Keywords

12.1 Introduction

Japan is not only known for its high life-expectancy and its increasingly aging society (Coulmas 2008), but also for its relatively traditional, rigid social structures with predetermined life courses and career paths (Sugimoto 2010). These features suggest that gender and age differences in subjective well-being are more evident in Japan than in other societies. The World Value Survey data show for example that Japan (20.8 %) ranks together with Bangladesh (16.5 %), Iran (22.7 %), the Philippines (15.5 %), Saudi Arabia (8.6 %) and Morocco (7.9 %) among the lowest countries regarding “norms on gender inequality” (see Tesch-Römer et al. 2007). The numbers in parenthesis show the percentage of women disagreeing with the statement that “When jobs are scarce, men should have more right to a job than women”. Further, Veenhoven (2008) develops a happiness indicator called Happy Life Years (HLY) which is the product of a country’s average life expectancy (L) multiplied by the happiness index taken from the World Database of Happiness, which is transformed into a scale ranging from “0” to “1” (H). HLY = L * H. Regarding Japan he observes that “[t]he rank-order of HLY is quite similar to the rank order in average happiness, the rank-order correlation being + 0.94. The correlation is not perfect however, in some countries people live long, but are not too happy (e.g. Japan)”, Veenhoven (2008).

The Japanese government has recently jumped onto the bandwagon of policy makers considering alternative indicators to Gross National Product (GNP) , such as Gross National Happiness (GNH) and set up a Commission on Measuring Well-Being in late 2010.Footnote 1 Earlier in the same year, the Cabinet Office has set the focus of the annual National Survey on Lifestyle Preferences (NSLP) on happiness and its impact factors.

Despite the government’s efforts to promote alternative well-being measures, the Japan-related literature in happiness economics, however, shows a relative scarcity of studies compared to other countries. International studies of macro-level happiness usually mention Japan in the context of the Easterlin Paradox (Easterlin 1995; Hagerty and Veenhoven 2003; Easterlin 2005; Veenhoven and Hagerty 2006; Stevenson and Wolfers 2008; Suzuki 2009). Existing studies on Japan at the micro-level, however, have a number of drawbacks. They are either based on limited samples such as students (Tafarodi et al. 2012a, b) or they are focused on narrow topics like social trust (Tokuda et al. 2010; Tokuda and Inoguchi 2008) or income inequality (Oshio and Kobayashi 2010, 2011; Oshio et al. 2011). One of the few broad-scale studies on Japan is Inoguchi and Fuji (2009). The study uses data from the AsiaBarometer. However, here too, the sample size is relatively small (n = 1,003) while the data from 2006 is already slightly dated. Moreover, the study is rather simple, since many controls are not included in the analysis.

Given this state of the field, we have conducted an econometric analysis of the Japanese National Survey on Lifestyle Preferences (NSLP) of the years 2010 to 2012. The purpose of our paper is twofold. First, based on a very recent and comprehensive dataset we add to the discussion of gender and age effects on happiness from a Japanese perspective. Second, since a subset of the data includes two different measures of well-being, we further examine differences in perceived life satisfaction and happiness from a gender and age perspective and by doing so add another dimension to the discussion.

12.2 Literature

While a number of international studies in the field of happiness research have shown remarkably similar results regarding the effects of income , unemployment and marriage (Blanchflower and Oswald 2011), the effects of basic control variables, such as age and gender, are still discussed controversially. It is an established finding in the international literature that women are happier than men (e.g. Blanchflower and Oswald 2004; Frey and Stutzer 2002; Praag and Ferrer-i-Carbonell 2008, only Hayo and Seifert 2003 report different findings for Eastern Europe). However, the size of the gender effect is in most studies either small or negligible.Footnote 2 Regarding the age affects on well-being the international literature finds a U-shaped pattern (Frey and Stutzer 2002; Oswald 1997; Blanchflower and Oswald 2008). Easterlin (2006) however points out that, looking at the raw data, without keeping all other variables constant, several psychological studies report an inverted U-shape (see also Myers 2000; Argyle 1999; Diener et al. 1999). Related to this observation, Blanchflower and Oswald (2008) admit that, when looking at the raw data the U-shape does not hold true for all countries, but for almost all countries it can be observed when controlling for other variables in a multiple regression. Using fixed-effects estimations on large panel datasets from Germany, the UK and Australia , Frijters and Beatton (2012) find almost no change in subjective well-being between the age of 20 and 50. Using the same fixed-effects estimations on the same UK panel data Clark (2007) as well as Baird et al. (2010) find a U-shaped age effect between the age of 20 and 50, indicating that even the analysis of panel data yields no consistent results.

