Abstract
Economic development strategies aimed at attracting highly skilled workers through investment in urban amenities are gaining momentum throughout the United States. However, most of the foundational research for the approach was tested in very large cities, both in the United States and abroad. Based on quality of place (QOP) variables suggested from previous research, confirmatory factor analysis was used to generate a set of factors for a selection of small and midsized U.S. cities, and linear regression was used to relate these factors to the presence of college-educated populations, younger college-educated populations, and adult population growth. The results indicated that some of the QOP factors associated with better human capital outcomes in prior literature focusing on larger cities were also significant predictors of better human capital outcomes in midsized cities. The relationship between these factors and development outcomes for small cities was much weaker.
Introduction
Creative class theory focuses on attracting and retaining younger, highly educated workers through amenities, diversity, and tolerance that lend to a unique sense of place. The attraction of these workers contributes to heightened economic activity by capitalizing on agglomeration and innovation. Creative class theory was largely ignored by the academic community for several years after Florida (2002, 2005) first posited it (Markusen 2006). According to Florida (2002, 2005), amenity-rich cities that are diverse and tolerant attract well-educated migrants. Spatial clustering of these migrants, along with retention of well-educated residents, promotes economic growth. These knowledge workers constitute the creative class. Creative class theory turns on two testable hypotheses: First, knowledge workers cluster in diverse, high-amenity cities, and second, cities with a higher proportion of knowledge workers experience a higher level of economic growth.
The causal link between creative class theory and economic growth is unclear. Persons with college degrees have long been recognized as higher earning and more mobile net contributors to the local economy. Footloose knowledge workers are unlikely to choose a city that cannot provide them stable employment (Scott 2006), so the classic chicken-and-egg problem persists. As the creative class is typically operationalized by level of educational attainment, the difference between creative class theory and human capital theory is subtle at best (Markusen 2006), with Florida’s contribution largely limited to the role of diversity (Rausch and Negrey 2006).
Despite untested causal links and gauzy specifications of the diversity and tolerance variables, the policy community took creative class theory seriously. Under the rubric of “placemaking,” cities invested in amenities to attract and retain young, talented workers (Hoyman and Faricy 2009). Although there is no consensus on the definition of placemaking in the urban literature (Palermo and Ponzini 2015), for the purpose here it will be used synonymously with quality of place (QOP). QOP focuses on the attributes of the physical space rather than outcomes of individuals in that space. One can see the interchangeable use of the two concepts in creative class theory, both in the scholarly and policy literature.
In 2012, Oklahoma City Mayor Mick Cornett stated, “The will to invest in our community was born of a need to attract and retain talent drawn to urban areas with a quality of place. It’s our belief that jobs follow this talent pool” (Cornett 2012). Louisville Mayor Greg Fischer announced the city’s new placemaking economic development strategy in April 2014 saying, “quality of place creates quantities of opportunities” (Fischer 2014). Many other small and midsized cities likewise identify QOP as their economic development driver, including Asheville, North Carolina; Indianapolis, Indiana; and Pittsburgh, Pennsylvania.
“Placemaking” is Michigan’s new statewide strategy. In his special message to the Michigan legislature, Governor Rick Snyder (2011) said,
Neighborhoods, cities and regions are awakening to the importance of “place” in economic development. They are planning for a future that recognizes the critical importance of quality of life to attracting talent, entrepreneurship and encouraging local businesses. Competing for success in a global marketplace means creating places where workers, entrepreneurs, and businesses want to locate, invest and expand. This work has been described as a “sense of place” or “place-based economic development” or simply “placemaking.”
Maine and Wyoming have also declared that attracting and retaining talent is the new focus of their economic development programs. Peck (2005) noted that cities have implemented policies specifically intended to enhance their ranking on the three Ts of creative class theory—technology, talent, and tolerance.
Florida’s early work was focused on very large cities, and more recent research has called for a test of creative class theory in cities of other size (Hoyman and Faricy 2009; Scott 2006). This effort uses QOP indicators from previous studies of creative class theory in large cities to examine the concentration of highly skilled workers in small—those with populations between 250,000 and 500,000—and midsized cities—those with populations greater than 500,000 and less than 2.5 million.
In their tests of creative class, human capital, and social capital theories, Hoyman and Faricy (2009) analyzed 276 metropolitan areas—which ostensibly include many small and midsized areas. However, these authors do not break these results down by the size of the metropolitan area, and differences between city types thus cannot be observed. Erickcek and McKinney (2006) found that some QOP indicators are associated with increased personal income in small cities—which they defined as those with metropolitan area populations less than 1 million. Importantly, they also note that their small city model results were statistically different from the same model run on all metropolitan areas. Results from a comparison of Scandinavian cities likewise suggest differences between large and small cities in the relationship between QOP and growth in the creative class (Andersen et al. 2010).
