Education, experience and urban wage premium *

1 Education, experience and urban premium * Fredrik Carlsen, Jørn Rattsø and Hildegunn E. Stokke ** Department of Economics, Norwegian University of Science and Technology fredrik.carlsen@svt.ntnu.no; jorn.rattso@svt.ntnu.no; hildegunnes@svt.ntnu.no Abstract Recent studies investigate the urban premium associated with education and work experience. We study the effect of work experience and firm tenure across education groups. All education groups benefit more from working in cities, and the extra city premium highly educated enjoy over less educated workers is increasing with city work experience. Interestingly, the city premium of less educated workers is increasing in firm tenure, while the highly educated gain more by shifting jobs between firms. The analysis is based on administrative registers covering all full time private sector workers in Norway, about 4.7 million worker-year observations. Keywords: Agglomeration economies, education, worker experience, firm tenure JEL codes: J24, J31, J61, R12, R23 Date: February 17, 2015 * We appreciate discussions at the 2012 European Meeting of the Urban Economics Association, the 2012 North American Meeting of the Regional Science Association International, the 2013 Meeting of the Western Regional Science Association, the 2013 Congress of the European Economic Association, the Urban and Regional Economics seminar at LSE, the Labour Economics/Empirical Microeconomics workshop at the Norwegian School of Economics, and at the department seminar at Umeå School of Business and Economics, and comments from Ragnhild Balsvik, Nate Baum-Snow, Paul Cheshire, Jorge De la Roca, Steve Gibbons, Christian Hilber, Ana I. Moreno-Monroy, Jarle Møen, Henry Overman, Kjell Gunnar Salvanes and Olmo Silva. We are grateful for the cooperation of Statistics Norway. An earlier version of this paper was titled Urban premium and the role of education: Identification of agglomeration effects for Norway. ** Corresponding author: Hildegunn E. Stokke, Department of Economics, Norwegian University of Science and Technology, 7491 Trondheim, Norway. E-mail: hildegunnes@svt.ntnu.no

2 1. Introduction Highly educated individuals have higher productivity and tend to live in cities. The observation has led to a literature dealing with skilled cities or smart cities (Glaeser and Saiz, 2004; Shapiro, 2006; Combes et al., 2008; Winters, 2011). The urban concentration of highly educated explains a large part of the observed urban rural gap. The more recent literature has addressed the role of place and type of experience and derived dynamic agglomeration effects (de la Roca and Puga, 2014; Matano and Naticchioni 2013). Our contribution is to investigate the effects of work experience and firm tenure in cities across education groups. The bottom line is that the advantage of having work experience in cities is increasing in the level of education. The highly educated take advantage of shifting between firms in cities, while the low educated benefit from long firm-specific tenure. A rapidly expanding literature has studied the education/skill gradient of the urban rural gap. Studies based on US data (NLSY, PSID or PUM) tend to find that the urban premium is increasing in education/skills (Glaeser and Mare, 2001; Wheeler, 2001; Gould, 2007; Rosenthal and Strange, 2008; Bacalod et al., 2009). Exceptions are Adamson et al. (2004) who find a nonlinear relation between the urban premium and education level, and Lee (2010) who finds that the urban premium is negative for high-skilled health workers and positive for less skilled health workers. A Swedish study based on panel data of all private sector workers finds that the urban premium is largest for workers in occupations that demand non-routine tasks (Andersson et al., 2014), whereas, in Italy, return to higher education seems to be negatively correlated with regional population size (Di Addario and Patacchini, 2008). Several studies have concluded that the urban rural gap is partly or completely due to differences in growth between urban and rural areas (Glaeser and Mare, 2001; Wheeler, 2006; Yankow, 2006; Gould, 2007; Baum-Snow and Pavan, 2012; Wang, 2013; De la Roca and Puga, 2014). Wage growth seems to have a substantial between-job component (Wheeler, 2006; Yankow, 2006), and workers at least partly keep gains when moving to other regions, indicating that returns to work experience in urban areas are portable

3 (D Costa and Overman, 2014; De la Roca and Puga, 2014). De la Roca (2011) expands the analysis by looking at both initial and return migration. Few studies have considered the relation between skills and return to work experience in urban areas. Using the National Longitudinal Survey of Youth, Gould (2007) finds that s of white collar workers increase faster in urban areas, whereas the growth of blue collar workers do not vary with urban scale. Using administrative panel data for Spain, De la Roca and Puga (2014) find that returns to work experience in big cities are highest for the most skilled workers, where skill is measured by worker fixed effects. Matano and Naticchioni (2013) also find that return to work experience in the largest cities is increasing along the distribution using panel data about Italian workers. Studies of returns to urban work experience have considered two reasons why s may increase faster in urban areas: faster accumulation of human capital due to higher rates of interactions and spillover of knowledge (Glaeser, 1999) and more efficient job matches due to thicker labor markets (Helsley and Strange, 1990). Yankow (2006) has suggested that faster learning causes higher within-job growth whereas improved matching results in higher between-job growth. Learning may take place on many arenas, and job change represents an opportunity to meet new people and learn new skills. Moreover, as pointed out by Wang (2013), better matches may be due to more efficient within-firm task assignments. Frequent job changes are thus not necessarily an indication of efficient matching; nor are long job tenures necessarily an indication of fast accumulation of human capital. Urban labor markets promote productivity by allowing workers to become more specialized (Duranton and Jayet, 2011). Two pieces of evidence suggests that specialization is more important for highly educated workers. First, the share of vacancies filled by promotions or applicants with experience from the same occupation is increasing in job rank and the average education level of the occupation (Kwon and Meyersson Milgrom, 2014). Second, Schwerdt et al. (2010) compare the long run costs of displacement after plant closure for blue collar workers and white collar workers, and find that only the latter have persistent employment and earnings losses; blue collar workers suffer only transient losses. These

