World Bank Reprint Series: Number 364 Kyu Sik Lee Decentralization Trends of Enploynient Location and Spatial Policies in LDC Cities Reprinted with permission fromr, Urban Stu4ies, vol. 22 (April 1985), pp. 151-162, published by the Longman Group Limited, U.K. Urban Studies (1985) 22, 151-162 C 1985 Urban Studies Decentralization Trends of Employment Location and Spatial Policies in L.,DC Cities Kyu Sik Lee [First received August 1983] Summary. This paper documents the observed decentralization trends of employment in Bogota and Cali using several sets of household survey data and the Colombian social security data. Then, it summarizes the theoretical and empirical framework developed for the location choice of manufacturing firms. Drawing on these behavioral underpinnings of the firm's location choice the paper evaluates several policy issues related with employment decentralization based on the policies implemented in the Seouil region. Introduction and congestion are greater than the benefits of agglomeration economies. It is more likely that the In the next two decades the urban population in concerns about the size of these large cities stem developing countries will continue to grow about from the decline in the quality of life of their high four times as fast as in the industrialized countries. income groups, from the frustrations of planners Between 1975 and 2000 the urban areas of develop- who have experienced enormous pressure in recent ing countries are expected to absorb close to one years to accommodate the rapid urban growth, and billion people. In the mid-1970s the net annual from a fear that large cities roay experience cata- addition to the population of Mexico City and Sao strophic failure of management. The second, and Paulo, for example, was over half a million each; the perhaps more easily understood, policy concerns number was over a quarter of a million in Jakarta focus on regional equity, since in most countries it and Seoul. By the year 2000 the developing world is politically and socially important to maintain will have 40 cities with 5 million or more people; 18 balance between regions in terms of income, edu- of them are expected to have more than 10 million cation and employment opportunities, and urban people. Until 1950 Buenos Aires was the only city amenities. Concerns about regional equity also often in the developing world with a population over underlie programs and policies that attempt to 5 million (World Bank, 1979, p. 72). redirect population growth from large cities. The rapid urbanization in these countries has In the middle income countries of Latin America produced a heavy concentration of population and and East Asia, including Mexico, Brazil, Venezuela, economic activity in a very few large urban centers. the Philippines, and Korea, policy makers have paid This pattern of urban concentration has generated a good deal of attention to various ways to reduce two major policy concerns. First, it is widely be- spatial biases toward the large cities. More recently, lieved that the largest cities in these countries are the governments of these countries have initiated getting 'too big.' This belief is not usually based on specific policies to decentralize economic activity evidence that negative externalities such as pollution away from the largest urban center to peripheral The Author is a Senior Economist at the Wlorld Bank Water Supplv and Urban Developmtient Departmtentl Washington DC, and he acknowledges the efficient research assistance provided by Wilhelm Wagner. An earlier version of this paper was presented at the conference on Ulrhanization Policies and Processes in Developing Countries at the University of Chicago, May 1982. 151 152 KYU SIK LEE areas or secondary cities. Policy packages with Bogota 'City Study' conducted at the World Bank, diveise instruments have been initiated and imple- analytical and empirical results wete established mented with varying degrees of success (Renaud, regarding firms' location decisions, but a study of 1981; Townroe, 1979). Among the decentralization policy effects was not undertaken due to the absence policies initiated in these countries, industrial loca- of explicit policy instruments implemented there. tion policies tend to be the most important ones. This Measuring such policy effects is the focus of our is not surprising since 'the true determinants of current research on employment location policies in urbanization and spatial concentration in developing Korea, which is a country with a long history of countries are found in the forces that determine the policy experiments. location of employment opportunities: the nature This paper reports the empirical findings from the and patterns of industrialization, the pace of agricul- Bogota City Study which had a heavy behavioral tural development, and the growth of transportation focus and draws some policy conclusions in the and communication networks' (World Bank, 1979, context of Korean spatial policy experience. The p. 76). More specifically, the decentralization policies following section documents the changing patterns include explicit policy ins'ruments intended to relo- of employment location in Bogota and Cali, the cate existing industries from the large urban centers, third largest city in Colombia; the third section or to induce new industries to new industrial zones, summarizes the econometric work on manufactur- the secondary cities, or lagging regions. ing firms' location choices; the fourth section out- The rationale for the policies to decentralize eco- lines a framework for measuring policy effects; and nomic activity is difficult to justify on economic finally, some policy conclusions are drawn in the last grounds alone, because these policies are an outcome section. of diverse social, political and economic objectives: interregional equity, political cohesion, national de- fense, and preventing further growth of large cities observed Patterns of Employment Loaion because of the fear of a catastrophic failure of urban This section describes the spatial distribution of management. Nevertheless, with the complex set of employment in Bogota and Cali; it documents loca- policy objectives, decentralization policies are very tion patterns of four major industry groups, manu- likely to be pursued in these middle income