EMPLOYMENT EFFECTS OF THE MINIMUM WAGE: PANEL DATA EVIDENCE FROM CANADIAN PROVINCES

Transcription

1 EMPLOYMENT EFFECTS OF THE MINIMUM WAGE: PANEL DATA EVIDENCE FROM CANADIAN PROVINCES KATE RYBCZYNSKI and ANINDYA SEN Recent U.S. studies offer conflicting evidence on minimum wage impacts. This paper studies the effects of 185 amendments to minimum wage on employment rates using panel data across Canadian provinces from 1981 to Ordinary least squares and instrumental variables (IV) estimates imply a 10% increase in minimum wage is associated with a 1% 4% reduction to employment rates for both male and female teens. We also find that an increase in the minimum wage is associated with lower employment of prime-aged immigrants. Our results are robust to a wide array of IV and the use of controls for spatial heterogeneity. (JEL J30, J71, J23) I. INTRODUCTION There is a considerable amount of U.S.-based research on minimum wage effects. Most studies focus on estimating the effects of minimum wage amendments on employment among teens or among workers in industries with a significant proportion of minimum wage earners (such as restaurants). Currently, there is a sharp divide among these studies on whether increases to the minimum wage adversely impact the employment of teens and/or unskilled workers. This study contributes to the literature by investigating the impacts of amendments to the minimum wage on employment rates of teens and primeaged adults, by gender, with a panel of Canadian provinces between 1981 and We also investigate the effects of changes in minimum wage legislation on older teens, and on immigrants by collecting data from the annual waves of the Survey of Labour and Income Dynamics (SLID) and Survey of Consumer Finances (SCF), which allows us to construct province-year cells of employment rates for these population subgroups. We acknowledge excellent research assistance by Wai Hong Choi, Scott Legree, and Andrew Alkhani, and useful comments from two knowledgeable referees. Rybczynski: Associate Professor, Department of Economics, University of Waterloo, Waterloo, ON N2L3G1, Canada. Phone Ex , Fax , E- mail krybczyn@uwaterloo.ca Sen: Professor, Department of Economics, University of Waterloo, Waterloo, ON N2L3G1, Canada. Phone Ex , Fax , asen@uwaterloo.ca The issue of minimum wage effects is of contemporary policy relevance in Canada given national discussions on a $15 per hour minimum wage and the recent commitment by the province of Alberta to implement a $15 minimum wage by Furthermore, the use of Canadian data offers a particularly interesting laboratory as the time period witnessed some rather steep increases to the minimum wage for multiple Canadian provinces, resulting in rich crossprovince and time-series variation with which to identify minimum wage effects. Specifically, our sample allows us to exploit identification offered by 185 amendments enacted by 10 provinces. As a result, our analysis may offer some insight on the current debate of the employment effects of increases to the minimum wage. We think 1. See Alberta makes $15 minimum wage regulations official by Dean Bennet, published by the Toronto Star at: alberta-makes-15-minimum-wage-regulations-official.html. Accessed September 13, ABBREVIATIONS CPS: Current Population Survey GDP: Gross Domestic Product IV: Instrumental Variables LFS: Labour Force Survey Lib: Liberal Party NDP: New Democratic Party OLS: Ordinary Least Squares PC: Progressive Conservative Party SCF: Survey of Consumer Finances SEPH: Survey of Employment, Payrolls, and Hours SLID: Survey of Labour and Income Dynamics Contemporary Economic Policy (ISSN ) Vol. 36, No. 1, January 2018, Online Early publication July 18, doi: /coep Western Economic Association International

2 RYBCZYNSKI & SEN: EMPLOYMENT EFFECTS OF THE MINIMUM WAGE 117 this is important given the sharp dichotomy in minimum wage elasticities suggested by recent U.S. studies, and the corresponding debate on the relevance of jurisdiction-specific trends in order to account for unobserved spatial heterogeneity. Our paper evaluates the sensitivity of estimates to the use of province and year-fixed effects as well as provincial linear trends and higher-order trends. We also assess the magnitude of potential simultaneity bias, through the use of instrumental variables (IV) analyses. Finally, there are relatively few studies that have estimated the effects of minimum wage amendments on employment rates across population subgroups such as women and immigrants (see Orrenius and Zavodny 2008). This is a relevant exercise given the historical objective of minimum wage legislation: protection for vulnerable groups. Canadian data are appropriate in this respect as the incidence of recent immigrants earning minimum wage has risen rather sharply from the late 1990s onward. Moreover, the Minimum Wage Advisory Panel (2014) finds that recent immigrants, defined as having arrived in Canada within the past 10 years, as a proportion of the total minimum wage workforce has risen from 6.9% in 1998 to 19.1% in Our estimates reveal that amendments to the minimum wage result in lower employment rates for male and female teens, with an absence of statistically significant gender differences. Specifically, our estimates imply that a 10% increase in the minimum wage is significantly correlated with a 1% 4% drop in teen employment rates for both genders. The results are robust to the use of province-specific linear trends and quadratic trends. IV regressions offer estimates that are comparable, and perhaps more importantly, are robust to different sets of instruments. We think this is an important finding, given the limited number of studies that have constructed plausible instruments in order to resolve endogeneity bias. While we find no evidence of minimum wage effects on prime-aged men, results based on SLID data suggest that amendments to the minimum wage may have a large impact on employment rates among prime-aged immigrants. In tandem, these results demonstrate that increases to the minimum wage set by Canadian provinces have had significant employment effects with respect to certain groups. From a more general perspective, these results offer an alternative viewpoint to recent U.S. studies, which assert that increases to the minimum wage have no employment effects. II. RECENT LITERATURE AND CONTROVERSY The U.S.