1# RECORD ddfcdefd81b294b9edfe4b97744d2d4725aeed232# TITLE: Urban location and housing prices within a hedonic model3# YEAR: 2008.0 VENUE: Journal of Regional Analysis & Policy DOI: 10.68479/001c.94644# AUTHORS: J. Ottensmann; Seth B. Payton; J. Man5# PAGES: 17 CANDIDATE TABLE PAGES (1-based): [4, 8, 9, 12, 14]6# ABSTRACT: A measure of location relative to employment is often included in hedonic housing price models. This is most often distance to the center, based on the monocentric model, which does not consider the decentralization of employment in urban areas. This paper tests the performance of alternative measures of location, considering both distance and time to the center and to multiple employment centers and measures of accessibility to employment and change in accessibility. The measures using multiple employment centers and accessibility perform better than simple distance to the center, with the combination of accessibility to employment and change in accessibility doing best.789===== PAGE 1 (score 6, numeric-paren 0, table-words 0, var-words 6) =====1011 12Urban Location and Housing Prices within a Hedonic Model 13 14 15John R. Ottensmanna, Seth Paytona, and Joyce Manb 16Indiana University–Purdue University Indianapolis a and Indiana University Bloomington b - USA 17 18 19 20Abstract. A measure of location relative to employment is often included in hedonic housing price 21models. This is most often distance to the center, based on the monocentric model, which does 22not consider the decentralization of employment in urban areas. This paper tests the perfor-23mance of alternative measures of location, considering both distance and time to the center 24and to multiple employment centers and measures of accessibility to employment and change 25in accessibility. The measures using multiple employment centers and accessibility perform 26better than simple distance to the center, with the combination of accessibility to employment 27and change in accessibility doing best. 28 29 30 311. Introduction 32 33Location within an urban area, particularly with 34respect to the location of employment, is assumed to 35be a determinant of land prices within standard urban 36economic models. And because land prices, ceteris pa-37ribus, will affect house prices, one or more measures of 38urban location are frequently included in hedonic 39housing price models. 40The standard urban economic model (Alonso 411964; Mills 1972; Muth 1969) makes the monocentric 42assumption that employment is located in the Central 43Business District (CBD). Following this, distance to the 44CBD has frequently been used as a measure of loca-45tion. With the continuing movement of employment in 46metropolitan areas away from the CBD, distances to 47multiple employment centers or even to all employ-48ment in the form of an accessibility measure is increa-49singly used as an alternative. 50This paper examines the performance of alterna-51tive measures of urban location within a hedonic 52housing price model for Indianapolis, Indiana. The 53goals of the research are to understand better how res-54idence location relative to employment location affects 55house prices and to provide guidance as to the most 56effective method for specifying location within hedon-57ic models. The research systematically compares 58measures of location using both distances and travel 59times, to the CBD, to multiple employment centers, 60and used in measures of accessibility to employment. 61The significance of changes in accessibility to em-62ployment is also assessed. 63 642. Location in Hedonic Models 65 66In their seminal research developing a hedonic 67model for the prediction of house prices, Kain and 68Quigley (1970) included structural characteristics of 69the housing unit, neighborhood characteristics, and 70distance to the CBD. Reviews of hedonic models (Bo-71wen et al. 2001; Malpezzi 2003) have described such 72models as including housing structure characteristics, 73the social and natural environment (neighborhood 74characteristics), and location within the market. 75Heikkila et al. (1989) cited numbers of studies of 76determinants of residential property or land values 77using hedonic models, saying that to the extent they 78have included location, it has generally been distance 79to the CBD, using the assumption of a monocentric 80city employed in the standard urban economic model 81(Alonso 1964; Mills 1972; Muth 1969). The early model 82developed by Kain and Quigley (1970) included dis-83tance to CBD, but it was not statistically significant. 84Witte et al. (1979), in one of the first attempts to expli-85citly apply Rosen’s (1974) theory of implicit markets to 86the development of a hedonic housing price model, 87included distance to the CBD among the neighbor-88hood characteristics, and this loaded heavily on an 89JRAP 38(1):19-35. © 2008 MCRSA. All rights reserved. 