Heterogeneous effects of minimum wage on labor market outcomes: A case study from Turkey

We assess the effects of a sharp minimum wage increase on wages, informality, and employment in Turkey, a large developing economy with one of the highest minimum wage-to-average wage ratios among OECD countries and widespread discrepancies between labor market outcomes of women and of men. We look at the quasi-experimental 2016 minimum wage increase and pay attention to identifying information coming from demographic groups. We find that the increase in the minimum wage had an economically substantial and statistically significant positive impact on wages. Despite the positive wage effects of the increase, we find no negative employment effects. However, we show that the minimum wage increase may have caused an increase in the share of informal employment among workers with less than tertiary education, especially for such workers working for small firms. Current version: September 16, 2020


Introduction
The literature on the effects of minimum wage on employment finds its roots in the standard labor market model, which posits a negative relationship between the demand for labor and the wage level and suggests that a minimum wage imposed above the labor market clearing wage level will lead to unemployment. Several empirical studies test the impact of minimum wage on employment, but no clear consensus emerges from these studies. While some argue that minimum wage increases have adverse impacts on less-skilled workers' employment (Brown et al., 1982;Neumark and Wascher, 2007); others find no employment effects (Card, 1992a;Card, 1992b;Katz and Krueger, 1992;Card and Krueger, 1994), even though the overwhelming majority of these studies find some evidence for the positive effects of minimum wage on average wages. A controversy about methodology also exists, and these different findings may be partly due to the use of different empirical approaches. As the drawbacks of the use of time series are well established (Card and Krueger, 1995), there is an ongoing debate about the use of panel data of case-study approaches and of alternative approaches such as the "bunching" estimator (Dube et al., 2010;Allegretto et al., 2011;Allegretto et al., 2017;Neumark et al., 2014;Neumark and Wascher, 2017;Neumark, 2017;Cengiz et al., 2019).
While most of this literature focuses on advanced economies, several additional issues emerge when it comes to analyzing the effects of minimum wages in developing economies.
First, weak compliance with the minimum wage law in these countries (Rani et al., 2013) may negate the employment effects of the minimum wage. It is possible that overall employment is not affected by minimum wage increases as employers are able to pay lower than the minimum wage by employing workers informally. For example, del Carpio et al. (2019) found that increases in the minimum wage raise the share of informal workers in the labor market in Thailand. Similar results are presented for Honduras (Ham, 2018), Brazil (Jales, 2018;Broecke and Vandeweyer, 2016), and Chile (Wedenoja, 2013). On the other hand, a negative impact on formal employment is reported for Indonesia (Comola and De Mello, 2011), Russia (Muravyev and Oshchepkov, 2016), and South Africa (Millea et al., 2017). Second, the relationship between minimum wage and average wages may be quite different in developing economies due to the existence of large informal sectors. 1 If an increase in minimum wage level causes a labor flow from the formal sectors to the informal sectors, or new entrants become informal workers rather than formal workers due to minimum wage increase, average wage effects of the increase will depend on the effects on informal wages. Khamis (2013) found that minimum wage increases have considerable positive impacts on both informal and formal workers' wages in Argentina. Lemos (2009) argued that an increase in the minimum wage is associated with a compression in the wage distribution of informal workers in Brazil. While Millea et al. (2017) found a positive relationship between minimum wages and informal workers' wages in South Africa, Ham (2018) found a negative relationship in Honduras.
Building on this literature, we investigate the labor market outcomes of the quasiexperimental sharp increase in the minimum wage level in Turkey in 2016. Turkey presents an interesting and important case considering that its minimum-wage-to-average-wage ratio is 1 In some developed countries where sector-specific collective agreements are prevalent, informal works, such as service provision, are margins of adjustment to wage increases arising from collective agreements because it is possible to pay informal service providers below the minimum wage rate (Martins, 2020). C o lo m b ia T u rk e y C h ile F ra n c e N e w Z e a la n d S lo ve n ia L it h u a n ia Is ra e l P o rt u g a l L u xe m b o u rg A u s tr a lia R o m a n ia P o la n d L a tv ia B e lg iu m H u n g a ry K o re a U n it e d K in g d o m G e rm a n y C a n a d a N e th e rl a n d s S lo va k R e p u b lic Ir e la n d G re e c e E s to n ia Ja p a n   of total employment. The labor market is characterized by regional inequalities, widespread discrepancies between female and male employment and wages, and wage discrepancies across employment by small and large firms.
In addition to its potential role in such disparities, the minimum wage may also be considered as a "surrogate social policy" used to support low-income groups. The labor market produces dynamics that lift individuals out of poverty (Şeker and Dayıoğlu, 2015), and minimum wage may thus play an active role in the income and labor market outcomes of the poor. However, there are few and limited studies examining the Turkish case. Acar et al. (2019) argued that the increase in minimum wage in 2016 led to an increase in firms' exit rates from formal sectors, and Gürsel et al. (2018) found a positive relationship between the minimum wage increase and the share of informal workers. Of the few earlier studies, Güven et al. (2011), using a time-series approach, found no employment effects. Pelek (2015), using a fixed-effect model and regional variation in Kaitz index, shows that minimum wage has a positive impact on informality among young workers. Papps (2012), using the variation in individual-level labor costs to detect employment effects of the social security taxes and minimum wage, found that minimum wage has a negative effect on employment outcomes. Yüncüler and Yüncüler (2016) utilized the variation in the share of minimum wage workers affected by the minimum wage increase in 2004 within various industry and occupation interactions and found that the minimum wage increase affected wages, working hours, and informality positively.
In this article, we follow the methodology used in Caliendo et al. (2018), Dolton et al. (2015), and Card (1992b) and exploit the regional variation in the share of workers affected by minimum wage increases. This allows us to control for the Russian economic sanctions and the coup attempt-two confounders in 2016 who coincided with the minimum wage increase. We also show that ignoring the gender discrepancy in labor markets may result in biased results when analyzing the effects of minimum wages. Our findings also demonstrate the importance of addressing the heterogeneous effects of the minimum wage increase across firms with different sizes. We find that the increase in the minimum wage in 2016 has been shown to have had economically substantial and statistically significant positive wage effects on most demographic groups. When it comes to least educated groups, these positive effects are mainly seen in the wages of men. We also find positive, statistically significant and economically important effects on informal wages, supporting the "lighthouse effect" of the quasi-experimental minimum wage increase. We find no negative impact on employment from the minimum wage increase. However, we also find that the increase induced an economically considerable positive effect on the share of less-educated workers working without social security. These positive informality effects rise when we focus on firms that employ less than ten employees.
The remainder of the article is organized as follows. In Section 2, we provide a brief overview of the labor market characteristics, the institutional setting, and the political background of the sharp minimum wage increase in 2016 to show that this increase was exogenous to labor market conditions. We present the data in Section 3. We identify the minimum wage workers, discuss whether the minimum wages are binding, and present our empirical model and identification strategy in Section 4. Results and their robustness are discussed in Section 5 before we conclude with Section 6.
2 Institutional details and political background