countries. facturing, commerce, finance, and services, which Hence, it will be extremely important to help govern- together account for about 85 per cent of total ments select policies that are least damaging to the employment in each city. Three main data sets are overall welfare of the economy. used: For analytical reasons, it is convenient to address 1. the 1978 household survey for both cities, employment location policies from two levels of 2. the 1972 household survey for Bogota, and spatial aggregation: "'deconcentration policies' aimed 3. the social security establishment files, 1976 for at modifying employmcnt location patterns within the Cali, and 1978 for Bogota. capital region; 2'decentralization policies' intended to Ideally, the data for the study of employment loca- influence spatial patterns at the national level. The tion should be establishment-based. Although the former involves a study of intrametropolitan phenom- social security files are such data sets, they have poor ena while the latter deals with interregional issues, coverage of small firms. The 1978 household survey, including the development of secondary cities and which was jointly prepared by the World Bank and lagging regions. This paper is confined to intra- the Colombian National Statistics Department metropolitan issues within the capital region. (DANE), had a module to obtain information on In order to formulate sound spatial policies and to establishments where the respondents worked. The implement plans and programs efficiently, it is essen- questions covered the present location of work tial for policy makers to understand the trend of place, employment size of the firm, the initial year of employment location patterns and how firms' loca- operation at the present location, the previous loca- tion behavior tends to produce such a trend. In the tion if moved, and the type of business. 'The material in this section was exttacted from Chapter 2 in Lee (1983). 2According to Bronitsky et al. (1975), large US cities have about 10 to 15 per cent of the city's total employment in the CBD. DECENTRALIZATION TRENDS OF EMPLOYMENT LOCATION 153 Table 1 Employment Share of Major Sectors in Selected Cities Sector Bogota Cali Seoul Kuala Manila Abidjan Tunis US US (1978) (1978) (1970) Lumpur (1970) (1970) (1972) Small Large % % % (1970) % % % (1960) (1960) Manufacturing 23.6 31.1 22.7 20.5 22.1 22.5 18.9 25.1 30.2 Commerce 20.3 21.6 28.9 17.6 13.8 18.6 17.5 28.4 32.6 Servicesa 40.9 32.6 33.0 35.5 37.3 42.8 39.2 28.8 24.8 aIncludes financial services. Source: The values for Bogota and Cali are based on the 1978 DANE household survey; those for other cities are from Renaud (1981). Since the household sample is based on a sample cent including finance) is on the high side, while frame designed for a household survey, making Cali's 33 per cent is on the low side compared to inferences based on such a survey about the charac- other LDC cities. This may reflect that Bogota is the teristics of establishments where the household nation's capital and financial center. members work requires some justification. We can To study the statial distribution of employment, make ex post comparisons of the household survey we have devised a ring system for each city based on results with the social security data regarding estab- comunas, the administrative units defined for sub- lishment characteristics important for this study, areas in Colombian cities. Bogota has 38 comunas because the household survey questionnaire also and Cali has 28. By aggregating comnunas, we divided asked the respondent whether or not his firm is Bogota into six 'rings' and Cali into five rings affiliated with the social security system. Therefore, (Figures 1 and 2). The ring system is useful for the distributions given by the household survey studying the spatial distribution of employment in regarding the establishments affiliated with the terms of the distance from the CBD. social security system can be compared with the Table 2 shows the spatial distribution of employ- corresponding distributions based on the social ment by ring for major industry groups in Bogota security data. We find that in terms of percentage and Cali. For all employment, Cali shows a greater distribution of several establishment characteristics, central tendency than Bogota. The peak concentra- namely, firm size, type of industry, and firm loca- tion in Cali is in Ring 3, while it is in Ring 5 in tion, the two data sets are strikingly similar. This Bogota. In Cali the share of employment drops corroboration between the two data sources in- markedly beyond Ring 3. The CBD's share of em- creases our confidence in drawing inferences from ployment is not much different between the two the household survey. Most of the analyses in this cities, 14 per cent in Bogota and 16 per cent in Cali, section are thus based on the information coming which is comparable to that of large US cities.2 The from the 1978 household survey, which covers all central tendency is, however, substantially different workers regardless of the size of firms where they between the industry groups. In both cities the work. concentration of employment in the central area is Employment shares of major sectors in Bogota highest in finance, followed by commerce, services, and Cali are similar to those of other cities of the and manufacturing. For all industry groups, how- world. As shown in Table 1, the share of manufac- ever, the central area in Cali has a much larger share turing employment is remarkably consistent among of employment than of Bogota. LDC cities, ranging from 20 to 25 per cent. Bogota's In order to examine the changing patterns of manufacturing employment share falls in this range. employment location, we need to look at the spatial Cali's share of 31 per cent however is similar to that distribution of employment at least for two points in of large US cities. The employment shares in com- time. For Bogota we use two household survey data merce, about 20 per cent for both Bogota and Cali, sets for the trend analysis: the 1972 household are comparable to those of other LDC cities. survey conducted for the Bogota Urban Develop- Bogota's share of employment in services (41 per ment Study, Phase II, and the 1978 DANE house- 154 KYU SIK LEE Table 2 Employment Distribution by Ring and Major Industry Groups, 1978 Comuna boundaries Ring Bogota Ring boundaries / / / < All' Manu- Commerce Finance Services 6 % facturing % % % 1(CBD) 13.95 6.01 15.75 41.43 12.91 5 2 17.74 13.47 19.77 29.38 18.68 3 16.40 21.54 14.8? 