Monocentric and Polycentric Density Functions and Their Required Commutes

Shunfeng Song

Working Paper UCTCNo. 198

The University of California TransportationCenter Universityof California Berkeley, CA94720

The University Transportation

of California Center

The University of California Transportation Center (UCTC) is one of ten regional units mandated by Congress and established in Fall 1988 to support research, education, and training in surface transportation. The UCCenter serves federal Region IX and is supported by matching grants from the U.S. Department of Transportation, the California Department of Transportation (Caltrans), and the University. Based on the Berkeley Campus, UCTC draws upon existing capabilities and resources of the Institutes of Transportation Studies at Berkeley, Davis, Irvine, and Los Angeles; the Institute of Urban and Regional Development at Berkeley; and several academic departments at the Berkeley, Davis, Irvine, and Los Angeles campuses. Faculty and students on other University of California campuses may participate in

Center activities. Researchers at other universities within the region also have opportunities to collaborate with UCfaculty on selected studies. UCTCseducational and research programs are focused on strategic planning for improving metropolitan accessibility, with emphasis on the special conditions in RegionIX. Particular attention is directed to strategies for using transportation as an instrument of economic development, while also accommodating to the regions persistent expansion and while maintaining and enhancing the quality of life there. The Center distributes reports on its research in working papers, monographs, and in reprints of published articles. It also publishes Access, a magazine presenting summaries of selected studies. For a list of publications in print, write to the address below.

University of California TransportationCenter 108 NavalArchitecture Building Berkeley,California 94720 Tel: 510/643-7378 FAX:510/643-5456

Thecontentsof tiffs report reflect the views the authorwho responsible of is for the facts andaccuracy the data presentedherein. Thecontentsdo not of necessarily reflect the official views policies of the Sta~e California the or of or U.S.Depaffment Transportation.Th~ of report does not constitute a standard, specification,or regulation.

Monocentric and Polycentric Density Functions and Their Required Commutes

ShunfengSong
Department of Economics University of Nevada at Reno Reno, Nevada 89557-0016

Working Paper October 1992

UCTCNo. 198 TheUniversity of California Transportation Center Universityof California at Berkeley

ABSTRACT This paper examines relationship betweenurban structure and the commuting behavior. Analyzingthe 1980 journey-to-work data for the Los Angelesregion, this paper has shown that polycen~cdensity functions fit the actual urban structure better than the conventional monocen~c model. This finding indicates the preeminence accessibility to majoremployment of centers in location choices. This paper also estimates commute flows implied by the polycentric and monocentricfunctions. It finds the monocen~c modelvery poor at explaining commuting behavior. The empirical results showthat polycentric urban structure increases the urban commute. This finding helps to preserve the assumptionthat urban workers economize commuting, suggests that efforts to promote on and moreefficient urban form, such as the jobs-housingbalance policy, have the potential to succeed.

I wouldlike to thank KennethA. Small, DavidBrownstone,Gordon (Pete) Fielding, and Genevieve Giuliano for their helpful comments suggestions. Financial and supports from the University of California Transportation Center and the OuJou Yi Scholarship are greatly acknowledged. views expressed here are the sole The responsibility of the author.

INTRODUCTION

The standard urban economicmodel was developed during the 1960s as economists soughtto providean analytic framework studyingurban spatial structure (Alonso[1], for Mills [25] and Muth[30]). The central features of the modelare monocentricityand the trade-off betweenhousingcosts and commuting costs. Usingthe indifference principle, the value of locational advantage (access to the city center) is capitalized in land rent. Locational equilibrium is achieved whenthe marginalhousing savings of moving farther from the city center equal the marginal increased commuting costs. The modelpredicts that both land rent and residencedensity decline with the distance fromthe center (Mills [27], chpts 6-7, Mills and Hamilton[28], appendixA, Muth[31] and Wheaton [40]). The standard urban model, however,has been thought to be a poor description of reality. Hamilton[16] showsthat the actual average commute typical U.S. metropolitan in areas is eight times as large as the minimum average commute consistent with the standard monocentricmodel. Hamiltoncalled this unexplained commute "wasteful commuting;" here I call i~t excess commuting avoid any normativeconnotations. Since 87 percent of the to actual commute excess~Hamilton is concludesthat the monocentri_c urban modelhas little predictive value concerning commuting behavior, and claims that "it is not clear that the trade-off between commuting land rent plays any significant role at all in location and decisio:as" (Hamilton[16], p. 1050). Small and Song[36] find that the monocentric required commute only about one-fifth of the actual commute Los AngelesCountyin 1980, is for verifying Hamiltonsfinding that the monocentric modelis very poor at explaining urban commuting.

