Exploring Wage Determination by Education Level

Abstract

The purpose of this study is to explain urban wage differentials with a special focus on educational levels. The authors explore whether the share of people with a bachelor’s degree or higher in the community matters to the wages of those within specific educational cohorts, accounting for cost of living, human capital externalities, consumer externalities, policy factors, and local labor market conditions. Using data for all U.S. Metropolitan Statistical Areas between 2005 and 2012, the authors find that the presence of more highly educated people will result in a higher median wage in the community overall, as do many studies, but that this factor does not significantly increase the wage for any individual education cohort. These results are hidden if we only look at the entire workforce in the aggregate.

Keywords

labor median wage urban wage differentials education human capital externalities MSA

Local leaders across the United States strive to create conditions for better employment opportunities, higher incomes, and less inequality for their constituents. Research has found evidence that productivity is higher in urban areas, which leads to higher wages, after accounting for higher costs of living (Glaeser & Maré, 2001). This urban wage premium has numerous proposed explanations, with concentrations of higher education and skills being key variables (see a comprehensive review by Heuermann, Halfdanarson, & Suedekum, 2010). Cities have been shown to be special in numerous ways. For example, cities “speed the accumulation of human capital” (Glaeser & Maré, 2001), cities have “something in the air” (Krupka & Noonan, 2013), and cities attract “creative” people (Florida, 2012).

Some research has suggested that the benefits of having an educated labor force appear to go beyond the higher incomes earned by knowledge workers concentrated in cities. Moretti (2013), for example, argues in his book, The New Geography of Jobs, that there is a positive effect between the share of an area’s population with at least a bachelor’s degree and the average wages earned by less educated workers in a particular geographic area. As he puts it, “Brain hubs pay high average salaries to unskilled workers too” (p. 97). This type of spillover, if and where it exists, makes the education factor even more valuable to communities.

In this study, we examine these knowledge spillovers further by analyzing wage differences across cities using median wages by educational cohorts. Specifically, our study examines the effect of the presence of human capital levels on annual median wages for workers with different education levels across 374 U.S. metropolitan statistical areas (MSAs). We also incorporate a substantial time frame covering 2005 to 2012. We explore the relationship between the overall educational attainment of a community and the median wage for different educational attainment groups. The dependent variable is the natural log of the real median annual wage for the working age population 25 and older with earnings by level of educational attainment from the American Community Survey (ACS).

One of the critiques of the numerous studies that have been done to explain wage differentials is that each tends to focus narrowly on one explanation with other factors left out (Hanson, 2000). A strength of our approach in this study is that we explore human capital externalities and their potential spillovers while also considering other variables found to be important in the literature, including cost of living, amenities, labor market conditions, and policy. We find evidence that the estimated impact of each of these explanations varies by educational cohort. Unlike many studies, we include a measure of labor market conditions to control for business cycle effects across regions, which turns out to be very important in explaining wage variation.

A second contribution of this study is that we use a partial cost-of-living adjustment, as suggested by Dumond, Hirsch, and MacPherson (1999), utilizing newly developed regional price parities (Aten, Figueroa, & Vengelen, 2015) and data that allow us to separate housing costs from other cost-of-living factors. This is the first time, to our knowledge, that this has been done for all U.S. MSAs. This approach avoids misleading estimates of real wages as demonstrated in Dumond et al. (1999), and suggests that housing should be treated partially as an amenity and not just a cost factor.

A third contribution of this study is to analyze the explanatory power of variables from the literature by educational attainment; in other words, our approach allows us to gain insights by looking at sublabor markets defined by education levels. Our results suggest that in terms of the aggregate median wage, adjusted properly for cost of living differences, having a better educated population does not significantly increase other people’s wages. Contrary to other studies and to what policy makers might hope, our results suggest that the wages of individual education cohorts are not substantively affected simply by the presence of a better educated population.

The Urban Wage Story

The strong linkage between education and wages receives a great deal of attention in the U.S. labor market. By one calculation, the inflation-adjusted wage gap between a family with high school degrees and one with college degrees increased $30,000 between 1979 and 2012 (Porter, 2014). Based on the ACS data, there are large differences in the annual median salaries across groups with different education attainment (Figure 1). The median annual salary for those with a graduate degree in the United States was $65,164 in 2012 compared with only $19,404 for those without a high school diploma, and the salary difference between someone with a high school degree and a bachelor’s degree increased $825 (in real 2012 dollars) in the 7 years between 2005 and 2012. The other trend, however, is that for all education levels, the inflation adjusted U.S. median wage has been falling in recent years. The declines have been larger for the less educated—between 1.6% and 1.9% per year for individuals with some college or an associate degree or less education—compared to an average decline of 0.7% per year for college graduates and 0.5% per year for those with graduate degrees.

Figure 1.

U.S. median annual salaries by education level, adjusted for inflation (constant 2012 dollars): 2005 to 2012.

Underlying these trends are important regional variations. For example, in 2012 the median wage for high school graduates in Milwaukee was $28,708 and only $25,693 in Los Angeles, while the median salary for a person with a graduate degree was $66,024 in Milwaukee and $73,642 in Los Angeles. Most people might simply assume that people living in Los Angeles had higher incomes than people living in Milwaukee, and for people with graduate degrees that is true, but not for people with only a high school diploma. So although more education is associated with higher incomes in both places, there are other factors that must explain why the median worker with just a high school degree is paid more in Milwaukee than the similarly educated worker in Los Angeles, while the person with a graduate degree earns more in Los Angeles. One explanation could be differences in the cost of living, but cost-of-living differences cannot explain all of the variance in income between Los Angeles and Milwaukee. Furthermore, in the case of a person with only a high school degree, incorporating those differences would widen Milwaukee’s advantage.¹

A study by the Federal Reserve Bank of Cleveland also found that metro areas with a higher share of workers with a bachelor’s degree or more had higher wages and lower unemployment rates, using data for 2011 for the 100 largest metropolitan areas (Richter & Nelson, 2014). They concluded that these correlations suggest that less-educated workers may benefit from working in the same geographical area with a more educated population. This is consistent with Moretti’s work based both on individual data (Moretti, 2004a) and average wages (Moretti, 2013).