While the international literature finds only moderate gender differences in happiness, most of the Japan-related studies report rather big gaps in happiness between men and women (Sano and Ohtake 2007; Kusago 2008; Urakawa and Matsuura 2007a; Tsuji 2011; Ohtake 2012). Exceptions are Yamane et al. (2008) and Oshio et al. (2011). The former find only a small coefficient of the gender variable, while the coefficient in the latter study does not turn out to be statistically significant. Tiefenbach and Kohlbacher (2013), however, find a coefficient of 0.45 (on a scale from 0 to 10) that lends further evidence to both the universal finding that women are happier than men and to the country specific result that the gender gap in happiness is rather big in Japan compared to other countries.

Regarding the age effects on well-being the findings in the Japan-related literature are not consistent. While the U-shaped pattern is reported by numerous studies (Oshio and Kobayashi 2011; Ohtake 2012; Kusago 2007; Tsuji 2011), others show different effects such as an inversed U-shape (Tsutsui et al. 2010), a downward sloping pattern (Yamane et al. 2008), or no significant correlation at all (e.g. Inoguchi and Fuji 2009; Sano and Ohtake 2007). Finally, Tiefenbach and Kohlbacher (2013) find in their study on Japan a U-shaped trend in both the raw data and the multivariate happiness estimation. A very unique result of their study is that the age effects can only be found among men, whereas the age groups dummies have almost no statistically significant coefficients for women.

Regarding the concepts of happiness and life satisfaction, two positions can be found in the literature. While one side argues that not only the label but also the measures of “happiness” and “life satisfaction” can be used interchangeably (Veenhoven 1991, 2008; Frey 2008; Frey and Stutzer 2002), other authors, argue that there are significant differences between those concepts, not only theoretically (Lane 2000), but also when analyzing empirical data (Gundelach 2004), especially when analyzing sub-groups like Russian students (Balatsky and Diener 1993) or older women with heart diseases (Friedman 1993).

12.3 Data and Analytical Strategy

The present study uses regression analysis to analyze data from the National Survey on Lifestyle Preferences (kokumin seikatsu senkôdo chôsa) of the years 2010 to 2012 in Japan .Footnote 3 This survey has been introduced in 1972 and is commissioned on an annual basis since 1984 by the Japanese cabinet office. Since 2010 the focus has been placed on individual happiness and its determinants.Footnote 4 The population of the survey includes men and women in Japan between 15–80 years of age and the sample is generated via a two-stage randomized stratified procedure and includes 4,000, 5,000 and 4,000 persons respectively for the years 2010, 2011, and 2012. Due to the relatively high response rates, there are 9,280 completed questionnaires available for analysis. Unfortunately, the NSLP is not a panel survey, and therefore only allows for a pooled cross-sectional analysis. Although pooling the three datasets from 2010 to 2012 yields a large sample size, it comes with the downside of limiting the number of control variables. This is due to the fact that the questionnaire items and their coding have been subsequently changed over the years, leaving only a limited number of variables which can be consistently included in a cross-sectional estimation.

Against this backdrop, our analytical strategy is as follows. In a first step we run a regression analysis using the pooled data of the years 2010–2012. Since pooling the datasets results in a substantial loss of a number of control variables, we then estimate regression models only using the 2010 data. This allows us to include more control variables (albeit with a smaller sample size (n = 2,802)). In each step we run three regression models: one using the whole sample followed by two separate models for men and women. The reason for focusing particularly on the 2010 dataset lies in the fact that it includes not only a happiness measure, but also a question regarding the general life satisfaction of the respondents. This allows us to control for differences in perceived happiness and life satisfaction levels stratified by gender. As analytical method we use ordinary least squares (OLS) as well as ordered probit regressions, but since the results of the latter are consistent with our findings only the OLS results are reported (see also Tiefenbach and Kohlbacher 2013).

12.4 Pooled Cross-Sectional Analysis of the Years 2010–2012

In the following sections we first describe the variables of interest used in the regression models for the pooled dataset of the years 2010–2012. We then look at the raw data regarding the average happiness level stratified by gender and age. Finally we introduce our regression models and discuss their results.

