Understanding Japan’s fertility rate: the possible
relation with the level of irregular employment
among men
Name: Zakaria Farrigh Student number: 10810099 Supervisor: Konstantinos Ioannidis
Contents
- Abstract p.1
- 1. Introduction p.1
- 2. Literature review pp.1-3
- 2.1 Variables from studied literature pp.3-6 - 3. Methodology and Hypothesis pp.6-7
- 4. Quantitative analysis pp.7-12 - 4.1 Descriptive Statistics pp.7-8 - 4.2 Regression analysis pp. 8-12 - 5. Discussion pp.12-14 - 6. Conclusion pp. 14-15 - References pp. 16-17
- Appendix A1: English Survey pp.18-19 - Appendix A2: Japanese survey pp.20-21 - Appendix A3: All variables, their type, and range of values. p.22 - Appendix A4: Breusch-Pagan / Cook-Weisberg output. p.23 - Appendix A5: Breusch-Godfrey LM test for autocorrelation output p.24 - Appendix A6: Regression 2 without Newey-West standard errors p.25 - Appendix A7: Breusch-Pagan / Cook-Weisberg test output regression 2 p.26 - Appendix A8: Breusch-Godfrey LM test output regression 2 p.27
Statement of Originality
This document is written by Student Zakaria Farrigh who declares to take full responsibility for the contents of this document.
I declare that the text and the work presented in this document are original and that no sources other than those mentioned in the text and its references have been used in
creating it.
The Faculty of Economics and Business is responsible solely for the supervision of completion of the work, not for the contents.
1 Abstract
This paper investigates whether there is a relation between irregular employment among men in Japan, and the fertility rate in Japan -and if so, whether this relation is positive or negative-. The sample under study consists of 243 subjects from the Japan Post Bank. The results show that there indeed is a relation between the two. To be more precise a negative relation. However, this conclusion is not definitive, more research needs to be done for a more definitive conclusion.
1. Introduction
One of the problems developed countries are facing is an insufficient birth rate. One of these countries is Japan. Japan currently has a birth rate equal to 1.46 births per woman (source: OECD), while the replacement level for developed countries is equal to 2.1 (DeStefano and Kabaklarli, 2011, p.68). An insufficient or decreasing birth rate brings several economic problems to a country, for example a reduction in the future workforce, which will create an imbalance between pension contributors and pensioners (Zambaa & Hassen, 2015, pp.233-234). Furthermore, a low fertility rate causes an increase in capital, output, and pension’s payout per worker (Zambaa & Hassen, 2015, pp.233-234).
Several papers and articles have been written regarding this topic, but the ones of interest are a paper written by Patricia Boling and an article written by Alana Semuels. Boling tries to explain the level of Japan’s fertility rate by looking at seven variables, and at the same time she compares Japan’s situation with that of France. The variable of interest here is the amount of hours worked by the spouse. This variable is based on the expectations of ideal workers in Japan. One of the main features of the Japanese employment system is the idea of lifetime employment, where Japanese starters on the labour market would receive training from a company and receive lifetime employment at the same company, paired with certain benefits (Ono, 2010, pp.1-3). On the contrary, Semuels (2017) explains Japan’s low fertility rate by looking at the increasing amount of irregular workers.
2. Literature review
According to Boling (2008, pp.315-316) Japanese companies expect complete devotion of their workers to their job, no questions asked -especially the male employees-. This includes overtime work, transfers -even if it means living away from the employee’s family for an extended period of time-, going out drinking, participating in company’s sports activities, going on trips, and so on (Boling, 2008, p.316). Particularly during times when the company is facing economic hardships, men feel a responsibility to completely devote themselves to the company and do not want their personal lives to stand in the way, even if this means that they cannot be
2
there for their family or help around the house (Boling, 2008, p.316). This means that all household tasks and other activities would fall upon the wife, but also all childrearing, like supervising their studies, managing their extracurricular activities, and more (Boling, 2008, p.316). All these responsibilities, which the wives will have to take, tend to demotivate women from starting families, thus also from giving birth (Boling, 2008, p.316-317).
Furthermore, women aiming for a successful career are also expected to put 10 to 15 hours a day of work, and are penalized when leaving work earlier to do the household tasks (Boling, 2008, p.316). Japan has a demanding corporate culture, which makes it difficult for women to either have their own family, or to maintain management track jobs when they start having children (Fackler, 2007). In other words, Boling suggests that there is a negative correlation between worked hours, or in other words regular employment -from now on noted as lifetime employment-, and the fertility rate. This would mean that, if men would work less hours, for example by having part time or irregular employment, the fertility rate would at the same time increase.