Using data from the ACS for 2005 to 2012, in comparison with the Cleveland Federal Reserve study (Richter & Nelson, 2014), we also find a consistent, positive correlation between the share of the population aged 25 to 64 with a bachelor’s degree and the median wage when all education cohorts are put together (Table 1). Table 1 shows the Pearson correlation coefficient between the share of the population aged 25 to 64 with at least a bachelor’s degree and the median wage in a metropolitan area by year. The positive correlation is very strong and highly significant for all workers with a correlation value between .618 and .688. The correlation is also fairly strong for different subgroups defined by their educational attainment. For the subgroups the correlation is positive and significant at the 1% level in 37 out of 40 possible year-by-educational-attainment cells, and it is significant at the 5% level in 2 other cells. In 2012, the correlation between the community’s educational attainment and the median wage was not significant at the 5% level for individuals who did not complete high school, but even this positive correlation was significant at the 10% level.

Table 1.

Pearson Correlation Coefficients Between Nominal Median Wage by Education Attainment With the Share of the Population Aged 25 to 64 With a Bachelor’s Degree or More: 2005 to 2012.

	2005	2006	2007	2008	2009	2010	2011	2012
Nominal median wage (N)	374	374	374	374	374	374	374	374
All education levels	.618***	.648***	.618***	.659***	.675***	.659***	.688***	.672***
Did not graduate high school	.206***	.253***	.108**	.145***	.159***	.159***	.250***	.101*
High school graduate or GED	.359***	.377***	.313***	.360***	.360***	.286***	.299***	.300***
Some college or associate	.326***	.320***	.295***	.362***	.347***	.359***	.388***	.338***
Bachelor’s only	.123**	.192***	.165***	.301***	.280***	.231***	.204***	.272***
Graduate degree	.260***	.236***	.234***	.309***	.264***	.258***	.291***	.334***

Note. ***, **, and * denote coefficients that are significant at the 1% level, 5% level, and 10% level, respectively.

These results lend support to the hypothesis that all workers benefit from having a well-educated community; however, as we shall see, this analysis does not hold up under more rigorous tests.

Although aggregate correlations between education and wages are evident, how robust this relationship actually is once local conditions and other factors are considered is a key question (Fontes, Simoes, & Hermeto Camilo de Oliveira, 2010). In addition, although evidence of an urban wage premium is fairly well accepted, the explanations for the premium continue to be explored (Heuermann et al., 2010). Most explanations focus on ways that education may increase productivity through social returns to education (Moretti, 2004b; Rauch, 1993), proximity to college-educated workers (Rosenthal & Strange, 2008), or utilization of skills (Combes, Duranton, & Gobillon, 2008; Rodriguez-Pose & Vilata-Bufi, 2005), for example, whereas other studies argue that the nature of cities themselves increase productivity due to scale from density, agglomeration effects, and other local conditions that enhance local resources (Fingleton, 2003; Glaeser & Maré, 2001; Wheaton & Lewis, 2002).

Results from studies looking at more generalized spillovers from the presence of educated workers to wages of less educated workers have been mixed (Bratti & Leombruni, 2014; Schumacher, Dias, & Tebaldi, 2014). In his Journal of Econometrics article, Moretti (2004a) found that for cities in the United States, an increase in the supply of college graduates raised the wages of college graduates and of those that did not graduate from high school, and in his book, Moretti (2013) found broader spillovers to noncollege workers. A recent study by Schumacher et al. (2014), following Acemoglu (1996), found that concentrations of highly educated employees have a positive spillover to all workers in the business services sector in the United States and Brazil, with a spillover to other sectors in Brazil but not in the United States. In contrast, other studies have found small or no spillover effects. For example, Bratti and Leombruni (2014) found a small spillover from the presence of college-educated workers to white-collar wages but not to blue-collar workers in Italian manufacturing. Using U.S. data, both Acemoglu and Angrist (2000) and Ciccone and Peri (2006) found insignificant spillovers from education levels to wages.

In this study, using the share of the population with a bachelor’s degree or more as our measure of human capital stock, we explore possible spillover effects to the wages of five levels of educated workers in U.S. MSAs.

Literature and Variable Choices

Heuermann et al. (2010) outline two sets of literatures focusing on explaining either urban wage premiums and or human capital externalities. The two are closely related. In both approaches, two sets of variables are typically used in explaining regional wage differences. The first set is individual characteristics of workers and the second is characteristics of the local economy where the individuals work.

In this study, we identify the “median” person by educational attainment in each metro area and attempt to explain what that person is earning by regressing the wages for that median person against community and state variables. Our equation can be represented as follows:

L n W_{i t} = α + β X_{i t} + γ_{1} Z_{1 i t} + γ_{2} Z_{2 i t} + ϕ_{i t} + ε_{i t},

(1)

where Ln W is the log of the real median annual wage for the group of workers with the same educational attainment level; X is the vector of local human capital characteristics, including the share of the population with a bachelor’s degree or more, and the share of young workers as a gauge of experience; Z₁ is the vector of local community characteristics including metro labor market conditions, the local cost of housing, and the presence of amenities; Z₂ is the vector of state-level policy characteristics, including the legally mandated minimum wage adjusted for local area nonhousing cost of living differences, and the statewide tax rate measured by total state and local taxes as a share of personal income; φ is the vector of unobservable individual characteristics; and ε is the error term.

Our dependent variable is the median wage, W, and is partially adjusted for cost-of-living differences, measured in U.S. 2012 dollars. We prefer using the median wage over the mean. First, to calculate the mean wage by educational attainment one would need to use the microdata sample from the ACS. Two problems with this data set are that the respondent’s location is only identified by Public Use Micro Area, which does not always correspond to a metropolitan area, and second, the income data are top coded based on the highest 0.5% in a state. In most states, this would imply a top-coded value of between $300,000 and $500,000. The wages of individuals whose actual wage exceeds the top code value are assigned a wage value equal to the top code value. The tabulated data for metropolitan areas published by the U.S. Census Bureau (n.d.) do not have these limitations.

