Open Access

Does the IAB employment sample reliably identify maternity leave taking? A data report

Zeitschrift für ArbeitsmarktForschungJournal for Labour Market Research200942:11

https://doi.org/10.1007/s12651-009-0011-0

Accepted: 20 October 2008

Published: 3 April 2009

Abstract

The data set that researchers have used most often to study career interruptions due to childbirth in the German context is the German Socio-Economic Panel (GSOEP). An alternative data source is the much larger IAB Employment Sample (IABS). Although this data set does not include direct information on childbirth, mothers on maternity leave can potentially be identified. There are, however, two problems. First, the leave variable in the IABS does not distinguish between maternity leave and other leave taking, such as sick leave. Second, the child's birth month has to be inferred from the month in which the mother goes on maternity leave, which is likely to lead to measurement error in the time that the mother spends at home after childbirth.

This paper investigates both problems, using an extended version of the IABS that supplements the social security records with direct information on childbirth from the German Pension Register. I find that for Western West German citizens, at least 90% of leave spells are due to maternity leave. The child's birth month is correctly estimated for at least 70%, and over- or underestimated by one month for about 25% of mothers.

I conclude that the most recent scientific use files of the IABS, the IABS 75-01 and IABS 75-04, provide a very valuable alternative data source to the GSOEP to study career interruptions due to childbirth, as long as the focus is on women who are attached to the labour market.

Kann die IAB-Beschäftigtenstichprobe benutzt werden, um Erziehungsurlaub verlässlich zu identifizieren? Ein Daten-Report

Zusammenfassung

Der Datensatz, der in Deutschland am häufigsten benutzt wurde, um Erwerbsunterbrechungen von jungen Müttern zu untersuchen, ist das sozio-ökonomische Panel. Ein alternativer Datensatz ist die wesentlich größere IAB-Beschäftigtenstichprobe (IABS). Dieser Datensatz enthält zwar keine direkten Informationen über das Geburtsdatum von Kindern. Mütter im Erziehungsurlaub können jedoch über Erwerbsunterbrechungen identifiziert werden. Hier gibt es jedoch zwei Probleme. Erstens, die Erwerbsunterbrechungsvariable in der IABS unterscheidet nicht zwischen einer Unterbrechung aufgrund von Erziehungsurlaub und einer Unterbrechung von z. B. Krankheit. Zweitens, der Geburtsmonat des Kindes muss vom Monat, in dem die Mutter in den Erziehungsurlaub geht, abgeleitet werden. Dies führt wahrscheinlich zu einem Messfehler in der Dauer der Erwerbsunterbrechung.

Dieser Datenreport untersucht beide Probleme basierend auf einer erweiterten Version der IABS, die zusätzlich zu den Sozialversicherungsangaben der IABS direkte Informationen über das Geburtsdatum der Kinder enthält. Diese Information stammt aus den Daten der Rentenversicherung. Meine Ergebnisse für westdeutsche Frauen zeigen, dass mindestens 90% der Erwerbsunterbrechungen in der IABS Unterbrechungen aufgrund von Erziehungsurlaub sind. Außerdem wird für mindestens 70% der Mütter der Geburtsmonat des Kindes in der IABS korrekt gemessen. Für weitere 25% wird der Geburtsmonat um einen Monat unter- oder überschätzt. Mein Fazit ist, dass die neueren Scientific Usefiles der IABS, die IABS 75-01 und die IABS 75-04, eine sehr wertvolle alternative Datenquelle zum sozio-ökonomischen Panel darstellen, um Erwerbsunterbrechungen aufgrund von Erziehungsurlaub zu studieren. Allerdings muss berücksichtigt werden, dass in der IABS nur Mütter im Erziehungsurlaub, und nicht generell die Geburt eines Kindes, beobachtet wird.

1 Introduction

Researchers have long been interested in questions like: when do mothers return to work after childbirth? What is the impact of career interruptions due to childbirth on subsequent wage growth? How does parental leave legislation affect the labour supply and wages of women? These questions are highly relevant in light of the recent developments in family leave policies around the world. For instance, Germany has recently increased maternity benefits (Elterngeld) after childbirth to 67% of the net pre-birth income during the child's first year (BMFSFJ 2007). Other countries that have recently expanded maternity leave coverage include Canada (2003) and the UK (2003, 2007). The data set that has been most commonly used to address these questions in the German context is the German Socio-Economic Panel (e.g. Weber 2004; Görlich and De Grip 2007; Vlasbloom and Schippers 2003). In addition to detailed information on fertility and employment, the GSOEP contains a large array of background characteristics, such as marital status and husband's income. It suffers, however, from a small sample size.

An alternative data set that has been used to address similar questions is the IAB Employment Sample (e.g. Schönberg and Ludsteck 2008; Ejrnæs and Kunze 2006). The two most recent scientific use files are the IABS 75-01 and IABS 75-04.1 The IABS has several advantages over the GSOEP, the most important of which is its much larger sample size. For instance, Vlasbloom and Schippers (2003) identify 649 mothers (on maternity leave) in the GSOEP, using data from 1984 to 2000. In contrast, I was able to identify 47,703 mothers on maternity leave between 1984 and 1993 in the IABS 75-01. A further advantage of the IABS Employment Sample is that information on the employment history and wages is measured more precisely than in the GSOEP.2

One disadvantage of the IABS, compared to the GSOEP, is that the IABS does not contain direct information on childbirth. The data set does, however, include a variable that indicates an interruption of the employment relationship, so women who go on maternity leave can potentially be identified. This variable has two limitations. First, not all leave spells may be due to maternity leave. Alternative reasons include sick or disability leave. Second, since the IABS does not include direct information on childbirth, the month in which the child was born has to be inferred from the month in which the mother goes on leave. This is likely to lead to measurement error in the child's birth month and therefore in the time that mothers spend at home after childbirth. Both types of measurement error may bias findings regarding the determinants of when women return to work after childbirth, or regarding the impact of the duration of the career interruption on subsequent wage growth.

This paper investigates whether the leave variable in the most recent scientific use files of the IAB Employment Sample, the IABS 75-01 and IABS 75-04, can be reliably used to identify maternity leave. This is made possible by an extended, weakly anonymous version of the IABS 75-95. The IABS 75-95 is an older version of the IABS 75-04 that includes a 1% random sample of men and women covered by the social security system. I refer to the extended version of this data set as the IABS 75-95 Plus. This data set supplements the social security records from the IABS with information on activities during employment gaps from the German Pension Register. In particular, since 1986 the extended version has included precise information on when a woman gave birth.

I use this data set to address four questions. First, I analyze how many and which mothers go on maternity leave. Between 1987 and 1994, about 50% of mothers in Western Germany and 59% in Eastern Germany took maternity leave. The share is likely to be considerably larger for first-time mothers. Not surprisingly, taking maternity leave is substantially more common for mothers who were employed around conception, i.e. around nine months prior to childbirth (around 90%). This illustrates that the IABS cannot be used to identify childbirth in general. However, the IABS is useful if the research focus is on women who are attached to the labour market.

Second, I analyze how many leave spells in the social security data can be linked to childbirth in the Pension Register. I show that for Western German women this is the case for at least 90% of the leave spells, but only after some sample restrictions have been imposed.

Third, I analyze how the child's month of birth that is inferred from the start of the leave spell in the social security data differs from the true month of birth in the Pension Register. I find that in most cases (70%), the two coincide. In 25% of the cases, the inferred birth month is over- or underestimated by one month. Measurement error in the birth month leads to measurement error in the time which the mother spends at home before returning to work. This may be a problem for instance if one wants to evaluate the impact of maternity leave legislation on women's decisions to return to work. One may expect that an unusually large share of women return to work after exactly 36 months if the job-protection period is 36 months. Due to the particular type of measurement error in the child's birth month, however, one would also expect an especially large share 35 or 37 months after childbirth. I confirm this in Sect. 5.2.

Fourth, I directly investigate the biases that may arise due to the two types of measurement error in the IABS. I focus on two issues: women's decisions as to when to return to work after childbirth, and the impact of the duration of the career interruption on subsequent wages. I first present results using only information from the social security data that is available in the scientific use files of the IABS 75-01 and IABS 75-04. I then report the “true” results, based on the information from the Pension Register in the IABS 75-95 Plus. Overall, the IABS and the Pension Register yield very similar findings. However, the IABS slightly underestimates the impact of education and age on the returning hazard. Probably most importantly, the IABS somewhat overestimates the cost of career interruptions.

I conclude that the most recent scientific use files of the IABS, the IABS 75-01 and IABS 75-04, provide a very valuable alternative data source to the GSOEP for studying career interruptions due to childbirth, as long as the focus is on women who are attached to the labour market.

The plan of this paper is as follows. I begin with a brief description of the expansions in parental leave coverage that have taken place in Germany since the late 1970s (Sect. 2). In Sect. 3, I describe the IABS 75-95 Plus, which is used to investigate the reliability of the leave variable in the scientific use file. Section 4 presents evidence on how many and which mothers take maternity leave. In Sect. 5, I address two sources of measurement error in the scientific use file: I first analyze how many leave spells are due to childbirth. I then turn to measurement error in the child's birth month. In Sect. 6, I investigate how a noisy measure of maternity leave in the scientific use file may bias findings regarding women's decision to stay at home, and regarding the impact of career interruptions due to childbirth on subsequent wages. I conclude in Sect. 7.

2 Background: maternity leave legislation in Germany

In this section, I briefly describe the main features of maternity leave legislation in Germany. A more detailed description can be found for instance in Schönberg and Ludsteck (2008) or Kreyenfeld (2001). Since 1968 mothers have been entitled to paid maternity leave six weeks before and eight weeks after the birth of a child. During this `maternity protection' period, the firm is not allowed to dismiss the mother, and the mother has the right to return to a job that is comparable to the job she held before the birth.

