sign in sign up
<<
Home , Economies of density, Economies of scale

AND IN INDUSTRIES: AN ANALYSIS OF PERSONAL SERVICE INDUSTRIES BASED ON ESTABLISHMENT-LEVEL DATA Masayuki Morikawa*

Abstract—This study aims to empirically investigate the determinants of the policy as a means to increase service sector productivity service industry productivity, such as and economies of density. By using establishment-level data related to personal service substantially depends on aggregate-level international com- industries in Japan, the study estimates the production functions for both parisons and a few case studies. -added and physical measures. In almost all the service indus- The simultaneous occurrence of production and con- tries examined, significant economies of scale and economies of density are observed, wherein productivity increases by 7% to 15% when the sumption is frequently identified as a characteristic of the municipality population density doubles. These findings are confirmed by service industry. This implies that most service industries an estimation in which the measures of physical output are considered cannot stockpile inventories to smooth production and that instead of value added. service industry productivity is affected by demand condi-

tions such as time series fluctuations of demand and spatial Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 difference in demand density. I. Introduction With regard to the time series aspect, by analyzing HIS study empirically investigates the determinants of industry-level time series data and correcting it for unob- T service industry productivity, such as economies of served input utilization, Basu, Fernald, and Kimball (2006) scale and economies of demand density. By using establish- reveal a larger discrepancy between the observed total fac- ment-level data related to personal service industries tor productivity (TFP; standard Solow residual) and the wherein the simultaneity of production and consumption is ‘‘purified’’ change in the nonmanufacturing sec- particularly prominent, the study estimates the production tor than that in the durable sector. In Japan, functions for 10 narrowly defined industries in Japan. I Kawamoto (2005) conducts a similar analysis by using data compare the results of the analyses for the service industries for 1973 to 1988 from the Japan Industrial Productivity with those for the manufacturing and industries in (JIP) database and demonstrates that the influence of the order to highlight the differences between them. input utilization rate on the output in the nonmanufacturing Recent studies emphasize the importance of service industry is three times greater than that in the durable man- industries with regard to a country’s per- ufacturing industry. formance. For example, van Ark, O’Mahony, and Timmer Concerning the spatial aspect, considerable research on (2008) analyze the industry-level productivity growth of agglomeration economies is related to the subjects covered Europe and the United States and conclude that the major in this paper. Rosenthal and Strange (2004), in their repre- factor causing the divergence of aggregate-level productiv- sentative survey in this area, conclude that productivity ity is the difference in service sector productivity. In Japan, increases by 3% to 8% if a ’s population density dou- the productivity of the service sector has become an impor- bles. However, most of the empirical research in this area tant policy agenda. In light of this, the Service Productivity uses municipal-level data1 or manufacturing industry plant- and Innovation for Growth (SPRING) organization and Ser- level data.2 For example, Nakamura (1985), in a representa- vice Engineering Research Center were established to pro- tive study on Japan’s agglomeration economies on produc- mote innovation and productivity in the service sector. tivity, estimates the translog of two- Numerous estimations of production and cost functions digit manufacturing industries by analyzing city-level for the manufacturing sector have been conducted and used aggregated data from 1979. The study shows that the value in the planning of industrial policy. However, in the case of of the of labor productivity with respect to the service industries, empirical studies on fundamental pro- population of is approximately 3.4%; however, the duction structure, such as those on scale economies, have results vary considerably across industries. Tabuchi (1986) been quite limited in number because of the lack of relevant also uses a cross-section of city-level manufacturing data firm- or establishment-level data. Therefore, the planning of from 1980 and conducts regression to explain labor produc- tivity. The result shows that the value of elasticity of labor Received for publication August 7, 2008. Revision accepted for publica- productivity with respect to the population density of a tion September 3, 2009. city ranges from 4% to 8% depending on the specification. * Research Institute of Economy, and Industry. I thank Masahisa Fujita, Kyoji Fukao, Yoko Konishi, Hyeog Ug Kwon, Toshiyuki Matsuura, Takanobu Nakajima, Yoshihiko Nishiyama, Masa- hiko Ozaki, and an anonymous referee for their helpful comments and 1 Ciccone and Hall (1996) is a representative study in the United States. suggestions. I am grateful to the Research and Statistics Department of Kanemoto, Ohkawara, and Suzuki (1996) and Davis and Weinstein the Ministry of Economy, Trade and Industry for providing access to the (2001) are Japanese studies that use municipal-level data. data used in this research. The views expressed in the paper are solely 2 Henderson, Kuncoro, and Turner (1995) and Henderson (2003) are the those of the author and are not necessarily those of the RIETI or the representative studies that use data on manufacturing plants in the United METI. States.

The Review of Economics and Statistics, February 2011, 93(1): 179–192 Ó 2011 by the President and Fellows of Harvard College and the Massachusetts Institute of Technology 180 THE REVIEW OF ECONOMICS AND STATISTICS

Analyses that are focused on nonmanufacturing industries theoretical implications of the empirical results. Section V are rarely conducted, whereas analyses based on service presents the conclusions, including policy implications. industry establishment-level data are almost nonexistent.3 Although the data used in the study concern a specific manufacturing industry, Syverson (2004) investigates the II. Data and Methods effect of spatial substitutability on the productivity of plants. The hypotheses are that consumers in a densely clus- A. Data tered can easily switch suppliers and that inefficient This paper uses data on personal service industries producers may find it difficult to survive. As a result, in obtained from the Survey of Selected Service Industries such markets, the average productivities should be higher conducted by the Ministry of Economy, Trade and Industry and the dispersion of productivities among plants should be (METI). Introduced in 1973, this survey is conducted to lower. Syverson (2004) tests these hypotheses using the clarify the actual conditions of service industries and obtain data concerning ready-mixed concrete plants in the United basic data for policies concerning them. The participants of

States and shows that the results are consistent with the Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 the survey are all establishments operating in Japan. Initi- hypotheses. Further, using data for ready-mixed concrete ally the survey considered only five service industries, but plants, Syverson (2007) demonstrates that dispersion the coverage has gradually been expanded and now covers decreases when spatial increases. This work more than twenty industries. Although the survey is con- also suggests the possibility that the same mechanism plays ducted annually for -leasing and information service a more significant role in service, retail, or wholesale indus- industries, most industries are surveyed every three to four tries. Although the data used in these analyses are not years. The survey compiles information related to establish- directly related to the service sector, the main concept ments or firms according to industry characteristics. The behind these findings is the most relevant element for my questionnaire items cover a variety of issues, and some analyses. items may differ depending on the industry being surveyed. On the basis of this discussion, this study investigates the In this paper, I use recent (2002–2005) establishment- factors influencing the TFP of service industries by analyz- level data concerning ten major personal service industries ing unique establishment-level data related to 10 personal in Japan: movie theaters, golf courses, tennis courts, bowl- service industries in Japan and focuses on the effects of spa- ing alleys, fitness clubs, golf driving ranges, culture centers, tial demand density. The major results can be summarized theaters (including rental halls), wedding ceremony halls, as follows: and aesthetic salons (for example, facial and body massage but not hair and nail salons). Table 1 shows the market sizes In almost all the service industries examined, econo- and numbers of employees of these industries. mies of scale in terms of establishment size and firm size are detected. Almost all the service industries exhibit significant B. Methodology economies of population density. The productivity of an establishment increases by approximately 7% to This paper estimates simple production functions. The 15% when the population density of the city doubles. dependent variables are value added and measures of physi- The effects of demand density on these industries are cal output. The independent variables are labor (number of far greater than those on manufacturing industries. employees), the proxy for tangible capital, the degree of diversification (share of main business sales), the dummy These results hold even if physical output measures, which remain unaffected by price differences among for multiple establishments, the population density of the different cities, are used in the analyses. location city, and the number of industry establishments in the same city: The rest of the paper is organized as follows. Section II ln Yi ¼ b0 þ b1 ln L þ b2 ln K þ b3 Diversification describes the data employed and the methods of analysis. þ b Dummy for Multiple Establishment Section III presents the results of the estimations of the pro- 4 ð1Þ duction functions. Section IV interprets and discusses the þ b5 Population Density