Second, and more important, the mean is skewed upward by the presence of very high wage individuals. Variation in mean and median wages can be seen from data available from the U.S. Census Bureau for all year-round, full-time workers aged 16 and older by gender. For example, for male workers in 2012 the mean wage in the United States was $64,650 and the median wage was $47,473, which is a difference of 36%. In 2012, the metropolitan area with the largest difference was Bridgeport-Stamford, Connecticut, where the mean wage for male workers was $124,784, whereas the median wage was only $70,970, a difference of 76%. In Los Angeles, the mean wage for male workers was 45% higher than the median wage, whereas in Milwaukee the mean was only 28% higher than the median.

Moreover, the degree of skewness in the ACS microdata sample varies by level of educational attainment. In Los Angeles, the mean wage for someone with only a high school degree in 2012 was 27% higher than the median wage, whereas in Milwaukee the mean wage for a high school graduate was only 8% higher than the median wage. The disproportionate presence of some high-wage, high school graduates tends to pull up the mean wage much more in Los Angeles than it does in Milwaukee. The mean wage for workers with a graduate degree is 31% higher than the median wage in Los Angeles, while in Milwaukee it is 33% higher. The variance across MSAs in the difference between the mean and the median is much higher for the lower educational attainment categories. To eliminate this distortion we use the ACS data on the median wage published by the U.S. Census Bureau.²

We adjust the median wage variable to account for nonhousing differences in the cost of living. Our cost-of-living adjustment is based on regional price parities developed by the Bureau of Economic Analysis (BEA; Aten et al., 2015). The adjustment is based on a weighted average of the goods and nonhousing related services cost indices. The housing cost index, which is based on the median rent data from the ACS, is included as an explanatory variable. Winters (2009) showed that rents were a better measure of housing costs than owner-occupied housing values. Furthermore, regional differences in wages and incomes should only partially adjust for differences in the local cost of living because the local price of housing incorporates some location amenity value (Dumond et al., 1999). The BEA data are also based on direct price measures rather than indirectly using wages (Riefler, 2007). We then extend this measure back in time to derive price parity index values for the period 2005 to 2007, since the BEA estimates start in 2008. (See Appendix A for further discussion of this procedure.)

We study the effect of human capital externalities on the real wage for our whole sample, which is comparable to many previous studies. We also study these relationships for five disaggregated groups distinguished by the standard categories of education attainment: (1) did not graduate from high school, (2) high school diploma or a graduate equivalent degree (GED), (3) associate college degree or some college, (4) bachelor’s degree only, and (5) graduate degree. Very few studies have used this type of disaggregated data, and none that we could find use them to analyze urban wage differentials.³

In addition to disaggregating by education level as a way to understand the human capital factors, for each education equation we measure the local human capital characteristics with two variables: human capital intensity and experience. We measure human capital intensity by the share of the population aged 25 to 64 that has a bachelor’s degree or more. If having more educated people nearby has a knowledge spillover effect on the wages of others, then we would expect to see a positive coefficient on this variable. We measure experience by the share of the working age population that is young. Specifically, we define “young” as the share of the working age population that is between 25 and 34 out of the total population aged 25 to 64. This variable is specific to each educational attainment cohort. For example, the population aged 25 to 34 with a graduate degree is divided by the total population aged 25 to 64 with a graduate degree. Younger workers tend to get paid less than older workers, presumably because they are less productive; thus a community where a bigger share of workers are younger, within each educational attainment cohort we would expect the workers in that cohort to be paid less.

Local labor market conditions are captured with the share of employed individuals (aged 25 to 64) relative to the total prime working age (25 to 64) population. This labor market tightness variable is specific to each educational attainment cohort. For example, the number of employed people who have not completed high school is divided by the total population that has not completed high school. For two of the regression equations, those for the median wage of individuals with a bachelor’s degree or a graduate degree, the independent variable used to measure labor market conditions is the number of employed people with a bachelor’s degree or more divided by the population with a bachelor’s degree or more. The employment data for the population aged 25 to 64 were not available separately for people with only a bachelor’s degree or a graduate degree.

If labor markets are tight, where a large proportion of the potential workforce is employed, then we would expect upward pressure on wages. Because labor market conditions in the aggregate tend to be positively correlated with an area’s educational attainment (in the aggregate the more educated communities have more people working), it is desirable to include this measure by educational attainment category (instead of simply looking at the aggregate labor market). This will ensure that it is the labor market conditions for a particular type of worker that are determining the wage for that particular type of worker. There is some concern that this variable could be endogenous to this equation, as there could be unobserved or unmeasured characteristics of a metro area’s industry and employment mix that could lead to both higher wages and a tighter labor market. We examine this issue empirically.

We include two policy variables that vary by state: the legislatively mandated minimum wage adjusted for local nonhousing costs of living and expressed in constant 2012 dollars, and a tax rate variable. These are included to capture public policy efforts to directly influence the local wage, or to indirectly influence the local gross wage by reducing or increasing the after-tax wage. We would expect that the minimum wage would have a positive effect on the median wage, especially for the less educated cohorts, and that the tax rate would have a positive effect on the median wage if the equilibrium wage rate is determined by after-tax income. Note that because the minimum wage variable is adjusted for differences in the local nonhousing cost of living, it varies by metro area.

Finally, we include a set of variables to capture various amenities. There are both positive and negative consumption externalities that can affect wages that employees will accept (Dimou, 2012; Gabriel & Rosenthal, 1999; Roback, 1982). For example, Florida’s (2012) work on the “creative class” poses the hypothesis that people are attracted to places with an array of activities with opportunities for interaction. In this study, we include three desirable amenities: presence of leisure and culture (includes employees at parks, museums, amusement parks, golf courses, ski resorts, nonhotel casinos, etc.), higher educational institutions, and public transportation. The presence of these amenities is measured by the share of employment in each of these industries in the MSA as reported by the Bureau of Labor Statistics.⁴ We also include the share of employment in durable manufacturing, expecting that this is a negative amenity in that people may require a higher wage to work in the challenging working conditions in factories or to live near such factories. We expect that the desirable amenities would be negatively related to the median wage for each educational attainment category, while the undesirable amenity would be positively related to wages.