However, according to an article by The Atlantic, written by Alana Semuels, it might be the other way around. According to Semuels (2017), the fertility rate is decreasing because more men are being irregularly employed instead of lifetime employed. This statement also follows from Houseman and Osawa (1995, p.10), who show that during the past years part-time employment/irregular employment among men has been increasing steadily. Semuels (2017) explains this by stating that Japan is a country where men are seen as the breadwinners of the family, and that a lack of steady jobs may be creating a class of men who cannot marry and start a family because they –and their future partners- are aware that they may not be able to afford this. Semuels supports this by stating that having a stable job is especially important since approximately 70% of women stop working after their first child, and have to rely on their husband’s salary for a while (Semuels, 2017). This means that Semuels suggests that there is a positive relationship between lifetime employment and the fertility rate, and a negative relation between irregular employment and the fertility rate.
These papers give two views which both would make sense. Boling’s view is sensible, because lifetime employment both among men and women comes with many hardships and pressure which mostly fall upon the women, which would demotivate women from starting families and thus giving birth. While Semuel’s view is sensible as well: if the expected breadwinners -in this case the men- are less stable, then it would be less attractive to start a family with them, since they would be unable to sufficiently support the family, thus leading to less women giving birth. This raises the following question: Does the rising part-time
3
employment among men in Japan influence Japan’s birth rate, and if so, positively or negatively?
The aim of this thesis is to look at whether there indeed is a positive correlation between irregular employment, as Boling suggests, or whether there is a negative correlation, as Semuel’s suggests. The data for this thesis has been gathered by performing surveys among a random sample of employees at the Japan Post Bank in and around Fukui prefecture. The data will be used to perform a regression on Stata.
To shine some more light on the importance of this research: another paper regarding a similar topic has been written by Rahmqvist. Rahmqvist (2006, pp.1263-1265) looked at birth rates, abortion rates, and permanent employment rates among men and women in Sweden from 1986 to 2003. According to Rahmqvist (2006, pp.1262-1265), in Sweden, there seems to be a positive correlation between permanent employment status and the birth rate, which shows to be statistically more significant for women compared to men. What also should be noted is that this only holds for young adults in the age group of 20 to 34 years old, again, in Sweden (Rahmqvist, 2006, pp. 1262-1266). This means that Rahmqvist is unable to draw a more generalized conclusion which would hold for every country, and for every age group in the workforce older than 18 years old. Furthermore, he also does not look at the case of non-permanent employed men, which means that this paper contradicts Boling, but does not necessarily support Semuels. Other papers like Salisbury’s also look at the fertility rate and male employment, but do not consider the type of employment in relation with the fertility rate (1998, pp.266-268). This shows that it’s not only interesting to look at Japan’s case, but also necessary since there has not been a definite conclusion regarding the correlation between the two variables under study.
2.1 Variables from studied literature
For the variables used for the regression, several papers have been studied. The survey has been composed in such a way that several variables can be used for the regression. All of these variables and their correlation with the dependent variable will be discussed paper by paper. These include papers by: Woofter, Chowdhury, Goto et al., Alesina and Giuliano, Boling, Destefano and Kabaklarli, Matsumoto and Yamabe, Zámkováa and Blaškováa, and Baghestani and Malcolm.
The first paper of interest is written by Woofter. In this paper there are three variables of interest, namely: the use of contraceptive measures, the death rate of partners, and sterility (Woofter, 1949, pp.357-365). According to Woofter (1949, pp.357-359), there is a -sensible-
4
negative correlation between the use of contraceptive measures and the fertility rate. Furthermore, there seems to a negative relation between the death rate of partners and the fertility rate, this is explained by showing that due to the decrease in mortality rates, a larger number remain unbroken by death, which has a direct effect on reproduction (Woofter, 1949, pp.361-363). Finally, sterility seems to have a negative correlation with the fertility rate, which of course makes sense (Woofter, 1949, pp.364-365).
Second is the paper by Chowdhury. In this paper infant mortality - so whether the family had a child which died early or not- seems to be a variable of interest (1988, pp.666-674). Chowdhury (1988, pp.666-674) shows that there is proof for the demographic transition theory, which states that as infant mortality decreases over time, fertility rates will fall, which may happen with a long lag. Thus meaning that Chowdhury suggests a positive correlation between the two.
Third is the paper by Goto et al.. The main topic of this paper is not the fertility rate, but to explore the potential influences of unintended pregnancy on child-rearing outcomes and to conclude what actions could improve child-rearing in Japan. The paper claims that unintended pregnancy is related with an increased risk of adverse child-rearing results. This includes a decreased mother-to-child attachment, an increase in negative feelings of mothers, and less participation of fathers in child-rearing activities. Intuitively, this should lead to parents having less children in the long run. So, despite that Goto et al. do not investigate the fertility, it seems that they still hint towards a negative relation between having an unintended first child and the fertility rate. So this paper will include the following variable: whether the first child was unintended or not.