This effort will test the first creative class theory hypothesis, that knowledge workers cluster in diverse, high-amenity cities, in those small and midsized cities that are increasingly embracing it. Refraining from causal speculation, we ask whether amenity-rich, diverse, and tolerant cities are correlated with a higher proportion of knowledge workers. Unlike most prior research in which the metropolitan statistical area (MSA) was the unit of analysis, this effort focuses instead on the primary core county of the MSA. This choice is motivated by the fact that placemaking efforts are typically targeted to the central city and its near suburbs rather than spread across the MSA.
Rather than consider the effects of QOP indicators on development outcomes for all small/midsize cities together, we maintain two distinct groups of cities based on the arbitrary population threshold of 500,000. Analyzing small cities separately from midsized cities is motivated by the fact that midsized cities may have additional resource advantages that allow them to better carry out QOP initiatives. There is scant literature suggesting possible relationships between QOP and economic development outcomes for small cities, but it is reasonable to hypothesize that small cities will generally offer a more limited variety of entertainment and housing options, and attract firms more slowly.
Role of Amenities in Theories of Urban Economic Development
In both the popular and scholarly literatures, urban economic vitality has been linked to growth in population (and hence the tax base) in the form of new businesses and new residents. Losing middle-class, low service-demanding households to the suburbs can concentrate lower-income, high service-demanding households in the city. One suggested antidote to urban decline is territorial expansion that unites cities with near suburbs (Rusk 2013). Another is to attract firms, especially high-technology, knowledge-based firms, to the city using traditional tools of economic development. Yet another is to develop capacity for successful entrepreneurship (human capital) through higher education and targeted support. Creative class theory focuses on attracting and retaining younger, highly educated workers through placemaking or enhancing QOP. Cities may pursue one or all of these economic development strategies depending on their civic culture (Reese and Rosenfeld 2001). In fact, most cities pursue multiple economic development strategies.
The location theory of traditional urban economic development holds that firms locate where costs are minimized (Blakely and Leigh 2010), so incentives and/or abatements were logical tools for the attraction of new businesses. Residents are attracted by jobs, and make location decisions based on proximity to work, cost of housing, and other personal criteria. Although amenities were acknowledged as relevant to both firms and residents, the contribution of amenities was not well understood because they are difficult to quantify. A more contemporary view of urban economic development suggests that location decisions of certain kinds of firms and of a certain group of well-educated and/or highly skilled workers may be codeterminant, and amenities may help explain both dynamics (Arora et al. 2000). More specifically, information and service firms may be attracted to locations where educated workers with high-demand skills cluster and where physical proximity to similar firms induces innovation (Glaeser and Gottlieb 2008). Places with a high concentration of human capital grow more rapidly and are more innovative than places with relatively lower concentrations of human capital (Lucas 1988). Amenities are important to the migration decisions of highly skilled workers, but in ways we still do not fully understand (Lambiri, Biagi, and Royuela 2007).
Interest in the location decisions of highly skilled workers preceded creative class theory (see Freid 2000; Gustafson 2001; Stokols and Schumaker 1982). Some urbanists hypothesized that amenities might be gaining importance relative to employment opportunity in location decisions of skilled workers. Others noted that knowledge workers still make location decisions based on work opportunities rather than lifestyle (Darchen and Tremblay 2010; Gottlieb 1995). However, the presence of amenities can influence migrants even when they might not personally utilize them, as amenities need not be consumed to add value to a location decision (Clark and Khan 1988). A knowledge worker may choose a high-amenity city regardless of individual preferences to consume the amenities offered. However, using data from the American Time Use Survey, Van Holm (2014) found little difference between the leisure activities of the creative class and the working and service classes.
Operationalizing Amenities—QOP Indicators
Most QOP indicators fall into two broad categories: social and economic. Social indicators measure individual perceptions of well-being or social conditions that contribute to well-being. Economic indicators include measures of economic growth and consumptive behaviors. Several scholars point out that social and economic indicators have rarely been used in concert (Diener and Suh 1997; Marans 2003), whereas others have noted that social and economic indicators do not always correlate well with each other (Wish 1986).
Despite limitations in measurement, rankings of cities based on QOP indicators captured the attention of the policy community (Rogerson 1999). Rankings of dubious scholarship were seized by national and local media, and portrayed as an authoritative evaluation of who is winning (and therefore losing) the battle of city competitiveness (McCann 2004). The Places Rated Almanac (Savageau 2007) was considered one of the most defensible of the type. It ranked MSAs on the basis of climate, housing, health care, transportation, education, arts and culture, recreation, crime, and personal economics.