Since the late 1970s, there have been several expansions in leave coverage. Figure 1, taken from Schönberg and Ludsteck (2008), provides a visual overview of the reforms. The first reform took place in May 1979. It increased the job-protected maternity leave period from two to six months. This reform also turned the right to a leave of absence during the first eight weeks following childbirth into an employment ban during this period. During the first two months following childbirth, mothers received their full salary, while payment between the third and sixth month following childbirth was roughly equal to 375 Euros per month (Zmarzlik et al. 1999). This corresponds to about one third of women's average pre-birth earnings.3 Only women who were employed before the birth were entitled to maternity benefits.
Fig. 1

Maternity Leave Legislation in Germany (Selected Reforms)

In January 1986, the job-protection period was increased from six to ten months and a further increase to 12 months starting in January 1988 was announced. An important component of this reform was that fathers became eligible to take paternity leave. However, the proportion of fathers taking parental leave is very small; in 2001 it was 1.6% (Engstler and Menning 2003). A further component of this reform was that all mothers, regardless of their employment status prior to childbirth, became eligible for maternity benefits. During the six weeks prior to and eight weeks following childbirth, maternity benefit remained at the level of the mother's pre-birth earnings (or 300 Euros if the mother was not employed before the birth). Until December 1993, maternity benefits were equal to 300 Euros from the third to the sixth month after childbirth, irrespective of the mother's (or the father's) income prior to the birth.4 This corresponds to about 20% of women's average pre-birth earnings. 3 From the seventh month onwards, maternity benefits were means-tested and depended on the annual net family income two years before the birth of the child. The majority of women received benefits longer than six months; in 1986, for instance, this proportion was 83.6% (Engstler and Menning 2003, BMFSFJ 2000).

In July 1989 and July 1990, the job-protected period of leave was further raised to 15 and 18 months respectively. In January 1992, the job-protected period of leave was increased from 18 to 36 months. Maternity benefit payments still ended at 18 months, but were to be extended to 24 months one year later. The most recent policy reform took place in January 2007. This reform increased maternity benefit to 67% of the net pre-birth income during the child's first year. If the father takes parental leave as well, benefits are paid for two additional months (BMFSFJ 2007b).

Several states, including Bavaria, Baden-Wuerttemberg, Saxony, and Thuringia, pay maternity benefits in addition to the federal benefits. For instance, since 1986, Baden-Wuerttemberg has paid 200 Euros per month for an additional 12 months, once the federal benefit has expired. Since July 1989, Bavaria has paid 250 Euros per month up until the child's second birthday, also starting with the expiration of the federal maternity benefit. Similar rules have existed in Saxony and Thuringia since 1992.

3 Data description and sample selection

The analysis in this report is based on an extended, weakly anonymous version of the IABS 75-95 (Bender et al. 2000). The weakly anonymous version of the IABS 75-95 differs from the scientific use files of the IAB Employment Samples in that very few steps have been undertaken to anonymize the data. Although the findings in this paper are based on the original social security data, they apply to the most recent scientific use files of the IAB Employment Sample, the IABS 75-01 and IABS 75-04. However, I caution against using the scientific use files of the IABS 75-95 and IABS 75-97 to identify maternity leave spells if the focus of the research is to analyze the impact of maternity leave policies on mothers' labour market outcomes. Please see Appendix B for details.

The IABS 75-95 is a 1% random sample of social security records, available for the years 1975 (1992 for Eastern Germany) to 1995. The data set includes all men and women who, during this period, held at least one job for which social security contributions had to be paid.5 In addition to a wide variety of background characteristics, such as age, education, industry and occupation, the data set includes precise information on wages, on when the individual switched employers, when he or she entered and left unemployment, and when he or she interrupted their current employment relationship. This information is reported by firms, and mis-reporting is subject to penalties. The data set does not contain information on the individual's activities during employment gaps. To mention only a few possibilities, a woman may be self-employed, retired, or taking care of her children.

The IABS 75-95 is extended in two ways. First, it is supplemented by information on activities during employment gaps from the German Pension Register. A detailed description of this data set with one extension and of the Pension Register can be found for instance in Wübbeke (2005a, b) or Prinz (1997).6 The Pension Register includes information on career interruptions if the activities during the career interruption entitle the individual to a pension. This is currently the case for employment gaps due to military service, full-time education, sick leave, disability, child care, and the activity variable in the Pension Register therefore distinguishes between these activities. More specifically, with regard to women the extended IABS 75-95 contains precise information on their children's dates of birth. Unfortunately, prior to 1986, data on fertility is incomplete. This is because child care constitutes a pension claim only for children born after December 31 1985. Before 1986, women could voluntarily report the birth of their child to the Pension Register, while after 1986 the registration offices (Einwohnermeldeämter) automatically forward this information to the Pension Register. The Pension Register does not contain (direct) information on whether the mother is on leave from her employer.

Is the information on childbirth in the Pension Register (after 1986) complete? In 1986, the Pension Register recorded 4,508 births to (West) German citizens. In that year, 567,310 children were born in West Germany to German citizens.7 Hence, the extrapolated number of births in the Pension Register is about 20% (4,508,000 versus 567,310) lower than the actual number of births. This is not surprising, as the IABS 75-95 only includes women who were employed and paid social security contributions at least once between 1975 and 1995. For instance children of civil servants (e.g. teachers) or self-employed mothers who were never in employment covered by social security are not recorded. Overall, the IABS Employment Samples cover approximately 77% of the workforce (Bundesagentur für Arbeit 2004), suggesting that for those women who were in employment covered by social security at least once, the information about childbirth in the Pension Register is (virtually) complete.

Unfortunately, the registration offices do not report information on the order of births. Hence, the first child observed in the data may in fact be the second or third child.

The second addition to the IABS 75-95 is that, for all women and men included in the IABS 75-95, social security records are extended to the year 2003. Hence, the extended version of the IABS 75-95 allows researchers to observe men's and women's employment and unemployment histories over the period from 1975 to 2003. Information on employment gaps from the German Pension Register, however, is available only up to 1995. I refer to the IABS 75-95 with two extensions as the IABS 75-95 Plus.8

The IABS contains a variable that was first used by Ejrnæs and Kunze (2006) to identify whether a woman has given birth and took maternity leave. Firms in Germany are required to report when a mother is on maternity leave because mothers are not allowed to work the first two months after childbirth. The leave variable (btyp) is created by the IAB based on information reported by the employer as to why an employment relationship was interrupted. The variable distinguishes between five values (Table 1), in addition to regular employment spells (btyp = 1) and unemployment spells (btyp = 7). The btyp variable is equal to 2 if the employer reports the employment relationship as interrupted, and the employment relationship stops on December 31, and is continued on January 1st the following year. The btyp variable takes the value 3 if employers report a wage equal to zero, but do not report an interruption of the employment relationship. The btyp variable is equal to 4 if the employer reports the relationship as interrupted and the employee continues to work with this employer, while it is equal to 5 if the employee returns to work with a different employer. Finally, the btyp variable takes the value 6 if the employee never returns to the labour market.
Table 1

The btyp variable in the social security data

1

Regular employment spell

2

Employer reports an interruption of the employment relationship. Employment ends on December 31, and starts January 1 the following year.

3

Employer does not report an interruption of the employment relationship. The wage reported is 0.

4

Employer reports an interruption of the employment relationship. The individual returns to the same employer.

5

Employer reports an interruption of the employment relationship. The individual returns to a different employer.

6

Employer reports an interruption of the employment relationship. The individual does not return to the labour market.

7

Receipt of unemployment benefits or other transfer payments.

Note: The table lists the definition of the btyp variable in the IABS that is used to identify career interruptions due to childbirth.

Three problems may arise when using this variable to identify childbirth. First, there may be women who give birth but do not go on maternity leave. Second, not all leave spells in the IABS are due to childbirth. Other reasons why a woman may take a leave of absence from her employer include sick or disability leave. Third, since the IABS does not contain direct information on children's dates of birth, the month of the child's birth has to be inferred from the month in which the mother goes on maternity leave. This is likely to lead to measurement error in the month of birth, and therefore in the time that the mother stays at home before returning to work.

The extended version of the IABS 75-95 allows me to address each of these problems. I construct three samples to do so. Sample A consists of all women who gave birth between January 1987 and December 1994.9 Here, the information on childbirth comes from the Pension Register, while the information on leave comes from the social security data of the IABS 75-95. I use this sample to analyze how many women who give birth take maternity leave.

Sample B consists of all women with at least one leave spell between January 1987 and December 1994 in the social security data of the IABS 75-95 Plus. I use this sample to analyze how many leave spells in the social security data are due to childbirth, rather than due to sick leave etc.

Sample C includes all women who gave birth between January 1987 and December 1994 according to the Pension Register and went on maternity leave according to the social security data. I use this sample to analyze the relationship between the month in which a woman goes on maternity leave and the month when she gives birth.

For Eastern German mothers, the sample is restricted to the years 1993 and 1994. A detailed description of the variables can be found in Appendix A.

4 How many mothers take maternity leave?

One shortcoming of the IABS is that although it allows researchers to identify whether a woman has taken (maternity) leave, it does not identify childbirth in general. For instance, if a woman terminates her job as soon as she finds out that she is pregnant and does not take a leave of absence from her employer, one would observe an employment gap in the IABS. Similarly, if a woman goes on maternity leave for her first child, and has her second child while still on leave (which is not uncommon, due to the long job-protection period in Germany), there will be no record of the birth of the second child. In this section, I analyze how many mothers go on maternity leave in Germany. The findings are based on Sample A.

The results can be found in Table 2. Columns 1 to 3 report the share of all women who take maternity leave, while columns 4 to 6 refer to women who were employed around conception, nine months prior to childbirth. The table distinguishes between Western and Eastern Germans, as well as between women of German and foreign nationality. Due to the small number of observations, I have dropped women of foreign nationality in Eastern Germany from the sample. For women in Western Germany, the share of women who take maternity leave increased from 47.8% in 1987 to 52.0% in 1994. Women of foreign nationality are somewhat less likely to go on maternity leave (about 47%), whereas in Eastern Germany taking maternity leave is more common (about 59%). These numbers may seem low. Note, however, that they refer to all births, and I will provide evidence in Table 3 that taking maternity leave is likely to be considerably more common for the first birth. It is important to bear in mind that if a mother has a second child while still on maternity leave for the first child, or while working in a marginal part-time job that is exempt from social security contributions, the IABS does not record a leave of absence for the second birth.
Table 2

How many women take maternity leave?