Fourth is the paper by Alesina and Giuliano, which looks at the relation between the divorce rate and fertility. According to Alesina and Giuliano (2006, pp.1-14) fertility goes down due to marriage instability, and that -threat of- divorces lead to less children being produced meaning that there is a negative relation between the two.
Fifth is the paper by Boling, one of the papers leading to the research question. Boling mentions two variables of interest, namely: gender in combination with the employment status (Boling, 2008, pp.312-317), and total number of hours worked per week by the subject (Boling, 2008, pp.315-316). As mentioned earlier, when it comes to lifetime male employment and the fertility rate, Boling states that there is a negative correlation between the two. When it comes to lifetime female employment and the fertility rate, Boling states that there is a negative correlation between the two as well. The last gender specific relation which has been incorporated in this thesis is the one between irregular employment among men/women and the
5
birth rate. While one is the variable under study -of which the potential correlation should be shown-, the other correlation-irregular employment among women and the birth rate- is as shown earlier positive. Again, women aiming for a successful career are also expected to put 10 to 15 hours a day of work, and are penalized when leaving work earlier to do the household tasks, which means that women will opt for irregular employment instead of lifetime employment so that they can start a family and focus on it (Boling, 2008, pp.315-317). Finally, it also has been noted earlier that the total number of hours worked per week by the subject is negatively correlated with the fertility rate.
Sixth is the paper by Destefano and Kabaklarli, which mentions a negative relation between education and fertility (2011, pp.71-72). Destefano and Kabaklarli state that this can be explained through multiple channels, but the one of interest is the cost-of-fertility hypothesis, which states that couples with a higher level of education understand, and have better access to birth control, which enables them to use it more effectively (2011, pp.71-72). A high level of education thus lowers the cost of birth control, which enables educated parents to utilize contraceptives more than others.
Seventh is the paper by Matsumoto and Yamabe. In this paper there are two variables of interest: the desired number of children, and the living area of the subject (Matsumoto & Yamabe, 2013, pp.1-8). According to Matsumoto and Yamabe (2013, pp.1-8) the desired number of children and the fertility rate are positively correlated, which makes sense. If you want four children and have only one, then you will obviously aim to produce more children. They also show that rural areas produce more children than urban areas, which means that there is also a relation between the living area of subjects and the fertility rate (2013, pp.1-8).
Eighth is the paper written by Zámkováa & Blaškováa, which states that wages -or total family income- is also a determining variable when looking at the fertility rate (2013, pp.1871-1873). They show that there is a positive correlation between income and fertility, and explain this by stating that if -household- incomes rise families are less fearful of existential problems and are not dispirited to have children (2013, pp.1873).
The final paper of interest is written by Baghestani and Malcolm, who show that the marriage rate is a related variable as well (2016, pp.432-435). They show that fluctuations in marriage, which fluctuate due to institutional changes to the structure of marriage, influence the fertility rate (2016, 443). The relation between the two is a positive one (2016, pp.435-443).
All these variables appear in the table in Appendix A3 and are easy measurable through a survey. As it may have become clear, most of the variables will be dummy variables, while
6
some other variables will be either continuous or discrete. See Appendix A3 for the table for the variables used and their details. The survey in both English and Japanese form are included at the end of the paper -Appendix A1 and A2-, as well .
3. Methodology and Hypothesis
As mentioned earlier the data has been gathered by a survey, distributed among a random sample of employees at the Japan Post Bank in and around Fukui prefecture, in Japan. Approximately 500 surveys have been distributed at offices through a high ranked representative of the Japan Post Bank. In total 270 persons filled these surveys out. However, of these 270 surveys, 13 contained blanks and 14 came in too late, which means that for the analysis 243 surveys have been used. Excluding both the incomplete surveys and the ones which were too late, gives a response rate of 48.6%. This could potentially have led to selection bias. For example, low income people don’t fill in the survey -despite it being completely anonymous- because they are too ashamed to share this, or high income people could have also not filled in the survey, because they would rather keep their wage a secret. It could also be that people with a lot of children didn’t fill this out as to not stand out, or the same for people without children. Ideally whether a survey is completed or not should be independent and completely random. However, if there really are systematic differences between the people who did fill out the survey and who didn’t, perhaps this sample is unable to capture certain effects.
The gathered data has been used to perform a regression on Stata where our dependent variable is the fertility rate per woman -or the amount of babies born per family, in Stata this dependent variable has been labelled as birth rate-, and our independent variables, which are the variables mentioned in section 2.1 and are described in Appendix A3.
After the regression we want to test whether the irregular employment status of a male influences the fertility rate or whether this effect is non-existent. Formally, our hypotheses are H1: βIRMALE ≠0 vs. H0: βIRMALE =0, where β is the coefficient of the dummy variable under study
and where “IRMALE” stands for “irregularly employed male”. The control group for this study is the full time employed men -which means that there is no dummy for full time employment among men-, while the group of interest will be the irregularly employed men. If the regression shows that all variables are efficient, and there are no signs of multicollinearity or omitted variable bias, a T-test can be performed to test the above hypothesis. If there are signs of either multicollinearity or omitted variable bias, either a variable has to be removed or added and the regression has to be performed again.