-based research offers an ambiguous picture on minimum wage effects. Neumark and Wascher (2007) and Card and Krueger (1995) offer excellent reviews of the literature from the perspective of two prominent sides of the debate. 2 The negative and significant association between minimum wage and employment found in many panel studies (Burkhauser, Couch, and Wittenburg 2000; Neumark and Wascher 1992, 1994, 2007; Sabia 2009a, 2009b; Thompson 2009), 3 contrasts with the positive association frequently reported in papers using the case study approach (Card 1992; Card and Krueger 1994; Card, Kramarz, and Lemieux 1999; Katz and Krueger 1992). A shortcoming shared by the case studies is a reliance on time-series variation in minimum wage amendments within a single or limited number of jurisdictions. This raises the possibility that corresponding estimates of the minimum wage may be biased by unobserved heterogeneity. Several recent studies call into question the negative relationship between the minimum wage and employment found in many panel data-based analyses, reporting that any statistically significant relationship disappears with the inclusion of jurisdiction-specific linear trends or regionyear interactions (see Addison, Blackburn, and Cotti 2012; Allegretto, Dube, and Reich 2011; Dube, Lester, and Reich 2010). 4 These studies suggest that the negative relationship between the minimum wage and employment trends for teens 2. Schmitt (2013) contains an excellent review of recent U.S. studies. 3. Sabia, Burkenhauser, and Hansen (2012) also find minimum wage increases to have statistically significant and negative effects on teen employment, but rely primarily on time-series data from New York. 4. Dube, Lester, and Reich (2010) employ data across U.S. counties from 1990 to 2006 and find that coefficient estimates of the minimum wage (with respect to restaurant employment) are in some cases significant; however, estimates become insignificant when they control for regionspecific linear trends. Minimum wage estimates are also insignificant when the authors construct cross-border or contiguous county pairs and includes county pair and period interactions in their empirical specifications (these are intended to control for economic shocks common to both counties). Addison, Blackburn, and Cotti (2012) employ comparable county level data for the restaurant-and-bar sector and also find that the negative employment effects from increases to the minimum wage disappear with controls for unobserved trends. Likewise, using teen employment rates constructed from the Current Population Surveys (CPS) for , Allegretto, Dube, and Reich (2011) obtain a statistically significant relationship between minimum wage hikes and lower employment which disappears with the inclusion of statespecific linear trends or region-year interactions.

3 118 CONTEMPORARY ECONOMIC POLICY and/or low skilled workers may be attributed to unobserved spatial heterogeneity, which arises (for example) when researchers do not properly control for the possibility that minimum wage amendments are more likely to be implemented in areas with more severe economic shocks. However, Neumark, Salas, and Wascher (2013) show that estimating linear trends when the underlying trend is not linear can lead to extremely biased results. 5 Moreover, using region-year interaction terms removes much of the legitimate identifying variation in minimum wages and is tantamount to throwing the baby out with the contaminated bathwater (Neumark, Salas, and Wascher 2013). Sabia, Burkenhauser, and Hansen (2012) note similar reservations on the results of including statespecific linear trends. Neumark, Salas, and Wascher (2013, 2014) also criticize Allegretto, Dube, and Reich (2011) for using data including recessions at the start ( ) and end of their sample period, which could result in biased minimum wage effects. Furthermore, the studies by Allegretto, Dube, Lester, and Reich do not acknowledge the possibility that insignificant minimum wage elasticities might also be a product of high frequency data. This is, in fact, one of the seminal results from Baker, Benjamin, and Stanger (1991). These authors estimate the effects of minimum wages across Canadian provinces and over time ( ) and find a statistically significant elasticity of 0.25 with respect to teen employment. However, perhaps more importantly, they use filters in order to decompose this elasticity into short- and long-run components. Their findings suggest that the statistical significance of the estimate is driven by long-run changes over time, while short-run responses are small and insignificant. The implication is that estimates based on higher frequency data are more likely to produce smaller estimates, while lower frequency data result in significant estimates Another paper offers an explanation for insignificant minimum wage estimates. Results from Thompson (2009) indicate that bias may be induced by the non-binding nature of U.S. minimum wage laws in many states. Thompson (2009) finds that minimum wages increases from 1996 to 2000 did not share any statistically significant relationship with trends in teenage employment. However, employing data for counties with binding minimum wages, he obtains minimum wage elasticities of 0.3 to 0.4 for all counties and 0.4 to 0.6 for small counties. 6. In fact, Neumark, Salas, and Wascher (2013) obtain significant minimum wage elasticities when they use a low frequency filter to aggregate the county level type data used by In general, Canadian studies based on panel data tend to find significant negative minimum wage effects. However, the majority of these papers, including Baker, Benjamin, and Stanger (1999), Campolieti, Fang, and Gunderson (2005), and Campolieti, Gunderson, and Riddell (2006) do not employ province linear trends. Sen, Rybczynski, and Van de Waal s (2011) study is the most similar to this study and does employ province linear trends, but uses data over a slightly shorter time period ( ). Furthermore, Sen, Rybczynski, and Van de Waal (2011) do not consider the effects of the minimum wage by gender, immigrant, or other population subgroups. Brochu and Green (2013) employ provincial data from but primarily focus on labor market transitions. In the regression where they estimate the effects of the minimum wage on teen employment they do not use linear trends. 