Phasing out payroll tax subsidies

Herget, Anna; Riphahn, Regina T. Article — Published Version Phasing out payroll tax subsidies International Tax and Public Finance Provided in Cooperation with: Springer Nature Suggested Citation: Herget, Anna; Riphahn, Regina T. (2025) : Phasing out payroll tax subsidies, International Tax and Public Finance, ISSN 1573-6970, Springer US, New York, NY, Vol. 32, Iss. 5, pp. 1501-1531, https://doi.org/10.1007/s10797-024-09879-0 This Version is available at: https://hdl.handle.net/10419/330619 Standard-Nutzungsbedingungen: Die Dokumente auf EconStor dürfen zu eigenen wissenschaftlichen Zwecken und zum Privatgebrauch gespeichert und kopiert werden. Sie dürfen die Dokumente nicht für öffentliche oder kommerzielle Zwecke vervielfältigen, öffentlich ausstellen, öffentlich zugänglich machen, vertreiben oder anderweitig nutzen. Sofern die Verfasser die Dokumente unter Open-Content-Lizenzen (insbesondere CC-Lizenzen) zur Verfügung gestellt haben sollten, gelten abweichend von diesen Nutzungsbedingungen die in der dort genannten Lizenz gewährten Nutzungsrechte. Terms of use: Documents in EconStor may be saved and copied for your personal and scholarly purposes. You are not to copy documents for public or commercial purposes, to exhibit the documents publicly, to make them publicly available on the internet, or to distribute or otherwise use the documents in public. If the documents have been made available under an Open Content Licence (especially Creative Commons Licences), you may exercise further usage rights as specified in the indicated licence. https://creativecommons.org/licenses/by/4.0/ Vol.:(0123456789) International Tax and Public Finance (2025) 32:1501–1531 https://doi.org/10.1007/s10797-024-09879-0 Phasing outpayroll tax subsidies AnnaHerget1· ReginaT.Riphahn1 Accepted: 24 December 2024 / Published online: 12 February 2025 © The Author(s) 2025 Abstract Many countries subsidize low-income employments or small jobs. These subsidies and their phasing out can generate labor market frictions and distort incentives. The German Minijob program subsidizes low-income jobs. It generates a ‘Minijob trap’ with substantial bunching along the earnings distribution. Since 2003, the newly introduced Midijob subsidy aims to reduce the Minijob-induced notch in the net earnings distribution. Midijobs reduce payroll taxes for employments above the Minijob earnings ceiling. We investigate whether introducing Midijobs reduced the Minijob trap. We apply a regression discontinuity design using administrative data and a difference-in-differences estimation using survey data. While in both cases our results show a small positive overall effect of Midijobs on transitions out of Minijobs, they are effective only for a narrow treatment group. Keywords Midijobs· Minijobs· Payroll tax subsidy· Causal effects· Difference-indifferences· Regression discontinuity· SOEP· SIAB JEL Classification J21· J38· H24 1 Introduction In recent decades, many countries have shifted minimum income protection from traditional means-tested transfers to make work pay policies to incentivize labor supply.1 Make work pay policies condition on employment and either provide subsidies to employers or to employees via negative income taxes or payroll tax subsidies. Well-known examples of such policies are the U.S. Earned Income Tax Credit (EITC) and the U.K. Working Family Tax Credit (WFTC). Numerous contributions * Regina T. Riphahn [email protected] 1 Friedrich-Alexander-University Erlangen-Nürnberg, Lange Gasse 20, 90403Nuremberg, Germany 1 See, e.g., Immervoll etal. (2007) or Jara etal. (2020). 1502 A.Herget, R.T.Riphahn evaluate the labor supply effects of these programs.2 This paper contributes to the make-work pay literature: we study the effect of a new program (Midijobs) aimed at phasing out a payroll tax subsidy (Minijobs) in Germany. The Midijob program was introduced to incentivize labor supply and earnings increases for workers with very low earnings. This is a challenge in many labor markets. The Midijob program was implemented on April 1, 2003. It reduces employee social insurance contributions for jobs within the Midijob earnings range, originally 400 to 800 Euro per month. It was introduced to address a labor supply disincentive generated by Minijobs; Minijobs are low-income employments paying up to 400 Euro per month at the time. They cause substantial bunching at the Minijob earnings limit (Gudgeon & Trenkle, 2024; Tazhitdinova, 2020) because Minijob employees pay neither income taxes nor social insurance contributions. Taxes and contributions fall due on total earnings when earnings exceed the Minijob earnings threshold; Tazhitdinova (2020) pointed out that an average woman’s combined tax and social insurance liabilities when earning 1 Euro beyond the pre 2003 Minijob earnings threshold reached 45 percent of her total gross income. Therefore, workers rarely extend their earnings beyond this limit which generates the so-called ‘Minijob trap’. Figure1 illustrates the uneven number of jobs across the earnings distribution. Midijobs were designed to address the Minijob trap and we investigate whether the Midijob program succeeds in this respect. Both programs, Minijobs and Midijobs are used extensively. At the end of 2021, 4.1 million individuals used Minijobs as their main employment and about 3 million workers were employed in Midijobs (see Appendix FigureA.2).3 Thus, the programs jointly covered 15.3 percent or one-sixth of the German labor force. Given the magnitude of the Midijob program, its expansions to earnings limits of 1,300 (in 2019), 1600 (in 2022), and even 2000 Euro in 2023, and the lack of prior evidence, the Minijob trap addressed in this paper is important. We use two complementary empirical strategies. First, we exploit the introduction of Midijobs on April 1, 2003, to evaluate their immediate effects on the propensity to transition from a Minijob to higher-earning, regular employment. For this purpose, we apply a regression discontinuity in time (RDiT) design to administrative data and investigate whether there is a jump in the transition probability at the cutoff date. However, this approach may suffer from shortcomings, e.g., by capturing only local effects. Our second approach exploits a unique feature of income tax regulations. Specifically, Midijobs reduced workers’ social insurance contribution rates in the earnings range just above the Minijob earnings limit leaving income tax rates untouched. This attenuates the kink in the net earnings schedule, particularly for individuals who pay low or no income taxes (e.g., single individuals). In 3 Another 3.1 million individuals use Minijobs as secondary employment. As secondary job holding is not in the focus of the Midijob regulation we do not discuss it further (for a detailed analysis see, e.g., Tazhitdinova 2022). − The utilization of Minijobs declined slightly in 2015 with the introduction of mandatory minimum wages and again during the Covid pandemic. FigureA.2 clearly shows that more employees were covered after the Midijob earnings ceiling increased on July 1, 2019 from 850 to 1300 Euro per month. 