We used a fixed effect estimation technique to account for the metro area factors that do not change over time, and we added year dummies to control for effects that vary by year but not by geography. We also estimate our equations using metro areas larger than 500,000 in 2012 (104 metro areas, thus a sample similar to Richter & Nelson, 2014), less than 500,000, and all MSAs together. (See Appendix B for a list of the 374 MSAs; available online at edq.sagepub.com/supplemental.) This allows us to determine if there are differences in the factors that influence wages in larger and smaller metro areas (Echeverri-Carroll & Ayala, 2011).

Results

The descriptive statistics for each of our variables are reported in Table 2.

Table 2.

Descriptive Statistics.

	Minimum	Maximum	Mean	Standard deviation
Median wage all, 2012$, adjusted for nonhousing cost of living	20,457	61,419	35,054	4,365
Median wage did not graduate from high school, 2012$, adjusted for nonhousing cost of living	5,865	58,169	20,732	3,888
Median wage high school graduate or GED, 2012$, adjusted for nonhousing cost of living	17,924	52,682	28,986	3,350
Median wage some college or associate’s degree, 2012$, adjusted for nonhousing cost of living	19,486	60,061	34,561	3,873
Median wage bachelor’s degree, 2012$, adjusted for nonhousing cost of living	22,773	80,990	47,525	5,950
Median wage graduate degree, 2012$, adjusted for nonhousing cost of living	6,053	114,106	62,020	8,361
Price parity index housing	48.0	195.5	88.8	24.9
Share of population 25 to 64 with bachelor’s or more	10.0	62.0	26.7	8.3
Share of population 25 to 64 who are 25 to 34, all	13.1	43.8	25.4	3.4
Share of population 25 to 64 who are 25 to 34, did not graduate from high school	2.7	66.6	26.3	6.5
Share of population 25 to 64 who are 25 to 34, high school graduate or GED	10.5	46.4	23.5	4.8
Share of population 25 to 64 who are 25 to 34, some college or associate’s	10.8	50.0	26.8	4.4
Share of population 25 to 64 who are 25 to 34, bachelor’s	5.8	51.9	28.4	6.0
Share of population 25 to 64 who are 25 to 34, graduate degree	0.4	39.5	19.0	5.9
Share of population 25 to 64 who are employed, all	50.6	86.7	72.3	5.4
Share of population 25 to 64 who are employed, did not graduate from high school	14.8	84.4	52.1	9.8
Share of population 25 to 64 who are employed, high school graduate or GED Share of population 25 to 64 who are employed, some college or associate’s Share of population 25 to 64 who are employed, bachelor’s or more Minimum wage, 2012$, adjusted for nonhousing cost of living State and local taxes as a share of personal income Share of total employment in durables manufacturing Share of total employment in arts and recreation services Share of total employment in 4-year colleges and universities Share of total employment in public transit	43.5 47.5 47.8 5.39 6.94 .32 .24 0 0	90.0 90.1 93.7 9.47 14.5 43.82 18.27 24.84 1.00	68.6 75.0 82.3 7.38 9.74 6.64 1.64 2.47 .79	6.0 5.1 4.0 .85 1.3 4.67 1.17 3.04 .10

We begin by estimating a base model that includes the human capital variables and the housing price cost of living variable as explanatory variables (Table 3), followed by estimations for our full model shown in Table 4. The results for the year effects are shown in Appendix C. Note that the dependent variable, the housing price variable, and the minimum wage variable used in our estimations are in natural logs, and the other variables are measured as shares. The results come from a fixed effects model estimated for all U.S. metropolitan areas over 8 years from 2005 through 2012.⁵ The standard errors are calculated from a covariance matrix estimator that adjusts for heteroscedasticity and serial correlation at the level of the metropolitan area.⁶ The estimated standard errors, clustered at the MSA level, are reported below each coefficient.

Table 3.

Results for Base Model.

Dependent variable: Natural log of median wage in 2012 U.S. dollars, partially adjusted for metro area cost of living differences
	1	2	3	4	5	6
Variable	All	Did not graduate from high school	High school graduate or GED	Some college or associates	Bachelor’s only	Graduate degree
Constant	8.53075***	7.98410***	8.25544***	8.73598***	9.49874***	10.57434***
	(.17516)	(.50944)	(.22652)	(.24029)	(.22266)	(.29897)
Ln (price parity index, housing)	0.40250***	0.45736***	0.46494***	0.41212***	0.30555***	0.10956
Ln (price parity index, housing)	(.03959)	(.11550)	(.05050)	(.05420)	(.05017)	(.07113)
Share of population (25-64) with bachelor’s or more	0.00626***	0.00152	0.00024	−0.00056	0.00009	0.00192
Share of population (25-64) with bachelor’s or more	(.00061)	(.00217)	(.00107)	(.00083)	(.00104)	(.00169)
Population (25-34)/population (25-64), education specific	−0.00004	−0.0026***	−0.00084	−0.0026***	−0.00250***	−0.00401***
Population (25-34)/population (25-64), education specific	(.00069)	(.00078)	(.00057)	(.00055)	(.00045)	(.00044)
R² overall	0.3785	.0803	.1529	.2517	.1067	.1164

Note. ***, **, and * denote coefficients that are significant at the 1% level, 5% level, and 10% level, respectively.

Table 4.

Results for Full Model.

