Differences in Employment Outcomes for College Town Stayers and Leavers

Areas surrounding colleges and universities are often able to build their local stock of human capital by retaining recent graduates in the area after they finish their education. This paper classifies 41 U.S. metropolitan areas as "college towns" and investigates differences in employment outcomes between college graduates who stay in the college town where they obtained their degree and college graduates who leave after completing their degree. We find that college town stayers experience less favorable employment outcomes along multiple dimensions. On average, stayers earn lower annual and hourly wages and work in less educated occupations.


Introduction
Researchers have consistently shown that human capital is an important ingredient for regional economic prosperity. Individuals who invest in their own human capital benefit themselves through higher productivity and wages (Psacharopoulos and Patrinos 2004;Abel and Gabe forthcoming). However, there is growing evidence that skilled workers also benefit other workers in the local labor market by making them more productive as well (Rauch 1993;Moretti 2004). In particular, researchers have suggested that college graduates play a key role. 1 Areas with a high percentage of adults with college degrees have higher wages and higher employment rates for both college graduates and non-graduates (Moretti 2004;Winters 2010). Shapiro (2006) also suggests that college graduates increase the quality of life in an area and make it a more desirable place to live. Because of these external benefits of human capital, highly educated areas also experience faster employment and population growth (Glaeser, Scheinkman and Shleifer 1995;Simon 1998;Simon and Nardinelli 2002).
Because of the many benefits of human capital, policymakers are often very interested in finding ways to attract and retain highly educated workers (Hoare and Corver 2010). Local colleges and universities often play an important role in building an area's stock of college educated workers for at least two reasons. First, they help educate local residents and keep them close to home for college (Hickman 2009). Second, colleges and universities also bring in students from outside the local area. While many of these recent student in-migrants will leave once they complete their education, some of them will stay and help build the area's local stock of human capital (Blackwell, Cobb, and Weinberg 2002;Huffman and Quigley 2002;Groen 2004;Groen and White 2004; Oosterbeek and Webbink forthcoming).
While research has shown that some recent graduates stay in the area where they obtained their education, there has been very little research examining how staying in the area affects individual employment outcomes for recent graduates. Several researchers examine the effects of migration on income and employment for certain groups (e.g., Axelsson and Westerlund 1998;Yankow 2003;Boheim and Taylor 2007;Anil, Sjoquist, and Wallace 2010;Blackburn 2010a,b), but none of these explicitly examine the effects of staying in a college town for recent graduates. Many recent college graduates might be better able to find gainful employment by relocating to a different labor market and of course many graduates do leave the area where they completed their education. Recent graduates who stay in the area where they earned their degree may pay a price for doing so, and this price is likely to be especially large in smaller labor markets that produce large numbers of college graduates, i.e., "college towns." This paper seeks to fill an important gap in the research literature by investigating differences in employment outcomes between recent college graduates who stay in the college town where they obtained their degree and those who leave after completing their degree. Using decennial census data, we classify 41 U.S. metropolitan areas as college towns and examine differences in the probability of employment, annual wage income, hourly wages, annual hours worked, and a measure of the education level of a worker's occupation. The results suggest that, on average, college town stayers earn lower annual and hourly wages and work in less educated occupations. Thus, it appears that college town stayers generally experience worse employment outcomes than those who leave. These results continue to hold when we employ a difference in 3 differences approach that treats recent graduates from large metropolitan areas as a control group and when we control for differences in the cost of living across labor markets.

Conceptual Framework
The conceptual framework for this paper proceeds by considering the migration decision of a recent college graduate. We follow a large literature and view migration as a human capital investment (Sjaastad 1962). In this framework individuals maximize their own expected utility and choose to live in the area that offers the highest expected utility. Thus, upon completing her degree, a college graduate will move to a new area if it gives her greater expected utility than the place where she earned her degree. The expected utility that an area offers will depend on a number of factors including wages and employment opportunities, the local cost of living, consumer amenities and quality of life, and the tastes and preferences of the individual making the migration decision (Clark and Hunter 1992;Franklin 2003;Arntz 2010). Chen and Rosenthal (2008) and Plane and Jurjevich (2009) suggest that young college graduates are often most strongly attracted to large metropolitan areas with high wages and a strong labor market. Whisler et al. (2008) argue that cultural and recreational amenities also play an important role in retaining college graduates in an area. Furthermore, Berry and Glaeser (2005) and Waldorf (2009) suggest that educated individuals are attracted to areas that already have large numbers of educated people.