2 For a recent survey see Neumark and Shirley (2020), also Bastian and Lochner (2022), Bastian (2020), Hoynes and Patel (2018), Azmat (2014), Chetty and Saez (2013), Chetty etal. (2013), Dahl etal. (2009), Francesconi and van der Klaauw (2007), Eissa and Hoynes (2004), and Eissa and Liebman (1996). 1503 Phasing outpayroll tax subsidies contrast, Midijobs hardly affect the net earnings schedule of individuals who at the kink additionally become subject to high income tax rates (e.g., secondary earners in marriages) because for them income taxes dominate social insurance contributions. We consider the former group to be treated by the reform whereas the latter group is our control group. We use survey data from the German Socioeconomic Panel (SOEP) which offers information on marital status and apply a difference-indifferences (DID) analysis to study the propensity to transition from a Minijob to higher-earning, regular employment. The DID analysis yields an average treatment effect on the treated which captures more sluggish labor supply responses than the discontinuity design. We find that while the program works on average, it does not work for all. The first strategy (RDiT) yields small increases of at most 15 percent in average monthly transition rates out of Minijobs into higher paying, regular employment at the time of the Midijob introduction. However, these changes are concentrated in the small group of male Minijob holders. The second strategy (DID) confirms these patterns: the reform effect on annual transitions out of Minijobs is significantly larger in the treatment than in the control group. Non-married individuals with lowincome tax burdens respond more strongly to the reform than secondary earners in marriages—typically females—who are subject to high-income tax rates. The results of both identification strategies are robust to various specification changes and sample adjustments. Overall, Midijobs were introduced to phase out Minijob subsidies and to reduce bunching at the Minijob earnings threshold. This objective was missed for most Minijob employees. The Minijob trap did not disappear after the introduction of Midijobs. These findings have a clear takeaway: in countries with household taxation, phase-out policies from Minijob-type notches must account for the entire tax domain (including the income tax-induced portion of the notch) to be effective. These findings complement the results of Tazhitdinova (2020) and Gudgeon and Trenkle (2024). Both papers focus on Minijob employment and study labor supply elasticities at the Minijob earnings ceiling. Tazhitdinova (2020) uses bunching approaches to determine annual labor supply elasticities between 1999 and 2010. She finds larger labor supply elasticities for men than for women and strong increases in labor supply elasticities for single individuals after 2003. Gudgeon and Trenkle (2024) focus on a sample of married women only and study the responsiveness of labor supply to shifts in the Minijob earnings ceilings in 2003 and 2013 over time. The authors find substantial delays in earnings responses and argue that labor demand frictions attenuate estimates of intensive margin labor supply elasticities. We add to these contributions by focusing on the effect of Midijobs; they may be a mechanism to phase out Minijob subsidies and to facilitate transitions to regular employment. Our research connects to several strands of the literature. First, we contribute to the international literature on the effectiveness of making work pay policies such as the EITC in the United States or the United Kingdom tax credit programs. Numerous studies investigate whether the subsidy programs enable beneficiaries to grow 1504 A.Herget, R.T.Riphahn Fig. 1 Distribution of gross monthly earnings (April 2003–December 2012). Note: The graph represents the bottom part of the gross earnings distribution for all employed individuals registered with the unemployment insurance. Using all jobs observed between April 2003 and December 2012 it shows the number of employees per 50 Euro bin of monthly gross earnings. For depictions of annual gross earnings distributions, see Appendix Fig.A.1. Source: SIAB, own calculations Fig. 2 Net earnings with and without Midijob subsidy by income tax burden (2003). Note: The graph sketches net monthly earnings along the development of monthly gross earnings. Up to gross earnings of 400 Euro per month Minijobs eliminate any difference between gross and net warnings. Beyond the Minijob earnings threshold the red (blue) lines indicate the situation before (after) the introduction of Midijobs. The dashed lines assume an average income tax rate of 30 percent whereas the straight lines assume zero income taxes. Source: Own illustration 1505 Phasing outpayroll tax subsidies out of their need for support by expanding their labor supply.4 Only few contributions address the role of benefit phase-out. Eissa and Liebman (1996) study benefit phase-out in the EITC. Interestingly, they find single mothers to be rather unresponsive to increasing marginal tax rates in the phase-out region of the program.5 Leigh (2007) investigates the role of phase-out rates using the 1999 reform of the United Kingdom’s tax credit. He confirms that lower phase-out rates have positive impacts on labor supply. Comparing the effects of the U.S. EITC and the German Minijob program Berthold and Coban (2013) conclude that in contrast to the EITC, the German program was ineffective in supporting low-income earners. However, so far, it is still unresolved whether the phasing out of the Minijob subsidy by means of Midijobs succeeds in encouraging higher-earning employment. We are the first to address this issue. Second, we contribute to international research on the employment effects of payroll tax subsidies. Most studies investigate the extensive margin of labor supply and find no employment effects in response to changes in payroll taxes. Saez etal., (2019, p.1) argue that it is “received wisdom” that the payroll tax incidence falls on workers’ net wages.6 We add to this literature by focusing on the intensive margin of labor supply. A third branch of studies investigates the stepping stone character of Minijobs themselves. The 2003 reform rendered Minijob employment more attractive to incentivize the labor market entry of those previously not employed: lawmakers hoped for Miniand Midijobs to become stepping stones into regular employment. Several studies evaluate the programs and conclude that it is unlikely that the programs act as a stepping stone.7 The literature on Midijobs is limited and largely descriptive. Most studies explain the Midijob instrument and its utilization.8 Bach etal., (2018a, 2018b) conclude that the 2019 expansion of the Midijob earnings ceiling from 850 to 1,300 Euro per month may even worsen the part-time employment trap. Our analyses contribute to the literature in several ways. First, while the literature on payroll tax subsidies focuses on the extensive margin of labor force participation, 6 This is broadly supported in the literature, see e.g., Gruber (1997), Anderson and Meyer (1997, 2000), Korkeamäki and Uusitalo (2009), Huttunen etal. (2013), and Bennmarker etal. (2009) who provide evidence from Chile, the United States, Finland, and Sweden, respectively. 