7
or smaller than zero, is because of the previously mentioned literature review. As stated earlier Boling suggests a sensible view that there should be a positive correlation between irregular employment and the fertility rate. Again having children comes with many hardships and pressure, which mostly fall upon the women if their partners are lifetime employed, since the men have to devote themselves to their work. This demotivates women from having children, showing a negative relation between lifetime employment among men and fertility, and thus suggesting a positive relation between irregular employment among men and fertility. Irregularly employed men should have more time to share the burdens with their partner, thus not demotivating women from having children, and thus increasing the fertility rate. Once more, at the other side, Semuels suggests that there is negative relation between irregular employment among men and the fertility rate. This view is sensible as well: if the expected breadwinners -in this case the men- are less stable, then it would be less attractive to start a family with them, since they would be unable to sufficiently support the family, thus leading to less women giving birth. This means that our alternative hypothesis has to be βIRMALE is not equal to zero, since the
sign of the variable could go both ways in the light of these two views.
4. Quantitative analysis 4.1 Descriptive Statistics VARIABLES mean IRMALE 0.276 IRFEMALE 0.267 LIFEFEMALE 0.103 IRPARTNER 0.317 LIFEPARTNER 0.490 HOURS 53.08 MARRIED 0.786 CHILDINTENDED 0.638 NUMBERCHILDREN 1.613 RURAL 0.547 EDUCATED 0.856 EDUCATEDPARTNER 0.580 INCOME 550,576 CONTRACEPTIVES 0.239 DIVORCED 0.0741 INFANTMORTALITY 0.107
8
As stated earlier, this analysis contains 243 subjects and they have all been questioned for the variables in tables 4.1 and 4.2. Looking at tables 4.1 and 4.2, the first thing which should be noticed is that the variables sterility and widowhood have been left out. This is because all subjects answered this question with ‘no’. This means that including these in the regression would be not useful. Furthermore, it seems that our sample consists of approximately 63% working males -where 67 males are irregularly employed and 86 are lifetime employed-, while only 37% consists of working females -65 irregularly employed and 25 lifetime employed-, which shows a seemingly large inequality in the work force. Most subjects reported to be married -78.6%-, and on average want more children than the fertility rate -1.61-, but still less than the replacement rate.
(1) (2) (3)
VARIABLES mean min max
HOURS 53.08 30 90 INCOME birthrate 550,576 1.416 210,000 0 1,000,000 5
Table 4.2: Hours worked per week and income
Looking at table 4.2, it seems that our sample mean for the birthrate -1.42- is approximately equal to the current population mean for the birth rate -1.46-. What’s remarkable as well, is that there is a substantial gap between the minimum and maximum values of hours worked per week, and the income -of which the gap is even relatively larger compared to the latter-. The subjects seem to work on average 53 hours per week, which is larger than in the Netherlands, where the average hours worked per week for women is 28.4 hours per week and for men 39.2 hours per week (source: Central Bureau of Statistics in the Netherlands). The average total family income seems to be ¥550,576 per month, which is approximately -against the exchange rate on 12/02/1028- €4,138.28, which is lower than the average in the Netherlands, which is €8041.67 for a 3.9 person household of which all children are younger than 18 years old (source: Central Bureau of Statistics in the Netherlands).
4.2 Regression analysis
In regression 1 we regress birthrate on whether the subject is an irregularly employed male or not, whether the subject is an irregularly employed female or not, whether the subject is a lifetime employed female or not, whether the subject’s partner is irregularly employed or
9
not, whether the subject’s partner is lifetime employed or not, hours worked per week,
whether the subject is married or not, whether the subject’s first child was intended or not, the number of desired children, whether the subject lives in a rural area or not, whether the subject had tertiary education or not, whether the subject’s partner had tertiary education or not, monthly income, whether the subject uses contraceptives or not, whether the subject has ever been divorced or not, and whether the family had a child which died early or not. The first regression gives the following output:
(1) VARIABLES birthrate IRMALE -0.747** (0.293) IRFEMALE -0.547*** (0.144) LIFEFEMALE 0.085 (0.219) IRPARTNER -0.312** (0.141) LIFEPARTNER -0.439*** (0.146) HOURS -0.031*** (0.006) MARRIED 0.224 (0.209) CHILDINTENDED 1.183*** (0.183) NUMBERCHILDREN 0.238*** (0.064) RURAL 0.187 (0.151) EDUCATED -0.738*** (0.149) EDUCATEDPARTNER 0.204* (0.108) INCOME -0.000 (0.000) CONTRACEPTIVES -0.174 (0.151) DIVORCED 0.866*** (0.186) INFANTMORTALITY 0.124 (0.156) Constant 2.884*** (0.503) Observations 243 R-squared 0.654
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.