While some areas are certainly better than others at attracting and retaining recent college graduates, the optimal location will not be the same for all individuals (Plane and 4 Heins 2003). Individuals have different skill sets and different preferences for various locations.
Furthermore, some graduates may develop human capital that is specific to the location where they earned their college degree (DaVanzo 1983;Krupka 2009;Winters forthcoming). For example, they may develop networks with professors and local employers that enhance their employment prospects in the area. Alternatively, recent graduates may have built strong friendships and developed a taste for local amenities that help tie them to the area where they attended college. These preferences for the place where they attended college may even incline some recent graduates to accept lower paying jobs to stay in the area that they have grown to love.

Empirical Approach and Data
This paper examines employment differences between recent college graduates who stay in the college town where they earned their degree and those who leave a college town after their education is complete. We focus on recent college graduates who earned their degrees in relatively small college towns because it is these graduates who are likely to have the greatest difficulty finding a good job in the area where they completed their degree. Large metropolitan areas like New York and Los Angeles have thicker labor markets and generally offer better employment prospects for recent graduates and are better able to retain recent college graduates (Whisler et al. 2008). This paper uses microdata from the 2000 Census available from the IPUMS (Ruggles et al. 2008) to classify 41 U.S. metropolitan areas as college towns. We define a metropolitan area as a college town based on the percentage of adults in the area who are enrolled in college. More specifically, an area is considered a college town if the share of the population age 18 and older enrolled in college is more than one standard deviation greater than 5 the mean across all metropolitan areas. For the year 2000, the mean share of adults enrolled in college across metropolitan areas was 0.092, and the standard deviation across metropolitan areas was 0.050. Thus an area is considered a college town if its share enrolled in college exceeds 0.142. The 41 metropolitan areas meeting this criterion and included in this study are listed in Table 1.
Admittedly, this definition of college towns is a fairly restrictive one, and a case could be made that many areas not included should be rightfully considered as college towns. The present definition is intended to limit the analysis to areas where colleges and universities have a relatively large influence. Furthermore, the 41 areas meeting this criterion are also relatively small metropolitan areas having a mean population of 203,919. Again, it is these smaller areas in which recent graduates may suffer the most from staying in the area where they went to college. Ideally, we might also like to include non-metropolitan college towns, but unfortunately the Census public use microdata do not allow separate identification of geographic areas with populations less than 100,000. Consequently, the lowest level of identifiable geography, the PUMA, lumps very small areas together with other nearby areas that often span large land masses. This makes it harder to identify whether recent college graduates remain in nonmetropolitan college towns because they might still be in the same PUMA but have moved to a new location a considerable distance from where they attended college.
These geographic identification problems also affect a few of the metropolitan areas included in this study, though to a lesser extent than would be true for non-metropolitan areas.
6 After defining the 41 college town metropolitan areas examined in this study, we next wish to consider differences in employment outcomes between recent graduates who stay in these college towns and those who leave after completing their degree. We define recent college graduates as persons between the ages of 23 and 27 with at least a four-year college degree. We do not observe when an individual completed her degree, so our definition of recent is based on age. College graduates ages 23-27 at the time of the census were ages 18-22 five years before the census and were likely to have been enrolled in college but not yet have earned a bachelor's degree. The main results presented below are qualitatively robust to altering the age range to include persons ages 28 and 29 at the time of the census or exclude persons age 23 at the time of the census. Our initial analysis also restricts the sample to recent college graduates who resided in one of the 41 college towns five years prior to the census. Unfortunately, there may be some misreporting of previous residence for recent college graduates. Some recent graduates living in a college town five years prior may instead report their residence five years ago as their parents' residence. It is unclear how much or in which direction the resulting sampling and measurement error would affect the results.
The sample also excludes two groups of recent graduates. First, we exclude persons who are still enrolled in higher education in 2000. Since these people are still in school it does not make sense to compare their employment outcomes to those who have completed their schooling. We also exclude all persons born outside the U.S. since many of these only come to the U.S. for an education and leave after completing their degree. As a practical matter, the results below are qualitatively robust to including recent college graduates who are currently enrolled and foreign born graduates who are still in the U.S. in 2000, but the results presented do 7 not include these groups of recent graduates. After restricting the sample along these dimensions we are left with 11,237 individuals. We next define the binary variable, ‫,ݎ݁ݕܽݐܵ‬ which is equal to one if the recent graduate is still in the college town where she resided five years prior and zero otherwise.
