Are Lone Mothers Responsive to Policy Changes? Evidence from a Workfare Reform in a Generous Welfare State* Authors: Magne Mogstad; Research Department, Statistics Norway/ESOP, Dept. of Economics, University of Oslo; e-mail:
[email protected] Chiara Pronzato; Dondena (Bocconi University), Research Department/Statistics Norway; CHILD Collegio Carlo Alberto; e-mail:
[email protected] Abstract: There is a heated debate in many developed countries about how to design a welfare system that moves lone mothers off welfare and into work. We analyze the consequences of a major Norwegian workfare reform of the generous welfare system for lone mothers. The reform imposed work requirements and time limits on welfare receipt, while raising in-work benefits. Our difference-in-differences estimates show that the reform was successful in improving labor market participation and increasing the earnings of lone mothers. However, the reform was associated with income loss and increased poverty among a sizeable subgroup of lone mothers who were unable to offset the loss of out-of-work benefits with gains in earnings.
Keywords: lone mothers, workfare reform, earnings, labor market participation, poverty, disposable income JEL classification: C23, I32, I38, J00
* We thank two anonymous referees, Rolf Aaberge, Tony Atkinson, Richard Blundell, Andrea Brandolini, Ugo Colombino, John Ermisch, Marco Francesconi, Tarjei Havnes, Terje Skjerpen, as well as participants at a number of seminars and conferences for helpful comments. The Norwegian Research Council and the European Research Council have provided financial support for this project.
1. Introduction Most of what we know about how lone mothers respond to policy changes that increase their incentives to move off welfare and into work comes from programme evaluations carried out in Canada, the UK, and the US. The results are striking: the welfare reforms had generally positive effects on employment, earnings and income, and reduced programme caseloads and government expenditure.1 The Anglo-Saxon experience has fuelled an increasing interest among European policymakers and researchers in policy changes providing stronger work incentives for lone mothers. Yet caution must be applied when drawing lessons from the Anglo-Saxon experience, as emphasized in several recent review articles (see e.g. Blank, 2002; Moffitt 2007; Brewer et al., 2009). In particular, the impact of welfare reforms may depend heavily on the broader institutional context and the economic environment in which they are implemented. On this background, Brewer et al. (2009) conclude that evidence from countries with a different economic and institutional structure than the Anglo-Saxon has significant value for research and policy. In this paper, we examine the consequences of a major Norwegian workfare reform of the generous welfare system for lone mothers. The reform imposed and enforced work requirements and time-limits on welfare receipt, while raising in-work benefits. The aim of the reform was to improve the labor market attachment of lone mothers, and thereby reduce welfare dependency and alleviate poverty. Our study provides the first evidence of the effects of the Norwegian workfare reform. To this end, we make use of a unique household panel data set based on administrative registers covering the entire resident population. To identify
1
Lone parent benefits in the US underwent a major workfare reform in 1996, when time limits and work requirements were imposed, the funding for childcare increased and, in many states, the benefit reduction rates were lowered. Moffitt (2007) summarizes the evidence on this much studied reform. In addition, there are a number of program evaluations of in-work benefit reforms, including Eissa and Liebman (1996) and Meyer and Rosenbaum (2001) of the Earned Income Tax Credit reform in the US, Brewer and Gregg (2001), Blundell et al. (2005) and Francesconi and van der Klaauw (2007) of the Working Families’ Tax Credit reform in the UK, as well as Michalopoulos et al. (2005), Card and Hyslop (2006), and Bitler et al. (2008) of the Canadian SelfSufficiency project. Brewer et al. (2009) review the evidence from the Working Families’ Tax Credit reform.
1
the reform effects on lone mothers, we use a difference-in-differences (DD) approach. Our baseline specification compares the outcome of interest of lone mothers to that of a comparison group of married mothers (with children of the same age), before and after the reform.2 To increase confidence in our identification strategy, we perform several specification checks. Unlike most previous studies of welfare reforms targeting lone mothers, we investigate the impacts not only on earnings and labor market participation, but also on disposable income and poverty. The insights may be summarized in two broad conclusions. First, the workfare reform was successful in increasing the average earnings and labor market participation of lone mothers. Second, the reform was associated with the side-effects of income loss and increased poverty among a sizeable subgroup of lone mothers who were unable to offset the loss of out-of-work benefits with gains in earnings. As theory predicts systematic heterogeneity in the impact of the reform, we also look closely into whether the policy affected different groups differently. The remainder of the paper is organized as follows. Section 2 describes the Norwegian welfare system for lone mothers and discusses the expected effects of the workfare reform. Section 3 outlines our empirical strategies, whereas Section 4 describes the data. Section 5 reports the empirical results, and Section 6 concludes.
2. Background This section describes in detail the workfare reform of the Norwegian welfare system for lone mothers, before commenting on predictions from labor supply theory. Policy changes Historically, the transitional benefit scheme has been a generous out-of-work welfare program targeted exclusively at lone mothers. A workfare reform of the transitional benefit scheme
2
Throughout this paper, we have included mothers who are cohabiting in the ‘married’ category.
2
was introduced on January 1, 1998. There were three important changes. First, work requirements were imposed, though only for lone mothers whose youngest child was three years old or more. Second, the age limit for eligibility on the youngest child was lowered, and time limits on welfare participation were introduced. And third, in-work benefit levels were raised. Table 1 provides more details of the transitional benefit scheme and the changes made in the 1998 reform. Table 1. Key features of the transitional benefit reform (€ – 1998) Characteristic
Before the reform
After the reform
Maximum benefit level
€ 695 per month
€ 855 per month
40 percent of earnings exceeding a
40 percent of earnings exceeding a threshold of €
threshold of € 215 per month
230 per month
None
If youngest child is 3 years old or more, the lone
Benefit reduction rate
Work requirements
parent has to work half time Time limit
None
Maximum 3 years of welfare receipt
Age limit
Youngest child less than 9–10
Youngest child less than 8 years old
years old (4th grade of primary school)
Means-testing of benefits depending on assets
None
None
As the aim of this paper is to evaluate the workfare aspects of the reform, we will focus attention on lone mothers whose youngest child is older than 3 years, since these were the parents faced with work requirements. Another reason for not focusing on lone mothers with younger children is that in 1998 the Norwegian government introduced a cash-for-care reform, which is a cash transfer to married and lone mothers with children aged one or two who did not make or only partly made use of government-subsidized day-care centres.3
3
To obtain consistent estimates of the transitional benefit reform on lone mothers with the youngest child less than 3 years of age, it would have been necessary to assume that the cash-for-care reform had the same impact on married and lone mothers with young children. Schøne (2003) and Naz (2003) find that the reform reduced employment among married mothers, in particular among those with high education. See also Havnes and
3
Economic impacts of the workfare reform Figure 1 draws the budget constraint before and after the welfare reform to illustrate how the work requirements and the increase in in-work benefits affected work incentives. Specifically, the figure shows how disposable income on the vertical axis varies with working hours per week on the horizontal axis; the earnings and welfare components (after tax) are above the zero-line, while the taxes and child care costs are below it. For brevity and with minimal loss of generality, we present only the work incentives for a lone mother with one child who has an hourly wage equal to 75 percent of the average wage in the labor force. Figure 1. Work incentives before and after the reform for a lone mother with one child 4-5 years of age Note: Figure 1 considers a lone mother with one child who has an hourly wage equal to 75 percent of the average wage in the labor force. The figure is based on an exact representation of the Norwegian tax-benefit system. The exceptions are social assistance and housing benefits, which are granted at the discretion of social security office staff supplementary to other social policies as last resorts of assistance. How these benefits vary with earnings is therefore imputed from the data. Children start school the year they turn six. To reflect child care costs, the figure assumes that the child is between 4 and 5 years of age. Primary school is compulsory and free of charge.
To assess the expected effects of the workfare reform, we begin with the usual static labor supply model assumptions: the woman can freely choose hours of work at the given offered wage, and offered wages are constant. In particular, we ignore any human capital, searchtheoretic, or related issues. We also assume that there is no time limit. Later we relax these assumptions. Consider first the case in which a lone mother prefers to work zero hours and therefore receives the maximum transitional benefit payment, when exposed to the pre-reform rules. Depending on the lone mother’s preferences, the reform could lead to either of two outcomes. First, she might continue not to work and receive zero transitional benefits. Second, she might enter the labor market; transitional benefit payment remains zero until she works half time. At this point, she satisfies the work requirement making her eligible for in-work benefits, generating a sharp jump in disposable income. Hence, the reform provides strong incentives for lone mothers to enter the labor market, and especially to work at least half time. Mogstad (2010a, 2010b) for a discussion of government subsidized child care arrangements in Norway, and their relationship to maternal labor supply and child development.