 

All

Working 9 months prior to childbirth

 

1

2

3

4

5

6

 

West, German

West, Foreign

East

West, German

West, Foreign

East

1987

47.82%

48.74%

 

88.94%

90.08%

 
 

4,724

238

 

(50.55%)

(50.84%)

 

1988

49.54%

45.42%

 

88.53%

87.50%

 
 

5,055

251

 

(52.42%)

(44.80%)

 

1989

49.73%

43.51%

 

89.54%

83.69%

 
 

4,977

285

 

(52.62%)

(49.65%)

 

1990

51.31%

45.85%

 

89.90%

84.78%

 
 

4,968

277

 

(53.20%)

(50.00%)

 

1991

54.21%

43.49%

 

89.68%

80.45%

 
 

4,874

269

 

(55.46%)

(49.81%)

 

1992

54.64%

49.25%

 

89.02%

83.69%

 
 

4,989

266

 

(55.68%)

(53.00%)

 

1993

55.04%

48.45%

58.11%

90.40%

78.89%

81.10%

 

4,809

322

518

(54.81%)

(56.07%)

(63.32%)

1994

52.01%

49.65%

59.44%

90.38%

86.81%

79.82%

 

4,620

286

503

(54.46%)

(51.06%)

(65.81%)

Total

51.79%

46.78%

58.74%

89.95%

83.52%

80.46%

 

39,016

2,194

1,021

(53.45%)

(51.43%)

(64.55%)

Note: The table reports the share of mothers who take maternity leave. Columns 1 to 3 refer to all mothers, and columns 4 to 9 refer to mothers who were employed 9 months prior to childbirth. Here, the share in parentheses displays the share of mothers who were employed 9 months prior to childbirth. The findings are based on Sample A.

Table 3

The determinants of taking maternity leave (dependent variable: 1 if mother takes maternity leave)

 

All

West, Germans

West, Foreigners

East

 

1

2

3

4

 

42,706

39,015

2,180

1,021

East

0.031

   
 

(0.014)

   

Foreign

0.022

   
 

(0.017)

   

Age

0.044

0.103

0.058

0.169

 

(0.006)

(0.006)

(0.019)

(0.036)

Age2

−0.001

–0.002

−0.001

–0.003

 

(0.000)

(0.000)

0.000

(0.001)

Medium-skilled

0.154

0.167

0.062

0.240

 

(0.007)

(0.008)

(0.023)

(0.045)

High-skilled

0.080

0.027

–0.008

0.231

 

(0.012)

(0.014)

(0.062)

(0.055)

2nd child

 

–0.368

–0.167

 
  

(0.005)

(0.025)

 

3rd child

 

–0.456

–0.198

 
  

(0.006)

(0.037)

 

4th or further child

 

–0.486

–0.194

 
  

(0.006)

(0.069)

 

Note: The table reports results from linear probability models where the dependent variable is equal to 1 if the mother takes maternity leave. The findings are based on Sample A. Robust standard errors in parentheses.

Not surprisingly, substantially more women take maternity leave if they were employed nine months prior to childbirth. For women in Western Germany, the share is about 90% and has remained roughly constant over time. The share is somewhat lower for women of foreign nationality and in particular for women in Eastern Germany. Table 2 also provides information on how many women are employed around conception (small number in parentheses). This is the case for about 50% of mothers in Western Germany, regardless of nationality, and about 65% of mothers in Eastern Germany.

It may seem surprising that some women who were attached to the labour market prior to giving birth do not use the option to take maternity leave, especially since taking maternity leave does not imply any obligations on the part of the mother. In particular, women on maternity leave are not required to return to their previous employer. Note, however, that since 1986 all mothers in Germany have been entitled to maternity benefit even if they were not employed prior to childbirth. Moreover, pregnant women are eligible for unemployment benefits. Hence, terminating employment soon after conception may be optimal for mothers who do not expect to return to the labour market (and to their current employer in particular) in the near future. In my sample, about one third of women who were working 9 months prior to childbirth but do not take maternity leave are observed to claim unemployment benefit prior to childbirth, and do not return to the labour market for at least six years.

I provide more information about the determinants of taking maternity leave in Table 3. The first column pools women (mothers?) in Eastern and Western Germany. The remaining columns report results separately for women of German and foreign nationality in Western Germany and women in Eastern Germany. In addition to the variables reported, I control for the year in which the woman gives birth. Eastern German women are more likely to go on maternity leave both conditional and unconditional on pre-birth characteristics, such as education and age. This is not the case for women of foreign nationality: although overall they are less likely to take maternity leave than German nationals (Table 2), the sign reverses if one conditions on pre-birth characteristics. For both foreign and German nationals and women in Eastern and Western Germany, older mothers are more likely to go on maternity leave. With the exception of Eastern German women, the relationship between taking maternity leave and education is non-monotone, and the medium-skilled who completed an apprenticeship are most likely to take maternity leave.

For women in Western Germany, I also include indicators for birth order as additional regressors. I do not do this for Eastern German women because here fertility data is only available for the years 1992 to 1995. Note that these estimates are likely to present a lower bound for the true impact of birth order on the taking of maternity leave, since fertility data is incomplete before 1986. Nonetheless, there is a strong negative relationship between birth order and the taking of maternity leave: among German nationals, women are 36.4% less likely to take maternity leave for their second than for their first (recorded) child, and 45.0% less likely for their third child. The pattern is the same for foreign nationals, although the impact is smaller in magnitude. This suggests that the share of first-time mothers who take maternity leave is considerably larger than the overall share of 50% reported in Table 2.

5 Leave spells in the IABS and maternity leave

In this section, I address the two sources of measurement error in the social security data: first, not all leave spells may be due to childbirth; and second, the child's birth month is likely to be measured with error.

5.1 How many leave spells in the IABS are due to maternity leave?

One problem of the social security data is that the leave variable does not distinguish between alternative reasons for taking leave. While maternity leave (for women of child-bearing age) is likely to be the most common reason, other reasons include illness, disability and military service. Next, I use the IABS 75-95 Plus to analyze how many leave spells in the IABS are due to maternity leave.

Table 4 displays the share of leave spells that are due to childbirth. The findings are based on Sample B. I report results separately for women of German and foreign nationality in Western Germany, as well as women in Eastern Germany. For women in Western Germany, I further distinguish between all leave spells and the first leave spell observed in the data.10 Researchers may be interested in the latter restriction if they would like to study the return to work after the birth of the first child. It is important to bear in mind, however, that the first leave spell in the social security data is only a proxy for the first birth.
Table 4

How many leave spells are due to maternity leave?

 

West, Germans

West, Germans, 1st spell

West, Foreigners

West, Foreigners, 1st spell

East

No restriction

55.40%

59.71%

44.80%

54.93%

36.12%

N

38,984

25,243

3,286

1,804

1,672

Age

Between 18 and 40

74.36%

77.55%

64.92%

72.30%

58.35%

N

28,752

19,280

2,218

1,350

1.018

True leave spells deleted

2.12%

2.03%

3.00%

3.30%

1.53%

Between 18 and 35

77.61%

79.93%

71.81%

76.51%

66.36%

N

26,072

18,026

1,820

1,192

868

True leave spells deleted

10.56%

9.21%

11.26%

12.91%

3.48%

Plus duration of leave spell

>2 months

84.04%

85.55%

75.62%

81.48%

68.82%

N

23,929

16,184

1,735

1,096

773

True leave spells deleted

11.53%

13.56%

10.32%

13.84%

7.01%

>3 months

84.80%

87.01%

77.81%

83.39%

70.98%

N

21,867

14,876

1,523

963

696

True leave spells deleted

17.84%

20.54%

16.28%

22.35%

11.27%

Plus leave spell not equal to 1st

88.74%

90.68%

80.81%

86.85%

78.23%

N

20,879

14,275

1,506

958

620

True leave spells deleted

16.96%

19.40%

14.33%

18.79%

11.31%

Plus spell not preceded by app.

89.19%

91.35%

80.85%

87.06%

78.20%

N

20,308

13,752

1,483

935

601

True leave spells deleted

18.65%

21.84%

15.14%

20.37%

12.51%

Note: The table reports the share of leave spells in the IABS that can be linked to childbirth in the Pension Register, after imposing more and more restrictions. After deleting women younger than 18 and older than 40, spells that are shorter than 2 months, spells that start on the first of a month, and spells that are preceded by apprenticeship training, 89.19% of all leave spells of western German women in the IABS are due to childbirth. For women of foreign nationality and women in eastern Germany, the shares are 80.69% and 78.20% respectively. The findings are based on Sample B.

When I impose no restrictions, only 55.37% of all leave spells and 58.55% of first leave spells for Western German women are due to childbirth. For women of foreign nationality and women in Eastern Germany, the share is even smaller. The share increases by about 20 percentage points if I restrict the sample to women of child-bearing age, between 18 and 40. Of the spells deleted due to this restriction, about 2% are due to childbirth. The share of “correct” leave spells increases if the age restriction is made more stringent. For instance, when I delete women older than 35 from the sample, 71.63% of leave spells of women of foreign nationality are due to childbirth, compared to 64.80% when I delete women older than 40. However, the more stringent age restriction also increases the share of leave spells that are due to childbirth but are erroneously deleted from the sample, from about 2% to about 10%. The remainder of this paper restricts the sample to women between 18 and 40. Depending on the research question, other researchers may prefer more stringent restrictions.

In Germany, mothers are not allowed to work eight weeks after childbirth, and may go on maternity leave six weeks before the baby is due. A second sensible restriction therefore is to delete “short” leave spells. When leave spells shorter than or equal to two months are deleted in addition to women younger than 18 and older than 40, the share of leave spells that are due to childbirth increases by roughly ten percentage points. However, note that the share of correct births that are wrongly deleted from the sample increases from about 2% to about 11%, suggesting that some maternity leave spells are shorter than two months, despite the employment ban during the first eight weeks after childbirth. Restricting the sample to spells longer than or equal to three months further increases the share of leave spells due to childbirth, but only slightly. This more stringent restriction also raises the probability of a true maternity leave spell being deleted from the sample from about 11% to about 18%. Throughout the remainder of this paper, I restrict the sample to spells longer than or equal to two months.

After these restrictions have been imposed, the share of “wrong” leave spells is considerably larger if the leave spell started in January than in any other month. This is mostly, but not entirely, due to leave spells for which the btyp variable takes the value 2, i.e. employment relationships which the employer reports as interrupted, and which end on December 31 and continue on January 1st the following year. This suggests that most of the spells for which the btyp variable is equal to 2 are not due to maternity leave. In addition, the share of leave spells that are not due to maternity leave is somewhat larger if the spell starts on the first of a month.