Onepossible explanationof Hamiltonsresults is that employment residences are and distributed in a pattern consistent with many employment centers, not just one. Several recent studies have demonstratedthe presence of employment subcenters in large American cities (Cervero [3], Giuhano Small [10] and McDonald and [23]), casting doubts on the assumption monocentdcity. theoretical basis for urban subcenters has also received of The attention (Helsley and Sullivan [I9], Odland[33], Sasaki [35] and White[41, 43]). A few empirical researches have examined impacts of urban subcenterson spatial patterns the (Gordon al. [12], Griffith [15] and Smalland Song[37]). et Polycentric density functions, however, have not been incorporatedinto the calculation of excess commute. a polycentric urban area, someoutwardand In circumferential commuting might be going on and excess commuting might not be as much as that found by Hamilton(Suh [38] and White[42]). Some studies, however,have argued that polycen~c spatial structure reduces rather than lengthens urban commuting (e.g., Gordon aI. [I 1] and Gordon Wong et and [13]). It is unclear whateffect that polycentricity has on urban commuting. Moreover,Hamiltoncalculates the excess commute assuminga monocentficform for U.S. cities. His result could be misleadingif the monocentric modelinadequatelyrepresents the spatial structure. Hence,weare unableto distinguish whetherhis result is an indictment of the monocentricityor of the morefundamentalassumptionon commuting behavior that householdstrade off betweenland rents and commuting costs. Anindictment of the monocentficity mightnot be surprising; but rejecting the trade-off assumption more is drastic. Thelatter strikes at the heart of urban economics hence implies an urgent need and

to :reformulate the analytical land-use modelmost commonly by urban economists;it used also indicates a need to re-assess those efforts to promote moreefficient urban form, such as the jobs-housingpolicy that pursues a jobs-housingbalance development urban areas so in that the amount commuting its consequentpeak period congestion and air pollution of and can be reduced. This papertests the existenceof polycentricityanddetermines effect on the spatial its structure and the estimate of excesscommuting; also examines behavioralvalidity it the reg;uding the trade-off assumption.Using1980small-zonejourney-to-work data for the Los Angeles region, I first estimate rnonocentricand polycentricdensity functions for both employment resident workers, and performhypothesistests on the fit of the monocentric and model explainingthe spatial distributions and the existence of polycentricity. I then in determinethe effect of polycentricity on the estimate of excess commuting estimating the by average minimum commutes required by the monocentricmodeland the polycentric model. Finally, I examine behavioral validity that householdsmakeattempts to economize the on commuting their location choices. in

DENSITY FUNCTIONS

In a rnonocentriccity, urban residents are assumed value access to the center and to trade off this access and housingcosts in their location decisions. In consequence, urban residents are distributed in a circularly symmetric manner with density function f(r), where

r is the distance from the center. Assuming someemployment decentralized, urban is economists have postulated the distribution of employment similar to that of resident workers (Hamilton[16], Mills [27], Mutht30] and Thurstonand Yezer[39]). The negative exponential density function is the most commonly modelin the used monocentric literature; andit is also used in this paper. For the case of resident worker city distribution, this function can be derived theoreticaUywith a maximizing model.For example,Mills and Hamilton[28] and Papageorgiou Pines [34] have derived the negative and exponential density function using a compensated demand housingwith a unitary price for elasticity; Bussiereand Snickars[2] derive the samefunction formwith an entropy maximizing method.Thenegative exponential density function is also supported by empirical evidence (Clark [5], Mills [27] and Math[30]). For the case of employment distribution, this functional formhas been also theoretically derived (McDonald and [22] Mills [26]) and commonly (Kemper Schmermer used and [20], Mills [27] and Mills and Ohta [291) Thenegativeexponentialdensity is written here as
Dm u"Doe-gre = , m = 1, 2 ..... M, (I)

whereD, is the workerresidence or employment density at distance r, to the single urban center; Mis the total number zones in aa urban area; e~ is a multiplicative error of term associated with zone m;t D and g are parameters to be estimated from the data by O

tAn additive error term has been also used in the literature on the negative exponential density function. Greeneand Bambrock [14], however,showsthat a multiplicative error term is moreappropriate. 4

ordinaryleast square after taking logarithmof equation(1). Theoretically, Dois the density exmapolated the urban center, and g is the density gradient measuring percentagefall to the off in density for an unit increase in distance fromthe CBD. Thenatural extension of the monocentric modelis to assume that access to all employment centers is of primaryimportance location decisions, and specify that resident in workersand employment functions of distances to all employment are centers. ~ Such density function has been suggestedand estimatedby Gordon aa. [12], Griffith [15] and et Small and Song[371, whichis written here as "b~- + v. , D. -- ~A,,e
11-1 N

m-- 1, 2 .....

~r,

(2)

whereN is the number employment of centers in an urban area; r.. is the distance from center n to zone m; v. is the error term associated with zone m; A. and b. are parametersto be estimated for each employment center n. This specification of polycentric modelassumes that the density at any location is the vertical summation the negative of expo.ential density functions, eachreflecting the influenceof a center on that location When intercepts of all centers exceptone are zero, the polycentfic formcollapses the to the, monocentric form. Therefore,wecan performstatistical tests on hypotheses that the polycentncmodelexplains the actual distributions better than the monocentfic model,and that the existenceof polycentricityin the spatial structure. We also test the significance can

2Apolycentricdensity function couldbe the upperenvelope the density gradients for the of centers if weassume influences fromall centers are completely the substitutable; it couldalso be a multiplicative function of centers influences if we assumecenters are complementary (I-Ieikkila et aa. [18]). Aaadditive functionis specifiedby assuming centers are in between that of these two extremecases. 5

of of center n in explainingthe overall density pattern by means t-test on its parameters and b. Estimatingthe polycentric density functions, weare able to determine impactthat the each center has on the overall distributions of employment resident workers. and estimated aggregate impact of center n is computed from the formula
The

where S~, is the area of zone m; ~ and b. are the estimated intercept and density is gradient. IMPACT, positively related to ~ and negatively related to b..