4
Consider next a lone mother who prefers to work positive hours and therefore receives less than the maximum transitional benefit payment, when exposed to the pre-reform rules. Depending on the preferred hours of work, the expected reform effects are qualitatively different. If she prefers to work less than half time, the substitution and income effects go in the same direction, inducing her to increase hours of work. In contrast, if she prefers to work half time or more, the income effect from the increase in in-work benefits reduces labor supply. Overall, the work requirements and the increase in in-work benefits imply a subsidy to part-time work. The reform effect on average earnings will depend on the size of the different responses, weighted by the relative numbers of lone mothers at different points along the budget constraint. The reform effect on disposable income depends on the extent to which the loss of out-of-work benefits is offset by gains in earnings and higher in-work benefits for part time workers. In particular, the disposable income will fall and the poverty rate rise among lone mothers facing insurmountable employment barriers, who after the reform lost eligibility to transitional benefit. Figure 2. Hours of work distribution of lone and married mothers with the youngest child 4-9 years of age Notes: Panel A shows the hours of work distribution in 1997 of lone and married mothers with the youngest child 4-9 years of age. Panel B shows the differences in population shares by hours of work between the lone mothers and married mothers in 2001, subtracted the differences in population shares by hours of work between the two groups in 1997. To compute the population shares of non-working mothers, we use administrative data on earnings for the entire population: zero hours of work is defined in accordance to our measure of labor force participation (see Section 4). To compute the distribution of hours of work among working mothers, we use information from the Wage Statistics Survey. This survey covers all employees in the public sector. For employees in the private sector, the data is based on an annual stratified random sampling of enterprises; all large enterprises, 40 percent of medium-sized enterprises, and less than 20 percent of small enterprises are sampled. Hours of work is defined as the contractual number of working hours per week, excluding meal breaks.
To get a perspective on the relative numbers of lone mothers at different points along the budget constraint, Panel A of Figure 2 draws the hours of work distributions of lone and married mothers. It should be noted, however, that the information on hours of work comes from survey data on an unrepresentative sample of women. With this caveat in mind, it seems that as much as 35 percent of all lone mothers were not participating in the labor market in 1997. However, we also see that that the vast majority of employed lone mothers were 5
working fairly long hours. In fact, as many as 88 percent of those who participated in the labor market were working at least half-time (19 hours per week), and almost 45 percent worked full-time (38 hours per week). In comparison, married mothers were participating in the labor market more often than lone mothers, although typically working fewer hours. Figure 1 does not reflect the introduction of welfare time limits and the reduction of the upper age limit for the youngest child to the welfare eligibility criteria. The long-term static effect of these measures is to eliminate welfare completely for certain lone mothers, which should increase labor supply for the same reasons that welfare decreases labor supply in the first place. In addition, there are some dynamic effects that unambiguously go in the same direction. First, one may expect lone mothers on welfare to anticipate the date when benefits will run out and begin to intensify their job search or even to accept job offers at an increasing rate when approaching this date.4 This implies that the time limits and the upper age limits do not have to be binding to affect the labor supply of welfare recipients. If there is uncertainty in terms of job opportunities or randomness in wage offers, one may want to accept an offer that is, in the short run, less attractive than staying on welfare even if it arrives in advance of the date when benefits will run out. Furthermore, the introduction of time limits should provide incentives for recipients who might need welfare in the future to delay the use of welfare benefits, or to leave welfare as rapidly as possible, in order to preserve future eligibility.5 The total impact of the change in time and age limits on disposable incomes and poverty depends on whether the earnings increases outweigh the loss of benefits. To get a sense of the likely responses to the workfare reform, we may compare the differences in the hours of work distributions of lone mothers and married mothers, before and after the reform. Panel B of Figure 2 displays the differences in population shares by
4
Moffitt (1985) and Røed and Zhang (2005) find this behavior for unemployment insurance recipients approaching the time their benefits will run out. 5 Grogger (2002), Grogger and Michalopoulos (2003) and Swann (2005) find that the introduction of time limits reduces welfare receipt substantially and that a significant part of this reduction occurs because recipients are forward-looking.
6
hours of work between lone mothers and married mothers in 2001, subtracted the differences in population shares by hours of work between two groups in 1997. In line with the predictions from labor supply theory, lone mothers appear to have increased their labor force participation but reduced part-time work with longer hours (31 hours per week), relative to married mothers. At the same time, the prevalence of both half-time work (19 hours) and full time work (38 hours) seem to have increased. The relatively large increase in full-time work is consistent with institutional constraints on hours of work choices (e.g. due to fixed costs of working). Phase-in provisions A final important feature of the transitional benefit reform is that phase-in provisions were introduced so that a subgroup of lone mothers who were entitled to and had applied for benefits by 1 January 1998 could continue to receive transitional benefits under the prereform rules. From 1 January 2001, benefits were paid exclusively according to the postreform rules. Note that women becoming lone mothers after 1 January 1998 were not entitled to the phase-in provisions. Table 2. Participation rates and average benefit amount for the transitional benefit scheme, 1993–2001 Participation rate
Average monthly transitional
on transitional benefits
benefit payment per recipient
Year
(%)
(€ – 1998)
1993
66
477
1994
65
469
1995
65
460
1996
65
465
1997
65
470
1998
66
524
1999
64
496
2000
61
492
2001
36
449
Notes: In each year, the sample consists of all lone mothers with the youngest child 4-9 years of age. The participation rate is defined as the fraction of the sample that receives transitional benefits. The transitional benefit amount is defined as the average monthly transitional benefit payment per recipient of transitional benefits.
7
Table 2 displays participation rates and average benefit payment for the transitional benefit scheme over the period 1993-2001. The time trends clearly mirror the fact that the reform was phased in gradually. The participation rates for lone mothers declined gradually after the reform in 1998, with a substantial drop in 2001 when the phase-in provisions were terminated. Further, the average benefit payment declined in 2001; this conforms to intuition, as lone mothers were then exposed to work requirements and benefit payments are reduced when earnings increase. In our estimations, we will pay close attention to the phase-in issue.
3. Empirical strategies This section outlines the empirical strategies that we use to evaluate the effects of the workfare reform. We will first discuss the main empirical strategy, before describing the methods used to investigate heterogeneity in the reform effects. Finally, we describe how our estimations deal with the phase-in issue. Main empirical strategy Our main empirical strategy is the following: We start by comparing the change in the outcome of interest from 1997 to 2001 for women (with children of the same age) who were either lone mothers or married mothers in both years. As discussed below, the reason for using 2001 as the last year of the comparison is that although the reform was undertaken in 1998, it was three years before the policy changes were fully implemented. Since married and lone mothers may react differently to time-specific factors, like changes in the labor market conditions, we add a comparison of the change in the outcome from 1993 to 1997 for women who were either lone mothers or married mothers in both years. This gives us a trend-adjusted DD estimator, exploiting that the reform creates variation along three dimensions: (a) between lone and married mothers; (b) between time periods before and after the reform (1997-2001 versus 1993-1997); (c) between first (1997 for the 1997-2001 period and 1993 for the 1993-
8
1997 period) and second years (2001 for the 1997-2001 period and 1997 for the 1993-1997 period). The trend-adjusted DD estimator can be defined as:
[
] [
]
(1) β = (Y01 − Y97 | LONE = 1) − (Y01 − Y97 | LONE = 0) − (Y97 − Y93 | LONE = 1) − (Y97 − Y93 | LONE = 0)
where Yt is the average outcome of interest in year t, and LONE is a binary assignment indicator equal to 1 if the woman is a lone mother in both years and 0 if she is a married mother in both years. The first bracket compares the time change in the average outcome for lone and married mothers in the time period after the reform. The second bracket makes the same comparison for the time period before the reform. Whereas the first bracket would be the standard DD estimator of the effect of the reform, the second bracket accounts for the possibility that lone and married mothers have some (unobserved) characteristics that make them react differently to time-specific factors. The identifying assumption is that the relative outcome of lone and married mothers would, on average, have changed in the same way in the period after the reform as in the period before the reform, in the absence of the reform. A trend-adjusted DD regression can be expressed as: (2) Yijt = α1 + α 2 REFORM j + α 3 SECONDt + a4 LONEij + a5 ( REFORM j × SECOND t ) + a6 ( REFORM j × LONEij ) + a7 ( SECONDt × LONEij ) + β ( REFORM j × SECOND t ×LONEij ) + θX ijt + ε ijt ,
where i indexes mother, t indexes year (1 = second and 0 = first), j indexes time period (1 if the period is 1997-2001, 0 if the period is 1993-1997), X is the local unemployment rate where the mother resides, and εijt is a composite residual consisting of an individual-specific fixed effect and a standard error term.6 The dummy variable SECOND is equal to 1 if the year is 2001 for the time period 1997-2001 or 1997 for the time period 1993-1997 (and 0 if the year is 1997 for the time period 1997-2001 or 1993 for the time period 1993-1997). The 6
Bertrand et al. (2004) show that the standard errors in DD regressions may be misstated in the presence of serial correlation of outcomes of the same mother over time. Although our fixed effects specification directly accounts for time-invariant unobserved heterogeneity, the fixed effects do not capture time-varying dependence. However, by limiting the time dimension to a few years we reduce the problem of serial correlation considerably.