90 91 929394===== PAGE 4 (score 30, numeric-paren 0, table-words 2, var-words 22) =====959622 97 Ottensmann et al. 98Table 1. Descriptive Statistics 99 100 101Standard 102Variable 103Mean 104Deviation 105 106 107 108Sales price 109110,888 11078,492 111Log sales price 11211.437 1130.621 114 115 116 117Number of bathrooms 1182.022 1190.872 120Total number of rooms 1217.152 1221.863 123Floor space (100s sq ft) 12416.371 1257.592 126Has basement 1270.416 1280.493 129Age 13037.276 13126.882 132Lot area <= 0.5 acre 1330.843 1340.364 135Lot area >= 1.0 acre 1360.033 1370.179 138Garage bays 1391.622 1400.742 141Has porch or deck 1420.562 1430.496 144Exterior brick or stone 1450.603 1460.489 147No air conditioning 1480.138 1490.345 150Effective tax rate 1511.185 1520.783 153School district SAT 154988.491 15536.466 156Neighborhood median income 15750,837 15818,166 159Neighborhood percent black 16020.736 16121.746 162Neighborhood percent vacant 1637.529 1644.161 165 166 167 168Distance to CBD (kilometers) 16911.710 1704.880 171Free-flow travel time to CBD 1728.670 1733.793 174Congested travel time to CBD 1758.630 1763.937 177 178 179 180Mean distance to 7 ZIP centers 18115.459 1823.923 183Mean distance to 5 TAZ centers 18415.840 1854.190 186Free-flow travel time to 5 TAZ cen-187ters 18812.937 1893.329 190Congested travel time to 5 TAZ 191centers 19212.581 1933.223 194 195 196 197 198 199The 200remaining 201neighborhood 202characteristics, 203neighborhood median income, percentage of the pop-204ulation African-American, and percentage of housing 205units vacant, have been derived from data from Cen-206sus 2000 (U.S. Bureau of the Census 2006). Most often, 207census neighborhood characteristics have been in-208cluded in hedonic models using values for the census 209 210al Progress (ISTEP), which is taken by all students. The SAT score 211variable was selected because it had a slightly higher level of signi-212ficance in the model. 213tract or block group in which the housing unit is lo-214cated. This approach has two problems. First, a house 215could be located near the boundary of a tract or block 216group so the relevant neighborhood characteristic is as 217much the characteristic of an adjacent area as the cha-218racteristic of the area in which the house is located. 219Second, census tracts and block groups vary greatly in 220their spatial extent so census tract or block group 221measures represent characteristics extending at vary-222ing distances from the housing unit. 223To address these problems in assigning census 224neighborhood characteristics, the model includes es-225timates of these values for neighborhoods of a uniform 226one-mile radius surrounding each of the housing 227units. These neighborhood variables were estimated 228from the census block group data, with the values 229being the means of the census characteristics for the 230block groups intersecting the one-mile radius neigh-231borhoods weighted by the proportion of the one-mile 232neighborhood area within each block group. Descrip-233tive statistics for these neighborhood characteristics 234are included in Table 1. 235Several of the models tested include an additional 236neighborhood variable, location within Center Town-237ship. Center Township is the older, inner-city portion 238of the urban area (see Figure 1 for the location). It is 239recognized locally as being a generally less-desirable 240area of the city. 241 242 243 244Figure 1. Location of Center Township 245246247===== PAGE 5 (score 6, numeric-paren 0, table-words 0, var-words 6) =====248249Location in a Hedonic Model 250 251 23 252 253 2543.2 Employment Data 255 256The analyses test location measures related to the 257location of employment. Two sources of employment 258data are used, one with employment for the traffic 259analysis zones (TAZs) used for transportation plan-260ning and a second with employment for ZIP code 261areas. Marion County is the inner part of the larger 262Indianapolis metropolitan area. Because accessibility 263to employment outside of Marion County is relevant 264to residents of the county, both sets of employment 265data include employment for Marion County and for 266the surrounding 8 counties. This area includes all of 267the counties in the Metropolitan Statistical Area (MSA) 268as defined in 2000 along with one additional county, 269Madison County, a separate MSA deemed potentially 270significant for employment accessibility. 271The first dataset has employment in 2000 for the 2721285 TAZs in the 9-county area. These data were de-273veloped from the ES-202 workforce data collected by 274the Indiana Department of Workforce Development. 275The data were assembled and provided by the Indiana 276Department of Transportation. The boundaries of the 277TAZs are shown in Figure 2. In addition to distance, 278the analyses also use the free-flow and congested road 279travel times between each pair of TAZs.2 These travel 280times were developed and provided by Cambridge 281Systematics, Inc. using a travel demand model origi-282nally created by the Indianapolis Metropolitan Plan-283ning Organization. These are travel times from the 284nodes within each TAZ at which trips to and from the 285TAZ are loaded onto the road network. The analyses 286involving travel times from houses to TAZ employ-287ment use the travel times from the TAZ node closest to 288the house to the employment TAZ nodes. For the ana-289lyses involving distance, the distances are the airline 290distances from the actual house locations to the em-291ployment TAZ nodes. 292The second employment dataset includes em-293ployment for both 1995 and 2000 for 100 ZIP code 294areas in the same 9-county area. These are also from 295the ES-202 workforce data collected by the Indiana 296Department of Workforce Development. The Indiana 297Business Research Center at IUPUI created the tabula-298tions of these data by ZIP code for the Center for Ur-299ban Policy and the Environment.3 The boundaries of 300 3012 The free-flow travel times are the times for the shortest paths over 302the road network based upon the travel speeds on the network links 303under non-congested conditions. The congested travel times are the 304shortest paths taking into account the assignment of all peak period 305trips to the network, with the consequent reduction in travel speeds 306on some links as a result of the assigned trip volumes. 