Regional Science and Urban Economics 46 (2014) 12–17

Contents lists available at ScienceDirect

Regional Science and Urban Economics

j ourna l homepage: www.e lsev ie r .com/ locate / regec

Does culture affect local productivity and urban amenities?☆,☆☆

Brahim BoualamUniversity of Geneva, Geneva School of Economics and Management, 40 Bd. du Pont D'Arve 1211, Geneva 4, Switzerland

☆ The author gratefully acknowledges very helpful coCéline Carrère, Ann Markusen, Marcelo Olarreaga, GioFrédéric Robert-Nicoud, Farid Toubal, two anonymous r17th International Conference on Cultural EconomicsNorth-American Meeting of the Regional Science Asso2012).☆☆ Appendix A and Appendix B at the end of the paperand data sources. Appendix C, which contains Tables C.1scriptive statistics and robustness checks, is not publishedpaper version at: http://boualamb.weebly.com/research.

E-mail address: Brahim.Boualam@unige.ch.

http://dx.doi.org/10.1016/j.regsciurbeco.2014.01.0080166-0462/© 2014 Elsevier B.V. All rights reserved.

a b s t r a c t

a r t i c l e i n f o

Article history:Received 27 March 2013Received in revised form 28 January 2014Accepted 28 January 2014Available online 13 February 2014

Keywords:Urban economicsLocation choiceLocal amenitiesCulture

Does a better cultural milieu make a city more livable for residents and improve its business environmentfor firms? I compute a measure of cultural specialization for 346 U.S. metropolitan areas and ask if differ-ences in cultural environment across cities capitalize into housing price and wage differentials. Simple cor-relations replicate standard results from the literature: cities that are more specialized in culturaloccupations enjoy higher factor prices. Estimations using time-series data, controlling for city characteris-tics and correcting for endogeneity weaken the magnitude of this effect. Even though the arts and culturemight be appealing to some people and firms, such determinants are not strong enough to affect factorprices at the city level.

© 2014 Elsevier B.V. All rights reserved.

1. Introduction

This paper asks if a better cultural milieu can improve the attrac-tiveness of a city. Cities like Paris, London and New York tend to bemore attractive partially because of their cultural vitality. But arethese differences strong enough to be considered as relevant deter-minants of the location of firms and residents? To answer this ques-tion, I evaluate how culture shapes the relative demand for a city byestimating hedonic rent and wage equations. Using a large sample ofU.S. metropolitan areas between 2005 and 2011, the empirical anal-ysis shows that cultural determinants are not strong enough to affectfactor prices at the city level.

Cultural policies are increasingly considered as drivers of eco-nomic growth and urban recovery. Famous contributions by Richard

mments and suggestions fromvanni Peri, Jordan Rappaport,eferees and participants at theACEI (Kyoto, 2012) and 59thciation International (Ottawa,

report the main tables, figuresto C.8 that provide further de-but is available in the discussion

Florida (Florida, 2002a,b; Florida and Mellander, 2010) popularizedthe idea that culture positively affects city attractiveness. Accord-ingly, the presence of artists and creative people tends to attracthighly-educated and talented workers, which in turns favor the ex-pansion of skill-intensive and innovative industries. Theories in thespirit of Florida (2002a,b) received extraordinary attention frompolicy makers and the media. However, they are also sharply criti-cized by academic researchers due to major limitations (Glaeser,2005; Markusen, 2006; Montgomery, 2005; Hoyman and Faricy,2009; Sawicki, 2003; Marcuse, 2003; Peck, 2005). First, they oftenfail to provide well-defined and exhaustive measures of the culturalintensity of places. Second, they rarely take account of the impact ofother city-specific characteristics that might also influence the loca-tion of economic agents. Third, they do not consider potential prob-lems of reverse causality between cultural variables and economicoutcomes.

This paper addresses these three limitations and questionsthe existing empirical literature on this topic. I propose a mea-sure of cultural employment based on the type of tasks per-formed by employees. I restrict this measure to occupationsthat are intrinsically oriented towards the production of non-tradable cultural goods and services since only these potential-ly affect the utility of residents and firms at the local scale.Next, I estimate hedonic wage and rent equations to evaluatethe impact of this variable on urban productivity and consump-tion amenities. I control for the impact of natural, political andother location-specific variables by including a full set of fac-tors that could influence the location of economic agents.

