Master thesis
Author: Ida Viklund
Thesis advisor: Karin Edmark Spring 2010
The Swedish parental benefit and its effects on childbearing
- An empirical study of whether raised compensation levels within the parental benefit causeABSTRACT
In this study I analyze the Swedish parental benefit system focusing on a certain component, the basic level, and its effects on childbearing. The basic level consists of a fixed monetary amount given to parents. The compensation level was raised three times between 2002 and 2004, which would, according to economic theories, imply increased incentives for having children. I use longitudinal data of women in fertile age between 1993 and 2005. Applying a difference-in-differences model on comparison groups based on region of birth I conclude that the policy changes probably did not cause increased probabilities of having children.
INDEX
1. Introduction ... 4
1.1 Disposition...5
2. The parental benefit system in Sweden... 6
2.1 The compensation levels...6
2.2 Alterations within the basic level ...7
2.3 Other alterations...8
3. Theories and earlier research ... 9
3.1 Fertility economics...9
3.2 Fertility rates in Sweden...9
3.3 Family policy and fertility rates ...10
4. Data procedure ...13
5. Methodology ...14
5.1 Before and after design and difference-in-differences ...14
5.1.1 Specifying the intervention periods ... 16
5.2 Logistic regression ...17
5.2.1 The logistic model without control variables ... 17
5.2.2 The logistic model including control variables ... 18
6. Results...20
6.1 Trends in fertility rate ...20
6.2 The logistic model without control variables...21
6.3 The logistic model including control variables...23
1. Introduction
Does family policy affect people’s decision to have children? The reasons why people have children may be as many as the number of citizens in the world. Still most of the OECD countries have family policies containing measures for stimulating fertility rates. The family policies of Sweden are diverse and include different purposes and measures regarding working life, parental benefit, childcare, and child allowances. In this thesis I will focus on the parental benefit and its impacts on childbearing.
When the Swedish parental insurance was introduced in 1974, replacing the earlier maternity allowance, the purpose was to secure women’s long-term participation in the labour force. Women (and men) were now entitled to compensation based on their loss of income when staying at home with their children. This construction would facilitate the combination of family life and working life. Put in economic terms, the alternative costs of having children were reduced through the reform (SOU 2005:73, p 107, 115ff).
Since the introduction the system has been revised a number of times and at present it consists of three different levels. The first level, the sickness benefit level, is based on the parent’s present income1. Because of that it’s the part best corresponding with the original idea of compensating the parent for the loss of income. The second level, the basic level, is not based on income. It is given to individuals that, in short, have not worked sufficiently for receiving the sickness benefit level. The basic level consists of a lower fixed amount and serves more as a cost-neutraliser of having children, than actually compensating for loss of income. The basic level is the issue of interest in this thesis. The third level, the minimum level, is equally given to everybody after one’s ordinary parental benefit days have ended. It also consists of a lower fixed amount. The parental benefit system will be further explained in the next chapter.
The compensation given at the basic level was more or less constant during the 1990’s (60 SEK per day) but it was raised three times, that is in 2002, 2003, and 2004. As from 2004 the compensation is 180 SEK per day. These alterations could affect fertility rates positively according to economic theories but it is unlikely that one would observe changes if studying the whole population on aggregate level. The reason is that most of the citizens use the sickness benefit level, and not the basic level, and therefore are not concerned by these policy changes (Försäkringskassan 2010).
Duvander and Eklund (SOU 2006:37, 54ff) studied usage of the basic level among groups of mothers divided on region of birth, for children born in 1999. They showed that women born outside Sweden, and particularly in regions outside Europe, utilized the basic level to a greater extent than native born women. This is the starting point for my study and what brings the comparison groups. Groups using the basic level to greater extent than others would, on aggregate level, be more affected by alterations in the compensation level. Accordingly, changes in fertility rates due to these alterations would be greatest for these groups. 2
By analyzing fertility rates (or probabilities of having children) for the different comparison groups before and after the policy change and comparing the group-specific differences, I am able to draw conclusions about whether this policy change have had any impact on childbearing. The underlying hypothesis is that possible changes in probabilities of having children between the pre intervention period and the post intervention period should be greater for certain immigrant groups.
Thus my research question is:
Does an increase in the compensation level within the basic level cause increased probabilities of having children among individuals that were eligible for the basic level only?
1.1 Disposition
After the introduction follows chapter 2 further describing the Swedish parental benefit system, the compensation changes within the basic level, and other important alterations. In chapter 3 I will discuss theory and earlier research regarding the parental benefit. The chapter consists of a theory section that connects childbearing to economics, a section that presents earlier research regarding fertility rates in Sweden, and lastly, a section that discuss the impacts of different family policies on the chosen comparison groups. The database and the data procedure are discussed in chapter 4. The methodology is presented in chapter 5. This chapter is divided into two parts that treat the method “before and after design and difference-in-differences” and the logistic regression models. The results of the regression models including a reprecification of the model are presented in chapter 6 followed by the analysis and conclusions in chapter 7 respectively chapter 8.
2
2. The parental benefit system in Sweden
The Swedish parental benefit has been considered a generous insurance on account of the compensation levels and the large amount of days off that parents are entitled to. Since the introduction it has been prolonged and extended in several steps, aims have been reformulated, and a fixed number of days for each parent have been stipulated. I will describe some of these policy changes below, including the changes within the basic level and other changes that are of importance for the analysis.