10
The first thing we should do is checking whether the above regression is valid or not. Essentially we need to test whether the residuals of the regression are homoskedastic and uncorrelated. Starting with checking for homokcedasticity, this can be checked with the Breusch-Pagan / Cook-Weisberg test. The output of the test can be found in Appendix A4. The test results in the following: The P-value, which is smaller than 0.0001, shows that the estimator is not homoskedastic, but heteroskedastic. The next step is to check whether there is serial correlation or not using the Breusch-Godfrey test for autocorrelation. The output of the test can be found in Appendix A5. The test shows again, a P-value smaller than 0.0001, meaning that there is serial correlation. If there was no serial correlation, but there was heteroscedasticity, it could be solved by using Robust standard errors. However, since there is not only heteroskedasticity, but serial correlation as well, it would be better to regress using Newey-West standard errors, accounting for both issues at the same time. This regression provides the following output: (1) VARIABLES birthrate IRMALE -0.747*** (0.279) IRFEMALE -0.547*** (0.135) LIFEFEMALE 0.0850 (0.219) IRPARTNER -0.312** (0.150) LIFEPARTNER -0.439** (0.185) HOURS -0.0309*** (0.00610) MARRIED 0.224 (0.157) CHILDINTENDED 1.183*** (0.212) NUMBERCHILDREN 0.238*** (0.0560) RURAL 0.187 (0.135) EDUCATED -0.738*** (0.202) EDUCATEDPARTNER 0.204 (0.132) INCOME -2.02e-07 (4.01e-07) CONTRACEPTIVES -0.174 (0.138) DIVORCED 0.866*** (0.232)
11 INFANTMORTALITY 0.124 (0.239) Constant 2.884*** (0.504) Observations 243
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 4.4: Regression 1 with Newey-West errors
Looking at the output in table 4.4, it seems that we have seven variables which are significant against α=0.01, two against α=0.05, and none against α=0.1 -which was the case for EDUCATEDPARTNER in table 4.3-. The variable of interest -IRMALE- shows a negative correlation with the dependent variable and seems to be significant against α=0.05. Showing that we can reject our null-hypothesis. This gives us reason to believe that irregular employment among men indeed does have a negative relation with the fertility rate, as Semuels suggests.
However, the regression also shows that the following variables are not significant: lifetime employment status of a female, marital status, living area, income, whether the subject uses contraceptive measures or not, infant mortality rate, and whether the subject’s partner is educated or not. To improve the model it would be best to remove these variables and to perform the regression again. This results in the output in Appendix A6, which shows us an ordinary linear regression, leaving out the insignificant variables found in regression 1. So in regression 2 we regress birthrate on whether the subject is an irregularly employed male or not, whether the subject is an irregularly employed female or not, whether the subject’s partner is irregularly employed or not, whether the subject’s partner is lifetime employed or not, hours worked per week, whether the subject’s first child was intended or not, the number of desired children, whether the subject had tertiary education or not, and whether the subject has ever been divorced or not. Performing both the Breusch-Pagan / Cook-Weisberg test for heteroskedasticity and the Breusch-Godfrey test for autocorrelation gives the outputs in Appendix A7 and A8 respectively. Again, the regression suffers from both heteroscedasticity and serial correlation. Which means that the Newey-West standard errors have to be used, resulting in the following:
12 (1) VARIABLES birthrate IRMALE -1.016*** (0.185) IRFEMALE -0.441*** (0.112) IRPARTNER -0.307** (0.120) LIFEPARTNER -0.373*** (0.122) HOURS -0.0273*** (0.00518) CHILDINTENDED 1.141*** (0.157) NUMBERCHILDREN 0.275*** (0.0518) EDUCATED -0.690*** (0.178) DIVORCED 0.755*** (0.163) Constant 2.904*** (0.391) Observations 243
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
Table 4.5: Regression 2 with Newey-west errors.
This time, the regression reports all variables -except IRPARTNER- to be even significant against α=0.01. This includes the variable of interest IRMALE, showing that if you are an irregularly employed male, you will on average have 1.016 child less than if you were either a lifetime employed male or female.
5. Discussion
Looking at our results, the first surprising aspect -and criticism- is that the regression suffered from both heteroscedasticity, and serial correlation. Despite having solved this using Newey-West standard errors, this is still odd. This could be due to the earlier mentioned selection bias and potential systematic differences. Again, ideally whether a survey is completed or not should be independent and completely random. However, if there really are systematic differences between the people who did fill out the survey and who didn’t, perhaps this sample is unable to capture certain effects. It could also be due to the fact that the surveys have been gathered firm by firm, which means that they have been gathered group by group, which means that errors could be correlated. Worst case scenario, this could also be due to a
13
non-existent correlation in our model, or because a sample of 243 is still too small.