þ b6 Number of Local Establishments:

3 Several papers analyze the productivity of nonmanufacturing sectors The dependent variables (Yi) are value added and the sev- by using geographically aggregated data, for example, Mera (1973) and eral physical output measures. Value added (va) is calcu- Dekle (2002). Both papers suggest a strong relationship between produc- tivity and spatial density in nonmanufacturing industries in Japan. lated as va ¼ sales – costs þ þ rents. Total rents are Although the subject is the effect of density on regional employment the sum of rents for land and buildings and rents for equip- growth, Combes (2000) shows a substantial difference between the results ment. It is preferable to add the user cost of capital ( for manufacturing and service sectors. Recently Graham (2009) estimated and localization economies for two-digit manufacturing and payment) explicitly, but since the value of tangible capital service industries by using data on British firms. stock is not surveyed, the data on interest payment are ECONOMIES OF DENSITY AND PRODUCTIVITY IN SERVICE INDUSTRIES 181

TABLE 1.—SIZE OF THE SELECTED PERSONAL SERVICE INDUSTRIES Number of Annual Sales Number of Industry Year Establishments (billion yen) Employees Movie theaters 2004 2,464 228.6 16,292 Golf courses 2004 2,026 975.8 132,570 Tennis courts 2004 1,531 55.2 14,516 Bowling alleys 2004 948 130.3 16,348 Fitness clubs 2005 1,881 385.8 67,874 Golf driving ranges 2004 2,707 167.5 27,670 Culture centers 2005 698 57.3 55,271 Theaters 2004 698 197.3 12,262 Wedding ceremony halls 2005 2,826 891.1 98,668 Aesthetic salons 2002 5,877 234.3 23,944 Source: Survey of Selected Service Industries (METI).

4 unavailable and impossible to estimate. With regard to the TABLE 2.—MEASURES OF FIXED CAPITAL STOCK Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 physical measures of outputs, data such as the total number Industry Measures of Capital Stock of users of movie theaters per year and the total number of Movie theaters Floor area games at bowling alleys per year are available. Although Golf courses Number of holes these physical output measures are not ideal, they have an Tennis courts Number of courts Bowling alleys Number of lanes advantage in that they remain unaffected by price differ- Fitness clubs Floor area ences among different establishments. I employ these mea- Golf driving ranges Number of boxes sures to compare the results by using the revenue-based out- Culture centers Floor area 5 Theaters Number of seats put measure. Wedding ceremony halls Floor area Among the explanatory variables, L is the number of Aesthetic salons Number of beds employees (emp) including part-time workers. Unfortunately, Source: Survey of Selected Service Industries (METI). the data regarding the number of hours and the measures of labor quality are unavailable. I discuss this issue later. cal economy of scale and includes both -side and In order to correctly estimate the production function, demand-side factors. information of fixed capital stock (K) is essential. However, The share of main business sales (mshare) is calculated the Survey of Selected Service Industries does not collect by dividing the sales of main services by the total sales of this information. Although the survey does collect informa- the establishment. Since the capital stock measures tion of equipment flow, it is impossible to esti- employed here represent only the main business and do not mate capital stock by using the perpetual inventory method cover all fixed capital stocks, the coefficients of capital because the survey is not conducted annually for each stocks in the estimation using value-added as the dependent industry. Nevertheless, the survey contains data on good variable may be biased. Moreover, in regressions using proxies for fixed capital of personal service industries—for physical output measures as the dependent variables, the example, the number of courts for tennis courts, number of coefficients of labor and capital may be biased because the lanes for bowling alleys, number of boxes for golf driving output measures represent only the main business. There- ranges, and number of seats for theaters. Although these fore, this variable (mshare) is used as a control variable. proxies for capital stock do not include buildings or The dummy for multiple establishments (multidum) equals , they incorporate the most important aspects of 1 if the establishment is part of a firm that has more than two each service. Furthermore, since these variables are physi- establishments in the same industry; in other cases, the cal measures of capital, it is not necessary to use price dummy equals 0. The coefficient of this variable indicates the deflators. Moreover, they reflect the measure of an impor- existence of firm-level economies of scale, which are differ- tant capital stock: land. Specific capital stock variables are ent from establishment-level economies of scale. For exam- listed in Table 2. ple, if a firm operates many establishments across Japan, the As is evident, the sum of the coefficients for L and K productivity of the establishments may be higher because of (d b b ) is the measure of scale elasticity. If the sum ¼ 1 þ 2 efficient management of common tasks, common procure- is larger than unity, there is an economy of scale at the 6 ment, or sharing of know-how among establishments. establishment level. This measure is not a pure technologi- The source for the population density figures of the cities is the 2005 Population Census.7 In the estimations, the log 4 Since the Survey of Selected Service Industries does not have infor- mation on capital depreciation, the value added used here is the ‘‘’’ value added. 5 Foster, Haltiwanger, and Syverson (2008) compare physical TFP 6 In order to check for robustness, I conduct regressions excluding measures and revenue-based TFP measures for several narrowly defined mshare and multidum. The estimated establishment-level scale economies manufacturing industries in the United States. This paper indicates that and the coefficients of population density are essentially unaffected. measured productivity dynamics differ significantly according to the TFP 7 Population density data for some establishments are unavailable measures used. because of recent active mergers among municipalities in Japan. 182 THE REVIEW OF ECONOMICS AND STATISTICS of this variable (lnpopdens) is used in order to compare the ferences, are superior to value added in this respect. As results with preceding studies. In the case of personal ser- shown in section III, the coefficients for population density vice industries, population density can be interpreted as are positive and highly significant for almost all industries demand density. The coefficients of this variable indicate even if physical output measures are used as the dependent the output elasticity with respect to demand density. variables. The gap between value added and physical output In addition to the variables already noted, the number of estimation can be interpreted as a bias resulting from price establishments of the same industry in the city (num)is differences among different municipalities. However, phy- included in a supplementary estimation. This variable acts sical output measures have the drawback that they cover as a proxy for the supply-side factors affecting productivity, only the output of the main business of the establishments which include the degree of competition, knowledge spill- and are not available for a specific industry (theater). More- over, sharing of locally common inputs, and so on.8 Although over, it is impossible to compare the results with manufac- the coefficient of this variable itself is important, the effect turing and retail industries that do not have physical output of the inclusion of this variable on the coefficient of popula- measures. Therefore, I present the value-added results first

