2.1 Introduction
In spite of abundant minimum wage studies, it remains an open question empirically as to how a minimum wage increase affects wage inequality. Recently, Autor et al. (2016) use three decades of data after 1980 to reassess the study of the minimum wage effect on the US earning inequality that was done by Lee (1999), who found that the falling relative level of the US federal minimum wage can account for the growth in inequality not only
in the lower tail but also the higher tail of the wage distribution. Although they use an IV strategy eliminating most of the ripple effects found by Lee, they still observe spillover effects which they argue are a result of a measurement error, but needs to be proved with ideally administrative payroll data.
In fact, other than spillover effects, influence of the minimum wage on wage inequality also depends on the effects—both the wage and employment effects—on workers who are directly affected by the minimum wage. Studies using the wage distribution to identify the minimum wage effect on the wage inequality, such as these two influential studies, need to be based on the assumption of no employment effect on workers bound by the minimum wage in their sample.15 However, the assumption may not hold, especially when minimum wages increase and affect more than teenagers and other minority workers.
In a more recent working paper, Cengiz et al. (2018) used a bunching-based method to estimate the wage frequency distribution and proved that there were spillover effects of minimum wage increases in the administrative data with no disemployment effect at the same time, which is inferred by tracking the changes in the number of jobs throughout the wage distribution. The drawback of this method is that under the accumulated number it failed to identify who gained or lost jobs, which is rather a critical issue in the minimum wage research. In other words, to answer the open question about the wage inequality effect resulting from minimum wage increases, other than keeping focusing on the wage distribution, it would be helpful to go back to the individual-based analysis—how are the lowest-wage workers affected by the minimum wage increases, and also what are the effects on workers who earn slightly above the new minimum?
However, previous literature fails to answer those questions due to combined issues in three key aspects. First, there is a paucity of research on the directly affected work-ers, which is the lowest-wage group (Belman et al., 2015; Manning, 2016; Jardim et al., 2017; Boffy-Ramirez, 2019). Second, similar to the first but not identical, previous stud-ies mostly include directly affected workers and indirectly affected workers with slightly higher wages than the minimum. As a result, both the direct and indirect effects, which may result from the labor-labor substitution, the scale effect, or the compensation hierar-chies, cannot be identified (Neumark, 2018). Third, the empirical methodology needs to
15Lee (1999) only focused on the observed wage distribution without accounting for employment effects caused by the minimum wage. Autor et al. (2016) argued that limiting their sample to 25-64 years old can prevent their findings from being affected by disemployment effects.
rely on a valid counterfactual, which could be contaminated by unobservable confounders, trends, or impacts caused by minimum wage changes (Allegretto et al., 2017; Neumark and Wascher, 2017; Totty, 2017). These three aspects have been discussed separately in prior studies more or less, in particular the methodology. But to my knowledge, there has not been a study dealing with all of them yet. Actually, without the separation of workers with different levels of wages, the methodology cannot prevent attenuated estimates of the effect as they blend workers and rule out unobservable impacts of minimum wage in the counterfactual.
This study aims to pin down the minimum wage effects on the earning inequality by providing firm- and individual-level evidence built on an unique method containing several features. First, the changes in the wage and employment of bound and unbound workers are separated and tracked in this study. This study has the advantage of being based on monthly administrative personnel longitudinal data of Taiwan, the monthly record of the mandatory labor and employment insurance run by the government. Second, rather than using the method of difference in difference, I employ the approach of sharp regression discontinuity (RD) design. This high frequency data along with there being only one increase in the national minimum wage during the research period allow me to use RD to avoid the need of creating a valid counterfactual. Third, this study also analyzes the difference between individuals rather than the difference between percentiles in the wage distribution.
To my best knowledge, this is the first study using high quality personnel administrative data to estimate the direct and indirect effects of a minimum wage increase simultaneously, hence providing individual-level evidence for the wage inequality effect and contributing to the literature examining the wage difference in the wage distribution (Dinardo et al., 1996; Lee, 1999; Autor et al., 2016; Cengiz et al., 2018). Meanwhile, the research also contributes to the literature on employment effects by dealing with the issues of research targets (Neumark and Wascher, 2007; Sabia et al., 2012; Belman et al., 2015; Manning, 2016; Jardim et al., 2017; Neumark, 2018; Boffy-Ramirez, 2019), the methodology (Alle-gretto et al., 2017; Neumark and Wascher, 2017; Totty, 2017; Boffy-Ramirez, 2019), and the employment flow (Brochu and Green, 2013; Dube et al., 2016). In addition, it also con-tributes to the literature on spillover effects (Gramlich, 1976; Grossman, 1983; Akerlof and Yellen, 1990; Katz and Krueger, 1992; Spriggs, 1993; Card and Krueger, 1994; Neumark
et al., 2004; Clemens et al., 2018; Cengiz et al., 2018; Dube et al., 2019), in which recent studies find positive wage spillover effect and refer to the equity effect (Cengiz et al., 2018;
Dube et al., 2019). Finally, since the inequality has worsened in most of the countries since the 1980s, and the minimum wage is a common tool universally, results from Taiwan also enrich the literature.