7 This study adds to existing literature through the following contributions. First, the use of Canadian data allows us to identify the effects of the minimum wage based on rich cross-province and time-series variation with a considerably large number of legislated minimum wage amendments. Second, we estimate minimum wage effects, across age groups, across gender, and with respect to immigrants. The latter is a population subgroup for which there is very Dube, Lester, and Reich (2010). They also demonstrate that the core findings of Dube, Lester, and Reich (2010) are sensitive to different and plausible sensitivity checks. For example, the core research methodology of Dube, Lester, and Reich (2010) is in the use of contiguous cross-border county pairs that might better control for unobserved economic shocks that are common to counties. However, Neumark, Salas, and Wascher (2013) suggest that the exclusion of other regions or counties as potential controls is generally not supported by the data. Furthermore, when Neumark, Salas, and Wascher (2013) use regions that are supported by the data, they do not find any evidence of disemployment effects. Neumark, Salas, and Wascher (2014) contains further arguments. 7. Studies based on United Kingdom, French, Spanish, and Portuguese data (see Abowd et al. 1997; Bazen and Skourias 1997; Dolado et al. 1996; Dolton, Bondibene, and Wadsworth 2010, 2012; Gorry 2013; Machin and Manning 1997; Machin and Wilson 2004; Neumark and Wascher 2004; Pereira 2003; Stewart 2002, 2004a, 2004b) report a variety of different minimum wage effects with respect to employment (ranging from no effect to significantly negative effects, or negative effects for teens and positive effects for youth). Most of these studies are based on time-series variation in a national minimum wage, making it difficult to control for unobserved regional or subnational policies that might also be responsible for local employment trends. The exception is Neumark and Wascher (2004). Stewart (2002), and Dolton, Bondibene, and Wadsworth (2010, 2012) do look at regional trends in employment, but the identification of the minimum wage is predominantly at the national level.

4 RYBCZYNSKI & SEN: EMPLOYMENT EFFECTS OF THE MINIMUM WAGE 119 limited empirical evidence of minimum wage effects. Canada is an especially relevant jurisdiction for this study, given the large number of immigrants it contains. Third, traditional panel data methods have been challenged by recent studies which argue the importance of controls for spatial heterogeneity. These recent papers usually do not find adverse employment consequences after controlling for area-specific time trends. The significant time-series variation in Canadian minimum wages across provinces allows us to evaluate the robustness of minimum wage effects, from a comparable jurisdiction, through the use of province and year-fixed effects and province linear trends. Finally, studying the Canadian experience with minimum wages should be informative for policymakers, in both Canada and in the United States, given similarities in government policy and cultural norms. III. THE MINIMUM WAGE IN CANADA Akyeampong (1989) notes that the original objective of minimum wage legislation was to prevent the exploitation of workers by firms, which then evolved to ensure workers received a livable wage and, subsequently, to protect women and young workers from discrimination. 8 Canada s initial attempt at a minimum wage was through the enactment of the federal Fair Wages Policy of 1900, which protected workers engaged on all public works and government contracts. However, more comprehensive coverage began later in the century and by 1920, six Canadian provinces had passed laws to protect working women and children from exploitation. By the mid-1950s, most Canadian provinces had enacted minimum wage legislation for male employees. As discussed above, Canada is an interesting laboratory to study the minimum wage, as it falls under provincial jurisdiction, resulting in attractive cross-province and time-series variation. Until 1996, there was a separate federal minimum wage for workers in federal jurisdiction industries, such as transportation, financial services, telecommunications, and broadcasting. The federal minimum wage was eliminated in 1996, with the provincial minimum wage defining the relevant wage floor for federal employees working in a specific province. 9 Thus, a relevant 8. For an excellent discussion of the role of minimum wage as a policy tool for social justice, refer to Green (2014). 9. Please refer to Kerr (2008) for further details. concern is whether minimum wage estimates might be confounded if provincial employment data prior to 1996 contain a significant number of federal employees who were subject to the federal minimum wage, and not the provincial minimum wage of residence. However, Kerr (2008) reports that only a very small number of federal employees were actually paid the federal minimum wage. Studies based on data over the pre- and post-1996 sample period (Akyeampong 1989; Galarneau and Fecteau 2014; Sussman 2005) suggest that a vast majority of minimum wage earners tend to be teens and young adults in the retail trade and services (food and accommodation) industries. Specifically, these Canadian estimates suggest that teens account for about half of all minimum wage workers, and that one in three teens work for the minimum wage (Galarneau and Fecteau 2014; Sussman 2005). Moreover, prime-aged women, immigrants, and low-educated workers are all disproportionately represented among minimum wage employment (Galarneau and Fecteau 2014; Minimum Wage Advisory Panel 2014; Sussman 2005). Finally, Galarneau and Fecteau (2014) consider trends in average hourly earnings in industries with significant numbers of minimum wage employees, between 1975 and 2013, and find that