11.43 16.87 4 20.60 24.89 19.37 10.88 23.18 5 24.94 28.25 27.72 5.65 21.79 6 3.43 2.19 1.51 0.60 4.41 na 2.96 3.66 1.05 0.63 2.15 Total 100.00 100.00 100.00 100.00 100.00 Ring Cali Alla Manu- Commerce Finance Services % facturing % % % 1(CBD) 16.36 9.83 25.23 67.41 14.09 2 26.16 25.71 23.68 19.69 34.57 3 32.67 39.64 31.64 4.70 30.24 4 14.37 11.81 11.84 1.57 15.26 5 2.78 1.38 4.53 0.00 2.12 na 7.66 11.62 3.08 6.64 3.73 Total 100.00 100.00 100.00 100.00 100.00 'Includes other sectors. Data Source: DANE Household Survey 1978. Fig. 1. Bogota. Ring System Based on Comunes. the CBD in the 1972 survey. It should be noted that the CBD, together with Ring 2, lost their share only hold survey. As in the case of the 1978 survey, the slightly from 36.6 per cent in 1972 to 31.7 per cent in 'Phase II' survey had also included questions about 1978.' the respondent's work place, including location, firm In Bogota the extent of employment decentraliza- size, and type of business. For Cali we use the Social tion in manufacturing is very similar to that of Security data file for 1976 with the 1978 DANE commerce, except that the CBD sustained a much household survey. larger share of employment in commerce. The ser- Table 3 reports Bogota's spatial distribution of vice sector also shows a clear trend of employment employment by major industry groups for 1972 and decentralization. The lozation pattern of financial 1978. We find evidence of decentralization of Bogo- establishments is differe'it from those of other sec- ta's total employment during the period. The sharp tors; it maintained the central tendency during the decline of the CBD's share (from 23 to 14 per cent) six year period. For this sector, however, it should is, however, suspect. This apparent overstatement be noted that Ring 2 had a substantial gain of was largely due to the problem with the definition of employment share, reflecting the shift of employ- 3Moreover, we find that based on the two data sets the total number of jobs in the CBD together with those in the 'International Center' (comuna 81, the area directly north of the CBD) stayed almost constant during the six year period as follows: Comuna 1972 1978 31 (CBD) 201,975 166,878 81 (International Center) 24,787 62,565 Sum 226,762 229,443 DECENTRALIZATION TRENDS OF EMPLOYMENT LOCATION 155 Table 3 Changes in Employment Location, Bogota, 1972-1978 Ring All' Manufacturing Commerce Finance Services 1972 1978 1972 1978 1972 1978 1972 1978 1972 1978 1(CBD) 23.03 13.95 18.20 6.01 19.43 15.75 42.11 41.43 22.62 12.91 2 13.61 17.74 16.07 13.47 12.18 19.77 13.69 29.38 12.74 18.68 3 14.62 16.40 18.94 21.54 13.35 14.83 6.89 11.43 15.88 16.87 4 18.80 20.60 20.27 24.89 21.83 19.37 10.00 10.88 20.74 23.18 5 18.61 24.94 21.76 28.25 21.52 27.72 14.64 5.65 17.89 21.79 6 1.67 3.43 1.04 2.19 3.42 1.51 0.55 0.60 1.88 4.41 na 9.67 2.96 3.72 3.66 8.27 1.05 12.12 0.63 8.25 2.15 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 aIncludes other sectors. Data Sources: DANE Household Survey 1978; Phase II 1972. Table 4 Changes in Employment' Location. Cali, 1976-1978 Ring Allb Manufacturing Commerce Finance Services 1976 1978 1976 1978 1976 1978 1976 1978 1976" 1978 1(CBD) 31.51 26.19 20.19 14.94 48.38 54.97 45.68 84.90 29.15 25.34 2 37.12 27.61 34.28 27.02 34.50 24.40 16.67 15.10 55.47 32.19 3 28.40 31.33 41.40 46.27 11.09 15,65 37.55 14.79 16.45 4 2.95 13.82 4.09 11.03 6.03 4.97 0.11 - 0.54 24.37 5 0.03 1.05 0.05 0.70 - - - - 0.06 1.65 Total 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 100.00 'Based on employment in firms with 10 or more qmployees. bIncludes other sectors. 'Excludes employment in the government sector. Data Sources: DANE Household Survey 1978; Social Security Data Files 1976. ment to the 'International Center' as discussed Table 4 indicate that although Cali experienced a above. trend of employment decentralization in manufac- In Table 4 we have attempted to examine evidence turing and services, it had an increasing central of employment decentralization in Cali using the tendency in commerce and finance. The sharp in- Social Security data for 1976 and the 1978 house- crease in the CBD's share of employment in finance hold survey data. As indicated earlier, there is a during this reriod is obviously due to sampling strong consistency between the Social Security data errors. The 1978 household survey must have and the subset of the 1978 household survey data undersampled individuals working in financial firms, which represents those individuals whose firms are especially those located in outer rings. Nevertheless, affiliated with the Social Security system. Since the it should be reasonable to conclude that these results Social Security files have poor c-overage of small do not support a decentralization trend in these two firms for all industry groups, in order to increase the sectors in Cali. comparability between the two data sets, in Table 4 Having examined the trend of employment decen- we have included only those jobs in firms with 10 or tralization, it will be useful to analyze the various more employees. components that contribute to the changing patterns For all industry groups together, we find clear of employment location. Such studies of changes at evidence of employment decentralization in Cali as the margin can be accomplished by investigating the was the case in Bogota. The results reported in location behavior of firms in the following four 156 KYU SIK LEE industry group the origin and destination ratio using the 'rings' as sub-areas. This ratio is obtained by dividing the number of jobs moving out of an area by the number of jobs moving into that area during /\5 0 the 1973-1978 period.4 A 4 The results reported in Table 5 indicate that in both Bogota and Cali the CBD experienced .a net loss of jobs in each industry group. The net outflow of jobs from the CBD was much greater in Bogota than in Cali in each industry group except for the service sector. In Cali all other rings outside the CBD had net gains of jobs except for the service employinent in Ring 3. However, although Cali's ratio values suggest an outflow from the CBD to the other rings, within