Looking at our variable of interest, both regressions hint towards a negative relation between irregular employment among men and fertility. Only difference between the two regressions is that in the second regression the estimated effect is larger. However, what’s important here is not the size of the coefficient, but the sign. When removing variables from the model, part of the effect of the old variable will be transferred to the other variables.
As stated earlier the reason for writing this paper is because of the two opposing views of both Semuels and Boling. Semuels believes there is a negative correlation between irregular employment among men, while Boling believes this relation should be positive. As observed both regression hint towards a negative relation, meaning that the results suggest that Semuels could be right. So the answer to the research question would be: yes, according to the model irregular employment among men does seem to influence the fertility rate, and according to this study negatively.
However, this answer is not definitive as there are some criticisms. As it may have become clear, a number of variables had to be removed from the regression, namely lifetime employment status of a female, marital status, living area, income, whether the subject uses contraceptive measures or not, infant mortality rate, and whether the subject’s partner is educated or not. These variables should be important determinants for the fertility rate according to previous literature. What should also be noted is that the variables did seem significant when they were regressed on their own against our dependent variable, but seem to lose their significance when more variables are added. Looking at LIFEFEMALE, it seems that it is not important whether the subject is a lifetime employed male or female, but whether he or she is lifetime employed in general or not. Looking at MARRIED and INCOME it seems that the effect of both is not significant in the studied sample, despite it being significant in other papers. This could be because there is some relation for example between EDUCATED and INCOME, since educated persons do earn more, or because INCOME is influenced by your household situation, where income should increase when you report to be married and perhaps have children. Also, INCOME may be insignificant because of a non-linearity between the dependent and independent variable, which should be solved using logs. However, using logs on the model only resulted into more insignificance.
What could also have influenced the significance of these variables is the fact that the sample was not 100% random, due to a another source of bias. As stated earlier the subjects work at the Japan Post Bank, which means that most people are educated people who went through the traditional process of “shūkatsu” -also known asjob hunting- after their studies and
14
wanted to start a steady life -perhaps with someone-. It would have been best to get a completely random sample which includes all kinds of subjects, this could be done by sending surveys to random people from the whole Japanese population, instead of just the Japan Post Bank around Fukui prefecture.
Another criticism which should be taken into account is that this paper did not take other important independent variables into account, like for example: the cost of raising and educating children, opportunity costs of raising children, future economic expectations by individuals, social pressure, corporate culture, and family support policies to name a few. This is due to several reasons, but the main two reasons are the difficulty of implementing those variables, and the difficulty of asking them/obtaining the date necessary through a survey.
Last big criticism which should be mentioned is that instead of a linear regression, it would have perhaps been better to use another type of regression. For example: using a logit or probit model, where the “odds” of having another child or not would have been looked at. This would also show whether irregular employment among men influences the fertility rate or not and would perhaps give more satisfying result.
The above criticisms lead to enough reason for a follow-up study to be conducted which may give a better model and a more conclusive answer to the research question. This should lead to a follow up study with a larger and completely random sample, a model with more variables -perhaps less dummies and more continuous or discrete variables- which are more significant, and a better analysis.
6. Conclusion
This research was the result of two contradictive views on the fertility rate in Japan. Boling hinted towards a positive relation between irregular employment and the fertility rate, while Semuels thought this relation should be negative. As mentioned earlier: Boling’s view is sensible, because lifetime employment both among men and women comes with many hardships and pressure which mostly fall upon the women, which would demotivate women from starting families and thus giving birth. While Semuel’s view is sensible as well: if the expected breadwinners -in this case the men- are less stable, then it would be less attractive to start a family with them, since they would be unable to sufficiently support the family, thus leading to less women giving birth.
These contradictory views raised the following question: Does the rising part-time employment among men in Japan influence Japan’s birth rate, and if so, positively or
15
negatively? Data was gathered through a survey and a model was set up to study whether the relationship exists, and if so, whether this relation is positive or negative. The data was analyzed using a linear regression, but showed heteroskedasticity and serial correlation. This has been adjusted for by using Newey-West standard errors. Both the regressions in section 4.2 show that irregular employment among men negatively influences the birth rate, which answers the research question.
However, looking at all the criticisms in section 5, these results are not definitive, meaning that there should be a follow-up study should yield better results and more definitive conclusions.
16 References
Alesina, A., & Giuliano, P. (2006). Divorce, Fertility and the Shot Gun Marriage (Working Paper No. 12375). Retrieved from National Bureau of Economic Research website: http://www.nber.org/papers/w12375.
Baghestani, H., & Malcolm, M. (2016). Factors predicting the US birth rate. Journal of
Economic Studies, 43(3).