12.4.1 Variables of Interest

Our dependent variable is the current happiness level of the respondent (HAP). The corresponding survey item reads: “How happy are you currently?”Footnote 5. Answer options range from 0 to 10 on an 11 point scale. On the right hand side of the equation we are mainly interested in the gender and age group variables. But in order to control for distortions we further introduce a number of control variables that are common in happiness estimations. Apart from basic socio-demographic variables (income, income squared, age groups, gender) we control for family relations (cohabitation with spouse, number of children, children under 6 years dummy), employment relations (student, housewife, without work) as well as standard control variables including survey year and prefecture dummies. Unfortunately, the items asking the age and number of children, as well as the items asking about the employment relations of the respondents as well as the class of their yearly household income, were changed over the years, which is the reason why we can only use a rather limited number of standard controls.Footnote 6 We also include an array of items regarding volunteer activities as these have gained in importance in the aftermath of the 3/11 disaster in 2011 in Japan (Avenell 2012), a category which has, fortunately, not been subject to change over the years in the survey instrument. The items ask the respondents, (i) how many times a month they a engage in volunteering activities, (ii) whether they are the beneficiary of voluntary services and (iii) whether they make donations.

12.4.2 A Look at the Raw Data: Age and Gender

The Fig. 12.1a–c show the average happiness over the age groups from 15–19 years to 75–80 years for the whole sample (a) as well as stratified by gender (b = women, c = men). Apart from the average happiness of each age group the graphs include the total average happiness as well as a bipolynomial trend line. The following observations can be made. With an average happiness level of 6.21 men are about 0.24 points less happy than the overall average (6.45) as well as 0.46 points less happy than women (6.67).Footnote 7 The raw data for all three graphs is W-shaped with peaks at the lower and higher age groups as well as in the years between 30 and 34. Looking at the bipolynomial trend line reveals that while the trend for men is U-shaped, the trend for women shows a downward slope. Taken together, the trend for the total sample shows only the left side of a U-shape flattening out around the age of 55 and rising slightly again in the age from 75 to 80.

Fig. 12.1
figure 1

a–c Average happiness stratified by gender and age. (Pooled raw data from the National Survey of Lifestyle Preferences of the years 2010–2012)

Full size image

12.4.3 Analysis and Discussion

Table 12.1 shows the regression results for the model specified above. As mentioned, we ran the regression not only for the whole sample (model 1), but also separately for women (2) and men (3). The variances explained as represented by the adjusted R-squared values are 0.127 (1), 0.106 (2), and 0.135 (3), respectively.

Table 12.1 OLS regression results using the NSLP 2010–2012 happiness data
Full size table

Looking at the estimation results, women are about 0.41 points happier than men (on a scale from 0 to 10), even when controlling for an array of other variables. As for age, the regression results basically confirm the descriptive findings presented above. For the whole sample, all age groups—except for the group 50–59 years—are significantly happier than the reference group (45–49 years). The male sample shows a similar picture, but the situation is less clearly pronounced for the female sample (since here in addition to the group of 50–59 year olds, the 35–39 as well as the 60–64 year age group are also not significantly happier than the reference group). Figure 12.2a–c show the marginal age effects visually, highlighting the classic U-shaped curve. It is interesting to note, that the U-shape of the marginal age effects is much more pronounced in the regression analysis compared to the bipolynomial trend depicted in the raw data. This is especially the case for the female subsample which shows a declining trend line in the raw data, but a rising marginal age effect in the older age groups in the regression analysis.

Fig. 12.2
figure 2

a–c Marginal age effects based on a pooled OLS regression with data from the National Survey on Lifestyle Preferences of the years 2010–2012. (The average happiness of the 45–49 year olds is used as reference group. Bars in light colors indicate marginal effects that do not meet the p < 0.05 condition)

Full size image

As for income effects , we find that income positively contributes to the happiness level in all three models, with the coefficient being slightly higher for men than for women. The income squared term is also significant, with a negative coefficient. This shows again that happiness increases with income, however at diminishing returns.

The question of whether the respondent lives together with his/her spouse indicates that those co-habiting are significantly happier than those who do not. This effect is much stronger for men than for women.Footnote 8

As for the effects of children, we find that the number of children contributes positively to the happiness level, but the statistical significance is only given for the whole sample and disappears when separating the sample into men and women. Exploring this issue further, we find however, that those with children under 6 years of age are significantly happier than those without children or with children over the age of 6 years.

Next, we examine working conditions. Being a student does not make a significant impact, but being without work has a negative impact on happiness. However, this finding is statistically significant only at the 10 % level (whole sample, men) or not significant at all (women). This fits with the finding that women are significantly happier if they are housewives. This finding is not statistically significant for (house)men, but this may be partly due to the small sample size (N = 85).