Dependent variable: Natural log of median wage in 2012 U.S. dollars, partially adjusted for metro area cost of living differences
	1	2	3	4	5	6
Variable	All	Did not graduate from high school	High school graduate or GED	Some college or associate’s	Bachelor’s only	Graduate degree
Constant	8.39161***	8.16493***	8.14249***	8.65223***	9.3965***	10.49469***
Constant	(.17225)	(.50593)	(.22372)	(.23316)	(.23353)	(.33875)
Ln (price parity index, housing)	0.34186***	0.32311***	0.40306***	0.34866***	0.27899***	0.09612
Ln (price parity index, housing)	(.03853)	(.11384)	(.04758)	(.05300)	(.05090)	(.07140)
Share of population (25-64) with bachelor’s or more	0.00546***	0.00173	−0.00013	−0.00052	0.00005	0.00184
Share of population (25-64) with bachelor’s or more	(.00056)	(.00215)	(.00103)	(.00084)	(.00105)	(.00170)
Population (25-34)/population (25-64), education specific	−0.00104	−0.0030***	−0.00117**	−0.00279***	−0.00265***	−0.00401***
Population (25-34)/population (25-64), education specific	(.00066)	(.00079)	(.00056)	(.00054)	(.00045)	(.00044)
Employment (25-64)/population (25-64), education specific	0.00454***	0.00340***	0.00418***	0.00392***	0.00223***	0.00081
Employment (25-64)/population (25-64), education specific	(.00056)	(.00075)	(.00049)	(.00052)	(.00085)	(.00088)
Ln (minimum wage)	0.03569**	0.09864**	0.04251*	0.01126	0.00898	−0.00967
Ln (minimum wage)	(.01653)	(.04633)	(.02280)	(.02099)	(.02284)	(.02977)
State and local taxes/personal income	0.00429*	0.00456	0.00324	0.00413	0.00076	0.00808*
State and local taxes/personal income	(.00260)	(.00898)	(.00360)	(.00386)	(.00434)	(.00490)
Employment share, durables manufacturing	0.00461***	0.01411***	0.00412**	0.00598***	0.00395	0.00104
Employment share, durables manufacturing	(.00142)	(.00451)	(.00190)	(.00209)	(.00246)	(.00238)
Employment share, arts and recreational services	−0.01120**	−0.0485***	−0.00929	−0.01645**	−0.00765	0.00027
Employment share, arts and recreational services	(.00527)	(.01723)	(.00645)	(.00725)	(.00761)	(.00956)
Employment share, 4-year colleges	−0.00110	−0.00633	−0.00920*	−0.00114	0.00044	0.00504
Employment share, 4-year colleges	(.00385)	(.01148)	(.00530)	(.00438)	(.00472)	(.00464)
Employment share, public transit	−0.00409	−0.12374	0.06329	−0.05902	0.01246	−0.06766
Employment share, public transit	(.04328)	(.11831)	(.05623)	(.06645)	(.09181)	(.08069)
R² overall	.4829	.1304	.2304	.3477	.1466	.0826

Note. ***, **, and * denote coefficients that are significant at the 1% level, 5% level, and 10% level, respectively.

The first column in Table 3 reports results for the whole sample, combining all education levels. Our primary interest is in the impact of human capital intensity measured as the share of the population with a bachelor’s degree or more; this variable has a positive and statistically significant coefficient. The more educated the community, the higher the median wage. In our base model, a 10% increase in an area’s share of its working age population with at least a bachelor’s degree, say from 30% to 33%, would result in almost a 2% increase in the overall real median wage from $35,000 a year to $35,657 a year.

The coefficient on the housing price variable suggests that an area’s median wage partially responds to an increase in housing costs (coefficient value of 0.40), indicating that some of the variance in housing prices reflects the amenity value of various communities. A coefficient estimate of 1.0 would imply that workers require a wage rate sufficient to completely offset higher housing costs. A point estimate of 0.0, on the other hand, would imply that the value of the place amenity would completely offset the higher cost of housing.

Our measure of experience is the share of the working age population that is relatively young. This variable is not significant when all education groups are combined but is significant at the 1% level, as expected, once we control for education attainment, except for high school graduates where it is insignificant.

The more interesting results come from the regressions that consider median wages stratified by educational attainment (Table 3; columns 2-5). Here the base case results show that the impact of a more highly educated workforce is generally quite small and is not statistically significantly different from zero. The higher level of wages in a better educated community appears to reflect each individual’s own educational attainment, with no measurable spillovers to the median wages of other workers. This result is inconsistent with the correlations presented in Table 1, and conflicts with the argument that Moretti (2013) presents in his book as well as with the results from the Cleveland Federal Reserve study (Richter & Nelson, 2014).

The full model presented in Table 4 includes several other conditioning variables, as described in the data section above. These variables are the log of the minimum wage, adjusted by metro-area cost of living but not by housing costs, the measure of labor market tightness, a measure of state tax rates, and the amenity variables. The estimated coefficients for the human capital and housing price variables are similar to those obtained in the base case models. The coefficient on the housing price variable varies by educational attainment category but remains significant among all educational levels, except for individuals with a graduate degree. This coefficient was also insignificant in the base regression for people with a graduate degree, which suggests that people with a graduate degree are not compensated for higher housing costs with higher wages. Perhaps people with a graduate degree consider that higher housing costs reflect the amenity value of place. People with less education do require higher pay to work in communities with higher housing costs, indicating, that for them, the cost of housing exceeds the amenity value of place, and wages partially adjust to compensate.

Similarly, adding the other explanatory variables does not change the results for the community educational attainment variable. Again, there appears to be no spillover wage benefits for workers because they are living in communities with a relatively large share of educated residents.

Results for the control variables are also interesting. We find, not surprisingly, that metro areas with younger workforces, stratified by the educational attainment of the cohort, have a significant and negative effect on the median wage. For example, an increase in the share of the work-age population with a bachelor’s degree that is relatively young (aged 25 to 34) from 25% to 30% would reduce the median wage of that segment of the workforce from about $45,000 to $44,406.

Our measure of labor market conditions is positive and significant except for individuals with a graduate degree. Because the job market for people with a graduate degree tends to be geographically larger than the local metro area, it is less surprising that their wages would not be influenced by conditions in the local labor market. However, an increase in the employment-to-population ratio from 55% to 58% for those with a high school degree would increase the median wage of high school graduates from about $30,000 to $30,379. There is some concern that this variable could be jointly determined along with the median wage, as there are perhaps unobserved factors that increase the tightness of a local labor market and also push up the median wage in that market. To mitigate this concern, we also estimated this model using a lagged measure. The results for all variables were virtually identical, so we present only one set of results.⁷

The minimum wage has a positive effect on median wages for all of the educational attainment groups (and the entire workforce) except for people with a graduate degree. It is significant at the 5% level for all workers without controlling for educational cohort, at the 5% level for high school dropouts, and at the 10% level for high school graduates. So the least educated workers are likely to see an increase in the median wage if there is an increase in the statutory minimum wage in their community. For individuals who did not complete high school, an increase in the minimum wage from $7.25 an hour to $8.25 an hour would raise the median wage for this group from about $20,000 a year to $20,257 a year.