non-CBD rings the values do not 3 lindicate decentralization. In the case of Bogota, however, the results show evidence of decentralization within the non-CBD > 4 rings due to the moving patterns of relocating firms. In Bogota's manufacturing the ratio gradually de- clines as the distance from the CBD increases; Rings 4 and 5, for instance, had a net gain of employment. The attraction of jobs into Ring 2 is also reflected in Table 5, indicating the growth of the 'International Center.' Table 6 reports the location patterns of jobs Fig. 2. Cali. Ring System Based on Comunas. created by firms established during the five year period (births). It is striking to see that there is a groups: strong regularity in the location patterns of new 1. relocating firms (movers), jobs, i.e. in both Cali and Bogota the employment 2. newly established firms (births), share of new jobs by ring increases with the distance 3. defunct firms (deaths), and from the CBD for almost all industry groups with 4. the expansion or contraction of stationary (non- the exception of the finance sector. This is an relocating) firms. indication that newly established firms tend to locate Evidence will be presented for the first two cat- in outer areas, contributing to the decentralization egories. The destination of movers and the location patterns. It is worth noting that in Bogota's finance patterns of births are important for understanding sector, 43 per cent of jobs which were created by new the changing patterns of employment location and firms during this period located in Ring 2, capturing for predicting the future spatial structure. Here, we the northward shift of financial activities from the use the relevant information collected in the 1978 CBD to the 'International Center.' household survey to examine the direction of mov- ments of jobs and the location patterns of births. Behavioral Underpinnings of Manufacturing Firm The 1978 household survey included questions Locatioii Choice about a firm's year of operation at the present location, and the previous location if it moved. Using various data sets, the location patterns of Based on this information, we calculated for each employment in Bogota and Cali and their changes 4Since the number of workers at the previous locations is estimated on the basis of the number of respondents of the survey, it is implicitly assumed that the number of jobs existing at the previous locations is the same as that of the 1978 locations. This assumption should not affect our conclusions. DECENTRALIZATION TRENDS OF EMPLOYMENT LOCATION 157 Table 5 Table 6 Moving Patterns of Jobs Measured by Origin-destination Ratioa Location Patterns of New Jobs, 1973-1978 for Rings, 1973-1978 Ring Bogota Ring Bogata Alla Manu- Commerce Finance Services Allb Manu- Commerce Finance Services facturing facturing I 13.78 7.55 14.09 29.20 18.61 1 2.37 2.51 3.15 3.32 1.45 2 17.80 13.47 20.31 43.04 14.05 2 0.66 1.54 0.45 0.33 0.87 3 14.42 19.98 8.15 13.86 17.38 3 1.33 1.58 1.32 1.04 1.36 4 20.87 24.66 23.00 9.43 17.55 4 0.55 0.76 0.51 iu.q6 0.58 5 27.67 30.62 31.97 4.47 26.78 5 0.44 0.32 0.20 2.70 0.35 6 2.64 2.82 1.72 0.00 2.32 6 0.59 c _cd _cd 0.29 na 2.82 0.90 0.75 0.00 3.32 Ring Cali Total 100.00 100.00 100.00 100.00 100.00 Allb Manu- Commerce Finance Services Ring Cali facturing Alla Manu- Commerce Finance Services 1 1.85 1.97 1.71 1.58 2.03 facturing 2 0.68 0.70 0.61 _d 0.45 3 0.86 0.93 0.84 _-d 1.21 1 14.70 9.27 13.61 0.00 26.27 4 0.67 0.70 _c _cd 0.33 2 25.62 19.18 24.05 60.00 30.88 5 3.44 -c _d d C 3 38.33 44.95 44.76 20.00 32.99 4 13.40 22.24 11.45 0.00 5.16 'The number of jobs moving out of a zone divided by the number 5 2.60 2.29 1.97 0.00 0.00 of jobs moving into the zone. na 5.35 2.06 4.15 20.00 4.70 bIncludes other sectors. CNo firms moved into the area. Total 100.00 100.00 100.00 100.00 100.00 dNo firms moved out of the area. 'Even though the ratio was high, Ring 5 lost only 651 jobs and aInc.ludes other sectors. gained 189 whereas the CBD lost 9,479 and gained 5,126. The Data Source: DANE Household Survey 1978. total number of jobs relocated was 28,170 in Cali and 84,755 in Bogota. Data Source: DANE Household Survey 1978. (Lee, 1982b). This section briefly describes the model and summarizes the empirical results.' over time were summarized in the previous section. Mills (1972), Solow (1972), and Muth (1969) The results reveal a high degree of employment provided the basic theoretical foundations for econ- location dynamics and strong evidence of the spatial omic analysis of residential location and employment decentralization of employment in both cities. In location in urban areas. The theoretical and empilical order to understand and explain' these changing work on housing and residential location has ad- location patterns, in 1978 we conducted a survey of vanced rapidly during the decade,6 but comparable manufacturing establishments in Bogota using the work has not followed in the field of employment DANE's industrial directory as the sample base. location. Mieszkowski and Straszheim (1979, page The survey questionnaire was designed to take no xiii) observe that 'among the most important issues more than one hour to complete and did not require not covered in this volume is the relative importance the respondents to refer to their accounting books; of different factors in employment location and however, the questionnaire provided a large amount decentralization patterns within metropolitan areas, of information with nearly 300 computer readable which deserves far more research. Principally because variables. The survey results, which provide the of data problems, little research has been conducted basis for understanding the changing location pat- on employment locations or urban labor markets in tems of manufacturing employment, were reported recent years. Effective use of the censuses of employ- in Lee (1982a), and a model of employment location ment is plagued by disclosure rules.' Nevertheless, was formulated and estimated using this survey data the work by Leone (1971), Schmenner (1973), and 5The discussion that follows in the remainder of this section is based on Lee (1982b). 6Major work in this area includes Kain and Quigley (1975), Straszheim (1974), and Ingram (1977). 