7 See e.g., Fertig etal. (2005), Fertig and Kluve (2006), and Freier and Steiner (2008) or more recently Caliendo etal. (2016), Lietzmann etal. (2017), and Carrillo-Tudela etal. (2021). 8 See, e.g., Fertig and Kluve (2006), Brandt (2005, 2006), Herzog-Stein and Sesselmeier (2012), Berthold and Coban (2013), Fichtl (2015), Keller and Seifert (2015), Seifert (2017), Dundler et al. (2019), Keller etal. (2021), and Herget and Riphahn (2022). 5 Browning (1995) calculates potential negative income effects induced by high marginal taxes in the phase-out region of the EITC. However, empirical studies did not support this rationale (see also Meyer 2002). LaLumia (2009) investigates the reporting behavior of self-employed in response to the incentives of the EITC program and confirms that the phase-out region generates less of a response than the phase-in region. Trampe (2007) summarizes the literature which hardly found negative effects of the EITC phase-out region. He finds small negative effects (however, for a discussion see Hoynes 2007 and Trampe 2008). 4 See, e.g., Bargain and Orsini (2006), Bargain etal. (2010), Blundell (2000), Francesconi and van der Klaauw (2007), or Grogger (2003). 1506 A.Herget, R.T.Riphahn we study the intensive margin of labor supply. Small jobs restrict overall labor supply, limit human capital investments and career prospects (Beckmann, 2020), and inhibit the accumulation of pension claims. Therefore, it is important to understand mechanisms that support transitions to regular employment. Second, the literatures on ‘making work pay’ and on the stepping stone character of small jobs have not yet addressed the relevance of the phasing out of payroll tax subsidy programs. If the Midijob program effectively supports transitions to extended labor supply it could constitute a useful policy for other (national) labor markets, as well.9 Finally, we are the first to study the effect of the introduction of Midijobs on the propensity to leave Minijob employment, i.e., the effectiveness of Midijobs as a labor market policy. The structure of this paper is as follows. In section two, we provide institutional detail on the Miniand Midijob programs and their development over time. We describe our empirical analysis based on the regulatory discontinuity in Sect.3 and our difference-in-differences analyses in Sect.4. In Sect.5 we offer a policy discussion and conclude. 2 Background 2.1 Institutions andreforms Minijob employees, i.e., those earning no more than the Minijob earnings threshold are exempt from otherwise mandatory social insurance contributions and income taxes. Instead, their employers pay a fixed share of gross Minijob earnings to social insurance and tax authorities (for details see, e.g., Collischon etal., 2021).10 This regulation has existed since the early days of the German social insurance system (1893) to limit the bureaucratic burden for small jobs (BMAS 2018, p.110). Over time, Minijob regulations were modified with varying objectives. At the Minijob earnings ceiling, the sudden fade-out of the wage subsidy inherent in Minijob employment generates a large, discontinuous change in tax and social insurance liabilities. When they earn below the Minijob earnings ceiling employees pay neither social insurance contributions nor income taxes. When they earn above the Minijob earnings ceiling, income taxes on total earnings plus social insurance contributions of about 20 percent become payable (see Fig.2). This leads to a (large) reduction in net monthly earnings. While social insurance contributions are relevant for all employees, income taxes differ for joint and individual filers; in Sect.4, our identification strategy takes advantage of this heterogeneity. Those filing individually benefit from a sizable initial tax allowance and do not face income taxes immediately after exceeding the earnings threshold. In contrast, joint filers, i.e., secondary 9 Di Porto etal. (2022), Dolado etal. (2021), and Scarfe (2021) recently studied zero-hours contracts and casual work in Italy, the United Kingdom, and Australia, respectively. Jobs based on these contracts are similar to German Minijobs. 10 Minijobs also take the form of short-term employment relationships which do not extend beyond (currently) 70days, independent of earnings. We disregard this second category of Minijob employment, which is much less prevalent. 1507 Phasing outpayroll tax subsidies earners in marriages, may be affected by sizeable tax rates on their entire earnings as soon as they exceed the Minijob earnings threshold. Thus, the Minijob earnings threshold generates a discontinuity in the level (a notch) and the slope (a kink) of the net earnings schedule (Kleven, 2016). This “Minijob trap” bars increases in labor supply and earnings and causes substantial bunching in the earnings distribution (see Fig.1) (Gudgeon & Trenkle, 2024; Tazhitdinova, 2020). The reforms implemented on April 1, 2003 raised the monthly Minijob earnings ceiling from 325 to 400 Euro, abolished a limit of 15 working hours per week, increased employer contribution rates from 22 to 25 percent of Minijob earnings, and–most interesting for us–newly introduced the Midijob program.11 The intention of the 2003 reform was (a) to reduce illicit moonlighting by making legal small jobs more attractive and (b) to offer stepping-stone employment opportunities for the unemployed and opportunities for upward mobility for those in marginal employment (Eichhorst etal., 2012). The introduction of Midijobs did not affect employers who continued to pay regular social insurance contributions of about 20 percent on earnings beyond the Minijob earnings ceiling. Midijobs were introduced to incentivize labor supply beyond the Minijob earnings ceiling; Midijobs entail payroll tax subsidies for employees earning between (then) 400 and 800 Euro per month. These subsidies phase