these are fairly stable, ranging from $18 to $24 (in constant 2013 dollars). However, with obvious caveats regarding dips and peaks within specific time periods, the real minimum wage has steadily increased over time, and specifically in the past decade. Employing data from the Survey of Employment, Payrolls, and Hours (SEPH), Galarneau and Fecteau (2014) suggest that the ratio of the real minimum wage to real average hourly earnings increased from roughly 37% 41% in 2005 to 45% 46% in Therefore, variation in the minimum wage relative to average hourly earnings ratio has been primarily driven by the minimum wage. In summary, these trends suggest that the minimum wage constitutes a significant portion of average hourly earnings for some Canadians, and importantly, that changes in the minimum wage to average earnings ratio are primarily driven by minimum wage amendments. IV. DATA We use provincial time series data to construct a standard panel over the time period. The majority of our data are extracted

5 120 CONTEMPORARY ECONOMIC POLICY from CANSIM, the socioeconomic database made available to the Canadian university community by Statistics Canada. The CANSIM time series data provides aggregates of genderspecific employment and population data, based on the Labour Force Survey (LFS). The LFS is a representative survey that covers the civilian, noninstitutionalized population 15 years of age and over (Statistics Canada 2014). This survey is conducted monthly, nationwide, in the provinces and the territories. Excluded from the survey s coverage are: persons living on reserves and other Aboriginal settlements in the provinces, full-time members of the Canadian Armed Forces, and the institutionalized population. These groups together represent an exclusion of less than 2% of the Canadian population aged 15 and over (Statistics Canada 2014). Our analysis involves aggregate statistics on the employment of older teens (aged 18 19), and school enrolment rates, but these are not available on CANSIM. 10 Moreover, data on immigrant population and employment status are only recorded in the LFS from 2006 onward. We manually construct aggregate enrolment rates and employment-to-population ratios for these groups using 19 waves, between 1990 and 2011, from the SLID and the SCF. 11 Both the SLID and SCF suppress immigration status for a sizable fraction of the observations, particularly in provinces with few immigrants. Therefore, the provinces of Saskatchewan, Prince Edward Island, Newfoundland, New Brunswick, and Nova Scotia are omitted from our analysis of immigrant employment rates. Furthermore, given the sometimes small sample sizes, we focus on employment ratios for teen and prime-aged adult immigrants, and do not disaggregate these groups by gender. 12 The reason we use two datasets is that the SLID replaced the SCF in The SLID and SCF are both nationally representative surveys. These survey samples are drawn from the (April) 10. Average wage data are also unavailable on CANSIM prior to We use linear interpolation, by province, to fill in these values. 11. Enrolment rates for teens aged are manually constructed using the public use microdata files of the monthly LFS from 1990 to 2008 (and 1997 to 2008). The rationale for using school enrolment rates is discussed later. 12. Another drawback of the SCF/SLID data is that we are unable to calculate immigrant employment rates based on year of arrival, because this information is suppressed for some waves. Furthermore, the recentness of immigration is categorized differently in the SLID and the SCF. LFS 13 and, therefore, have the same sampling design and target population. The SCF was used to produce annual statistics on income in Canada until 1997, whereas the SLID is the primary source for Canadian income data from 1997 to In order to evaluate the correspondence of these surveys to the LFS, we compared overall employment rates from the SLID/SCF to population employment rates from the LFS. The employment rates across surveys are very similar (averages differed by about one percentage point). As a final check, we compared employment rates for prime-aged immigrants constructed from the 2001, 2006, and 2011 waves of the SLID against the 2001 and 2006 Census of Canada, and the 2011 Labour Force Survey estimates. 15 We obtained comparable employment rates and, thus, we are confident that our data are representative of the population of immigrants within Canada. Information on provincial minimum wages is obtained from Human Resource and Social Development Canada s (HRSDC) minimum wage database. 16 This database lists the exact date of changes in the adult general minimum wage for each of the Canadian provinces and of federal employees. The last posted federal employee minimum wage occurred in As of 1988, all provincial minima exceeded the federal minimum wage, and the federal minimum was eliminated in We note that the 13. The SCF began as a supplement to the April LFS survey, in which four out of six households were mailed an additional questionnaire on income. The SLID was conducted by computer-assisted telephone interview, and again, was drawn from the April LFS. The LFS is conducted via computer-assisted telephone interviews, normally on the 15th day of each month, and uses a probability sample with a stratified multistage design. The LFS has six rotation groups (each nationally representative itself), one of which is rotated out (replaced) each month. See imdb/p2sv.pl?function=getsurvey&sdds=3701, www23.statcan.gc.ca/imdb/p2sv.pl?function=getsurvey& SDDS=3502, and for further details on these surveys. 14. The SLID was phased out in 2011 and income data are now available in the Canadian Income Survey and National Household Survey. However, microdata from the Canadian Income Survey are not publicly available, and the National Household Survey is only conducted every 5 years. 15. Because of data quality concerns with the National Household Survey, which replaced the mandatory long-form census questionnaire in 2011, we employ the 2011 LFS Estimates (Cansim table ) to check the comparability of the 2011 SLID public use micro data. 16. Available at: sm-mw/menu.aspx?lang=eng. Accessed May 20, 2016.