158 KYU SIK LEE Kemper (1973), followed by Struyk and James protection, and zonal amenities such as air quality (1975), pushed back the frontier in developing an and the extent of congestion. empirical basis for employment location study. The relevant cost components in calculating the Moreover, recent efforts for actual modeling of optimum combination of inputs, hence in selecting a employment location and econometric estimation of particular location, are then wages, capital cost, such models have been promising: Hanushek and input materials costs, delivery costs of inputs and Song (1978) developed a framework for analyzing the output, and land rent. Following the standard spatial structure of employment in the Boston metro- theory in urban economics, a particular plant site is politan area; Erickson and Wasylenko (1980) esti- then occupied by the firm which bids the highest mated a model of relocating firms in the Milwaukee price for it. The bid price will depend on the metropolitan area; Schmenner (1973, 1982) provided attractiveness of the particular site, which depends evidence from his econometric work on Cincinnati in part on the availability of local public goods, for a and New Engiand. Carlton's work (1977, 1979) firm of particular type. In locational equilibrium, all focused on the intercity location decisions of new firms in an urban area make the same profits and no firms. firm has any incentive to relocate. This equilibrium In this study the basic premise for modeling configuration is an outcome of cost trade-off employment location is drawn from already well- calculations by individual firms given the spatial established theoretical and empirical foundations variations of relevant costs. For example, a large in the housing and residential location literature. manufacturing firm may choose a site in a low rent Residential location studies usually assume that a area near the periphery to meet its need for more consumer chooses a particular dwelling unit at a plant space at the expense of longer delivery particular location where he maximizes his utility distance. On the other hand, small firms may prefer given the budget constraint. In a similar optimiza- a central location where the high rent is more than tion framework, we assume that the firm, as a price compensated by various externalities available for taker, locates where it maximizes profits. In making production and shipment. The high rent in the the location decision, the attributes of the particular central area can also be explained by high intensity plant site as well as the lot size enter into the process of variable input use. The empirically observed rent of determining the optimum combination of inputs gradient reflects the increasing marginal productiv- for production. This means that the site attributes ity of land as the distance to the CBD becomes directly enter into the firm's production function in shorter. the following way: The stochastic specification of the above model Q = f(L, X; Z), (1) should provide a framework for predicting the prob- ability that a firm of particular type will occupy a where Q= output, L = lot size, X = a vector of site with particular attributes (Z). Since a given site variable inputs such as labor, and plant and equip- will be occupied by a firm with the highest bid, the ment; a -d Z = a vector of exogenous site character- relevant random variable for determining the proba- istics. bility is the maximum bid given by a group of firms The site characteristics are independent of lot size with similar attributes. The probability distribution and represent 'local public goods' available to that of a random variable associated with the maximum particular location. In a recent work on housing, bid leads to a multinomial logit specification for the Burstein (1980) explicitly introduces local public firm location model. goods into the consumer's utility function; Ellickson The application of the multinomial logit method (1981) also integrates the theory of residential loca- to economic research became popular with McFad- tion choice with that of local public goods by den's work on travel demand studies (1973; 1974; including in the consumer's utility function the 1976). Such a model was used to predict an indi- attributes of the surrounding neighborhood and vidual's choice of travel mode among a finite num- public schools. In the case of firm location study, ber of alternatives such as by car, bus, taxi, transit, local public goods include the quality of public or foot, given the characteristics of the individual or utility services such as electricity and water, the household. Subsequently, the multinomial logit quality of municipal services such as police and fire framework was applied in housing and residential DECENTRALIZATION TRENDS OF EMPLOYMENT LOCATION 159 location studies, for instance, Friedman (1975), type will occupy a site with particular attributes, and Lerman (1977), and Quigley (1976). These studies the site attributes are the independent variables applied the travel mode choice model to variations subject to policy manipulations. They include of residential choice models: a utility maximizing measures of access to output and input markets, consumer of type t chooses a house of type Z, which employees' commuting distance, the provision of is analogous to the consumer's choosing a travel public services, zonal amenities, and external (and/ mode (Z) among several alternatives. or scale) economies. This means that the model can In his recent work Ellickson (1977, 1981) made an predict incremental effects of alternative policy in- important departure from the above-mentioned ap- struments on the firm's likelihood of choosing a proach and offered an alternative multinomial logit particular site. specification for the residential choice model using Policy makers in developing countries often at- the bid-rent theory. According to Ellickson (1981, tempt to relocate industries from traditional indus- page 63), 'the most natural way to interpret such trial districts of a large city to outer areas or to models is in terms of a prediction of what sort of smaller cities. The government's plans may include consumer is most likely to occupy a house with a developing new industrial towns or new industrial specified set of characteristics,' hence, the probabil- estates or expanding existing ones to induce new or ity that a house with characteristics Z will be