Boling, P. (2008). Demography, Culture, and Policy: Understanding Japan's Low Fertility.
Population and Development Review, 34(2).’
Chowdhury, A., R. (1988).The Infant Mortality-Fertility Debate: Some International Evidence. Southern Economic Journal, 54(3).
DeStefano, T., & Kabaklarli, E. (2011). Economic Determinants of Japan’s Low Fertility Rate: Cointegration Analysis. Population Review, 50(2).
Fackler, M. (2007, August 6). Career women in Japan find a blocked path. Retrieved from the New York Times.
Goto, A., Yasumura, S., Yabe, J., & Reich, M. (2006). Addressing Japan's Fertility Decline: Influences of Unintended Pregnancy on Child Rearing. Reproductive Health Matters,
14(27).
Houseman, S., & Osawa, M. (1995). Part-time and temporary employment in Japan. Monthly
Labor Review, 118(10).
Matsumoto, Y., & Yamabe, S. (2013). Family size preference and factors affecting the fertility rate in Hyogo, Japan. Reproductive Health 10(1).
Ono, H. (2009). Lifetime employment in Japan: Concepts and measurements. Journal of The
Japanese and International Economies, 24(1).
Rahmqvist, M. (2006). The close relation between birth, abortion and employment rates in Sweden from 1980 to 2004. Social Science & Medicine, 63(5).
Salisbury, P., S. (1998). Factors Affecting Birth Rates among White Women 20–24 Years of Age: A Trend Analysis (January 1972–march 1992). Social Indicators Research,
43(3).
Semuels, A. (2017, July 20). The Mystery of Why Japanese People Are Having So Few Babies. Retrieved from https://www.theatlantic.com/
17
Zambaa, A., A., & Ben Hassen, L. (2014). The impact of low fertlity rate on the level of pension. Socioeconomica 3(6).
Zámkováa, M., & Blaškováa, V. (2013). Identification of factors affecting birth rate in Czech Republic. American Institute of Physics, 1558(1).
18 Appendix A1: English Survey
This survey contains two pages and has been distributed in order to gather data for the thesis by a student of the University of Amsterdam. This survey is COMPLETELY anonymous and the results are only used for research purposes.
1. Are you male or female? Male
Female
2. Are you currently married, in a relationship, or single? Married
Relationship Single
3. If you are single, were you previously married or in a relationship? Yes, but divorced
Yes, but widowed No
4. Are you and your partner regularly employed, irregularly employed, or unemployed? You: Regularly Irregularly Unemployed Your partner: Regularly Irregularly Unemployed
5. How much hours do you and your partner work per week -in numbers-?
You: Your partner:
6. What is your and your partner’s monthly salary -in numbers-?
You: Your partner:
7. What is the highest level of education you and your partner completed? You:
Junior High School High School Community college Medical School University: Undergraduate University: Graduate PhD Your Partner:
Junior High School High School
Community College Medical School
19 University: Graduate
PhD
8. How many years of education did you and your partner complete after Junior High School -in numbers-?
You: Your partner:
9. Do you live in a rural or urban area? Rural
Urban
10. Do you and your partner use contraceptive measures? Yes
No
11. Do you or your partner suffer from sterility? Yes
No
12. How many children do you have -in numbers-? Answer:
13. How many children would you want to have in total -in numbers-? Answer:
14. Do you plan on having another child? Yes
No
15. Was your first child intended or not? Yes
No
16. Did you have a child which passed away before reaching adulthood? Yes
No
20 Appendix A2: Japanese survey
このアンケートは2ページあります。また、アムステル大学の卒業論文に使うデータを集めるためにお配りし ています。この調査は完全に匿名であり、結果は卒業論文のみで利用することにし、それ以外に活用することは ありません。 １． あなたの性別をお答えください。 □男性 □女性 ２．あなたは現在結婚していますか。または、交際中ですか。 □結婚している □交際中 □独身 ３．独身と答えた方にお聞きします。あなたは以前結婚、または、交際していましたか。 □はい、しかし離婚または破局した □はい、しかし死別した □いいえ ４．あなたとあなたの配偶者（恋人）の雇用形態をお答えください。 あなた □正規雇用 □非正規雇用（派遣社員、パート、フリーターを含む） □働いていない 配偶者（恋人） □正規雇用 □非正規雇用（派遣社員、パート、フリーターを含む） □働いていない ５．あなたとあなたの配偶者（恋人）は週に何時間働きますか。数字でお答えください。 ・あなた （ 時間） ・配偶者（恋人）（ 時間） ６．あなたとあなたの配偶者（恋人）の1か月の収入はいくらですか。数字でお答えください。 ・あなた （ 円） ・配偶者（恋人）（ 円） 次のページに続きます。
21 ７．あなたとあなたの配偶者（恋人）の最終学歴をお答えください。 ・あなた ・配偶者（恋人） □中学校 □中学校 □高校 □高校 □専門学校 □専門学校 □大学（医学部） □大学（医学部） □大学 □大学 □修士 □修士 □博士 □博士 ８．あなたとあなたの配偶者（恋人）は中学校卒業後、何年間学校に行きましたか。 ・あなた（ 年間） ・配偶者（恋人）（ 年間） ９．あなたは田舎、都市部どちらに住んでいますか。 □田舎 □都市部 １０．あなたとあなたの配偶者（恋人）は避妊用具を使用していますか。 □はい □いいえ １１．あなたとあなたの配偶者（恋人）は不妊の問題を抱えていますか。 □はい □いいえ １２．子どもは何人いますか。 （ 人） １３．最終的に何人の子どもが欲しいですか。 （ 人） １４．今後、子どもをつくる予定はありますか。 □はい □いいえ １５．あなたの最初のお子さんは計画的なものでしたか。 □はい □いいえ １６．大人になる前に亡くなったお子さんはいますか。 □はい □いいえ 質問は以上です。ご協力ありがとうございました。