3073 Employment for non-spatially-extensive ZIP codes such as post 308offices and individual firms and buildings (point locations) were 309the ZIP code areas are shown in Figure 3. Distances of 310house sales to ZIP code employment are the direct air-311line distances from the house locations to point loca-312tions within each ZIP code.4 313 314 315 316Figure 2. TAZ Boundaries and Employment Centers 317 318 319The two employment datasets are used for several 320reasons. Only the ZIP code dataset includes employ-321ment at two points in time, allowing examination of 322the effect of recent changes in accessibility to employ-323ment. The travel time data correspond to the TAZ da-324ta, requiring the use of those data for the analyses in-325volving travel time. Comparison of the distance ana-326lyses using the two datasets also provides for consid-327eration of the effect of the spatial resolution of em-328ployment data on the performance of measures of lo-329cation. 330 331 332added to the employment for the ZIP code areas in which those 333points were located. 3344 The point locations used for the ZIP codes are from the file of 5-335digit ZIP points provided by ESRI with the ArcView software rather 336than the ZIP code centroids. Prior research has found that these 337points correspond more closely to the centers of the population and 338employment distribution within the ZIP code areas than the centro-339ids. No documentation is provided, however, as to exactly how 340those point locations were established. 341342343===== PAGE 8 (score 179, numeric-paren 139, table-words 2, var-words 22) =====34434526 346 Ottensmann et al. 347 348Table 2. Models with and without Distance to CBD and Center Township (t-values) 349 350 351Model 1 352Model 2 353Model 3 354Model 4 355Independent 356Base 357CBD 358Center 359Center + 360Variables 361Model 362Distance 363Township 364CBD 365 366 367 368 369 370Number of bathrooms 3710.1104* 3720.1103* 3730.1132* 3740.1121* 375 376(21.19) 377(21.14) 378(22.08) 379(21.88) 380Total number of rooms 3810.0130* 3820.0131* 3830.0110* 3840.0114* 385 386(6.15) 387(6.16) 388(5.28) 389(5.48) 390Floor space (100s sq ft) 3910.0211* 3920.0210* 3930.0214* 3940.0214* 395 396(34.06) 397(34.04) 398(35.23) 399(35.21) 400Has basement 4010.0974* 4020.0970* 4030.1087* 4040.1036* 405 406(15.44) 407(15.13) 408(17.43) 409(16.43) 410Age 411-0.0024* 412-0.0024* 413-0.0021* 414-0.0024* 415 416(15.71) 417(14.65) 418(14.17) 419(15.07) 420Lot area <= 0.5 acre 421-0.0492* 422-0.0493* 423-0.0416* 424-0.0426* 425 426(6.23) 427(6.24) 428(5.35) 429(5.49) 430Lot area >= 1.0 acre 4310.0956* 4320.0959* 4330.0951* 4340.0996* 435 436(6.10) 437(6.11) 438(6.18) 439(6.47) 440Has garage 4410.0855* 4420.0855* 4430.0842* 4440.0843* 445 446(19.81) 447(19.81) 448(19.83) 449(19.88) 450Has porch or deck 4510.0477* 4520.0478* 4530.0456* 4540.0470* 455 456(8.66) 457(8.67) 458(8.42) 459(8.68) 460Exterior brick or stone 4610.0840* 4620.0841* 4630.0777* 4640.0784* 465 466(13.95) 467(13.96) 468(13.12) 469(13.25) 470No air conditioning 471-0.2511* 472-0.2512* 473-0.2444* 474-0.2451* 475 476(29.12) 477(29.11) 478(28.80) 479(28.91) 480Effective tax rate 481-0.2393* 482-0.2393* 483-0.2385* 484-0.2376* 485 486(70.28) 487(70.20) 488(71.24) 489(71.01) 490School district SAT 4910.0004* 4920.0004* 4930.0003* 4940.0003* 495 496(5.56) 497(5.45) 498(4.60) 499(3.66) 500Neighborhood median income 5010.0051* 5020.0051* 5030.0049* 5040.0058* 505 506(22.36) 507(17.88) 508(22.15) 509(20.54) 510Neighborhood percent black 511-0.0005* 512-0.0005* 513-0.0009* 514-0.0010* 515 516(3.31) 517(3.31) 518(6.73) 519(6.99) 520Neighborhood percent vacant 521-0.0216* 522-0.0216* 523-0.0135* 524-0.0129* 525 526(22.34) 527(22.34) 528(12.74) 529(12.08) 530 531 532 533 534 535Center Township 536 537 538-0.1748* 539-0.1886* 540 541 542 543(17.50) 544(18.25) 545Distance to CBD 546 547-0.0004 548 549-0.0055* 550 551 552(0.35) 553 554(5.10) 555 556 557 558 559 560R2 5610.8579* 5620.8579* 5630.8627* 5640.8631* 565 566 567 568 569 570*Significant at p<0.001 571 572 573 574 575 576and without the Center Township dummy. The subs-577tantive conclusions one would make regarding the 578performance of the alternative location measures are 579the same in nearly all instances. Since the focus of this 580paper is on the performance of the measures of loca-581tion, the results without the additional measure of lo-582cation provided by the Center Township variable are 583reported. 584585586===== PAGE 9 (score 25, numeric-paren 0, table-words 6, var-words 6) =====587588Location in a Hedonic Model 589 590 27 591 592 593Travel times to the CBD provide alternative mo-594nocentric measures of location that may more accu-595rately represent accessibility to employment. Since 95 596percent of workers in Marion County who commuted 597to work in 2000 used cars or trucks, and only 2.3 per-598cent took public transit, travel time on the road net-599work is the relevant measure for nearly all workers 600(U.S. Bureau of the Census 2006). Free-flow travel 601times are those estimated for conditions under which 602traffic volumes do not adversely affect speeds. Con-603gested travel times are estimated for the peak period, 604taking into account traffic volumes that reduce speeds. 605Congested travel times thus may be more appropriate 606for accessibility to employment, with a high propor-607tion of commuting taking place during the peak pe-608riods. 