The next row of Table 4 reports the share of true leave spells after spells that start on the first of a month have been deleted. The results are similar if I delete leave spells where the btyp variable is equal to 2 instead. This restriction increases the share of leave spells due to childbirth by about five percentage points for all groups. However, it also increases the share of erroneously deleted true leave spells by about five percentage points.

As a final restriction, I delete leave spells that are preceded by a spell in apprenticeship training. The reason for this is the different maternity leave legislation for regular employees and apprentices. The final share of leave spells in the IABS that can be linked to childbirth in the Pension Register is 89.19% for Western German women for all leave spells, and 91.36% for first leave spells. The shares are up to ten percentage points smaller for women of foreign nationality and for Eastern German women.

Note that these findings treat the information in the Pension Register as the true data, and the social security data as measured with error. If the information on children's dates of birth in the Pension Register is incomplete or measured with error, then a leave spell in the social security data of the IABS could be due to childbirth, although there is no record of childbirth around the start of the leave spell in the Pension Register. The shares in Table 4 are therefore best interpreted as lower bounds for the true shares.

What explains the “wrong” leave spells in the IABS? The first row of Table 5 reports the share of erroneous leave spells for which we observe in the Pension Register an activity other than maternity leave during employment gaps. The results are based on Sample B, but restricted to leave spells that are not due to childbirth. When I impose no additional restrictions, 49.35% of leave spells for Western German women can be linked to an activity in the Pension Register. The share increases to 55.38% when I impose my preferred restrictions, i.e. when I restrict the sample to women between 18 and 40, to spells longer than two months, to spells that do not start on the first of a month, and to spells that are not preceded by a spell in apprenticeship training. By far the most common activity is sick leave (about 40%), followed by disability leave (about 9%). These figures imply that in about 5% of leave spells in the IABS (i.e. 0.1 × 0.5)11, the social security data in the IABS and the Pension Register provide inconsistent information. This may be the case either because the information in the Pension Register is incomplete, or because the IABS data contains employment interruptions that are in fact permanent separations.
Table 5

Wrong leave spells

 

West, Germans

West, Foreigners

East

No restrictions

49.43%

50.00%

45.22%

N

17,387

1,814

1,068

Preferred restrictions

55.40%

52.46%

56.80%

N

2,195

284

131

Note: The table restricts the sample to leave spells in the IABS that are not due to childbirth, and reports the share of spells that can be linked to an activity other than childbirth (such as sick leave) in the Pension Register. The first row refers to all leave spells that are not due to childbirth. The second row imposes the preferred restrictions, i.e. women younger than 18 and older than 40, spells shorter than 2 months, spells that start on the first day of a month, and spells that are preceded by a spell of unemployment are deleted from the sample. The third row displays the share for spells that start on the first of a month.

5.2 Measurement error in the month of birth

A third shortcoming of the social security data is that the child's birth month has to be inferred from the month in which the mother goes on maternity leave. This is likely to lead to measurement error in the time which mothers spend at home after childbirth. Next, I provide evidence of this type of measurement error. The findings are based on Sample C. I additionally impose my preferred restrictions; i.e. I restrict the sample to women between 18 and 40, to spells longer than two months, to spells that do not start on the first of a month, and to spells that are not preceded by a spell in apprenticeship training.

Since in Germany mothers are allowed to go on maternity leave six weeks before the expected birth date, I approximate the child's birth month as six weeks after the mother went on leave. Table 6 reports the share of births where the birth month imputed from the IABS coincides with that observed in the Pension Register, or occurs one or two months before or after. In 69.28% of all leave spells of Western German women (column 1), the IABS measures the birth month correctly. In about 12%, I either over- or underestimate the true birth month by one month. The table also reveals that I am more likely to underestimate the birth month in the IABS than to overestimate it (17.7% versus 13.0%). This is not surprising, as women who are sick during pregnancy are likely to go on leave earlier. The shares are similar in Eastern Germany (column 5), or when I consider the first spell only (column 2). However, in the IABS measurement error in the month of birth is somewhat larger for women of foreign nationality (columns 3 and 4).
Table 6

Measurement error in the birth month variable (birth month in the pension register minus imputed birth month in the IABS)

 

West, Germans

West, Germans, 1st spell

West, Foreigners

West Foreigners, 1st spell

East

 

1

2

3

4

5

<−1

0.63%

0.71%

0.48%

0.54%

0.75%

−1

12.37%

12.41%

10.53%

12.27%

10.83%

0

69.28%

68.94%

61.62%

61.73%

70.28%

1

12.08%

12.55%

12.71%

12.64%

10.83%

2

2.12%

1.98%

4.24%

3.97%

1.26%

>2

3.51%

3.40%

10.41%

8.84%

6.05%

N

16,532

11,454

826

554

397

Note: The table reports the difference between the birth month observed in the Pension Register and that observed in the IABS, based on the start of the maternity leave spell. The findings are based on Sample C.

I would again like to stress that these findings treat the information in the Pension Register as the true data, and the social security data as measured with error. If the information on children's dates of birth in the Pension Register is measured with error, then the imputed birth month from the IABS could be the correct birth month, although the Pension Register indicates otherwise. Consequently, the share of births for which the birth month imputed from the IABS and the birth month recorded in the Pension Register coincide is again best interpreted as a lower bound for the share of births for which the IABS measures the birth month correctly.

6 The consequences of using a noisy measure of maternity leave

The analysis so far has shown that, after some appropriate restrictions have been imposed, the vast majority of leave spells in the IABS are indeed due to childbirth. However, any sample based on the IABS is likely to contain some erroneous leave spells that cannot be linked to childbirth. For Western German women, the share of erroneous leave spells is at most 10%, while it may be as large as 20% for women of foreign nationality or for women in Eastern Germany. In addition, the month in which a woman gives birth, and therefore the time she spends at home before returning to work, is measured with error in the IABS. In this section, I provide a first analysis of whether measurement error in the IABS may lead to serious biases. I concentrate on two key issues: the woman's decision as to when to return to work, and the impact of the career interruption on wages after childbirth.

6.1 True maternity leave spells and observable characteristics

The extent to which measurement error in the IABS biases estimates crucially depends on how it is related to observable characteristics, such as birth month, education or age. I investigate this in Tables 7 and 8. The results are based on Sample B. Additionally, I impose my preferred sample restrictions. In particular, leave spells that start on the first of a month have been dropped from the sample. Panel A of Table 7 displays the share of correct leave spells by imputed birth month. I distinguish between women in Western and Eastern Germany, between women of foreign nationality or with German citizenship, and between all spells and the first spell. For Western German women, the share of correct leave spells varies from 87.15% in May to 91.77% in September. I just fail to reject the hypothesis that the birth month has no impact on whether the leave spell is due to childbirth or not at a 10% level (p-value 0.103). Importantly, although on average the share of correct leave spells is somewhat smaller in May than in other months, this pattern is not observed every single year. For women of foreign nationality and Eastern German women, the variation in the share of correct leave spells across birth months is considerably larger, due to the smaller sample size. Again, there is no clear pattern across years.
Table 7

True maternity leave spells and observable characteristics (preferred sample restrictions)

Panel A: True spells and birth month

 

All

All, 1st spell

West, Germans

West, Germans, 1st spell

West, Foreigners

West, Foreigners, 1st spell

East

1

88.20%

90.96%

88.79%

91.34%

81.90%

89.61%

83.33%

2

87.96%

91.43%

88.96%

92.11%

76.45%

86.49%

84.78%

3

89.04%

91.58%

89.80%

92.48%

81.45%

84.06%

81.48%

4

88.05%

89.62%

88.94%

90.47%

81.95%

87.18%

73.47%

5

86.05%

89.18%

87.15%

90.04%

76.56%

86.59%

75.41%

6

88.08%

89.53%

88.73%

90.25%

86.13%

88.30%

67.50%

7

89.55%

91.59%

90.14%

92.20%

80.00%

83.05%

89.83%

8

89.45%

90.91%

90.39%

92.11%

83.85%

86.67%

68.63%

9

90.69%

92.69%

91.77%

93.93%

84.00%

87.78%

71.15%

10

90.08%

92.77%

90.37%

92.89%

87.39%

94.59%

87.23%

11

86.75%

89.73%

87.98%

90.20%

75.78%

89.39%

75.00%

12

88.01%

89.40%

88.07%

90.24%

74.80%

80.49%

77.27%

N

22.397

15.277

20.308

13.752

1.483

935

601

p-value

0.076

0.088

0.103

0.097

0.149

0.384

0.115

Panel B: Other observable pre-birth characteristics

 

All

All, 1st spell

West, Germans

West, Germans, 1st spell

West, Foreigners

West, Foreigners, 1st spell

East

East

−0.115

−0.097

     
 

(0.016)

(0.016)

     

Foreign

0.000

0.016

     
 

(0.010)

(0.012)

     

Medium-skilled

0.072

0.045

0.075

0.048

0.027

0.022

0.076

 

(0.007)

(0.008)

(0.007)

(0.008)

(0.020)

(0.023)

(0.063)

High-skilled

0.127

0.075

0.130

0.077

0.150

0.133

0.052

 

(0.012)

(0.014)

(0.013)

(0.014)

(0.057)

(0.053)

(0.084)

Log-wage

0.011

0.038

0.014

0.042

–0.018

–0.004

–0.043

 

(0.006)

(0.007)

(0.006)

(0.008)

(0.024)

(0.028)

(0.047)

Age

0.146

0.138

0.147

0.136

0.120

0.095

0.185

 

(0.006)

(0.008)

(0.007)

(0.008)

(0.019)

(0.024)

(0.032)

Age2

–0.003

–0.003

–0.003

–0.003

–0.002

–0.002

–0.004

 

(0.000)

0.000

0.000

(0.000)

(0.000)

(0.000)

(0.001)

Full-time

0.025

–0.001

0.022

–0.003

0.079

0.052

–0.071

 

(0.006)

(0.008)

(0.007)

(0.009)

(0.027)

(0.033)

(0.042)

2nd spell

–0.021

 

–0.021

 

–0.043

 

–0.085

 

(0.005)

 

(0.006)

 

(0.025)

 

(0.102)

3rd spell

–0.090

 

–0.083

 

–0.180

  
 

(0.012)

 

(0.013)

 

(0.042)

  

4th spell

–0.240

 

–0.224

 

–0.319

  
 

(0.030)

 

(0.034)

 

(0.065)

  

N

22,219

15,137

20,226

13,692

1,400

868

588

Note: Panel A displays the share of leave spells in the IABS that can be linked to childbirth in the Pension Register by month of birth (in the IABS data), after imposing the preferred restrictions. The last row reports the p-value for the hypothesis that the birth month dummies in a linear probability model are jointly equal to zero. Panel B reports results from linear probability models where the dependent variable is equal to 1 if the leave spell in the IABS is due to childbirth. The findings are based on Sample B. Robust standard errors in parentheses.