The minimum wage policy and labor markets
Minimum wage regulations in Turkey began with region-varying levels in 1951, was replaced by a sector-specific minimum wage statute in 1974, and became national in 1989 (Korkmaz, 2004). The minimum wage legislation states that a tripartite committee representing workers, employers, and the government must set a monthly minimum wage at least every two years.
Most of the time, the committee announces different minimum wage levels for the first and second half of the following year in December of the current year, taking into account the inflation expectations, and the updated wage floors go into effect on January 1 and July 1 of the following year. In some years, like 2016, when the minimum wage increase is relatively high, the committee sets one minimum wage level for the entire year.
The share of wage employment in total employment and the share of minimum wage workers within the wage employment has been on the rise since the early 2000s; however, on the other hand, the rate of unemployment has been persistently high even in periods of high economic growth (Orhangazi, 2019). Increasing wage employment, higher share of minimum wage workers, and the persistent high unemployment render labor market regulations, relevant in the debates on unemployment, such as employment protections and minimum wage, and non-wage costs to employers (Ayhan, 2013). Moreover, it is possible to place the minimum wage regulations within the context of social policy in Turkey, since crucial social policy indicators such as the poverty threshold are determined according to the current minimum wage level (Tekgüç, 2018). Given the active role of the labor market in income and poverty dynamics in Turkey (Şeker and Dayıoğlu, 2015), the minimum wage policy, through its effects in the labor market, may also be considered of as one of "surrogate social policies" used by the government to lift individuals out of poverty and to provide a modest amount of social assistance.
Three characteristics of the labor markets in Turkey should be taken into account when analyzing the impacts of minimum wages on labor market outcomes. First, even though the minimum wage is legally binding, some employers, especially small firms, in practice avoid paying the minimum wage (and other employment regulations) by employing workers informally, that is, without registering them with the social security system. Although the share of informal employment has declined during the 2000s, it is still estimated that around one-third of overall employment and nearly 20% of full-time wage employment in the private sector are informal (Orhangazi, 2019). Therefore, small firms/higher informality-larger firms/ lower informality segmentation in the labor market may also be debated.
Second, an important segmentation in the Turkish labor market is observed along the widespread discrepancies between labor force participation rates, unemployment rates, and the wages of women and men. The participation rate of women in Turkey in the labour force (34% in 2018) was much lower than that of men (72% in 2018). Compared to most other developing countries, it is also much lower (e.g. 48% in South Africa and Argentina, 55% in the Russian Federation, 52% in Brazil in 2018). Moreover, the rate of unemployment is higher for women and there is a persistent gender wage gap. For example, less educated women earn 24% less than less-educated men, while women with tertiary education earn 9% less than men with tertiary education (Tekgüç et al., 2017). Third, regional inequalities also characterize the labor market. Unemployment is concentrated in large migrant-receiving provinces such as Istanbul, Izmir, Adana, Mersin, and Van, while the Northeastern region displays low unemployment and relatively high labor force participation due to the higher share of agricultural employment.
Share of manufacturing employment in the formal sector is the highest in Tekirdağ, Bilecik, and Düzce regions and the lowest in Iğdır, Hakkari, and Ardahan regions in Southeastern Turkey, which is also economically the most underdeveloped with an ongoing armed conflict (Ministry of Development, 2011).
In the empirical analysis below, we pay specific attention to the heterogeneities presented by these characteristics and show that the effects of the minimum wage increase may be different for women and men, formal and informal sectors, small and large firms, different education groups, different age groups, and across regions.