CALCULATION OF REQUIRED COMMUTE

The"wasteful commuting" literature has used two methods estimate the excess to commute. uses estimated monocentdc One density gradients, here called Hamiltons calculation. The other uses a linear assignmentmodel,called Wh#es calculation. Theyare reviewed turn. in

HamiltonsCalculation Hamilton[16] estimates the minimum average commuting distances required by the monocentric modelfor 14 U.S. cities and 21 Japanesecities in the late 1970s. He shows that the minimum commute required by the monocentric the average distance of residents is

to the center minusthe averagedistance of jobs to the center. To estimate the required monocentric commute, Hamilton first calculates the average commute if all jobs were (,4) located in the center, A= -~f0rr~(r) ar, (,,

whereP is the metropolitantotal populationand P(r) is the populationdensity at distance r fromthe citys center. Hethen calculates the amount reduction (B) in average of commute job decentralization fromthe center, by
I

J .Io

(s)

whereJ is the metropolitan total employment J(r) is the employment and density at dis~ace r from the citys center. Finally, he calculates the required commute = A - B), (C whicl~is the differenceof the averagedistance of population distribution to the center and the averagedistance of employment the center. to Assuming both employment populationdensities decline exponentially from that and an urban center, Hamilton[16] found that the average minimum commuting distance is only 1.12 miles (C = 1.12 miles), based on the Mills-type density gradients whichare updated by Macauley [21]. But the average actual commute 8.7 miles in his samplecities. Hence, is 87 percent of the actual commute excess. Basedon this finding, Hamilton is concludesthat the monocen~c modelis inadequate in explaining actual commuting urban areas. in Applying Hamiltonscalculation to the samesampleof U.S. cities for 1980census data, Thurstonand Yezer[39] recornpute the minimum commute predicted by the monocentric model.In their calculation, heterogeneity(in term of occupationalgroups)

employment resident workersis considered, and individual density functions for each and occupationalgroup are estimated. Their results showthat about 60 percent of actual commute excess. Small and Song[36] apply Hamiltonscalculation to Los Angeles is Countyfor 1980journey-to-work census data. Theyfind that about four-fifths of the actual commute excess, confirmingthe general order of magnitude Hamiltonsoriginal is of estimate and verifying his original argument that the monocentric modelis very poor at explaining urban commuting.

Wh#esCalculation White[42] uses a linear programto calculate the minimum commute the actual for distributions of resident workersand employment. linear program used to assign trip A is flows among locations (swaphomesor jobs) so that the aggregate commute minimized, is knowing exogenous the numbers jobs and resident workersat each location and the of commuting costs (time or distance matrix) among locations. Let nij be the numberof commuters from zone i to zone j, ci~ be the

corresponding networkcommuting (travel distance or time). A linear programis used cost to find nii to , subjectto the constraints
Enij 2 = Ni , ~nly = E] , nlj ~ 0 , (i,j =i, 2 ..... I) , (7)

(6)

whereN=Z;.dVi=X;.,Ej the number commuters the urban area, while Ni is the is of in

number resident workers in zone i and E is the number jobs in zone j. of of i Applying calculation to the actual distributions of resident workers jobs, this and White[42] finds that the average mir~mum commuting time is 20 minutes for a sample whichoverlaps Hamiltons. Comparing average actual commuting the time of 22.5 minutes in these sample cities, she concludes that there is little excess commuting American in cities, with only 11 percent is excess. Whitetherefore concludesthat the urban commuting not is "wasteful" and claims "that the monacentric urban modelsare in better shape that Hamiltons gloomy diagnosis wouldimply." (p. 1109) Whitescalculation has been also used by I-Iamflton [17], Cropper Gordon and and [6] Small:and Song[36]. Usingtravel dis~~ce rather than travel time as by White, Hamilton [17] showsthat 47 percent of urban commuting excess in Bostonarea. Comparing is 16 percent found by Whitein the samearea, Hamilton concludesthat this discrepancyis mainly due to the difference betweenusing distance and using time as a measureof commuting cost Croppe.rand Gordon use a modifiedWhitescalculation to microdata in Baltimore, [6] whichreallocates householdsto houses to minimize aggregate commuting the distance subject to the ,constraint that no households utility is lowered.They find that about50 percent of the actual commute excess. Small and Song[36] apply Whites calculation to small-zone is 1980journey-to-work data for Los AngelesCounty.Theyfred that about two-thirds of the actual commute excess, with little difference between is using distance and using time; they also showthat Whites estimate of excess commute downward is biased due to aggregation bias fromusing large zonesin her data set. This paper uses both Hamiltonsand Whitescalculations to estimate the minimum

commute required by the monocentricand polycentric models. In principle+ Whites calculation can be used to anyestimatedset of density functions whichserves as the basis for calculating a minimum required commute ApplyingWhites calculation to the predicted monocentric densities, weare able to examine bias of Hamiltons the calculation due to its assumptions that there exists a radial transportation network commuting and trips proceed along radial routes. Applying Whitescalculation to the predicted polycentric densities, we are able to determinethe effect of polycentricity on the estimate of minimum required urban commuting.