4 results are consistent with the idea that internal labor markets and firm-specific human capital is more important for highly educated workers, suggesting that large urban markets should increase productivity, and therefore s, more for highly educated than for less educated workers. Highly educated workers may also benefit more from knowledge spillovers in urban regions, for instance by interacting with university and research institutes. On the other hand, low educated persons may have a higher capacity for learning because they have a lower stock of capital (Wheeler, 2004; Di Addario and Patacchini, 2008; Rosenthal and Strange, 2008; Matano and Naticchioni, 2012). The size of the labor market affects returns to tenure via the transferability of human capital (Stevens, 1994; Di Addario and Patacchini, 2008, Lazear, 2009). In large urban labor markets, on-the-job training will improve the outside options and improve the bargaining position of workers versus firms. Workers have also incentives to investment more in general human capital if outside options are better (Wasmer, 2006). These effects create a positive correlation between urban scale and return to tenure. Since the geographic mobility in Norway is increasing in education level (Machin et al., 2012), the size of the local labor market is more important for outside opportunities of workers with low education. Therefore, the association between urban scale and returns to tenure is likely to be more positive (or less negative) for low educated workers. To our knowledge, only Matano and Naticchioni (2013) have studied how firm tenure affects the urban premium for different levels of skills. They consider migrants to the largest Italian cities and find that s of high skill workers increase faster after migration during job changes whereas s of low skill workers increase faster when workers remain in the same firm. We use administrative register data for the whole working population of Norway during 2003 to 2010. We exclude part time workers and workers in the public and primary sectors, producing a dataset with about 4.7 million worker-year observations in 54 industrial sectors, 350 occupations, 89 labor market regions and about 150 000 firms. To study the effect of urban scale on s, we distinguish between the 7 largest labor market regions with more than 150.000 inhabitants, denoted cities and the other 82 regions. We study how the city premium and returns to work experience in cities depend on education level and firm

5 tenure when controlling for industry, occupation and unobservable time-invariant worker characteristics. We find that highly educated workers benefit most from working in cities, and the extra city premium they enjoy over low educated workers is increasing with city work experience. The result is consistent with variation across skill levels shown by De la Roca and Puga (2014) and Matano and Naticchioni (2013). Furthermore, we find that the city premium of low educated workers is increasing in firm tenure, whereas the premium of highly educated workers is increasing with job shifts. This finding is in accordance with the effects of firm tenure for high and low skill workers shown by Matano and Naticchioni (2013). It is of interest to try to separate education and skills as sources of premia. Methodologically the main future challenge is to look in more detail at the individual choices of firm and sector affiliation. The rest of the paper is organized as follows. Section 2 discusses our econometric strategy and data. The estimates of static agglomeration effects across education groups are presented in section 3. Section 4 moves on to dynamic agglomeration effects based on city and firm experience effects. The robustness of the results is investigated in section 5. Section 6 summarizes our conclusions and indicates future research. 2. Econometric strategy and data Norwegian cities are small by international comparison, and most regions have large unpopulated areas. Based on information about commuting flows between municipalities, Statistics Norway has divided Norway into 89 travel-to-work areas, denoted economic regions. The economic regions conform to NUTS-4 regions, as defined by the European Union standard of regional levels. This level of aggregation captures functional regions understood as common labor markets. We define cities as labor market regions with more than 150 000 inhabitants in 2010 and thereby include the 7 largest city regions. Our dataset is computed from three administrative registers: the employment, tax and education registers. The employment register links workers and firms, and gives information

6 on work contracts for all employees during 1993-2010. It includes the number of days worked, the type of contract 1, and from 2001 onwards it also has information on the exact number of hours worked per week. We calculate the number of hours worked per year, which is combined with data on annual income from the tax register to give a measure of hourly s for all employees. 2 We concentrate on workers with full time contracts (at least 30 hours per week). Workers with more than two contracts during a year, as well as workers with one full time and one part time contract are excluded. Workers with two full time contracts are excluded if the number of days worked that year exceeds 455 days. This means that we allow for a maximum of 3 months overlap between the two contracts. To avoid extreme observations, we exclude individuals working less than 50 hours or more than 3500 hours per year. Similar, workers with hourly below 70 NOK or above 1250 NOK are also excluded. Finally, we focus on workers between 25 and 65 years old. Information about the work contracts back to 1993 is used to calculate a measure of work experience for each worker, both overall experience, experience by type of region (cities vs. the rest) and experience in the worker s present firm (firm tenure). The dataset links workers and firms and includes about 150 000 distinct firms. The workers are allocated to 60 industrial sectors. Since the productivity of resource based sectors are unrelated to urbanization, we exclude the primary sectors agriculture, fishing and forestry. In the public sector s are determined by national regulation, and public sector workers are therefore excluded (sectors public administration, education and health care). We are left with 54 sectors; the largest are construction, domestic trade, retail sales and business services. The education register covers the whole adult population and gives information about the highest completed education level in the beginning of October each year. We also have information on the age, gender, immigration status and home region of all individuals. Finally, information on occupation group is available from 2003 onwards at the 4-digit level and includes 350 occupations. 1 The employment register separates between three contract types: Full time contracts with at least 30 hours work per week, part time contracts with 20 29 hours work per week, and part time contracts with less than 20 hours work per week. 2 Self-employed workers are not included.