out as earnings increase. Instead of full regular social insurance contribution rates of 20 percent, the Midijob rates increased on a sliding scale starting at about 4 percent for monthly earnings of 400 Euro and reaching the unsubsidized level of 20 percent for monthly earnings of 800 Euro. At the same time, Midijob employees are fully liable for income taxes on their total earnings (see Fig.2).12 In 2013, the monthly Minijob and Midijob earnings limits were raised to 450 and 850 Euro, respectively (for later adjustments see Appendix TableA.1). After this reform, social insurance contribution rates for Midijobs commenced at about 10 percent for monthly earnings starting at 451 Euro and increased to 20 percent at monthly earnings of 850 Euro. Minijobs and Midijobs have been used intensely. The introduction of the Midijob subsidy of social insurance contribution rates was intended to attenuate the notch in the net earnings distribution, phase out the Minijob subsidy, and incentivize labor supply and earnings beyond the Minijob earnings ceiling. We investigate whether Midijobs effectively reduced the barriers to exiting subsidized Minijob employment and entering higher-earning employment. 11 The relevant legislation (Zweites Gesetz für Moderne Dienstleistungen am Arbeitsmarkt, Hartz II) was passed on December 23, 2002 as an early element of a bundle of labor market reforms. For a compact review of the reforms see, e.g., Carrillo-Tudela etal. (2021). 12 FigureA.3 compares the implicit Minijob and Midijob subsidy to the patterns of the US EITC. FigureA.3.1 shows the situation before the 2003 reform: Minijob subsidies accrued in the 0-325 Euro earnings range. They increased with household tax rates and terminated abruptly at the Minijob earnings limit; depending on the tax regime they amount to about 50 percent of gross earnings. FigureA.3.2 shows the monthly credit after the 2003 reform. Here, the Midijob subsidy for earnings beyond the Minijob earnings limit (now 400 Euro) is introduced which is independent of tax rates. It starts at about 16 percent of gross earnings (about 64 Euro) and declines along the Midijob earnings range. Figure A.3.3 contrasts these patterns with the EITC regime of 2003 where subsidies are phased out without abrupt breakpoints as implied by the Minijob earnings threshold. 1508 A.Herget, R.T.Riphahn A priori, as the Midijob subsidy only covers social insurance contributions but not income taxes, the policy left a substantial tax notch for secondary earners. Therefore, we should not expect the Midijob reform to affect transitions from Minijobs for secondary earners much. 2.2 Utilization patterns andchanges intheearnings distribution While Miniand Midijobs cover different earnings ranges and feature different utilization patterns, both are characterized by a prevalence of female workers (see, e.g., Oschmiansky and Berthold (2020), Tazhitdinova (2020), Herget and Riphahn (2022)). Based on weighted survey data and in the age group that we study (30–59) more than 90 percent of Minijobbers are female. Among these females, about 90 percent are married and thus generally secondary earners who face income taxes when earning above the Minijob earnings ceiling. Typically, Minijobs pay low hourly wages. Classic employers of Minijobbers are in the hospitality industry (bars, restaurants), cleaning and building services, or retail. Minijob employment is concentrated in small establishments (0–9 employees) (Collischon etal., 2021). Minijobs are often informal with limited duration, no written contracts, irregular work hours, and on-call employment (Bruckmeier etal., 2018). Bachmann etal. (2012) asked Minijobbers why they use a Minijob (with multiple answers possible); almost 60 percent were motivated by additional earnings, 15 percent by gathering work experience, 14 percent indicated that this was the only job they could find, and 14 percent were motivated by the possibility to work flexible hours. For female Minijobbers being able to combine work and family life as well as flexible hours were substantially more important than for male Minijobbers (31 percent vs. 17 percent). Midijobs are typically part-time positions. About 62 percent are held by females. Males use Midijobs typically when they are young. 46 (61) percent of female (male) Midijobbers are younger than 35 and 44 (29) percent are aged 35–54 (Herget & Riphahn, 2022). In FigureA.4, we describe for the calendar years 2002–2017 from which states of origin individuals enter Midijob employment; for the period before the Midijob introduction, we use the Midijob earnings range. Based on administrative data, no more than 20 percent of Midijobs originate from Minijob employment. The reform of April 1, 2003 not only introduced the Midijob subsidy for social insurance contributions. It also shifted the threshold of the Minijob earnings limit from 325 to 400 Euros per month. We present the distribution of monthly gross earnings after the reform (2003–2012) in Fig.1. Figure3 shows the change in the distribution of monthly gross earnings between 2002 and 2004. The frequency of jobs earning 325–400 Euro increased, that of jobs earning 250–325 Euro declined. Interestingly, the frequency of employments in the Midijob earnings range between 400 and 800 Euro did not increase. 1515 Phasing outpayroll tax subsidies Figure2 shows the relationship between gross and net income. The solid lines describe the situation for a person with no income tax obligation, and the dashed lines describe the situation for a person with relatively high income tax rates, both before and after the reform. For both groups, net income falls at the Minijob earnings ceiling. The decline is larger for the individual with income tax obligations. The notch characterizes the Minijob employment trap. In both cases, the reform (blue lines) attenuates the drop in net earnings and reduces the disincentives to expand labor supply. After the reform, the notch almost disappears for individuals without income tax obligations but not for those subject to income taxes. Therefore and based on the results of the discontinuity analyses in Table3 and Fig.5, we hypothesize that individuals with low income tax rates (e.g., non-married individuals) respond more strongly to the reform than those with high income tax rates (e.g., secondary earners in marriages) because the reform made a relevant difference for the former but not for the latter (Fig.6). To test this hypothesis our outcome of interest (Y) again is a dichotomous indicator of whether an individual i in a Minijob in period t leaves the Minijob between period t and t + 1 and transits to employment with earnings above the Minijob threshold. We distinguish the situation before and after the reform of April 1, 2003 Table 2 RDiT Estimates of Transitions out of Minijobs–SIAB. Source: SIAB (2017) and own calculations Note: Linear regressions, standard errors clustered at the individual level; * p < 0.10, ** p < 0.05, ***p < 0.01. The columns entitled “Coeff” provide the estimate of α2 in Eq.