STUDY AREA AND IDENTIFICATION OF EMPLOYMENT CENTERS

This section first describes the study area and data. It then showshowemployment centersare identified.

Study Area and Data Thestudy area contains mostof urbanizedportion of five counties in the greater Los Angeles region, namelyLos Angeles, Orange, Riverside, San Bemardino,and Ventura Counties. Journey-to-work data from the 1980Censusare used, that are provided by the Southern California Association of Governments (SCAG). data include aggregate zoneThe to-zone commute flows. Informationon zoneoto-zone travel distances is extracted from the data created for the UrbanTransportation Planning Package(UTPP), whichis calibrated

10

based on a peak-periodrepresentation of the road network. Thegeographicunit is the transportation analysis zone (AZ)defined by the SCAG. Like census tracts, Azs are aggregates of census blocks but have their boundariesdetermined by functional traffic characteristics and need not have a fixed population, and hencereduce the "census-tract delineation bias~ observedby Frankena ha density estimation. [8] Thestudy area consists of 1124Azs, after deleting 161 very low-densityzones for simplicityo3 The 1124zones cover 3,401 square miles. This paper analyzes the 4.53 rrfitlion workerswhoboth live and workin the 1124-AZ study area. Becausethe standard location modelonly considers resident workersand only employed individuals commute to work,this paper analyzesresident workersrather than population

Identification of Employment Centers A number peaks of densities mayexist in an urban area. A definition of of employment center, however,should capture the notion that the concentration of employment in a zone of a group of adjacent zones are large enoughto have somediscernible impact on the overall spatial structure. Recentempiricalstudies have provideda variety of criteria on defining employment centers. Gordon al. [12] identified density peaks via visual et inspectionof density maps.Heikkilaet al. [18] treat the centroids of the Regional Statistical Area (defined by a planning agency) as the proxies of subcenters. Cervero[3] and Dunphy [7] definedsubcenterswith a sequentialprocess, using criteria suchas the size of

:~AI1are remotefromthe highly developedparts of the region, with the exception of 11 largely undeveloped zones ha the Santa Monica mountain that separates the densely developed WestLos Angeles corridor from the moresuburbanSan FernandoValley. 11

employment specialization of employers. and McDonald discussedseveral empirical criteria for the identification of urban [23] employment subcenters. He suggested that local peaks in gross employment density and the employment-population ratio are the best hadicators of employment subcenters. GiulJanoand Small[10] use a version of McDonalds definition and present a simple systematic identification of employment subcenters. Theydefine a center as a contiguousset of zones, each with density aboveD, that together have at least E total emptoyment. Using980 censusjourney-to-work data for the Los Angeles region, they have identified 32 centers with criteria /9 = 10 employeesper acre and E = 10,000 employees. This paper uses the approachsuggestedby Giuliano and Small, becauseit incorporatesadjacent high-densityzonesand restricts attention to centers large enough to exert potentially significant influenceson the overall urbanstructure in a metropolitan area. In order to have a manageable number employment of centers in the density function estimation, criteria become = 15 and E = 35,000. Usingthese criteria, seven D employment centers are identified; they are listed in Table 1. Among seven centers, five the locate in Los AngelesCountyand two locate in OrangeCounty.

EMPIRICAL RESULTS

Monocentric Density Estimates The monocea~c density functions of resident workersand employment estimated are

12

Table 1. Employment Centers Identified in 1980 County Total Emp. 429869 208166 48510 47459 43761 37305 35675 Emp. Den. Dist. from (Emp/Acre) CBD(miles) 42.26 25.31 18.77 16.01 23.01 17.18 25.14 0.1 15.8 18.8 40.7 7.3 32.9 12.6

DowntownLA (CBD) UCLA/Santa Mortica LAAirport Orange Co. Airport West Hollywood Santa Ann Pasadena

L.A. L.A. L.A. Orange L.A. Orange L.A.