Table 8

Correct birth month and observable characteristics (preferred restrictions)

Panel A: Correct birth month and birth month

 

West, Germans

West, Germans, 1st spell

West, Foreigners

West, Foreigners, 1st spell

East

1

68.63%

67.79%

60.27%

60.00%

76.19%

2

68.65%

69.26%

51.79%

51.35%

63.89%

3

69.35%

69.73%

70.77%

69.23%

64.52%

4

68.96%

67.32%

66.20%

59.57%

68.97%

5

70.23%

69.02%

54.29%

53.06%

66.67%

6

67.29%

67.50%

60.98%

61.54%

75.00%

7

70.52%

70.78%

52.73%

51.61%

76.09%

8

70.10%

69.91%

62.67%

67.24%

67.74%

9

70.85%

70.18%

61.33%

60.34%

74.19%

10

69.43%

68.30%

65.22%

68.75%

69.44%

11

69.26%

69.46%

62.12%

64.29%

84.00%

12

67.79%

67.78%

68.12%

69.77%

57.14%

N

16,532

11,454

826

554

397

p-value

0.692

0.821

0.496

0.650

0.655

Panel B: Other observable pre-birth characteristics

 

West, Germans

West, Germans, 1st spell

West, Foreigners

West, Foreigners, 1st spell

East

Medium-skilled

0.041

0.029

0.107

0.101

–0.038

 

(0.012)

(0.015)

(0.037)

(0.046)

(0.098)

High-skilled

0.075

0.052

0.165

0.210

–0.188

 

(0.021)

(0.025)

(0.096)

(0.098)

(0.140)

Log-wage

0.024

0.048

–0.028

0.004

0.015

 

(0.010)

(0.014)

(0.046)

(0.060)

(0.074)

Age

0.052

0.055

0.032

0.076

–0.029

 

(0.011)

(0.013)

(0.040)

(0.049)

(0.069)

Age2

–0.001

–0.001

–0.001

–0.001

0.000

 

0.000

0.000

(0.001)

(0.001)

(0.001)

Full-time

0.020

0.015

0.044

0.011

–0.023

 

(0.011)

(0.016)

(0.051)

(0.070)

(0.074)

2nd spell

0.007

 

0.025

 

–0.139

 

(0.009)

 

(0.044)

 

(0.155)

3rd spell

0.005

 

–0.116

  
 

(0.018)

 

(0.074)

  

4th spell

0.054

 

–0.182

  
 

(0.042)

 

(0.131)

  

N

16,479

11,413

787

522

388

Note: Panel A reports the share of leave spells where the birth month imputed from the start of the leave spell in the IABS coincides with that in the Pension Register, by birth month in the IABS data. The last row reports the p-value for the hypothesis that the birth month dummies in a linear probability model are jointly equal to zero. Panel B reports results from linear probability models where the dependent variable is equal to 1 if the birth month in the IABS is the same as that in the Pension Register. The findings are based on Sample C. Robust standard errors in parentheses.

In Panel B, I report results from linear probability models where the dependent variable is equal to 1 if the leave spell in the IABS can be linked to childbirth in the Pension Register. In line with the findings in Table 4, leave spells are less likely to be correct for women in Eastern Germany, whereas foreign nationality no longer has a negative impact on whether or not the leave spell in the IABS is due to childbirth. For all groups, the probability of the leave spell being due to childbirth increases with education and age, and decreases with the number of leave spells.

Table 8 presents a similar analysis, with an indicator variable for whether or not the month of birth is measured correctly as the dependent variable. The findings are based on Sample C. I again impose the following restrictions in addition to those mentioned in Sect. 3: women must be between 18 and 40, spells must be longer than two months, spells must not start on the first of a month, and spells must not be preceded by a spell in apprenticeship training. There appears to be no clear relationship between the month of birth and a correct measurement of the birth month. With the exception of women in Eastern Germany, the incidence of a correct measurement increases with education and age.

To sum up, the incidence of whether or not a leave spell in the IABS is due to childbirth is not random, but is correlated with, for instance, education, age, and the number of the leave spell. The share of wrong leave spells is, however, not strongly correlated with the month of birth. The same holds for measurement error in the month of birth.

6.2 True versus estimated time away from work

Since several observable characteristics, such as education, help to predict both types of measurement error in the IABS, one may worry that the IABS cannot be reliably used to identify career interruptions due to childbirth. Next, I use descriptive as well as regression tools to compare the time that mothers spend at home after childbirth in the IABS (where career interruptions due to childbirth can only be approximated) with that in the Pension Register (which includes precise information on childbirth). The findings are based on Sample B, and I impose my preferred sample restrictions. In Fig. 2a, I plot the share of women who return to work t months after childbirth in the IABS. It is important to bear in mind that throughout the time period considered, the IABS does not include information on jobs for which social security contributions do not have to be paid. Hence, if a mother accepts a so-called marginal part-time job, which is exempt from social security contributions (i.e. jobs with a monthly salary (in 2008) of less than 400 Euros), after childbirth, there is no record of this employment in the IABS. Maternity leave duration is computed as the time between the month in which the mother returns to work and the approximated month in which she gives birth (based on the month when she takes maternity leave). Hence, maternity leave duration is based on the social security data only, and can therefore be computed from the scientific use file of the IABS 75-01 and IABS 75-04. I also plot the share when the sample is restricted to spells that are due to childbirth. Here, I include spells that start on the first of a month in the sample, since there is no economic reason for excluding them. The analysis is now based on the month of birth observed in the Pension Register, as opposed to the imputed month of birth in the IABS. Maternity leave duration is computed as the time between the month when the mother returns to work, obtained from the social security records of the IABS, and the month in which she gives birth, obtained from the Pension Register. Hence, the approximation based on the IABS contains two sources of measurement error: first, the sample includes some spells that are not due to childbirth, and second, the month of birth – and thus time at home – is measured with error.
Fig. 2

True versus approximated time at home: share returning t months after childbirth (preferred sample restrictions)

I display the results separately for Western Germans, women of foreign nationality, and women in Eastern Germany. Clearly, the shares of mothers who return to the labour market 2, 10, 12, 15, 18, 24 and 36 months after childbirth exceed those in any other months. These dates coincide with some important dates in German maternity leave legislation. Throughout the time period considered, the job-protection period was gradually increased from ten months in 1986 to 36 months in 1992. Moreover, after two months, maternity benefit drops sharply from the full salary to 300 Euros per month, which is about 25% of the average salary. Furthermore, since 1993 (1988 in Baden-Wuerttemberg and 1989 in Bavaria), maternity benefits have been paid for 24 months, while the job-protection period is 36 months. Finally, two Eastern German states (Saxony and Thuringia) pay maternity benefits for an additional six months, which may explain the unusually large share of women returning to work 30 months after childbirth in Eastern Germany.

Clearly, the share of women returning in the 10th, 12th, … month is larger based on the “true” data from the Pension Register than that based on the approximated social security data in the IABS. This is expected because of measurement error in the month of birth in the IABS. The second most important difference between the true and the approximated data is that the approximated data overestimates the share of women who return to work very early, within four months after childbirth. A further inspection shows that this is because of the inclusion of leave spells that are not due to childbirth in the IABS data, rather than due to measurement error in the month of birth. That is, erroneous leave spells in the IABS tend to be shorter than leave spells that are due to childbirth.

I provide further findings in Fig. 2b, again separately for Western German women, women of foreign nationality, and women in Eastern Germany. The figure compares the Kaplan–Maier survival estimates based on the approximated IABS data with those based on the true data from the Pension Register. The figure confirms that an unusually large number of mothers return to work around the time when the job-protection or maternity benefit period ends, and this share is larger based on the Pension data than based on the IABS data. More importantly, the approximated survival curve always lies below the true survival curve, and runs roughly parallel to the true survival curve. Hence, the IABS data overestimates the share of women who return to the labour market early, within the first four months after childbirth, whereas the share of women returning later on is estimated more or less correctly.

Researchers and policy makers may also be interested in how observable characteristics, such as education, age or wages, affect women's return decision. In Table 9, I report results from (non-parametric) proportional hazard models, and compare estimates based on the approximated IABS data with those based on the Pension Register. For the Pension Register, I distinguish two samples: the first sample includes spells that start on the first of a month, while these spells are excluded in the second sample. Panel A pools women in Eastern and Western Germany and women of German and foreign nationality; Panels B to D display results for each group separately. In addition, I report results separately for all spells and first spells. The reason for this is the previous finding that the share of leave spells in the IABS that can be linked to childbirth in the Pension Register is somewhat larger for the first spell (Table 4). A coefficient of greater than 1 implies that the variable increases the hazard rate, while a coefficient of smaller than 1 implies that the variable decreases the hazard rate.
Table 9

True versus approximated leave spells: Proportional hazard models

Panel A: All

 

All spells

1st spell

 

True, 1st incl.

True, 1st excl.

Approx.

True, 1st incl.

True, 1st excl.

Approx.