22 Appendix A3: All variables, their type, and range of values.
Name of variable Variable
Type
Range of values
IRMALE (irregularly employed male) dummy 1=yes, 0=no IRFEMALE (irregularly employed female) dummy 1=yes, 0=no LIFEFEMALE (lifetime employed female) dummy 1=yes, 0=no IRPARTNER(irregularly employed partner) dummy 1=yes, 0=no LIFEPARTNER (lifetime employed partner) dummy 1=yes, 0=no HOURS (hours worked per week by subject) discrete (0, 168) MARRIED (whether subject is married or not) dummy 1=yes, 0=no CHILDINTENDED (whether the first child is intended
or not)
dummy 1=yes, 0=no
NUMBERCHILDREN (the desired number of children) discrete all positive whole numbers
RURAL (living in a rural area or not) dummy 1=yes, 0=no EDUCATED (whether the subject had tertiary education
or not)
dummy 1=yes, 0=no
EDUCATEDPARTNER (whether the partner had tertiary education or not)
dummy 1=yes, 0=no
INCOME (total family income) continuous (0, infinity) CONTRACEPTIVES (whether person/partners are
using
contraceptive measures or not)
dummy 1=yes, 0=no
DIVORCED (whether he/she has ever divorced or not) dummy 1=yes, 0=no INFANTMORTALITY (whether the family had a child
which died early or not)
dummy 1=yes, 0=no
STERILITY (whether one of the partners suffers from sterility or not)
dummy 1=yes, 0=no
WIDOW (whether the subject has had a previously deceased partner)
23 Appendix A4: Breusch-Pagan / Cook-Weisberg test for heteroskedasticity output
Breusch-Pagan / Cook-Weisberg test for heteroskedasticity : Ho: Constant variance
Variables: IRMALE IRFEMALE LIFEFEMALE IRPARTNER LIFEPARTNER HOURS MARRIED CHILDINTENDED NUMBERCHILDREN RURAL EDUCATED EDUCATEDPARTNER
INCOME CONTRACEPTIVES DIVORCED INFANTMORTALITY Chi2(16) = 57.98
24 Appendix A5: Breusch-Godfrey LM test for autocorrelation output
Breusch-Godfrey LM test for autocorrelation
--- lags(p) | Chi2 df Prob > Chi2
---+--- 1 | 163.405 1 0.0000
--- H0: no serial correlation
25 Appendix A6: Regression 2 without Newey-West standard errors
(1) VARIABLES birthrate IRMALE -1.016*** (0.174) IRFEMALE -0.441*** (0.124) IRPARTNER -0.307** (0.129) LIFEPARTNER -0.373*** (0.121) HOURS -0.0273*** (0.00544) CHILDINTENDED 1.141*** (0.150) NUMBERCHILDREN 0.275*** (0.0585) EDUCATED -0.690*** (0.137) DIVORCED 0.755*** (0.164) Constant 2.904*** (0.379) Observations 243 R-squared 0.638
Standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1
26 Appendix A7: Breusch-Pagan / Cook-Weisberg test for heteroskedasticity output for regression 2
Breusch-Pagan / Cook-Weisberg test for heteroskedasticity Ho: Constant variance
Variables: IRMALE IRFEMALE IRPARTNER LIFEPARTNER HOURS CHILDINTENDED NUMBERCHILDREN EDUCATED DIVORCED
Chi2(9) = 25.63 Prob > chi2 = 0.0023
27 Appendix A8: Breusch-Godfrey LM test for autocorrelation output for regression 2
Breusch-Godfrey LM test for autocorrelation
--- lags(p) | chi2 df Prob > chi2 ---+--- 1 | 163.126 1 0.0000
--- H0: no serial correlation