13

by ordinary least square, after taking logarithm of equation (I). Table 2 presents the results. As expected, the density gradient of employment greater than that of resident workers, is implying that resident workers are more dispersed than the employment.The gradient estimates showthat both resident worker and employment distributions were quite flat in 1980 in Los Angeles region; resident worker density falls off at 46 percent and employment density falls off at 5.0 percent by one mile increase in distance from the CBD.These two density gradients, however, are not muchdifferent. This result suggests that the degrees of employment decentralization and resident worker suburbanizafion are close, implying a general balance between housing and jobs over the distance from the CBD. Table 2 showsthat the monocentricdensity function fits the resident worker distribution better than the employment distribution, based on the criteria of maximum

explanatory powerin standard regression analysis and accuracy in predicting the total employmentand resident workers of the urbanized area (McDonaldand Bowman [24]). The monocentricfunction explains 38.6 percent of the variance in resident worker distribution; whereas it explains 34.3 percent in the case of employment distribution. Integrating the predicted densities over the whole region, I found that the monocentficmodelpredicts 99.1 percent of the total resident workers but only 79 percent of the total employment. To examinea possible crater at the CBD the case of resident worker distribution, in the monocentricdensity function is also fitted with a quadratic distance term addedto grin,
D~ = D~e-~~g~deu-, m = I, 2 ..... M, (8)

where g2 is the coefficient of the quadratic term When both ~ and 82 are negative, a crest of density is predicted at location of g/(2g2) miles from the CBD. results, The 14

Table 2. Estimates on MonocentricDensity Functions Intercept (logDo) Employment 8.5321" (0.0688) ResMent Worker 8.6574" (0.0586) 0.0457" (0.0017) 0.386 4,487 0.0498" (0.0021) 0.343 3,577 Gradient (g) a /i Integration (1000s)

Note:

Standarderrors are in parentheses. There are 1117and 1102observations for resident workerand employment equations. Thedata set has 4,528 thousandactual jobs (and resident workers). * Estimate statistically significantat 0.05level, l-sided test. is

15

however, show that ~=0.0222 and ~2=-0.0003, implying that the (predicted) resident worker density declining monotonically from the center. Hence, no crater is predicted at the center.

Polycentric Density Estimates

The polycentric density functions are estimated by non-linear least squares. Tables 3 and 4 show the results, with the Orange County Airport center excluded. Regressions on the

polycentric density functions reveal that the Orange CountyAirport center has very low tvalues on its intercept and gradient estimates. For example, t-values are less than 0.52 for the employment equation. Moreover, this center has a negative intercept in the resident worker equation; F-test (discussed below) showsthat the six-center modelcannot be rejected even at a 25 percent significance level in the case of employment distribution. For these reasons, the OrangeCountyAirport center is eliminated. Tables 3 and 4 suggest the existence of polycentricity in the Los Angeles region. Consider the number centers with both intercept and gradient statistically of significant at a 5

percent level (1-sided test, i.e., with t> 1.64). Five centers pass tiffs test in the case resident workerdistribution (Table 3); four centers satisfied this criterion in the case employment distribution, with Pasadena close to the margin (Table 4).

The estimates of total impact, IMPACT, that each center n has on the regionwide distributions of resident workersand employment also reveal the polycentricity in the overall spatial structure. Theyare shownin the last columnof Tables 3 and 4, with t-values in

16

Table 3. Estimates on Polycentric Density Function; Resident Workers

Center Location

Intercept

Gradient

Impact (1000s)

Downtown LA (CBD)

2125" (2.76) 2673" (3.53) 2517" (2.80) 6936" (7.63) 2493" (6.49) 701 (1.08)

0.3361" (1.87) 0.0665" (1.69) 0.0270" (2.15) 0.2484* (4.86) 0.0614 (2.83) 0.0344 (0.55)

58.1 (1.09) 937.4 (1.46) 2789.6" (2.69) 311.1" (2.63) 1269.0" (2.24) 707.9 (0.71)

UCLA/SantaMortica

LA Airport

West Hollywood

Santa Aria

Pasadena

2 R = 0.481

Note: t-values are in parentheses. There are 1 I24 observations. * Estimateis statistically significant at 0.05 level, 1-sided test.

17

Table 4. Estimates on Polycentric Density Function: Employment

Center Location

Intercept

Gradient

Impact (1000s)

Downtown LA (CBD)

254990" (35.03) 76020" (1.69) 3925" (5.09) 12539" (3.16) 6467 (1.19) 33516 (1.51)

1.2199" (38.98) 1.5472" (2.22) 0.0234" (3.30) 0.2910" (3.15) 0.3114 (1o31) 1.4203" (1.67)

422.5" (21.19) 39.5" (2.07) 4957.6" (5.92) 400.8" (2.08) 211.0 (1.02) 33.2 (1.28)

UCLA/Santa Monica

LA Airport

West HoUywood

Santa Aria

Pasadena

R2 = 0.705

Note:t-values are in parentheses. There are 1124 observations. * Estimateis statistically significant at 0.05 level, I-sided test.