(1), “N” and “Pre-reform mean Y” provide the number of observations and pre-reform mean of the outcome for each sample. “RE” is the relative effect, i.e., the ratio of the coefficient estimate of α2 and the pre-reform mean of the dependent variable and characterizes the relative effect size. The models without added controls (in rows 1 and 4) control for ‘after’ indicator, the re-centered time trend (linear or quadratic), their interaction(s), and a constant term. The vector of basic control variables (in rows 2 and 5) additionally account for gender, 5 age group indicators, East German residence, and German citizenship. The vector of extended controls (in rows 3 and 6) additionally accounts for 3 indicators of educational attainment, 7 indicators of tenure, 8 indicators of occupation, 4 indicators of firm size, and 9 indicators of industry. Sample A considers all Minijob-months observed between April 1, 2002 and March 31, 2004. Sample B drops those Minijobs that were started after April 1, 2003 and Sample C drops those Minijobs that were started after Dec. 31, 2002. Sample A-2013 replicates Sample A-2003 around the reform date of January 1, 2013 Sample A—2003 Sample B—2003 Sample C—2003 Sample A—2013 (1) (2) (3) (4) Specification Coeff RE Coeff RE Coeff RE Coeff RE 1 Linear no controls 0.0005 6.5% 0.0001 0.7% −0.0007* -8.2% −0.0003 −1.9% 2 Linear basic controls 0.0004 4.7% 0.0001 1.1% −0.0006 −7.1% −0.0003 −1.7% 3 Linear ext. controls 0.0001 1.2% 0.0003 3.5% 0.0001 1.2% −0.0005 −3.3% 4 Quadratic no controls 0.0013** 14.8% 0.0011* 13.1% 0.0004 5.3% −0.0014* −8.4% 5 Quadr. basic controls 0.0012* 14.1% 0.0011* 13.0% 0.0005 5.5% −0.0014* −8.4% 6 Quadr. ext. controls 0.0009 10.6% 0.0013** 15.3% 0.0007 8.6% −0.0015** −8.9% N 853,241 751,217 715,313 1,062,892 Pre-reform mean Y 0.0085 0.0085 0.0084 0.0167 1516 A.Herget, R.T.Riphahn (after) and we differentiate groups that were affected to different extents: secondary earners in married couples are subject to high income tax rates; we consider them as our control group observations (treat = 0).23 For them, the Midijob subsidy of social insurance contributions hardly reduces the relevant notch in the earnings distribution. In contrast, non-married individuals with or without a partner in the household enjoy an individual tax allowance that exempts annual earnings of up to about 9,000 Euro from income tax payments (Grundfreibetrag). For these individuals marginal and average income tax rates at the Minijob earnings threshold are low and—depending on other sources of income—may even be zero. Therefore, the introduction of the Midijob subsidy constitutes a relevant reduction in the notch in their net earnings distribution. We consider them as the treatment group of the reform (treat = 1). We use a standard DID model and consider covariates (X) to reduce the residual variance. Let μ be a random error. We estimate the coefficient vectors α and β in this model: Our effect of interest is the estimate of α3. It indicates whether individuals with low income tax burdens (treat = 1) changed their propensity to transition out of the Minijob earnings range after the reform more than individuals with high income tax burdens (treat = 0). If such a difference exists it suggests that the reform reduced the Minijob trap at least for part of the population. The DID estimate represents a causal reform effect if several conditions are met: first, without the reform, the time trend in the propensity to leave a Minijob for higher-earning employment should have developed in parallel for individuals in the treatment and the control group. We inspect the evidence on pre-reform trends in the next section. Second, the reform should not affect treatment or control groups in ways other than through the introduction of the Midijob subsidy. This requirement is violated as the reform not only introduced the Midijob payroll tax subsidy for earnings above the Minijob earnings threshold but also increased the Minijob earnings threshold itself from 325 to 400 Euro per month. However, as the increased earnings threshold affected treatment as well as control group observations, it will bias the estimate of the effect of the Midijob introduction only if the two groups respond differently to the threshold shift. In that case, α3 partly reflects heterogeneous responses (2) Y it+1=𝛼0+𝛼1afterit +𝛼2treatit +𝛼3 ( afterit ⋅treatit ) +𝛽1Xit +𝜇 it Fig. 4 Monthly transition rate from Minijob Employment (April 2002-March 2004). a Sample A (all Minijob observations in the observation window). b Sample B (Sample A without Minijobs started after April 1, 2003). c Sample C (Sample B without Minijobs started after Dec. 31, 2002). Note: The graphs represent the development of monthly average transition rates from Minijob to regular employment. The sample includes all who hold a Minijob as their main employment without being registered unemployed in a given month. The monthly transition rates are de-seasonalized. For a representation of these graphs by gender see Appendix FigureA.6. Source: SIAB (2017) and own calculations ▸ 23 Gudgeon and Trenkle (2024) focus their analysis of Minijob reforms on married women exactly because they are subject to a large notch in their potential earnings. Unfortunately, these authors’ administrative data is not available to us. 1517 Phasing outpayroll tax subsidies 1518 A.Herget, R.T.Riphahn Table 3 RDiT Estimates—SIAB—by gender and predicted secondary earner status. Source: SIAB (2017) and own calculations Note: See Table2. Columns 3 and 4 split the sample based on the predicted probability of being non-married, i.e., single earner vs. married, i.e., likely secondary earner, which was based on the linear regression estimated on SOEP data (see Appendix B) Sample A—2003 Sample A—2003 (1) Men (2) Women (3) Pred. Non-Married (4) Pred. Married Specification Coeff RE Coeff RE Coeff RE Coeff RE 1 Linear no controls 0.0083 *** 73.6% − 0.0004 − 5.2% 0.0152 *** 85.9% − 0.0001 − 0.9% 2 Linear basic controls 0.0078 *** 69.0% − 0.0005 − 6.1% 0.0149 *** 84.2% − 0.0002 − 2.2% 3 Linear ext. controls 0.0061 *** 54.0% − 0.0007 − 8.5% 0.0129 *** 72.9% − 0.0005 − 5.5% 4 Quadratic no controls 0.0063 *** 55.7% 0.0003 3.3% 0.0135 *** 76.3% 0.0005 5.4% 5 Quadr. basic controls 0.0092 *** 81.4% 0.0002 2.4% 0.0136 *** 76.8% 0.0005 5.0% 6 Quadr. ext. controls 0.0071 *** 62.8% 0.0001 1.2% 0.0118 *** 66.7% 0.0002 2.1% N 95,277 757,964 54,183 890,793 Pre-reform mean Y 0.0113 0.0082 0.0177 0.0092 1519 Phasing outpayroll tax subsidies to the change in the