tion density is also relevant to this study. and then the results concerning physical output measures in Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 Estimations are made by using the latest cross-sectional order to check for robustness. Both regression results are data for each industry. The most recent years for which data shown in the appendix. are available vary by industry. In addition to these regres- With regard to the difference in the quality of output, sions, I estimate by pooling data from two years of observa- establishments located in dense markets may produce more tions. In these estimations, I add a year dummy (yeardum) differentiated or higher-quality services. The data used in for the observations of the latest year. However, since the this study do not contain direct information on the quality results of pooling regressions are essentially the same as of services. However, if establishments in dense markets those of cross-section regressions, I report the regression produce higher-quality services, the true productivity in results of only a single year. such markets is higher than the measured productivity. In this case, the conclusion of the paper will be reinforced. C. Measurement Issues The quality of inputs, particularly the labor quality differ- ence among cities, is an important issue because the quality In estimating economies of density (or agglomeration), of labor (or capital) may influence the measured relation- we must consider various measurement issues. Specifically, ship between density and productivity. Recent research ana- it is important to consider the effects of price differentials lyzing the relationship between density and wages empha- among cities, difference in the quality of outputs and inputs sizes the endogeneity bias or reverse causality. Typically (particularly, worker skills), possible endogeneity bias, and the endogeneity bias arises when workers possessing differ- 9 choice of functional form. In this section, I will briefly dis- ent skills are sorted across cities (Combes, Duranton, & cuss these potential problems. Gobillon, 2008) or the work environment in cities causes Estimations using the value-added as the dependent vari- the workers to be more productive (Glaeser & Mare´, 2001; able may be affected by service price differences among Gould, 2007). Most of the related previous literature dis- cities. If intense competition lowers the markup in densely cusses the causality from density to productivity (Ciccone populated cities, the elasticity of productivity with respect & Hall, 1996, among others). However, population density to density will be lowered. In this case, the coefficient of or labor quality becomes endogenous when regional charac- population density has a downward bias in the estimation of teristics affect both the productivity and the migration of the value-added production function. On the other hand, if people and firms. In this case, ordinary least square (OLS) service are higher in densely populated cities because estimates may show an upward bias.12 Unfortunately, it is of strong demand, the coefficient of population density impossible to control the quality of labor because the estab- 10 exhibits an upward bias. The effect of the price differ- lishment-level data used here do not contain information ences cannot be quantified because the data used here do about labor quality, such as the education of the workers 11 not include establishment-level price information. Physi- and their tenure with the firm.13 However, the analysis in cal output measures, which remain unaffected by price dif- this paper, which is conducted for the narrowly defined industries, is not affected by regional differences in indus- 8 For example, Henderson (2003) shows that the number of plants in trial structure. As a result, the possible bias arising from the same industry and the same region positively affects productivity among high-tech manufacturing industries. 9 Combes, Mayer, and Thisse (2008) provide a useful survey on the effects of output prices, workers’ skill, and endogeneity bias on the esti- 12 Combes, Duranton, and Gobillon (2008) and Mion and Naticchioni mation of local productivity. (2009) estimate the endogenous quality of labor bias by using individual- 10 The difference in factor prices (wages) among cities may bias the level panel data about wages. Bacolod, Blum, and Strange (2009) present results conversely. direct evidence that the relationship between city size and worker skills in 11 According to the 2006 Consumer Price (CPI) for major cities, the United States is positive but weak. the is higher in larger (denser) cities. However, the elasticity 13 The quality of capital can be discussed in a similar fashion. However, of price level with respect to population density is only 0.013. Apparently I have not found any empirical study on the difference in capital quality the price differential relative to the city size is very small in Japan. among firms in cities. ECONOMIES OF DENSITY AND PRODUCTIVITY IN SERVICE INDUSTRIES 183 skill differences among workers in cities will be smaller TABLE 3.—ESTABLISHMENT-LEVEL SCALE ELASTICITY than that in analyses that use the wages of the workers in Industry Scale Elasticity cities as the dependent variable. In addition, as shown in Movie theaters 1.125 section III, personal service industries exhibit considerably Golf courses 1.278 larger economies of population density than manufacturing Tennis courts 1.287 Bowling alleys 1.154 industries do. Unless skill differences among workers in Fitness clubs 1.102 cities are extremely large in service industries as compared Golf driving ranges 1.412 to those in manufacturing industries, the skill differences Culture centers 0.953 Theaters 1.149 are not the chief reason for the observed economies of den- Wedding ceremony halls 1.076 sity in service industries. Aesthetic salons 1.499 Selecting an appropriate functional form has always been (Simple average) 1.204 Retail 1.324 a controversial issue in estimating production functions. Manufacturing 1.163 Although the Cobb-Douglas function is useful because of Scale elasticity indicates the sum of the coefficients for L and K. Area of floor is used as a measure of 14 capital stock for retail, and the value of tangible capital is used for manufacturing.

its simplicity, it is a restrictive functional form. Therefore, Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 I also estimate the translog production function, a represen- tative flexible production function, to check the robustness TABLE 4.—COEFFICIENTS OF MULTIPLE ESTABLISHMENTS DUMMY (MULTIDUM) of the results:15 Industry Coefficients 2 ln Y ¼ a0 þ aL ln L þ aK ln K þð1=2ÞaLLðln LÞ Movie theaters 0.260*** 2 Golf courses 0.115*** þð1=2ÞaKKð ln KÞ þ aLK ln L ln K Tennis courts 0.147** þ b Diversification ð2Þ Bowling alleys 0.112*** 1 Fitness clubs 0.358*** Golf driving ranges 0.083** þ b2 Dummy for Multiple Establishment Culture centers 0.360*** þ b3 Population Density Theaters 0.083 Wedding ceremony halls 0.141*** As shown further, the coefficients of population density Aesthetic salons 0.578*** (Simple average) 0.239 are still highly significant and quite similar in size to the Retail 0.374*** results achieved using the Cobb-Douglas production func- Manufacturing 0.113*** tion. **Significant at 5%. ***Significant at 1%. Simple average excludes theaters.