18

4 parentheses. Theresults show that three centers have statistically significant impactson resident workerdistribution and four centers on the employment distr/bution. Onemightthink that distance to the oceanhas large impactat least on resident workerdistribution, and that the large impactof the Los Angeles Airport center is due to its location on the oceanshore. Estimationon the resident workerequation with inclusion of dis~ce to the ocean, however,showsthat the oceanhas a very localized impact; its density gradient equals -6.7979 per mile (;=4.37), implyingthat its influence on resident worker distribution fails off 99.9 percent for one mile increase fromthe ocean shore. Moreover, the LosAnglesAirport center still has the largest impact, althoughits intercept decreasesfrom 2517to 2205and its gradient increases from0.0270to 0.0335. Theformaltest for the existence of polycentricity, however, basedon the statistic is E = (SSa" SSR~)/q SSR (M-p) ~/ whereSSt~ and SStP are the restricted (monocentric)and unrestr/cted (polycentric) of squared residuals, with samedependentvariable (density D); Mis the samplesize; is the number parametersbeing estimated in the unrestricted estimate; and q is the of number restrictions on these parameters the restricted estimate. Underthe null of in hypothesis, F is approximately distributed accordingto a central F-distribution with (9)

*IMPACT. a nonlinear function of two randomvariables, ~ and bo. Knowing is their estimated variances and covariance we can computean approximatevariance for IMPACT,~ from the formula o~A~ = d~d. where~ is the variance-covariancematrix of (A,, b,) and d is the vector of derivatives equation (3) with respect to A~ and b,. See Chow pp. 182-183. [4] 19

degrees of freedom (q, M-p),5 where p=2N,q=2(N-l), and is the(unrestricted) number centers (Small and Song[37]). of Performing F-tes~ to the unrestricted (polycentric) modelwith six centers, I find that the F-statistic has values of 27.68 for resident workerdistribution and 25.11 for employment distribution. Theseresulting tests, with (10,11 I2) degreesof freedom, indicate that the null hypothesis(monocentric model)is soundlyrejected at a significance level of 0.0001in both ~ cases. Thepolycentricmodel,therefore, explainsstatistically better the distributions of resident workersand employment. Rejecting the monocentricmodelindicates that the overall access to major employment centers is much moreimportant in location choices than the accessto the central businessdistrict.

Monocentric Required Commute UsingHamiltonscomputation taking these monocentric and functions to represent smoothly varyingdistributions, I find that the averageworkerlives at 25.02 miles (A=25.02, equation 4) fromthe center and the average job locates 24.00 miles (B=24.00,equation from the center. The difference, C=1.02 miles, is the average minimum commute required by the monocentric model.It accounts for only about one-tenth of the average actuM commute 10.81 miles; i.e., 90.56 percent of actual commute excess. The first rowin of is Table 5 showsthe result, verifying Hamiltons original finding that the standard monocentric

5See Chow [4], pp. 229-230and Gallant [9], pp. 78-79. 6With(10,1112)degreesof freedom,the F-valuefor rejecting the null hypothesisat 0.0001 significancelevel is 3.5564 20

model greatly underpredicts the actual commuting observed in urban areas. Hamiltons computation assumesthat there exists a radial transportation network which is everywhere dense. The actual road network, however, is not ubiquitous; commutingtrips do not always proceed along radial routes. The calculation of minimum required commute with equations (4) and (5) might therefore be downward biased. determine the effect of actual road network on the estimate of excess commuting, also I calculate the minimum commuterequired by the monocentric model by applying Whites calculation to the predicted densities of employment resident workersinstead of actual and derisity pattern in previous studies (Hamilton [17], Small and Song[36] and White [42]). this case, a l~eax programis used to assign trip flows among zones so that the aggregate commuting distance is minimized, given the predicted numbersof jobs and resident workers 7 each zone. at The second row in Table 5 presents the results, showing that the minimum required commute 1.99 miles and about 81 percent of the average actual commute excess. The is is actual road network, therefore, results in a minimum commute that is about twice as large as that obtained by Hamiltons computation, increasing from 1.02 miles to 1.99 miles. This finding indicates that the assumptionof radial aansportation networkconsiderably underpredicts the minimum commuterequired by the monocentric model.

7The monocentric density functions have different accuracies in predicting the total employmentand resident workers. A linear program, however, requires same aggregation for employment resident workers. Here I scale the predicted aggregation downor up to match and the actual total (which is samefor both employment resident workers). and 21

Table 5. Estimates of AverageRequired and Excess Commute Required Excess

(miles)

(~)

Monocentric Model" b Monocentric Model b Polycentric Model

1.02 1.99 4.35

90.56 81.59 59.76

distance is 10.81 miles. Note: The average actual commuting "Estimateswith Hamiltonscalculation. b Estimateswith Whitescalculation.

22

Polycentric Required Commute The polycentric required commute estimated by applying Whitescalculation to the is predicted densities of the polycentric functions. Thethird rowin Table 5 presents the results, showingthat the polycentric modelhas a minimum average required commute of 4.35 miles. This result indicates that the polycentric required commute considerablylarger is than that required by the monocentric model.The latter requires 1.02 miles by Hamiltons calculation and 1.99 miles by Whites calculation. Comparing with the monocentric excess commutes 90.6 percent and 81.6 percent, the polycentric modelhas a much of smaller amount excess commute, of with less than 60 percent of the actual commute being excess. Hence,it explains the observedcommuting patterns much better than the monocentricmodel. Thesefindings have two implications. First, the polycentric modelrequires more urban commuting the monocentric than model.Put differently, polycentricity has a positive effect on the estimate of minimum required urban commute. This implication is consistent with the argument Sub[38] andWhite[42] but conflicts with that of Gordon al. [I 1] of et and Gordon Wong and [13]? As shownearlier, the monocentricdensity gradients are quite simil~a"for employment resident workerdistributions, indicating that the degreesof and employment decentralization and resident workersuburbanizationare close. This general balance betweenhousing and jobs, in turn, requires a smaller amountof minimum