609Results for distance, free-flow travel time, and 610congested travel time to the CBD are presented in the 611top portion of Table 3. (From this point forward, re-612sults are presented only for the location variables and 613the overall R2 value for the model, as the regression 614coefficients for the other variables in the model are not 615significantly affected by the location measures.) As 616observed above, distance to the CBD was not signifi-617cant. However, both free-flow and congested travel 618times to the CBD were significant and have the ex-619pected negative signs for the regression coefficients. In 620addition, the regression coefficient and t value for 621congested travel time are twice as large as for free-622flow travel times. So travel times to the CBD (especial-623ly congested times) are significantly better predictors 624of house prices and better measures of accessibility to 625employment. 626 627 628Table 3. Models for Distance/Travel Time to CBD and Multiple Employment Centers 629 630Regression 631Standard 632 633 634 635Location Variable(s) 636Coefficient 637Error 638t 639Sig 640R2 641 642 643 644 645 646 647Base Model 648 649 650 651 6520.8579 653 654 655 656 657 658 659Distance/Travel Time to CBD 660 661 662 663 664 665Distance to CBD 666-0.0004 6670.0011 668-0.3471 6690.7285 6700.8579 671Free-flow travel time to CBD 672-0.0033 6730.0013 674-2.6385 6750.0083 6760.8580 677Congested travel time to CBD 678-0.0066 6790.0012 680-5.2895 6810.0000 6820.8583 683 684 685 686 687 688 689Distance/Travel Time to 3 TAZ Employment Centers 690 691 692 693Distance to CBD 6940.0005 6950.0012 6960.4542 6970.6497 6980.8605 699Distance to Keystone/Castleton 700-0.0008 7010.0006 702-1.2444 7030.2134 704 705Distance to West 86th Street 706-0.0048 7070.0006 708-7.9205 7090.0000 710 711Free-flow travel time to CBD 712-0.0032 7130.0014 714-2.3359 7150.0195 7160.8608 717Free-flow time to Keyston/Castleton 718-0.0015 7190.0006 720-2.3661 7210.0180 722 723Free-flow time to West 86th Street 724-0.0060 7250.0006 726-9.2329 7270.0000 728 729Congested travel time to CBD 730-0.0050 7310.0013 732-3.8608 7330.0001 7340.8610 735Congested time to Keystone/Castleton 736-0.0010 7370.0006 738-1.7612 7390.0782 740 741Congested time to West 86th Street 742-0.0063 7430.0007 744-9.4799 7450.0000 746 747 748 749 750 751 752 753Mean Distance/Travel Time to Employment Centers 754 755 756 757Mean distance to 7 ZIP centers 758-0.0096 7590.0008 760-11.6072 7610.0000 7620.8601 763Mean distance to 5 TAZ centers 764-0.0089 7650.0008 766-11.8679 7670.0000 7680.8601 769Mean free-flow time to 5 TAZ centers 770-0.0129 7710.0010 772-13.4577 7730.0000 7740.8608 775Mean congested time to 5 TAZ centers 776-0.0138 7770.0010 778-14.0707 7790.0000 7800.8610 781 782 783 784 785 786 787788789===== PAGE 10 (score 12, numeric-paren 0, table-words 2, var-words 8) =====79079128 792 Ottensmann et al. 793 7944.2. Distance/Travel Time to Employment Centers 795 796The increasing decentralization of employment 797within metropolitan areas has led to the questioning of 798the continued appropriateness of the monocentric as-799sumption of the standard model. Observation of the 800emergence of multiple employment centers within 801metropolitan areas has led to suggestions that accessi-802bility to employment in those multiple centers would 803provide a better measure of location within hedonic 804models. 805With distances or times to multiple employment 806centers, the issue arises as to how those centers would 807collectively influence house prices. Small and Song 808(1994) and Anas et al. (1998) lay out three alternatives 809(focusing on population densities rather than house 810prices). Assume that the effect of any employment 811center on house prices is an estimated house price val-812ue at that center times a negative exponential function 813of distance or time from that center, as generally as-814sumed with the monocentric model. In the first alter-815native, if the centers are perfect substitutes, the effect 816on house prices is the maximum of the values pre-817dicted for the various centers. In other words, each 818center independently affects house prices for those 819houses within its “sphere of influence.” At the other 820extreme, if the centers are complements, with the resi-821dents of each house valuing access to every employ-822ment center, the effect of the centers is the product of 823their individual effects. This has the very convenient 824property that when taking the log of the dependent 825variable, the effects of the distances or times are addi-826tive, making the model easy to estimate. Finally, the 827third alternative represents an intermediate case, with 828the negative exponential effects of distance or time on 829house prices being additive. This method requires 830nonlinear estimation. It has been employed by Griffith 831(1981), Gordon et al. (1986), Song (1994), and Small 832and Song (1994) in predicting population densities but 833not within the more complex context of a hedonic 834housing price model. Most commonly used is the 835second formulation assuming that the centers are 836complements, undoubtedly due to the ease of estima-837tion. This paper employs that approach, followed by a 838brief discussion of (unsuccessful) attempts to estimate 839the model for the other extreme, assuming the centers 840are substitutes. 841The obvious first attempts involved estimation of 842models for multiple employment centers by including 843the distances or times from the house sales locations to 844the all of the centers. The results for such models for 845both the TAZ and ZIP code centers were seriously 846compromised by multicollinearity. Regression coeffi-847cient values and signs for the distance and time va-848riables were virtually random. Variance Inflation Fac-849tors were extremely high. The relatively close proximi-850ty of the employment centers in Indianapolis appar-851ently produced these levels of multicollinearity, as 852opposed to, for example, the more widely separated 853employment centers for which similar models were 854successfully estimated for Los Angeles (Heikkila et al. 8551989). 