The parental benefit system was introduced in 1974 and it implied vital changes in the view of women’s participation in the labour force and parents’ equal responsibility for their children. The previous maternity allowance consisted of a basic amount given to all mothers, irrespective of whether they were part of the labour force. But through the introduction of the parental insurance fathers were also entitled to use the parental benefit and women were further encouraged to participate in the labour force. It was appreciated a necessity to compensate the parents for their loss of income when staying at home. The overall purpose of the parental benefit was, and still is, to facilitate for both parents the combination of family life and gainful employment or studies (SOU 2005:73, 107-115).
2.1 The compensation levels
In the introduction I briefly described the different levels of the parental benefit; the sickness benefit level, the basic level, and the minimum level. I will discuss them a bit further and give closer details for a better understanding of the system. All parts are taxable. The sickness benefit level is calculated on one’s sickness benefit qualifying annual income and gives the parent compensation at 80 % of this income. The sickness benefit qualifying annual income is based on one’s actual income. It does need to be the same sum. This special income forms the basis of calculations of reimbursements given within the general social insurance in Sweden. This includes reimbursements such as parental benefit, and sickness benefit if one falls ill. The sickness benefit level given within the parental benefit is a generous compensation but it is only given to those who fulfil certain requirements.
parental benefit on basic level (SOU 2005:73, 81). The compensation given within the basic level consists of a fixed amount. Since 2004 this amount is SEK 180 per day, which is little compared to the recompenses within the sickness benefit level. The construction of the system therefore strongly encourages the citizens to have an employment before having children. The third part, the minimum level, is SEK 180 per day for children born after July 1st 2006, and was SEK 60 per day prior to that. The minimum level is equally given to everybody (i.e. it is not based on one’s income level) after the first 390 days, for a maximum of 90 days. This means that the parents together obtain the sickness benefit level or the basic level for the first 390 days, and afterwards may utilize the minimum level for some 90 days extra. Accordingly, the maximum number of days granted and paid within the parental benefit system is 480 days.3
2.2 Alterations within the basic level
The changes in the basic level are described below. Note that it was formerly called guarantee level. Subsequently I will solely use the notation basic level.
1987: the guarantee level is raised to SEK 60 per day. 1994: the guarantee level is raised to SEK 64 per day. 1995: the guarantee level is lowered to SEK 60 per day.
2002: the basic level is introduced and guarantee level is ceased. Raise up to SEK 120 per day.
2003: the basic level is raised to SEK 150 per day. 2004: the basic level is raised to SEK 180 per day.
The raises in 2002, 2003, and 2004 are of interest in this study and investigated period is 1993 to 2005. The compensation raise in 2002 entailed a doubling of the previous level.4 Until 2002 the maximum compensation received each month was SEK 1 800. From 2004 this amount was, and still is, SEK 5 400. The sums are small compared to the amounts received within the
3
That is 480 full days. One may also have the benefit spread and prolonged over time by being compensated a lower amount each month.
4 Strictly speaking, due to inflation, it’s not an exact doubling. Between 1993 and 2005 the inflation rate was
sickness benefit level. Yet the relative increases are large and possible effects on childbearing are not unlikely. Obviously other factors affect childbearing as well, but I am interested in the specific effects of the raises in compensation level.
2.3 Other alterations
One needs to discuss other changes in the parental benefit system that may affect the incentives for childbearing within the control group. If not considering these aspects a possible impact on childbearing would be either overestimated or underestimated.
The parental benefit allows the parent entitled to the sickness benefit level 80 % of the sickness benefit qualifying annual income. Until 1995 the level was 90 %. In 1995 it was reduced to 80 % and during a short period it was further reduced to 75 % before the current level was stated in 1998 (SOU 2005:73, 113f). It is possible that these reductions affected the incentives for childbearing within the control groups. However, if these changes had coincided with the increases in basic level it would have been problematic, since the regression model would overestimate the effect of the raises in basic level.
Another alteration that actually coincided was the introduction of the second month of paternity leave in 2002. Equivalently the number of granted days off was prolonged one month, to 390 days. The 90 days with minimum level were unchanged. The effects on childbearing due to this policy change are not certain. I have not found any reason for assuming that fertility rates would be changed in neither direction.
In 2002 there was a policy reform that further increased the existing possibilities of shorten working hours and correspondingly receive compensation from the parental benefit for parents with small children (SOU 2005:73, 114). This reform should facilitate the combining of family life and working life, which might imply increased incentives for childbearing. However, both treatment groups and control groups would be concerned by the reform, wherefore I assume that changes in incentives are equal.
3. Theories and earlier research
3.1 Fertility economics
In the beginning of the 1960’s the economist Gary Becker discussed fertility economics, explaining how women and men make calculations about fertility (or the number of children wanted) on economic grounds (Becker 1991, 94ff). Referring to microeconomic theories childbearing and children are associated with utility and two distinct costs. The first type of costs, direct costs, is costs for housing, food, clothes, and other daily expenditure. Indirect costs are loss of income, or the alternative cost. The greater one’s income the greater the alternative cost of having children. This relationship is especially true for women, since they in general are the ones having the main responsibility for raising children. The main purpose of the Swedish parental benefit system is to reduce the indirect costs of having children by providing compensation for the loss of income. Yet, the basic level does not function equally, since it provides compensation for those who do not have an income, or for those having a very low income. Therefore it rather functions as a neutraliser of the direct costs. Assuming children being a “good”, which Becker does, costs affect the willingness to have children negatively. If reducing the costs associated with children, like the basic level does, the willingness for having children should inversely increase. This relationship is a basic assumption for the hypothesis that is tested in the thesis.
3.2 Fertility rates in Sweden
The total fertility rate5, which is the most common used measure of fertility, has varied a lot in Sweden during the past decades, and the fluctuations tend to follow economic trends in the society. The figure from SCB (Statistics Sweden) below illustrates this relationship.