We then use linear regression to estimate differences between stayers and leavers for several employment outcomes controlling for differences in individual characteristics, ܺ, i.e, we estimate regressions of the following form: The individual characteristics include separate dummy variables for the highest degree earned, age, whether the individual is female, Black, Hispanic, Asian, Other Nonwhite, married, or has children. One unfortunate limitation of the current paper is that there may be unobservable differences between stayers and leavers that also separately affect employment outcomes. 2 For example, leavers may be more ambitious and career driven than stayers. If drive and ambition also improve employment prospects, then failing to observe these will induce a negative correlation between employment outcomes and staying. Several previous studies have attempted to account for whether migrants are generally a self-selected group (e.g., Axelsson and Westerlund 1998;Yankow 2003;Boheim and Taylor 2007;Blackburn 2010). This usually involves trying to identify a variable that affects the probability of migration but does not separately affect employment outcomes such as proximity to family or housing tenure in the previous period. The census data used in this study do not offer information on either of these and we are thus unable to examine whether migration is endogenous for recent college graduates 8 who earned their degree in a college town. 3 Instead, we examine the sensitivity of the results to employing a difference in differences (DD) technique that treats recent graduates from large metropolitan areas as a control group. More details are discussed later. 4 This paper considers five employment outcomes: 1) whether an individual is employed at the time of the census, 2) the log of annual wage income earned in the previous year, 3) the log average hourly wage in the previous year computed as annual wages divided by the number of hours worked in the previous year, 4) the log of hours worked in the previous year, and 5) the percentage of workers in the same occupation with a bachelor's degree or higher for employed recent graduates. This fifth outcome is intended to capture whether stayers are more likely to take jobs for which they are overeducated. The hypothesis of this paper is that college town stayers may experience worse employment outcomes than those who leave along a number of dimensions. Summary statistics for the dependent and independent variables for the full sample are reported in Table 2. A few things are worth mentioning. First, 92.3 percent of the sample is employed, reflecting the high rates of labor force participation and low rates of unemployment for young college graduates as well as the relatively strong labor market in 2000. Second, only 19.4 percent of recent graduates who lived in a college town five years prior are still in the college town where they completed their degree. The large majority, therefore, have either moved back to a previous location or moved on to a new location. Table 3 presents the mean values of the dependent and independent variables separately for college town stayers and leavers and also reports the differences in the means and whether 9 the differences are statistically significantly different from zero. This provides a first look at a number of differences between those who stay in a college town after completing a degree and those who leave. According to these raw differences, stayers perform significantly worse than leavers in all five of the employment outcomes that we study. Stayers have lower employment rates, earn lower annual and hourly wages, work fewer annual hours, and work in less educated occupations. However, the table also shows that there are important differences between stayers and leavers in individual characteristics. Stayers are significantly less likely to hold a master's or professional degree, and there are some differences in age between the two groups. Females are significantly more likely than males to stay in the college town where they completed their degree. Blacks are significantly more likely to stay than Whites, but Asians are significantly less likely to stay than Whites. Stayers are also significantly more likely to be married and more likely to have children. Being married and having children, therefore, appear to increase the costs of leaving and tie recent graduates to the area where they completed their degree. Because of these significant differences in individual characteristics, we cannot accurately assess the effect of staying on employment outcomes simply by taking differences in means. Instead, we need to use multivariate regression techniques to control for the individual characteristics. The next section presents the results of regressing the five employment outcomes on the variable ‫ݎ݁ݕܽݐܵ‬ and the individual characteristics.

Differences between College Town Stayers and Leavers
Regression results for differences between college town stayers and leavers are presented in Table 4. The first column reports results for the probability of employment estimated via a linear probability model. The results are qualitatively robust to estimating probit and logit models. The estimated coefficient for ‫ݎ݁ݕܽݐܵ‬ is small and not statistically different from zero.