SWithoutquestion, Gordon al. [11] and Gordon Wong et and [13] are correct within the %trong" monocentric model that assumes all jobs are located at the CBD.The standard monocentric model, however, has incorporated the assumption that some employmentis decentralized. Mills and Hamilton[28] and Muth[32] showthat the standard monocentric predictions remainregarding the rent function and density function, assuming that firms pay wagescompensating commuting the costs. 23

commuting than the polycen~cmodel. Hence, it is the decentralization of employment rather than the polycentric structure that makes possible to reducethe urban commuting. it The second implication suggests that householdsdo makeattempts to economize on commuting. Like the standard monocentricmodel, the polycentdc modelassumesthat accessibility to workplace the primarydeterminant the residential location choices. is of Since it is shown earlier that the polycentric modelis superior to the monocentric model in explainingthe spatial structure~ the trade-off assumption expectsthat the polycentric model is also superior to the monocentric modelin explaining the actual urban commute. The findings that the polycentric modelexplains much better both the actual commuting distance and the actual distribution patterns, therefore, supportsthe assumption the standardurban in modelregarding the commuting behavior. It is worthnoting that Hamilton [16] also presents evidencewhichsupports the assumptionthat urban householdseconomize their commuting. on Supposingthat households are indifferent to commuting select their homes job sites at random,Hamilton and and found that the meancommute 12.09 miles. Comparing is with the average actual commuting distance of 8.7 miles, the actual commute less than 72 percent of the commute would is that emergeif householdschose their homes job sites at random and Clearly, urban households makeattempts to economizeon commuting.

24

CONCLUSIONS AND IMPLICATIONS

Analyzing the I980 journey-to-workCensusdata in the Los Angelesregion, this of paperhas shown that the polycentricdensity functionsfit the actual spatial structure statistically better than the monocentric model.For both resident workerand employment distributions, the monocentric model rejected at a significancelevel of 0.0001in favor of is the polycentric model.This result suggeststhat polycentricity clearly exists in large urban areas and indicates that the overall access to major employment centers is much more important location choicesthan the access to the central businessdistrict. in This paperhas determined positive effect of polycentricity on the estimate of a minimum requ/red commute, comparingthe monocentricand polycentric required by commutes. results showthat the monocentricmodelrequires a small amountof The commute: 1.02 miles by Hamiltonscalculation and 1.99 miles by Whites calculation; whereasthe polycentdc modelrequires 4.35 miles. Comparing with the actual average commute 10.81 miles, they account for 9.4 percent, 18.4 percent, and 40 percent of the of actual averagecommuting distance. Theseresults indicate that the polycentric structure increases the estimate of minimum requh-edurban commuting, that the polycentfic model and explains better the observedcommuting pattern than the monoeentric model. Hamiltons original finding, that the monocentric modeldoes a very bad job in predicting the actual commuting behavior, is moreaa indictment of the monocentricity assumption than a rejection of the assumption commuting on behaviorin location choices. Themonocentficmodelgreatly underpredicts the actual commute because it inadequately

25

represents the spatial structure in large urban areas. Harniltonscalculation, whichassumes there exists a radial transportation network commuting and trips proceedalong radial routes, also contributes to this bias. Once polycentricity is substituted for monocentricity the and actual road networkis used, actual commuting behavior is much better explained. Thefindings that the polycentric modelexplains much better both the observed commuting distance and the actual distribution patterns help to preserve the assumption that urban households makestrong attempts to economize commuting. turn, they suggest on In the policy implication that efforts to promotemoreefficient urban form, such as the jobshousingpolicy mayhave the potential to succeed.

26

REFERENCES

[1]

W. Alonso. Location and Land Use. Harvard University Press, 1964. R. Bussiere and F. Snickars. Derivation of the Negative Exponential Model By an Entropy Maximizing Method. Environment and Planning A 2: 295-301, 1970. R. Cervero. Americas Suburban Centers: The Land Use Transportation Link. Unwin Hyman, 1989. G.C. Chow. Econometrics. McGraw-Hill Book Company, 1983.

[2] [3]
[4]

[51
[6]

C. Clark. UrbanPopulation Densities. Journal of Royal Statistical Society, Serial A, 114: 490-494, 1951. M. L. Cropper arid P. L. Gordon. Wasteful Commuting:A Re-Examination. Journal of Urban Economics 29(1): 2-13, 1991. R. T. Dtmphy. Defining Regional EmploymentCenters in an Urban Area. Transportation Research Record 861: 13-15, 1982. M. W. Frankena. A Bias in Estimating Urban Population Density Functions. Journal of Urban Economics 5: 35-45, 1978. A.R. Gallant. Nonlinear Regression. The AmericanStatistician 29(2): 73-81, 1975.