13B. Boualam / Regional Science and Urban Economics 46 (2014) 12–17

Lastly, I correct for the endogenous determination of culturalsupply by implementing an instrumental variable strategy. Iuse the annual amount of federal grants received by individualartists and art organizations in each city as an instrument for cul-tural vitality.

The simplest specifications recover findings from existing literature(Florida and Mellander, 2010; Sheppard et al., 2006; Clark and Kahn,1988). In the cross-section of American cities, estimates report apositive effect of culture on city attractiveness. Further empiricalinvestigations using time-series data and controlling for city-specificcharacteristics reduce the estimated effect of culture on factor prices.Estimations that correct for endogeneity ultimately reveal that the effectof culture becomes negligible. I conclude that the standard positiveeffect associated with culture captures the impact of omitted variablesand results from the simultaneous determination of culture, wages andrents. This interpretation is supported by various additional checks.

2. Identification strategy

To determine how differences in the cultural landscape across citiesaffect the location of people and firms, I rely on the identification strat-egy proposed by Roback (1982). In thismodel,firms and households areperfectly mobile across locations. At a spatial equilibrium, therefore, allagents are indifferent among locations.

This inter-city model of location is used to evaluate how residentsand local firms value localized amenities. We can indeed allow indirectutility and profits to be possibly affected by a city specific attribute thatvaries across cities. The latter is defined as a consumption amenity aslong as it positively affects consumer's utility and as a productiveamenity when it enhances firms productivity and therefore economicprofits. In the presence of a consumption amenity, any increase in thelevel of amenity must be compensated either by a decrease in wagesor by a rise in land rents to eliminate workers' incentives to move tothis city. Residents then incur a loss in purchasing power but benefitfrom the presence of the amenity in the city. In the case of a productionamenity, incentives to move are arbitraged away either by a rise inwages or in rents: firms' costs are higher in the city but the locationremains attractive thanks to the valuable amenity.

We can easily derive from this framework the hedonic rentand wage equations that describe the relationship betweenurban amenities and these two factor prices. The analysis ofthe rent equation allows determining the overall effect of cul-ture: the fact that high-amenity cities experience higher rentssuggests that demand for these locations is higher. The wageequation helps us to determine if this overall effect is dominat-ed by the impact on firms or households. A negative impact onwages mirrors the fact that workers are willing to give upwages to live in high-amenity cities. In contrast, higher wagesimply that firms can sustain higher land and labor costs insuch high-amenity cities, reflecting the fact that it is a produc-tive amenity. Therefore, the Roback (1982) model provides avery suitable identification procedure for assessing how house-holds and firms are affected by inter-city differences in theircultural environment. Table A.1 summarizes the main resultsof this model in the case of positive or negative amenities inboth consumption or production.

Table A.1Analysis of the hedonic wage and rent equations.

Wages

N0 b0

Rents N0 Production amenity Consumption amenityb0 Consumption disamenity Production disamenity

3. Data

I use recent Occupational Employment Statistics (OES) fromthe U.S. Bureau of Labor Statistics (BLS) describing employmentfor 372 U.S. Metropolitan Statistical Areas between 2005 and2011. The occupational classification system (SOC) reports em-ployment statistics based on the type of tasks performed byworkers. Employment data are then disaggregated into 22major working activities which are in turn broken down into840 detailed occupations. Markusen (2006) or Glaeser et al.(2001) show that existing measures of cultural employmentbased on broad occupational categories can face major limita-tions. Therefore, I delineate cultural employment by inspectingeach detailed occupation. In addition, I restrict cultural occupa-tions to tasks that are mainly oriented towards the productionof goods and services that are non traded and therefore mostlylocally consumed. This includes for instance art teachers, curators,museum technicians, conservators or librarians.1 I compute theannual share of cultural workers in total labor force for 346Metropolitan Statistical Areas2 (MSA), covering 82% of total USpopulation. Using this definition, the share of cultural employmentranges between 0 and 0.8% with a median value of 0.2%.