9
dummy variable REFORM is equal to 1 if the time period is 1997-2001 (and 0 if the time period is 1993-1997). Equation (2) allows for different intercepts and time-specific effects for lone and married mothers. The reform effect is given by β, identified from the time change in the average outcome of lone mothers relative to married mothers, in the post-reform period relative to the pre-reform period. In comparison, the conventional DD regression imposes α2=α5=α6=β=0, and takes α7 to be the reform effect. Previous studies of welfare reforms have generally applied a DD regression to repeated cross-sectional data.7 As we have access to panel data, we improve on this by restricting the sample to the same lone and married mothers in the first and second year of each time period; this removes biases from comparison over time within each group due to unobserved compositional changes from the first to the second year. Heterogeneous effects As discussed above, theory predicts systematic heterogeneity in the responses to the reform. This raises two concerns with the empirical strategy discussed above. The first concern is that the estimated mean impacts may average together effects of different sign and magnitude. In particular, lone mothers with little or no labor market attachment are likely to be affected differently by the reform compared to those with stronger attachments to the labor market. On the one hand, to gain eligibility to benefits these mothers need to increase labor supply considerably. On the other hand, if they are unable to find work and thus lose eligibility, their disposable income will fall substantially. The most common way to address heterogeneous responses is to estimate mean impacts for subgroups thought to be likely to respond differently to the intervention. In line with this tradition, we will estimate equation (2) separately according to the characteristics of the lone mothers, such as educational attainment, labor market experience, and local labor market conditions.
7
A notable exception is Francesconi and van der Klaauw (2007).
10
The second concern is that the above empirical strategy may not necessarily produce a representative picture of the average effect for the entire population of lone mothers. In particular, equation (2) identifies the mean impact of the reform on women who had endured at least 4 years as lone mothers by 2001, hereafter labelled lasting lone mothers, and is silent on the responses of women with shorter spells as lone mothers. In the spirit of Heckman (1991), there are two possible explanations for why the responses to the reform are likely to vary systematically with the length of time spent as a lone mother. The first explanation is that as a consequence of experiencing a long spell as a lone mother, preferences or the budget constraint relevant to future choices (or outcomes) are altered. In this case, time as a lone mother has a genuine behavioral effect, in the sense that a woman who has experienced a short spell would behave differently in the future than an otherwise identical woman who has experienced a longer spell as a lone mother. There are a number of different mechanisms that might explain such a difference. One is that being the sole caregiver of a child may over time result in loss of motivation or depreciation of human capital, which may make one less likely to find work. A similar mechanism is at work if longer spells as a lone mother aggravate social exclusion or stigma, and thereby reduces labor market opportunities. A second explanation is that women may differ in certain characteristics that influence their probability of experiencing a long spell as a lone mother, but that are not influenced by the length of the spell in and of itself. In this case, time spent as a lone mother may affect the estimated reform effects because women who remarry (quickly) differ from those who stay single (longer). For example, in Becker’s seminal model of the marriage market, it is less likely that low-skilled women (re)marry because they are less attractive as spouses. Hence, the fraction of low-skilled lone mothers might be expected to rise with time spent in lone motherhood.
11
Both these explanations suggest that the responses to the reform vary systematically with the length of time spent as a lone mother. And if the differences in the responses are not captured by some observable characteristic, subsample estimation of equation (2) will fail to reveal the heterogeneous effects associated with time spent as a lone mother. To complement the subsample estimation of equation (2), we will therefore use the following empirical strategy: We start by comparing the change in the outcome of interest from 1997 to 1999 of mothers who are married in 1997, split up in 1998, and are lone mothers in 1999, to the change in the outcome from 1995 to 1997 of mothers with children of the same age who are married in 1995, split up in 1996, and are lone mothers in 1997. We will label these women as newly lone mothers, as opposed to the women who have endured at least 4 years as lone mothers when faced with the post-reform rules.8 Because this strategy samples from the flow of lone mothers, it avoids oversampling those with long spells as lone mothers (i.e. length-biased sampling). The reasons for considering the year immediately after the married mother splits up and becomes a lone mother, and not the year of change itself, are that we have annual data only on the outcomes, and that we want to allow the women some time to adjust to their new situation. In order to take into account time-specific factors, such as economic fluctuations, that might otherwise confound the estimates of the reform effects, we add a comparison of the change in the outcome of married mothers who stay married from 1997 to 1999 with that of married mothers who stay married from 1995 to 1997. A DD estimator of the reform effect on newly lone mothers can be defined as: (3) δ =
[(Y
99
− Y97 | NEW = 1) − (Y97 − Y95 | NEW = 1)] − [(Y99 − Y97 | NEW = 0 ) − (Y97 − Y95 | NEW = 0 )]
8
Another difference between newly and lasting lone mothers is that the former group consists only of lone mothers who were formerly married or cohabiting. Since models of the marriage market predict that it is less likely that low-skilled women marry, it is likely that lone mothers who were formerly married or cohabiting have higher education and earnings potential. It should be noted, however, that the vast majority of Norwegian lone mothers were formerly married or cohabiting (see e.g. Noack og Keilman, 1993), indicating that a large fraction of the lasting lone mothers were also formerly married or cohabiting. Unfortunately, our data do not allow us to identify the lasting lone mothers who were formerly married or cohabiting.
12
where NEW is a binary assignment indicator equal to 1 if a mother who is married in the first year of the comparison (1997 or 1995) makes the transition to lone mother before the last year of the comparison (1999 or 1997), and 0 if she stays married. The first bracket compares the time change in the average outcome of newly lone mothers in the time period before and after the reform. The second bracket makes the same comparison for married mothers staying married. The impact of the reform on newly lone mothers is then identified as the effect of becoming a lone mother after the reform, relative to the effect before the reform. The identifying assumption is that the relative outcome of married mothers making the transition to lone motherhood and those staying married would, on average, have changed in the same way in the period after the reform as in the period before the reform, in the absence of the reform. A DD regression of the reform effect on newly lone mothers can be expressed as: (4) Yijt = γ 1 + γ 2 REFORM j + γ 3 SECONDt + γ 4 NEWij + γ 5 ( REFORM j × SECOND t ) + γ 6 ( REFORM j × NEWij ) + γ 7 ( SECONDt × NEWij ) + δ ( REFORM j × SECOND t × NEWij ) + λX ijt + uijt ,
where i indexes mother, t indexes year (1 = second and 0 = first), and j indexes time period (1 if the period is 1997-1999, 0 if the period is 1995-1997), X is the local unemployment rate where the mother resides, and uijt is a composite residual consisting of an individual-specific fixed effect and a standard error term. Taking advantage of our panel data, in each time period we sample only mothers who are married in the first year and either stay married or make the transition to lone motherhood before the second year. The dummy variable SECOND is equal to 1 if the year is 1999 for the time period 1997-1999 or 1997 for the time period 1995-1997 (and 0 if the year is 1997 for the time period 1997-1999 or 1995 for the time period 19951997). The dummy variable REFORM is equal to 1 if the time period is 1997-1999 (and 0 if the time period is 1995-1997). Equation (4) allows for different intercepts and time-specific effects for married mothers making the transition into lone motherhood and those staying married. The reform effect is given by δ, identified from the time change in the average 13
outcome of the two groups, in the period after the reform relative to the period before the reform. Phase-in period A complicating issue in many program evaluations using a DD approach is that welfare reforms are seldom retroactive, so temporary provisions are often introduced during a phasein period. In this phase-in period, some or all welfare recipients may continue to receive benefits according to pre-reform rules or receive some form of compensation for inconveniences caused by the reform, which blurs the before and after distinction that forms the basis for the DD methods. While several past program evaluations employing the DD approach have simply ignored the potentially confounding effects of a gradual phase-in of reforms, Blundell et al. (2005) decided to exclude observations from a one-year phase-in period in their evaluation of the Working Families’ Tax Credit reform in the UK. In our case, phase-in provisions were introduced so that lone mothers who had applied for and were entitled to benefits before 1998 could continue to receive them under the prereform rules for up to three years. As suggested by Table 2, the phase-in period provides limited information about the incentive effects of the reform on lasting lone mothers. To circumvent the phase-in issue, we exclude the observations from the phase-in years 1998, 1999 and 2000 when estimating the reform effects on lasting lone mothers. In comparison, the empirical strategy for newly lone mothers has the advantage of avoiding the phase-in issue, as the newly lone mothers in 1999 will not be entitled to phase-in provisions.