Volunteering and donating are two activities that contribute significantly to the happiness of our respondents. It is interesting to note that while doing volunteering (for others) makes a significant impact, receiving volunteering services (from others) does not.

Finally, we have included two dummy variables about the time of the survey to control for the potential impact of the March 2011 triple disaster in Japan . The findings show that respondents were significantly unhappier in 2012 (i.e. after the disaster) than they were in 2010 and 2011 (i.e. before the disaster).Footnote 9 However, there is no significant difference between the individual happiness levels of 2010 and 2011, lending further support to the assumption that 3/11 had a significant (negative) impact.

12.5 The 2010 Data: Life Satisfaction and Happiness in a Gender Perspective

In the following section we refine our analysis by focusing on the 2010 data which allows us to add more controls to the regression models. This section is structured as follows. We first describe what kind of variables we added in the new model compared to the regression of the pooled dataset. Then we will have a look at the raw data and see how happiness and life satisfaction differ between men and women as well as how they develop with increasing age. Finally, we present and discuss the results of the refined regression models.

12.5.1 Variables of Interest

Since the 2010 data include a life satisfaction question, we ran two series of regressions: one using the happiness variable described under 3.1 as dependent variable and one series using the life satisfaction variable instead. Both series of regression models contain one model for the whole sample and two models splitting the sample up between men and women. Unlike the happiness question, the life satisfaction question is placed in the middle of the survey and reads “How satisfied are you regarding the following items” and here the first out of a list of 61 items is “life as a whole” (seikatsu zenpan).Footnote 10 The answer options range from “satisfied” to “dissatisfied” on a five-point scale. To compare the happiness level (HAP) with the level of life satisfaction (LS) we had to bring both measures onto the same scale.Footnote 11

On the right hand side of the equation the following variables remain unchanged compared to the models of the pooled dataset: gender, co-habitation, student, housewife, without work, as well as the set of variables regarding volunteering and the prefecture controls. The income variables are slightly changed. We now include five classes of annual household income and five classes of total household savings. Footnote 12 For the sake of parsimony we did not use age groups but rather the age and age squared variable to shorten the list of variables. We further use four variables regarding the number of children in different age groups.Footnote 13 We also extend the employment relations variables with the following categories: normal employee, managerial position, member of board of directors of a private company, employee of a non-private organization (including board members), irregular employee (including agency work, temporary jobs and part time work), civil servant and entrepreneur. Further, we add a dummy variable taking “one” in case that the respondent or one of his family members is currently unemployed. Finally, the year dummies are dropped and control variables regarding different city sizes are added.

12.5.2 A Look at the Raw Data: Happiness and Life Satisfaction Stratified by Gender

The graphs below show the average happiness (Fig. 12.3a–c) and the average life satisfaction (Fig. 12.3d–f) over the age groups from 15–19 years to 75–80 years for the total sample (Fig. 12.3a, d) as well as for women (Fig. 12.3b, e) and men (Fig. 12.3c, f) separately. Apart from the average happiness or life satisfaction of each age group, the graphs include the total average as well as a bipolynomial trend line. The following observations can be made. Men are not only less happy (6.24) but also less satisfied with their lives (6.09) than women (6.69 and 6.35).Footnote 14 Looking at the levels of life satisfaction and happiness across different age groups among men reveals a stark contrast. Although both graphs (Fig. 12.3c, f) can be described as W-shaped, the middle peak in the happiness data covers a wide range from 30 to 59 years and doesn’t differ very much from the peaks at the youngest and oldest age groups. The life satisfaction levels, however, show extreme peaks at the upper and lower end of the age groups compared to the much smaller peak from 40 to 59 years. Given those differences, the bipolynomial trend line for men describes a clear U-shape for life satisfaction, while it shows a somewhat downward sloping trend for happiness.

Fig. 12.3
figure 3

a–f Happiness and life satisfaction stratified by gender and age. (Raw data from the National Survey of Lifestyle Preferences of the year 2010)

Full size image

In contrast to this, the happiness and life satisfaction levels for women are rather similar (Fig. 12.3b, e). They start with a peak at the youngest age group followed by a substantial drop in the age group from 20 to 24 years. The levels then rise until they reach their peak between 30–34 years and then slope downwards. Two significant differences between the two groups are that (i) the peak of the youngest age group is higher in the life satisfaction than in the happiness data and further that (ii) the life satisfaction levels start to rise again after the age of 60 years, while the happiness levels show no such trend. According to the difference in the two data for women the trend in the happiness data describes a downward sloping curve, while the trend in life satisfaction describes a skewed U-shape.