5This is an estimation of the number of children an imaginary woman would have during her lifetime.
Figure 1. Total fertility rate (number of children per woman) in Sweden between 1988 and 2009.
Source: SCB
At the end of the 80’s and the beginning of the 90’s there was a baby boom in Sweden, which was followed by a decline parallel with the economic downturn in the society. In 1998 and 1999 the rates were as low as 1,51 respectively 1,50 children per woman. The so-called replacement level, the level at which the population reproduces itself, is 2,1 children per woman. A smaller number implies a decline of the population, if there is no migration that covers the deficit. Thanks to migration the Swedish population is not declining. Since 1999 the rates have steadily increased and in 2009 the fertility rate was 1,94 children per woman (SCB 2010b).
Studies by SCB have indeed shown that the childbearing patterns in Sweden are connected to the situation on the labour market (SCB 2008:1, 1998:1). The SCB study from 2008 showed that the propensity to have the first child increases when employment level is higher. The authors also concluded that native born and foreign born women in the labour force had a larger probability of having their first child at a given point in time, compared to women that were not in the labour force. An implication would be that periods associated with large unemployment rates and economic downturn would have declining or lower fertility rates. This assumption seems to be supported by figure 1 above.
3.3 Family policy and fertility rates
explanation is the so-called “migration effect”. This effect may possibly originate from the fact that a larger share of the women migrating from these countries is asylum seekers or relative immigrants, compared to immigrants from other regions. These women might have had to postpone childbearing earlier, wherefore one observes high fertility rates within these groups. Notable is that this effect is applicable only if one focuses on the first child. Referring to third and fourth childbirths the “migration effect” is no longer valid. The levels might instead be explained by the childbearing period being initiated in younger ages, which allows the woman to have more children during lifetime.
Studies have also shown that women from different birth regions might not respond equally on child-stimulating measures introduced by the state (SOU 2005:73). In the 1980’s a “speed-premium” reform was introduced destined to improve the economy of families with children. Because the sickness benefit level of the parental benefit is based on the parent’s income, and because particularly women tend to work part-time after the first child is born, this would imply that the compensation level for the following child would be low. These consequences were neutralized by the “speed-premium” reform, which allowed the parents to keep the level of the parental benefit received for the first child also for the following children. The reform was called the “speed premium” among the public as it required that the following child was born within a specific time period. This reform did not only shorter the intervals between the births, but was also a contributing factor to the high fertility rates in Sweden in the end of the 1980’s and beginning of the 1990’s (SOU 2005:73, 158).
The “speed premium” did however not have effects on all immigrant groups (Andersson et al 2006). Based on their quite broad division of immigrant groups they concluded that Nordic born women changed their behaviour in a similar way as native born, but non-Nordic born did not adapt their childbearing due to the reform. The authors concluded that explanations for the non-occurring effect might be that they were not concerned by these changes, since unemployment levels are higher and part time work is less common, or that they for some reasons did not obtain the information, or that they were just less willing to adapt their behaviour for other reasons (Andersson et al 2006, 51-70).
benefit on this level was 4 %. As a comparison 34 % of the women born in North America, 55 % of women originating from Asia, and 63 % of women born in Africa, had parental leave on basic level. Table 1 below illustrates these differences.
Table 1. The share of women, divided on region of birth, receiving the basic level. The selection consists of all mothers to children born in Sweden in 1999.
Region of birth
Share of women receiving basic level (%)
Sweden 3,6
Other Nordic countries 6,2
EU15 except Nordic countries 17,9
Europe except EU15 45,7
Africa 62,8 Asia 55,2 North America 33,8 South America 28,2 Oceania 18,4 Source: SOU 2006:37
As can be seen from the table the share of women using the basic level varies with region of birth. The authors discuss institutional factors explaining the divergent use of basic level, such as structures within the labour market and access to information. They further discuss family strategies and attitudes to childcare that possibly may be explaining factors (SOU 2006:37, 37 ff). Yet, they underline the deficient knowledge regarding explaining factors for the divergence.
The study by Duvander and Eklund does not say anything about the utilization during an extended time period, and I have not found other studies investigating this issue either. It is however not unlikely that these patterns of unequal usage would be valid other years, bearing in mind the unequal situation at the labour market (SOU 2006:37, 38).
4. Data procedure
In the thesis I use the database LINDA, a longitudinal individual database, maintained by SCB. This database contains a selection of 3 % of the Swedish population and in addition a selection of 20 % of the foreign born citizens. The selections are not overlapping. The data on the selected individuals have been supplemented with the individuals’ family members, which makes it possible to considerate the qualities of the partner in the regression. The shortfall of the dataset is the definition of a family. It is not possible to connect individuals registered at the same address if they are not married or do not have common children. This is unfortunate since a large share of all births in Sweden takes place in families with non-married parents. As a consequence I’m not able to take the income of the other parent into consideration. Due to other technical difficulties I cannot base the analysis solely on married women. Therefore I will found the analysis only on the qualities of the women themselves.
The studied period is 1993 to 2005. The upper limit is set as data is not available after 2005. Females between 19 and 49 years old are in focus of this study, i.e. women in their fertile period. Each individual in the dataset has a personal identification and likewise a family identification, which allows the connecting of newborn children to their mothers. Through this procedure the binary dependent variable is derived. Thus, if a woman has a child a specific year she is coded 1, and otherwise 0. The data constitutes of approximately 150 000 women every year in the specified age interval. From one year to another individuals will fall out of this defined age interval and new ones will be included. The number of studied cases during the whole period is close to 2 million.