4. Data and definitions The empirical analysis is based on administrative registers covering the entire resident population of Norway in the period 1993–2001. The register panel data set with household and demographic information is merged with detailed income data from the Tax Assessment Files through unique individual identifiers. The income data are collected from tax records 14
and other administrative registers rather than interviews and self-assessment methods. The coverage and reliability of Norwegian register data are considered to be exceptional, as is documented by the fact that the quality of such national data sets received the highest rating in a data quality survey in the Luxembourg Income Study database (Atkinson et al., 1995). Our empirical analysis focuses on lone mothers who are at least 18 years old and not more than 55, whose youngest child is between 4 and 9 years of age. In our baseline specification, our comparison group consists of married mothers who are between 18 and 55 years of age, whose youngest child is between 4 and 9 years of age. In one of the robustness checks, we will also consider single women in the same age range. Following closely the previous literature, students, self-employed, as well as individuals receiving permanent disability benefits are dropped from the analysis.9 To account for variations in local labor market conditions, we make use of data on local unemployment rates for 90 economic regions. Specifically, the economic regions constitute a regional level between country and municipality. The main criteria used for defining the economic regions are labor market, trade and service patterns, as well as commuting and internal migration patterns. Letting economic regions rather than municipalities form the basis for measuring unemployment rates may provide a better predictor of local labor market conditions. The dependent variables are defined as follows. Our measure of earnings is defined as the woman’s annual gross earnings. Disposable income is defined in close agreement with international recommendations (see Expert Group on Household Income Statistics, 2001) and incorporates the woman’s (but not her spouse’s) annual wages, capital income and all public cash transfers, less taxes. To evaluate the effects of the reform on annual gross earnings and disposable income, we use the consumer price index to make earnings and incomes from different periods comparable; throughout this paper the reference year is 1998, and €1 is set 9
Eissa and Liebman (1996) and Francesconi and van der Klaauw (2007) use similar sample selection criteria in their reform evaluations of lone-parent benefits.
15
equal to NOK 8.4. The fixed time-specific effects account for general income and earnings growth. In addition, we construct a measure of labor market participation based on the basic amount thresholds for earnings used by the Norwegian Social Insurance Scheme to determine labor market status (in order to determine eligibility for unemployment benefits, disability benefits, and old-age pension). Specifically, a mother is defined as working if her annual earnings exceed one basic amount (in 1998, about € 5,300). Our final outcome is poverty, in which case we follow common practice and define the annual poverty thresholds as 50 percent of the median in the distribution of household equivalent disposable income. To enable comparison of household disposable income between individuals belonging to households of varying size and composition, the OECD equivalence scale is applied; the weight of the first adult in the household is set to 1, each additional adult gets a weight of 0.7 and each child gets a weight equal to 0.5. In a given year, a woman is defined as poor if her household equivalent income is lower than the poverty threshold. The choices of poverty threshold and equivalence scale correspond to Norwegian official poverty statistics, as well as the 2002 Poverty White Paper (Ministry of Social Affairs, 2002). Descriptive statistics Figure 3 draws labor market participation and mean earnings by year, among lone mothers, married mothers, and single women without children. As expected, labor market participation and mean earnings are considerably lower for lone mothers compared to married mothers and single women. We also see a fairly good coherence between the time trends of lone and married mothers before the reform, whereas single women display a substantially different time trend in labor market participation. For this reason, married mothers appear to be the most suitable comparison group. Nevertheless, to take into account that the pre-reform trend in labor market participation is steeper for lone mothers, we will apply trend-adjusted DD
16
estimators. In this regard, it is reassuring that the pre-reform data establish a clear trend that can be “extrapolated” to the post-reform period. Figure 3. Labor force participation and mean earnings of lone mothers, married mothers, and single women Note: In each year, the sample consists of lone and married mothers with the youngest child between 4 and 9 years of age, as well as single women without children.
From Figure 3, we further see that the gap in labor market participation between lone and married mothers is reduced significantly after the reform, whereas the change in the relative earnings of the two groups is less distinct. This conforms to the theoretical predictions as the reform subsidizes part-time work; labor market participation should therefore unambiguously increase, whereas the mean impact on earnings could be averaging together positive effects on low-earning lone mothers and possibly negative effects on high-earning lone mothers.
Table 3 displays detailed descriptive statistics for the sample of lasting lone mothers, which forms the basis for the estimation of equation (2). We can see that lone mothers have substantially lower earnings and labor market participation than married mothers, but at the same time higher individual disposable income, simply because of the generous welfare system for lone parents. Yet lone mothers are more prone to poverty (based on equivalent household disposable income, in line with common practice and Norwegian official poverty statistics), since lone parent benefits do not appear to fully compensate for being in a singleearner rather than a dual-earner family. More importantly, it is evident from Table 3 that there are distinct differences before and after the reform in the changes over time in the outcomes of lone and married mothers. In our trend-adjusted DD framework laid out above, this is suggestive of substantial reform effects on lasting lone mothers (cf. Column 5). For example, the difference in labor market participation between married mothers and lasting lone mothers is considerably smaller in the second year than in the first year, in the period after the reform relative to the period before the reform; specifically, equation (1) suggests that the workfare
17
reform increased the labor market participation of lasting lone mothers by 4.3 percentage points. Table 3 also shows descriptive characteristics of married and lone mothers before and after the reform. As is clear from Table 3, married mothers are, on average, older and have more children, higher education, and more labor market experience than lasting lone mothers. However, we are not too concerned with differences in the characteristics of these two groups per se. Our primary concern is that there could be differences in how time-varying characteristics of lone and married mothers change over time, in the period after the reform relative to the period before the reform. It is therefore reassuring that none of the variables listed in Table 3 raises such concerns. Consistent with this, there are no significant reform effects when the time-varying characteristics are used as the dependent variable in the DD estimation (cf. Column 5). Table 3. Pre-reform and post-reform descriptive statistics of married mothers and lasting lone mothers –– Level –– Lone mothers Pre-reform First year (1993) Outcomes (mean) Earnings (€ – 1998) Labor market part. (percent) Disposable income (€ – 1998) Poverty (percent) Characteristics (mean) Age Years of schooling Labor market experience Non-western immigrant % Number of children Age of the youngest child Unemployment rate % Observations Pop. share of lone mothers
––––––––––––––– Difference ––––––––––––––– ––– DD ––– Lone mothers – Married mothers Pre-reform Post-reform First year Second year First year Second year Estimate (1993) (1997) (1997) (2001) (No controls)
9958 52,6 17676 4,7
-4101 -22,8 5123 2
-4220 -17,2 5447 2,8
-4523 -20,2 6222 3,3
-3514 -10,3 5720 7,6
1128*** 4,3*** -826*** 3,5***
29,9 11,1 7,7 2,6 1,6 3 4,3
-3,4 -1 -3,3 0,4 -0,7 0,2 0,1
-3,4 -1 -4,1 0,4 -0,7 0,3 0,1
-2,9 -1,1 -3,7 1 -0,6 0,4 0,1
-2,9 -1,1 -4,2 1 -0,6 0,4 0,1
0,0 0,0 0,3 0,0 0,0 -0,1 0,0
10,992
90,171 0,122
90,171 0,122
93,116 0,113
93,116 0,113
366,574 0,117
*** significant at 1% level, ** significant at 5% level, *significant at 10% level. Notes: Pre-reform sample consists of women who were either lone mothers or married mothers in both 1993 (first year) and 1997 (second year). Post-reform sample consists of women who were either lone mother or married mother in both 1997
18
(first year) and 2001 (second year). Level refers to means for lone mothers in first year of the pre-reform period. Difference refers to differences in means between lone and married mothers in first/second year in pre-reform/post-reform period. Labor market experience is defined as years of employment in previous years, according to the standard of the National Insurance Administration. DD estimate is based on equation (1).