6 RYBCZYNSKI & SEN: EMPLOYMENT EFFECTS OF THE MINIMUM WAGE 121 TABLE 1 Sample Characteristics Variables Mean Standard Deviation Min Max Male/Female sample ( ) Ln (real minimum wage) Ln (average hourly wage) Prime-aged male unemployment rate Real provincial GDP in billions No. observations 310 Immigrant sample ( ) Ln (real minimum wage) Ln (average hourly wage) Prime-aged male unemployment rate Real provincial GDP in billions No. observations 110 Employment-to-population ratios for adults (25 54) Women Men Immigrant Employment-to-population ratios for youth (15 19/16 24) Women age Men age Immigrant age Adult (25 54) to total population ratios by subgroup Women Men Immigrant Youth (15 19/16 24) to total population ratios by subgroup Women age Men age Immigrant age Notes: The above sample is obtained from pooling data across 10 Canadian provinces between 1981 and 2011 in the case of men/women and between 1990 and 2011 for 5 Canadian provinces (excluding Saskatchewan and all of the eastern maritime provinces) in the case of immigrants. adult minimum wage does not cover all jobs. For example, some jurisdictions have separate occupation-specific minima. 17 However, youthspecific minimum wages have been phased out of most provinces following the 1982 adoption of the Charter of Rights and Freedoms. The hourly nominal minimum wage ranged from $3 to almost $4 in 1981, with Quebec and Saskatchewan possessing the highest minimum wages and Nova Scotia the lowest. All provinces implemented several amendments over the sample period (a total of 185 amendments), and by 2011 the nominal minimum wage ranged from $9.40 to $ Manitoba and Nova Scotia had the highest and Alberta the lowest. In terms of specific legislative amendments: 23 amendments to the minimum wage were enacted by Quebec, 22 by Nova Scotia, 23 each by New Brunswick and Prince Edward Island, 18 by Ontario, 20 by Manitoba, 19 by 17. Some of these job-specific minima exceed the adult minimum wage and/or are defined at daily or monthly rates. Historical values of the exceptions to the general adult minimum wage are not available in the minimum wage database. Newfoundland, 16 by Saskatchewan, 10 each by British Columbia and 20 Alberta, yielding a total of 185 amendments. In terms of magnitude, increases in minimum wage ranged from 10 cents to over one dollar. Specifically, of the 185 amendments, 41 increased minimum wages by less than 25 cents, 105 increased between 25 and 49 cents, 30 increased between 50 and 74 cents, 6 increased between 75 and 99 cents, and 3 increased by one dollar. In terms of percentage change, 92 amendments represented a minimum wage hike of less than 5%, 76 amendments represented an increase of 5% 9%, 16 represented a hike of 10% 19%, and 1 represented a change of more than 20%. These trends suggest that the use of Canadian data offers some interesting cross-province and time-series variation in order to identify the effects of the minimum wage. In sum, our primary data are from 1981 to 2011 for 10 Canadian provinces for men and women, yielding 310 observations. Using the SCF and SLID to construct further group-specific employment ratios between 1990 and 2011, we

7 122 CONTEMPORARY ECONOMIC POLICY obtained 220 observations for older teens. Data on immigrants are limited to five Canadian provinces (excluding Saskatchewan, Prince Edward Island, Newfoundland, Nova Scotia, and New Brunswick), yielding 110 observations. Summary statistics for our data are available in Table 1. V. EMPIRICAL MODEL Consistent with previous studies we employ the following semi-logarithmic base specification in order to estimate the impact of minimum wages on employment-to-population ratios 18 : (1) EMP POP itg =β ln MW it +γx itg +μ i +τ t +ε itg where EMP/POP itg represents the employmentpopulation ratio for group g in province i at time t. ln MW it is the natural logarithm of the real minimum wage, X itg is a standard vector of time-varying covariates (some of which are group specific) that impact trends in employment-population ratios independently of the real provincial minimum hourly wage, μ i are province-specific fixed effects, τ t are yearspecific fixed effects, and ε itg is the error term. Employment-to-population ratios are the standard dependent variables in the literature. An employment-to-population ratio denotes the proportion of a specific group employed as a fraction of the total population of that group in a given province in a given year. For example, with respect to prime-aged women, the employmentto-population ratio is the proportion of employed prime-aged women as a fraction of the total population of prime-aged women, in province i and year t. We use employment-to-population ratios for women, men, and immigrants as dependent variables to assess the impact of provincial minimum wage policies on each of these groups. This variable is in levels. The key covariate is the natural logarithm of the nominal legislated minimum wage deflated by 18. We obtain similar results whether we use log-log or semi-log specifications. Different studies have relied on either of these models. For brevity, and consistency with Sen, Rybczynski, and Van de Waal (2011) we restrict our focus to estimates from the semi-log model denoted by Equation (1). However, we report estimates from log-log models in Table A1. We note that the coefficient estimates are statistically significant on all covariates with the exception of log average hourly wage, and GDP in the log-log specification. Consistent with previous studies, the coefficients on the prime-aged male unemployment rate and on the population share of teens are negative, GDP is mildly negative and log average hourly wage is positive. the provincial consumer price index (ln MW it ). Because minimum wage changes do not always occur at the start of a year, and in 25 cases there are two hikes within a year, we use linear interpolation to generate one minimum wage value per year. While real minimum wage is commonly employed in the literature (see, e.g., AAddison, Blackburn, and Cotti 2012; Burkhauser, Couch, and Wittenburg 2000; Card and Krueger 1995; Neumark, Salas, and Wascher 2013; Sen, Rybczynski, and Van de Waal 