The 2016 minimum wage increase
The real minimum wage level was gradually increased through the 2000s and 2010s. The sharpest minimum wage increase came in 2016 with a 33% nominal, 25% real increase.
The main driving force behind this sharp increase was the electoral competition in 2015.
In the days leading to the June 7 general elections, the most popular election promise by the competing parties became the promise to increase the minimum wage, when the main opposition party, the Republican People's Party's (CHP), pledged to raise the monthly minimum wage from 949 Turkish liras to 1500 Turkish liras. The other opposition parties quickly jumped in and the Nationalist Movement Party (MHP) promised an increase to 1400 Turkish liras and the People's Democratic Party (HDP) to 1800 Turkish liras. The ruling Justice and Development Party (AKP) refrained from joining this competition and kept the position that these promises were economically unrealistic. However, AKP lost its parliamentary majority in the June 7 elections, and, unwilling to form a coalition government, pushed for renewed elections. Between the June 7 and the renewed elections on November 1, AKP also entered the minimum wage pledge competition and promised a 1300 Turkish lira minimum wage, which, after AKP's election victory in November, went into effect on January 1, 2016. Hence, the increase in the minimum wage can be seen as exogenous to the internal dynamics of the economy as it originated mainly from the exogenous political competition rather than the dynamics of the economy itself.
As Figure 2 shows, this is one of the highest minimum wage increases in Turkey during the 2000s. Although a similar increase took place in 2004, the increase in the minimum wage cost to the employer, which is the sum of unemployment premium, social security tax, and net minimum wage, was smaller in 2004. As such, the 2016 increase received widespread complaints from the employers, 2 which led the government to temporarily reduce the social 2 Just to cite a few examples, an employer in textile sector was making the following complaint: "I rented a new facility. I was planning to increase my production by hiring 500 workers. In response to announcement [of the minimum wage increase], however, I dropped the idea" (Hürriyet, 2015). A CEO in textile manufacturing business argued that "the labor-intensive sector is not able to cope with a 30% increase in minimum wage. Some factories will be liquated." (ibid.). But complaints were not limited to the textile sector. Before the enactment of the increase, the vice chairman of the Independent Industrialists' and Businessmen's Association (MÜSIAD) maintained that "business world is very sensitive to such costs" (IHA, 2015). security premiums paid by the employers for the already employed for 2016. In other words, the nominal increase in the total labor cost to the employers was less than the 33% increase in the minimum wage but still significant at 26%. Moreover, this temporary reduction was only valid for the number of employees who were employed full-time in 2015 and did not cover new hires.
The year 2016 also witnessed two political events that had significant economic consequences. First, following a military rift in Syria toward the end of 2015, Russia began imposing economic sanctions on Turkey. These sanctions included stopping agricultural imports from Turkey, rescinding the work permits of Turkish construction companies in Russia, and preventing Russian tour operators from organizing tours to Turkey. These sanctions hit regions that either produced agricultural products for the Russian markets or were dependent on tourists from Russia. As the sanctions and the minimum wage increase occurred around the same time, a minimum wage study based on a sectoral identification strategy would not be able to distinguish their effects. Therefore, in the analysis below, we rely on an identification strategy based on the regional variation in the potential minimum wage workers. The second political event in 2016 was the coup attempt of July 15. This attempt and the ensuing state of emergency led to a contraction of the economy in the third quarter of 2016. It is thus possible to falsely attribute the effects of this turmoil to the minimum wage's effects. Therefore, in the analysis below, we capture such yearspecific effects and address the possibility that the turmoil's and the sanctions' effects vary across regions.

Data
We use the 2009-2016 annual Household Labor Force Survey (HLFS), which presents individual-level cross-sectional data gathered by Turkstat to produce official labor market indicators and provides information about demographic characteristics, employment status, income, and past work experience of household members. Even though some consider the quarterly HLFS to be more informative than the annual HLFS, especially when there are two different minimum wage levels in a year (Gürsel et al., 2018), the quarterly HLFS lacks regional information, which is essential for the empirical strategy of this article. Therefore, we opt for the annual HLFS.
The annual HLFS is representative at the national and the NUTS-2 region level. Our main sample is non-institutional working-age (15-64) population. The main outcome variables we use are income, employment, and informality status; and the main demographic variables such as age, education, and gender. We focus on the full-time wage employment in the private sector rather than the overall employment, since the public sector employees are already paid more than the minimum wage. Therefore, we start our analysis from 2009, the first wave that introduces public-private sector distinction in the data set. Finally, we analyze employment outcomes rather than unemployment outcomes, since minimum wage may affect households' labor force participation decisions, leading to spurious increases in unemployment figures.
Our analysis is thus not affected by the change in unemployment and labor force participation definitions that took place in the data in 2014.

Distribution of affected workers by demographic groups
Heterogeneous workers models predict that minimum wage's effects on the labor market outcomes of the demographic groups with a higher share of the minimum wage workers will be stronger (Brown et al., 1982). As a result, most studied groups in the minimum wage literature are teenagers (13-19 years), young adults (20-24 years), retail trade workers, and workers in restaurants. Studies seeking to discern the effects of the minimum wage in Turkey generally focus on the youth (e.g. Bakış et al., 2015;Pelek, 2015), conduct an industry-occupation based analysis (e.g. Yüncüler and Yüncüler, 2016), or examine the manufacturing sector (e.g. Güven et al., 2011). We identify workers potentially affected by the minimum wage increase in year t using the fraction of workers whose wages fall below the new minimum wage level in year t, but equal or surpass the 75% of the minimum wage level in year t-1 within nearly all education and age groups. 3 Figure 3 shows the share of full-time private sector workers earning lower than 711 Turkish liras, workers earning lower than 1300 Turkish liras but higher than 711 Turkish liras, workers 3 Data shows that the share of wage workers with wages less than 95% of minimum wage fluctuates from year to year (see Figure A1 and A2 in appendix), suggesting that even some of these workers below minimum wage is impacted by it. The share of workers with wages below 75% of minimum wage is stable over the years.  Meanwhile, workers with a tertiary education seem to be the least affected group, as the share of those earning higher than the minimum wage was higher than 50% in both years.
The share of workers with a lower than minimum wage was very low among higher educated workers in both years. All in all, Figure 3 suggests that the minimum wage increase in 2016 had an enforcement power on wages and that less-educated workers appear to be the most affected group. 4 Teenagers and young adults are also among the prime suspects who are most likely to be affected by an increase in the minimum wage. Neumark and Wascher (2007) present an extensive review of the literature, which shows that the vast majority of the minimum wage literature considers the effects of minimum wage on teenagers and young adults. We thus look at the share of workers earning around the minimum wage in 2015 and 2016 by age cohorts in Figure 4. The figure shows that the only group in which more than 50% of workers earned around minimum wage in 2015 was the 15-24 age cohort. There was a considerable increase in the share of workers with a wage around the new minimum wage level (between 1300 Turkish liras and 1600 Turkish liras) in 2016 among the youth. But nearly 35% of the youth was not able to climb up the wage ladder after the increase, most likely due to informality. It is also important to note that the share of workers earning between 711 Turkish liras and 1300 Turkish liras in 2015 was approximately 40% within all other age cohorts, showing that the share of workers 4 In Figure A1, we adopt a more conservative approach to identify minimum wage workers by taking the minimum wage level in the first half of the year as the minimum wage threshold, and the minimum wage level in the second half of the year as the maximum threshold, and allowing 5% error margin to observe time trends in the share of minimum wage workers in both formal and informal sectors. The figure shows that the share of minimum wage workers has been highest among formal workers with no high school degree and lowest among formal workers with college education. Considering only formal workers, for example, around 33% of the workers with less than high school education earned a minimum wage in 2015, while the same figure was nearly 25% for the high school group, and around 13% for workers with college education. Also, the share of minimum wage workers has increased considerably over time among all formal workers and there was a noticeable increase in the share of workers working at the minimum wage level among workers with no high school degree and high school degree in 2016. When we look at the same figures by age groups in Figure A2, we observe that in the formal sector, the share of minimum wage workers within the 15-24 age group was approximately 38% in 2015. By contrast, 22% of workers aged 25-34, for example, earned the minimum wage in 2015. whose wages could be directly affected by the minimum wage increase in 2016 (i.e. those earned between 711 Turkish liras and 1300 Turkish liras) were also high among young adult and adult workers expected to be more experienced and hence less affected by the increase.
In short, the evidence presented in Figures 3 and 4 indicates that the minimum wage has more impact on the less-educated and the younger workers (ages 15 to 24). But the figures show also the importance of addressing the identifying information coming from remaining demographic groups. For this reason, differing from the previous studies focusing on specific groups such as youth, restaurant workers, or manufacturing workers, we analyze the effects of minimum wage on all age and education groups in our sample. Since the figures illustrate also that it is crucial to take into account the labor market segmentation (formal vs. informal) to capture the effects of the minimum wage policy on workers working without social security, we present the wage distribution of workers working full time in the formal and informal sectors starting from 2013 in Figure 5 to quantify the enforcement power of the minimum wage policy more carefully.