 

N = 18.937

N = 17.655

N = 22.219

N = 13.180

N = 12.314

N = 15.137

East

2.108

(0.095)

2.100

(0.100)

1.850

(0.080)

2.456

(0.119)

2.453

(0.124)

2.139

(0.099)

Foreign

1.313

(0.057)

1.339

(0.060)

1.309

(0.046)

1.331

(0.071)

1.366

(0.076)

1.292

(0.057)

Medium-skilled

1.023

(0.029)

1.037

(0.030)

0.990

(0.024)

0.962

(0.034)

0.974

(0.035)

0.954

(0.029)

High-skilled

1.276

(0.066)

1.312

(0.070)

1.163

(0.054)

1.235

(0.077)

1.271

(0.082)

1.180

(0.068)

Log-wage

1.190

(0.029)

1.194

(0.030)

1.204

(0.026)

1.504

(0.053)

1.512

(0.055)

1.448

(0.046)

Full-time

1.221

(0.030)

1.221

(0.031)

1.169

(0.027)

1.214

(0.046)

1.215

(0.048)

1.185

(0.041)

Age

0.940

(0.023)

0.948

(0.024)

0.898

(0.019)

0.876

(0.026)

0.884

(0.028)

0.828

(0.022)

Age2

1.001

0.000

1.001

0.000

1.002

0.000

1.002

(0.001)

1.002

(0.001)

1.003

0.000

ML

12 months

0.960

(0.033)

0.942

(0.033)

0.949

(0.030)

0.953

(0.040)

0.925

(0.040)

0.926

(0.036)

15 months

0.908

(0.033)

0.887

(0.033)

0.915

(0.031)

0.924

(0.041)

0.902

(0.042)

0.913

(0.038)

18 months

0.843

(0.028)

0.828

(0.028)

0.836

(0.026)

0.847

(0.035)

0.830

(0.035)

0.827

(0.031)

36 (18) months

0.779

(0.028)

0.769

(0.028)

0.776

(0.025)

0.778

(0.034)

0.768

(0.035)

0.772

(0.031)

36 (24) months

0.723

(0.024)

0.710

(0.024)

0.737

(0.022)

0.717

(0.029)

0.703

(0.029)

0.724

(0.027)

Panel B: West, Germans

 

All spells

1st spell

 

True, 1st incl.

True, 1st excl.

Approx.

True, 1st incl.

True, 1st excl.

Approx.

 

N = 17.663

N = 16.479

N = 20.226

N = 12.206

N = 11.413

N = 13.692

Medium-skilled

1.034

(0.031)

1.046

(0.033)

1.019

(0.027)

0.971

(0.037)

0.979

(0.038)

0.978

(0.033)

High-skilled

1.307

(0.071)

1.334

(0.075)

1.198

(0.059)

1.264

(0.084)

1.288

(0.089)

1.201

(0.074)

Log-wage

1.157

(0.029)

1.162

(0.030)

1.163

(0.027)

1.481

(0.055)

1.492

(0.058)

1.410

(0.048)

Full-time

1.217

(0.031)

1.213

(0.033)

1.156

(0.028)

1.199

(0.049)

1.190

(0.051)

1.152

(0.044)

Age

0.951

(0.024)

0.958

(0.025)

0.895

(0.020)

0.882

(0.028)

0.889

(0.030)

0.815

(0.023)

Age2

1.001

0.000

1.001

0.000

1.002

(0.000)

1.002

(0.001)

1.002

(0.001)

1.004

0.000

ML

12 months

0.974

(0.034)

0.957

(0.034)

0.957

(0.031)

0.965

(0.041)

0.937

(0.041)

0.935

(0.037)

15 months

0.919

(0.034)

0.900

(0.035)

0.925

(0.032)

0.927

(0.042)

0.907

(0.042)

0.926

(0.040)

18 months

0.852

(0.029)

0.837

(0.029)

0.839

(0.027)

0.852

(0.036)

0.834

(0.036)

0.830

(0.032)

36 (18) months

0.786

(0.029)

0.778

(0.029)

0.771

(0.026)

0.786

(0.035)

0.776

(0.036)

0.774

(0.032)

36 (24) months

0.743

(0.025)

0.730

(0.025)

0.750

(0.023)

0.736

(0.030)

0.721

(0.030)

0.744

(0.028)

Panel C: West, Foreigners

 

All spells

1st spell

 

True, 1st incl.

True, 1st excl.

Approx.

True, 1st incl.

True, 1st excl.

Approx.

 

N = 851

N = 787

N = 1.400

N = 564

N = 522

N = 868

Medium-skilled

0.984

(0.080)

0.984

(0.083)

0.870

(0.056)

0.956

(0.096)

0.974

(0.102)

0.874

(0.072)

High-skilled

0.813

(0.214)

0.981

(0.257)

0.741

(0.180)

0.872

(0.260)

1.157

(0.328)

1.040

(0.274)

Log-wage

1.582

(0.181)

1.547

(0.182)

1.494

(0.130)

1.488

(0.215)

1.417

(0.207)

1.401

(0.152)

Full-time

1.171

(0.131)

1.214

(0.140)

1.172

(0.098)

1.566

(0.219)

1.725

(0.238)

1.446

(0.158)

Age

0.899

(0.075)

0.931

(0.082)

0.932

(0.059)

0.838

(0.086)

0.850

(0.091)

0.921

(0.078)

Age2

1.002

(0.002)

1.001

(0.002)

1.001

(0.001)

1.003

(0.002)

1.003

(0.002)

1.001

(0.002)

ML

12 months

0.701

(0.111)

0.697

(0.116)

0.837

(0.101)

0.683

(0.150)

0.684

(0.156)

0.750

(0.133)

15 months

0.718

(0.126)

0.674

(0.126)

0.790

(0.109)

0.867

(0.261)

0.811

(0.204)

0.668

(0.135)

18 months

0.667

(0.101)

0.673

(0.105)

0.779

(0.089)

0.681

(0.136)

0.679

(0.137)

0.717

(0.116)

36 (18) months

0.628

(0.101)

0.623

(0.103)

0.823

(0.098)

0.602

(0.126)

0.598

(0.127)

0.698

(0.116)

36 (24) months

0.455

(0.068)

0.451

(0.070)

0.607

(0.068)

0.445

(0.087)

0.434

(0.086)

0.487

(0.077)

Panel D: East

 

All spells

      
 

True, 1st incl.

True, 1st excl.

Approx.

      
 

N = 422

N = 388

N = 588

      

Medium-skilled

1.027

(0.255)

1.099

(0.293)

0.810

(0.165)

      

High-skilled

1.319

(0.468)

1.379

(0.511)

0.978

(0.292)

      

Log-wage

2.376

(0.383)

2.412

(0.410)

2.652

(0.372)

      

Full-time

1.292

(0.206)

1.337

(0.219)

1.377

(0.180)

      

Age

0.834

(0.132)

0.834

(0.156)

0.891

(0.102)

      

Age2

1.002

(0.003)

1.002

(0.003)

1.002

(0.002)

      

1994

1.114

(0.116)

1.077

(0.117)

1.034

(0.094)

      

Note: The table compares results from proportional hazard models in the IABS, after imposing my preferred restrictions (“Approx.”), with those in the Pension Register (“true”). Here, I distinguish samples. In the first column, spells that start on the first of a month are included in the sample. In the second column, they are excluded. Robust standard errors in parentheses.

Table 9 reveals several interesting patterns. First, all of the variables have the expected signs, in both the true and the approximated data. Education, wages prior to childbirth, and working full-time prior to childbirth all increase the hazard rate, while age decreases it. Moreover, the hazard rate declines with the expansion of the maternity leave period over the sample period.

Second, the approximation based on the IABS slightly underestimates the impact of education on the hazard rate for all groups. This may be because both sources of measurement error are less severe for better educated workers. Similarly, age has a more negative impact on the hazard rate in the approximated IABS data than in the true Pension Register data; again, this could be because both types of measurement error are more severe for younger workers. The IABS data also underestimates the impact of living in Eastern Germany on the hazard rate. All of the other coefficients are very similar for the approximated IABS data and the true Pension Register data. In particular, the estimates for the impact of the leave period are almost identical for the true and the approximated data.

Third, the biases tends to be slightly larger for women of foreign nationality and for Eastern Germans than for Western German women, for whom both types of measurement error are slightly less severe. Fourth, excluding spells that start on the first of a month in the Pension Register data has very little impact on the estimates. This suggests that dropping these spells in the IABS is unlikely to pose any problems.

6.3 True versus estimated impact of time away from work on wages

Next, I compare the impact of the time spent at home after childbirth on the wage drop following childbirth in the IABS data (where career interruptions due to childbirth can only be approximated) with that in the Pension Register (which includes precise information on childbirth). The findings are based on Sample B, and I impose my preferred sample restrictions. I further restrict the sample to women who return to the labour market within six years of childbirth. As in the previous section, the approximation based on the IABS contains two sources of measurement error: first, the sample includes some spells that are not due to childbirth, and second, the month of birth, and thus the time spent at home, is measured with error.

I estimate first difference models. My dependent variable is the difference between the first log-wage observed after childbirth and the pre-birth log-wage. I regress this variable on the number of months which the mother stays at home after childbirth, her education, age and age squared, the change in full-time status before and after childbirth, and dummy variables for the birth year. The results can be found in Table 10. Panel A displays the results for Western German women, while Panels B and C report the results for women of foreign nationality and Eastern German women respectively.
Table 10

True versus approximated leave spells: The impact of career interruptions on post-birth wages

Panel A: West, Germans

 

All spells

1st spell

 

True, 1st incl.

True, 1st excl.

Approx.

True, 1st incl.

True, 1st excl.

Approx.

 

N = 11.005

N = 10.283

N = 12.976

N = 7.232

N = 6.766

N = 8.362

Time at home

–0.004

(0.000)

–0.004

(0.000)

–0.006

(0.000)

–0.004

(0.000)

–0.004

(0.000)

–0.005

(0.000)

Medium-skilled

–0.109

(0.015)

–0.109

(0.016)

–0.117

(0.012)

–0.150

(0.019)

–0.151

(0.020)

–0.162

(0.016)

High-skilled

–0.035

(0.027)

–0.031

(0.028)

–0.061

(0.025)

0.019

(0.034)

0.023

(0.035)

–0.022

(0.031)

Change full-time

–0.005

(0.007)

–0.003

(0.007)

0.013

(0.006)

0.089

(0.010)

0.089

(0.011)

0.096

(0.009)

Age

–0.104

(0.013)

–0.099

(0.014)

–0.116

(0.010)

–0.156

(0.017)

–0.148

(0.017)

–0.173

(0.013)

Age2

0.002

(0.000)

0.002

(0.000)

0.002

(0.000)

0.003

(0.000)

0.002

(0.000)

0.003

(0.000)

Panel B: West, Foreigners

 

All spells

1st spell

 

True, 1st incl.

True, 1st excl.

Approx.