2011), results are substantively similar, if we use the ratio of minimum wage to average hourly wage. Our vector X it includes a set of covariates that is common in the literature (e.g., Addison, Blackburn, and Cotti 2012; Burkhauser, Couch, and Wittenburg 2000; Card and Krueger 1995; Green and Harrison 2010; Lemos 2005a; Neumark, Salas, and Wascher 2013). These covariates include average hourly wage for workers paid on an hourly basis, in real dollars, to capture the effects of changes in the minimum wage controlling for growth patterns observed in average wages. Unemployment rates of prime-aged males and the real provincial gross domestic product (GDP) are used to control for provincial labor market conditions and economic cycles, respectively. For each group (e.g., women, men, and immigrant) the population of the group as a proportion of their total working-age population is used to account for effects on employment that are due to an increasing or decreasing proportion of a particular group eligible for labor force participation. 19 In order to maintain consistency with how we empirically measure the minimum wage, average hourly wage is also in natural logarithms, whereas the rest of the covariates are in levels. Province and year-fixed effects are used to control for any unobserved policy shocks that may be correlated with movements in the minimum wage and are either specific to a province over time or for a year across provinces. In such specifications, the effects of the minimum wage are identified by exploiting within province timeseries variation, holding constant unobserved provincial differences that remain fixed over time- and year-specific shocks common to all jurisdictions. We assess the sensitivity of our findings to the use of province-specific linear and quadratic trends. 19. We also constructed a covariate for federal and provincial transfers (in real dollars) to the poorest quintile of population employed to account for shifts in welfare transfers which may occur more frequently in minority or eligible groups. Results are substantively similar if this covariate is employed in the specification.

8 RYBCZYNSKI & SEN: EMPLOYMENT EFFECTS OF THE MINIMUM WAGE 123 TABLE 2 Estimates of the Effects of the Minimum Wage on Employment of Men and Women Aged (1) (2) (3) (4) Women Men Women Men Women Men Women Men Ln (real minimum wage) *** *** ** *** ** *** (0.084) (0.088) (0.043) (0.039) (0.045) (0.034) (0.056) (0.041) {elasticity} { 0.172} { 0.094} { 0.260} { 0.256} { 0.211} { 0.207} { 0.306} { 0.322} Newey-West standard errors (0.053) (0.055) (0.032) (0.031) (0.026) (0.024) (0.030) (0.027) Ln (average houly wage), real Yes Yes Yes Yes Yes Yes provincial GDP, subgroup pop. over total population, prime-aged male unemployment rate Province linear trends Yes Yes Yes Yes Quadratic in prov. linear trends Yes Yes Province and year FE Yes Yes Yes Yes Yes Yes Yes Yes R Obs Notes: Dependent variables are employment-to-population ratios for each subgroup: women and men aged The above estimates are obtained from pooling data across 10 Canadian provinces between 1981 and Coefficient estimates are presented with two-way (province year) cluster-robust standard errors in parentheses underneath. Elasticity is calculated based on the mean employment rates, by each group, across its entire sample. Newey-West standard errors, correcting for unknown heteroskedasticity and first-order autocorrelation, are presented as well. Column (1) presents results from the parsimonious specification, column (2) from the baseline specification, column (3) includes provincial linear trends, and column (4) includes quadratic trends. VI. RESULTS A. Baseline Estimates for Teens and Prime-Aged Adults Table 2 contains ordinary least squares (OLS) coefficient estimates of the minimum wage with respect to the employment-to-population ratio, by gender, for four specifications. 20 Specification (1) is parsimonious, controlling only for the log real minimum wage, province and year-fixed effects. Specification (2), our baseline, evaluates the minimum wage effects controlling for a standard set of time-varying province level covariates. Province-specific linear trends and a quadratic in the provincial linear trends are added specifications (3) and (4). We present cluster-robust standard errors, clustering on province and year, using the two-way clustering method proposed by Cameron, Gelbach, and Miller (2006). Newey-West standard errors, correcting for unknown heteroskedasticity and 20. The full baseline results (excluding year- and province-fixed effects) are reported in Table A1. This table also contains results for an estimate of a log-log specification (logged dependent variable and all nonbinary baseline covariates logged). In both cases, the prime-aged male unemployment rate and the population share (subgroup population over total population) are strong and significant predictors of teen employment rates. first-order autocorrelation are presented in the fourth row. Coefficient estimates of the minimum wage from Table 2 are negative and statistically significant (at either the 1% or 5% levels) for both sexes, across all but the first specification. Amendments to the minimum wage appear to have a comparable impact on the employmentto-population ratios of male and female teens and coefficient estimates suggest minimum wage elasticities ranging from to in line with previous U.S. studies, and in the lower range of results from previous Canadian studies. On average, these results suggest an absence of significant gender differences in the minimum wage effect among teens. Another important finding is that coefficient estimates of the minimum wage remain negative and statistically significant even after employing province and yearfixed effects, province-specific linear trends and quadratic trends. The coefficient estimates are mildly diminished with linear trends, but higher with a quadratic in province-specific trends. Table 3 is organized similar to Table 2, but contains estimates of the effects of the minimum wage with respect to employment rates of adult men and women aged (prime-aged). Specifications (1) through (3) are as in Table 2. Specification (4) excludes the prime-aged male