However, the Almanac asserted that these features were important rather than demonstrated it (Landis and Sawicki 1988). Tests of actual migration against the characteristics found only four (housing cost, crime, education, and recreation) to be significant (Herzog and Schlottmann 1986). Even though social scientists were quick to point out that QOP rankings neither explain why differences exist among cities nor link policy choices to those differences, two ideas became firmly rooted: first, that positive QOP factors contribute to economic growth and, second, that those factors, however specified, are tied to place.
Richard Florida is credited for distilling the streams of related social science research into a concise and accessible idea with an action agenda. Florida asserted that the emerging knowledge economy shifted local economic development strategy from low cost to high quality, placing a premium on innovation and new business development. Florida (2002) found that environmental quality surpassed housing cost, cost of living, commuting patterns, schools, climate, government service, and public safety as a location factor for high-technology firms. He operationalized environmental quality as a composite of air quality, water quality, and sprawl, and compared 35 regions on their high-technology industries and knowledge worker attraction. Water and air quality showed a significant relationship, whereas sprawl was mixed.
Using the same 35 cities, Florida (2000) examined the relationship between high-technology regions and high-amenity regions. He distinguished between “new economy” and “old economy” amenities. Old economy amenities include professional sports teams, fine arts (opera, symphony, etc.), and cultural amenities (museums and exhibits). New economy amenities include nightlife (bars and restaurants), outdoor recreational amenities, and diversity. Although old economy amenities did not correlate with high-technology firms or concentration of knowledge workers and new economy amenities produced mixed results, diversity showed a strong correlation with concentration of high-tech firms and workers.
Florida used the concentration of gay couples within a region as a proxy for the openness and attractiveness to alternative lifestyles. The degree of correlation between the “gay index” (Black et al. 2000) and high-tech firms and workers was higher than all the other factors. Two other diversity indicators, the “bohemian index” and the “melting pot index,” also correlated with concentration of high-tech industries (Florida and Gates 2001). The bohemian index is a measure of the proportion of persons in creative occupations in the region compared with the nation, whereas the melting pot index correlates to the proportion of foreign-born residents. When Florida regressed his bohemian index on knowledge workers (operationalized as the percentage of persons holding a bachelor’s degree or higher), controlling for other amenity measures and openness (including diversity and melting pot), he found that the bohemian index explained approximately 70% of the variation.
In addition to environmental quality and diversity identified by the Florida specifications, several authors recommended additional QOP indicators. Andrews (2001) suggested environmental threats to human health, recreational amenities, aesthetics of landscape and streetscape, availability and diversity of housing, stability of property values, transportation options, employment and education opportunities, crime rates, a sense of community, trust in government, and civic engagement. Arora et al. (2000) included culture, recreation, climate, and housing price. Trip (2007) used creativity and talent, diversity, tolerance, safety, and specific cultural and leisure amenities. Türksever and Atalik (2001) chose shopping facilities, environmental pollution, education provision, cost of living, noise levels, climate, job opportunities, travel to work, crowding, relations with neighbors, housing conditions, parks, green areas, health, leisure opportunities, sporting venues, crime rate, accessibility to public transportation, and traffic congestion.
Reilly and Renski (2008) recommended combining factors into a single index or grouping them along dimensions. Suggested dimensions included natural environment, built environment, culture and recreation, and civic traditions. Sawicki (2002) identified four broad categories of measures that are suitable to capture QOP: micro- and macro measures of persons and households, spatial measures of subpopulations, and measures of the geographic area’s characteristics.
This effort used the QOP indicators identified by previous research to generate broad QOP indexes. An important consideration in evaluating the relationship between these QOP indicators and positive outcomes for cities is the issue of endogeneity between the indicators and the outcomes. To provide a stronger argument that placemaking initiatives are tied to outcomes, it was necessary to address the distinction between QOP factors leading to positive outcomes and positive outcomes leading to higher QOP. We tackled this identification problem through the use of QOP indicators that were measured a decade prior to the measurement of the outcome.
Data and Method
Based on specific indicators from previous research and guided by recommendations in prior research regarding categories of indicators, six broad QOP factors were identified: crime, entertainment, density, diversity, housing, and knowledge workers. Broad factors measuring climate or environment were excluded, as these aspects of place, such as annual precipitation or average January temperature, are not clearly modifiable through policy mechanisms (National Research Council 2002). County-level data on 23 potential indicators for these six factors were collected for 81 counties that comprise the core of an MSA with a population between 250,000 and 500,000, and for 83 counties that comprise the core of an MSA with a population between 500,000 and 2.5 million. All of the potential QOP indicators were measured in 2000, or as closely as possible to 2000. Each of these variables, or a very closely constructed approximation, was used in prior QOP research. Every indicator was then assigned to the factor with which it had the clearest theoretical relationship. A list of the initial indicator variables and their factor groupings, along with the source of the data, appears in the appendix. Table 1 displays descriptive statistics for the 23 potential indicators for the small and midsized MSAs in the sample, and Figure 1 shows the geographic distribution of these MSAs.