True, 1st incl.

True, 1st excl.

Approx.

 

N = 605

N = 560

N = 1.035

N = 381

N = 354

N = 602

Time at home

–0.001

(0.002)

–0.001

(0.002)

–0.002

(0.001)

0.001

(0.002)

0.001

(0.002)

0.000

(0.001)

Medium-skilled

–0.032

(0.039)

–0.019

(0.040)

–0.043

(0.029)

–0.060

(0.051)

–0.048

(0.048)

–0.039

(0.040)

High-skilled

0.115

(0.063)

0.101

(0.061)

0.053

(0.047)

0.082

(0.065)

0.066

(0.064)

0.035

(0.051)

Change full-time

0.038

(0.024)

0.044

(0.025)

0.023

(0.019)

0.018

(0.031)

0.019

(0.032)

0.004

(0.026)

Age

0.019

(0.039)

–0.003

(0.039)

–0.036

(0.028)

–0.017

(0.049)

–0.028

(0.051)

–0.060

(0.036)

Age2

0.000

(0.001)

0.000

(0.001)

0.001

(0.000)

0.000

(0.001)

0.000

(0.001)

0.001

(0.001)

Panel C: East

      
 

All spells

      
 

True, 1st incl.

True, 1st excl.

Approx.

      
 

N = 368

N = 338

N = 495

      

Time at home

0.005

(0.002)

0.004

(0.002)

0.002

(0.001)

      

Medium-skilled

–0.025

(0.080)

–0.070

(0.086)

–0.082

(0.058)

      

High-skilled

0.007

(0.121)

–0.047

(0.128)

–0.051

(0.085)

      

Change full-time

0.034

(0.031)

0.028

(0.032)

0.010

(0.024)

      

Age

0.048

(0.062)

0.018

(0.068)

–0.038

(0.039)

      

Age2

–0.001

(0.001)

0.000

(0.001)

0.001

(0.001)

      

Note: The table compares results from first difference models (i.e. the difference between the logarithm of the post-birth and pre-birth wage) in the IABS, after imposing my preferred restrictions (“Approx.”) with those in the Pension Register (“true”). Here, I distinguish two samples. In the first column, spells that start on the first of a month are included in the sample. In the second column, they are excluded. Robust standard errors in parentheses.

The most important pattern that emerges from the table is that the IABS data somewhat overestimates the negative impact of career interruptions on the wage drop following childbirth, in particular for Western German women. For this group, the estimate based on the IABS data implies that spending one additional year at home after childbirth reduces wages by 7.0%. In contrast, the true estimate based on the Pension Register is only 5.3%. The bias is similar when the sample is restricted to the first spell (6.5% versus 4.7%). The pattern is similar for women of foreign nationality and women in Eastern Germany: the impact of career interruptions is more negative (or, in the case of Eastern German women, less positive) in the IABS data than in the Pension Register. One explanation for this finding is that the IABS sample includes women who are on leave from their employer because they are sick, and not because they have given birth. Wage losses due to sick leave may exceed those due to maternity leave.

Also note that it makes little difference in the Pension Register whether or not we include maternity leave spells that start on the first of a month. This provides further evidence that this restriction is unlikely to bias results.

A further interesting finding that emerges from Table 10 is that the wage penalty associated with longer career interruptions is substantially larger for Western German women than for women of foreign nationality or for Eastern German women, in both the IABS data and the Pension Register. In fact, the estimates in Table 10 suggest that (in Eastern Germany?) staying at home longer has a positive impact on the wage drop following childbirth. However, this finding has to be interpreted with considerable caution. This is because Eastern Germany experienced substantial wage growth throughout the early and mid-1990s. Since fertility data for Eastern Germany is only available from 1992 to 1995, it is difficult to disentangle the impact of career interruptions from that of aggregate wage growth.

7 Conclusion

The two German micro data sets most commonly used to study the effects of career interruptions due to childbirth or taking maternity leave are the IAB Employment Sample and the German Socio-Economic Panel. I summarize the main advantages and disadvantages of each data set in Table 11. The main advantage of the IABS is the large sample size and the precise measurement of earnings. The main advantage of the German Socio-Economic Panel is the inclusion of a wide variety of background characteristics. In particular, the IABS does not include direct information on childbirth. It does, however, contain a variable that indicates an interruption of the employment relationship, so women who go on maternity leave can potentially be identified.
Table 11

Advantages and disadvantages of the IABS and the GSOEP

IABS

GSOEP

Advantages

Disadvantages

Advantages

Disadvantages

1) Large sample size

1) No direct information on childbirth

1) Complete fertility history available

1) Small sample Size

2) Precise information on wages and employment

2) No information on marital status, spousal income and labour supply, child care usage, etc.

2) Information on e.g. marital status, spousal income and labour supply, child care usage, etc.

2) Wages and employment are likely to be measured with error

 

3) No information on parental leave of fathers

3) Parental leave of fathers can be analyzed

 
 

4) Information on marginal employment since 1999 only

4) Information on marginal employment

 
 

5) No information on hours worked, other than full-time and part-time work

5) Information on (actual and usual) hours worked

 
 

6) Civil servants (e.g. teachers) and self-employed are excluded

6) Covers all individuals

 

Note: The table provides an overview of the main advantages and disadvantages of the IABS and GSOEP for studying the impact of childbirth on mothers' (and fathers') careers.

In this paper, I analyze the measurement error associated with the maternity leave variable in the IABS. My overall conclusions are positive: the vast majority of leave spells in the IABS are due to maternity leave (at least 90% for Western German women), but only after certain restrictions have been imposed. The child's birth month is measured correctly for at least 70% of births, and over- or underestimated by one month for a further 25%. I therefore conclude that the scientific use files of the IABS 75-01 and IABS 75-04 provide a very valuable alternative data source to the GSOEP to analyze career interruptions due to childbirth. Questions that can be addressed using the IABS include: When do women who take maternity leave return to the labour market after childbirth? How does the time away from work after childbirth affect the future wage growth of mothers? Which mothers experience the largest wage losses? Have there been changes over time? How have the expansions in maternity leave coverage that have taken place in Germany since the late 70s affected the labour supply and wages of mothers? Despite some caveats, the IABS data also has advantages over some widely used data sets in other countries. For instance, the Panel Study of Income Dynamics (PSID), the National Longitudinal Survey of Youth (NLSY), or the Survey of Income and Program Participation (SIPP) suffer, just like the GSOEP, from a small sample size, or follow individuals only for a few years so that long-term consequences of career interruptions due to childbirth cannot be analyzed.

There are, however, a number of questions that cannot be addressed with the IABS, but could possibly be addressed with the GSOEP, if the sample size is large enough (see also Table 11). Most importantly, the IABS does not allow researchers to identify childbirth, but only leave from employment. I show that between 1987 and 1994, about 50% of Western German mothers went on leave, although the share is likely to be considerably larger for first-time mothers. Taking maternity leave is substantially more common among mothers who were employed prior to childbirth. The IABS should therefore only be used if the focus is on mothers who are attached to the labour market.

Second, for fathers, only a small share of leave spells is due to paternity leave. Hence, the data set cannot be used to analyze how many fathers take paternity leave, and what consequences taking paternity leave has on their future careers. This is a highly relevant question in light of recent policy developments. For instance, the 2007 reform in Germany provided strong incentives for fathers and mothers to share parental leave. There are similar incentives for instance in Sweden, Iceland and Austria.

Third, although the IABS includes more background information than most other administrative data sets, some limitations remain. For instance, it is not possible to identify single women, or to link husbands and wives, in the IABS. Both are possible in the GSOEP. Moreover, unlike the GSOEP, the IABS does not include information on hours worked (other than full- or part-time work), and provides no information on women who are not covered by social security such as civil servants and the self-employed. A further disadvantage of the IABS, compared to the GSOEP, is that it only includes information on marginal part-time employment (i.e. jobs with a monthly salary (in 2008) of less than 400 Euros) from 1999. These jobs are likely to be common among mothers with young children. Hence, whether the GSOEP or the IABS is better suited to analyze a research problem depends on the research question.

Executive summary

The data set that researchers have used most often to study career interruptions due to childbirth in the German context is the German Socio-Economic Panel (GSOEP). An alternative data source is the much larger IAB Employment Sample (IABS). Although this data set does not include direct information on childbirth, mothers on maternity leave can potentially be identified. There are, however, two problems. First, the leave variable in the IABS does not distinguish between maternity leave and other leave, such as sick leave. Second, the child's birth month has to be inferred from the month in which the mother goes on maternity leave, which is likely to lead to measurement error in the time which the mother spends at home after childbirth.

This paper investigates both problems, using an extended version of the IABS that supplements the social security records with direct information on childbirth from the German Pension Register. My conclusion is positive: I find that for Western German women, at least 90% of leave spells in the IABS are due to maternity leave, but only after some sample restrictions have been imposed. The child's birth month is correctly estimated for at least 70%, and over- or underestimated by one month for about 25% of mothers.

I also directly investigate the biases that may arise due to the two types of measurement error in the IABS. I focus on two issues: women's decisions as to when to return to work after childbirth, and the impact of the duration of the career interruption on subsequent wages. Overall, both the IABS and the Pension Register (where the two types of measurement error are absent) yield very similar findings. However, the IABS slightly underestimates the impact of education and age on the returning hazard. Probably most importantly, the IABS somewhat overestimates the cost of career interruptions: one year at home is associated with a wage loss of 7% in the IABS, but only 5.3% in the Pension Register.

Finally, I analyze how many and which mothers take maternity leave. I find that between 1987 and 1994, about 50% of mothers in Western Germany, and 59% in Eastern Germany took maternity leave. The share is likely to be considerably larger for first-time mothers. Not surprisingly, taking maternity leave is substantially more common for mothers who were employed around conception, i.e. around nine months prior to childbirth (around 90%).

I conclude that the most recent scientific use files of the IABS, the IABS 75-01 and IABS 75-04, provide a very valuable alternative data source to the GSOEP to study career interruptions due to childbirth, as long as the focus is on women who are attached to the labour market.