The state and local tax rate variable tended to be positive across the educational attainment categories, but is only significant at the 10% level for the all worker categories and individuals with a graduate degree. The point estimates indicate that much of an increase in state and local taxes are offset by an increase in the gross wage. For example, if state and local taxes were increased from 10% to 11%, then the typical worker with a graduate degree with a median income of $62,000 would pay an additional $620 per year in taxes. Our model, however, estimates that their gross income would increase from $62,000 to $62,503, offsetting 81% of those higher taxes.⁸

Among our amenity variables the strongest results were reported for the share of total wage and salary employment in durables manufacturing. This coefficient was positive for all groups and was significant at the 5% level of all educational attainment categories, except for people with a bachelor’s degree and people with a graduate degree. This indicates that wages for lower educational attainment workers tend to be higher in areas with a relatively large share of employment in durables manufacturing. For workers with some college education or an associate’s degree, an increase in the share of employment in durables manufacturing from 5% to 8% would increase the median annual wage from about $35,000 to $35,638. This result is consistent with an interpretation that the presence of durable manufacturing results in a higher median wage to compensate for the negative amenity. The results could also reflect the fact that durables manufacturing tends to be relatively well paid and employs a disproportionate share of less educated workers. Durables manufacturing, however, represented a very small share of employment in most metropolitan areas and thus this explanation seemed unlikely.⁹

The estimated year effects are also quite interesting (see Appendix C). For example, the dummy variables for the year 2007 were generally insignificant. This year, of course, is the peak year before the Great Recession and the weak national labor market conditions that followed. In almost every other year, the negative effects were significant at the 1% level for every education group except those with a graduate degree. In general, the year dummy variable was not significant or was less significant for those with a graduate degree, suggesting that the downturn had less impact on median wages for this group. There were larger-year effects for those with less education and in general they became more negative as we moved through the estimation period. For example, the coefficient value for 2012 was −.087 for individuals who have not completed high school and −.028 for individuals holding a graduate degree (both results were statistically significant at the 1% level). In 2006, the coefficient value for individuals who had not completed high school was −.027 and for people with graduate degrees it was −.010. The fall in real wages over time, even controlling for local labor market conditions and other factors, is a matter of socioeconomic concern as it has primarily affected the less educated.

Robustness Checks

We examined the robustness of our results in several ways. First, we estimated several models that varied in how the cost of living was incorporated into the estimation procedure, including using nominal wages and including all price variables as independent variables. The coefficient estimates on the other independent variables were remarkably similar in all of these alternative models.

Second, we estimated a random effects model, which was soundly rejected in favor of fixed effects, for both the base model and the full model. Interestingly, the random effects model was the only estimation that resulted in a positive and significant effect of the community education attainment variable on the median wage for people with a bachelor’s degree and people with a graduate degree.

Third, to examine the importance of city size, we divided our data set into MSAs greater than 500,000 people and those with fewer. The parameter estimates between these and our estimates on the full data set were very similar. The standard errors increased in the MSA subsets as one would expect, and this reduced the statistical significance of the variables and in a couple of instances made the coefficient estimates insignificantly different from zero at the 5% level. One exception to these overall results was that the housing price variable for the population with a graduate degree in the larger MSAs became significantly positive. Furthermore, the coefficient on the community education attainment variable was not statistically significant for any of the educational attainment cohorts, even when we restricted our sample to the larger MSAs.

As mentioned in the Results section, we also examined the sensitivity of our results to our choice of the labor market tightness variable. Our measure was the share of the adult population that was employed. We found no qualitative differences in the parameter estimates when either a lagged version or a lead version of our variable was used.

Finally, we estimated a hierarchical linear model type specification that used the two levels of the metro area and the state (Fontes et al., 2010). Our state-level variable was the tax rate variable. This specification differs slightly from a fully specified hierarchical linear model in that our error structure allowed for nonparametric correlation within each metro area, rather than metro- and state-level random effects. Again, this specification did not produce parameter estimates that were qualitatively different from the simpler specifications. We found no evidence to support the existence of positive spillovers from an educated workforce to the wages of less educated workers.

Discussion

These results suggest two overall scenarios, defined by educational group—those who are relatively educated and those who are not. For those with a graduate degree, local factors do not seem to matter much, and instead wages in this cohort are probably determined more by national and international opportunities. Only the share of young people with graduate degrees was significant, indicating that a more youthful and less experienced community will have lower wages once education is accounted for. Interestingly, even the cost of housing did not matter for this group; that is, people with a graduate degree did not need to be compensated for higher urban housing costs, suggesting that living in an expensive community includes an amenity value to them that almost completely offsets higher housing costs. For those with a bachelor’s degree, in addition to the share of young workers, the median wage was affected by the local labor market—the tighter the market for those with bachelor’s degrees, the higher the wage. For this group, higher housing costs must be partially compensated with higher wages, indicating that living in a higher-cost area has benefits of its own for these workers as well.

At the other end of educational attainment, the wages of workers who did not graduate from high school appear quite sensitive to local factors. Their wages are lower the younger the population, influenced by labor market tightness, and need to be paid more than those with more education to compensate for housing costs. In addition, the wages in this group benefit from a mandated minimum wage. The minimum wage was not statistically significant for any other group, although it was almost significant for those individuals with just a high school diploma. The wages for people who did not complete high school are also higher the larger the share of durable manufacturing. The wages of those that graduated with a high school degree and those with some college or an associate’s degree are affected similarly.

Hence clear differences between those with a bachelor’s degree and graduate degree, and those with less education, are suggested by these results. Furthermore, as indicated throughout this study, we do not find statistically significant evidence that the presence of a relatively educated group in the aggregate has any effect on the wages of others.

Finally, anecdotal evidence indicates that real wages generally have been on the decline over time (Figure 1, Irwin, 2015). Our results are consistent with this observation, with less educated workers losing more than educated workers. The unexplained decline in the median wage is a topic for future research.