Descriptive Statistics for Central Counties of MSAs in Sample, by Size of MSA.
Note. M = mean; SD = standard deviation; MSA = metropolitan statistical area; STEM = science, technology, engineering, and mathematics.
Geographic distribution of the MSAs in sample.
Confirmatory factor analysis (CFA) was used to reduce the dimensionality of the QOP problem and to evaluate whether the indicators are appropriate measures of the underlying factors to which they have been assigned. Although a factor analysis will generate a number of factors equal to the number of variables included in the analysis, many of the generated factors will be inconsequential in explaining the variation in the input data. A significant determination in the CFA methodology is thus identifying whether the theoretically grouped variables actually load well on a single significant factor. This analysis followed Kaiser’s rule (Jolliffe 2002), in which factors that have an eigenvalue greater than one are deemed significant and retained. The generated factors were then rotated using a varimax rotation to allow for easier interpretation. In most cases, all of the indicators used in the construction of the factor loaded heavily on the factor. Table 2 displays the breakdown of the indicators assigned to each of the six factors, as well as the factor loading for the indicator on the constructed factor. Table 3 shows the cities scoring the highest and the cities scoring the lowest among each of the constructed factors.
Factor Loadings for Variables Included in CFA.
Note. CFA = confirmatory factor analysis; CRM = crime factor; DEN = density factor; DIV = diversity factor; ENT = entertainment, culture, and recreation factor; HOU = housing factor; KNW = knowledge worker factor; STEM = science, technology, engineering, and mathematics.
MSA Central Counties with High/Low Factor Values, by MSA Size.
Note. MSA = metropolitan statistical area.
The crime indicators had a fairly high bivariate correlation within the MSAs in the study (ρ = .64), so it is unsurprising that both loaded heavily on the constructed crime factor. The indicators included in the density factor were those expected to be salient to younger generations that value urban living and eschew a culture of automobiles. All of these indicators loaded heavily on the density factor, with the lowest factor loading of 0.45 for the population without a vehicle. The utilization of variables measuring diversity is well established within the QOP framework (Black et al. 2000; Florida and Gates 2001). The three indicators used here loaded moderately on the diversity factor, with factor loadings ranging from 0.43 (gay index) to 0.77 (% foreign born). The indicators that comprised the entertainment factor included common measures of “things to do,” including recreation and dining options. Although the loadings for the full-service restaurant variable and developed open-space land variable were notably lower than the others, these variables were retained. Higher values of the indicators incorporated into the housing factor were expected to inhibit future growth and development. The percentage of housing units that are inferior (e.g., lack plumbing or full kitchen facilities) and the percentage that are expensive loaded very highly, whereas the rental occupancy rate had a much lower factor loading. However, rental occupancy was retained as it measures an important aspect of the housing market that was not captured in any of the other factors. The knowledge worker indicators loaded moderately to highly with the exception of the self-employment variable, which displays both small magnitude and reversed directionality. It was removed from the final construction of the factor.
These six factors were incorporated into regression models consistent with creative class theory, in which the outcomes were growth in different population segments. The outcomes considered were the percentage point change between 2000 and 2013 in the population aged 25 to 34 with a bachelor’s degree, the percentage point change between 2000 and 2013 in the adult population that has a bachelor’s degree, and the percentage growth between 2000 and 2013 in the total population aged 25+. The use of the percentage point change in the educational attainment outcomes rather than the percentage change is meant to minimize the risk of overstating any observed change in the dependent variable when the starting point is very low. Based on the prior literature and theory described above, the crime and housing factors are expected to show a negative association with each of the outcomes, and the other four factors are expected to exhibit a positive association with each of the outcomes.
Population counts and counts of population by educational attainment were obtained from Summary File 3 of the 2000 Decennial Census and from the 2013 American Community Survey five-year estimates, both available from the U.S. Census Bureau. The outcomes were measured a decade later than the QOP factors, which will diminish the endogeneity problem noted previously. All of the regression models also include dummy variables for the census region of the United States in which the MSA is located (Figure 1), to control for regional population trends during the analysis period. These regional indicators are meant to capture unobserved heterogeneity between the different U.S. regions that is not ascribable to any of the included factors.