III. Results As Basu et al. (2006) and Kawamoto (2005) indicated, the measured Solow residuals are affected by mismeasure- A. Economies of Scale ments of the rate, which implies that the The estimated results are presented in the appendix. The estimated scale elasticity is also affected by demand condi- estimated scale elasticity for each industry is shown in tions. For example, the capacity utilization rate of a service Table 3.16 Surprisingly, the figures for all industries except establishment located in a shrinking city is expected to be for culture centers exceed unity; this suggests the existence lower than that of an establishment located in an expanding of establishment-level economies of scale. The simple aver- city. In this case, the scale elasticity may be overestimated. age is approximately 1.2. Most of the economies of scale in Therefore, the estimated figures here should not be inter- service industries are larger than those in average manufac- preted as indicators of a pure technological economy of turing plants.17 scale. The purpose of this study is to estimate not the pure technological factor but the effects of both supply- and 14 Eberts and McMillen (1999) provide a comprehensive survey on the demand-side factors. In the case of service industries, capa- issue of the functional form of the production function for estimating city utilization is the most important factor in determining agglomeration economies. 15 However, using the translog production function occasionally leads productivity because it is impossible to stockpile inventory. to the problem of multicollinearity. Disregarding this factor and focusing on only the pure tech- 16 In the analysis of the retail industry, microdata from the Census of nological factor would cause an essential aspect of the ser- Commerce (2004) are used. Microdata from the Census of Manufactures (2004) are also used for the analysis of the manufacturing industry. vice sector to be overlooked. Although the specifications of the production functions are essentially the After controlling for establishment-level economies of same as the personal service industries, some variations exist because of scale, we observe that the coefficients of the dummy for the differences in the original data. In the estimation for the retail indus- try, the dependent variable is sales (not value added), and the independent multiple establishments (multidum) indicate the existence variables include a dummy for self-service and dummies for a two-digit of firm-level economies of scale. Table 4 summarizes the industry. In the case of the manufacturing industry, explanatory variables results according to the industry. For fitness clubs, culture do not include the measure of diversification. 17 The corresponding figure for manufacturing plants is approximately centers, and aesthetic salons, the coefficients exceed 0.3 1.16 (estimated using data from the Census of Manufactures for 2004). and are statistically significant. This implies that, ceteris 184 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 5.—ECONOMY OF POPULATION DENSITY:COEFFICIENTS OF LNPOPDENS TABLE 6.—EFFECT OF POPULATION DENSITY,INCLUDING INDIRECT EFFECT Industry Coefficients Direct Indirect Total Industry Effect Effect Effect Movie theaters 0.132*** Golf courses 0.141*** Movie theaters 0.132 0.018 0.150 Tennis courts 0.191*** Golf courses 0.141 0.032 0.174 Bowling alleys 0.096*** Tennis courts 0.191 0.048 0.239 Fitness clubs 0.103*** Bowling alleys 0.096 0.034 0.130 Golf driving ranges 0.200*** Fitness clubs 0.103 0.023 0.126 Culture centers 0.166*** Golf driving ranges 0.200 0.095 0.295 Theaters 0.433*** Culture centers 0.166 0.006 0.160 Wedding ceremony halls 0.081*** Theaters 0.433 0.015 0.448 Aesthetic salons 0.088*** Wedding ceremony halls 0.081 0.009 0.091 (Simple average) 0.133 Aesthetic salons 0.088 0.047 0.135 Retail 0.048*** (Simple average) 0.133 0.033 0.167 Manufacturing 0.027*** Retail 0.048 0.014 0.063 ***Significant at 1%. The theaters where the direct effect is extremely large are excluded from the cal- Manufacturing 0.027 0.000 0.027 culation of simple average. The total effect is the sum of the direct effect of population density and the indirect effect in the form of the scale effect. The theaters where the direct effect is extremely large are excluded from the calcula- Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 tion of simple average. paribus, the productivity of a single establishment of a mul- tiple-establishment firm is more than 30% higher than that economies of scale. In fact, the denser the population, the of a single-establishment firm. For golf courses, tennis larger is the average size of establishments in all ten indus- courts, bowling alleys, and wedding ceremony halls, the tries. In order to determine the magnitude of these indirect coefficients are around 0.10 to 0.15; this also indicates sig- effects of population density, I calculate the elasticity of the nificant firm-level economies of scale. These results reflect average size of establishments with respect to population the fact that large firms are efficient in managing common density and multiply the figure by the scale elasticity (d –1). tasks or sharing know-how among establishments, suggest- Table 6 shows the total effects of population density as the ing that the expansion of superior firms into several busi- sum of the direct productivity effect and indirect effect ness domains or chain-type operations may have a positive 18 through the scale economy. Although the indirect effects influence on the productivity growth of service industries. are smaller in magnitude than the direct effects, they increase the total effect of population density on productiv- B. Economies of Demand Density ity by 1% to 7%. As a result, by considering this indirect effect, the difference between service industries and manu- The coefficients of population density (lnpopdens) are facturing industries increases.20 the focus of this study. Table 5 displays the summary of the Syverson (2004) argues that in a denser market, higher results. In most industries, the elasticity of output with substitutability among producers truncates productivity dis- respect to population density is around 0.1 to 0.2, and the tribution from the lower end, resulting in higher minimum differences among industries are small. This implies that and average productivity levels as well as less productivity when the population density of a city doubles, the produc- dispersion. To comprehend this, I calculate the index of tivity increases from 7% to 15%. The coefficients for the productivity dispersion (P90/P10, P75/P25, and log var- personal service industries are considerably larger than iance) individually for cities with high population density 19 those for the manufacturing and retail industries. Accord- and those with low population density (Table 7). The results ing to a representative survey by Rosenthal and Strange differ by industry. For golf courses, fitness clubs, and aes- (2004), doubling the city size increases the productivity of thetic salons, the dispersion is larger in cities with low establishments in the city from approximately 3% to 8%. population density; this is consistent with Syverson’s The figures for the service industries estimated in this case (2004) result for the ready-mixed concrete industry in the are larger than those estimated by the ‘‘consensus,’’ wherein United States. However, for theaters and wedding ceremony aggregate-level data or manufacturing industry data were halls, the productivity dispersion is larger in densely popu- used. lated cities. Apparently Syverson’s (2004) argument cannot Because the average size of establishments may be large be generalized to service industries as a whole. From an in densely populated cities, such cities also benefit from unprejudiced perspective, it can be inferred that while pro- ductivity distributions are not necessarily truncated, the 18 For multiple establishments, measured productivity may be upward overall is higher in densely populated cities. biased simply because common management and administration costs are borne by the headquarters. To address this issue, I conduct analyses by Next, the number of establishments of the same industry dividing the multiple establishments dummy into a headquarters dummy (num) is added as a right-hand-side variable. The coefficients and a branch dummy. According to the result, although the statistical sig- nificance is lower in case of the headquarters, it generally still is positive and significant. 20 In the case of manufacturing, there is no significant relationship 19 The result for the manufacturing industry (0.027) is similar to that of between the population density of the location and the average size of Nakamura (1985) and lesser in value than that of Tabuchi (1986). plants. ECONOMIES OF DENSITY AND PRODUCTIVITY IN SERVICE INDUSTRIES 185

TABLE 7.—DISTRIBUTION OF TFP BY POPULATION DENSITY P75-P25 P90-P10 Log Variance Industry Low High Low High Low High Movie theaters 0.960 0.868 1.849 1.879 0.675 1.003 Golf courses 0.651 0.571 1.383 1.286 0.419 0.333 Tennis courts 1.027 1.101 2.231 2.131 1.220 0.956 Bowling alleys 0.624 0.600 1.238 1.297 0.339 0.297 Fitness clubs 0.783 0.769 1.571 1.475 0.526 0.432 Golf driving ranges 0.855 0.904 1.734 1.695 0.596 0.503 Culture centers 1.337 1.228 2.242 2.566 0.952 1.155 Theaters 1.401 1.613 3.282 3.585 2.048 2.646 Wedding ceremony halls 0.676 0.743 1.471 1.593 0.434 0.500 Aesthetic salons 1.037 0.974 2.131 1.984 0.859 0.754 Retail 1.204 1.198 2.495 2.486 1.571 1.756 Manufacturing 0.728 0.721 1.509 1.515 0.456 0.485 Low-population-density cities and high-population-density cities are classified by the average population density. Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021

TABLE 8.—EFFECT OF THE NUMBER OF ESTABLISHMENTS IN THE SAME CITY TABLE 9.—COEFFICIENTS OF CITY AND PREFECTURE POPULATION DENSITY Coefficients Coefficients Industry num lnpopdens Industry City Population Prefecture Movie theaters 0.030*** 0.087*** Movie theaters 0.0874** 0.0609 Golf courses 0.005 0.141*** Golf courses 0.0604*** 0.1916*** Tennis courts 0.004 0.188*** Tennis courts 0.2084*** 0.0278 Bowling alleys 0.025* 0.102*** Bowling alleys 0.0773*** 0.0285 Fitness clubs 0.006 0.098*** Fitness clubs 0.0587*** 0.0634*** Golf driving ranges 0.004 0.197*** Golf driving ranges 0.1552*** 0.0742*** Culture centers 0.021 0.175*** Culture centers 0.1106*** 0.0867** Wedding ceremony halls 0.007*** 0.066*** Theaters 0.3288*** 0.1511 Aesthetic salons 0.003*** 0.075*** Wedding ceremony halls 0.0719*** 0.0177 Retail 0.000*** 0.038*** Aesthetic salons 0.0837*** 0.0062 Manufacturing 0.000*** 0.025*** Retail 0.0464*** 0.0044*** Theaters, where many establishments do not have competiting establishments in the same city, are Results from a regression in which both city and prefecture population densities are included as inde- excluded. *Significant at 10%. ***Significant at 1%. pendent variables. **Significant at 5%. ***Significant at 1%.