856For the TAZ-based employment centers, three of 857the five centers identified had substantially greater 858employment than the other centers. These are the cen-859ters shown darker and labeled in Figure 2. Models 860were estimated using the distances and times to just 861these three major employment centers. The results are 862shown in the middle of Table 3. Looking first at the 863overall performance of the models, the R2 values show 864increases over the base model, rising from 0.858 to 865about 0.860 to 0.861, as opposed to negligible increases 866for the models using only distances or times to the 867CBD. (It should be noted that the objective sought in 868considering the alternative measures of location was 869not, however, to substantially improve the predictive 870power of the model, which is already very high. Ra-871ther, it was to provide for a more complete specifica-872tion of the model.) As with distance to the CBD, the 873travel times performed better than distance, with con-874gested travel times slightly better than free-flow times. 875For the two models using travel times, the regres-876sion coefficients for the times to each of the three cen-877ters were statistically significant. For the model using 878distance, only one of the coefficients, distance to Keys-879tone/Castleton, was significant. Distance to the CBD 880and the other employment center were not. This is the 881one additional case in which differences are observed 882when the Center Township dummy variable was in-883cluded in the model. With Center Township included 884to account for the negative neighborhood effects, the 885regression coefficients for both distance to the CBD 886and to the West 86th Street center were significant, 887paralleling the results obtained for the models using 888only distance to the CBD. 889The ZIP code-based centers outside of the CBD 890did not have the significant differences in total em-891ployment levels observed for the TAZ centers. It was 892therefore not possible to identify a subset of major 893employment centers that could be used for estimating 894comparable models. 895Distances and times to all of the employment cen-896ters can be included in models if one makes the as-897sumption that the coefficients for the negative expo-898nential decline of house prices are the same for each of 899the centers. With this assumption, models can be esti-900901902===== PAGE 12 (score 34, numeric-paren 0, table-words 12, var-words 5) =====90390430 905 Ottensmann et al. 906 907Table 4 Models for Accessibility to Employment and Change in Employment. 908 909Location Variable 910Regression 911Coefficient 912Standard 913Error 914t 915Sig 916R2 917 918 919 920 921 922 923Accessibility to Employment 924 925 926 927 928 929Distance accessibility to ZIP employment 930(coefficient = 1.1) 9310.0059 9320.0006 93310.1156 9340.0000 9350.8595 936Distance accessibility to TAZ employment 937(coefficient = 1.6) 9380.0205 9390.0014 94014.5388 9410.0000 9420.8612 943Free-flow travel time accessibility to TAZ 944employment (coefficient = 2.2) 9450.0273 9460.0019 94714.5371 9480.0000 9490.8612 950Congested travel time accessibility to TAZ 951employment (coefficient = 2.1) 9520.0232 9530.0017 95413.8811 9550.0000 9560.8610 957 958 959 960 961 962 963Accessibility to Change in Employment 964 965 966 967 968Distance accessibility to change in ZIP 969employment (coefficient = 0.081) 9700.0081 9710.0006 97213.5473 9730.0000 9740.8608 975 976 977 978 979 980 981Accessibility to Employment and Change in Employment 982 983 984 985Distance accessibility to ZIP employment 986(coefficient = 1.3) 9870.0048 9880.0007 9896.9897 9900.0000 9910.8616 992Distance accessibility to change in ZIP 993employment (coefficient = 0.067) 9940.0069 9950.0006 99611.4801 9970.0000 998 999 1000 1001 1002 1003 1004 1005Accessibility to Employment, Change in Employment, Mean Distance to ZIP Centers 1006Distance accessibility to ZIP employment 1007(coefficient = 1.3) 10080.0048 10090.0007 10106.9896 10110.0000 10120.8616 1013Distance accessibility to change in ZIP 1014employment (coefficient = 0.067) 10150.0075 10160.0012 10176.0354 10180.0000 1019 1020Mean distance to 7 ZIP employment cen-1021ters 10220.0008 10230.0017 10240.4736 10250.6358 1026 1027 1028 1029 1030In his comparison of the performance of alterna-1031tive accessibility measures in predicting population 1032densities, Song (1996) chose to use the natural log of 1033accessibility to predict density rather than using the 1034natural form as was done here. Tests were made using 1035the log of the accessibilities, and the fits of the models 1036(after finding best-fit accessibility coefficients asso-1037ciated with using the logs) were no better. 1038 10394.4. Change in Accessibility to Employment 1040All of the applications involving the use of meas-1041ures of accessibility to employment to predict densities 1042or house prices have used accessibility to levels of em-1043ployment at a single point in time. Attention has not 1044been given to the possible effect of changes in accessi-1045bility to employment over time. However, if accessibil-1046ity to employment has a significant effect on house 1047prices, then changes in accessibility to employment in 1048the recent past, which may be seen as being predictive 1049of future changes in accessibility, might also affect 1050house prices. A study examining changes in accessibil-1051ity to employment in Milwaukee over more than three 1052decades found these changes were associated with a 1053range of neighborhood characteristics (Ottensmann 10541980). In the estimation of predictive models for an 1055urban simulation model, change in accessibility to 1056employment was included along with current accessi-1057bility in the prediction of the change from nonurban to 1058residential land use (Ottensmann 2005). 