7 The final dataset includes almost 600 000 workers every year during 2003-2010, giving a total of about 4.7 million worker-year observations. Workers can enter and leave during the 8-year period, and in total about 1 million different workers are included. We estimate equations for the whole sample of workers, as well as for three subgroups of workers according to the level of education: tertiary (workers that have completed at least one year at college/university), secondary (workers that have completed at least one year of secondary education) and primary (workers with not more than compulsory schooling). Workers with secondary education account for the largest share with about 2.5 million observations. The subgroups of workers with primary and tertiary education contain 0.9 and 1.3 million observations, respectively. We start out with a hedonic regression of hourly s for the period 2003-2010 that controls for observable worker characteristics, and includes sector, year and occupation fixed effects: ln w isot city it s o t X it isot (1) where wisot is the hourly income for worker i in sector s, occupation o and year t, city it is the city dummy and is the city premium. Sector, occupation and year fixed effects are represented by, s o and, respectively. t is a vector of parameters and isot is an error term. The vector of observable worker characteristics ( X it ) includes dummies for age (5-year intervals), gender, immigration status and education level (primary, secondary, tertiary), as well as aggregate work experience since 1993 (calculated in days and expressed in years). Next, we exploit the panel dimension of the data to control for unobservable worker characteristics by adding worker fixed effects ( i ): ln w isot city it s o t i X it isot (2)

8 As pointed out by De la Roca and Puga (2014), the city premium ( ) represents a mix of static and dynamic effects of working in cities. In the next step, we allow the value of experience to vary between cities and smaller regions: ln w city X exp_ c (3) isot it s o t i it 1 it isot where exp_ c it is work experience in cities acquired by worker i up until time t. 3 If 1 > 0 ( 1 < 0), growth will be higher (lower) in cities than in smaller regions. The immediate static city premium is given by the estimated coefficient on the city dummy ( ), whereas the city premium after years will be 1 Finally, we add firm tenure and allow the effect of firm tenure to vary between regions: ln w city X exp_ c ten ten city (4) isot it s o t i it 1 it 2 it 3 it it isot where ten represents years of experience in the worker s present firm. 4 Returns on firm it tenure is higher (lower) in cities than in other regions if 3 > 0 ( 3 < 0). The city premium after years of work experience and 1 years in the present firm will now be 2 1 1 3 2. Table 1 presents descriptive statistics for the main variables used in the individual level regressions. As shown in panel a, the average worker in our dataset has an hourly of 243.7 NOK, is about 42 years old, and has about 8.5 years of work experience. Average firm tenure is 4.3 years, and average experience in city regions is 3.9 years. Panel b offers mean values of hourly s, age and work experience for the three education groups. Tertiary educated workers have higher s, are somewhat younger, have less overall work experience and lower firm tenure, but more experience in cities. According to panel c, about 19% of the workers have only primary education, while workers with secondary and tertiary education account for about 53% and 28.5% of the sample, respectively. About 72% of the 3 The regression also includes quadratic experience terms. 4 The regression also includes quadratic experience and tenure terms.

9 workers are male and 12% are immigrants. The allocation of workers between city regions and the rest is roughly equal, but cities have higher share of tertiary educated workers (38% vs. 20%), lower share of male workers (67% vs. 75%) and more immigrants (15% vs. 9%). Table 1 about here 3. Static agglomeration effects across education groups We start out presenting the raw agglomeration effect of cities of 14.1% in column 1 of Table 2. Interestingly, this is exactly the same raw effect as in the dataset for the UK by D Costa and Overman (2014, Table 2). Their definition of city region is 100.000 inhabitants in a travel to work area, somewhat lower than the 150.000 cutoff chosen here. Although city size definition varies according to country setting, the urban gap is typically in this order of magnitude. Glaeser and Mare (2001) find a higher US rural-urban gap of 25%. Bütikofer et al. (2014) apply a longer time series for Norway and estimate a gap of 16%. To control for observed heterogeneity we run an individual level regression over the whole sample including all individual characteristics, as well as year, sector and occupation fixed effects, as described by equation (1) in section 2. The city effect is reduced to 7.5% in column 2. The education gap is about 5% from primary to secondary education and 19% from primary to tertiary education. The male advantage is 13.6%. Non-western immigrants have lower s on average. Experience matters, and the effect is non-linear. Wages increase with experience for the first 25 years, and on average one extra year of experience adds 1% to s. The importance of unobserved characteristics has been a source of concern and only a few studies have been able to follow movers between regions to identify the ability factor. The regression in column 3 includes worker fixed effects (as described by equation (2) in section 2). The city effect is 5.1% when observable and unobservable individual characteristics are taken into account. It follows that about 2/3 of the city gap is accounted for by observable and unobservable individual characteristics. The result is similar to Bütikofer et

10 al. (2014) using a different dataset for Norway. The comparable city effect of D Costa and Overman (2014) controlling for sorting is 2.3%, but their controls are different and they do not include experience variables. Gould (2007) finds somewhat less reduction of the premium for white-collar workers, but larger for blue-collar. Glaeser and Mare (2001) produced worker fixed effect estimates that were about 1/3 of the initial gap. The inference of the effect of education now is based on change of situation for the individuals. The effect of a year of experience increases to almost 8% on average indicating a negative correlation between ability and experience. Table 2 about here Our main interest is the separate city effects for each education group. The raw agglomeration effect (regressions not shown) is distinctly different for the three education groups, increasing from 6.1% for primary educated workers to 12.4% for workers with tertiary education. When controlling for observed heterogeneity (columns 1-3, Table 3), the city effect increases from 4.6% for primary educated workers to 11.1% for workers with tertiary education. When also controlling for unobservable characteristics of workers (columns 4-6, Table 3), the city effects are still increasing in education level, from 3.2% to 6%. Hence, consistent with most US studies as well as the Swedish study by Andersson et al. (2014), we find that the city effect is increasing in education level / skills. This result survives when worker fixed effects are included, but educational differences become smaller. Unobserved abilities are thus somewhat more important for tertiary educated, and explain about half the raw urban premium for this education group. The estimates also show that overall experience is more valuable for highly educated individuals. On average, one extra year of experience increases s of primary and secondary educated by about 7%, while the highly educated get a increase of 9.3%. Table 3 about here