of this variable and those of population density, along with six industries (movie theaters, tennis courts, bowling alleys, this additional variable, are indicated in Table 8. The coeffi- theaters, wedding ceremony halls, and aesthetic salons), the cients of the number of establishments are mostly positive coefficients for population density of the city are significant, (though frequently insignificant), which suggests that the but those of the prefecture are not. For four industries (golf number of establishments in direct competition has a posi- courses, fitness clubs, golf driving ranges, and culture cen- tive effect on productivity. Particularly for movie theaters, ters), both city population density and prefecture population wedding ceremony halls, and aesthetic salons, the coeffi- density are positive and significant, but the values of the cients of the number of establishments in the city are posi- coefficients of the city population are greater for golf driving tive and significant. However, even after controlling for this ranges and culture centers. Barring a few exceptions, these variable (for example, localization economies from knowl- results demonstrate that most of the markets of personal ser- edge or efficiency arising from intense competi- vice industries are geographically localized. tion), the values of the coefficients of population density are not significantly affected. C. Estimations Using Physical Output Measures Overall, these results suggest that the agglomeration economies of service industries mainly arise from demand I have thus far used value added as an explanatory vari- density. able for the estimation of the production function. The Sur- I have used cities as units for calculating population den- vey of Selected Service Industries provides information on sity. However, the appropriate areas for analysis may differ physical output measures for most of the personal service according to the industry. For some industries, consumers industries. Since the estimations of the production function may have to visit a distant establishment in order to find a in this paper are based on a narrowly defined industry, these service. To verify this, I simultaneously use the population physical output measures can be used as dependent vari- density of the city and prefecture (equivalent to a state in the ables. In the analyses based on value added, price differ- United States) as an explanatory variable. The corresponding ences among cities may overstate the effect of population results for population density are presented in Table 9. For density. Generally the larger the size of the city, the higher 186 THE REVIEW OF ECONOMICS AND STATISTICS

TABLE 10.—RESULTS USING PHYSICAL OUTPUT MEASURES Establishment-Level Scale Elasticities Coefficients of multidum Coefficients of lnpopdens Revenue-Based Physical Output Revenue-Based Physical Output Revenue-Based Physical Output Industry Models Models Models Models Models Models Movie theaters 1.125 1.076 0.260 0.452 0.132 0.122 Golf courses 1.278 0.964 0.115 0.071 0.141 0.119 Tennis courts 1.287 1.002 0.147 0.463 0.191 0.222 Bowling alleys 1.154 1.272 0.112 0.123 0.096 0.071 Fitness clubs 1.102 1.185 0.358 0.412 0.103 (0.025) Golf driving ranges 1.412 1.427 0.083 0.133 0.200 0.116 Culture centers 0.953 1.020 0.360 (0.154) 0.166 0.080 Wedding ceremony halls 1.076 0.933 0.141 (0.052) 0.081 0.181 Aesthetic salons 1.499 1.240 0.578 0.315 0.088 0.066 Figures in parentheses are insignificant at the 10% level. Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 is the price level. For example, Foster, Haltiwanger, and industries (bowling alleys, golf driving ranges, and culture Syverson (2008) compare physical TFP and revenue-based centers), the elasticity almost halves in estimations using TFP for several U.S. manufacturing plants that produce physical output measures. For these three industries, the homogeneous products. They find that revenue-based TFP higher service prices in densely populated large cities is positively correlated with the product prices at plant slightly overstate the effects of population density on pro- level. In order to prevent this bias and to check the robust- ductivity estimations in which value added is used as an out- ness of the results, I conduct the analyses by using physical put measure. However, even after resolving this bias, these output measures. industries exhibit significant economies of demand density. The output measures used in the analyses are the total The size of the estimated coefficient is larger when a physi- number of users (lnuser) for movie theaters, tennis courts, cal output measure is used for wedding ceremony halls. golf driving ranges, fitness clubs, and aesthetic salons; total To summarize, almost all of the results obtained by using number of wedding parties (lnuser) for wedding ceremony physical output measures correspond to the results obtained halls; total number of games played (lngame) for bowling by using value added. Although the effects of higher service alleys; and total number of participants multiplied by the prices in large cities are observed in some service indus- terms of the classes (lnuserterm) for culture centers. All tries, these effects are not a dominant factor affecting these measures are for an annual basis and are expressed in economies of density. logarithmic form.21 Table 10 is a summary of the major results. Here, I report D. Estimation Using Translog Production Function only a simple form of regression: the number of establish- ments in a city (num) is not added as an explanatory vari- As discussed in section II, selecting an appropriate func- able. The estimated scale elasticity exceeds unity for seven tional form is always an issue in estimating the production industries except for golf courses and wedding ceremony function. In order to check the robustness of the results halls. Most of the service industries still exhibit economies obtained by using the Cobb-Douglas function, estimated of scale at the establishment level. The coefficients of the coefficients of population density obtained by using the dummy of multiple establishments (multidum) are also translog production function are presented in Table 11. For positive and significant for eight industries, except for cul- both the instances of value-added measures and physical ture centers. This also confirms the results obtained by output measures being used as dependent variables, not using revenue-based value added as the dependent variable. only are the coefficients of population density significant in The reason for the higher productivity of larger establish- most cases, but also the values of the coefficients are similar ments or larger firms is not that their service prices are in magnitude to the results of the Cobb-Douglas specifica- higher but that their physical output quantities per input are tion.22 larger. The coefficients for population density are positive and significant for eight industries, except for fitness clubs. IV. Interpretation and Discussion A careful observation of the output elasticity with respect to population density reveals that in four industries (movie The results presented in the previous section indicate a theaters, golf courses, tennis courts, and aesthetic salons), strong and positive correlation between service sector pro- the results are similar in magnitude for estimations wherein ductivity and population density. Theoretically, economies value-added and physical output measures are used. In three 22 The multicollinearity problem arises in the estimation of translog production function because explanatory variables are highly correlated. 21 The appropriate physical output measure is not available for the case Therefore, occasionally the coefficients of labor, capital, their square of theaters only. terms, and the interaction term are imprecisely estimated. ECONOMIES OF DENSITY AND PRODUCTIVITY IN SERVICE INDUSTRIES 187