One implication of these results is that they do not point to any policy shortcuts to raising workers’ wages. On the other hand, importantly, metro areas where employment opportunities are better have been able to sustain higher wages than metro areas where employment opportunities are more limited. Hence, one possible way to increase wages would be to increase the demand for labor with the relevant educational skills (see also Groen, 2011; Nolan, Morrison, Kumar, Galloway, & Cordes, 2011) as well as to find ways to teach the skills needed by businesses (see, e.g., Holzer, 2012). From an individual’s point of view, moving to an area with more demand for his or her particular skill level might be necessary. Moretti (2013) suggests a national program of vouchers to help low-income people around the country move to places that better match their skills. From a city’s point of view, this means that a strategy targeting diversification might make more sense than one of specialization. A diversified economy would more likely accommodate a range of education and skill levels. In our results, adding durables manufacturing jobs, for example, increased wages for people with a high school degree and some college or an associate’s degree. This approach is quite different than trying to replicate the Silicon Valley phenomenon.

Conclusion

In contrast to anecdotal evidence using simple correlations and some more formal studies that have been done to date, using a representative individual we were not able to find significant human capital spillovers to wages due to having an educated community. These results were consistent across several different estimations and robustness checks. Standard variables from the literature performed mostly as expected, especially when all workers were estimated together, and gave us added insight as a result of using the labor market segments defined by educational attainment. Looking at the aggregate population with all education groups together, we find, as do many others, that the presence of a larger share of educated workers has a positive effect on wages. However, explaining wage levels across metro areas reveals new insights when we disaggregate the analysis by educational cohort and correct appropriately for cost of living differences. First, local market conditions matter especially to less educated workers. Second, and most important, the wages of each educational cohort are determined primarily by their own characteristics and are not significantly influenced by the educational level of the community at large.

Footnotes

Appendix A

Appendix C

Results for Dummy Variables: Base and Full Model.

	All education levels	Did not graduate from high school	High school graduate or GED	Some college or associate’s degree	Bachelor’s only	Graduate degree
Base model
Dummy 2006	−0.0102***	−0.0327***	−0.0122***	−0.00650*	−0.0164***	−0.00746
	(.00232)	(.00984)	(.00422)	(.00355)	(.00471)	(.00556)
Dummy 2007	−0.00375	−0.0334***	−0.00723*	−0.00472	−0.00522	−0.00698
	(.00275)	(.01036)	(.00427)	(.00406)	(.00491)	(.00820)
Dummy 2008	−0.0169***	−0.0270**	−0.0240***	−0.0285***	−0.0343***	−0.0150***
	(.00301)	(.01116)	(.00480)	(.00406)	(.00500)	(.00563)
Dummy 2009	−0.0532***	−0.1112***	−0.0732***	−0.0697***	−0.0348***	−0.00476
	(.00317)	(.01092)	(.00467)	(.00432)	(.00535)	(.00553)
Dummy 2010	−0.0436***	−0.1156***	−0.0654***	−0.0614***	−0.0355***	−0.00143
	(.00334)	(.01188)	(.00518)	(.00478)	(.00601)	(.00547)
Dummy 2011	−0.0625***	−0.1291***	−0.0870***	−0.0812***	−0.0526***	−0.0191***
	(.00363)	(.01171)	(.00552)	(.00482)	(.00554)	(.00669)
Dummy 2012	−0.0669***	−0.1139***	−0.0859***	−0.0987***	−0.0574***	−0.0338***
	(.00384)	(.01171)	(.00527)	(.00502)	(.00537)	(.00598)
Full model
Dummy 2006	−0.0106***	−0.0266**	−0.0117***	−0.0082***	−0.0179***	−0.01048*
	(.00248)	(.01030)	(.00426)	(.00372)	(.00474)	(.00592)
Dummy 2007	−0.0041	−0.02773**	−0.00595	−0.00480	−0.00668	−0.00900
	(.00298)	(.01107)	(.00454)	(.00449)	(.00535)	(.00730)
Dummy 2008	−0.0238***	−0.0268***	−0.0282***	−0.0317***	−0.0378***	−0.01635**
	(.00353)	(.01228)	(.00531)	(.00460)	(.00555)	(.00656)
Dummy 2009	−0.0437***	−0.0868***	−0.0597***	−0.0544***	−0.0316***	−0.00133
	(.00462)	(.01388)	(.00635)	(.00590)	(.00710)	(.00791)
Dummy 2010	−0.0294***	−0.0887***	−0.0457***	−0.0404***	−0.0311***	0.00303
	(.00515)	(.01535)	(.00750)	(.00673)	(.00805)	(.00858)
Dummy 2011	−0.0473***	−0.1020***	−0.0676***	−0.0586***	−0.0478***	−0.01486*
	(.00519)	(.01417)	(.00735)	(.00644)	(.00709)	(.00860)
Dummy 2012	−0.0522***	−0.0870***	−0.0654***	−0.0771***	−0.0538***	−0.0284***
	(.00516)	(.01390)	(.00690)	(.00646)	(.00683)	(.00736)

Note. ***, **, and * denote coefficients significant at the 1%, 5%, and 10% levels, respectively.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

Notes

Author Biographies

Penelope B. Prime is a clinical professor in the Institute of International Business, J. Mack Robinson College of Business, Georgia State University. Her research focuses on applying trade and growth frameworks to both regional and international issues.

Donald Grimes is a senior research area specialist at the Institute for Research on Labor, Employment, and the Economy, University of Michigan. His research focuses on the forecasting and analysis of local labor market activity.

Mary Beth Walker is dean of the Andrew Young School of Policy Studies at Georgia State University. Her research interests are in policy analysis and program evaluation. She has worked on applications in a variety of fields, including taxation issues, education, and health care utilization.

References

Acemoglu

(1996). A microfoundation for social increasing returns in human capital accumulation. Quarterly Journal of Economics, 111, 779-804. doi:10.2307/2946672

Acemoglu

Angrist

(2000). How large are human-capital externalities? Evidence from compulsory-schooling laws. In Bernanke

B. S.

Rogoff

(Eds.), NBER macroeconomics annual 2000 (Vol. 15, pp. 9-59). Cambridge: MIT Press.

Angrist

J. D.

Pischke

J.-S.

(2009). Mostly harmless econometrics. Princeton, NJ: Princeton University Press.