Results
Prior to carrying out the regression analyses, collinearity in the constructed factor variables and regional dummy variables was assessed. Table 4 displays the bivariate correlation coefficients between each of the variables included in the models. Although many of the correlation coefficients are quite low—as might be expected based on the deliberate generation of the factors—there are some notably high correlations (e.g., between the housing factor and diversity factor). However, none of the variables exhibited variance inflation factors greater than 5—a commonly used threshold for problematic collinearity—when the models were estimated, so all of the factors were retained in the models (Berman and Wang 2012).
Correlation Matrices for Factor Variables and Regions.
Note. MSA = metropolitan statistical area; CRM = crime factor; DEN = density factor; DIV = diversity factor; ENT = entertainment, culture, and recreation factor; HOU = housing factor; KNW = knowledge worker factor; NE = Northeast region; MW = Midwest region; SO = South region; WE = West region.
Preliminary regression results suggested that there was an outlier in the midsized MSA equations that could potentially be exerting undue influence on the estimated coefficients. This outlying observation—the city of San Jose—exhibited large Cook’s D values in each of the regressions. This exceptional influence appeared to be the result of an extreme value in the knowledge worker factor, which was more than 2.5 times larger than the second-largest value among the midsized MSAs. Based on the unwarranted influence that this MSA was having on the regression results, it was eliminated from the analysis. However, it should be noted that the removal of this observation makes no difference in the directions or signs of the estimated coefficients reported below; the effect is in the estimated standard errors and the consequent statistical significance of the observed factors.
The top panel of Table 5 displays the standardized coefficients from the regression of the six QOP factors on the percentage point change in the population aged 25 to 34 that has a bachelor’s degree. These results provide conflicting evidence that QOP factors are associated with growth in the population that is often considered the most attractive in QOP efforts. The density factor was associated with growth in the young adult college-educated population in the midsized cities, a result that is consistent with expectations. However, in both the midsized and small MSAs, the presence of knowledge workers in the earlier period exhibited a negative relationship with the college attainment for young adults. This result is inconsistent with conventional wisdom on what attracts college-educated young people to an area. In this model of change in the presence of college-educated population aged 25 to 34, the regional effects dominate the QOP effects. The negative coefficients suggest that growth in the younger educated population is occurring most rapidly in the Northeast; this is true for both small and midsized MSAs.
Standardized Coefficients from Regression of Population Change Outcomes on Six QOP Factors, by Size of MSA.
Note. The t-statistics are in parentheses. All models also include a constant term. QOP = quality of place; MSA = metropolitan statistical area.
p < .05. **p < .01. ***p < .001.
The second outcome measure was the percentage point change in the population aged 25+ with a college degree; the standardized coefficients from this model are shown in the middle panel of Table 5. Several of the factors, including the crime factor, the density factor, and the entertainment factor, were significantly associated (p < .05) with future growth in the college-educated population. The directions of these associations were consistent with expectations, suggesting that several of the primary focuses of QOP initiatives may indeed have a beneficial outcome for this particular measure. For the small MSAs, higher values of the entertainment factor were also significantly related to increasing college-educated populations, although crime and density did not show statistically significant relationships in this group. Except for the West region in the small MSA model, regional effects did not explain a significant amount of the between-MSA variation in college-educated population.
The standardized coefficients from the regression of the QOP factors on the change in the total population aged 25 years or older are shown in the bottom panel of Table 5. In midsized MSAs, lower rates of age 25+ population growth during the 2000–2013 period were significantly associated (p < .05) with greater density, whereas positive population growth was significantly associated with higher scores on the diversity factor and the knowledge worker factor. Although these positive relationships were expected, the negative relationship between density and age 25+ population growth was surprising. Although the density factor showed a similar association with age 25+ population growth in small MSAs as it did in the midsized MSAs, neither the diversity nor the knowledge worker factors had a relationship with population growth in the smaller areas.
The observed effects of the QOP factors in midsized cities were in addition to the regional effects, which show increased population growth in the South and West relative to Northeast and Midwest. The magnitudes of the standardized coefficients suggest that regional effects surpassed the QOP factors in explaining the variation in adult population growth observed in these cities over this period. However, density, diversity, and knowledge worker presence remain significant predictors of population change even after accounting for these secular population changes. Interestingly, there are no statistically significant regional effects among the small cities.
In each of the models, the amount of variation in the outcome variable explained by the six QOP factors was moderate but not negligible. For the midsized MSAs, the QOP factors (along with the regional fixed effects) explained between 38% and 58% of the variation in the outcome, depending on the specific outcome in question. Within the small MSAs, the constructed factors explained noticeably lower amounts of the variation in outcomes.