Table 4 shows the same descriptive statistics as Table 3, except that newly lone mothers replace lasting lone mothers. Again, there are signs of substantial reform effects (cf. Column 5). For example, equation (3) suggests that the workfare reform increased labor market participation of newly lone mothers by 2.2 percentage points. It is also evident that married mothers are, on average, older and have slightly more children and somewhat higher labor market experience and education than the newly lone mothers. However, the time-varying characteristics of lone and married mothers change very little over time, in the period after the reform relative to the period before the reform. Consistent with this, there are no significant reform effects when the time-varying characteristics are used as the dependent variable in the DD estimation (cf. Column 5). When comparing Tables 3 and 4, it is clear that the lasting lone mothers are younger and have considerably lower education and less labor market experience than the newly lone mothers. This manifests itself in newly lone mothers having more similar labor market participation and earnings, compared to married mothers (cf. Column 2 displaying second year averages before the reform, but after the woman has become a lone mother). Table 4. Pre-reform and post-reform descriptive statistics of married mothers and newly lone mothers –– Level –– Lone mothers Pre-reform First year (1995) Outcomes (mean) Earnings (€ – 1998) Labor market part. (percent) Disposable income (€ – 1998) Poverty (percent) Characteristics (mean) Age Years of schooling Labor market experience
––––––––––––––– Difference ––––––––––––––– ––– DD ––– Lone mothers – Married mothers Pre-reform Post-reform First year Second year First year Second year Estimate (1995) (1997) (1997) (1999) (No controls)
16962 82,1 15762 2,3
972 1,9 1528 -0,1
368 -2,4 8530 2,3
486 1,8 225 0,7
289 -0,3 9674 2,2
407* 2,2** 2447*** -0,9
33 11,8 11
-2,3 -0,3 -1,5
-2,3 -0,3 -1,6
-2,2 -0,4 -1,8
-2,2 -0,4 -1,7
0,0 0,0 0,2
19
Non-western immigrant (%) Number of children Age of the youngest child Unemployment rate (%) Observations Pop. share of lone mothers
2,4 2,1 6,6 3,6
0 -0,3 -0,2 0
0 -0,3 -0,2 0
0,1 -0,2 -0,2 0,1
0,1 -0,2 -0,2 0
0,0 0,0 0,0 -0,1
2,276
83,186 0,027
83,186 0,027
87,402 0,027
87,402 0,027
341,176 0,027
*** significant at 1% level, ** significant at 5% level, *significant at 10% level. Notes: The pre-reform sample of lone mothers consists of women who were married in 1995 (first year), split up in 1996, and are lone mothers in 1997 (second year). The pre-reform sample of married mothers consists of mothers who stay married from 1995 (first year) to 1997 (second year). The post-reform sample of lone mothers consists of women who were married in 1997 (first year), split up in 1998, and are lone mothers in 1999 (second year). The pPost-reform sample of married mothers consists of mothers who stay married from 1997 (first year) to 1999 (second year). Level refers to the means of lone mothers in first year of the pre-reform period. Difference refers to differences between the means of lone and married mothers in first/second year in pre-reform/post-reform periods. Labor market experience is defined as years of employment in previous years, according to the standard of the National Insurance Administration. DD estimate is based on equation (3).
5. Empirical results This section investigates the responses of lone mothers to the reform. We first discuss our main results, before reporting estimates of the reform effects during the phase-in period and results from several robustness checks.
Main results Table 5 shows the estimated mean impact of the workfare reform on earnings, labor market participation, disposable income and poverty of lasting and newly lone mothers. As discussed above, the reform is expected to stimulate the earnings of lone mothers as long as the positive effects of the time and the age limits as well as the work requirements dominate the negative effect induced by the increase in in-work benefit levels of those working at least part-time. Indeed, Table 5 shows positive and significant reform effects on earnings and labor force participation. Specifically, the reform led to a 2.2 percentage points increase in the labor market participation of newly lone mothers and a € 406 increase in their earnings. In comparison, the reform is estimated to have had a substantially larger impact on lasting lone mothers. Specifically, the reform increased their earnings by € 1,116 and their labor market participation by 4.3 percentage points. To get a perspective on the magnitudes, these parameters estimates suggest that the workfare reform reduced the gaps between married
20
mothers and lasting lone mothers in labor force participation and earnings by around 25 percent. The differences in the labor market responses of newly and lasting lone mother are consistent with the theoretical predictions. Since the reform essentially subsidizes part-time work, the estimated mean impacts will depend on the relative number of lone mothers at different points along the budget constraint (as well as their elasticity). As the lasting lone mothers had relatively weak labor market attachment before the reform they will, on average, have stronger incentives to increase their labor supply in response to the policy changes, compared to newly lone mothers. By computing the weighted average of the estimated reform effects on lasting and new lone mothers, we find that the ‘aggregate’ mean effects are around € 985 for earnings and 3.9 percentage points for labor market participation. The relatively large overall effects reflect that there are nearly five times as many lasting lone mothers as newly lone mothers. In comparison to the evidence from the Anglo-Saxon countries, our results on labor market participation are fairly similar. For example, Francesconi and van der Klaauw (2007) estimate that the WFTC reform in the UK increased employment of lone mothers by around 5 percentage points, and Moffitt (2007) reports that employment increased by about 4 percent as a result of the US workfare reform in 1996. However, our estimated reform effects on earnings appear to be somewhat lower than what is typically found in studies of similar reforms in Canada, the UK, and the US (see e.g. Moffitt, 2007). Table 5. Estimated reform effects on newly and lasting lone mothers Panel A: Lasting lone mothers Dependent variable: Earnings (€ – 1998) Labor market part. (perc. points) Disposable income (€ – 1998) Poverty (perc. points)
Estimate 1116*** (128) 4.3*** (0.7) -844*** (114) 3.5*** (0.4)
Mean 13,872
Obs. 366,574
67.0
366,574
21,974
366,574
4.7
366,574
*** significant at 1% level, ** significant at 5% level, *significant at 10% level.
21
Panel B: Newly lone mothers Estimate 406* (207) 2.2** (1.0) 2447*** (205) -0.9 (0.7)
Mean 18,557
Obs. 341,176
82.1
341,176
25,053
341,176
4.3
341,176
Notes: Each column in each panel is a separate estimation. All estimations include individual-specific and time-specific fixed effects, as well as controls for local unemployment rate where the mother resides. Outcomes are defined in Section 4. The standard errors in parentheses are robust to heteroskedasticity and clustered at individual level. Mean refers to average outcome of lone mothers in the second year in the pre-reform period (1997). In Panel A, estimations are based on OLS on equation (2), with 1993-1997 as the pre-reform period and 1997-2001 as the post-reform period. In Panel B, estimations are based on OLS on equation (4), with 1995-1997 as the pre-reform period and 1997-1999 as the post-reform period.
Turning our attention to disposable income and poverty, the last two columns of Table 5 suggest a qualitative difference in the reform effects on newly and lasting lone mothers. On the one hand, the policy changes led to a substantial increase in the disposable income of newly lone mothers, in part because of higher earnings but also due to higher in-work benefit levels. On the other hand, the workfare reform caused a considerable decrease in disposable income and an economically significant increase in poverty among lasting lone mothers. The reason is that a sizeable group was unable to offset the loss of out-of-work benefits with gains in earnings, even though lone mothers were offered highly subsidized child care, after-school programs were widely available, and the economy was fairly strong. This suggests that the desired effects of the workfare reform on earnings and labor market attachment were associated with income loss and increased poverty among a subgroup of lasting lone mothers that face overwhelming employment barriers. The difference in the estimated reform effects on poverty and disposable income align well with the fact that lasting lone mothers have considerably less labor market experience and lower education than newly lone mothers. The weighted averages of the reform effects on lasting and newly lone mothers are around € -241 for income and 2.7 percentage points for poverty. In contrast, studies of workfare reforms in Anglo-Saxon countries suggest positive effects on income, and reductions in poverty.10 Table 6 investigates to what extent differences in their observable characteristics capture the differences in the reform effects between lasting and newly lone mothers. We start by estimating equation (2) separately according to characteristics of the lasting lone mothers, including educational attainment, labor market experience and local unemployment rates. The 10
See e.g. Blank and Schoeni (2003), who investigate changes in the distribution of family income in the US over the 1990s. Their findings suggest that the welfare reforms led to a rise in family income and a decline in poverty, which align well with Meyer and Sullivan’s (2004) study of the consumption pattern of lone mothers during this period.