The study of childbearing is done with comparison groups based on region of birth, as was discussed in the previous chapter. In the dataset the women have been coded accordingly. Clearly, coding was not possible in those cases where no information was found regarding the woman’s country of birth. In these cases the woman has been excluded from the selection. This comes for approximately 40 000 observations.
5. Methodology
The framework of this study is a quasi-experiment, a feature common in social science (Meyer 1995). It implies that one uses a policy change as an exogenous or a natural variation in the explanatory variable. In this study the policy change is equivalent to the raises of the compensation level in the basic level, and childbearing is the dependent variable. Another characteristic of this study and most of the studies within social science is that the different comparison groups are not randomly assigned.
The main methodology of the thesis are the before and after design and difference-in-differences, and a logistic regression model. I have discussed childbearing in terms of fertility rates and probabilities of having children. When referring to childbearing in the public debate one often uses the term fertility rate. Many studies that I have referred to likewise use fertility rate. Probabilities of having children follow the fertility rates in many aspects, although strictly speaking they are not equal. In this thesis the outcome of the logistic regression model will be expressed in odds ratios, which are built upon probabilities of having children. These concepts will be explained in section 5.2. To facilitate the description of the difference-in-differences model I will continue to use the term fertility rate in section 5.1.
5.1 Before and after design and difference-‐in-‐differences
Table 2. Difference-in-differences
Groups Before After After-before
Increased incentives A B (B-A)
No increased incentives C D (D-C)
Difference-in-differences (B-A) - (D-C)
Source: Edmark 2007
The model will control for all trends in fertility that are equal among the groups. The equal trends in fertility is a basic assumption of the model, and one may discuss whether it is valid for my comparison groups. A demographic report by SCB showed that in general fertility rates are larger for foreign born women than for native born, although the trends in fertility are quite similar (SCB 2008:2, 18ff). The division of women in the referred study differs slightly from that used in my study. SCB use categories of women born in: the Nordic countries, EU-states, other European countries, and other countries of high, medium respectively low Human Development Index6. Since fertility trends in the listed groups are
similar I will assume this state also for my comparison groups.
The comparison groups in this study are based on region of birth and accordingly are: Sweden, Other Nordic countries, EU15 except the Nordic countries, other European countries, Africa, North America, South America, and Asia. As mentioned earlier Oceania has been excluded from the analysis due to very small selection. An advantage of founding the comparison groups on region of birth is that it’s exogenous in respect to the reform, i.e. region of birth is not affected by alterations in the reform. Swedish born women serves as a reference group. It’s originating in that they on aggregate level are not influenced by policy changes within the basic level, or more precisely, they are barely affected by policy alterations. Approximately 4 % of these women use the basic level, see table 1 at page 13. Nevertheless, the great differences in use of basic level between the comparison groups motivates native born women used as a control group. The expected outcome of the model is that groups that are more dependent on the basic level consequently would show greater divergence in fertility between the periods.
The econometric specification of the model with only two groups (treatment group and reference group) is listed accordingly:
6 Human Development Index is a measure used to compare countries in terms of development level.
yjit = α + α1dt + α1dj + βdjt + εjit
where y is the dependent variable (fertility) and is indexed j representing the outcome for the treatment group or the comparison group (j = 1 respectively 0).
t stands for the time period, t = 1 after the treatment (post intervention period), and t = 0
before the treatment (pre intervention period). Accordingly dt is given the value of 1 if t = 1
and otherwise 0. The corresponding coefficient, α1, captures the effect of time on fertility on all groups.
α1 on the other hand represents the time-invariant difference in overall means of fertility between the different groups. As was explained earlier fertility rates differ between the comparison groups (although the trends may be equal) and this feature is taken into consideration by the variable dj.
djt is a dummy variable and would have the value of 1 if j=1 and t=1. In other words djt = 1 signifies the experimental group after the treatment. β is accordingly the difference-in-differences estimator and the causal effect of the treatment. The econometric specification of β is expressed below:
β^dd = Δy10 – Δy00 = y11 – y10 – (y01 – y00)
The subscript indicates the time period whereas the subscript indicates the group. The bar signifies an average over unit i. This specification is exactly the same as was expressed in table 2.
5.1.1 Specifying the intervention periods
pre period and the post period. In calculations of the difference-in-difference estimator (djt) dt remains 1 or 0.
5.2 Logistic regression
The second part of the methodology chapter treats the logistic regression models. These models are the statistical procedure through which the difference-in-differences estimator is calculated. Sections 5.2.1 and 5.2.2 are based on Gujarati and Porter (2009, 541-81), and a study by Arbetsgivarverket (Swedish Agency for Government Employers) (2007:1, 13 f).
5.2.1 The logistic model without control variables
The dependent variable of this study is fertility rate or childbearing. In the statistical procedure this variable slightly changes character. One calculates probabilities of having children. This comes as the regressand in my dataset is dichotomous and can take only two values, that is, 1 if the woman has a child, or 0 if the woman does not have a child a specific year. Therefore the basic assumptions of the ordinary least square model (OLS) are not valid, and a logistic binary regression model is used. The theoretical principle of the linkage between difference-in-differences and logistic regression is that one calculates the probability of having a child in each period, and studies the change in probability from one period to the next. The results will however be presented in odds ratios, which is the ratio of two probabilities; the probability that the event will occur divided by the probability that it will not occur. Basically the odds ratio signifies the risk of having children given certain factors or circumstances such as region of birth, pre or post intervention period and other control variables.