Identification strategy
We showed that minimum wage has enforcement power in Turkey and workers with no high school degree and younger workers (ages 15 to 24) in the formal sector seem to be more likely to be affected by an increase in the minimum wage. Many minimum wage studies examine the minimum wages' effects on demographic groups by exploiting the state-level variation in the minimum wage statute and labor market outcomes. While an identification strategy based on the variation in state-specific minimum wage regulation is inapplicable in countries with a national minimum wage level, a regional identification strategy can still be valuable since the effects of a national minimum wage level can differ across regions. Card (1992b), for example, captures the effects of the federal minimum wage increases in 1990 and 1991 in the U.S. by utilizing the state-level variation in the fraction of workers affected by the increases. Dolton et al. (2015) and Caliendo et al. (2018) use a similar methodology to examine the effects 5 Figure A3 supports this evidence from a different perspective. It shows that some informal workers were able to earn the new minimum wage following the minimum wage increase in 2016. To support these descriptive evidences, we construct below a causal relationship between the minimum wage increase and informal employment wages. We follow a similar methodology and exploit the regional variation in the fraction of affected workers whose wages fall below the new minimum wage level in year t, but equal or surpass the 75% of the minimum wage level in year t-1 in region j in demographic group s.
For example, if an increase in the minimum wage from year t-1 to t affects the labor market outcomes of the workers with no high school degree, then the regions with a higher fraction of affected workers within this demographic group in year t-1 should experience a relatively larger change in less-than-high-school-educated workers' labor market outcomes in year t.
Our data provides spatial information at NUTS-1 and NUTS-2 levels. The provinces are classified into 12 (NUTS-1) and 26 (NUTS-2) regions according to their geographical, demographic, economic, and socio-cultural proximity. We use NUTS-2 level classification to utilize a higher variation and define the fraction of affected workers in each NUTS-2 region in year t-1 as the following, where n sjt-1 is the number of affected workers whose wages are equal to or higher than the 75% of the minimum wage level in year t-1 but lower than the minimum wage level in year t, in demographic group s in NUTS-2 region j in year t-1; E sjt-1 is the employment rate of salaried, waged, or casual workers working in private sector in demographic group s in region j in year t-1. 6 We have three education groups, and five age groups.
For example, the upper panel of Figure 6 shows that the ratio of salaried, waged, or casual young workers working in the private sector whose wages equal or exceed 711 Turkish liras

Model
To capture the relationship between the fraction of workers affected by minimum wage increases and the changes in labor market outcomes for demographic groups, we use Dolton where Y isjt is labor market outcomes, i.e. wages, employment, and informality status of worker i with demographic characteristic s in region j in year t. J j is a dummy variable for region j, 6 To check the robustness of this key variable of interest, we take into account informality issue and use a different numerator including only formal workers. We also exploit more conservative variable of interest, "fraction at the minimum wage", by taking the minimum wage level in the first half of the year as the minimum threshold and the minimum wage level in the second half of the year as the maximum threshold, and allowing a 5% error margin to capture minimum wage workers. See the notes in Figure A2 and Section 5 for a more detailed discussion of alternative exposure variables. Notes: Own calculations based on the 2015 wave of the Turkish Household Labor Force Survey. The sample includes full-time wage employment in the private sector. The figures show the regional distribution of young or less than high school educated workers earning higher than 711TL but lower than 1300TL in 2015.
capturing the effects of the time-invariant characteristics of the region on dependent variables.  More Impacted regions are ones in which the fraction of workers affected by the increase (workers earning less than 1300TL but higher than 711TL in 2015) in the minimum wage in 2016 is higher than the median fraction. There are 13 NUTS-2 More Impacted and 13 NUTS-2 Less Impacted regions.
We assume that labor market outcomes in regions with a higher fraction of workers affected by minimum wage increases and in regions with a lower fraction of workers affected by minimum wage increase would have followed a similar trend in the absence of a minimum wage shock. If the assumption is satisfied, then a difference between labor market outcomes of more impacted regions and that of less impacted regions in 2016 can be attributed to the effects of the minimum wage increase. The violation of the assumption is likely, however, because the regions with a lower fraction of affected workers were in the Northeast and Mediterranean regions (see Figure 6). The eastern region of the country has been subject to region-specific shocks such as armed conflicts or a region-specific state of emergency. Since the fraction of affected workers was relatively low in these regions, a negative shock coinciding with the minimum wage increase in these regions can create negative employment outcomes independently of the minimum wage increase, causing the spurious positive relationship between employment outcomes and the fraction. Given that 2015-2016 was a period in which two events that were likely to affect labor market outcomes took place (see section 2.2), it is possible that we falsely attribute the effects of such time-variant unobservable heterogeneities across regions to interaction term T k(t) F sjt-1 .
To investigate such heterogeneities, we define the regions in which the fraction of affected workers with no high school education in 2015 was higher than the median fraction as more impacted regions. Figure 7 shows that although the increase in average female wages in more impacted regions exceeded the increase in average female wages in less impacted regions in 2016, the divergence started in 2015. Likewise, the assumption that less-than-high-schooleducated male and female workers' employment would have followed a similar trend in the absence of the minimum wage increase does not hold. The parallel trend assumption is satisfied for only informal employment of male workers with no high school education. We change the definition and define the regions in which the fraction of affected young workers in 2015 was higher than the median fraction of more impacted regions. Figure 8 indicates that there was no considerable divergence in the average wages of both female and male young workers until 2016. For employment and informality outcomes of young workers, however, the parallel trend assumption is violated.
In short, both More Impacted regions are ones in which the fraction of workers affected by the increase (workers earning less than 1300TL but higher than 711TL in 2015) in the minimum wage in 2016 is higher than the median fraction. There are 13 NUTS-2 More Impacted and 13 NUTS-2 Less Impacted regions.
market outcomes. Similarly, if the effects of the Russian sanctions or the coup attempt vary across regions, we are able to control for such heterogeneity across regions.