11 4. Dynamic agglomeration effects experience by type of region and firm The dynamics of agglomeration are related to the accumulation of experience. We separate between experience in cities versus smaller regions. Table 4 presents regressions with worker fixed effects and with experience disaggregated based on where it is accumulated (equation 3 in section 2). For all workers (column 1, Table 4), the initial/static city effect now is 5.2%, about the same as the estimated city effect without control for city experience (see column 3, Table 2). The effect of having experience from cities is of economic importance, and given the average city experience of 3.9 years in the data, the dynamic effect adds about 5% to the urban premium. The total city effect consequently is about 10%. Hence, in our dataset half the gain from working in a city comes from the static effect and the other half is a dynamic effect. Table 4 about here Columns 2-4 of Table 4 present separate results for education groups. The effect of city experience on s is positive and increasing in education level during the first 13-14 years. Figure 1 shows the trajectories of the city premium for primary and tertiary educated workers. The city premium of primary educated workers starts at 3.3% (the static city effect) and increases to about 8% after 12 years, whereas the city premium of workers with tertiary education increases from 6.1 % to above 14 %. Hence, the difference between primary and tertiary educated workers increases over time and is after 12 years approximately twice the difference of the static effects. Figure 1 about here We calculate medium-term city premium based on average years of experience in our data set. The average city experience is 4.5 years for the highly educated and the total city premium is 11.1%. The primary educated have on average 3.3 years of experience and their city premium is 5.1%. The difference is statistically significant at the 1% level. The city premium is compared across empirical specifications in Table 7, and the results reported in Table 4 are presented in the first row.

12 Table 5 presents results where firm tenure is added as explanatory variable and the effect of firm tenure is allowed to vary between cities and other regions (described by equation 4, section 2). The effect of firm tenure is U-shaped in all regions and for all education groups. However, the interaction between firm tenure and the city dummy is significantly positive for the primary and secondary educated and significantly negative for workers with tertiary education. Hence, working in cities makes experience in the same firm more valuable for workers without tertiary education and less valuable for workers with tertiary education. A consequence of this result is that the city premium trajectories depend on firm tenure. Table 5 about here A weakness of the analysis is that we have data about worker experience since 1993, which is not the full history of worker experience for many workers. The analysis is repeated for a sample of workers where we have full history of experience, workers born later than 1967. The results reported in Table 6 would have differed from Table 5 if worker experience early in life is important. But it is not. The estimates are very similar to those presented in Table 5. Firm tenure in cities is still positive for primary educated workers and negative for workers with tertiary education, but the difference between the education groups is larger than in Table 5. This might indicate that the gains from firm tenure among primary educated and job switching among tertiary educated are more prominent in the early stages of the career. Table 6 about here Figure 2 is based on the estimates for young workers in Table 6 and shows that the city premium trajectory of tertiary educated workers that change jobs frequently is located above the trajectory of tertiary educated workers that remain in the same firm. The opposite is the case for primary educated workers; for them, the city premium is higher and increases faster when the worker stays in one firm. The trajectories of tertiary educated workers are always located above those of primary educated workers, reflecting the city premium is increasing in education level independent of firm tenure. But the

13 additional city premium of tertiary educated workers is decreasing in firm tenure largest when jobs are changed frequently and smallest when workers stay in the same firm. Figure 2 about here The combined static and dynamic agglomeration effects for all workers and the sample of young workers are shown in rows 2 and 3, respectively, of Table 7. Even though firm tenure affects return on city experience, as shown above, the narrowing down of the data to young workers does not influence the size of the city premiums or differences between education groups significantly (when evaluated at mean city and firm experience). 5. Robustness check Given the volume of data involved here, there are many ways of specifying the models estimated. We have studied the heterogeneity of the population with respect to gender and ethnicity. The results are robust when we study men only and when we exclude foreign immigrants. The importance of other robustness checks are presented in Table 7, and the first three rows summarize the urban premiums in the models above. Table 7 about here We have chosen to investigate the city agglomeration effect in the 7 largest city regions in Norway (above 150.000 inhabitants). The robustness of the cutoff has been investigated. In row 4 of Table 7 (and in Appendix Table 1) we introduce a separation between cities (above 150.000 inhabitants) and small cities (between 65.000 and 150.000 inhabitants) and analyze the importance of experience and agglomeration effect in the two types of cities. The agglomeration effect of the cities is somewhat higher, but is negligible in the smaller cities. The literature has been concerned with the endogeneity of population size and density since the contribution of Ciccone and Hall (1996) handling the endogeneity with instrumentation based on historical population numbers. In an early working paper version we have reported results with instrumentation of a continuous population size variable, and the results are

14 broadly in accordance with the analysis presented here. The challenge with the cutoff procedure is that the endogenous migration may affect what regions are above the cutoff. To investigate this we have thrown out observations from smaller cities close to the cutoff. The result is reported in row 5 of Table 7 (and Appendix Table 2). The city agglomeration effect estimated is hardly affected by this exclusion compared to the situation where a small city effect is separated out. Finally, a major concern in the estimation of city size effects is the role of amenities motivating migration. Four types of amenities have been extensively studied in the literature school quality, cultural services, crime, and climate. We have checked the robustness of the results with respect to a set of amenity variables. The measure of school quality is based on Borge and Naper (2006). They have estimated municipal fixed effects based on individual data of student achievement in English and with other relevant controls. The weighted municipal effects are aggregated to regional school quality. Cultural amenities are measured as net per capita regional spending on museums in the year 2010. Public safety is measured by number of violence related crimes per 1000 inhabitant and as an average over the period 1994-2001. Finally, climate is represented by the average winter temperature during 1971-2000. Descriptive statistics of the amenity variables are given in Table 1 in section 2. The city agglomeration effect including the amenity variables is shown in the last row of Table 7 and documented in Appendix Table 3. The city effect is not much affected and the differences between education groups remain. 6. Concluding remarks We have used register data for all full time workers in the private sector in Norway (about 4.7 million worker-year observations) to study the city premium. The individual panel data include observations of education levels and occupations, as well as labor market, employment sector and firm level affiliation, and also allow for identification of unobserved individual effects based on migration between regions. The main focus is the analysis of differences in city effects for the s across education groups taking into account firm tenure.