9 124 CONTEMPORARY ECONOMIC POLICY TABLE 3 Estimates of the Effects of the Minimum Wage on Employment of Men and Women Aged (1) (2) (3) (4) Women Men Women Men Women Men Women Men Ln (real minimum wage) (0.055) (0.035) (0.045) (0.022) (0.026) (0.019) (0.044) (0.042) {elasticity} {0.032} {0.011} {0.039} { 0.002} { 0.025} { 0.002} {0.046} {0.017} Newey-West standard (0.032) (0.021) (0.026) (0.012) (0.015) (0.010) (0.027) (0.023) errors Ln (average hourly wage), No No Yes Yes Yes Yes Yes Yes real provincial GDP, subgroup pop. over total population Prime-aged male Yes Yes Yes Yes Yes Yes unemployment rate Province linear trends Yes Yes Province and year FE Yes Yes Yes Yes Yes Yes Yes Yes R Obs Notes: Dependent variables are employment-to-population ratios for each subgroup: women and men aged The above estimates are obtained from pooling data across 10 Canadian provinces between 1981 and Coefficient estimates are presented with two-way (province year) cluster-robust standard errors in parentheses underneath. Elasticity is calculated based on the mean employment rates, by each group, across its entire sample. Newey-West standard errors, correcting for unknown heteroskedasticity and first-order autocorrelation, are presented as well. Column (1) presents results from the parsimonious specification, column (2) from the baseline specification, column (3) includes provincial linear trends, and column (4) presents results from the baseline specification but omits prime-aged male unemployment rate as a covariate. unemployment rate covariate. 21 While coefficient estimates of the minimum wage are negative in some columns, they are predominantly small and statistically insignificant. Although women are more likely to work minimum wage jobs than men, results from Table 3 provide little evidence of disproportionate employment effects on women. 22 B. Estimates for Older Teens and for Immigrants Thus far, we have focused on teens and women, population subgroups that are disproportionately represented among minimum wage workers. In this section, and the next, we further explore subgroup analysis by considering minimum wage effects for other vulnerable 21. Prime-aged male unemployment rate is incorporated in our baseline to control for underlying labor market conditions. Although the coefficient is both economically and statistically significant, one might be concerned that male unemployment rates capture a large part of the variation in male employment rates that is driven by changes to real minimum wage. However, we find that results are nearly identical (for women as well as men, and teens as well as adults) whether we include prime-aged male unemployment rates or not. 22. We also considered minimum wage effects on young men and women aged (results omitted for brevity). For both men and women, the coefficient estimates were negative, smaller in magnitude than the teen regressions, and tended to be statistically insignificant for men and women. groups: older teens (Table 4) and immigrants (Table 5). Table 4 contains estimates of the effects of the minimum wage with respect to employment rates of older teens, aged These teens are more likely to have completed high school, entered the labor market, and are subject to the general adult minimum wage across provinces. 23 Table 4 also includes results for teens aged in a comparable time frame ( ). The first set of estimates contains the baseline specification (with province- and year-fixed effects), while the second set of estimates incorporate a control for the enrolment rate. The relationship between school enrolment and employment outcomes represents an interesting relationship. On the one hand, if employment prospects are poor, then students may opt to return to school (riding out poor labor conditions and also improving credentials). Conversely, enrolled students who wish to work part time are very likely to take jobs which pay at or near the minimum wage, representing a sort of captive audience. Results presented in Table 4 are quite comparable to those presented in Table 2. Coefficient estimates on teens aged are larger in 23. As with the rest of the adult population, these teens could still be subject to different occupation-specific minima for some jobs (e.g., serving alcohol/food).

10 RYBCZYNSKI & SEN: EMPLOYMENT EFFECTS OF THE MINIMUM WAGE 125 TABLE 4 Estimates of Minimum Wage and Enrolment Rates on Employment of Men and Women Aged and Baseline Baseline + Enrolment Teens Teens Teens Teens Women Men Women Men Women Men Women Men Ln real min wage *** *** *** *** # (0.064) (0.061) (0.176) (0.129) (0.067) (0.060) (0.169) (0.129) Enrolment rate *** (0.179) (0.148) (0.083) (0.089) Ln (average hourly wage), Yes Yes Yes Yes Yes Yes Yes Yes real provincial GDP, subgroup pop. over total population, prime-aged male unemployment rate, province FE, year FE R Obs Notes: Dependent variables are employment-to-population ratios for each subgroup: women and men aged The above estimates are obtained from pooling data across 10 Canadian provinces between 1990 and Coefficient estimates are presented with two-way (province year) cluster-robust standard errors in parentheses underneath. # Significant at 11%; *significant at 10%; **significant at 5%; ***significant at 1%. the post-1990 sample (with elasticities between and 0.4), and remain statistically significant at the 1% level. Results for older teens exhibit a similar pattern and are larger for women than for men; however, these estimates are less precise. For both age groups, minimum wage coefficient estimates are stable to the inclusion of school enrolment rates. The coefficient on enrolment rates is negative, but the estimates are imprecise for all groups except older teen women. Finally, in Table 5, we present results for the overlooked but increasingly relevant vulnerable population group: immigrants. Estimates of the minimum wage coefficients are presented for immigrants aged and aged across four model specifications. Estimates in all columns are conditioned on the use of province and time varying covariates employed in the previous regressions (including province and yearfixed effects). The