When comparing the differences between happiness and life satisfaction in the total sample, the most prominent feature is the downward-sloping trend in happiness compared to the U-shaped trend seen in the life satisfaction graph. This indicates that (a) the drop in well-being in middle age as well as (b) the increase in well-being in later life is more visible in life satisfaction than in happiness. Since women show only marginal differences between happiness and life-satisfaction the different shapes of the graphs of the total sample are significantly influenced by the male subsample.

12.5.3 Analysis and Discussion

Table 12.2 shows the regression results for the model specified above. As mentioned, we ran the regression separately for happiness as the dependent variable (models 1–3) and for life satisfaction (models 4–6) as well as the whole sample (model 1; model 4), and separately for women (2; 5) and men (3; 6). The variances explained as represented by the adjusted R-squared values range from 0.116 to 0.203. Below we point out the most significant differences between happiness and life satisfaction as well as between the subsamples of men and women.

Table 12.2 OLS regression results using the NSLP 2010 happiness and life satisfaction data
Full size table

Looking at the estimation results, women are about 0.496 points happier and 0.355 points more satisfied than men. As for age, the regression results show a negative age effect and a significant age-squared term indicating U-shaped marginal age effects . As for income , we basically find the same positive effect for all models. The effects of savings are only visible in the lowest and highest savings groups. It is interesting to note that the effects of no savings are statistically stronger on life satisfaction than on happiness, and here the effects are further significantly stronger for women than for men.Footnote 15 The highest savings group, however, shows opposite gender effects: here the coefficient on the life satisfaction of men is significantly higher.Footnote 16 In total the savings effects suggest the interpretation that women are rather concerned about having a minimum amount of savings, while a larger amount doesn’t increase their well-being (neither life satisfaction nor happiness).

We further find that people who are cohabiting with their spouse are significantly happier and satisfied than those who are not. Cohabitation also contributes more strongly to being happy than to being satisfied.Footnote 17 Looking at the separate analyses for the subsamples, we find however that this is only true for men, but not for women.

As for children, we find significant results only for elementary and junior high school age children and the children below the age of 6 years. Both make the respondents significantly happier, but only children under age 6 make them also significantly more satisfied. The impact of children under 6 years on happiness is also much stronger than on life satisfaction.Footnote 18 Looking at the gender differences , the results show that these relationships are only significant for women but not for men.

Regarding the working conditions we find that people in management positions are significantly happier and more satisfied than regular employees. In terms of gender differences however, this finding is only significant for the life satisfaction of men. Footnote 19 Male irregular employees are unhappier and less satisfied than male regular employees. Further, people who have an unemployed person in their family are also unhappier and less satisfied (whole sample as well as men and women separately), with the effect being more strongly pronounced for life satisfaction than for happiness.Footnote 20

12.6 Conclusion

Our analysis of the National Survey on Lifestyle Preferences of the years 2010–2012 contributes to the discussion regarding gender and age effects in the following ways. First, the universality of U-shaped age effects reported by Blanchflower and Oswald (2008) could be confirmed.Footnote 21 While a U-shape can be seen already in the raw data of the male subsample, the regression analysis shows that the marginal age effects of women also describe a U-shaped pattern. Second, regarding the gender effects our analyses lend further support to our previous finding (reported in Tiefenbach and Kohlbacher 2013) that women in Japan are significantly happier than men. These results indicate further that women are not only happier, but also more satisfied with their lives. Third, regarding gender differences in the control variables we find for (a) family relations that while the happiness effects of co-habitation are larger for men than for women, the opposite holds true for the effects of children. Children under the age of 6 years as well as children of elementary and junior high school age only affect the subjective well-being of women. We further find for (b) financial relations that having no savings has a very high negative impact on Japanese women, while they do not show a significant positive happiness effect when having very high amounts of savings. Finally we find for (c) the work relations, especially in the case of men, that being a manager (as compared to a regular employee) has positive happiness effects; while being an irregular employee has a highly negative effects on happiness and life satisfaction. Fourth, we contribute to the discussion whether happiness and life satisfaction should be treated and analyzed differently. Although we find significant difference in the effect sizes of some variables (i.e. savings, children, marriage, and unemployment) the differences do not affect or distort the overall picture of the key well-being indicators. In analogy to Veenhoven (1991 and 2008) and Frey (2008) we conclude that the concepts can be used interchangeably, at least in the case of Japan.