Results and discussion
We present descriptive statistics in Table 1  are clustered at the region (NUTS-2) level. When we focus on education groups, we control for age using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. We control for education when we focus on age groups using dummies for less than high school, high school, and tertiary-educated groups.
Before jumping to the interpretation of the baseline estimations, Table 1 provides some useful information to make sense of the results. It divides the regions into more impacted regions in which the fraction of workers earned lower than 1300 Turkish liras but higher than 711 Turkish liras in 2015 is higher than the median fraction and less impacted regions in which the fraction is lower than the median fraction. Some time-invariant differences stand out. First, less impacted regions are more populous. Second, the employment-to-population ratio among both men and women is higher in less impacted regions. Third, average wages are lower in more impacted regions, while informality is considerably higher, especially for women. Finally, the raw gender wage gap seems to be higher in regions expected to be more impacted by the minimum wage increase. We try to capture such time-invariant heterogeneities across regions by using the NUTS-2 level region fixed effects. We try to discern the effects of unobservable time-varying heterogeneities across regions using 5 aggregated region fixed effects x year fixed effects. 7 The estimates for the increase in remaining years are presented in Figure A4 to A6.   The key variable of interest is the fraction of affected workers (the ratio of workers earning higher than 75% of the minimum wage level in year t-1 but lower than the new minimum wage level in year t). Each cell shows the estimates for the interaction of the 2016 dummy and the fraction of affected workers in 2015. Workers with Higher degree are those with 2-or 3-years higher education, or 4 years higher education, or master's degree (5-or 6-years faculty included) or doctorate. There are 26 NUTS-2 regions. All coefficients are relative to the interaction of the 2010 dummy and the fraction of affected workers in the 2009 variable. In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations. ***, **, and * indicate a statistical significance level at the 0.01, 0.05, and 0.1 level, respectively. Table 2 presents the wage effects of the minimum wage increase in 2016. It shows the baseline wage estimates for the key variable of interest, i.e. regional fraction of workers earned lower than 1300 Turkish liras but higher than 711 Turkish liras in 2015. Since the minimum wage committee sets a monthly wage floor, employers may offset the effects of the minimum wage increases by raising working hours. We thus calculate hourly wages using weekly working hour's information in the HLFS and look at the effects on both monthly and hourly wages. We repeat the main specifications for women and men to capture the gender segmentation in the labor market.

Wages
The first row in column (1) shows that a 1% increase in the regional fraction of workers with no high school degree who earned less than 1300 Turkish liras but more than 711 Turkish liras in 2015 increased the average monthly wages of workers with no high school degree in 2016 by 0.32%. Columns (2) and (3) show that this overall monthly wage effect is mainly due to positive and substantial wage effects on less-educated men. When we look at hourly wages in columns (4)-(6), we observe that the hourly wage effects on less-educated men are larger than the monthly wage effects. When it comes to the wage effects on workers with a high school degree, we observe larger overall wage effects despite smaller wage effects on men, hence the positive and noticeable wage effects on women with a high school degree. Among workers with tertiary education, wage coefficients are considerably larger, especially for men.
Given that the proportion of affected workers is lower within this education group, the result may seem unexpected. Several explanations may be proposed. First, the minimum wage is still a reference wage for tertiary-educated but informal workers. Second, the positive wage estimations for this group partly reflect a phenomenon of the Turkish labor market. Figure 3 shows that a noticeable share of the tertiary educated group earns around the minimum wage level. Third, although we control for age and time-varying heterogeneities, the largest wage effect estimated for this group is likely to be spurious. To avoid an ad-hoc selection of sample period, we preferred to start the analysis from 2010, as 2009 is the first year where we can identify public sector employees in the data set. 8 When we start the analysis from 2012, instead of 2010, as shown in Figure B1 in Appendix, wage coefficients for men with tertiary education (panel A) decline and are no more than other education groups. Figure B1 in Appendix also shows that the use of a different sample does not lead to a significant change in the wage coefficients estimated for no high school (panel B) and high school sub-groups (panel C). In other words, our sample period choice results in a potential bias in wage effects estimated for the tertiary-educated group, while this is not the case for workers with no high school degree or high school-educated workers. We further elaborate on this issue in Figures B2, B3, and Accompanying Discussion in Appendix.
We examine the wage effects by age groups in the lower panel. Columns (1) and (2) highlight positive wage effects on young (15-24) workers. The positive wage effects are due to young men's wages, and the coefficients are somewhat smaller. Hourly wage effects are larger than monthly wage effects. Small wage effects on young workers stem from relatively large informality effects on these groups. Below, we provide suggestive evidence to that end.
Meanwhile, overall wage effects are larger within other age cohorts and increase generally as age increases. The two age groups in which the wage effects are larger for women than that for men are 25-34 and 35-44 cohorts. Hourly wage effects on men are larger than monthly wage effects on men in all cases. But for women, this is generally true for elder women.
To sum up, we find economically large and statistically significant wage effects on most groups. Our findings suggest no increase in the wages of women with no high school degree, young (15-24) women, and elder women (45-64), but noticeable increases in the wages of other demographic groups. Hourly wage effects are generally larger, implying that employers may not be able to surpass the monthly minimum wage statute by inducing higher working hours. Supplementary evidence presented in Figure A4 in Appendix shows that, in most cases, there were no such considerable increases in the wages following the minimum wage increases that took place before 2016, hence the 33% increase in the minimum wage in 2016 may be seen as a positive wage shock. Table A4 in Appendix suggests that the shock had positive and relatively large wage effects on also informal men, supporting the "lighthouse effect" of the minimum wage increase. Overall, it is possible to suggest that a 1% increase in the fraction of workers supposed to be affected by the shock allows, on average, around a 0.4% increase in average wages. Table 3 shows the effects on full-time wage employment in the private sector. We replicate the same structure as in Table 1