15 The analysis takes into account the location and firm-specific work experience of the individuals. The experience is distinguished between cities (more than 150 000 inhabitants) and smaller regions. The initial static premium is not affected by the inclusion of worker experience history, but the experience effect adds to the medium-term effect since experience in large regions is found to be more valuable. The experience effect differs with respect to education, and in particular the highly educated gain from agglomeration. Firm tenure in cities is found to be to the advantage of primary and secondary educated, while tertiary educated workers gain from shifting between firms, especially for young workers. Future work will look in more detail at the individual choices of firm and sector affiliation. References Adamson, D., D. Clark, M. Partridge (2004), Do urban agglomeration effects and household amenities have a skill bias? Journal of Regional Science 44, 201-223. Andersson, M., J. Klaesson and J.P. Larsson (2014), The sources of urban premium by worker skills: Spatial sorting or agglomeration economies? Papers in Regional Science 93, 727-747. Bacolod, M., B. Blum and W. Strange (2009), Skills in the City, Journal of Urban Economics 65, 136-153. Baum-Snow, N. and R. Pavan (2012), Understanding the city size gap, Review of Economic Studies 79, 1, 88-127. Borge, L. and L. Naper (2006), Efficiency potential and efficiency variation in Norwegian lower secondary schools, FinanzArchiv 62, 221-249. Bütikofer, A., D. Polovkova and K.G. Salvanes (2014), What is the city but the people? Changes in urban premium and migrant selection since 1967, mimeo, Norwegian School of Economics. Ciccone, A. and R. Hall (1996), Productivity and the density of economic activity, American Economic Review 86, 1, 54-70. Combes, P-P., G. Duranton and L. Gobillon (2008), Spatial disparities: Sorting matters! Journal of Urban Economics 63, 723-742.

16 D Costa, S. and H. Overman (2014), The urban growth premium: Sorting or learning? Regional Science and Urban Economics 48, 168-179. De la Roca, J. (2011), Selection in initial and return migration: Evidence from moves across Spanish cities, IMDEA Working Paper No. 2011-21. De la Roca, J. and D. Puga (2014), Learning by working in big cities, mimeo CEMFI. Di Addario, S. and E. Patacchini (2008), Wages and the city. Evidence from Italy, Labour Economics 15, 1040-1061. Duranton, G. and H. Jayet (2011), Is the division of labour limited by the extent of the market? Evidence from French cities, Journal of Urban Economics 69, 56-71. Glaeser, E. (1999), Learning in cities, Journal of Urban Economics 46, 254-277. Glaeser, E. and D. Mare (2001), Cities and skills, Journal of Labor Economics 19, 2, 316-342. Glaeser, E. and A. Saiz (2004), The rise of the skilled city, Brookings-Wharton Papers on Urban Affairs 5, 47-94. Gould, E.D. (2007), Cities, workers, and s: A structural analysis of the urban premium, Review of Economic Studies 74, 477-506. Helsley, R. and W. Strange (1990), Matching and agglomeration economics in a system of cities, Regional Science and Urban Economics 20, 189-212. Kwon, I. and E. Meyersson Milgrom (2014), The significance of firm and occupation specific human capital for hiring and promotions, Labour Economics 31, 162-173. Lazear, E. (2009), Firm-specific human capital: A skill-weights approach, Journal of Political Economy 117, 914-940. Lee, S. (2010), Ability sorting and consumer city, Journal of Urban Economics 68, 20-33. Machin, S., K.G. Salvanes and P. Pelkonen (2012), Education and mobility, Journal of the European Economic Association 10, 417-450. Matano, A. and P. Naticchioni (2012), Wage distribution and the spatial sorting of workers, Journal of Economic Geography 12, 379-408. Matano, A. and P. Naticchioni (2013), What drives the urban premium? Evidence along the distribution, IZA Discussion Paper 7811. Rosenthal, S. and W. Strange (2008), The attenuation of human capital spillovers, Journal of Urban Economics 64, 373-389.

17 Schwerdt, G., A. Ichino, O. Ruf, R. Winter-Ebmer and J. Zweimüller (2010) Does the color of the collar matter? Employment and earnings after plan closure, Economics Letters 108, 137-140. Shapiro, J. (2006), Smart cities quality of life, productivity, and the growth effects of human capital, Review of Economics and Statistics 88, 324-335. Stevens, M. (1994), A theoretical model of on-the-job training with imperfect competition, Oxford Economic Papers 46, 537-562. Wang, Z. (2013), Smart city: Learning effects and labor force entry, mimeo, Department of Economics, Brown University. Wasmer, E. (2006), General versus specific skills in labor markets with search frictions and firing costs, American Economic Review 96, 811-831. Wheeler, C. (2001), Search, sorting and urban agglomeration, Journal of Labor Economics 19, 4, 879-899. Wheeler, C. (2004), Wage inequality and urban density, Journal of Economic Geography 4, 421-437. Wheeler, C. (2006), Cities and the growth of s among young workers: Evidence from the NLSY, Journal of Urban Economics 60, 162-184. Winters, J. (2011), Why are smart cities growing? Who moves and who stays, Journal of Regional Science 51, 2, 253-270. Yankow, J. (2006), Why do cities pay more? An empirical examination of some competing theories of the urban premium, Journal of Urban Economics 60, 139-161.