105910601061===== PAGE 13 (score 32, numeric-paren 0, table-words 10, var-words 12) =====10621063Location in a Hedonic Model 1064 1065 31 1066 1067 1068A model was estimated examining the effect on 1069house prices in the hedonic model of changes in acces-1070sibility to employment from 1995 to 2000 using the 1071distances to ZIP code employment. The assumption is 1072made that the accessibility coefficient is the same in 1073both years, with the calculation actually being of ac-1074cessibility to change in employment, which is mathe-1075matically equivalent using this assumption. The best-1076fit accessibility coefficient was determined as for ac-1077cessibility to employment. The results are shown in 1078the next portion of Table 4. The fit of the model was 1079about as good as for the various accessibility measures 1080(and even better than for accessibility to ZIP code em-1081ployment). Accessibility to employment change was 1082statistically significant. 1083If both accessibility to employment and change in 1084accessibility to employment are effective measures of 1085location for use in the prediction of house prices, it is 1086not unreasonable to consider that the combination of 1087the two might result in even better predictions. A 1088model including both accessibility measures was esti-1089mated, with the results presented in the next section of 1090Table 4. Estimation began by using the best-fit accessi-1091bility coefficients estimated for accessibility and 1092change in accessibility when used separately. Both 1093coefficients were then adjusted iteratively until the 1094combination of accessibility coefficients producing the 1095lowest standard error of estimate was found.7 This 1096model has an R2 value of 0.862, which is higher than 1097for any of the other models examined. The regression 1098coefficients for both accessibility to employment and 1099change in accessibility to employment are statistically 1100significant, with the latter having the higher t value. So 1101change in accessibility to employment is a significant 1102predictor of house prices after controlling for the ef-1103fects of the level of accessibility to employment. To-1104gether, the two provide the best measure of location 1105for predicting house prices. 1106With both accessibility to employment and change 1107in accessibility significant predictors of house prices, 1108the question arises as to whether distances to the mul-1109tiple employment centers would also contribute in 1110measuring location. The mean distance to the ZIP em-1111ployment centers was added to the previous model, 1112and the results are shown in the final section of Table 11134. Mean distance to the employment centers was not 1114 11157 In searching for the best combination of accessibility coefficients, 1116both the high and low estimates of the accessibility to employment 1117coefficients as described in the previous footnote were used as start-1118ing points. Once again, different combinations of accessibility coeffi-1119cients yielded two different minima with nearly identical standard 1120errors. One had a negative regression coefficient for accessibility. 1121Results for the model in which both coefficients were positive, as 1122expected, are reported. 1123statistically significant, and the regression coefficients 1124for the accessibility measures changed little. The com-1125bination of the two accessibility measures appears to 1126capture the effect of employment location, with the 1127location of multiple employment centers adding noth-1128ing. 1129 11305. Conclusions 1131 1132The measures of location were statistically signifi-1133cant predictors of sales price in the hedonic model. 1134However, their addition to the model produced only 1135small increases in R2. This raises the question as to the 1136relative importance of location as opposed to house 1137and neighborhood characteristics in determining 1138house prices. Table 5 presents the predicted changes in 1139sales price for changes in the independent variables 1140for a house with the mean sales price of $110,888. 1141Changes for the house and neighborhood variables 1142have been calculated using the regression coefficients 1143estimated for the base model without location. For the 1144changes in distances and times, changes are shown for 114510 kilometer and 10 minute changes in those meas-1146ures. Changes for the house and neighborhood meas-1147ures are for unit changes in those measures, with the 1148exception of the effective tax rate and the school dis-1149trict SAT, for which more intuitive levels of change 1150were used. 1151As expected, the house characteristics are major 1152determinants of sales price. Unit changes in four of the 1153variables—number of bathrooms, lot size greater than 1154one acre, and presence of a basement and air condi-1155tioning—resulted in changes in sales price greater than 115610 percent. The effects of the neighborhood variables 1157were smaller, though the percentage of units in the 1158neighborhood vacant and, especially, school district 1159SAT scores, had substantial effects. 