In other words, the results suggest that there is no difference in the probability of employment for stayers and leavers. Recent college graduates from college towns have an equal probability of finding employment regardless of whether they stay in the area where they earned their degree or move to a different area. Note that this is in contrast to what was observed in Table 3 The second column of Table 4 reports results for the log of annual income from wages and salaries. First note that this regression and the following regressions have fewer observations than in the first column because some individuals did not work in the previous calendar year. Taking logs forces us to drop those with zero wages and hours because the log of zero is not defined. Turning to the results, Stayer has a statistically significant negative effect with a coefficient of -0.267. This confirms that recent graduates who stay in the college town where they completed their degree on average accept much lower paying jobs than those who leave. Converting logs to percentages, the coefficient estimate suggests that college town stayers earn roughly 30 percent lower wages than those who leave. This is quite a price to pay for staying and helps explain why so many recent Log annual wages can essentially be broken down into log hourly wages and log annual hours worked. The third and fourth columns of Table 4 report the results for these separate outcomes. According to the results, college town stayers both earn significantly lower hourly wages and work significantly fewer annual hours with coefficients of -0.209 and -0.058, respectively. Notice that these two coefficients add up to the coefficient in the second column by definition. While both are significantly negative, the magnitude is much greater for log hourly wages and suggests that lower hourly wages account for 78 percent of the lower annual wages of stayers. The results for advanced education are again somewhat interesting.
Master's degree holders have significantly higher hourly wages but significantly lower annual

Difference in Differences Approach with Recent Graduates from Large Metro Areas
The results thus far suggest that recent college graduates who stay in the college town where they completed their degree experience significantly worse employment outcomes than those who leave college towns. Furthermore, the magnitudes of these effects are often quite large. The difference in annual wages between stayers and leavers is roughly 30 percent, a very large effect. A major concern, however, is that stayers and leavers are self-selected groups and might differ along unobserved dimensions. For example, leavers might have more drive and ambition that also makes them more productive workers. Unfortunately, the census microdata do not provide much information that would be useful in examining the endogeneity of migration. However, we can offer some additional insight about the effects on recent graduates of staying in a college town by employing a difference in differences (DD) regression technique that treats recent graduates from large metropolitan areas as a control group.
The DD approach for each employment outcome variable, Y, involves estimating the following regression for recent graduates who either resided in a college town or a large metropolitan area five years prior to the Census: (2) where ‫ݎ݁ݕܽݐܵ‬ is an indicator variable equal to one if a recent graduate still resides in the same area as they did five years prior and zero otherwise, and ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ is an indicator variable equal to one if a recent graduate resided in a college town five years prior and equal to zero if the recent graduate was in a large metro area five years prior. For the results presented, we define large metropolitan areas as those with a year 2000 population of at least 5,000,000, but the results are qualitatively robust to defining large metros as all those with at least 1,000,000 people. The DD regressions also include a variable for the interaction of ‫ݎ݁ݕܽݐܵ‬ and ‫,݊ݓ݈݈ܶ݁݃݁ܥ‬ which measures the difference in stayer-leaver differences between college towns and large metro areas. This interaction term is now the primary variable of interest. The DD procedure is an attempt to account for selection on unobservables by assuming that recent graduates select into migration in similar ways in college towns and large metro areas. If this assumption is violated, the DD procedure may also lead to invalid estimates.
Before proceeding to the DD results, it is first informative to simply compare raw differences in means between recent graduate stayers and leavers from large metropolitan areas.
These comparisons are reported in Table 5. As seen, the raw differences between stayers and leavers from large metros are small and insignificant for employment probabilities, annual wages, and occupational education levels. Large metro stayers do earn significantly higher hourly wages but work significantly fewer annual hours. Importantly, there are also some significant differences in individual characteristics. Also, though not explicitly shown, there is a considerable difference between college towns and large metro areas in the percentage of recent graduates who stay in the area after completing their degree. While only 19.4 percent of recent graduates from college towns are classified as stayers, 73.8 percent of recent graduates from large metros are considered stayers.
The DD regression results for the five employment outcomes are reported in Table 6. For the employment regression in the first column, ‫ݎ݁ݕܽݐܵ‬ has a small and insignificant coefficient, while ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ has a significantly positive coefficient of 0.013. The interaction of ‫ݎ݁ݕܽݐܵ‬ and ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ has a small and insignificant coefficient, reaffirming the results in Table 4 that staying in a college town after graduating does not affect the probability of employment.

16
The results for log annual wages are presented in the second column of Table 6; ‫ݎ݁ݕܽݐܵ‬ has a statistically significant coefficient of 0.024, suggesting that recent graduates who stay in the large metropolitan area where they completed their degree actually earn slightly higher annual wages than recent graduates from large metros who leave. ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ has a significant coefficient of -0.024, suggesting that college town leavers on average earn slightly lower annual wages than big city leavers. Again, the interaction of ‫ݎ݁ݕܽݐܵ‬ and ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ is the main explanatory variable of interest and has a significant negative effect on log wages with a coefficient of -0.292. This reaffirms that staying in a college town results in significantly lower annual wages for recent graduates even when we treat recent graduates from large metro areas as a control group.