The basic logit regression model is specified accordingly:
Logit(Pi) = ln(Pi /(1- Pi)) = β1 + β2Xi +ui
where i=1,….,n and symbolize each individual. P stands for probability and the term Pi /(1- Pi) is the odds ratio. However, the function is expressed in terms of logarithmic odds ratio. The expression may be simplified by taking the natural logarithm of the function, see below:
This expression clarifies that one calculates the probability of an event occurring (i.e. y=1)
given Xi , which in this case stands for all explanatory variables. Using the variables from the
difference-in-differences function the united model would, basically, be expressed as:
Pi = Prob(Yi= 1⏐Xi) = F(time, region, time*region) (1)
5.2.2 The logistic model including control variables
There might be systematic covariance between certain factors and childbearing that could explain varying outcomes for the comparison groups. To isolate the effect of the raise in compensation level on fertility (or probability of having a child) one needs to control for these differently impacting factors.
In chapter 3 I discussed the “migration effect” – that women recently immigrated tend to have initial high fertility rates, if studying rates of first child births. In particular, this effect is large for women that have emigrated from medium and less developed countries. This will be considered in the extended model with control variables. The treatment groups within this study are not divided according to level of development, and region of birth instead comes to serve as an approximation. Since the effect is valid if the woman immigrated to Sweden maximum two years earlier, I have created a dummy variable taking the value of 1 if the woman migrated within this period, and otherwise it takes the value of 0. The migration dummy is tied to each region.
A relatively large share of women in their most fertile period in the pre intervention period or post intervention period could affect on the results, wherefore age is controlled for. The impact of age is not assumed to be linear wherefore I use categories of age classes as dummy variables. The categories are divided on five year classes.
Further, the woman’s income is used as a control variable. It is inserted as a lagged value. Ideally one would have information on the income of the partner, since a household income would probably give a better estimation. Due to limitations in the dataset (discussed in chapter 4) I’m not able to use a household income.
may however obtain other subventions, such as social assistance and housing allowance. The former is based on established norms differentiated upon the size of the household (number of persons) and it aims at covering the household’s daily expenditure. Housing allowance is given to youths aged 18-29 and families with children. According to a report by Dahlberg et al (2009, 18 f) the propensity of receiving social assistance is augmented if the individual has a low education level, is born outside Sweden, or is single – in particular a single mother. It is therefore likely that individuals receiving parental benefit on basic level are entitled to other kind of support.
The increase in the basic level in 2002 might, for some individuals, be defrayed with an equally large decrease of the social assistance due to changed compensation rules (Swedbank 2001). The implications are that some individuals in the dataset might not experience an improved economic situation nor increased incentives for childbearing. This may make the effect of a raise in compensation level within the basic level smaller (in aggregate) than would otherwise have been.
To conclude the above discussion the basic specification of the logistic model with inserted control variables is:
Pi = Prob(Yi= 1⏐Xi) = F(time, region, time*region, age, migration, income, unemployment)
6. Results
This chapter is divided in three parts. Initially, in 6.1, I will illustrate the overall trends in fertility rate for the comparison groups between 1993 and 2005. Sections 6.2 and 6.3 present the results of the logistic regression models. The latter section also includes certain adjustments of the specified model.
6.1 Trends in fertility rate
The figure is based on data of frequencies of newborn each year and their mothers’ region of birth. The underlying data can be found in appendix 2.
Figure 2. Fertility rates between 1993 and 2005. The y-axis measures the number of children born per 1 000 women.
Source: LINDA (SCB)
As the graph illustrates most of the group-divided rates are between 40 and 80 children per 1000 women each year. The fluctuations impede comprehensive analysis of individual years, although trends are visible. The basic assumption of similar trends in fertility in the difference-in-differences model seems to be fulfilled to some extent.
The rates tend to fluctuate a lot for women originating from the EU15-states, Asia, and Africa. It’s wise to interpret the fluctuations carefully. The size of the selections from one year to another might play an explaining role for the variations. In an earlier referred report by SCB it was likewise shown that the fertility rates tend to fluctuate a lot between different periods (SCB 2008:2, 21). However, it is very interesting that women born in Africa show a declining (and not increasing) fertility rate in the post intervention period. Also note that the fertility patterns for native born women confirm the cyclic fluctuations in fertility rates shown in the earlier figure 1 at page 10.
One may discuss whether the reform in 2002 was expected among the public. Expectations would possibly lead to a drop in fertility rate prior to the reform. A reasonable assumption is that such a drop would be rather similar for the comparison groups. This seems not to be the case if studying the figure above. For women born in Asia and EU15 there seem to be a declining rate in fertility prior to the reform, however, the direct linkage to possible expectations is vague. From figure 2 it’s difficult to see a notable increase in fertility rates from 2003 onwards, the period in which alterations should be observed if the compensation rise had had any impact on fertility.
6.2 The logistic model without control variables
The results of the binary logistic regression are presented below. Note that descriptive data for the variables are presented in appendix 3.
Table 3. Results of the logistic regression model. n =1 979 861
Variables
Log odds ratio (B-coefficient)
Standard
error Odds ratio
Nordic countries -0,333 0,021 0,717* EU15 (non-Nordic) -0,078 0,037 0,925** Europe (non-EU15) -0,172 0,021 0,842* Africa 0,933 0,024 2,542* North America 0,088 0,059 1,092 South America 0,018 0,037 1,018 Asia 0,383 0,015 1,647* Sweden 1 (ref) Nordic countries -0,007 0,045 0,993 EU15 (non-Nordic) 0,049 0,070 1,051 Europe (non-EU15) -0,011 0,037 0,989 Africa -0,318 0,043 0,728* North America -0,018 0,104 0,983 South America -0,003 0,068 0,997 Diff-in-diff (djt) Asia -0,119 0,025 0,888* Constant -2,782 0,010 0,076* Source: LINDA (SCB)
* = significant at 0,01 level, ** = significant at 0,05 level, *** = significant at 0,10 level
Of interest in this table is mainly the odds ratio column, which is more easily interpreted than the log odds ratio (or the beta coefficient). The odds ratio is the ratio of the probability that an event will occur divided by the probability that it will not occur, i.e. the probability of having a child to the probability of not having a child given the explanatory variables. Odds ratios larger than 1 for a specific variable signifies an increased risk or chance compared to the reference group (or point), ceteris paribus. Similar, odds ratio less than 1 signifies a reduced likelihood compared to the reference group.