relocating firms to settle in a desired area. In all occupied by a household of type t.' Ellickson points cases implementing such plans and programs re- out that his method has several advantages such as quires the selection of particular types of industries specifying endogenously the properties of the dis- to occupy sites having particular attributes. Hence, turbance term (as Weibull). it is important for the policy makers to understand Ellickson's approach is particularly relevant for the requirements of firms for attaining equilibrium modeling employment location. As discussed above, at new locations, and to be able to assess the level of our problem is to predict the probability that given a government subsidies and infrastructure investment site with particular characteristics (Z), a firm of for meeting such requirements and their costs. particular type will occupy that site. This means that The Bogota study described earlier did not test the such a site may exist at any location in an urban effectiveness of explicit policy instruments partly area, and there can be a large number of site because such instruments were not implemented in variations given by combinations of site character- that city. However, the behavioral underpinnings istics. On the other hand, the application of the established in the study provide sufficient clues McFadden approach will require a definition of a regarding those policy instruments most appropriate finite number of locations (or sub-areas) in a city for influencing the location choice of particular from which a firm of particular type is supposed to types of firms, and for influencing aggregate location choose. Defining alternative sub-areas in that fash- patterns. ion will be arbitrary, however, and the subsequent The estimation results reported in Lee (1982b) analysis will tend to be descriptive rather than show which independent variables (i.e. site attri- analytical, suffering from the idiosyncracy intro- butes) are most important for the location choice of duced by the arbitrariness of the sub-area so defined firms of particular types. In Table 7, according to as alternative locations. the values of elasticity coefficients, the independent variables are ranked by order of importance for each A Frameworki for Evaluating Policy Effects type of firm. For small firms in both textile and fabricated metal industries, the accessibility to local A study of the effects of employment location input markets (INPUTBT), the commutilg distance policies requires a theoretical and empirical frame- of production workers (WKSOUTH), and the work for testing the effectiveness of policy instru- population density of the zone of plant location ments with respect to the firm's location choice. The (POPDENS) are important site characteristics. model described above provides such a framework. These three variables alone carry more than half of It predicts the probability that a firm of a particular total coefficient weight in the equation. It should be 'The mathematical derivation of this result can be found in Lee (1982b). 160 KYU SIK LEE Table 7 identified as 'incubator areas,' all of which are not in Ranking of Independent Variables for Firm Location Chloice the industrial district but are adjacent to the CBD. Variables Small Small Large From this analysis, it is apparent that government Textile Fabr. Metal Textile policies intended to influence employment location Rank Rank Rank patterns can be effective if such policies influence the INPUTBT 1 1 4 site attributes which are important to firms. The WKSOUTH 2 4 7 analysis further suggests that different mixes of POPDENS 3 3 8 policy instruments should be considered for different ELECINT 4 8 2 LOCQT 5 7 1 groups of firms and industries. PRODSOLD 6 2 5 ADMNORTH 7 5 3 DISTCBD 8 6 6 Policy Implications and Concluding Remarks Definition of variables: The strong decentralization of employment in INPUTBT = A measure ol' accessibility to local input markets; Colombian cities, summarized in Section 2, is com- WKSOUTH = The commuting distance of production workers; ' t POPDENS = Population density of the cummnta where the parable to the trend observed for large US cities establishment is located; during the past several decades (Hoover and ELECINT = Frequency of electricity interruption; Vernon, 1959; Leone, 1971). The main aim of spatial LOCQT = Location quotient; ' ' t PRODSOLD = A measure of accessibility to local product Policy in the US regarding intra-metropolitan decen- markets; tralization, however, has been quite different from ADMNORTH = The commuting distance of administrative that of LDCs. In the US the major policy objective workers; DISTCBD = Airline distance from the CBD. has been to reduce urban decentralization in the hope Source: Lee (1982b). of preventing central city decay. Although this policy objective is based on the belief that decentralization noted, however, that the accessibility to local product occurs because of deteriorating conditions in the markets (PRODSOLD) was the second most import- central city, little empirical evidence supports this ant variable for small, metal-fabricating firms. while it view. Most decentralization is attributed to trans- was not so important for small textile hrms. portation improvements, suburban (new town) For large textile establishments the location quo- development programs, and federal subsidies to tient (LOCQT), which is a measure of scale econom- homeownership (Muth, 1969). Muth's findings sug- ies in the zone of plant location, the quality of gest that federal programs intended to reduce electricity supply (ELECINT), and the commuting decentralization, such as urban renewal programs distance of administrative workers (ADMNORTH) which tend to lower a central city's housing density, are the three most important site attributes. These have expedited the trend rather than reduced it. three variables alone carry more than two-thirds of Therefore, attempting to reverse the tide of total weight in the equation. Access to local markets decentralization, which occurs from increased popu- and population density are relatively unimportant lation and incomes, often results in economic ineffici- for this group of firms. ency since transportation improvements have The results for large textiles are consistent with the reduced the central city's comparative