Urban wages also come from the U.S. Bureau of Labor Statis-tics and are defined as the median hourly wage of all non-cultural workers in each metropolitan area. Data on rentalprices come from the “50th percentile series” of the U.S.Department of Housing and Urban Development (HUD). Thesecorrespond to gross rent estimates, including utilities, at the50th percentile point of the rent distribution of rental housingunits. To control for a major difference in housing characteristics,I use the median gross rent of a 2-bedroom rental unit (Saiz,2007). Sources and definitions of control variables are provided inAppendix B. A statistical summary of the data is given in the onlineTable C.2.

4. Empirical results

4.1. Baseline regressions

To determine the effect of culture on production and consumptionamenities, I estimate the following reduced form equation using ordi-nary least squares (OLS) on the panel of U.S. cities over the period2005–2011:

ln Yctð Þ ¼ β0 þ β1 � ln Sctð Þ þ β2 � ln Zctð Þ þ β3 � ln Xcð Þ þ γc þ ϵct ð1Þ

where ln(Yct) alternatively stands for the logarithm of the mediangross rent and the median hourly wage in city c in year t. ln(Sct)is the measure of cultural specialization of city c (in log) while Zctand Xc respectively correspond to vectors of time-varying and invari-ant city attributes. I also include in several estimations a set of city

1 Estimations using a wider definition of culture including all culture-related occupa-tions provide similar results. A description of cultural occupations is given in the onlineTable C.1.

2 These geographical units correspond to local labor markets with strong commutingties between each component. I restrict the sample to cities located in contiguous conti-nental US states and exclude NewEngland City and TownAreas (NECTAs) forwhich a cor-responding MSA cannot be found.

5 The Kleinbergen–Paap statistics on the instruments shows that the F-statistics isabove the 15% Stock and Yogo (2005) critical size suggesting that the instrument is pretty

14 B. Boualam / Regional Science and Urban Economics 46 (2014) 12–17

fixed effects γc to control for unobserved city characteristics. Finally,ϵct is an error term.

Column (1) of Tables A.2 and A.3 displays correlations betweencultural employment and rents and wages, respectively. Americancities specialized in cultural occupations seem to experience higherfactor prices as the estimated coefficients are positive and highly sig-nificant in both regressions. A 10% increase in cultural employmentshare is associated with a 1.1% rise in median rents and a 0.9%increase in median wages. These results are consistent with a pre-vailing positive effect of culture on production amenities, in linewith Florida and Mellander (2010). Fig. A.1 provides a graphical rep-resentation of these relationships.

Column (2) of Tables A.2 and A.3 includes additional city covari-ates to control for major determinants of rents and wages that arenot otherwise related to culture. These specifications turn out toexplain about 75% of the total variance of median rents and 61% ofmedian wages in American cities. Most of the control variables arestatistically significant and consistent with existing findings.3 Over-all, the effect of culture on factor prices remains positive but themagnitude of the coefficients (and the significance level in the rentequation) decreases with the number of controls. One step furtheris taken by using the time dimension of the data in order to controlfor all time invariant and unobserved city-specific features. Esti-mates of Eq. (1) with city fixed effects are reported in columns (3).Regressions also include a fixed effect for the highest cultural citiesin the panel. Fixed effect estimations confirm findings of columns(2) as the estimated impact of culture turns out to be smaller inboth equations and remains significant only at a 5% significancelevel. Coefficients associated with time-varying control variablesremain stable in the within estimations except for a few number ofcovariates. These results cast additional doubt on the potentialpositive impact of culture on city attractiveness and confirm thesuspicion that cultural variables overlap with other city specificattributes.

4.2. Instrumental variables

OLS results cannot be interpreted as causal relationships sinceendogeneity problems may severely bias the estimates. We mayindeed suspect cultural employment to be affected by all determi-nants of local employment, wages or factor prices: if culture reactsto changes in factor prices at the local scale, estimates are biasedmaking statistical inferences subject to severe criticisms. I applyinstrumental variable techniques (IV) to correct this issue. I usethe (per capita) amount of Federal grants annually awarded bythe National Endowment for the Arts (NEA) to local art projectsas an instrument for cultural supply in each MSA. Federal grantssatisfy the requirements of a valid instrument: they are quite sub-stantial and constitute a significant shock to the production of cul-ture in the city. Besides, any idiosyncratic shock to factor prices ina city is very unlikely to affect NEA's financial endowment. Grantsare not determined by current values of wages and rents butinstead, are shown to mirror the average quality of art work ineach city thanks to a strict peer review grant making system.4

Grants are therefore correlated with a city's ability to produce cul-tural goods and services but are not otherwise correlated with theerror term.