Discussion
The first goal of this article was to assess relationships between various variables that might be used as measurable indicators of QOP, and to evaluate whether these variables load heavily onto theoretically recognized and interpretable factors. The results of this CFA were reasonably successful, with the 23 initial variables initially assigned to six QOP-relevant factors loading appropriately on the assigned factor. There were a few inconsistencies in the original designation of the indicator variables. For example, the variable representing the proportion of restaurants that are full-service, meant to capture an aspect of culture, did not load as highly on the entertainment and culture factors as did the other variables used to create that factor. The percentage of workers self-employed, which was expected to load heavily on the knowledge worker factor, did not fit into that factor at all and was dropped from the analysis. The remaining indicator variables fit well with the factor into which they were assigned, with factor loadings ranging from 0.20 to 0.92. The use of theoretically structured factors makes the broad interpretation of the six factors straightforward.
In general, the results from the regression models suggest that some QOP-based placemaking strategies may be effective in attracting knowledge workers in midsized cities; the results for the smaller cities are less certain. Although this article does not address the exact degree to which individual QOP indicators are associated with increases in the concentration of college-educated individuals, it does provide evidence that some broadly defined QOP investments may be more successful than others.
Turning first to the QOP effects on college degree attainment among the population aged 25 to 34, the most prominent result is that the presence of knowledge workers in the earlier period is not associated with increased future growth in this population. This result is true for both small and midsized cities, and is in direct opposition to place-based theories of growth. An examination of the data reveals that the bivariate correlation between the knowledge factor and change in percentage of college-educated population aged 25 to 34 is insignificant (ρ = .05) for cities of both size, suggesting the observed negative effect is the result of mediation or moderation by other variables in the model. However, the lack of a significant bivariate correlation is still surprising, given the importance of the presence of knowledge workers in the QOP discourse. This outcome may reflect the fact that cities with high knowledge worker factor values already have high college attainment of the population aged 25 to 34 and may thus expect smaller percentage point gains in the future.
The density factor is positively associated with change in the college-educated population aged 25 to 34 in midsized MSAs but not in small MSAs. Higher density is commonly associated with walkability and a variety of transportation options in a city. Small cities may not have the transportation options implicit in the density factor, as they may lack the critical mass to make light rail or efficient bus routes attainable.
The remainder of the QOP factors that were hypothesized to have an effect on changes in the educated young adult population show no significant effects; however, the regional effects are statistically significant and indicate greater growth in this young population in the Northeast region. The importance of the regional effects in this model suggests that there are unmeasured characteristics that are affecting college attainment in the small and midsized cities in this study. For example, states in this region may be investing more heavily in higher education, or migrants into this region may be skewed toward the more highly educated.
Looking at the results for all college-educated adults, several factors emerge as important in explaining variation in the midsized MSAs. As it was for the population aged 25 to 34, the density factor is positively associated with change in the college-educated population in midsized MSAs but not small MSAs. The crime/safety factor also exhibits the expected negative relationship with the QOP outcomes in midsized cities but not small cities. This outcome might be explained by the fact that small cities exhibited somewhat lower violent and property crime rates than their midsized brethren, making the crime issue less salient in these places. Of all the QOP factors, entertainment may be the one most identified with placemaking policies. The findings for the growth in the college-educated population suggest that the attention paid to this aspect of QOP may be warranted in some respects, as the entertainment factor exhibits a significant positive relationship with growth in the college-educated group in both small and midsized cities. This fact may be especially beneficial for cities undertaking QOP efforts, as increasing some of the indicators included in the entertainment factor (e.g., recreational facilities) may be more easily achieved than increasing indicators in the other factors (e.g., crime reduction).
The knowledge worker factor, which was expected to lead to greater growth in the college-educated population, shows no significant association in the model. However, unlike the case for the percentage of college-educated population aged 25 to 34, the bivariate correlation between the knowledge worker factor and change in the total percent of college-educated population is positive and significant (ρ = .05) for cities of both size. This implies that the expected underlying relationship between the presence of knowledge workers and college-educated growth is present in these MSAs but is ultimately mediated by other variables in the regression model.
Notably among the results for the college-educated outcomes, diversity plays no role here. This is generally inconsistent with creative class theory, in which diversity is a central component, although the expectation of a relationship between racial diversity and economic development—as well as the mechanism through which the relationship would operate—remains an open question (Thomas and Darnton 2006). It is important to note, however, that the specific operationalization of diversity used in this study—in which the presence of non-Whites, immigrants, and gays and lesbians are combined into a single index—merges distinct groups that were considered separately in Florida’s (2002) original creative class work. This merged diversity factor may mask the effects of the individual indicators of which it is comprised. In fact, the bivariate correlation between the percent non-White and the growth in the college-educated population is negative and significant (ρ = .05) within both small and midsized cities, and is the only component indicator within this factor that exhibits a significant bivariate association with this outcome. This suggests the complexity of capturing the human capital effects of diversity when a broad conception of diversity is used.