Informality
We examine informality outcomes in Table 4. The dependent variable is a dummy variable taking 1 if salaried, casual, or waged individual is working full-time in the private sector without social security, 0 otherwise. We replicate the main specification for small firms with less than 10 employees, since it may be easier for smaller firms to avoid regulations while larger firms are expected to be subject to closer scrutiny by the government. Hence, we want to investigate whether small firms have a higher ability to avoid the minimum wage regulations by employing workers without social security in case of an increase in the minimum wage and to quantify the labor market discrepancies between small and large firms.
Column (1) in the upper panel shows that the regions with a higher fraction of lessthan-high-school-educated workers earning less than 1300 Turkish liras but higher than 711 Turkish liras in 2015 experience a relative increase in informality of workers with no high school degree following the minimum wage increase in 2016. Columns (2) and (3) suggest that this relative increase in informality of workers with no high school degree is not attributable to a particular gender. When we focus on only small firms, we observe an increase in the positive informality effects on workers with no high school degree. More importantly, this relative increase is mainly in the informal employment of women with no high school degree. The informality effect is smaller when it comes to workers with a high school degree.
In all cases, the minimum wage effect on informality is statistically insignificant. However, statistical insignificance is driven by large standard errors and not because of small coefficient estimates. As a result, we do not want to disregard the potential effect on informality based solely on statistical significance.
The lower panel shows that the informality coefficients are considerably larger for young workers aged 15 to 24 and become even larger following the exclusion of firms employing more than 10 workers. Moreover, in this sample young women's informality increases disproportionally. These are economically large effects, though the standard errors are relatively large. Within other age cohorts, we find neither statistically nor The key variable of interest is the fraction of affected workers (the ratio of workers earning higher than 75% of the minimum wage level in year t-1 but lower than the new minimum wage level in year t). Each cell shows the estimates for the interaction of the 2016 dummy and the fraction of affected workers in 2015.
Workers with Higher degree are those with 2-or 3-years higher education, or 4 years higher education, or master's degree (5-or 6-years faculty included) or doctorate. There are 26 NUTS-2 regions. All coefficients are relative to the interaction of the 2010 dummy and the fraction of affected workers in the 2009 variable. In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations. ***, **, and * indicate a statistical significance level at the 0.01, 0.05, and 0.1 level, respectively. economically significant informality effects. Following the exclusion of large firms with more than 10 employees, however, the increase in informality effect on workers aged 45 to 54 becomes worth noting.
In short, our evidence suggests positive and relatively large informality effects on workers with no high school degree. The effect becomes economically more important when we focus on only small firms, especially for women with no high school degree. We find also positive but smaller informality effects on workers with a high school degree. The key variable of interest is the fraction of affected workers (the ratio of workers earning higher than 75% of the minimum wage level in year t-1 but lower than the new minimum wage level in year t). Each cell shows the estimates for the interaction of the 2016 dummy and the fraction of affected workers in 2015. Workers with Higher degree are those with 2-or 3-years higher education, or 4 years higher education, or master's degree (5-or 6-years faculty included) or doctorate. There are 26 NUTS-2 regions. All coefficients are relative to the interaction of the 2010 dummy and the fraction of affected workers in the 2009 variable. In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations. ***, **, and * indicate a statistical significance level at the 0.01, 0.05, and 0.1 level, respectively.

Robustness
Our main exposure variable, the fraction of workers earning lower than the new minimum wage level in year t but higher than 75% of the minimum wage level in year t-1, include informally employed. Figure A7 in Appendix shows that regions such as TRA2, TRB2, TRC2, and TRC3 had a very large share of informal workers and a higher share of workers earning lower than 711 Turkish liras in 2015 in full-time wage employment. Hence, it is reasonable to expect that the variation in the fraction of workers may be affected by the cross-region differences in the share of the informal sector, leading to biased estimates. To handle this issue, first, in the first two columns of Tables A1-A3 in Appendix, we define an alternative exposure variable "fraction of formal affected workers" by excluding informal workers from the numerator of the exposure variable. Second, in columns (3)-(5), we adopt a more conservative exposure variable, "fraction at minimum wage", i.e. the proportion of employees working at the minimum wage in year t-1. For example, in this case, workers earned between 902TL and 1050TL, not between 711TL and 1300TL, in 2015 are expected to be affected by the minimum wage increase in 2016.
In doing so, we eliminate the possibility that the variation in the proportion of workers who earn lower than minimum wage level in year t-1 is affected by the variation in the proportion of informal workers. Finally, in the models presented in the last two columns in Tables A1-A3 in Appendix, we exclude outlier regions TRA2, TRB2, TRC2, and TRC3 in which the share of informality is noticeably higher. 9 Comparison of the baseline results presented in Tables 2-4