advantage for findings that large firms are export-oriented and tend production and other economic activities. to locate near the periphery where land space needed Large US cities also have municipal fiscal prob- for modern facilities is available at lower costs and lems that encourage decentralization. The high cen- highway access is better than in the central area (Lee, tral city per capita tax burden relative to suburbs 1982a). Moreover, evidence on small firms' local provides incentives to move to the suburbs. It has market orientation and their tendency to locate in high been observed that 'the lower the average income density areas is consistent with the 'incubator hy- level of the central city relative to its suburbs, the pothesis' which was supported by a test in earlier work smaller is the central city's population, and the (Lee, 1981). The hypothesis states that small, new larger is the land area occupied by the urbanized manufacturing firms tend to locate in centralized areas area' (Muth, 1969). Evidence from Bogota and Cali, that provide essential services for production and sales however, does not suggest an increasing concentra- activities. In fact, specific areas in Bogota were tion of low income population in the central city, DECENTRALIZATION TRENDS OF EMPLOYMENT LOCATION 161 and there is no signs of central city decay. In fact, the was tested for Bogota, based on the industrial policies to decentralize economic activity from large directory data (Lee, 1981). Data from the establish- cities in LDCs mainly stem from the increasing ment survey conducted for the Bogota City Study concentration of economic activity in the central city, do in fact indicate that small firms start up in central accompanied by the perceived problemns of congestion locations (Lee, 1982a). The logit results summarized and pollution as the city's population grows rapidly. earlier also support this hypothesis (1982b). There is Also, municipal fiscal relationships in LDCs are evidence that Korean firms also follow this pattern different from those of the US. In Korea, for example, (Meyer, 1981). If central city incubation were indeed electric utility service charges, which are uniform prevalent in Seoul, the 1977 Industrial Location Act, nationwide, tend to subsidize central cities over out- which prohibits starting new manufacturing firms lying areas, hence encouraging centralizations. within the city limits, should have restricted such During the past decade various spatial policies to incubation. A related issue is whether incubation control the growth of Seoul and to disperse its can be replicated in new industrial estates or existing population have been implemented. For example, in towns outside Seoul. 1971 the greenbelt surrounding Seoul was estab- In 1978 as a major program to decentralize popu- lished. Six years later the 1977 Industrial Location lation and economic activity from Seoul the govern- Act in effect prevented new manufacturing firms ment established a new industrial town, Banweol, from locating within Seoul and enabled the govern- located less than 30 kilometers away. Although ment to issue relocation orders to establishments nearly 1,000 plant sites for small and medium sized already set up there. That same year the government firms were prepared, the occupancy rate has been initiated a ten-year comprehensive plan for popula- less than 20 per cent as of 1981. Many firms that tion and industrial redistribution from Seoul. The moved to Banweol suffered excess capacity and plan included a large number of policy instruments financial losses, resulting from in five major categories: 1. over investment in plant and land area, induced 1. the relocation of industries from Seoul; by the government incentive schemes; 2. inducements to relocate population to southern 2. increased operating costs after relocation; and provinces; 3. the 1980 general recession. 3. the decentralization of education facilities; According to a case study on Banweol (Choe and 4. the relocation of various urban functions within Song, 1982), the most serious problems facing the the capital region and the improvement of city relocated firms have been reduced accessibility to plan implementation, particularly in dealing with product markets and input suppliers, the unavail- clandestine housing construction in Seoul; and ability of production workers, and difficulties in 5. tax and credit incentive schemes to aid relocating obtaining day-to-day business information (poor firms. telephone service and person-to-person contacts). The majority of the instruments stipulated in the Poor access to Seoul and Incheon is largely respon- plan address the location and relocation of m, nu- sible for these problems. A related problem is the facturing establishments. A comprehensive review of reluctance of production wo,lkers to relocate to employment location policies in Korea appears in Banweol or to commute from Seoul. Attrition of Choe and Song (1982), and a theoretical aiiaiysis of skilled workers has been high, and it is difficult for policy efficiency is discussed in Murray (1982). The firms to replace them af er they quit. empirical evaluation of such policies is the main task That such a seemingly short distance thwarted the of the current Korean spatial study. Several other development of Banweol is striking. Logit analysis recent policy measures in Korea and Colombia are used to study the Bogota data helps explain the discussed below. Korean experience: Small and medium sized firms The 'incubator hypothesis' mentioned earlier prefer central locations. Accessibility to local states that small, new manufacturing firms start in markets and proximity to production workers are central locations that provide needed services and the most important site attributes for them. infrastructure and then move to less central loca- Unlike the bias against decentralization in the US tions as they grow and need more space for expan- several developing countries have aimed at decen- sion (Hoover and Vernon, 1959). This hypothesis tralizing economic activity from the central city, 162 KYU SIK LEE Nevertheless, the desirability of decentralization Discrimination: A Microeconomic Analysis. NBER: New policies on economic grounds has not been estab- York. KEMPER, P. (1 973), The Lotation Decisions of Manufacturing lished, and little