The third and fourth columns of Table 6 report the results for log hourly wages and log annual hours, respectively. ‫ݎ݁ݕܽݐܵ‬ has a significantly positive effect on log hourly wages but a significantly negative effect on log annual hours with coefficients of 0.063 and -0.039, respectively. Thus, large metro stayers earn higher hourly wages but work fewer hours than large metro leavers. ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ has a significantly negative coefficient for log hourly wages of -0.020 but an insignificant effect on log annual hours. The interaction term has a significantly negative effect on hourly wages with a coefficient of -0.270 but an insignificant effect on log annual hours. These results suggest that nearly all of the adverse effect of staying in a college town on annual wages results from lower hourly wages.
The fifth column of Table 6 presents results for the percent of workers in an occupation with a bachelor's degree or higher. ‫ݎ݁ݕܽݐܵ‬ has a significantly positive effect with a coefficient of 0.014 suggesting that large metro stayers work in slightly more educated occupations than large metro leavers. ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ also has a significantly positive coefficient of 0.015, suggesting that college town leavers also work in slightly more educated occupations than large metro stayers. The interaction term is negative and significant with a coefficient of -0.068, reaffirming that staying in a college town increases the likelihood of recent graduates taking jobs in occupations that do not require a four-year college degree.

Difference in Differences Approach Controlling for the Cost of Living
The DD results in the previous sub-section show that staying in a college town worsens employment outcomes for recent graduates along multiple dimensions, even when we use recent graduates from large metropolitan areas as a control group.
However, a legitimate concern is that the effects on individual well-being might be overstated by the effects on nominal wages if college towns have lower costs of living than the destination areas of college town leavers. In other words, it might not make sense to compare the nominal wages of college town stayers to the nominal wages of college town leavers, large metro stayers, and large metro leavers if these latter groups live in more expensive areas. Assessing individual well-being requires comparing "real wages", that is wages adjusted for the local cost of living. Therefore, we next re-estimate the DD regressions in Table 6 while adding a control for the cost of living. We control for the cost of living by adding to the regressions a measure of average logarithmic differences in housing rental payments for each individual's current location computed using microdata from the 2000 Census. This measure, referred to as ‫݊ܽ݁ܯ‬ ‫݈݃‬ ‫,ݐ݊݁ݎ‬ is computed by regressing the log of gross rents, ܴ, for each rental housing unit ݅ in area ݆ on a vector of housing characteristics, ‫,ܪ‬ and a vector of area fixed-effects 5 , ߨ: (3) ln ܴ = ‫ܪ‬ Γ + ߨ + ‫ݑ‬ .
The housing characteristics included are dummy variables for the number of bedrooms, the total number of rooms, the age of the structure, the number of units in the building, modern plumbing, modern kitchen facilities, and lot size for single-family homes. The results for this regression are available upon request. The area fixed-effects from (3) measure quality-adjusted average logarithmic differences in rents across areas and are used to control for differences in the cost of living across areas. One also could use housing values instead of rents to measure cost of living differences across areas, but Winters (2009) suggests that rents are a more appropriate measure of the cost of living. Table 7 presents the results of adding ‫݊ܽ݁ܯ‬ ‫݈݃‬ ‫ݐ݊݁ݎ‬ to the DD regressions to control for the cost of living in the current location. As seen in the first column of Table 7, ‫݊ܽ݁ܯ‬ ‫݈݃‬ ‫ݐ݊݁ݎ‬ has a significant effect on the probability of employment with a coefficient of -0.025, but controlling for the cost of living does not meaningfully affect the main results for the probability of employment. ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ is again significant with a coefficient of 0.011, but ‫ݎ݁ݕܽݐܵ‬ and the interaction term are again statistically insignificant. The result for the interaction term again suggests that staying in a college town after graduating does not affect the probability of employment.
The results of controlling for the cost of living in the log annual wage DD regression are reported in the second column of Table 7. As expected, ‫݊ܽ݁ܯ‬ ‫݈݃‬ ‫ݐ݊݁ݎ‬ has a significantly positive effect on log annual wages with a coefficient of 0.389. This result suggests that a one percent increase in housing rents results in a roughly 0.4 percent increase in annual wages to compensate workers for the higher cost of living. Controlling for the cost of living also changes the effect of ‫ݎ݁ݕܽݐܵ‬ on log annual wages. ‫ݎ݁ݕܽݐܵ‬ had a significantly positive effect on log annual wages in Table 6, but after controlling for the cost of living the effect is significantly negative with a coefficient of -0.018. This suggests that large metro stayers earn slightly lower annual wages than large metro leavers after controlling for the higher cost of living in large metro areas. Controlling for the cost of living, prior residence in a college town has a small and insignificant effect on annual wages. The interaction of ‫ݎ݁ݕܽݐܵ‬ and ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ again has a significantly negative effect on log annual wages, but the coefficient of -0.203 is somewhat smaller than the corresponding estimate in Table 6.