18 Figure 1: Urban premium for primary and tertiary educated workers years after move to city Urban premium primary and tertiary educated - Years after move to city 0.16 0.14 0.12 0.1 0.08 0.06 0.04 0.02 0 0 1 2 3 4 5 6 7 8 9 10 11 12 primary tertiary Figure 2: Urban premium for young workers dependent on education and firm tenure years after move to city 0.18 0.15 0.12 0.09 0.06 0.03 Urban premium young workers dependent on education and firm tenure - years after move to city 0 0 1 2 3 4 5 6 7 8 9 10 11 12 primary - staying in same firm tertiary - staying in same firm primary - shifting firm every year tertiary - shifting firm every year

19 Table 1: Descriptive statistics Panel a Mean St dev Min Max Hourly (in NOK) 243.7 125.5 70 1250 Age 41.8 10.4 25 65 Total work experience (in years) 8.5 4.3 0 17.6 Work experience in cities 3.9 4.9 0 17.6 Firm tenure (in years) 4.3 4.2 0 17.4 Work experience in small cities 2.2 4.3 0 17.5 Average winter temperature (Celsius) -2.3 3.1-12.8 3.0 School quality (English grade) 3.5 0.1 3.3 3.7 Crime (violence incidents per 1000 inhabitants) 5.1 1.7 2.5 9.3 Public expenditure museums (in NOK) 64.7 44.6-19.0 344.9 Panel b Mean values Primary educated Secondary educated Tertiary educated Hourly (in NOK) 199.5 227.5 302.8 Age 42.6 42.4 40.1 Total work experience (in years) 8.2 9.1 7.8 Work experience in cities 3.3 3.6 4.7 Firm tenure 4.3 4.7 3.6 Work experience in small cities 2.2 2.5 1.7 Panel c Share of observations Full sample Cities Rest of country All workers 1 0.463 0.537 Primary education 0.186 0.16 0.209 Secondary education 0.529 0.458 0.59 Tertiary education 0.285 0.381 0.201 Male 0.716 0.674 0.753 Immigrant 0.118 0.148 0.091 Immigrant, western 0.088 0.105 0.074 Immigrant, non-western 0.029 0.044 0.017 Notes: Work experience is calculated in days from 1993 onwards, and expressed in years. The city group is defined as regions with more than 150.000 inhabitants in 2010 (7 regions), while small cities are defined as regions with 2010 population in the range 65.000 150.000 (13 regions). Secondary education corresponds to workers that have completed at least one year of secondary education, while tertiary education includes workers with at least one year at university/college. Western immigrants are defined as immigrants from Europe, Japan, North America, Australia or New Zealand. The average winter temperature is given in Celsius degrees and is the average during 1971-2000. The measure of school quality is based on student performance in English adjusted for student and family characteristics (estimated by Borge and Naper, 2006), and is given on a scale from 1 to 6 with 6 as the best. The number of violence related crimes per 1000 inhabitants is measured as an average during 1994-2001, while net per capita public expenditures on museums is from the year 2010.

20 Table 2: Estimation of urban premium (not including dynamic effects) (1) Dependent variable City 0.141*** (0.0004) (2) 0.075*** (0.0003) (3) 0.051*** (0.0008) 0.092*** (0.0004) -0.0008*** 0.016*** Experience 0.015*** (0.0001) (Experience) 2-0.0003*** Secondary education 0.052*** (0.0004) (0.002) Tertiary education 0.186*** 0.098*** (0.0005) (0.0034) Immigrant, western 0.008*** (0.0005) Immigrant, non-western -0.027*** (0.0009) Male 0.136*** (0.0004) Year fixed effects Yes Yes Yes Sector fixed effects No Yes Yes Occupation fixed effects No Yes Yes Age controls No Yes Yes Worker fixed effects No No Yes Observations 4.710.077 4.710.077 4.710.077 R 2 0.09 0.44 0.78 Notes: The regressions are based on yearly data for all full time workers in the private sector during 2003-2010. The dependent variable is log hourly s. The city group is defined as regions with more than 150.000 inhabitants in 2010, which includes 7 out of 89 regions. Work experience is calculated in days from 1993 onwards, and expressed in years. Sector fixed effects are at the 2-digit level and include 54 sectors. Occupation fixed effects are at the 4-digit level and include 350 occupations. The age controls are given as 5-year intervals. The dummies for education and immigrant status are defined in the notes to Table 1. Standard errors are given in parenthesis. *** indicates significance at the 1 percent level. All regressions include a constant term.

21 Table 3: Estimation of urban premium by education groups (not including dynamic effects) Dependent variable (1) (2) (3) (4) (5) (6) Education group Primary Secondary Tertiary Primary Secondary Tertiary City 0.046*** (0.0007) 0.066*** (0.0004) 0.111*** (0.0006) 0.032*** (0.0024) 0.04*** (0.0013) 0.06*** (0.0013) Experience 0.007*** (0.0003) 0.006*** (0.0002) 0.02*** (0.0003) 0.079*** (0.0008) 0.082*** (0.0005) 0.112*** (0.0007) (Experience) 2-0.0000 0.0000*** -0.0003*** -0.0006*** -0.0005*** -0.0012*** Immigrant, western 0.006*** (0.0012) 0.007*** (0.0007) 0.005*** (0.0009) Immigrant, non-western -0.007*** (0.0015) -0.024*** (0.0014) -0.06*** (0.0017) Male 0.124*** (0.0009) 0.142*** (0.0006) 0.123*** (0.0007) Year fixed effects Yes Yes Yes Yes Yes Yes Sector fixed effects Yes Yes Yes Yes Yes Yes Occupation fixed effects Yes Yes Yes Yes Yes Yes Age controls Yes Yes Yes Yes Yes Yes Worker fixed effects No No No Yes Yes Yes Observations 877.391 2.492.656 1.340.030 877.391 2.492.656 1.340.030 R 2 0.28 0.38 0.41 0.67 0.75 0.79 Notes: The regressions are based on yearly data during 2003-2010 for three subgroups of full time workers in the private sector according to their level of education (primary, secondary, tertiary). The dependent variable is log hourly s. Explanatory variables are defined in the notes to Table 2. Standard errors are given in parenthesis. *** indicates significance at the 1 percent level. All regressions include a constant term.