The odds ratios for the comparison groups (dj) have turned out as expected, based on figure 2 presented above and descriptive data in appendix 3. Women born in Africa and Asia show the greatest odds ratios and do likewise show the greatest fertility rates in figure 2. Women originating from other Nordic countries than Sweden have lower odds ratios. The factor is 0,717, which is statistically significant. Figure 2 likewise confirms that women born in the Nordic countries have lower fertility rates. All regions are significant but North and South America. The reasons for the insignificances are probably the close fertility rates to Sweden and the great fluctuations shown.
The difference-in-differences estimator, dj
table. In terms of odds ratios the estimator measures the difference in probability of having a child to not having a child, from post intervention period to pre intervention period. The comparison for each region is made to Sweden. A bit surprising the outcomes of Africa and Asia do not turn out as expected and instead show statistically significant smaller odds ratios compared to Sweden. A value less than 1 for the difference-in-differences estimators implies that the difference (in probabilities of having children to not having children) is smaller than for Sweden. Figure 2 also clearly showed that the fertility rates actually declined in the post intervention period for African born women. In other words, the outcomes of Africa and Asia contradict the stated hypothesis and theories.
The region EU15 shows a greater increase in odds ratio compared to Sweden. The results are however not significant at the common significance levels. The other regions show odds ratios slightly less than 1, although not significant.
6.3 The logistic model including control variables
The results of the logistic regression model where control variables have been inserted are presented below.
Table 4. The logistic regression model with control variables. N: 1 630 944, missing cases: 348 917. The missing cases derive from difficulties in linking certain unemployment rates to the municipality codes due to changed ditto. I do not believe that the missing cases correlate with fertility rates, wherefore this should not be a problem for the analysis.
Variables
log odds ratio
(B-coefficient) standard error Odds ratio
Africa 0,834 0,030 2,303* North America 0,003 0,082 1,003 South America -0,003 0,050 0,997 Asia 0,292 0,019 1,340* Sweden 1 (ref) Nordic countries 0,071 0,069 1,073 EU15 (non-Nordic) -0,039 0,101 0,962 Europe (non-EU15) -0,070 0,053 0,932 Africa -0,191 0,060 0,826* North America 0,016 0,152 1,017 South America -0,087 0,099 0,917 Diff-in-diff (djt) Asia -0,101 0,034 0,904* 0,000
Sweden not valid
Nordic countries 0,231 0,086 1,260* EU15 (non-Nordic) 0,244 0,108 1,277** Europe (non-EU15) 0,603 0,057 1,828* Africa 0,523 0,066 1,687* North America 0,132 0,171 1,141 South America 0,488 0,126 1,629* Migration effect Asia 0,741 0,037 2,099* 19-24 1 (ref) 25-29 1,039 0,015 2,825* 30-34 0,850 0,015 2,339* 35-39 -0,013 0,016 0,987 40-44 -1,569 0,023 0,208* Age 45-49 -4,327 0,078 0,013* Income (t-1) 0,000 0,000 1,000* Unemployment (t-1) -0,005 0,003 0,995*** Constant -3,089 0,021 0,046* Source: LINDA (SCB)
* = significant at 0,01 level, ** = significant at 0,05 level, *** = significant at 0,10 level
EU15) used the basic level to a large degree, or 46 % of all women. According to the results of the logistic regression model women originating from this region did not respond to the policy change either. The odds ratio is smaller than that for the reference group, although not significant. Contrary to the first model without control variables, women originating from the Nordic countries and North America show a greater increase in probability of having children from pre intervention period to post intervention period, compared to native born women (=odds ratios are greater than 1). The estimates are not significant. The difference-in-differences estimators are, to conclude, on one hand contradictory, but on the other hand they are unison regarding African born and Asian born women.
The migration dummy, taking the value of 1 if the woman has immigrated within two years, measures how very recent immigration affects the probability of having children, compared to women from the same region that have immigrated earlier. All variables show odds ratios greater than 1, and all but the North American migration dummy are significant. It means that recent immigration involves a greater likelihood of having children. The effect is greatest for women emigrating from Asian countries, followed by women emigrating from, in order, European countries other than EU15, Africa, and South America.
The age variable measures how different age classes influence on the probability of having a child, compared to the reference group of women between 19 and 24 years. The results are highly credible. Women between 25 and 34 are the ones most likely to have a child, withholding all other factors. All outcomes but the age class 35 to 39 are significant. The odds ratio of income is 1, implying that one unit’s increase in income does not affect the odds ratio of having a child, in neither direction. It is possible that assuming non-linearity in the variable and introducing income classes would yield other results. One unit’s increase in unemployment rate does however imply a slightly reduced odds ratio of having a child compared to not having one, and is statistically significant at 0,10 level. This outcome is not surprising and has been shown earlier, for instance in the demographic report by SCB investigating how labour market status affects the probability of having children (SCB 2008:1). In this study the authors concluded that employed women had a greater propensity of having a first child, compared to women outside the labour force.