The implication is that after controlling for the cost of living, staying in a college town reduces annual wages for recent graduates by roughly 22.5 percent.
The third and fourth columns of Table 7 present the DD results for log hourly wages and log annual hours worked controlling for the cost of living. ‫݊ܽ݁ܯ‬ ‫݈݃‬ ‫ݐ݊݁ݎ‬ has a significantly positive effect on log hourly wages with a coefficient of 0.416 but an insignificant effect on log annual hours, suggesting that for recent graduates the cost of living affects hourly wages but not hours worked. ‫ݎ݁ݕܽݐܵ‬ again has a significantly positive effect on log hourly wages and a significantly negative effect on log annual hours with coefficients of 0.018 and -0.036, respectively. ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ now has an insignificant effect on both log hourly wages and log annual hours. The interaction of ‫ݎ݁ݕܽݐܵ‬ and ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ has a statistically negative effect on log hourly wages with a coefficient of -0.175 and an insignificant effect on log annual hours.
The fifth column of Table 7 presents the results for the percent of workers in an occupation with a bachelor's degree or higher controlling for the cost of living. As seen, ‫݊ܽ݁ܯ‬ ‫݈݃‬ ‫ݐ݊݁ݎ‬ has a statistically insignificant effect on the occupational education measure and controlling for the cost of living has no meaningful effect on the main results in the fifth column.
‫ݎ݁ݕܽݐܵ‬ and ‫݊ݓ݈݈ܶ݁݃݁ܥ‬ again have significantly positive effects with coefficients of 0.014 and 0.015, respectively. The interaction of the two again has a significantly negative effect with 20 a coefficient of -0.067, suggesting that staying in a college town forces recent graduates to accept jobs in less educated occupations.

Conclusion
College educated workers are an important ingredient for regional prosperity, and academics and policymakers are often interested in how areas can grow their stocks of human capital. Areas with a strong relative presence of colleges and universities are often thought to have an advantage in building their stocks of human capital because some recent graduates are likely to stay in the area where they attended college after they are done with school. Individuals choose to live and work in the location that gives them the highest possible utility, and many recent graduates wish to stay in the area where they completed their degree because they have developed location-specific human capital that increases the area's desirability. However, staying in the area where one attended college may require some sacrifice in employment opportunities, especially for persons attending college in relatively small college towns.
Using decennial census microdata for the U.S., this paper finds that recent graduates who stay in the college town where they earned their degree pay a considerable price for doing so.
They earn lower annual and hourly wages and accept jobs in less educated occupations.
Furthermore, the magnitudes of these effects are often quite large. The nominal difference in annual wages between college town stayers and leavers is roughly 30 percent. Even after controlling for the cost of living and treating recent graduates from large metro areas as a control group in an attempt to account for selection effects, staying in a college town still reduces annual wages by roughly 22.5 percent.

21
The implications for researchers and policymakers are clear. Colleges and universities can play a role in building the local stock of educated workers, but relatively small college towns often do not offer good employment prospects for recent college graduates. Some recent graduates will likely stay in the area because of attachments that they have developed, but those that stay are often forced to take lower paying jobs than they could get elsewhere Without a strong thick local market for educated labor, many recent graduates will leave the area after completing their education. Thus, while educational institutions can help build the local stock of human capital, the area needs to offer good employment opportunities for educated workers or else most recent graduates will leave once their education is complete. Waco, TX CBSA Ithaca-Cortland, NY CSA Note: Metropolitan areas are classified as "college towns" if the share of the adult population (age 18 and older) enrolled in college is more than one standard deviation greater than the mean across all metropolitan areas.    Large metropolitan areas are defined as metro areas with a year 2000 total population of 5,000,000 or greater. *Significant at 10%; **Significant at 5%; ***Significant at 1%.   Note: Heteroskedasticity robust standard errors in parentheses. *Significant at 10%; **Significant at 5%; ***Significant at 1%.