22 Table 4: Estimation of urban premium, including experience by type of region Dependent variable (1) (2) (3) (4) Education group All Primary Secondary Tertiary City 0.052*** (0.0008) 0.033*** (0.0024) 0.041*** (0.0013) 0.061*** (0.0013) Experience 0.085*** (0.0004) 0.076*** (0.0009) 0.078*** (0.0005) 0.104*** (0.0008) (Experience) 2-0.0005*** -0.0005*** -0.0004*** -0.0009*** Experience cities 0.015*** (0.0003) 0.006*** (0.0007) 0.008*** (0.0004) 0.013*** (0.0005) (Experience cities) 2-0.0006*** -0.0002*** -0.0003*** -0.0005*** Year fixed effects Yes Yes Yes Yes Sector fixed effects Yes Yes Yes Yes Occupation fixed effects Yes Yes Yes Yes Age controls Yes Yes Yes Yes Worker fixed effects Yes Yes Yes Yes Observations 4.710.077 877.391 2.492.656 1.340.030 R 2 0.78 0.67 0.75 0.79 Notes: The regression in column (1) is based on yearly data for all full time workers in the private sector during 2003-2010, while columns (2)-(4) consider subgroups according to workers level of education. The dependent variable is log hourly s. Experience in cities refers to work experience accumulated in the city group, defined as regions with more than 150.000 inhabitants in 2010. Other explanatory variables are defined in the notes to Table 2. Standard errors are given in parenthesis. *** indicates significance at the 1 percent level. All regressions include a constant term.

23 Table 5: Estimation of urban premium, including experience by type of region and firm tenure Dependent variable (1) (2) (3) Education group Primary Secondary Tertiary City 0.032*** (0.0025) 0.039*** (0.0013) 0.064*** (0.0014) Experience 0.086*** (0.0009) 0.086*** (0.0005) 0.109*** (0.0008) (Experience) 2-0.0007*** -0.0005*** -0.001*** Experience cities 0.005*** (0.0007) 0.007*** (0.0004) 0.013*** (0.0005) (Experience cities) 2-0.0002*** -0.0003*** -0.0005*** Firm tenure -0.006*** (0.0003) -0.005*** (0.0001) -0.003*** (0.0002) (Firm tenure) 2 0.0002*** 0.0001*** 0.0001*** Firm tenure x City 0.001** (0.0003) 0.001*** (0.0001) -0.001*** (0.0002) Year fixed effects Yes Yes Yes Sector fixed effects Yes Yes Yes Occupation fixed effects Yes Yes Yes Age controls Yes Yes Yes Worker fixed effects Yes Yes Yes Observations 877.391 2.492.656 1.340.030 R 2 0.67 0.75 0.79 Notes: The regressions are based on yearly data during 2003-2010 for three subgroups of full time workers in the private sector according to their level of education (primary, secondary, tertiary). The dependent variable is log hourly s. Experience in cities refers to work experience accumulated in the city group, defined as regions with more than 150.000 inhabitants in 2010. Firm tenure is defined as years of experience in the worker s present firm. Other explanatory variables are defined in the notes to Table 2. Standard errors are given in parenthesis. ***, ** and * indicate significance at the 1, 5 and 10 percent level, respectively. All regressions include a constant term.

24 Table 6: Estimation of urban premium young workers (born after 1967) Dependent variable (1) (2) (3) Education group Primary Secondary Tertiary City 0.031*** (0.0038) 0.042*** (0.0018) 0.068*** (0.0017) Experience 0.109*** (0.0016) 0.098*** (0.0008) 0.13*** (0.0012) (Experience) 2-0.001*** (0.0001) -0.0006*** -0.0012*** Experience cities 0.005*** (0.0013) 0.008*** (0.0006) 0.014*** (0.0007) (Experience cities) 2-0.0002** (0.0001) -0.0003*** -0.0005*** Firm tenure -0.011*** (0.0006) -0.006*** (0.0003) -0.004*** (0.0005) (Firm tenure) 2 0.0004*** (0.0001) 0.0001*** 0.0001* Firm tenure x City 0.002*** (0.0006) 0.000 (0.0002) -0.003*** (0.0004) Year fixed effects Yes Yes Yes Sector fixed effects Yes Yes Yes Occupation fixed effects Yes Yes Yes Age controls Yes Yes Yes Worker fixed effects Yes Yes Yes Observations 336.678 1.072.027 717.452 R 2 0.58 0.67 0.70 Notes: The regressions are based on yearly data during 2003-2010 for young workers, born after 1967, separated based on their level of education (primary, secondary, tertiary). The dependent variable is log hourly s. Explanatory variables are defined in the notes to Table 5. Standard errors are given in parenthesis. ***, ** and * indicate significance at the 1, 5 and 10 percent level, respectively. All regressions include a constant term.

25 Table 7: Urban premium in different regressions (including dynamic effects) Primary Secondary Tertiary Experience in city 0.051 0.066 0.111 Experience in city and firm tenure 0.051 0.065 0.11 Experience in city and firm tenure young workers 0.048 0.063 0.102 Experience in city, firm tenure and experience in small city 0.059 0.076 0.127 Experience in city and firm tenure excl. small city workers 0.061 0.073 0.123 Experience in city and firm tenure controlling for amenities 0.047 0.063 0.10 Notes: The table shows the urban premium for the three education groups in different regressions. All estimates include both static and dynamic agglomeration effects.