TABLE 11.—COEFFICIENTS OF POPULATION DENSITY USING TRANSLOG PRODUCTION FUNCTION Revenue-Based Models Physical Output Models Industry Cobb-Douglas Translog Cobb-Douglas Translog Movie theaters 0.132 0.142 0.122 0.123 Golf courses 0.141 0.147 0.119 0.121 Tennis courts 0.191 0.191 0.222 0.216 Bowling alleys 0.096 0.097 0.071 0.067 Fitness clubs 0.103 0.108 (0.025) 0.042 Golf driving ranges 0.200 0.200 0.116 0.115 Culture centers 0.166 0.165 0.080 (0.047) Theaters 0.433 0.406 – – Wedding ceremony halls 0.081 0.072 0.181 0.164 Aesthetic salons 0.088 0.088 0.066 0.067 Figures in parentheses are insignificant at the 10% level. Explanatory variables are mshare, multidum, and lnpopdens. Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 of density may arise from various sources. In this section, I spillover internal to the industry. However, the values of interpret the results from a theoretical perspective. the coefficients are relatively lower and the coefficients of Traditionally the typical sources of the economy of den- population density remain fairly unaltered by the inclusion sity (or agglomeration) are internal scale economies in pro- of this variable. Furthermore, when several measures of duction and positive . The positive externalities productivity dispersion between densely populated and are often classified into localization economies (or Mar- sparsely populated cities are compared (see Table 7), the shall-Arrow-Romer economies), that is, economies that are dispersion is not necessarily less in densely populated cities, external to the establishment but internal to the industry and as Syverson (2004) suggested. In summary, although the urbanization economies (or Jacobs economies), which are analyses do not disregard the selection or competition or external to both the establishment and industry but internal the effect as a possible source of the to densely populated areas. These externalities arise from observed economies of density, the impacts of these labor market pooling, input sharing, and knowledge spill- mechanisms are not quantitatively significant. overs (see Duranton & Puga, 2004; Moretti, 2004; Rosenthal What are the sources of the remaining significant effects & Strange, 2004; Strange, 2009). In addition, a third possi- of population density on productivity? Although this analy- ble source of the positive relationship between density and sis cannot directly test this hypothesis, a possible source is productivity is the selection mechanism through intense the effect of urbanization economies—knowledge or human competition (Syverson, 2004). capital spillover—which is not specific to an industry. These alternative explanations are not mutually exclu- Another possible reason is the effect of demand. It is impor- sive, and formally testing the competing theories is beyond tant to note that the coefficients of density are considerably the scope of this study. However, the empirical results pre- larger for service industries than for manufacturing indus- sented here illustrate the relative importance of alternative tries. Strong local demand results in a high-input utilization theoretical arguments. rate and efficient production planning. This may explain With regard to internal economies of scale arising from why economies of density are larger in service industries fixed costs or other sources, this paper directly estimates the where simultaneous production and consumption is promi- establishment-level scale elasticity, which simultaneously nent. From the perspective of evaluating TFP as pure tech- results in the estimation of the direct effect of population nological efficiency, this may be a measurement issue. density on productivity. As shown in Table 6, the effects of However, in most service industries, where stockpiled internal scale economies on productivity arising from local inventory does not exist, efficient input utilization is the population density (‘‘indirect effects’’) are generally posi- most important factor affecting the productivity and profit- tive, but their magnitude is considerably smaller than that ability of firms and establishments. To overlook this practi- of the direct productivity effects of population density. Spe- cally important aspect and to focus on only pure technologi- cifically, although internal economies of scale contribute to cal efficiency would imply disregarding the fundamental the productivity of service industries in densely populated question to be resolved. cities, their economic significance is not large enough to Various potential channels lead to the observed relation- justify the total effects of density. ship between density and service sector productivity. In When the number of establishments of the same industry addition to the explanation already offered, difference in (num) is added as an independent variable, the coefficients labor quality or in public infrastructure among cities may of this variable are generally positive and statistically sig- also be related. Although the analysis in this paper cannot nificant for some industries (see Table 8). This suggests the completely distinguish between these alternative explana- existence of the selection or competition effect or the tions, the crux is that service industries exhibit significant effects of localization economies, such as spatial knowledge economies of density; this is not observed in the existing 188 THE REVIEW OF ECONOMICS AND STATISTICS literature, which focuses on only the manufacturing indus- This paper is an attempt to analyze the establishment- try. While considering the importance of the service sector level productivity of personal service industries, which has in advanced economies, it is important to accumulate both previously not been studied. However, there are various empirical and theoretical researches on the productivity of limitations related to data: the data quality issue on capital service industries. This paper is the first step toward that stock measure, lack of information of labor quality and direction. working hours, and difficulty in constructing panel data, among others. This paper analyzes only a small number of V. Conclusion personal service industries. There is an urgent need for the government statistical agency to expand the statistics for This paper empirically investigates the determinants of the establishment-level service industry. service industry productivity, such as economies of scale and economies of density in Japan by analyzing unique Japanese microdata related to personal service industries. REFERENCES

Specifically, by using establishment-level data from the Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 Survey of Selected Service Industries, I estimate the cross- Bacolod, Marigee, Bernardo S. Blum, and William C. Strange, ‘‘Skills in sectional production functions for ten industries. the City,’’ Journal of Urban Economics 65:2 (2009), 136–153. Basu, Susanto, John G. Fernald, and Miles S. Kimball, ‘‘Are Technology The major findings obtained from the analyses are as fol- Improvements Contractionary?’’ American Economic Review 96:5 lows: (2006), 1418–1448. Ciccone, Antonio, and Robert E. Hall, ‘‘Productivity and the Density of Economic Activity,’’ American Economic Review 86:1 (1996), 54– In almost all the service industries examined, establish- 70. ment- and firm-level economies of scale are detected. Combes, Pierre-Philippe, ‘‘Economic Structure and Local Growth: The sizes of the effects of these economies are signifi- France, 1984–1993,’’ Journal of Urban Economics 47:3 (2000), 329–355. cant. Combes, Pierre-Philippe, Gilles Duranton, and Laurent Gobillon, ‘‘Spatial In almost all the service industries examined, signifi- Disparities: Sorting Matters!’’ Journal of Urban Economics cant economies of population density are observed, 63:2 (2008), 723–742. Combes, Pierre-Philippe, Thierry Mayer, and Jacques-Franc¸ois Thisse, where the productivity increases from 7% to 15% Economic Geography: The Integration of Regions and Nations when municipality population density doubles. The (Princeton, NJ: Princeton University Press, 2008). values of these coefficients are substantially greater Davis, Donald R., and David E. Weinstein, ‘‘Market Size, Linkages, and Productivity: A Study of Japanese Regions,’’ NBER working paper than those observed in manufacturing industries for no. 8518 (2001). which sales destinations are far less restricted geogra- Dekle, Robert, ‘‘Industrial Concentration and Regional Growth: Evidence phically. This demonstrates the importance of demand from the Prefectures,’’ this REVIEW 84:2 (2002), 310–315. Duranton, Gilles, and Diego Puga, ‘‘Micro-Foundations of Urban density in the productivity of service industries. Agglomeration Economies’’ (pp. 2063–2117), in J. V. Henderson The above findings are confirmed by an estimation and J. F. Thisse (Eds.), Handbook of Regional Science and Urban wherein measures of physical output are used instead Economics, Vol. 4 (Amsterdam: Elsevier, 2004). Eberts, Randall W., and Daniel P. McMillen, ‘‘Agglomeration Economies of the measure of value added. and Urban Public Infrastructure’’ (pp. 1455–1495) in E. S. Mills and P. Cheshire (Eds.) Handbook of Regional Science and Urban Economics, Vol. 3 (Amsterdam: Elsevier, 1999). These findings suggest the possibility that consolidation Foster, Lucia, John Haltiwanger, and Chad Syverson, ‘‘Reallocation, Firm and expansion at the establishment level, as well as multi- Turnover, and Efficiency: Selection on Productivity or Profitabil- store and chain store operations at the firm level, may help ity?’’ American Economic Review 98:1 (2008), 394–425. Glaeser, Edward L., and David C. Mare´, ‘‘Cities and Skills,’’ Journal of improve the productivity of personal service industries. Labor Economics, 19:2 (2001), 316–342. Creation of a densely populated area is also suggested, as Gould, Eric D., ‘‘Cities, Workers, and Wages: A Structural Analysis of this would have a positive effect on productivity. However, the Urban Wage Premium,’’ Review of Economic Studies 74:2 (2007), 477–506. this may involve a trade-off between productivity improve- Graham, Daniel J., ‘‘Identifying Urbanization and Localization External- ment and other social or economic policy goals. ities in Manufacturing and Service Industries,’’ Papers in Regional The population of Japan is declining. The long-term Science 88:1 (2009), 63–84. Henderson, Vernon, ‘‘Marshall’s Scale Economies,’’ Journal of Urban population projection (National Institute of Population and Economics 53:1 (2003), 1–28. Social Security Research) estimates a 13.4% decrease over Henderson, Vernon, Ari Kuncoro, and Matt Turner, ‘‘Industrial Develop- the next thirty years and a 29.6% decrease over the next ment in Cities,’’ Journal of 103:5 (1995), 1067–1090. fifty years. This may work to reduce the productivity of the Kanemoto, Yoshitsugu, Toru Ohkawara, and Tsutomu Suzuki, ‘‘Agglom- service sector arising from the decline in population den- eration Economies and a Test for Optimal City Sizes in Japan,’’ sity. It is important for urban planning to consider the Journal of the Japanese and International Economies 10:4 (1996), 379–398. impact of population distribution on service sector produc- Kawamoto, Takuji, ‘‘What Do the Purified Solow Residuals Tell Us about tivity. In order to enhance the productivity of the service Japan’s Lost Decade?’’ Monetary and Economic Studies 23:1 sector, some trade-offs with other values, such as regionally (2005), 113–148. Mera, Koichi, ‘‘Regional Production Functions and Social Overhead balanced growth or the survival of existing firms and estab- Capital: An Analysis of the Japanese Case,’’ Regional and Urban lishments, are inevitable. Economics 3:2 (1973), 157–186. ECONOMIES OF DENSITY AND PRODUCTIVITY IN SERVICE INDUSTRIES 189