threshold. To gauge the relevance of this shift in the Minijob earnings ceiling for Minijob exits we again exploit a later adjustment in the Minijob earnings threshold: on January 1, 2013, the monthly Minijob/Midijob earnings thresholds increased from 400/800 to 450/850. We test this reform’s effect on transitions out of Minijobs to approximate the impact of the 2003 change in the earnings ceiling from 325 to 400 Euros. A third identification requirement is that there are no systematic changes in the composition of the workforce in response to the treatment. First, we examined whether individuals might switch between treatment and control groups in response to the reform. We found that individuals changed their marital status in both directions in similar magnitudes before and after the reform. Second, we evaluated whether there were compositional changes with respect to observable characteristics within the treatment or control group over time. Such changes might indicate differential selection into the two groups in response to the reform. TableA.10 presents the p-values for tests of the hypothesis that the mean values of treatment and control Fig. 5 Monthly transition rate from Minijob Employment (April 2002-March 2004). a By gender (males left panel, females right panel). b By predicted single earner status (single earners left panel, secondary earners right panel). Note: The graphs represent the development of monthly average transition rates from Minijob to regular employment. The sample includes all who hold a Minijob as their main employment without being registered unemployed in a given month. The monthly transition rates are de-seasonalized. Both figures use subsamples from Sample A. Please note that the y-axis of Panel 5b is scaled different from Panel 5a. Source: SIAB (2017) and own calculations 1520 A.Herget, R.T.Riphahn group characteristics changed over time. We observe some minor adjustments but no major changes in characteristics.24 The final identification requirement is the absence of anticipation effects. If in response to the reform, Minijobbers postponed their transition to higher-earning employment or changed the take up of Minijobs altogether this biases our estimates if it affects treatment and control group observations differently. The reform was passed into law on December 23, 2002, and became effective on April 1, 2003, which does allow for potential anticipatory behavior (see our discussion in Sect.3.1). Even though there is no rationale as to why treatment and control group observations might differ in their response, we nevertheless inspect whether our estimates are sensitive to the time window of our sample. This identification strategy exploits potential heterogeneity in treatment effects. Comparing the behavioral responses of more (treatment group) and less (control group) strongly affected individuals does not indicate the overall average effect of the Midijob introduction. However, it can offer evidence as to whether there is an effect at all if those most affected respond differently from those least affected. Fig. 6 Event study analysis of parallel trends (SOEP). Note: The figure depicts estimated coefficients from linear regressions and their confidence intervals. The outcome of transitions from Minijobs to regular employment is regressed on year indicators and their interaction with the treatment indicator. We depict 90 percent confidence intervals of the interaction term coefficients using standard errors that are clustered at the individual level. Source: SOEP wave 35 and own calculations 24 In the control group, the share of female observations drops from 97 to 95 percent and the share with tertiary education increases from 7 to 9 percent. In the treatment group, the share of observations in East Germany increases and the grouping of Minijobbers in the smaller firm size categories is shifted somewhat. 1521 Phasing outpayroll tax subsidies 4.2 Survey data fordifference‑in‑differences analysis (SOEP) Our survey data are taken from the German Socio-Economic Panel Study (SOEP), an annual household panel survey collected since 1984 (Goebel etal., 2019).25 We use data covering the years 2001–2006 to evaluate the 2003 reform. Again, we restrict our sample to individuals aged 30–59 to omit students and retirees.26 We are interested in whether the reform affected the propensity to transition from Minijob employment to regular employment differently for more and less affected individuals. Our sample considers individuals employed in a Minijob as their main employment at the time of the annual survey. Since 2001, the survey asks directly about Minijob employment. We use this self-reported information and consider only those individuals to be in a Minijob who additionally indicate to earn no more than the Minijob earnings threshold.27 This leaves us with 2,736 person-year observations of 1,255 different individuals in Minijob employment in the years 2001–2006. The 1,255 individuals are observed in 1,604 different Minijob employment relationships over time. Unfortunately, the sample size is much smaller than in the administrative data. However, as the administrative data do not provide information on marital status this analysis requires survey data. Table4 provides descriptive statistics on our sample. The vast majority of our Minijobbers is female (94.4 percent) with the largest group aged 35–44 (mean age is 43). Relative to aggregate population shares Minijobs are used relatively more intensely in West than in East Germany and more by German citizens than by non-citizens. In our sample, about 90 percent of the observations are married and thus in our control group. The treatment group comprises those who are single (including those in cohabiting couples), divorced, or widowed. We observe 173 (276) and 1,082 (2,460) different individuals (personyear observations) in the treatment vs. control groups, respectively (for descriptive statistics on additional controls see Appendix TableA.11).28 Our dependent variable indicates whether a person held a Minijob in period t and in period t + 1 transited to regular employment paying social insurance contributions and earning above the Minijob earnings limit; we evaluate transitions between 2001 and 2006. The average annual transition rate is 10.3 percent. The last two columns of Table4 describe the mean transition rates for different groups. We observe higher rates after than before the reform date. As expected, married individuals (treat = 0) 27 Due to the second restriction, we lose 663 of 3,399 observations or 19.5 percent of those who indicated to work in a Minijob. We also drop two individuals for whom information on marital status, our treatment indicator, is missing. We do not use information from earlier survey years because they applied a different survey question to collect employment status information. 