and the results presented in
Tables A1-A3 in Appendix reveals that monthly wage coefficients are still statistically different from zero and become larger in most cases where we exclude informally employed from the numerator of exposure variable. When it comes to employment outcomes, in two of three alternative definitions of fractions, we observe negative coefficients for men with less than high school education (−0.068 (column 5 in Table A2 of the Appendix) and −0.078 (column 8 in Table A2 of the Appendix)). Only the latter is statistically significant. These negative employment effects for men with no high school degree are accompanied by positive coefficients for men with high school or higher degrees (albeit none of the coefficient estimates for these later groups are statistically significant). At first, these findings suggest that less-skilled men are substituted by more skilled men after the minimum wage increase. However, we do not observe a similar pattern for women. On the contrary, coefficient estimates for women are almost always positive for every education level (albeit mostly insignificant) and coefficient sizes are largest for women with less than high school education. Taking these findings together does not support a skillbiased substitution explanation. If men with less than a high school education are more likely than other groups (more educated men and women in general) to work in tradable goods sectors, a minimum wage increase can explain their relative employment losses. However, such an investigation is beyond the scope of this paper.
The alternative exposure variable excluding informal workers from the numerator results in a small decrease in the positive informality effect on workers with no high school degree, an increase in the informality of workers with a high school degree, and of workers aged 25 to 34.
And the exclusion of outlier regions leads to larger positive informality on workers aged 15 to 24.
In conclusion, large positive wage effects seem to be robust to alternative estimations. We find also that positive informality effects on less-experienced or less-educated groups are not sensitive to alternative estimations in most cases. For employment outcomes, only one of the alternative models contrasts with no employment findings on demographic groups supposed to be more affected by an increase in the minimum wage. Considering negative employment elasticities reported by the studies suggesting large and negative employment effects of minimum wage policy, however, statistically significant negative employment elasticities reported in our alternative specifications seem to be economically insignificant.

Concluding remarks
Using an identification strategy based on regional variation in the fraction of workers affected by minimum wage increases, we investigated labor market outcomes of the quasi-experimental minimum wage increase of 2016 in Turkey. Turkey presents an interesting case as its economy displayed a low employment generation capacity in recent decades; a large number of workers are employed at the minimum wage; and the national minimum-wage-to-average-wage ratio is one of the highest among OECD countries. Moreover, employment without social security is still widespread, the gender wage gap is persistent, there are noticeable regional inequalities in labor market outcomes, and the labor market plays an active role in income and poverty dynamics.
We showed that an important share of young workers (ages between 15 and 24) and workers with no high school degree in the formal sector work around the minimum wage level, but also that there are many demographic groups in which a considerable share of workers worked around the minimum wage level and expected to be affected by an increase in the minimum wage. Armed with this information, we find that the 2016 minimum wage increase had economically substantial and statistically significant positive effects on wages of most demographic groups. This is true also for informal workers, supporting the "lighthouse effect" of the minimum wage increase. Our wage specifications show also that hourly wage coefficients are larger in most cases, especially for men, hence suggest that employers may not able to surpass the monthly minimum wage statute by imposing longer working hours. We find substantial wage effects particularly on men aged 45-64 and on men with tertiary education, that is on demographic groups expected to be less affected by the minimum wage increase.
Although such large wage coefficients are partly spurious because of our sample period choice, the analysis presented in this study reveals that a considerable share of higher-skilled groups works around minimum wage, hence that the positive wage effects on demographic groups expected to be less affected by an increase in the minimum wage are likely to reflect a phenomenon of the Turkish labor market. This probable lack of skill premium deserves further investigation however it is beyond the scope of this paper.
Our findings indicate that the minimum wage increase did not have a negative impact on employment outcomes. However, the increase induced an economically considerable positive effect on the share of workers working without social security among workers with a higher school degree and no high school degree, potentially indicating that the minimum wage increase led to an increase in informal employment. This informality effect is stronger when we focus on small firms, employing less than ten employees.
These findings appear to be robust to alternative estimations that take into account the potential biases stemming from our key variable of interest. Yet, some limitations should be emphasized. First, we focus on the short-run effects of the minimum wage increase, though it is recognized that the short-run effects may differ from the effects in the longrun. Second, since the data we use do not allow us to estimate the effects on consumption and income dynamics, our findings on the large and positive wage effects of the minimum wage increase provide only indirect information about the effect of the minimum wage policy on income. Given the studies that emphasized the important role played by the labor market in income and poverty dynamics in Turkey (e.g. Şeker and Dayıoğlu, 2015), however, the importance of this finding in addressing the role of active labor market policies in income and poverty trends should not be underestimated. More importantly, our study has important implications for the minimum wage debate in the developing country context, since we discover noticeable heterogeneities in the effects of the minimum wage increase and find that ignoring gender discrepancy in a developing country labor market when analyzing the effects of the minimum wage may lead to biased results; that the effects of the minimum wage policy may be closely related to cross-firm differences in productivity and labor market regulations; and that previously ignored demographic groups may provide invaluable identifying information.

Declarations Availability of data and material
We use the annual Household Labor Force Survey (HLFS) for 2009-2016, a cross-sectional restricted-usemicro-data gathered by Turkstat to produce official labor market indicators and to provide information about demographic characteristics, employment status, income, and past work experience of household members. Since our estimations and calculations are based on this restricted-use-micro-data, we are not able to provide full access to the data. Access to our replication package is available upon request. Taking the minimum wage level in the first half of the year as the minimum threshold and the minimum wage level in the second half of the year as the maximum threshold, and allowing a 5% error margin, we discern minimum wage workers. Generalization of the approach yields the following wage condition for minimum wage workers: MW 1t -(MW 1t * 0.05) < Y it < MW 2t + (MW 2t *0.05), where MW 1t is the minimum wage level of the first half of the year t, MW 2t is the minimum wage level of second half of the year t, Y it . is the wage of the worker i in year t. There are some legitimate reasons for workers in the formal sector to earn less than the minimum wage during the survey month. The question relating to earnings asks respondents their earnings in the previous month and some full-time workers might have worked less than whole month because i) they started a new job in the middle of the previous month, ii) they got an unpaid leave to deal with family emergencies; or iii) the workplace was temporarily closed. Moreover, if the respondent was interviewed in January, she is reporting her wages from December of previous year, which is probably less than the 95% of new minimum wage. As can be seen from the figure, the share of formal sector workers with less than minimum wage is stable over the years for all education groups suggesting that these are idiosyncratic issues independent from macroeconomic trends. We thank İnsan Tunalı for pointing out these data issues. There are some legitimate reasons for workers in the formal sector to earn less than the minimum wage during the survey month. The question relating to earnings asks respondents their earnings in the previous month and some full-time workers might have worked less than whole month because i) they started a new job in the middle of the previous month, ii) they got an unpaid leave to deal with family emergencies; or iii) the workplace was temporarily closed. Moreover, if the respondent was interviewed in January, she is reporting her wages from December of previous year, which is probably less than the 95% of new minimum wage. As can be seen from the figure, the share of formal sector workers with less than minimum wage is stable over the years for all education groups suggesting that these are idiosyncratic issues independent from macroeconomic trends. We thank İnsan Tunalı for pointing out these data issues.  In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations. Each row shows the estimates for the interaction of the year t dummy and the fraction of affected workers in year t-1. All coefficients are relative to the interaction of the 2010 dummy and the fraction of affected workers in 2009 variable. In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations. In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations

Figure A7
The regional fraction of workers earned less than 711 Turkish liras and the regional informality ratio .   TR10   TR21   TR22   TR31   TR32   TR33   TR41   TR42   TR51   TR52   TR61   TR62   TR63   TR71   TR72   TR81   TR82   TR83   TR90   TRA1   TRA2   TRB1   TRB2   TRC1   TRC2   TRC3   0  10  20  30  40   Formal numerator is the ratio of formal workers earned higher than 75% of minimum wage level in year t-1 but lower than new minimum wage level in year t in full time wage employment in private sector. Fraction at is the ratio of minimum wage workers in t-1 whose wages are lower than 110% of the minimum wage but higher than 90% of the minimum wage. The last three columns show the models where TRA2, TRB2, TRC2, and TRC3 NUTS-2 regions are excluded. Each cell shows the estimates for the interaction of the 2016 dummy and the fraction of affected workers in 2015. Workers with Higher degree are those with 2-or 3-years higher education, or 4 years higher education, or master's degree (5-or 6-years faculty included) or doctorate. There are 26 NUTS-2 regions. All coefficients are relative to the interaction of the 2010 dummy and the fraction of affected workers in 2009 variable. In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations. ***, **, and * indicates statistical significance level at the 0.01, 0.05, and 0.1 level, respectively. Formal numerator is the ratio of formal workers earned higher than 75% of minimum wage level in year t-1 but lower than new minimum wage level in year t in full time wage employment in private sector. Fraction at is the ratio of minimum wage workers in t-1 whose wages are lower than 110% of the minimum wage but higher than 90% of the minimum wage. The last three columns show the models where TRA2, TRB2, TRC2, and TRC3 NUTS-2 regions are excluded. Workers with Higher degree are those with 2-or 3-years higher education, or 4 years higher education, or master's degree (5-or 6-years) or doctorate. There are 26 NUTS-2 regions. All coefficients are relative to the interaction of the 2010 dummy and the fraction of affected workers in 2009 variable. In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations. ***, **, and * indicates statistical significance level at the 0.01, 0.05, and 0.1 level, respectively. . Dependent variable is a dummy variable taking 1 if an individual is a wage earner and working without registered in the Social Security Institution in full-time in private sector. Formal numerator is the ratio of formal workers earned higher than 75% of minimum wage level in year t-1 but lower than new minimum wage level in year t in full time wage employment in private sector. Fraction at is the ratio of minimum wage workers in t-1 whose wages are lower than 110% of the minimum wage but higher than 90% of the minimum wage. The last three columns show the models where TRA2, TRB2, TRC2, and TRC3 NUTS-2 regions are excluded. Workers with Higher degree are those with 2-or 3-years higher education, or 4 years higher education, or master's degree (5-or 6-years) or doctorate. Each cell shows the estimates for the interaction of the 2016 dummy and the fraction of affected workers in 2015. Workers with Higher degree are those with 2-or 3-years higher education or faculty, or 4 years higher education, or master's degree (5-or 6-years faculty included) or doctorate. There are 26 NUTS-2 regions. All coefficients are relative to the interaction of the 2010 dummy and the fraction of affected workers in 2009 variable. In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations. ***, **, and * indicates statistical significance level at the 0.01, 0.05, and 0.1 level, respectively. The key variable of interest is the fraction of affected workers (the ratio of workers earned higher than 75% of minimum wage level in year t-1 but lower than new minimum wage level in year t). Each cell shows the estimates for the interaction of the 2016 dummy and the fraction of affected workers in 2015. Workers with Higher degree are those with 2-or 3-years higher education, or 4 years higher education, or master's degree (5-or 6-years) or doctorate. There are 26 NUTS-2 regions. All coefficients are relative to the interaction of the 2010 dummy and the fraction of affected workers in 2009 variable. In "by education" estimations, age is controlled using dummies for 15-24, 25-34, 35-44, 45-54, and 55-64 age cohorts. Education is controlled using dummies for less than high school, high school, and tertiary education groups in "by age" estimations. ***, **, and * indicates statistical significance level at the 0.01, 0.05, and 0.1 level, respectively.   TR10  TR21   TR22  TR31  TR32   TR33   TR41  TR42   TR51   TR52   TR61   TR62 TR63   TR71   TR72   TR81   TR82   TR83   TR90  TRA1   TRA2  TRB1   TRB2   TRC1  TRC2  TRC3   TR10   TR21   TR22   TR31   TR32   TR33   TR41   TR42   TR51   TR52   TR61   TR62   TR63   TR71   TR72   TR81   TR82   TR83   TR90  TRA1   TRA2   TRB1   TRB2   TRC1   TRC2   TRC3   TR10   TR21   TR22   TR31   TR32   TR33   TR41  TR42   TR51   TR52   TR61   TR62   TR63   TR71   TR72   TR81   TR82   TR83  TR90   TRA1  TRA2   TRB1   TRB2   TRC1   TRC2   TRC3   TR10   TR21   TR22   TR31   TR32   TR33   TR41   TR42   TR51  TR52  TR61   TR62   TR63  TR71   TR72   TR81   TR82   TR83   TR90  TRA1   TRA2   TRB1   TRB2   TRC1   TRC2  To read the figure for example, around 60% of men with no high school degree earned higher than 711TL but lower than 1300TL in 2015 and wage growth from 2015 to 2016 for this group was around 25%.