Details on the operationalization of the factor notwithstanding, there is a lingering question regarding the effect of contemporary U.S. population processes on the expected relationship between diversity and growth in human capital outcomes. Within the central counties used in this study (and within the United States as a whole), non-Hispanic Whites have higher levels of bachelor’s degree attainment than Blacks or non-Hispanics, the two largest groups in the non-White category. According to the 2013 American Community Survey five-year estimates for the midsized cities in the study, 36.6% of non-Hispanic Whites aged 25+ had a bachelor’s degree or higher compared with 18.1% of Blacks and 13.6% of Hispanics. This disparity in educational attainment by race and ethnicity is also present in the small cities, although overall rates of attainment are somewhat lower in the smaller cities.
However, much of the aged 25+ growth that these cities are experiencing is in the non-White population, and this is true regardless of city size or region of the country. Within the central counties of midsized MSAs in the Northeast and Midwest regions, the total population of non-Hispanic Whites aged 25+ decreased between 2000 and 2013. In these regions, the central county adult population growth observed is a result of growth in the adult Hispanic population and, to a lesser extent, the Black and Asian populations. In Southern and Western midsized MSAs, central county non-Hispanic White populations aged 25+ are growing, but their growth is dwarfed by growth in the adult Hispanic population. Overall, the non-Hispanic White population aged 25+ grew by 3.1% in midsized cities and 5.8% in small cities between 2000 and 2013. During the same time period, the midsized city’s Black population aged 25+ grew by 23.5% (23.2% in small cities), and the midsized city’s Hispanic population aged 25+ grew by 64.1% (58.2% in small cities). Thus, there may be some inconsistency in concurrently achieving the goals of greater diversity and greater educational attainment, given the changing racial and ethnic structure of the population and the present disparities in human capital achievement between racial and ethnic groups.
A similar (and related) phenomenon is related to change in population nativity within these cities. In the central counties of midsized MSAs in the Northeast and Midwest, absolute growth in the foreign-born population was greater than absolute growth in the native-born population between 2000 and 2013. Although foreign-born population growth was very large within midsized cities in the South and West regions, the change in native-born populations within these cities was substantially larger. Because native-born educational attainment is somewhat higher than foreign-born educational attainment in the northeastern MSAs, growth in the foreign-born populations within these cities might not be expected to lead to higher percentages of bachelor’s degrees.
Although growth in the total adult population is not commonly cited as an anticipated outcome in the creative class literature, greater growth is a testament to the general health of a place. Growth in the total adult population is relevant as it is the denominator in measures of college attainment. Consistent with the above discussion of changing population structure, total population growth exhibits the expected positive associations with the diversity factor in midsized cities: Cities that were more diverse in terms of non-White, immigrant, and gay populations in 2000 grew faster over the subsequent decade. Interestingly, population growth in midsized cities also exhibited the expected positive association with the knowledge worker factor. However, there exists a negative relationship between the density factor and total adult population growth, a result which may indicate variation in the preference for density among different population segments. For example, young, educated, single people may exhibit a preference for higher density—as presupposed by creative class theory—but older populations and families with children may not.
A primary motivation for evaluating the relationship between QOP factors in small and midsized cities was that much of the extant QOP literature focuses on larger places and the observed associations (or lack thereof) in those analyses may not hold for all cities. This analysis suggests that concern over differences in the effects of QOP factors on knowledge worker and population outcomes for various sized cities may be warranted. Although some of the factors—including diversity, density, and entertainment—displayed the expected association with selected development outcomes in midsized MSAs, this was much less true for smaller MSAs. Given that the numbers of cities in the small and midsized categories are roughly equal, this would not seem to be an issue of statistical power. Rather, there appears to be something about smaller cities that obstructs the expected positive effects of QOP efforts in these cities. Given that QOP ideals are being applied without regard to city size, this issue deserves further research attention.
Limitations of the Methodology
CFA is highly dependent on the set of initially chosen variables, and the individual factor loadings will vary with different sets of input variables included. Although a comprehensive selection of the measures that have been included in prior studies of this type was included, it is possible that some additional QOP indicators could be added or that some of the variables that were included could be discarded. Regardless of this potential limitation, the fact that the included indicators loaded heavily on six constructed factors that have been identified as broad QOP concepts is encouraging.
An additional limitation of this analysis is the lag time between the measurement of the QOP factor and the measurement of the outcome variables. Although every effort was made to include only explanatory variables that could be measured for periods temporally preceding the measure of the dependent variables, this was not always possible, and not every variable could be measured at exactly the same time period. Moreover, the time required for QOP initiatives to trigger an increase in knowledge workers or population growth is unknown.