22
subsample results reported in Columns 1-6 demonstrate that there is considerable heterogeneity in the responses of lasting lone mothers to the reform, in line with the theoretical predictions. Moreover, they indicate that at least some of the differential effects between newly and lasting lone mothers could be attributable to differences in observable characteristics. Specifically, the increases in labor market participation and earnings were strongest among lasting lone mothers with relatively low education or little labor market experience. At the same time, these subgroups of lone mothers experienced the largest income loss and rise in poverty incidence, most likely because a considerable proportion could not find work and thus lost benefits. And as expected, the reform was most successful in low unemployment areas where the rise in labor market participation and earnings were the largest and the income loss and increased poverty were the smallest. To overcome the curse of dimensionality that we would face if we were to use several characteristics to define subgroups, we use a standard survey weighting method to make
23
Table 6. Heterogeneity in the reform effects among lasting lone mothers Dependent variable: Earnings (€ – 1998) Labor market part. (perc. points) Disposable income (€ – 1998) Poverty (perc. points) Obs.
Characteristics (mean) Years of schooling Labor market experience Unemployment rate (%)
Newly lone mothers
Age Non-western immigrant (%) Number of children Age of the youngest child
(1) Low Educated
(2) High Educated
(3) Low labor market exp.
(4) High labor market exp.
1324*** (186) 5.3*** (1.1) -1228*** (143) 5.1*** (0.7) 118,594
885*** (180) 3.4*** (0.9) -560*** (177) 2.4*** (0.5) 247,980
1197*** (187) 4.2*** (1.2) -1133*** (197) 5.8*** (0.8) 97,592
672*** (179) 1.7** (0.8) -673*** (135) 2.0*** (0.4) 268,982
(5) Low local unempl. rate 1110*** (155) 4.7*** (0.9) -852*** (124) 3.5*** (0.5)
(6) High local unempl. rate 1205*** (301) 2.8 (1.7) -1234*** (243) 4.8*** (1.1)
(7) Re-weighted sample 975*** (181) 3.0*** (0.8) -661*** (165) 2.6*** (0.5)
274,594
91,980
369,574
–––––––––––––––––––––––––––––––––––––Lasting lone mothers –––––––––––––––––––––––––––––––––––––––– Low High Low labor High labor Low local High local Re-weighted Educated Educated market exp. market exp. unempl. rate unempl. rate sample
11,8
8,9
13,7
10,6
12,7
12,2
12
11,9
12,6
10,5
13,9
5,2
15,5
13
12,2
12,1
2,6
2,4
2,3
2,4
2,3
2,1
3,1
2,8
35
33,8
35,9
32,1
36,3
35,3
34,9
35,8
2,4
6,3
1,2
8,3
0,9
2,8
3
2,2
2,1
2,3
2,2
2,4
2,2
2,3
2,3
2,1
6,6
6,9
6,8
6,7
6,9
6,8
6,8
6,6
*** significant at 1% level, ** significant at 5% level, *significant at 10% level. Notes: In the upper panel, each cell reports a reform effect from a separate regression. The regressions are based on OLS on equation (2), with 1993-1997 as the pre-reform period and 1997-2001 as the postreform period. Columns 1-6 report subsample results: For each characteristic, the sample is split into two groups, above and below the median in the distribution of lone mothers in the first year in the prereform period. Column 7 estimates equation (2) on the re-weighted sample. All estimations include individual-specific and time-specific fixed effects, as well as controls for local unemployment rate where the mother resides. Outcomes are defined in Section 4. The standard errors in parentheses are robust to heteroskedasticity and clustered at individual level. The lower panel reports the characteristics in year 1997 of newly lone mothers, the subsamples of lasting lone mothers, and the re-weighted sample of lone mothers.
24
lasting lone mothers similar to the newly lone mothers in all the observable characteristics (see e.g. Yansaneh, 2005). The sampling weights are equal to the inverse probability of sampling a lasting lone mother (or a married mother) with a given set of characteristics (reported in Tables 3 and 4), in the period before and after the reform. The last column of Table 6 estimates equation (2) in the re-weighted sample of lasting lone mothers. The results show that the differential effects between newly and lasting lone mothers are reduced but far from eliminated by the re-weighting procedure. In particular, the evidence continues to suggest that the reform led to income loss and a rise in poverty incidence among lasting lone mothers. Phase-in issue As discussed in detail above, the workfare reform was gradually phased in, so that lone mothers who had applied for and were entitled to benefits before 1998 could continue to receive them under pre-reform rules for up to three years. While several past program evaluations have ignored the potentially confounding effects of a gradual phase-in of reforms, we have above followed Blundell et al. (2005) in discounting the observations from the phase-in years when estimating the reform effects on lasting lone mothers. In comparison, the empirical strategy for newly lone mothers sidesteps these issues, as they were not entitled to phase-in provisions if they split up after the reform. Table 7 reports estimates of reform effects on lasting lone mothers in 1998, 1999 and 2000, clearly suggesting that we would seriously underestimate the impact of the policy changes if we were to disregard or were unaware of the gradual phase-in of the reform. As expected, the reform effects on earnings and labour market participation increase as the end of the phase-in period (January 1, 2001) approaches. From a policy point of view, it is also interesting to see that there are no detrimental reform effects on poverty and disposable income among lasting lone mothers during the phase-in period, when they could self-select into the new system with higher in-work benefits or stay on the old-system without work 25
requirements. Table 7 also displays reform effects on newly lone mothers in the phase-in years, showing strong consistency over time in the estimates. This finding illustrates that it is unnecessary to wait until the phase-in period is over to evaluate the reform effects on newly lone mothers, which can be attractive from a policy perspective. Also, the consistency in the estimates over time increases our confidence in this empirical strategy. Table 7. Estimates of reform effects on newly and lasting lone mothers in the phase-in period Lasting lone mothers Last year in post-reform period Panel A: 1998 Earnings (€ – 1998)
Newly lone mothers
Estimate
Mean
Obs.
193*** (57)
15,810
425,732
Labor market part. (perc. points)
0.6* (0.3)
73.5
425,732
Disposable income (€ – 1998)
260 (309) -0.2 (0.2)
23,183
425,732
4.5
425,732
337*** (84) 1.3*** (0.5) 112 (78) -0.3
15,116
396,746
71.1
396,746
22,633
396,746
4.3
396,746
14,510
379,836
68.9
379,836
22,396
379,836
4.4
379,836
Poverty (perc. points) Panel B: 1999 Earnings (€ – 1998) Labor market part. (perc. points) Disposable income (€ – 1998)
Poverty (perc. points)
Estimate
Mean
Obs.
445** (217) 1.6 (1.0) 2,029*** (206) -1.1 (0.7)
18,557
338,360
82.1
338,360
25,053
338,360
4.3
338,360
406* (207) 2.2** (1.0) 2,447*** (205) -0.9 (0.7)
18,557
341,176
82.1
341,176
25,053
341,176
4.3
341,176
18,557
340,856
82.1
340,856
25,053
340,856
4.3
340,856
(0.3) Panel C: 2000 Earnings (€ – 1998) Labor market part. (perc. points) Disposable income (€ – 1998) Poverty (perc. points)
413*** (105) 2.4*** (0.6) -247* (130) 0.1 (0.3)
428** (207) 1.5* (0.9) 2,193*** (247) -0.1 (0.7)
*** significant at 1% level, ** significant at 5% level, *significant at 10% level. Notes: Each column in each panel for newly and lasting lone mothers is a separate estimation. All estimations include individualspecific and time-specific fixed effects, as well as controls for local unemployment rate where the mother resides. Outcomes are defined in Section 4. The standard errors in parentheses are robust to heteroskedasticity and clustered at individual level. Mean refers to average outcome of lone-mothers in the second year in the pre-reform period (1997). Lasting lone mothers: All estimations are based on OLS on equation (2). In Panel (A), 1996-1997 is the pre-reform period and 1997-1998 is the post-reform period. In Panel (B), 1995-1997 is the pre-reform period and 1997-1999 is the post-reform period. In Panel (C), 1994-1997 is the pre-reform period and 1997-2000 is the post-reform period. Newly lone mothers: All estimations are based on OLS on equation (4), with 1995-1997 as the pre-reform period. In Panel (A), 1996-1998 is the post-reform period. In Panel (B), 1997-1999 is the post-reform period. In Panel (C), 1998-2000 is the post-reform period.