26 App Table 1: Estimation of urban premium controlling for experience in small cities Dependent variable (1) (2) (3) Education group Primary Secondary Tertiary City 0.037*** (0.0027) 0.046*** (0.0015) 0.074*** (0.0017) Small city 0.016*** (0.0031) 0.016*** (0.0017) 0.019*** (0.002) Experience 0.086*** (0.001) 0.085*** (0.0006) 0.107*** (0.0008) (Experience) 2-0.0006*** -0.0004*** -0.0009*** Experience cities 0.006*** (0.0008) 0.008*** (0.0004) 0.015*** (0.0006) (Experience cities) 2-0.0002*** -0.0003*** -0.0006*** Experience small cities 0.002** (0.0008) 0.002*** (0.0004) 0.004*** (0.0006) (Experience small cities) 2-0.0001*** -0.0001*** -0.0002*** Firm tenure -0.006*** (0.0003) -0.005*** (0.0001) -0.003*** (0.0002) (Firm tenure) 2 0.0002*** 0.0001*** 0.0001*** Firm tenure x City 0.001** (0.0003) 0.001*** (0.0001) -0.001*** (0.0002) Year fixed effects Yes Yes Yes Sector fixed effects Yes Yes Yes Occupation fixed effects Yes Yes Yes Age controls Yes Yes Yes Worker fixed effects Yes Yes Yes Observations 877.391 2.492.656 1.340.030 R 2 0.67 0.75 0.79 Notes: The regressions are based on yearly data during 2003-2010 for three subgroups of full time workers in the private sector according to their level of education (primary, secondary, tertiary). The dependent variable is log hourly s. The city group is defined as regions with more than 150.000 inhabitants in 2010, while small cities are regions with population in the range 65.000-150.000 in 2010. Experience in cities and small cities refer to work experience accumulated in the respective city groups. Firm tenure is defined as years of experience in the worker s present firm. Other explanatory variables are defined in the notes to Table 2. Standard errors are given in parenthesis. ***, ** and * indicate significance at the 1, 5 and 10 percent level, respectively. All regressions include a constant term.

27 App Table 2: Estimation of urban premium excluding workers in small cities Dependent variable (1) (2) (3) Education group Primary Secondary Tertiary City 0.034*** (0.0031) 0.042*** (0.0017) 0.072*** (0.0019) Experience 0.085*** (0.0011) 0.085*** (0.0006) 0.111*** (0.001) (Experience) 2-0.0006*** -0.0004*** -0.001*** Experience cities 0.006*** (0.0008) 0.007*** (0.0004) 0.013*** (0.0006) (Experience cities) 2-0.0002*** -0.0003*** -0.0006*** Firm tenure -0.007*** (0.0003) -0.005*** (0.0002) -0.004*** (0.0003) (Firm tenure) 2 0.0002*** 0.0001*** 0.0002*** Firm tenure x City 0.001** (0.0003) 0.001*** (0.0001) -0.001* (0.0003) Year fixed effects Yes Yes Yes Sector fixed effects Yes Yes Yes Occupation fixed effects Yes Yes Yes Age controls Yes Yes Yes Worker fixed effects Yes Yes Yes Observations 647.189 1.830.635 1.067.734 R 2 0.67 0.75 0.79 Notes: The regressions are based on yearly data during 2003-2010 for three subgroups of full time workers in the private sector according to their level of education (primary, secondary, tertiary). The dependent variable is log hourly s. Workers in small cities (defined as regions with population in the range 65.000-150.000 in 2010) are excluded. Explanatory variables are defined in the notes to Table 5. Standard errors are given in parenthesis. ***, ** and * indicate significance at the 1, 5 and 10 percent level, respectively. All regressions include a constant term.

28 App Table 3: Estimation of urban premium controlling for regional amenity level Dependent variable (1) (2) (3) Education group Primary Secondary Tertiary City 0.028*** (0.0028) 0.037*** (0.0016) 0.056*** (0.0016) Experience 0.086*** (0.0009) 0.086*** (0.0005) 0.109*** (0.0008) (Experience) 2-0.0006*** -0.0005*** -0.001*** Experience cities 0.005*** (0.0007) 0.007*** (0.0004) 0.013*** (0.0005) (Experience cities) 2-0.0002*** -0.0003*** -0.0006*** Firm tenure -0.006*** (0.0003) -0.005*** (0.0001) -0.003*** (0.0002) (Firm tenure) 2 0.0002*** 0.0001*** 0.0001*** Firm tenure x City 0.001** 0.001*** (0.0003) (0.0001) Regional amenities: Winter temperature Yes Yes Yes School quality Yes Yes Yes Crime Yes Yes Yes Public expenditure museums Yes Yes Yes -0.001*** (0.0002) Year fixed effects Yes Yes Yes Sector fixed effects Yes Yes Yes Occupation fixed effects Yes Yes Yes Age controls Yes Yes Yes Worker fixed effects Yes Yes Yes Observations 877.391 2.492.656 1.340.030 R 2 0.67 0.75 0.79 Notes: The regressions are based on yearly data during 2003-2010 for three subgroups of full time workers in the private sector according to their level of education (primary, secondary, tertiary). The dependent variable is log hourly s. Explanatory variables are defined in the notes to Table 5. The regressions control for four regional amenity variables. Climate is measured by the average winter temperature during 1971-2000 (in Celsius degrees). The measure of school quality is based on student performance in English adjusted for student and family characteristics (estimated by Borge and Naper, 2006), and is given on a scale from 1 to 6 with 6 as the best. The average number of violence related crimes per 1000 inhabitants during 1994-2001 measures public safety, while net per capita public expenditures on museums in 2010 represents cultural amenities. Standard errors are given in parenthesis. ***, ** and * indicate significance at the 1, 5 and 10 percent level, respectively. All regressions include a constant term.