Aten

B. H.

Figueroa

E. B.

Vengelen

B. M.

(2015). Real personal income and regional price parities for states and metropolitan areas, 2008-2012. Retrieved from http://www.bea.gov/scb/pdf/2015/07%20July/0715_real_personal_income_and_regional_price_parities.pdf

Bratti

Leombruni

(2014). Local human capital externalities and wages at the firm level: The case of Italian manufacturing. Economics of Education Review, 41, 161-175.

Ciccone

Peri

(2006). Identifying human capital externalities: Theory with applications. Review of Economic Studies, 73, 381-412.

Combes

P.-P.

Duranton

Gobillon

(2008). Spatial wage disparities: Sorting matters! Journal of Urban Economics, 63, 723-742.

Díaz

A. M.

(2013). The employment advantages of skilled urban municipalities in Colombia. Ensayos Sobre Política Económica, 31, 315-366.

Dimou

(2012). Do local amenities affect urban wage premium? Some evidence from the Greek case. Sustainable Cities and Society, 2, 63-69.

10.

Dumond

J. M.

Hirsch

B. T.

MacPherson

D. A.

(1999). Wage differentials across labor markets and workers: Does cost of living matter? Economic Inquiry, 37, 577-598.

11.

Echeverri-Carroll

E. L.

Ayala

S. G.

(2011). Urban wages: Does city size matter. Urban Studies, 48, 253-271. doi:10.1177/0042098010369393

12.

Fingleton

(2003). Increasing returns: Evidence from local wage rates in Great Britain. Oxford Economic Papers, 55, 716-739.

13.

Florida

R. L.

(2012). The rise of the creative class: Revisited. New York, NY: Basic books.

14.

Fontes

G. G.

Simoes

R. F.

Hermeto Camilo de Oliveira

A. M.

(2010). Urban attributes and wage disparities in Brazil: A multilevel hierarchical model. Regional Studies, 44, 595-607.

15.

Gabriel

Rosenthal

(1999). Location and the effect of demographic traits on earnings. Regional Science and Urban Economics, 29, 445-461.

16.

Glaeser

E. L.

Maré

D. C.

(2001). Cities and skills. Journal of Labor Economics, 19, 316-342.

17.

Groen

J. A.

(2011). Building knowledge stocks locally: Consequences of geographic mobility for the effectiveness of state higher education policies. Economic Development Quarterly, 25, 316-329.

18.

Hanson

G. H.

(2000). Firms, workers, and the geographic concentration of economic activity. In Clark

G. L.

Feldman

M. P.

Gertler

M. S.

(Eds.), Handbook of economic geography (pp. 477-494). New York, NY: Oxford University Press.

19.

Heuermann

Halfdanarson

Suedekum

(2010). Human capital externalities and the urban wage premium: Two literatures and their interrelations. Urban Studies, 47, 749-767. doi:10.1177/0042098009352363

20.

Holzer

H. J.

(2012). Good workers for good jobs: Improving education and workforce systems in the U.S. IZA Journal of Labor Policy, 1(5). doi:10.1186/2193-9004-1-5. Retrieved from http://izajolp.com/content/1/1/5

21.

Irwin

(2015, April 23). Why less educated workers are losing ground on wages. New York Times, p. A3.

22.

Krupka

D. J.

Noonan

D. S.

(2013). City air and city markets: Worker productivity gains across city size. International Regional Science Review, 36, 183-206.

23.

Moretti

(2004a). Estimating the social return to higher education: Evidence from longitudinal and repeated cross-sectional data. Journal of Econometrics, 121, 175-212.

24.

Moretti

(2004b). Human capital externalities in cities. Handbook of Regional and Urban Economics, 4, 2243-2291.

25.

Moretti

(2013). The new geography of jobs. Boston, MA: Mariner Books, Houghton Mifflin Harcourt.

26.

Nolan

Morrison

Kumar

Galloway

Cordes

(2011). Linking industry and occupation clusters in regional economic development. Economic Development Quarterly, 25(1), 26-35.

27.

Porter

(2014, September 11). Equation is simple: Education = income. New York Times, pp. B1, B5.

28.

Rauch

J. E.

(1993). Productivity gains from geographic concentration of human capital: Evidence from cities. Journal of Urban Economics, 34, 380-400.

29.

Richter

F. G. C.

Nelson

(2014, February 10). The prospects of non-college bound workers in the fourth district. Retrieved from https://www.clevelandfed.org/en/Newsroom%20and%20Events/Publications/A%20Look%20Behind%20The%20Numbers.aspx

30.

Riefler

R. F.

(2007). State patterns of occupational earnings: Implications for long-term growth. Economic Development Quarterly, 21, 34-48. doi:10.1177/0891242406294927

31.

Roback

(1982). Wages, rents, and the quality of life. Journal of Political Economy, 90, 1257-1278.

32.

Rodriguez-Pose

Vilata-Bufi

(2005). Education, migration, and job satisfaction: The regional returns of human capital in the EU. Journal of Economic Geography, 5, 545-566.

33.

Rosenthal

S. S.

Strange

W. C.

(2008). The attenuation of human capital spillovers. Journal of Urban Economics, 64, 373-389.

34.

Ruggles

Alexander

J. T.

Genadek

Goeken

Schroeder

M. B.

Sobek

(2010). Integrated Public Use Microdata Series: Version 5.0 [Machine-readable database] (IPUMS). Minneapolis: University of Minnesota.

35.

Schumacher

F. I.

Dias

Tebaldi

(2014). Two tales on human capital and knowledge spillovers: The case of US and Brazil. Applied Economics, 46, 2733-2743.

36.

Wheaton

W. C.

Lewis

M. J.

(2002). Urban wages and labor market agglomeration. Journal of Urban Economics, 51, 542-562.

37.

Winters

J. V.

(2009). Wages and prices: Are workers fully compensated for cost of living differences? Regional Science and Urban Economics, 39, 632-643.

38.

U.S. Census Bureau. (n.d.). American Consumer Survey, Table S2001; Table B20004; Table B23006 Table B15001. Retrieved from http://factfinder2.census.gov/faces/nav/jsf/pages/index.xhtml

39.

Yankow

(2006). Why do cities pay more? An empirical examination of some competing theories of the urban wage premium. Journal of Urban Economics, 60, 139-161.

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.00 MB

0.10 MB