1160Considering distance and travel time to the CBD, 1161the effect of distance was small, but this variable was 1162not significant in the model. Travel times to the CBD 1163had a much larger effect, with 10 minute increases in 1164travel times being associated with 3.3 and 6.4 percent 1165declines in sales price for the free-flow and congested 1166travel time models respectively. Mean distance and 1167travel times to the multiple employment centers had 1168greater effects on sales price, with 10 kilometer or 10 1169minutes increases in the means predicting 9.1 to 12.9 1170percent decreases in sales prices, decreases in prices 1171for a house with the mean price ranging from $9,500 to 1172$12,900. These are substantial effects of location with 1173respect to employment on house prices. 1174 1175 1176 117711781179===== PAGE 14 (score 32, numeric-paren 0, table-words 2, var-words 26) =====1180118132 1182 Ottensmann et al. 1183Table 5. Change in Sales Price at Mean Sales Price 1184 1185Sales 1186 1187 1188Price 1189Percent 1190Dependent Variable Change 1191Change 1192Change 1193 1194 1195 1196Number of bathrooms 1197$12,945 119811.67 1199Total number of rooms 1200$1,453 12011.31 1202Floor space (100 sq ft) 1203$2,359 12042.13 1205Has basement 1206$11,347 120710.23 1208Age 1209-$263 1210-0.24 1211Lot area <= 0.5 acre 1212-$5,321 1213-4.80 1214Lot area >= 1.0 acre 1215$11,119 121610.03 1217Garage bays 1218$9,902 12198.93 1220Has porch or deck 1221$5,420 12224.89 1223Exterior brick or stone 1224$9,714 12258.76 1226No air conditioning 1227-$24,626 1228-22.21 1229Effective tax rate (tax per $100) 1230-$2,622 1231-2.36 1232School district SAT (100 points) 1233$4,592 12344.14 1235Neighborhood median income 1236$564 12370.51 1238Neighborhood percent black 1239-$51 1240-0.05 1241Neighborhood percent vacant 1242-$2,373 1243-2.14 1244 1245 1246 1247Distance to CBD (10 kilometers) 1248-$408 1249-0.37 1250Free-flow travel time to CBD (10 1251minutes) 1252-$3,642 1253-3.28 1254Congested travel time to CBD (10 1255minutes) 1256-$7,039 1257-6.35 1258 1259 1260 1261Mean distance to 7 ZIP centers (10 1262kilometers) 1263-$10,144 1264-9.15 1265Mean distance to 5 TAZ centers (10 1266kilometers) 1267-$9,483 1268-8.55 1269Mean free-flow travel time to 5 1270TAZ centers (10 minutes) 1271-$13,426 1272-12.11 1273Mean congested travel time to 5 1274TAZ centers (10 minutes) 1275-$14,307 1276-12.90 1277 1278 1279 1280 1281 1282Distance and travel times to the CBD were the 1283least effective measures of location in predicting house 1284prices in the hedonic model. Distance to the CBD was 1285not even statistically significant in the base model. It 1286was, however, significant when location in Center 1287Township (the inner city) was included in the model 1288to control for externalities negatively affecting house 1289prices. These results may provide some insight into 1290the frequent results of insignificant or even positive 1291coefficients for distance to the CBD in hedonic models 1292(Bender and Hwang 1985; Coulson 1991). If there are 1293negative neighborhood effects in inner-city areas that 1294are not completely reflected in the neighborhood cha-1295racteristics included in a hedonic model, effects that 1296are related to distance to the center, the finding of no 1297significant negative decline of house price with dis-1298tance may reflect incomplete specification of the 1299neighborhood effects in the model. This suggests cau-1300tion is in order in the interpretation of findings of a 1301lack of significant decline in house prices with distance 1302from the CBD. For example, Heikkila et al. (1991) em-1303phasized the result that, in their hedonic model in-1304cluding distances to multiple employment centers in 1305Los Angeles, distance to the CBD is the one measure 1306that was not statistically significant. While it certainly 1307may be true that the Los Angeles CBD does not affect 1308house prices in the area, it is also possible that the fail-1309ure to observe an effect resulted from an inadequate 1310specification of neighborhood effects that are asso-1311ciated with distance to the CBD. This finding empha-1312sizes the importance of correctly specifying neighbor-1313hood effects in hedonic models. 1314Measures of location including employment out-1315side of the CBD yielded better predictions and more 1316significant measures of location. Measures of distance 1317or time to multiple employment centers performed 1318about as well as the measures of accessibility to em-1319ployment. These results do not show clearly better 1320performance for the accessibility measures as found by 1321Song (1994; 1996) in models predicting population 1322density. 1323Across the models using distances or times to the 1324CBD and to multiple employment centers, travel times 1325consistently performed better than distances. Among 1326the alternative travel time measures, congested travel 1327times did better than free-flow travel times. This pat-1328tern did not hold, however, for the accessibility to em-1329ployment measures. For accessibility to employment, 1330the measures of accessibility to the more spatially-1331disaggregated TAZ employment performed better 1332than accessibility to employment in the ZIP code areas. 