To explain the effect on wage inequality, the hypothesis of the "equity effect" for the spillover effect plays a main role in recent studies, such as Giupponi and Machin (2018) and Dube et al. (2019). This effect reflects the firm’s response to maintain the wage structure, given workers’ efforts depend on relative wages rather than absolute ones. Under this hypothesis, firms simply raise overall wages and product prices to mitigate the increase in the minimum wage (Gramlich, 1976; Grossman, 1983; Akerlof and Yellen, 1990; Katz and Krueger, 1992). Recently, Dube et al. (2019) used data from a large US retailer showing that workers with wages up to 15% higher than the new minimum had raises after the minimum wage increase because of the firm’s policy. If it is the case generally, there should be no employment effect but wage increases, then using the wage distribution to analyze earning inequality would not be problematic as described above.
In this study, I began by generating the matched employee-employer data to analyze the changes of employment and average wages in firms to see if this hypothesis is supported by the data. By using the RD approach with the monthly periods as the running variable, I do not have to assume a counterfactual. Instead, the assumption behind the methodology is that the labor market outcomes are continuous around the cutoff time, which is the time of the implementation of the new minimum. If a discontinuity exists, it reflects the exogenous policy impacts. In other words, I use the time cutoff as an instrument variable to replace the minimum wage variable when identifying the causal effects of the minimum wage increase. Compared to assuming the counterfactual has a similar trend but is totally unrelated to the minimum wage change, the assumption in this study is weaker.
The estimates from the firm-level analysis initially show that there is a significant positive effect on the average nominal wage and no significant employment effect on full-time workers of all firms. Those results are in line with the expectation of the hypothesis.
However, the results do not hold when it comes to the real wage and the employment of small scale firms, which were more likely to be affected by the minimum wage increase. My best estimates show that the average real wage significantly decreased by 1% compared to
the prior level, and the employment effect is mixed but apparently not null. In a short bandwidth, about six months before and after the minimum wage increase, the lowest-wage full-time employment including the bound workers and the new minimum was increased by 1%, and the slightly higher-wage employment was increased by 2%; in a longer bandwidth, about 10 months before and after, the lowest-wage employment was decreased by 1%
but may result from minimum wage employers moving to higher-wage positions rather than disemployment, since the higher-wage employment was increased more and the total employment was increased by 2%. In addition, the part-time employment was increased by 12% to 25% from short bandwidth to longer bandwidth.
Except for the employment change, since the increase in the minimum wage exceeded the consumer price index, the decline of the average real wage may come from the unbound workers, which is opposite to the hypothesis. I then transfer my focus from employment of wage positions to individual workers to clarify from where the effects came.
To identify the effects on individuals, I firstly show the challenge of classifying work-ers by using observable wages. Since the workwork-ers’ wage information only can be observed given they are employed, using wages just prior to the increase can lead to missing the effects of minimum wages on transitions from non-employment to employment (Neumark, 2018). Previous studies usually used workers’ previous average wage to do classification (Clemens and Wither, 2014, 2017). However, it may mix the effects from directly af-fected and indirectly afaf-fected workers. In addition, since the outcomes—the wage and the employment—are also used to classify groups of workers, I show that using more restrictive information to select groups may lead to artifacts.
Workers with previous wages bound by the new minimum all the time or just bound before the cutoff time had a dramatic decline in employment after the increase. This is not due to the policy effect, but instead is a result of the target group being restricted to those employed with low wages previously instead. It also leads to an artifact of wage changes.
To avoid the influence from the restriction, I use the earliest wage information, about forty months before the increase, to classify workers, and then track their wages and status of employment overtime. The results show that the former bound workers experienced signif-icant increases on their nominal and real wages and even on their employment probability.
As a result, their real earnings, including the change of their employment, were increased.
On the other hand, the former unbound workers had their employment probability raised
but with an adverse effect on their real wages. The two opposite effects lead to total real earnings having no significant change.
The drawback of this classification strategy is the mixing of the direct and indirect effects, since the former status may change after forty months. To solve this problem, I then use the wage distributions before the minimum wage increase for those two groups of workers to separate both direct and indirect effects. The additional assumptions are that there is no effect on workers with much higher wages than the unbound workers I define, and all the workers share the same unemployment rate. Even after this adjustment, all the directions of results stay the same but with larger and more distinctive estimates of effects on the two groups of workers. My estimates show that the employment of full-time workers bound by the minimum wage is significantly increased by 1% to 4% due to the minimum wage increase, while their real wage is increased by 1% to 3%. On the other hand, the employment of workers who had wages slightly higher than the new minimum wage is also increased by 3% to 7%, but their real wage is decreased by 6% to 11%. The decrease reflects that they lose their potential nominal wage increase and suffer because of the hike of the CPI rather than having their wages cut. As a combining result of the employment and wage effect, the bound workers have their real earnings increased by 4%
to 5% but the unbound workers have no significant change on their real earnings.