Mion, Giordano, and Paolo Naticchioni, ‘‘The Spatial Sorting and Match- Strange, William C., ‘‘Viewpoint: Agglomeration Research in the Age of ing of Skills and Firms,’’ Canadian Journal of Economics 42:1 Disaggregation,’’ Canadian Journal of Economics 42:1 (2009), 1– (2009), 28–55. 27. Moretti, Enrico, ‘‘Human Capital Externalities in Cities’’ (pp. 2243–2291) Syverson, Chad, ‘‘ and Productivity: A Concrete in J. V. Henderson and J. F. Thisse (Eds.), Handbook of Regional Example,’’ Journal of Political Economy 112:6 (2004), 1181– Science and Urban Economics, Vol. 4 (Amsterdam: Elsevier, 1222. 2004). _____ ‘‘Prices, Spatial Competition and Heterogeneous Producers: An Nakamura, Ryohei, ‘‘Agglomeration Economies in Urban Manufacturing Empirical Test,’’ Journal of Industrial Economics 55:2 (2007), Industries: A Case of Japanese Cities,’’ Journal of Urban Econom- 197–222. ics 17:1 (1985), 108–124. Tabuchi, Takatoshi, ‘‘Urban Agglomeration, Capital Augmenting Tech- Rosenthal, Stuart S., and William C. Strange, ‘‘Evidence on the Nature nology, and Labor Market Equilibrium,’’ Journal of Urban Eco- and Sources of Agglomeration Economies’’ (pp. 2119–2171) in nomics 20:2 (1986), 211–228. J. V. Henderson and J. F. Thisse (Eds.) Handbook of Regional van Ark, Bart, Mary O’Mahony, and Marcel P. Timmer, ‘‘The Productiv- Science and Urban Economics, Vol. 4 (Amsterdam: Elsevier, ity Gap between Europe and the United States: Trends and 2004). Causes,’’ Journal of Economic Perspectives 22:1 (2008), 25–44. Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021

APPENDIX

Regression Results

MOVIE THEATERS (1) (2) (3) (4) Value Added Value Added Physical Physical lnemp 0.7899*** 0.8078*** 0.6753*** 0.6909*** (15.83) (16.20) (15.69) (15.99) lnfloor 0.3355*** 0.3345*** 0.4011*** 0.3988*** (7.20) (7.23) (10.13) (10.12) mshare 1.6429*** 1.6576*** 0.3452 0.3419 (5.12) (5.21) (1.27) (1.26) multidum 0.2601*** 0.2249*** 0.4516*** 0.4250*** (3.16) (2.73) (6.12) (5.74) lnpopdens 0.1321*** 0.0871*** 0.1221*** 0.0865*** (5.25) (3.03) (5.45) (3.36) num – 0.0296*** 0.0233*** (3.18) (2.78) _cons 4.3478*** 4.5578*** 5.2351*** 5.3987*** (10.39) (10.83) (14.75) (15.08) Number of observations 641 641 686 686 Adjusted R2 0.6845 0.6889 0.7144 0.7172 t-values in parentheses. ***Significant at 1%.

GOLF COURSES (1) (2) (3) (4) Value Added Value Added Physical Physical lnemp 0.8210*** 0.8197*** 0.2449*** 0.2439*** (31.10) (31.03) (13.09) (13.03) lnhole 0.4565*** 0.4546*** 0.7193*** 0.7183*** (7.30) (7.27) (16.12) (16.10) mshare 0.3560** 0.3591** 0.0354 0.0326 (2.52) (2.55) (0.35) (0.32) multidum 0.1151*** 0.1088*** 0.0711*** 0.0669*** (3.56) (3.33) (3.10) (2.88) lnpopdens 0.1413*** 0.1410*** 0.1189*** 0.1186*** (10.90) (10.88) (12.83) (12.80) num – 0.0045 0.0030 (1.32) (1.22) _cons 4.6317*** 4.6304*** 6.5038*** 6.5038*** (22.04) (22.04) (43.21) (43.21) Number of observations 1,389 1,389 1,437 1,437 Adjusted R2 0.6254 0.6256 0.4758 0.4761 t-values in parentheses. **Significant at 5%. ***Significant at 1%. 190 THE REVIEW OF ECONOMICS AND STATISTICS

TENNIS COURTS (1) (2) (3) (4) Value Added Value Added Physical Physical lnemp 1.0013*** 1.0023*** 0.3738*** 0.3754*** (37.90) (37.85) (10.68) (10.71) lncourt 0.2857*** 0.2849*** 0.6280*** 0.6269*** (7.11) (7.09) (11.93) (11.90) mshare 0.6451*** 0.6444*** 0.0843 0.0834 (6.14) (6.13) (0.61) (0.60) multidum 0.1465** 0.1466** 0.4626*** 0.4625*** (2.34) (2.34) (5.65) (5.65) lnpopdens 0.1911*** 0.1877*** 0.2222*** 0.2153*** (8.82) (8.31) (7.76) (7.22) num – 0.0035 0.0073 (0.54) (0.83) _cons 4.0330*** 4.0395*** 4.9863*** 4.9990*** (22.17) (22.15) (20.85) (20.86) Number of observations 1,314 1,314 1,310 2,226 Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 Adjusted R2 0.6569 0.6568 0.3070 0.3068 t-values in parentheses. **Significant at 5%. ***Significant at 1%.