parsimonious (1), baseline (2), provincial linear trends (3), and quadratic in provincial linear trends (4) specifications are identical to those of Table Coefficient estimates of the effects of the minimum wage on employment rates among 24. Mobility across regions might also influence minimum wage effects, particularly for immigrants. Ideally, we would control for migrants, as a share of the population stock, across province and across time for our sample period. However, because we do not have such rich data, we simply note that we are unable to control for this potentially confounding migration effect. young immigrants (aged 16 24) are statistically insignificant across all specifications. However, for all but the final specification, results indicate that increases to the minimum wage are negatively and significantly correlated with a decline in the employment ratios of prime-aged immigrants (aged 25 54). Both the magnitude and statistical significance of this estimate decrease when we use the specification with a quadratic in provincial linear trends. This result is not surprising given the small sample size and the issues related to spacial controls. Due to the small sample size, caution should be exerted in the emphasis of these results. However, to the best of our knowledge, this is novel empirical evidence, as we are unaware of any previous empirical evidence, which demonstrates that a higher minimum wage may have specific employment effects on prime-aged immigrants. 25 In contrast, Table 3 suggests that amendments to the minimum wage have no statistically significant impacts on prime-aged adult employment ratios for the Canadian population as a whole. C. Sensitivity Checks Given our large sample and generally robust findings in Table 2, we focus on teen employment rates in the sensitivity analysis. Table 6 contains 25. Orrenius and Zavodny (2008) report no statistically significant effect of minimum wage on employment, among adult immigrants in the United States.

11 126 CONTEMPORARY ECONOMIC POLICY TABLE 5 OLS Estimates of Minimum Wage on Employment of Foreign Born, Aged and (1) (2) (3) (4) Ln real min wage *** *** ** (0.197) (0.041) (0.217) (0.078) (0.270) (0.084) (0.325) (0.148) {elasticity} { 0.247} { 0.174} {0.307} { 0.235} {0.228} { 0.232} {0.195} { 0.085} Newey-west standard errors (0.129) (0.065) (0.172) (0.083) (0.154) (0.061) (0.193) (0.100) Ln(average houly wage), Yes Yes Yes Yes Yes Yes Real provincial GDP, Subgroup pop. over total population, Prime-aged male unemployment rate Province linear trends Yes Yes Yes Yes Quadratic in prov. linear Yes Yes trends Province & year FE Yes Yes Yes Yes Yes Yes Yes Yes R Obs Notes: Dependent variables are employment-to-population ratios for each subgroup: immigrants aged and The above estimates are obtained from pooling data across 1990 and 2011 for five Canadian provinces (excluding Saskatchewan and the maritime provinces). Coefficient estimates are presented with two-way (province year) cluster-robust standard errors in parentheses underneath. Elasticity is calculated based on the mean employment rates, by each group, across its entire sample. Newey-West standard errors, correcting for unknown heteroskedasticity and first-order autocorrelation, are presented as well. Column (1) presents results from the parsimonious specification, column (2) from the baseline specification, column (3) includes provincial linear trends, and column (4) includes quadratic trends. the results of a series of dynamic checks with respect to minimum wage employment effects for teen women and men. The first specification evaluates the sensitivity of estimates to the inclusion of 2-, 3-, and 4-year leaded values of the minimum wage. 26 Intuitively, minimum wage amendments that are sufficiently in the future should not share a statistically significant relationship with current employment rates. Any statistical significance would then most likely be attributable to unobserved spatial heterogeneity. For example, it is possible that jurisdictions which frequently enact amendments to the minimum wage are also more likely to implement other policies, which impact local labor market conditions. If estimates of the current minimum wage are not robust to the inclusion of leaded values, this would then suggest that the statistically significant correlation between teen employment and the minimum wage we have so far obtained may simply be an artifact of unobserved provincespecific heterogeneity. Empirical results in the first two columns reveal that coefficient estimates of the minimum wage with respect to teen male and female 26. All specifications in Table 6 also contain the same number of leads and lags for average hourly wage as the minimum wage covariate. employment rates remain statistically significant, even with the inclusion of leaded values, with coefficient estimates and 0.093, comparable to the results in Table 2. Alternative strategies to investigate the presence of provincespecific heterogeneity, using a distributed lag and lead model along the lines of Dube, Lester, and Reich (2010) and Sen, Rybczynski, and Van de Waal (2011), produce similar results. Specifically, the coefficient on minimum wage remains negative (large) and statistically significant. As suggested by Baker, Benjamin, and Stanger (1999) and Campolieti, Gunderson, and Riddell (2006), it may take more than a year to reach the full impact of minimum-wage amendments. Thus, in columns 3 4, we consider a specification with both lagged and leaded values of the minimum wage. While the contemporaneous minimum wage effect is now smaller and insignificant for both male and female teens, the 2- and 3-year lagged effects are substantial, negative, and statistically significant We conduct further sensitivity analysis on the immigrant sample looking at a leads and lags and leads model, and the results are similar to that of the non-immigrant population. Estimates are less precisely estimated, but the p value remains just above 0.11 on the contemporaneous minimum wage in the leads model. Statistically insignificant, but still large and negative, estimates in the lags and leads model. Results are available upon request.