Column (4) of Tables A.2 and A.3 describes the results of the IVestimations for the rent andwage equation respectively. All reported

3 See Shapiro (2006) or Glaeser and Saiz (2003) for the coefficient on education;Ottaviano and Peri (2006), Saiz (2007) and Glaeser et al. (2001) for migration and theshare of non-white people in total population; Rosenthal and Strange (2004) and Meloet al. (2009) for density; Albouy (2008) and Rappaport (2007) for natural amenities andSaiz (2007) andRoback (1982) for unemployment. See the discussion paper for a full anal-ysis of these estimates.

4 A full description of the instrument is provided in the discussion paper.

statistical tests uniformly show that NEA's grants is a valid instrumentfor cultural employment.5 When controlling for endogeneity issueswith this valid instrument, culture does not play a significant role inexplaining factor prices at the local scale. Indeed, the coefficient associ-ated with culture turns out to be insignificant in both equations. Thisresult suggests that previous OLS estimates are upwardly biased,consistent with the intuition that artists tend to locate in citieswith high wages and high rents. The sign of both coefficients wouldsuggest that culture may be a consumption amenity that positivelyaffects city livability — since it affects wages negatively and rentspositively — but the impact on factor prices is too low to significantlycapitalize into factor price differences. The estimates on the otherexplanatory variables are not affected by the IV estimations.

IV estimations ultimately show that strong positive relationshipsthat can be easily and constantly found using simple correlationsare harder to replicate when controlling for city characteristicsand solving for endogeneity problems that undoubtedly affect he-donic estimations. As robustness checks, I replicate this analysiswith alternative measures of cultural employment such as a con-centration index of cultural employment and a measure of rela-tive specialization compared to the national average.6 I also useseveral measures of access to commercial and non-commercial artestablishments defined as the number of theaters, book, musicand video stores, museums and libraries per inhabitants.7 I similar-ly examine the effect of culture on employment and populationgrowth and implement a Seemingly Unrelated Regression tech-nique (SUR) to take into account the correlation of errors acrossequations.8 Overall, these estimations go in the same directionand unambiguously weaken the estimated impact of culture.These findings systematically support the idea that simple correla-tions or naive estimations are actually marred by omitted variablebiases, measurement problems and reverse causality issues. Thisempirical exercise reveals that we can easily question the existingliterature on this topic.

5. Conclusion

Research on urban economics and economic geography is in-creasingly introducing culture related variables to explain propertyvalues, urban growth or quality of life (see for example Glaeseret al., 2001; Albouy, 2008; Sheppard et al., 2006; Ahlfeldt andMastro, 2012; Glaeser and Saiz, 2003; Clark and Kahn, 1988).These studies account for a restricted number of cultural infra-structures – such as museums, theaters or art venues – and there-fore do not seek to provide broad evidence on the role of culture.The most striking findings followed on from Richard Florida's cre-ative class theory. However, this approach is strongly and easilycriticized because of methodological and statistical limitations.

This article is a work of caution on the empirical literature. It con-tributes to research on culture and urban amenities by addressing itsmain limitations. It notably shows that potential reverse causalityproblems between cultural variables and economic outcomes mustbe carefully considered to ensure that if any relationship betweenthese variables is found, it can only be explained by the direct contri-bution of the arts.

strong, i.e. correlated with the endogenous regressor. Similarly, the F-statistics of the firststage regression is above 10which is the traditional rule of thumb for testing the relevanceof instruments in the first stage regression in case of a single endogenous regressor. Sim-ilarly, the underidentification KP-LM test suggests that the instruments are relevant andtherefore that the model is identified.

6 Results are provided in online Tables C.6 and C.7.7 See online Tables C.3 and C.4.8 See online Tables C.8 and C.5 respectively.

15B. Boualam / Regional Science and Urban Economics 46 (2014) 12–17

Appendix A. Figures and tables

(a) Median rents (b) Median wages

Fig. A.1. Rents, wages and culture.

Table A.2Rent regressions.