26
Robustness analysis Table 8 investigates the robustness of our baseline results reported in Table 5, to changes in the specification of control variables. It is re-assuring to find that the results from these specification checks are qualitatively the same as the baseline results, and, moreover, that the point estimates are generally quite similar. Column 2 reports the estimates when dropping the control for local unemployment rates, and Column 1 reports our baseline results from Table 5 for comparison. In Column 3, we address the concern that other changes in the local environment may bias our reform estimates: For each municipality in every year, we have included detailed demographic controls for age and gender composition, local government spending, disposable income per capita, and the population share of full-time employed. Column 4 demonstrates that our results are robust to relaxing the linearity restriction in local unemployment rates, controlling for local unemployment rate and local unemployment rate squared. In Column 5, we also control for the interaction between local unemployment rate and lone mother status, as well as interactions between the local unemployment rate and timefixed effects. This specification check allows the effect of being a lone mother to differ depending on local unemployment rates, as well as the effect of economic shocks in the country to differ depending on local labor market conditions. Finally, Column 6 controls for interactions between the local unemployment rate and the individuals’ age, education, and labor market experience. This specification check allows unemployment rates to affect different groups differently. To further increase the confidence in our results, Table 9 performs several other specification checks. In Column 2, we drop the trend adjustment in the DD regressions (i.e. the second bracket of equations (1) and (3)). In Panel A, this specification check examines how our reform estimates are affected by assuming that married and lone mothers react in the same way to time-specific factors, like changes in macroeconomic conditions. In Panel B, this specification check corresponds to restricting the time-specific effects to be the same for lone 27
mothers before and after the reform. In both panels, we see that the results are qualitatively the same as the baseline results. However, the point estimates differ somewhat, in particular for the labor market participation of lasting lone mothers. The reason is that our baseline specification, unlike the standard DD approach, takes into account that the pre-reform trend in labor market participation is steeper for lone mothers than for married mothers. Table 8. Robustness analysis of changes in the set of control variables (1) Baseline specification
Dependent variable: Earnings (€ – 1998) Labor market part. (perc. points) Disposable income (€ – 1998) Poverty (perc. points) Dependent variable: Earnings (€ – 1998) Labor market part. (perc. points) Disposable income (€ – 1998) Poverty (perc. points)
1116*** (128) 4.3*** (0.7) -844*** (114) 3.5*** (0.4) 406* (207) 2.2** (1.0) 2447*** (205) -0.9 (0.7)
(2)
(3)
No Controls
More Controls
(4)
(5)
(6)
Quadratic unempl. rate
Lone mother unempl. interactions
Controls Unempl. Interactions
Panel A: Lasting lone mothers 1125*** 1128*** 1089*** 1113*** (129) (128) (128) (136) 4.2*** 4.3*** 4.3*** 4.4*** (0.7) (0.7) (0.7) (0.7) -832*** -826*** -880*** -898*** (114) (114) (116) (114) 3.5*** 3.5*** 3.5*** 3.8*** (0.4) (0.4) (0.4) (0.4) Panel B: Newly lone mothers 405* 407* 402* 124 (207) (208) (207) (220) 2.2** 2.2** 2.2** 1.1 (1.0) (1.0) (1.0) (1.1) 2448*** 2444*** 2135*** 2447*** (205) (205) (217) (205) -0.9 -0.9 -1.0 -0.9 (0.7) (0.7) (0.7) (0.7)
1059*** (125) 5.4*** (0.7) -896*** (114) 3.3*** (0.4) 294 (204) 2.0** (1.0) 2430*** (205) -0.9 (0.7)
*** significant at 1% level, ** significant at 5% level, *significant at 10% level. Notes: Each cell in every panel reports a reform effect from a separate regression. Column (1) repeats the results from Table 5. Column (2) drops the control for local unemployment rate. Column (3) adds municipality-year controls for age (5-year age groups) and gender composition, proportion of population full-time employed, mean disposable income, and mean local government expenditure. Column (4) adds a quadratic term in local unemployment rate. Column (5) interacts local unemployment rate with lone mother status and time fixed effects. Column (6) interacts local unemployment rate with age, education and labor market experience. The standard errors in parentheses are robust to heteroskedasticity and clustered at individual level.
In Column 3, we perform a triple-difference analysis, adding lone mothers with older children (age 10-14) as an additional comparison group, exploiting the fact that these were not affected by the reform. The comparison group of married mothers enables us to control for timevarying changes common to mothers with the youngest child aged 4-9 years, whereas the comparison group of lone mothers with older children is used to control for time-varying changes common to lone mothers. The fact that our results are robust to adding the comparison group of lone mothers with older children indicates that changes in, say, 28
macroeconomic conditions that affected married and single mothers differently, is an unlikely source of bias in our estimates. In Column 4, we use single women without children as the comparison group, instead of married mothers. It is heartening to find that the estimated reform effects are quite similar. Table 9. Specification checks (1)
(2)
(3)
(4)
(5)
(6)
Baseline specif.
No control for prereform trend
Additional comp. group: Lone mother, child 10-14
Alternative comp. group: Single women, no child
Pre-reform placebo test
Log dependent variable
Panel A: Lasting lone mothers 1421*** 1192*** (139) (128) 4.5*** 4.8*** (0.7) (0.7) -797*** -800*** (124) (113) 2.8*** 3.4*** (0.4) (0.4)
116 (95) 0.9 (0.6) 703 (717) 0.4 (0.3)
0.045** (0.021)
Panel B: Newly lone mothers 573*** 406* (211) (208) 2.2** 2.7*** (1.0) (1.0) 2287*** 2433*** (481) (205) -1.2* -0.9 (0.7) (0.7)
-107 (196) 0.2 (1.0) 346 (214) 1.2 (0.7)
Dependent variable: Earnings (€ – 1998)
1116*** 1,004*** (128) (99) Labor market part. (perc. points) 4.3*** 10.1*** (0.7) (0.5) Disposable income (€ – 1998) -844*** -530*** (114) (89) 3.5*** Poverty (perc. points) 4.3*** (0.4) (0.3) Dependent variable: Earnings (€ – 1998) Labor market part. (perc.points) Disposable income (€ – 1998) Poverty (perc. points)
406* (207) 2.2** (1.0) 2447*** (205) -0.9 (0.7)
636*** (244) 3.3*** (1.0) 1622*** (278) -0.3 (0.6)
-0.132*** (0.006)
0.040 (0.026)
0.061*** (0.017)
*** significant at 1% level, ** significant at 5% level, *significant at 10% level. Note: Each cell in every panel reports a reform effect from a separate regression. Column (1) repeats the results from Table 5. Column (2) of Panel A drops the observations of women who were lone and married mothers in 1993 and 1997 from the estimation of equation (2), thereby imposing α2=α5=α6=β=0 and taking α7 to be the reform effect (cf. first bracket, equation (1)). Column (2) of Panel A drops the observations of married mothers from the estimation of equation (4), thereby imposing γ4= γ6= γ7 = δ=0 and taking γ 5 to be the reform effect (cf. first bracket, equation (3)). Column (3) adds lone mothers with the youngest child 10-14 years of age as an additional comparison group. Column (4) uses single mothers without children as the comparison group, instead of married mothers. Column (5) of Panel A estimates equation (2) with 1993-1996 as the pre-reform period, and 1994-1997 as the post-reform period. Column (5) of Panel B estimates equation (4) with 1993-1995 as the pre-reform period, and 1995-1997 as the post-reform period. Column (6) measures the continuous dependent variables in logs rather than levels, excluding observations with zero earnings or zero income. The standard errors in parentheses are robust to heteroskedasticity and clustered at individual level.
As a further specification check, Column 5 pretends that the workfare reform took place before the actual implementation, and then estimates the effects of this placebo reform using data from the pre-reform period. If we find an effect of the actual timing of the reform, but no effect of the placebo reform, we will be more confident in our empirical strategies. Further, the sign of the effect of the placebo reform may give an indication on which direction our welfare reform estimates could be biased. To perform the placebo reform, we change the 29
definition of the dummy variables SECOND and REFORM in equations (2) and (4), so that only observations from the pre-reform period are used. We see that the estimates are insignificant and generally rather small. If anything, the sizeable positive point estimate for disposable income of lasting lone mothers indicates that the actual reform impact reported in Table 8 understates the reduction in disposable income. As pointed out by Athey and Imbens (2006), even if the identifying assumption in DD analysis is satisfied in levels, it does not imply that it is satisfied in logs, and vice versa. In Column 6, we therefore examine the sensitivity of our results to specifying the dependent variable in logarithmic form for our two continuous outcomes, earnings and income.11 It is reassuring to find that the results are qualitatively the same and quantitatively quite similar, when measuring earnings and income in logs rather than levels. Another concern is that the transition to lone motherhood (that is, inclusion in our treatment group) could be endogenous to the policy reform. The workfare reform may influence the incentives to become a lone mother in two conflicting ways. On the one hand, the work requirements and the time and age limits may have made it more costly to be a lone mother, thereby providing weaker incentives for divorce. On the other hand, the increase in in-work benefit levels may strengthen the incentives for divorce. Our DD approach controls for this insofar as the fixed effects or the observable time-varying covariates capture the source of endogeneity. To investigate this further, we have examined the time trend in the probability of becoming a lone mother for married mothers with the youngest child aged 1014 and married mothers with the youngest child aged 4-9. Only the latter group was affected by the reform. We find that there is a very good coherence between the time trends of the groups before the reform, and more importantly, it is heartening to find no change in the relative probability of lone motherhood at the time or after the reform. This conforms well to
11
When running the regression with the dependent variable logged, we exclude a substantial number of observations with zero earnings (and a handful of observations with zero income), as log of zero is not defined.