1333Recent change in accessibility to employment 1334proved to be a significant measure of location affecting 1335house prices. This variable, calculated using ZIP code 1336employment change, actually performed better than 1337accessibility to ZIP code area employment. The com-1338bination of accessibility to employment with change in 1339accessibility was the best overall measure of location 1340in predicting house prices, with both measures signifi-1341cant. 1342Location with respect to employment is a signifi-1343cant predictor of house prices within a hedonic model, 1344supporting the general assumption embedded in the 1345standard urban economic model. With the decentrali-134613471348===== PAGE 15 (score 9, numeric-paren 0, table-words 0, var-words 9) =====13491350Location in a Hedonic Model 1351 1352 33 1353 1354 1355zation of employment that has occurred, the findings 1356do not, however, support the specific assumption of 1357the standard monocentric model of employment lo-1358cated only in the CBD. Measures of location that in-1359clude employment outside of the CBD do better than 1360the simple measures of distance or time to the CBD. 1361These results are not only important for under-1362standing the nature of the effects of accessibility to 1363employment on house prices. They also have impor-1364tant implications for public policy. Most obvious is the 1365use of hedonic house price models and other types of 1366models for property assessment for tax purposes. In-1367creasing use is being made of such models for the ap-1368praisal of real property and for real estate assessment 1369(see, e.g., Detweiler and Radigan 1996; Pagourtzi et al. 13702003; Renne 2003). Obtaining accurate assessment de-1371pends upon the proper specification of the models 1372used. Because location with respect to employment 1373has a significant effect on house values, failure to in-1374clude appropriate measures of such accessibility can 1375result in significant errors in assessment. 1376Hedonic house price models are frequently em-1377ployed to estimate the benefits or costs associated with 1378location-based amenities or disamenities, often with 1379the goal of informing public policymaking to address 1380the amenities or disamenities. It is a truism that the 1381misspecification of a model can result in errors in the 1382estimation of the model coefficients, though often the 1383direction and magnitude of such errors is unknown. 1384However, in at least some applications of hedonic 1385models to estimate the negative effects of disamenities, 1386the effect of failure to properly specify location with 1387respect to employment can be quite clear. Consider the 1388case of the effect of airport noise on property values. 1389(See Nelson 2004 for references to a large number of 1390such studies.) Areas near airports are frequently major 1391employment centers. If accessibility to that employ-1392ment is not included in the hedonic model, predicted 1393house prices near the airport will be underestimated. 1394This may then result in an underestimate of the nega-1395tive effect of airport noise on those properties affected. 1396Two studies (Espey and Lopez 2000; Tomkins et al. 13971998) have used hedonic models to simultaneously 1398estimate the effects on property values of both airport 1399noise and proximity to the airport. Both studies consi-1400dered the airport as a potential amenity, not as an em-1401ployment center, and no measures of location with 1402respect to employment were included in either model. 1403Both found negative effects of airport noise on proper-1404ty values, but they differed as to whether proximity 1405had a positive or negative effect. Tomkins et al. (1998) 1406found proximity to the Manchester airport led to high-1407er values. Espey and Lopez (2000) found a negative 1408effect for the Reno-Tahoe airport, but they noted that 1409the area surrounding that airport is primarily residen-1410tial, so it may not be a significant employment center. 1411Another example where disamenities might be asso-1412ciated with employment concentrations would be air 1413pollution and certain types of manufacturing em-1414ployment. 1415Hedonic models using measures of accessibility 1416based on travel times provide an opportunity to esti-1417mate the value people place on travel time savings. 1418Such estimates can be used in valuing the benefits as-1419sociated with travel time reductions associated with 1420planned transportation system improvements. At least 1421one author (Nelson 1977) has directly employed a he-1422donic house price model for this purpose. 1423Because it is a significant predictor of house prices, 1424location with respect to employment should be in-1425cluded for the proper specification of hedonic housing 1426price models. The combination of accessibility to em-1427ployment and change in accessibility to employment 1428provides the best specification of location, but the ne-1429cessary data may not always be available. But certain-1430ly location with respect to employment should include 1431either distances or times to multiple employment cen-1432ters or a measure of accessibility to all employment, as 1433the assumption that access to employment in the CBD 1434would be sufficient is not supported. 1435 1436 1437References 1438 1439Adair, A., S. McGreal, A. Smyth, J. Cooper and T. Ry-1440ley. 2000. House prices and accessibility: The test-1441ing of relationships within the Belfast urban area. 1442Housing Studies 15(5): 699-716. 1443Alonso, W. 1964. Location and Land Use: Toward a Gen-1444eral Theory of Land Rent. Cambridge, MA: Harvard 1445University Press. 1446Anas, A., R. Arnott and K. Small. 1998. Urban spatial 1447structure. Journal of Economic Literature 36(3): 1426-14481464. 1449Bender, B. and H. Hwang. 1985. Hedonic house price 1450indices and secondary employment centers. Journal 1451of Urban Economics 17(1): 90-107. 1452Bowen, W. M., B. A. Mikelbank and D. M. Preste-1453gaard. 2001. Theoretical and empirical considera-1454tions regarding space in hedonic housing price 1455model applications. Growth and Change 32(4): 466-1456490. 1457Burnell, J. D. 1985. Industrial land use, externalities, 1458and residential location. Urban Studies 22(5): 399-1459408. 1460Coulson, N. E. 1991. Really useful tests of the mono-1461centric model. Land Economics 67(3): 299-307. 1462