Stratifying the MSAs in the sample into small and midsized groups permitted a comparison of placemaking effects on population growth outcomes for cities of different sizes (and, presumably, available resources). Unfortunately, it also resulted in smaller samples within each group. However, this is unlikely to be the reason for the lack of statistically significant results among the small MSA regressions, as that group has a similar sample size as the midsized MSA group. As described above, it seems to be the case that the QOP indicators, as defined here, may not be as relevant in smaller cities as they are in midsized cities.
Conclusion
The findings from this effort are slightly more optimistic than other tests of creative class theory on midsized cities. Hoyman and Faricy (2009) found no relationship between their creative class components and the migration of young knowledge workers. Sands and Reese (2008) examined 40 midsized cities in Canada and found essentially no correlation between talent, tolerance, and technology and economic growth. These results also confirm that the specification of factors associated with QOP is critical and drives results. That said, there is evidence that some QOP factors are positively related to growth in the college-educated and the younger college-educated population. These results lend support to the notion that QOP investments may be an effective development strategy for midsized cities and that such investments may be less successful in smaller cities. However, more research is needed to identify the factors relevant to the location decision of knowledge workers. Some amenity investments may indeed be associated with attracting and retaining residents that make a positive net economic contribution to cities of all sizes.
Footnotes
Appendix
Variables Used in Confirmatory Factor Analysis, Justification for the Variable, Data Source, and Year of Data.
| Variable | Citation/Justification | Source(s) | Year |
|---|---|---|---|
| Crime factor | |||
| Property crime rate | Gottlieb (1995) | FBI Uniform Crime Reports | 2000 |
| Violent crime rate |
Gottlieb (1995)
Lambiri, Biagi, and Royuela (2007) |
FBI Uniform Crime Reports | 2000 |
| Density factor | |||
| % of housing renter-occupied | Andrews (2001) (availability and diversity of housing options) | Decennial Census | 2000 |
| % of households with no vehicle | Lambiri et al. (2007) | Decennial Census | 2000 |
| % of housing in structures with 10+ units | Andrews (2001) (availability and diversity of housing options) | Decennial Census | 2000 |
| Population density within developed land | Lambiri et al. (2007) (population density) | Decennial Census National Land Cover Database |
2000–2001 |
| Diversity factor | |||
| % foreign born |
Florida (2002)
Sands and Reese (2008) Trip (2007) |
Decennial Census | 2000 |
| % non-White | Gottlieb (1995) (% Black) Sands and Reese (2008) |
Decennial Census | 2000 |
| Gay index |
Florida (2002)
Trip (2007) |
Decennial Census | 2000 |
| Entertainment, culture, and recreation factor | |||
| NAICS 711–13 (arts and entertainment) establishments per capita | Trip (2007) | County Business Patterns | 2000 |
| % of restaurants full-service | Trip (2007) (restaurants per capita) | County Business Patterns | 2000 |
| Recreational facilities per capita | Trip (2007) | County Business Patterns | 2000 |
| Library circulation per capita | Florida (2002) (book acquisitions) | Public Libraries Survey (NCES) | 2000 |
| % of developed open-space land (e.g., parks) | Gottlieb (1995) (acres of state parks) National Research Council (2002) |
National Land Cover Database | 2001 |
| Housing factor | |||
| % of housing units with problems |
Sufian (1993)
Lambiri et al. (2007) |
Decennial Census | 2000 |
| % of households spending >30% of income on housing |
Sufian (1993)
Lambiri et al. (2007) |
Decennial Census | 2000 |
| Vacancy rate for rental housing | Andrews (2001) (availability and diversity of housing options) | Decennial Census | 2000 |
| Knowledge worker factor | |||
| Median weekly goods-producing wage | Sands and Reese (2008) | Quarterly Census of Employment and Wages | 2000 |
| Median weekly service-producing wage | Sands and Reese (2008) | Quarterly Census of Employment and Wages | 2000 |
| % of workers in STEM occupations | Sands and Reese (2008) | Decennial Census | 2000 |
| Patents issued per capita | Trip (2007) | U.S. Patent and Trademark Office | 2000–2005 |
| Research university full-time equivalent enrollment | Gottlieb (1995) (graduate students) Andrews (2001) (educational opportunities) |
NCES (IPEDS) | 2000 |
| % of workers self-employed | Florida (2000) | Decennial Census | 2000 |
Note. NAICS = North American Industry Classification System; NCES = National Center for Education Statistics; STEM = science, technology, engineering, and mathematics; IPEDS = Integrated Postsecondary Education Data System.
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.