utility function. Table contains odds ratio’s.
* p<0.05, ** p<0.01, *** p<0.001
Exponentiated coefficients; z statistics in parentheses R-squared Observations 1180 1180 (-1.03) education 0.923 (3.14) net income 1.215** (3.37) age 1.286*** (-2.18) (-1.66) Expo 0.000302* 0.00187 div Model 1 Model 2 Logit Divorced vs rest(Expo, only female )