BOWLING ALLEYS (1) (2) (3) (4) Value Added Value Added Physical Physical lnemp 0.7731*** 0.7771*** 0.4783*** 0.4784*** (21.64) (21.73) (12.19) (12.15) lnlane 0.3805*** 0.3809*** 0.7936*** 0.7936*** (7.12) (7.14) (13.43) (13.42) mshare 0.5811*** 0.5754*** 0.7031*** 0.7032*** (5.57) (5.52) (6.11) (6.10) multidum 0.1117*** 0.1129*** 0.1230*** 0.1230*** (2.85) (2.88) (2.84) (2.84) lnpopdens 0.0962*** 0.1016*** 0.0709*** 0.0709*** (6.64) (6.86) (4.43) (4.33) num – 0.0252* 0.0002 (1.72) (0.01) _cons 5.1502*** 5.1494*** 6.9348*** 6.9348*** (30.21) (30.24) (37.03) (37.01) Number of observations 861 861 864 864 Adjusted R2 0.7037 0.7044 0.5466 0.5461 t-values in parentheses. *Significant at 10%. ***Significant at 1%.

FITNESS CLUBS (1) (2) (3) (4) Value Added Value Added Physical Physical lnemp 0.6813*** 0.6818*** 0.4929*** 0.4921*** (31.32) (31.35) (15.94) (15.93) lnfloor 0.4211*** 0.4216*** 0.6925*** 0.6915*** (19.14) (19.17) (22.12) (22.10) mshare 1.1076*** 1.1007*** 1.1483*** 1.1367*** (12.29) (12.20) (8.93) (8.83) multidum 0.3575*** 0.3579*** 0.4121*** 0.4116*** (9.30) (9.32) (7.52) (7.52) lnpopdens 0.1034*** 0.0978*** 0.0252 0.0348** (8.78) (7.94) (1.51) (1.98) num – 0.0062 0.0105* (1.53) (1.83) _cons 3.7211*** 3.7181*** 2.8304*** 2.8358*** (24.34) (24.33) (13.02) (13.05) Number of observations 1,832 1,832 1,847 1,847 Adjusted R2 0.8038 0.8040 0.6805 0.6810 t-values in parentheses. *Significant at 10%. **Significant at 5%. ***Significant at 1%. ECONOMIES OF DENSITY AND PRODUCTIVITY IN SERVICE INDUSTRIES 191

GOLF DRIVING RANGES (1) (2) (3) (4) Value Added Value Added Physical Physical lnemp 0.6209*** 0.6210*** 0.4309*** 0.4309*** (24.38) (24.38) (17.26) (17.26) lnbox 0.7915*** 0.7916*** 0.9965*** 0.9967*** (26.15) (26.16) (33.57) (33.57) mshare 1.1364*** 1.1363*** 0.4662*** 0.4663*** (12.48) (12.48) (5.21) (5.21) multidum 0.0829** 0.0828** 0.1330*** 0.1330*** (2.38) (2.37) (3.89) (3.88) lnpopdens 0.1998*** 0.1974*** 0.1164*** 0.1143*** (18.45) (17.80) (10.95) (10.49) num – 0.0037 0.0033 (1.01) (0.91) _cons 3.1764*** 3.1747*** 4.1539*** 4.1524*** (24.91) (24.89) (33.28) (33.27) Number of observations 2,299 2,299 2,330 2,330 Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 Adjusted R2 0.7253 0.7253 0.6880 0.6880 t-values in parentheses. **Significant at 5%. ***Significant at 1%.

CULTURE CENTERS (1) (2) (3) (4) Value Added Value Added Physical Physical lnemp 0.5982*** 0.5975*** 0.3364*** 0.3366*** (17.85) (17.83) (6.94) (6.94) lnfloor 0.3548*** 0.3579*** 0.6833*** 0.6822*** (9.62) (9.68) (12.92) (12.86) mshare 1.9895*** 1.9934*** 0.9088*** 0.9102*** (17.34) (17.37) (5.40) (5.40) multidum 0.3597*** 0.3610*** 0.1543 0.1548 (4.14) (4.16) (1.22) (1.22) lnpopdens 0.1664*** 0.1747*** 0.0798* 0.0768* (5.68) (5.80) (1.86) (1.74) num 0.0210 0.0075 (1.17) (0.29) _cons 3.9476*** 3.9411*** 2.7933*** 2.7957*** (12.68) (12.67) (6.21) (6.21) Number of observations 665 665 690 690 Adjusted R2 0.6346 0.6348 0.3473 0.3464 t-values in parentheses. *Significant at 10%. ***Significant at 1%.

THEATERS,INCLUDING RENTAL HALLS (1) (2) Value Added Value Added lnemp 0.5898*** 0.5449*** (6.09) (5.83) lnseat 0.5592*** 0.6091*** (5.02) (5.67) mshare 1.1860*** 1.2815*** (3.69) (4.15) multidum 0.0827 0.1099 (0.39) (0.54) lnpopdens 0.4333*** 0.2690*** (6.06) (3.51) num – 0.0972*** (4.78) _cons 0.3298 0.9879 (0.32) (0.99) Number of observations 266 266 Adjusted R2 0.4352 0.479 t-values in parentheses. ***Significant at 1%. 192 THE REVIEW OF ECONOMICS AND STATISTICS

WEDDING CEREMONY HALLS (1) (2) (3) (4) Value Added Value Added Physical Physical lnemp 0.8763*** 0.8711*** 0.6132*** 0.6083*** (58.54) (58.21) (31.11) (30.82) lnfloor 0.1996*** 0.2014*** 0.3198*** 0.3214*** (10.35) (10.48) (12.57) (12.65) mshare 0.1513*** 0.1527*** 2.8085*** 2.8077*** (3.42) (3.46) (48.17) (48.23) multidum 0.1405*** 0.1409*** 0.0516 0.0520 (5.07) (5.10) (1.41) (1.43) lnpopdens 0.0814*** 0.0659*** 0.1811*** 0.1668*** (9.42) (7.10) (15.92) (13.61) num – 0.0065*** 0.0060*** (4.45) (3.12) _cons 4.7188*** 4.7751*** 2.8091*** 2.7563*** (41.35) (41.74) (18.66) (18.22) Number of observations 2,693 2,693 2,714 2,714 Downloaded from http://direct.mit.edu/rest/article-pdf/93/1/179/1919143/rest_a_00065.pdf by guest on 28 September 2021 Adjusted R2 0.7332 0.7350 0.6608 0.6619 t-values in parentheses. ***Significant at 1%.

AESTHETIC SALONS (1) (2) (3) (4) Value Added Value Added Physical Physical lnemp 0.8148*** 0.8111*** 0.5197*** 0.5188*** (41.44) (41.29) (26.17) (26.09) lnbed 0.6845*** 0.6794*** 0.7200*** 0.7189*** (28.57) (28.38) (29.54) (29.46) mshare 0.6531*** 0.6509*** 0.7735*** 0.7739*** (10.93) (10.91) (12.67) (12.68) multidum 0.5775*** 0.5721*** 0.3153*** 0.3141*** (18.54) (18.39) (9.95) (9.90) lnpopdens 0.0878*** 0.0752*** 0.0664*** 0.0636*** (9.87) (8.08) (7.36) (6.73) num – 0.0030*** 0.0007 (4.55) (1.01) _cons 4.6111*** 4.6511*** 4.1241*** 4.1331*** (54.12) (54.40) (47.69) (47.54) Number of observations 5,348 5,348 5,532 5,532 Adjusted R2 0.6764 0.6776 0.5539 0.5539 t-values in parentheses. ***Significant at 1%.