6.3.1 Additional tests
7. Analysis
The underlying basic assumption for my research question is the relationship between costs and childbearing, and the hypothesis that reduced costs should affect fertility (or the probability of having children) positively. The results of the logistic regression models are however contradictory. Some variables have turned out as in prior expected and others have not.
In the logistic regression the outcomes of the difference-in-differences estimators do not confirm the theory of increased incentives for childbearing due to the raise in compensation levels. The odds ratios for those groups that, in aggregate, should be more affected by the policy change were less than 1, instead of greater than 1. This implies that the difference in probability of having children between post and pre intervention period is smaller for the region of interest, compared to the reference group Sweden. This was the case for Africa and Asia. The outcomes were significant. The other groups showed slightly smaller odds ratios than Sweden, but these results were not significant. The only groups having odds ratios greater than 1 were women originating from the Nordic countries and North America. None of these outcomes were however statistically significant, although the Nordic countries interestingly were closest to the 0,10 level (0,22).
There might be several reasons for the unexpected outcomes of the difference-in-differences estimators.
In chapter 5 I discussed the fact that individuals receiving basic level might be entitled to other kinds of state subsidies, such as social assistance and housing allowance. In the earlier referred report by Dahlberg et al (2009), the authors showed that the propensity of receiving social assistance is augmented if the individual is born outside Sweden, or is single. Particularly single mothers have increased propensity of receiving social assistance. Reforms within these state subsidies should ideally be controlled for. A possible situation of non-improved household economy, due to reduced recompenses within the other state subsidies, has however not been taken into consideration in the model due to lacking data of these households. Not considering these households leads to an underestimation of the probabilities of having children for the treatment groups.
that the individuals might not be aware of the compensation raises, and naturally would not alter their behaviour. There is no easy resolution of how this would be considered in the model.
If some time is required for the compensation changes to be generally known data stretching longer than 2005 would be desirable. Partly I have investigated this situation by changing the years linked to the pre and post intervention periods. 2003 was accordingly considered a pre intervention year. In this model the significant odds ratios of less than 1 remained for Africa but not for Asia. The Nordic countries and North America, and also in this model: EU15, had odds ratios greater than Sweden, although not significant. The uncertainty in the results motivates further analysis of possible unconsciousness of the reform among the treatment groups. And likewise include some years after 2005 in the analysis. Another reason for the outcomes of the difference-in-differences estimators might be that the treatment groups actually did not, for other reasons, experience the raised compensation levels as increased incentives. The main purpose of raising the levels was not to increase incentives for childbearing, but to improve the economic conditions of the households.
The model used for answering the research question, difference-in-differences, assumes equal trends in fertility, which is a quite reasonable assumption. The demographic report by SCB indeed showed that the fertility rates of foreign born women, in general, tends to follow the fertility pattern of native born women (SCB 2008:2). Still, a further improvement of the model would be to consider regression discontinuity, allowing different linear trends in fertility between the comparison groups.
8. Conclusions
Initially I repeat the research question:
Does an increase in the compensation level within the basic level cause increased probabilities of having children among individuals that were eligible for the basic level only? Economic theories on childbearing and earlier observed fertility changes due to policy alterations give reasons to believe that raised compensation levels within the basic level would cause increased incentives for childbearing, or precisely, increased probabilities of having children. Hence, this issue has been investigated through the use of a “before and after design and difference-in-differences” method in a logistic regression model, where the difference-in-differences estimator is the causal effect of the reform on childbearing.
The difference-in-differences estimators for the specific treatment groups based on regions of birth did however not turn out as expected. Women that should have been most affected by the policy changes, due to their relatively large use of the basic level, showed odds ratios less than 1 instead of greater than 1. Odds ratios less than 1 imply that the difference in probability of having children between post and pre intervention period is smaller for the region of interest, compared to the reference group Sweden. Thereby the women seem to be unaffected by the policy changes.
Altering the pre intervention and post intervention period to allow a later arisen effect changed the results slightly. Yet, in the modified model it’s not possible to draw certain conclusions regarding increased probabilities of having children.
9. References
Andersson, G., Hoem J. M. and Duvander, A.: 2006, Social differentials in speed-premium effects in childbearing in Sweden. Max Planck Institute for Demographic Research. Demographic research. Vol 14, article 4, pp 51-70.
Arbetsgivarverket: 2007, Långtidssjukskrivnas sannolikhet att övergå till sjukpension eller arbete. En jämförelse av de stora avtalsområdena åren 2000-2004. Rapportserie 2007:1.
Becker, G.: 1991. A treatise on the family. Harvard University Press.
Dahlberg, M., Edmark, K., Hansen J. and Mörk, E.: 2009, Fattigdom i folkhemmet - Från socialbidrag till självförsörjning. Report 2009:4, The Institute for Labour Market Policy Evaluations (IFAU).
Edmark, K.: 2007, Strategic interactions among Swedish Local Governments. Economic Studies 105, Department of Economics, Uppsala University. Universitetstryckeriet, Uppsala.
Försäkringskassan: 2010, Department of Statistics and analysis. Information obtained 2010-04-20. http://statistik.forsakringskassan.se/rfv/html/FP_Tab_1_4_2009.html
Gujarati, D. N. and Porter, D. C.: 2009. Basic Econometrics. Fifth Edition. McGraw-Hill.
Meyer, B.: 1995, Natural and Quasi-Experiments in Economics. Journal of Business & Economic Statistics, Vol. 13, No. 2. pp 151-16.
SCB: 1998, Barnafödande och sysselsatta. Demografiska rapporter 1998:1. SCB-Tryck Örebro.