Kurzfassung

Der Datensatz, der in Deutschland am häufigsten benutzt wurde, um Erwerbsunterbrechungen von jungen Müttern zu untersuchen, ist das sozio-ökonomische Panel. Ein alternativer Datensatz ist die wesentlich größere IAB-Beschäftigtenstichprobe (IABS). Dieser Datensatz enthält zwar keine direkten Informationen über das Geburtsdatum von Kindern. Mütter im Erziehungsurlaub können jedoch über Erwerbsunterbrechungen identifiziert werden. Hier gibt es jedoch zwei Probleme. Erstens, die Erwerbsunterbrechungsvariable in der IABS unterscheidet nicht zwischen einer Unterbrechung aufgrund von Erziehungsurlaub und einer Unterbrechung von z. B. Krankheit. Zweitens, der Geburtsmonat des Kindes muss vom Monat, in dem die Mutter in den Erziehungsurlaub geht, abgeleitet werden. Dies führt wahrscheinlich zu einem Messfehler in der Dauer der Erwerbsunterbrechung.

Dieser Datenreport untersucht beide Probleme basierend auf einer erweiterten Version der IABS, die zusätzlich zu den Sozialversicherungsangaben der IABS direkte Informationen über das Geburtsdatum der Kinder enthält. Diese Information stammt aus den Daten der Rentenversicherung. Meine Ergebnisse für westdeutsche Frauen zeigen, dass mindestens 90% der Erwerbsunterbrechungen in der IABS Unterbrechungen aufgrund von Erziehungsurlaub sind. Außerdem wird für mindestens 70% der Mütter der Geburtsmonat des Kindes in der IABS korrekt gemessen. Für weitere 25% wird der Geburtsmonat um einen Monat unter- oder überschätzt.

Außerdem untersuche ich, ob und wie sich die beiden Messfehlerarten in der IABS auf verschiedene Schätzungen auswirken. Hier konzentriere ich mich auf zwei Fragestellungen: die Entscheidung von Müttern, wann sie nach der Geburt ihres Kindes wieder in den Arbeitsmarkt zurückkehren, und die Auswirkungen der Länge der Erwerbsunterbrechung auf die Löhne der Mütter. Im Großen und Ganzen sind die Ergebnisse, die auf der IABS und auf den Daten der Rentenversicherung (in denen es die beiden Messfehler nicht gibt) basieren, sehr ähnlich. Allerdings ist zu berücksichtigen, dass die IABS den Effekt von Ausbildung und Alter auf die Rückkehrrate in den Arbeitsmarkt leicht unterschätzt. Außerdem werden die Kosten einer Erwerbsunterbrechung in der IABS leicht überschätzt: Eine Erwerbsunterbrechung von einem Jahr führt zu einem Lohnverlust von 7% basierend auf den Daten der IABS, verglichen mit nur 5,3% basierend auf den Daten der Rentenversicherung.

Schließlich untersuche ich, wie viele Mütter vom Erziehungsurlaub Gebrauch machen. Dies ist für etwa 50% der Mütter in Westdeutschland und 59% der Mütter in Ostdeutschland der Fall. Für Erstgebärende ist der Anteil wahrscheinlich deutlich höher. Es sollte nicht überraschen, dass unter Frauen, die 9 Monate vor der Geburt ihres Kindes beschäftigt waren, die Inanspruchnahme vom Erziehungsurlaub wesentlich höher ist (etwa 90%).

Mein Fazit ist, dass die neueren Scientific Usefiles der IABS, die IABS 75-01 und die IABS 75-04, eine sehr wertvolle alternative Datenquelle zum sozio-ökonomischen Panel darstellen, um Erwerbsunterbrechungen aufgrund von Erziehungsurlaub zu studieren. Allerdings muss berücksichtigt werden, dass in der IABS nur Mütter im Erziehungsurlaub, und nicht generell die Geburt eines Kindes, beobachtet wird.

Footnotes
1

A detailed overview of the various versions of the IAB Employment Samples can be found in Appendix B.

 
2

I provide a detailed comparison of the advantages of the GSOEP and the IABS in Table 11.

 
3

Own calculations based on the IABS 75-01.

 
4

In January 1994, an income cap was introduced and couples whose gross annual income exceeded approximately 70,000 Euros (≈ 50,000 for singles) no longer qualified for maternity benefit between the third and sixth month after the birth of the child (Zmarzlik et al. 1999). The income cap has been reduced several times since 1994.

 
5

Information on marginal part-time jobs that are exempt from social security contributions (i.e. jobs that (in 2008) pay less than 400 Euros per month) is included in the IAB Employment Samples from 1999 onwards only. Marginal part-time employment may be particularly common for women with young children.

 
6

This data set with one extension of the IABS 75-95 has been used for example by Beblo and Wolf (2002), Bender et al. (2003), and Beblo et al. (2006) to analyze the impact of career interruptions on wage growth.

 
7

Source: Federal Statistical Office, Germany.

 
8

This data set with two extensions of the IABS 75-95 has been used by Müller (2007). She refers to the two extensions as supplements I and II.

 
9

I exclude the first (1986) and last (1995) years of the Pension Register for the following reason. As I describe in Appendix A, I define a leave spell as being due to maternity leave if there is a birth six months before or after the start of the leave spell. This is not possible for (all) births in the first and last years of the Pension Register. There is no reason to delete these years from the analysis for researchers who work with the IABS only.

 
10

I do not do this for Eastern German women since 96% of all leave spells refer to the first leave spell. This is because the social security data is only available from 1992 onwards.

 
11

I.e. around 10% of leave spells are not due to childbirth, and of those, half cannot be linked to any other reason for taking leave in the Pension Register.

 

Notes

Declarations

Acknowledgements

The author would like to thank Stefan Bender, Nils Drews, Astrid Kunze and Dana Müller for helpful comments and suggestions.

Authors’ Affiliations

(1)
Department of Economics, University College London

References

  1. Beblo, M., Wolf, E.: Wage Penalties for Career Interruptions. An Empirical Analysis for West Germany. ZEW Discussion paper, 02-45, Mannheim (2002)Google Scholar
  2. Beblo, M., Bender, S., Wolf E.: The Wage Effects of Entering Motherhood. A Within-Firm Matching Approach. IAB Discussion Paper 13/2006 (2006)Google Scholar
  3. Bender, S., Haas A., Klose C.: IAB Employment Subsample 1975–95: Opportunities for Analysis by the Anonymised Subsample. IZA Discussion Paper No. 117 (2000)Google Scholar
  4. Bender, S., Kohlmann, A., Lang, S.: Women, Work, and Motherhood. Changing Employment Penalties for Motherhood in West Germany After 1945. A Comparative Analysis of Cohorts Born in 1934–1971. MPIDR Working paper, 2003–006, Rostock (2003)Google Scholar
  5. Bundesagentur für Arbeit (2004): Arbeitsmarkt 2003, Amtliche Nachrichten der Bundesagentur für ArbeitGoogle Scholar
  6. Bundesministerium für Familie, Senioren, Frauen und Jugend (BMFSFJ): Bundesstatistik Erziehungsgeld 2000 (2000)Google Scholar
  7. Bundesministerium für Familie, Senioren, Frauen und Jugend (BMFSFJ): Mutterschutzgesetz. Leitfaden zum Mutterschutz. Bonn (2007a)Google Scholar
  8. Bundesministerium für Familie, Senioren, Frauen und Jugend (BMFSFJ): Elterngeld und Elternzeit. Das Bundeselterngeld- und Elternzeitgesetz. Bonn (2007b)Google Scholar
  9. Ejrnæs, M., Kunze A.: What Is Driving the Family Gap in Women's Wages? Norwegian School of Economics and Business Administration, mimeo (2006)Google Scholar
  10. Engstler, H., Menning S.: Die Familie im Spiegel der amtlichen Statistik. Bundesministerium für Familie, Senioren, Frauen und Jugend (BMFSFJ), Bonn (2003)Google Scholar
  11. Görlich, D., De Grip, A.: Human Capital Depreciation During Family-Related Career Interruptions in Male and Female Occupations. Maastricht University, mimeo (2007)Google Scholar
  12. Kunze, A.: The Timing of Careers and Human Capital Depreciation. IZA Discussion Paper No. 509 (2002)Google Scholar
  13. Kreyenfeld, M.: Employment and fertility – East Germany in the 1990s. Doctoral Thesis. Rostock University, Rostock (2001)Google Scholar
  14. Müller, D.: Der Traum einer kontinuierlichen Beschäftigung. Erwerbsunterbrechungen bei Männern und Frauen. In: Szydlik, M. (Hrsg.) Flexibilisierung. Folgen für Arbeit und Familie. Sozialstrukturanalyse, S. 47–67. VS Verlag für Sozialwissenschaften, Wiesbaden (2007)Google Scholar
  15. Prinz, K.: Versicherungsverläufe von Frauen und Männern der Geburtsjahrgänge 1931 bis 1960. Deutsche Rentenversicherung 3–4, 220–241 (1997)Google Scholar
  16. Schönberg, U., Ludsteck, J.: Maternity Leave Legislation, Female Labor Supply, and the Family Wage Gap. University College London, mimeo (2008)Google Scholar
  17. Vlasbloom, J.D., Schippers J.J.: The Dynamics of Female Employment around Childbirth. Tjalling C. Koopmans Research Institute, Discussion Paper No. 03–10 (2003)Google Scholar
  18. Weber, A.M.: Wann kehren Mütter auf dem Arbeitsmarkt zurück? Eine Verweildaueranalyse für Deutschland. ZEW Discussion Paper 04–08 (2004)Google Scholar
  19. Wübbeke, C.: Der Übergang in den Rentenbezug im Spannungsfeld betrieblicher Personal- und staatlicher Sozialpolitik (Textband). In: Beiträge zur Arbeitsmarkt- und Berufsforschung (BeitrAB) 290.1, Nürnberg (2005a)Google Scholar
  20. Wübbeke, C.: Der Übergang in den Rentenbezug im Spannungsfeld betrieblicher Personal- und staatlicher Sozialpolitik (Anlageband). In: Beiträge zur Arbeitsmarkt- und Berufsforschung (BeitrAB) 290.2, Nürnberg (2005b)Google Scholar
  21. Zmarzlik, J., Zipperer, M., Viethen, H.P.: Mutterschutzgesetz, Mutterschaftsleistungen, Bundeserziehungsgesetz, Bd. 8. Carl Heymanns Verlag, Bonn (1999)Google Scholar

Copyright

© Institut für Arbeitsmarkt- und Berufsforschung 2009