Dependent variable: Rents (1) (2) (3) (4)

OLS OLS Within IV

Culture 0.110*** 0.012** 0.010** 0.028(0.009) (0.005) (0.005) (0.056)

Education 0.174*** 0.069*** 0.073**(0.012) (0.027) (0.029)

Migration 0.082*** 0.033*** 0.034***(0.005) (0.009) (0.009)

Racial composition 0.042*** −0.063*** −0.062***(0.007) (0.010) (0.011)

Unemployment 0.097*** 0.089*** 0.086***(0.007) (0.005) (0.011)

Density 0.047*** 0.949*** 0.923***(0.004) (0.112) (0.126)

Service sector 0.060*** 0.039*** 0.040***(0.010) (0.007) (0.006)

Violence rate −0.031*** 0.004 0.004(0.007) (0.012) (0.010)

Amusement and recreation 0.074*** −0.065** −0.064***(0.013) (0.027) (0.020)

Food and drinking places 0.045** 0.385*** 0.378***(0.021) (0.056) (0.049)

Public schools 0.170***(0.013)

Humidity −0.019*(0.010)

Temp. January 0.146***(0.010)

Temp. July −1.016***(0.059)

Sunlight 0.194***(0.012)

Observations 2331 1949 1949 1943R-squared 0.061 0.754 0.601 0.595City fixed effects No No Yes YesKleibergen–Paap Weak Instr. 9.465Underidentification KP-LM 0.00850First stage F-statistics 11.23

Robust standard errors in parentheses; *** Significant at the 1 percent level, ** 5 percent level, and * 10 percent level.

Table A.3Wages regressions.

Dependent variable: Wages (1) (2) (3) (4)

OLS OLS Within IV

Culture 0.092*** 0.011*** 0.005** −0.016(0.005) (0.004) (0.002) (0.034)

Education 0.149*** 0.039** 0.035**(0.011) (0.016) (0.017)

Migration 0.009** 0.024*** 0.023***(0.004) (0.007) (0.006)

Racial composition 0.032*** −0.053*** −0.054***(0.005) (0.007) (0.007)

Unemployment 0.075*** 0.062*** 0.066***(0.005) (0.004) (0.007)

Density 0.041*** 0.881*** 0.911***(0.003) (0.095) (0.085)

Service sector 0.028*** 0.022*** 0.021***(0.006) (0.005) (0.004)

Violence rate −0.002 0.002 0.002(0.005) (0.008) (0.007)

Amusement and recreation −0.001 −0.049*** −0.051***(0.009) (0.017) (0.014)

Food and drinking places −0.023 0.209*** 0.218***(0.016) (0.035) (0.034)

Public schools 0.075***(0.009)

Humidity −0.030***(0.008)

Temp. January −0.041***(0.007)

Temp. July −0.524***(0.045)

Sunlight −0.002(0.009)

Observations 2340 1947 1947 1941R-squared 0.125 0.614 0.699 0.685City fixed effects No No Yes YesKleibergen–Paap Weak Instr. 9.500Underidentification KP-LM 0.00839First stage F-statistics 11.25

Robust standard errors in parentheses; *** Significant at the 1 percent level, ** 5 percent level, and * 10 percent level.

16 B. Boualam / Regional Science and Urban Economics 46 (2014) 12–17

Appendix B. Data sources

Education: Share of population of 25 years or more having at least abachelor degree using data from the American Community Survey esti-mates (ACS).

Migration: Share of foreign born residents in total population usingACS data.

Racial composition: Share of ‘non-white’ workers in total populationusing the Population by Race and Hispanic Origin Table derived fromU.S. Census and the Selected Social Characteristics from the ACSestimates.

Unemployment: Annual rate of unemployment by metropolitan areaprovided by the Smoothed Seasonally Adjusted Metropolitan Area Esti-mates series from the BLS.

Density: Total population per square mile of land area.Service sector: Annual share of employment in the service sector ex-

tracted from the U.S. County Business Patterns.Violence rate: Number of murders, forcible rapes, robberies and ag-

gravated assault per 100,000 inhabitants. Data come from the FederalBureau of Investigation (FBI).

Amusement and recreation: Number of amusement, gambling andrecreational establishments per inhabitant including amusementparks, golf courses, fitness and recreational sports centers or bowling al-leys from the County Business Patterns.

Food and drinking places: Number of food and beverage establish-ments per capita extracted from the County Business Patterns.