25 We use SOEP v35 (1984–2018), DOI https:// doi. org/ 10. 5684/ soepcore. v35. 26 We account for oversampling and non-response in the data by applying the cross-sectional sample weights provided with the SOEP data. In contrast to our analysis of register data we do not omit Minijobbers who are unemployed here due to the resulting small sample sizes. 28 We use a time varying treatment assignment where individuals enter the control group upon marriage. Ideally, the group assignment would be fixed prior to the treatment, e.g., based on marital status in 2002 or 2001. However, this reduces our sample size by about half. The estimates are robust in terms of their signs but have large standard errors. We consider it implausible that changes in marital status are connected to the introduction of Midijobs. 1522 A.Herget, R.T.Riphahn have a much lower average transition rate than non-married individuals (treat = 1). The transition rate of men is higher than that of women (16 vs. 11 percent). As in the SIAB data, the likelihood of leaving Minijobs declines with age. FigureA.7 shows the development of transition rates separately for our treatment and control groups using weighted data. Unfortunately, our data provide only two annual observations prior to the reform, i.e., 2001 and 2002. However, in these years the development of transition rates out of Minijobs is similar for treatment and control groups which suggests parallel paths prior to the reform.29 4.3 Results: difference‑in‑differences analyses Table5 shows our first set of DID results. Column 1 offers results without control variables and shows that the estimate of α3 is statistically significant and positive. It suggests that after the reform the treatment group of non-married individuals increased their propensity to leave a Minijob for regular employment by about 14 percentage points more than the control group of married persons; relative to a mean transition rate of about 10 percent this is a rather large unconditional effect. The result is confirmed in columns 2 and 3, where we first consider controls for basic demographics (gender, age group, East German residence, and foreign citizenship) and then add an extended set of controls (see table notes and Appendix TableA.11 for descriptive statistics). Column 4 shows that the estimation results are robust when we replace the overall ‘post’ effect with a set of calendar year fixed effects.30 TableA.12 presents the results of additional tests and describes effect heterogeneities. Column 1 shows the results after omitting those observations for which we cannot be sure whether a potential transition happened before or after the reform on April 1, 2003.31 The estimates on the thus reduced sample confirm the significant positive treatment effect. In columns 2 and 3 we evaluate the sensitivity of the results to the considered time window of observations. First, we omit two years of postreform observations (column 2) and then we add an additional post-reform observation year (column 3): our main result hardly changes. In column 4 we omit male observations; the result shows that women respond substantially less to the reform than men, confirming the results or our RDiT analyses. Omitting observations with 29 In a separate analysis, we estimated an event study to describe the preand post-treatment differences between treatment and control groups. We regressed the outcome on year indicators and their interactions with the treatment indicator but without additional controls. Figure6 shows the differences between treatment and control groups. Again, outcome differences prior to the reform are insignificant, and transition rates for the non-married increase faster after the reform than those for married Minijobbers. 30 We use sampling weights in the analyses of SOEP data. The results in Table5 are sensitive to this choice. 31 The uncertainty is due to the annual interview which informs only about the status at the time of the interview but not about when a status change occurred. We omit two groups of observations: those for whom we know the Minijob status in 2002 but do not know whether the transition to the 2003 status took place before or after the reform date of April 1, 2003 and those for whom we know their Minijob status prior to April 1, 2003 but do not know whether their transition to the 2004 status took place before or after the reform date of April 1, 2003. 1523 Phasing outpayroll tax subsidies East German residence as in column 5 reveals somewhat larger effects in West than in East Germany.32 Finally, we need to account for the fact that the reform of April 1, 2003 shifted the Minijob earnings ceiling from 325 to 400 Euro per month. If treatment (the non-married) and control (the married) groups responded differently to this change this may bias our finding. In order to gauge the overall relevance of the ceiling shift we consider the reform of January 1, 2013 when the Minijob earnings ceiling was increased again, this time from 400 to 450 Euro per month. We evaluate the impact of this reform on changes in transition propensities. We use the same sample and treatment definitions as before just shifting the observation period to 10years later (see Tables A.11 and A.13 for descriptive statistics).33 Table A.14 shows the results on Minijob transitions for the period 2011 to 2016. We evaluate the reform heterogeneity for treatment and control groups around the increase in the Minijob earnings ceiling from 400 to 450 Euros on January 1, 2013. In this case, the results yield a negative and statistically insignificant estimate of α3. Overall, the transition rates Table 4 Descriptive Statistics—Basic Controls: 2003 Reform Sample (SOEP). Source: SOEP wave 35 and own calculations Note: The descriptive statistics describe the sample of 2,736 person-year observations. The data are not weighted Descriptives Mean transition rate when variable has Variable Mean Std. Dev Value 0 Value 1 Transition (0/1) 0.1031 0.3041 0.0000 1.0000 Post (0/1) 0.5808 0.4935 0.0850 0.1179 Treat (0/1) 0.1009 0.3012 0.0995 0.1809 Female (0/1) 0.9441 0.2298 0.0952 0.2353 Age: 30–34 (0/1) 0.1648 0.3711 – 0.1537 Age: 35–39 (0/1) 0.2288 0.4201 – 0.1347 Age: 40–44 (0/1) 0.1988 0.3992 – 0.1116 Age: 45–49 (0/1) 0.1659 0.3721 – 0.1037 Age: 50–54 (0/1) 0.1411 0.3482 – 0.1108 Age: 55–59 (0/1) 0.1005 0.3007 – 0.0678 East Germany (0/1) 0.0757 0.2645 0.1085 0.1667 Foreign Nationality (0/1) 0.1137 0.3175 0.1148 0.1068 32 Marital status may be a weak proxy for the income tax burden. Therefore, we investigated whether it might represent alternative mechanisms, instead. We replaced our treatment indicator of not married (T) vs. married (C) individuals by several alternatives. First, for the sample of married persons (N = 2,460) we used an indicator of whether a person has children (C) or not (T). Second, we considered non-married individuals with (C) vs. without stable partners (T). In neither case did we obtain statistically significant treatment effects. Third, we compared non-married (T) individuals only to married individuals without children (C) to safeguard against effects of childcare. Here, we continue to find significant positive treatment effects, supporting our main results. 