SCB: 2008, Barnafödande bland inrikes och utrikesfödda. Demografiska rapporter 2008:2. SCB-Tryck Örebro.
SCB: 2010a, Konsumentprisindex. Information obtained 2010-05-20 http://www.scb.se/Pages/TableAndChart____35666.aspx
SCB: 2010b, Press release. Information obtained 2010-04-28 http://www.scb.se/Pages/PressRelease____290189.aspx
SOU 2005:73: Reformerad föräldraförsäkring – Kärlek, omvårdnad, trygghet. Fritzes, Stockholm.
SOU 2006:37. Duvander, A. and Eklund, S.: Utrikesfödda och svenskfödda föräldrars föräldrapenninganvändande. Om välfärdens gränser och det villkorade medborgarskapet. Fritzes, Stockholm.
Swedbank: 2001, Press release. Information obtained 2010-05-01
Appendices
Appendix 1
Table A1.1 Birth regions and number of individuals.
Appendix 2
The following tables show frequencies of newborn and their mothers each year. The tables are divided upon the women’s region of birth. Note that the data for 1998 seems questionable for some regions.
Table A2.1 Sweden.
Newborn Women Newborn/ 1000 women 1993 9 054 127 990 71 1994 8 232 125 989 65 1995 7 378 124 159 59 1996 6 666 122 454 54 1997 6 026 118 516 51 1998 6 209 117 374 53 1999 8 337 137 530 61 2000 8 614 136 042 63 2001 8 436 134 926 63 2002 9 072 134 931 67 2003 9 265 134 208 69 2004 9 427 134 103 70 2005 9 437 133 795 71 Source: LINDA (SCB)
Table A2.2 The Nordic countries except Sweden.
Table A2.3 EU15 except the Nordic countries. Newborn Women Newborn/ 1000 women 1993 74 1408 53 1994 72 1341 54 1995 80 1332 60 1996 52 1280 41 1997 67 1210 55 1998 54 1168 46 1999 90 1466 61 2000 103 1456 71 2001 93 1455 64 2002 89 1508 59 2003 87 1489 58 2004 117 1518 77 2005 104 1513 69 Source: LINDA (SCB)
Table A2.4 European countries except EU15.
Newborn Women Newborn/ 1000 women 1993 171 3 381 51 1994 181 3 888 47 1995 210 3 963 53 1996 191 3 929 49 1997 164 3 823 43 1998 21 532 39 1999 348 6 299 55 2000 313 6 413 49 2001 382 6 546 58 2002 405 6 622 61 2003 393 6 735 58 2004 402 6 851 59 2005 417 6 960 60 Source: LINDA (SCB)
Table A2.5 Africa.
Table A2.6 North America. Newborn Women Newborn/ 1000 women 1993 33 377 88 1994 27 367 74 1995 27 377 72 1996 32 396 81 1997 19 408 47 1998 29 427 68 1999 26 545 48 2000 33 561 59 2001 36 566 64 2002 45 610 74 2003 48 640 75 2004 49 661 74 2005 51 680 75 Source: LINDA (SCB)
Table A2.7 South America.
Newborn Women Newborn/ 1000 women 1993 83 1086 76 1994 84 1110 76 1995 84 1138 74 1996 59 1140 52 1997 48 1097 44 1998 76 1072 71 1999 78 1409 55 2000 81 1441 56 2001 85 1488 57 2002 100 1544 65 2003 109 1565 70 2004 121 1587 76 2005 106 1581 67 Source: LINDA (SCB)
Table A2.8 Asia.
Appendix 3
Table A3.1 Frequencies
Sweden
Nordic
countries EU15 Europe Africa
North America
South
America Asia Total
Appendix 4
Table A4.1 Results of the logistic regression where the post intervention period has been specified as 2004 and 2005. n: 1 630 944, missing cases: 348 917
Variables
log odds ratio
(B-coefficient) standard error Odds ratio
1993 1 (ref) 1994 -0,045 0,022 0,956** 1995 -0,240 0,022 0,787* 1996 -0,349 0,022 0,706* 1997 -0,422 0,022 0,656* 1998 -0,389 0,021 0,678* 1999 -0,181 0,019 0,834* 2000 -0,172 0,018 0,842* 2001 -0,173 0,018 0,841* 2002 -0,112 0,018 0,894* 2003 -0,088 0,017 0,915* 2004 -0,032 0,017 0,968*** (dt) 2005 Sweden 1 (ref) Nordic countries -0,051 0,029 0,950*** EU15 (non-Nordic) 0,098 0,052 1,103** Europe (non-EU15) -0,134 0,031 0,874* Africa 0,808 0,032 2,244* North America 0,000 0,088 1 South America -0,024 0,053 0,997 (dj) Asia 0,267 0,020 1,306* Sweden 1 (ref) Nordic countries 0,009 0,058 1,009 EU15 (non-Nordic) 0,002 0,089 1,002 Europe (non-EU15) -0,053 0,047 0,948 Africa -0,169 0,053 0,845* North America 0,038 0,137 1,039 South America -0,035 0,089 0,966 Diff-in-diff (djt) Asia -0,045 0,031 0,956
Sweden not valid
19-24 1 (ref) 25-29 1,039 0,015 2,825* 30-34 0,850 0,015 2,339* 35-39 -0,013 0,016 0,987 40-44 -1,569 0,023 0,208* Age 45-49 -4,327 0,078 0,013* Income (t-1) 0,000 0,000 1,000* Unemployment (t-1) -0,005 0,003 0,995*** Constant -3,089 0,021 0,045* Source: LINDA (SCB)
* = significant at 0,01 level, ** = significant at 0,05 level, *** = significant at 0,10 level