30
vast evidence from program evaluations carried out in the US, showing insignificant effects of welfare reforms on family composition (Moffitt, 2007).
6. Conclusion The generous Nordic model of welfare is commonly viewed as an exceptional success, in terms of equality, economic growth, and female labor force participation.12 However, it recently became evident that subgroups of the population with weak labor market attachment and high welfare dependency, such as lone mothers, were overrepresented among the poor. This prompted a major workfare reform of the Norwegian welfare system for lone mothers: work requirements were established, time limits imposed, and in-work benefit levels raised. In this paper, we have used a DD approach to produce the first evidence of the impacts of this workfare reform. A main finding is that the policy changes were successful in increasing the labor market participation and earnings of lone mothers. However, the desired effects of the workfare reform were associated with the side-effects of income loss and increased poverty among a subgroup of lone mothers who were unable to offset the loss of out-of-work benefits with gains in earnings. As most of what we know about the responses to such policy changes comes from Canada, the UK, and the US, evidence from a generous welfare state should be of particular interest. With regards to labor market participation and earnings, our results align well with the Anglo-Saxon experience. However, the finding of income loss and increased poverty among a subgroup of lone mothers with overwhelming employment barriers stands in contrast to the evidence from a similar workfare reform implemented in the US in 1996: In a survey of the research on this reform, Moffitt (2007; p31) concludes “that the 1996 welfare reform was a success, in overall terms and on average, is almost universally accepted by policy analysts 12
For example, the EU summit at Hampton Court in October 2005 on globalization and social models had an agenda that centered on the "Nordic model". The International Herald Tribune in an article in September 2005 says, "European leaders want to know how Sweden and its Nordic neighbors, so heavily laden with cradle-tograve welfare systems, float high above the struggling economies of much of the rest of the Continent."
31
and researchers”. Whether the presence of detrimental effects of workfare reforms in Norway but not in the US is related to cross-country differences in preferences, institutional factors like labor market regulations and unions, or the design of the tax system is an open question.13 However, the disparity in the responses of lone mothers to workfare reforms across two of Esping-Andersen’s (1990) highly differentiated worlds of welfare capitalism underscores that policymakers in other developed countries should be cautious when drawing lessons from the Anglo-Saxon experience.
7. References Alesina, A., E. Glaeser, and B. Sacerdote (2005): “Work and Leisure in the US and Europe: Why so Different?”, in M. Gertler and K. Rogoff (eds.) NBER Macroeconomics Annual, Cambridge, MA: MIT Press. Athey, S. and G. Imbens (2006): "Identification and Inference in Nonlinear Difference-InDifference Models". Econometrica, 74, 431–498. Atkinson, A., L. Rainwater and T. Smeeding (1995): Income Distribution in OECD Countries, Paris, OECD. Bertrand, M., E. Duflo and S. Mullainathan (2004): “How Much Should We Trust Differences-in-Differences Estimates?”, Quarterly Journal of Economics, 119, 249–75. Bitler, M., J. Gelbach, and H. Hoyes (2008): “Distributional Impacts of the Self-Sufficiency Project”, Journal of Public Economics, 2, 748-765. Blanchard, O. (2004): “The Economic Future of Europe”, NBER Working Paper, 10310. Blank, R. (2002): “US Welfare Reform: What’s Relevant for Europe?”, CESifo Working Paper, 753.
13
Whereas Blanchard (2004) emphasises the preference argument for why labor supply is lower in Europe than in the US, Olovsson (2009) argues that differences in taxes can account for the discrepancy and Alesina et al. (2005) stress the role of labor regulations and unions.
32
Blank, R. and R. Schoeni (2003): “Changes in the Distribution of Children’s Family Income over the 1990’s”, American Economic Review Papers and Proceedings, 304-308. Blundell, R., M. Brewer and A. Shephard (2005): “Evaluating the Labour Market Impact of Working Families’ Tax Credit Using Difference-in-Differences”, Working Paper, Institute for Fiscal Studies. Brewer, M., and P. Gregg (2001): ”Lone Parents, the WFTC and Employment in Households with Children”, in R. Dickens, J. Wadsworth and P. Gregg (eds), The State of Working Britain, London, LSE Centre for Economic Performance. Brewer, M., M. Francesconi, P. Gregg, and J. Grogger (2009): “Feature: In-Work Benefit Reform in a Cross-National Perspective – Introduction”, Economic Journal, 119, 1-14. Card, D. and D. Hyslop (2006): ”The Dynamic Effects of an Earnings Subsidy for Long-Term Welfare Recipients: Evidence from the SSP Applicant Experiment”, NBER Working Paper, 12774. Eissa, N., and J. Liebman (1996): ”Labor Supply Response to the Earned Income Tax Credit”, Quarterly Journal of Economics, 111, 605–37. Esping-Andersen, G. (1990): The Three Worlds of Welfare Capitalism, Princeton, Princeton University Press. Expert Group on Household Income Statistics (2001): Final Report and Recommendations, Ottawa. Francesconi, M., and W. van der Klaauw (2007): ”The Socioeconomic Consequences of “InWork” Benefit Reform for British Lone mothers”, Journal of Human Resources, 17, 1–31 Grogger, J. (2002): ”The Behavioral Effects of Welfare Time Limits”, American Economic Review, 98, 385–9. Grogger, J., and C. Michalopoulos (2003): ”Welfare Dynamics under Time Limits”, Journal of Political Economy, 111, 530-554.
33
Havnes, T. and M. Mogstad (2010a): “No Child Left Behind: Subsidized Child Care and Children’s Long-Run Outcomes” forthcoming American Economic Journal: Economic Policy Havnes, T. and M. Mogstad (2010b): “Money for Nothing? Universal Child Care and Maternal Employment”, IZA Discussion Paper, 4978. Heckman, J. (1991): “Identifying the hand of the past: distinguishing state dependence from heterogeneity”, American Economic Review, 81, 71–79. Meyer, B., and D. Rosenbaum (2001): ”Welfare, the Earned Income Tax Credit, and the Labor Supply of Single Mothers”, Quarterly Journal of Economics, 117, 1063–114. Meyer, B. and D. Sullivan (2004): ”The Effects of Welfare and Tax Reforms: The Material Well-Being of Single Mothers in the 1980s and the 1990s”, Journal of Public Economics, 88, 1387-1420. Michalopoulos, C., P. K. Robins and D. Card (2005): ”When financial work incentives pay for themselves: evidence from a randomised social experiment for welfare recipients”, Journal of Public Economics, 89, 5-29. Ministry of Social Affairs (2002): ”Plan of Action for Combating Poverty”, Poverty White Paper, Norwegian Ministry of Social Affairs. Moffitt, R. (1985): ”Unemployment Insurance and the Distribution of Unemployment Spells”, Journal of Econometrics, 28, 85–101. Moffitt, R. (2007): ”Welfare Reform: The US Experience”, forthcoming in Swedish Economic Review. Naz, G. (2004): ”The Impact of Cash-Benefit Reform on Parents’ Labour Force Participation”, Journal of Population Economics, 17, 369–83. Noack, T. and N. Keilman (1993): ”Familie og Husholdning”, [Families and Households] Statistical Analyses, 1, Statistics Norway.
34
Olovsson, C. (2009): “Why Europeans Work so Little?”, International Economic Review, 50, 39-61. Røed, K., and T. Zhang (2005): ”Unemployment Duration and Economic Incentives: A Quasi-Random Assignment Approach”, European Economic Review, 49, 1799–825. Schøne, P. (2003): ”Labor Supply Response to a Cash-for-Care Reform”, Journal of Population Economics, 17, 703–27. Swann, C. (2005): ”Welfare Reform When Recipients Are Forward-Looking”, Journal of Human Resources, 40, 31–56. Yansaneh, I. S. (2005): ”Construction and Use of Sample Weights”, in United Nation Statistics Division (ed.), Designing Household Surveys Samples: Practical Guidelines, New York, United Nation.
35