Public schools: Total current spending in elementary and second-ary schools per pupil. Primary data are extracted from the Public El-ementary — Secondary Education Finance Data from the U.S. Census

Bureau for the year 2009. Data reported at the school districts levelare aggregated to compute an average value at the MSA level.

Natural amenities: Monthly average temperatures in January andJuly, average number of hours of sunlight in January and average rela-tive humidity. These average data are computed over 30 years(1941–1970) and extracted from the Natural Amenities Scale Datasetfrom the U.S. Department of Agriculture.

References

Ahlfeldt, G., Mastro, A., 2012. Valuing iconic design: Frank Lloyd wright architecture inOak Park, Illinois. Hous. Stud. 27 (8), 1079–1099.

Albouy, D., 2008. Are big cities bad places to live? Estimating quality of life across metro-politan areas. NBER Working Papers, no 14472.

Clark, D.E., Kahn, J.R., 1988. The social benefits of urban cultural amenities. J. Reg. Sci. 28,363–377.

Florida, R., 2002a. Bohemia and economic geography. J. Econ. Geogr. 2 (1), 55–71.Florida, R., 2002b. The Rise of the Creative Class and How it is TransformingWork, Leisure,

Community and Everyday Life. Basic Books, New York.Florida, R., Mellander, C., 2010. There goes the metro: how and why Bohemians, artists

and gays affect regional housing values. J. Econ. Geogr. 10 (2), 167–188.Glaeser, E., 2005. Review of Richard Florida's The Rise of the Creative Class. Reg. Sci. Urban

Econ. 35 (5), 593–596.Glaeser, E.L., Saiz, A., 2003. The Rise of the Skilled City. Federal Reserve Bank of

Philadelphia.Glaeser, E.L., Kolko, J., Saiz, A., 2001. Consumer city. J. Econ. Geogr. 1 (1), 27–50.Hoyman, M., Faricy, C., 2009. It takes a village: a test of the creative class, social capital,

and human capital theories. Urban Aff. Rev. 44 (3), 311–333.Marcuse, P., 2003. Review of The Rise of the Creative Class, by Richard Florida. Urban Land

62, 40–41.Markusen, A., 2006. Urban development and the politics of a creative class: evidence from

a study of artists. Environ. Plan. 38 (10), 1921–1940.Melo, P.C., Graham, D.J., Noland, R.B., 2009. A meta-analysis of estimates of urban agglom-

eration economies. Reg. Sci. Urban Econ. 39 (3), 332–342.

17B. Boualam / Regional Science and Urban Economics 46 (2014) 12–17

Montgomery, J., 2005. Beware “the Creative Class”. Creativity and wealth creationrevisited. Local Econ. 20, 337–343.

Ottaviano, G.I., Peri, G., 2006. The economic value of cultural diversity: evidence from UScities. J. Econ. Geogr. 6 (1), 9–44.

Peck, J., 2005. Struggling with the creative class. Int. J. Urban Reg. Res. 29 (4),740–770.

Rappaport, J., 2007. Moving to nice weather. Reg. Sci. Urban Econ. 37 (3), 375–398.Roback, J., 1982. Wages, rents, and the quality of life. J. Polit. Econ. 90 (6), 1257–1278.Rosenthal, S.S., Strange, W.C., 2004. Evidence on the nature and sources of agglomeration

economies. In: Henderson, J.V., Thisse, J.F. (Eds.), Handbook of Regional and UrbanEconomics. Handbook of Regional and Urban Economics, Vol. 4. Elsevier, pp. 2119–2171 (chapter 49).

Saiz, A., 2007. Immigration and housing rents in American cities. J. Urban Econ. 61 (2),345–371.

Sawicki, D., 2003. Review of The Rise of the Creative Class: And How It's Transforming Work,Leisure, Community and Everyday Life, by Richard Florida. J. Am. Plan. Assoc. 69 (1), 90–91.

Shapiro, J.M., 2006. Smart cities: quality of life, productivity, and the growth effects ofhuman capital. Rev. Econ. Stat. 88 (2), 324–335.

Sheppard, S.C., Oehler, K., Benjamin, B., 2006. Buying into Bohemia: The Impact of CulturalAmenities on Property Values. Center for Creative Community Development.

Stock, J., Yogo, M., 2005. Testing forWeak Instruments in Linear IV Regression. CambridgeUniversity Press, New York 80–108.