33 We drop 32 individuals for whom information on marital status, our treatment indicator, is missing in the data. 1524 A.Herget, R.T.Riphahn increased slightly for the control group of married individuals after the reform (see row 1) and transition rates of the treatment group are generally significantly higher (row 2). Importantly, the reform did not affect the relative transition rates of the two groups. If these patterns similarly held in 2003 our findings of a significant increase in transition rates after the 2003 reform as reported in Tables5 and A.12 are not likely to be biased by the change in Minijob earnings ceiling that happened simultaneously with the introduction of the Midijob subsidy. The findings in Table A.14 corroborate our finding of a significant and large increase in transitions out of Minijobs after the 2003 reform for our treatment group, the non-married. 4.4 Difference indifferences analyses using administrative data Compared to the survey data, our administrative data offer larger samples, more precise information, and are observed at monthly instead of annual frequency. Even though the administrative data lack information on marital status, we can replicate the difference-in-differences analyses using this larger database. In this section, we show two approaches to approximate the survey-based analyses. In the first approach, we replace the non-married treatment indicator from Sect.4.3 with an indicator of male gender; in the survey data, the share of males among the nonmarried (22.1 percent) is six times as high as the share of males among the married (3.7 percent). Therefore, this might be a reasonable first approximation. In the second approach, we estimate a prediction equation for being non-married with the survey data and use it to predict non-married status in the administrative data. The Table 5 DID Estimates—2003 reform baseline results (SOEP). Source: SOEP wave 35 and own calculations Note: All estimations use 2,736 person-year observations. Linear regressions, standard errors clustered at the individual level are in parentheses. The vector of basic controls accounts for an indicator of gender, 5 indicators of age group, an indicator of East German residence, and an indicator of non-German citizenship. The vector of extended controls accounts for 5 education indicators, 7 tenure indicators, 4 firm size indicators, 9 industry indicators, and 8 occupation indicators. Column 4 replaces the post indicator with a set of calendar year fixed effects. The estimations use cross-sectional sample weights to account for nonresponse and oversampling. * p < 0.10, ** p < 0.05, ***p < 0.01 Full sample Full sample Full sample Full sample (1) (2) (3) (4) Post 0.004 0.004 −0.000 – (0.015) (0.015) (0.015) Treat 0.002 −0.022 −0.038 -0.0041 (0.029) (0.031) (0.032) (0.033) Post * treat 0.139*** 0.123** 0.131*** 0.134*** (0.052) (0.051) (0.050) (0.050) Controls—basic No Yes Yes Yes Controls—extended No No Yes Yes Year fixed effects No No No Yes 1531 Phasing outpayroll tax subsidies Huttunen, K., Pirttilä, J., & Uusitalo, R. (2013). The employment effects of low-wage subsidies. Journal of Public Economics, 97, 49–60. Immervoll, H., Kleven, H. J., Kreiner, C. T., & Saez, E. (2007). Welfare reform in European Countries: A microsimulation analysis. Economic Journal, 117, 1–44. Jara, H. X., Gasior, K., & Makovec, M. (2020). Work incentives at the extensive and intensive margin in Europe: The role of taxes benefits and population characteristics. Social Indicators Research, 152, 705–778. Keller, Berndt and Hartmut Seifert (2015). Atypical forms of employment in the public sector - are there any? WSI-Diskussionspapier No. 199, Düsseldorf: Wirtschaftsund Sozialwissenschaftliches Institut in der Hans-Böckler-Stiftung. Keller, B., Pusch, T., & Seifert, H. (2021). Midijobs: Die unbekannte Variante der atypischen Beschäftigung. WSI Mitteilungen, 14(2), 160–170. Kleven, H. (2016). Bunching. Annual Review of Economics, 8, 435–464. Korkeamäki, O., & Uusitalo, R. (2009). Employment and wage effects of a payroll-tax cut - Evidence from a regional experiment. International Tax and Public Finance, 16, 753–772. LaLumia, S. (2009). The earned income tax credit and reported self-employment income. National Tax Journal, 62(2), 191–217. Leigh, A. (2007). Earned income tax credits and labor supply: New evidence from a British natural experiment. National Tax Journal, 60(2), 205–224. Lietzmann, T., Schmelzer, P., & Wiemers, J. (2017). Marginal employment for welfare recipients: Stepping stone or obstacle? Labour, 31(4), 394–414. Meyer, B. D. (2002). Labor supply at the extensive and intensive margins: The EITC, welfare, and hours worked. American Economic Review, 92(2), 373–379. Neumark, D. & Shirley, P. (2020). The long-run effects of the Earned Income Tax Credit on Women’s Earnings, NBER Working Paper No. 24114, Cambridge, Mass. Oschmiansky, Frank and Julia Berthold (2020), Minijobs und Midijobs, available at https:// www. bpb. de/ themen/ arbeit/ arbei tsmar ktpol itik/ 317249/ minij obsundmidij obs/ (last access: August 9, 2022). Saez, E., Schoefer, B., & Seim, D. (2019). Payroll taxes, firm behavior, and rent sharing: Evidence from a young workers’ tax cut in Sweden. American Economic Review, 109(5), 1717–1763. Scarfe, Rachel (2021). Flexibility or certainty? The aggregate effects of casual jobs on labour markets, Edinburgh School of Economics Discussion Paper Series No. 294, University of Edinburgh. Seifert, H. (2017). Wie lassen sich Entwicklung und Strukturen atypischer Beschäftigungsverhältnisse erklären? WSI-Mitteilungen, 2017–01, 5–15. Tazhitdinova, A. (2020). Do only tax incentives matter? Labor supply and demand responses to an unusually large and salient tax break. Journal of Public Economics, 184(April), 104162. Tazhitdinova, A. (2022). Increasing Hours Worked: Moonlighting Responses to a Large Tax Reform. American Economic Journal: Economic Policy, 14(1), 473–500. Trampe, P. (2007). The EITC disincentive: The effects on hours worked from the phase-out of the earned income tax credit. Econ Journal Watch, 4(3), 308–320. Trampe, P. (2008). The EITC disincentive: A reply to Dr Hilary Hoynes. Econ Journal Watch, 5(1), 59–65. Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.