The positive employment effect on the low-wage workers is not so striking, since it has also been found by some recent studies (Dolton et al., 2012; Giuliano, 2013). The most surprising part is that there is no wage spillover effect but a negative influence on the unbound workers’ real wages. The result is opposite to the results shown by Dube et al.
(2019), which used data from a large individual firm, and Cengiz et al. (2018), in which they focused on the wage distribution. By tracking individuals to identify effects, this study at least points out an alternative expectation compared to those results. Meanwhile, the results show that there is no spillover effect on the earning inequality, also different from the studies focusing on the wage distribution (Lee, 1999; Autor et al., 2016).
In addition to estimating separated effects on bound and unbound workers, I further analyze the difference between the former bound and unbound workers across time. Com-pared to separated estimates, this method can rule out factors that may influence both kinds of workers. I show that differences in nominal wages, real wages, and earnings be-tween them had discontinuous declines around the cutoff time, which means the wage and
the earning inequalities were narrowed. The effects persisted during the remainder of our data period, forty months after the minimum wage increase. Since the wage increase of bound workers was determined by their previous nominal wages and the new minimum, I then test whether the unbound workers had any nominal wage change after the increase by using the wage distribution prior to the wage increase to generate the expected wage increase amount. The result also shows that rather than increasing, the unbound work-ers’ nominal wage was significantly lower than before, which still proves that there is no spillover effect on the wage inequality.
Since the results I show are not in line with the hypothesis of the equity effect (Gramlich, 1976; Grossman, 1983; Akerlof and Yellen, 1990; Katz and Krueger, 1992; Dube et al., 2019) or the demand shift resulting from the labor-labor substitution effect expected by standard economic theory, the next question is how did it happen. The most influential alternative theories behind the minimum wage studies include monopsony power (Burdett and Mortensen, 1998; Bhaskar and To, 1999; Manning, 2003; Flinn, 2006; Ahn et al., 2011) and search models (Ahn et al., 2011; Brochu and Green, 2013; Giuliano, 2013; Gittings and Schmutte, 2016). In more recent studies, Brochu and Green (2013) and Dube et al.
(2016) found that separations and accessions rates among potentially affected workers were reduced because of minimum wage increases, which is consistent to search models’
expectations. Following the studies, I use the matched employee-employer data to test the employment flow to check if results are in line with the search models.
Different from Dube et al. (2016), in which they used border discontinuity design, I use RD to observe the changes of the separation and new hire rates of the full-time and part-time workers among small scale firms. Rather than finding a robust decline on those rates, the results are mixed and show robustly contrary effects on part-time workers. The full-time separation rate was decreased by 3.3% in a longer bandwidth, but there was no significant change in a shorter one; the new hire rate had no significant change in a longer bandwidth but was increased by 39% in a shorter bandwidth. Turning to part-time employment, results show that there were higher separation and new hire rates after the minimum wage increase no matter the length of the bandwidth. Compared to the prior level, the separation rate on part-time workers was increased by a percent in the low 30s, and the new hire rate was increased by a percent in the high 30s to the low 40s. Apparently, the search friction previous studies expected may not explain this result.
Through the advantage of identifying individual workers, I further generate the transi-tion matrix of the former bound workers during the research period. I classify five employ-ment statuses, including full-time employed with the lowest wage, full-time employed with a low wage, full-time employed with a high wage, part-time employed, and non-employment.
I track the individuals’ transitions among the five statuses between every month, hence there would be twenty-five situations at every time. This approach helps to identify the transition changes more directly.
I find that the full-time employed workers have a steady probability of staying in the original statuses after the minimum wage increase. It means that their separation does not decline, which is in line with the finding in the short run firm-level analysis. On the other hand, full-time workers have a higher probability of transferring to part-time jobs although the scale of transferring to part-time jobs is too small to affect the main transition.
However, one month after the full-time workers transit to part-time jobs, the probabilities of part-time workers transferring to full-time employment and non-employment are increased.
These higher inflow and outflow on the part-time employment are also consistent to the firm-level finding. In other words, even the individual transition still rejects the possibility of the search friction hypothesis.
Although the transition matrix cannot directly prove the non-employed but potentially low-wage workers’ (including workers bound by the minimum and slightly above the new minimum) transition, their higher employment along with the steady separation rates indicate the increased employment should be from non-employment. The entry of the bound full-time workers and the transition from full-time to part-time are more close to the theory of the traditional monopsony theory, in which the labor market is "supply-constrained" (Card and Krueger, 1994).
However, the increased entry to full-time employment with wages slightly higher than the new minimum wage may be from the contribution of labor demand or labor supply.
Since their growth in employment coincides with a decrease in their real wages, there seems to be an outward shift of the labor supply. Previously, Neumark et al. (2004) found similar results but attributed it to slightly higher-wage workers needing to work more to cover their lower-wage family members’ disemployment due to the minimum wage increase.
That explanation is obviously not valid here since there is no apparent disemployment effect on the bound workers. That is, even though the phenomenon is really from a shift of the