[7] [8]
[9] [10]

G. Giuliano and K. A. Small. Subcenters in the Los Angeles Region. Regional Science and Urban Economics 21: 163-182, 1991. P. Gordon, A. Kumar, and H. Richardson. The Influence of Metropolitan Spatial Structure on CommutingTime. Journal of Urban Economics 26: 138- 151, 1989. , H. Richardson and H. L. Wong.The Distribution of Population and Employment a Polycentric City: The Case of Los Angeles. Environment and in Planning A 18: 161-173, 1986. and H.L. Wong. The Costs of Urban Sprawl: Some NewEvidence. Environment and Planning A 17: 661-666, 1985. D. L. Greene and J. Barnbrock. A Note on Problem in Estimating Exponential Urban ]Density Models. Journal of Urban Economics 5: 285-290, 1978. Modelling Urban Population Density in a Multi-Centered City. 27

[11]

[12]

[13]

[14]

[15] ]D. A. Griffith.

Journal of Urban Economics 9: 298-310, 1981o [16] B. W. Hamilton. Wasteful Commuting.Journal of Political 1982. . Wasteful Commuting Again. Journal of Pol#ical 1989. Economy90: 1035-1053,

[17]

Economy 96: 1497-1504,

[18]

E. Heikkila, P. Gordon, J. I. Kim, R. B. Peiser, H. Richardson and D. DaleJohnson. What Happenedto the CBD-DistanceGradient?: Land Values in a Policentric City. Environment and Planning A 21: 221-232, 1989. R. W. Helsley and A. M. Sullivan. Urban Subcenter Formation. Regional Science and Urban Economics 21: 255-275, 1991. P. Kemperand R. Schmenner. The Density Gradient for Manufacturing Industry. Journal of Urban Economics 1: 410-427, 1974. M. K. Macauley. Estimation and Recent Behavior of Urban Population and EmploymentDensity Gradients. Journal of Urban Economics 18: 251-260,

[19]

[20]
[2i]

1985.

[22]

J. F. Mcdonald. The Intensity of Land Use in Urban Employment Sectors: Chicago 1956-1970. Journal of Urban Economics 18: 261-277, 1985o . The Identification of Urban Employment Subcenters. Economics 21: 242-258, 1987. Journal of Urban

[23]
[24]

and H.W. Bowman.Some Tests of Alternative Urban Population Density Functions. Journal of Urban Economics3: 242-252, 1976. E.S. Mills. An Aggregative Model of Resources Allocation in a Metropolitan Area. American Economic Review 57: 192-210, 1967. . The Value of Urban Land. In H. S. Perloff, Environment. Johns Hopkins Press, 1969. editor, The Quality of the Urban

[25]

[26]
[27]

. Studies in the Structure of the UrbanEconomy.Johns Hopkins Press, 1972. and B. W. Hamilton Urban Economics. Scott, Foresman and Co., fourth edition, 1989. and K. Ohta. Urbanization and Urban Problems. In H. Patrick and H. Rosovsky, editors, Asias New Giant: Howthe Japanese EconomyWorks. The Brookings Institution, 1976. 28

[281 [29]

[30] R.F. Muth. Cities and Housing. University of Chicago Press, 1969. [3i] . Numerical Solution of Urban Residential Urban Economics 2: 307-332, 1975. Land-Use Models. Journal of

[32]

. Models of Land-Use, Housing, and Rent: An Evaluation. Regional Science 25(4): 593-606, 1985.

Journal of

[33]

3. Odland. The Conditions for Multi-Centered Cities. Economic Geography54: 234245, 1978. Y. Y. Papageorgiou and D. Pines. The Exponential Density Function: First Principles, Comparative Static, and Empirical Evidence. Journal of Urban Economics 26: 264-268, 1989. K. Sasaki. The Establishment of a Subcenter and Urban Spatial Structure. Environment and Planning A 22: 369-383, 1990. K. A. Small and S. Song. Wasteful Commuting:A Resolution. Journal of Political Economy 100: 888-898, 1992. . Population and Employment Densities: Structure and Change. Working Paper, Departmentof Economics, University of California, Irvine, 1992. S. H. Suh. Wasteful Commuting:An Alternative Economics 28: 277-286, 1990. Approach. Journal of Urban

[341

[351
[36]

[371 [381
[39]

L. Thurston and A. M. Yezer. Testing the Monocentric Urban Model: Evidence Based on Wasteful Commuting. AREUEA Journal 19: 41-51, 1991. W. C. Wheaton. A Comparative Static Analysis of Urban Spatial Structure. Journal of Economic Theory 9: 223-237, 1974. White. Firm Suburbanization and Urban Subcenters. Journal of Urban Economics 3: 323-343, 1976. Urban Journey Are Not Wasteful. Journal of Political Economy 96: 10971110, 1988. . Commuting and Congestion: A Simulation Model of a Decentralized Metropolitan Area. AREUEA Journal 18: 335-368, 1990.

[401

[411M. J. [421
[43]

29