is known of their effects or their Firms within the Newv York- Metropolitan Area, unpublished welfare implications. The key policy question is how Ph.D. dissertation, Yale University. to guard against excessive spatial policies relative to LEE, K. S. (1981). Intra-urban Location of Manufacturing Em- ployment in Colombia. Journal of Urban Economics. prevalent trends, since excessive measures might LEE, K. S. (1982a). Determinants of Intra-urban Location of result in serious welfare losses. In developing coun- Manufacturing Employment: An Analysis of Survey Results tries the absence of empirical inforrmation on decen- for Bogota, Colombia, Urban and Regional Report No. 81-83, n d p y e d The World Bank. tralization8 and policy effects does not yet permit LEE, K. S. (1982b). A Model of Intra-urban Employment Loca- the making of more efficient spatial policies. How- tion: An Application to Bogota, Colombia. Journal of Urban ever, policies to decentralize population and eco- Economics. c a y . LEE, K. S. (1982c). Changing Location Patterns of Manufactur- ing Employment in the Seoul Region. Urban Development better internal management of city growth. For Discussion Paper, The World Bank. example, the effect on air pollution or on traffic LEE, K. S. (1983). Employment Location in Colombian Cities, The of reducing the population or employ- City Study Research Monograph: The World Bank (Draft). congetionLEONE, R. (1971). Location of Manuqfacturing Activity in the New ment in a large city by a certain amount is likely to York Metropolitan Area, unpublished Ph.D. dissertation: Yale be very small (Tolley, 1979; Henderson, 1980). University. LERMAN, S. R. (1977), Location, Housing, Automobile Owner- ship and the Mode to Work: A Joint Choice Model, Transpor- REFERENCES tation Research Board Record: No. 610, 6-1. McFADDEN, D. (1973), Conditional Logit Analysis of Qualitative BRONITSKY, L. and others (1975). C'rban Data Book, US Depart- Choice Behavior, in P. Zarembka ed., Frontiers in Econo- ment of Transportation (Report No. DOT-TSC-OS5-75-45-1): metrics, Academic Press. Washington, DC. MCFADDEN, D. (1974). The Measurement of Urban Travel BURSTEIN, N. R. (1980). Voluntary Income Clustering and the Demand. Journal oJ Public Economics. Demand for Housing and Local Public Goods. Journal of MEYER, J. R. (1981). Report on Proposed Korean Spatial Study. Urban Economics. Urban and Regional Report No. 81-1: The World Bank. CARLTON, D. (1977). Models of New Business Location. Univer- MIESZKOWSKI, P. and STRASZI-EIM, M. (1979), ed., Current Issues sity of Chicago, Center for Mathematical Studies in Business in UJrban Economics, Johns Hopkins Press. and Economics, Report 7756. MILLS, E. S. (1972), Urban Economics, Scott, Foresman, Glen- CARLTON, D. (1979). Why New Firms Locate Where They Do.: view, Ill. An Econometric Model. in W. Wheaton ed., Interregional MURRAY, M. P. (1982). Here, There, Where? A Strategy for Movements and Regional Growth: The Urban Institute. Evaluating Industrial Relocation Policies in Korea, Urban CHOE, S. C. and SONG, B. N. (1982). An Evaluation of Industrial Development Discussion Paper, No. 6; The World Bank. Location Policies for Urban Deconcentration in Seoul Region. MUTH, R. (1969). Cities and Housing, University of Chicago U'rban Development Discussion Paper, No. 7, The World Bank. Press. ELLICKSON, B. (1977), Economic Analysis of Urban Housing QUIGLEY, J. (1976). Housing Demand in the Short Run: An Markets: A New Approach. The Rand Corporation, July Analysis of Polytomous. Choice, Explorations in Economic (R-2042-NSF). Research, 3. ELLICKSON, B. (1981). An Alternative Test of the Hedonic Theory RENAUD, B. (1981). National Urbanization Policy in Developing of Housing Markets. Journal of UJrban Economics. Countries, Oxford University Press. ERICKSON, R. A. and WASYLENKO, M. (1980). Firm Relocation SCHMENNER, R. (1973). City Taxes and Industry Location, unpub- and Site Selection in Suburban Municipalities. Journal of lished Ph.D. dissertation: Yale University. Urban Economics. SCHMENNER, R. (1982). Making Business Location Decisions, FRIEDMAN, J. (1975). Housing Locztion and the Supply of Local Prentice-Hall. Public Services. Rand Corporation, (P-542 1). SOLOw, R. (1972). On Equilibrium Models of Urban Location, in HANUSHEK, E. A. and SONG, B. N. (1978). The Dynamics ofPostwar M. Parkin, ed., Essays in Modern Economics: Longman. Industrial Location. Review of Economics and Statistics. STRASZHEIM, M. R. (1974). An Econometric Analysis of the Urban HENDERSON, J. V. (1980). A Framework for International Com- Housing Market, NBER: New York. parisons of Systems of Cities. Urban and Regional Report No. STRUYK, R. and JAMES, F. (1975). Intrametropolitan Industrial 80-3, The World Bank. Location, Lexington Books. HOXVER, E. M. and VERNON. R. (1959). Anatom.y of A Metropolis, TOLLFY, G. (1979). Urban Growth Policy in A Market Economy: Harvard University Press. Academic Press. INGRAm, G. K. (1977), Residential Location and Urban Housing TOWNROE, D. M. (1979). Employment Decentralization: Policy Markets, N3ER: New York. Instruments for Large Cities in LDC'S, Pergamon Press. KAIN, J. and QUIGLEY, J. (1975), Housing Markets and Racial World Bank (1979). World Developnment Report 1979. 8Based on preliminary estimates, Lee (1982c) shows that in the 1970's the CBD of Seoul experienced a net loss of a manufacturing employment at an annual rate of -- 7.6 per cent, while the periphery of Gyeonggi province (where Seoul is located) gained employment at a rate of 34 per cent. THE WORLD BANK Headquarters U 1818 H Street, N.W. Washington, D.C. 20433, U.S.A. Telephone: (202) 477-1234 Telex: WUI 64145 WORLDBANK RCA 248423 WORLDBK Cable address: INTBAFRAD WASHINGTONDC European Office 66, avenue d'1ena 75116 Paris, France Telephone: (1) 47.23.54.21 Telex: 842-620628 Tokyo Office Kokusai Building 1-1, Marunouchi 3-chome Chiyoda-ku, Tokyo 100, Japan Telephone: (03) 214-5001 Telex: 781-26838 The full range of World Bank publications, both free and for sale, is described in the World Bank Catalog of Publications, and of the continuing research program of the World Bank, in World Bank Research Program: Abstracts of Current Studies, IThe mosr recent edition of each is available without charge from: PUBLICATIONS SALES UNIT THE WORLD BANK 1818 H STREET, N.W. WASHINGTON, D.C. 20433 U.S.A. ISSN 0253-2131