1# RECORD b8c37b8ea0a12e9cbed953c9880d16a2673b21592# TITLE: Measures of labour market accessibility. What can we learn from observed commuting patterns?3# YEAR: 2020.0 VENUE: nan DOI: 10.18335/region.v7i1.2614# AUTHORS: Liv Osland; Arnstein Gjestland; I. Thorsen5# PAGES: 22 CANDIDATE TABLE PAGES (1-based): [13, 17, 22]6# ABSTRACT: It is well known that measures of labour market accessibility explains spatial variation in housing prices even in markets with polycentric labour market structures. This paper examines whether data on observed commuting patterns can replace or supplement gravity-based measures representing the commuting potential at specific locations. We use data from a region in Western Norway,and we find that measures based on observed commuting flows and commuting time cannot replace a gravity-based measure of labour market accessibility. Based on, inter alia, the spatial Durbin estimator we find that measures of observed commuting flows increase the explanatory power of a hedonic house price model.789===== PAGE 1 (score 7, numeric-paren 0, table-words 0, var-words 7) =====1011Volume 7, Number 1, 2020, 49–7012journal homepage: region.ersa.org13DOI: 10.18335/region.v7i1.26114Measures of labour market accessibility. What can we15learn from observed commuting patterns?∗16Arnstein Gjestland1, Liv Osland1 and Inge Thorsen1171 Western Norway University of Applied Sciences18Received: 13 February 2019/Accepted: 28 February 202019Abstract. It is well known that measures of labour market accessibility explain spatial20variation in housing prices, even in markets with polycentric labour market structures.21This paper examines whether data on observed commuting patterns can replace or22supplement gravity-based measures representing the commuting potential at specific23locations. We use data from a region in Western Norway, and we find that measures24based on observed commuting flows and commuting time cannot replace a gravity-based25measure of labour market accessibility. Based on, inter alia, the spatial Durbin estimator26we find that measures of observed commuting flows increase the explanatory power of a27hedonic house price model.28129Introduction30The relationship between house prices and access to workplaces is a central theme in31both theoretical and empirical housing market research. There are many reasons why this32relationship is important. Travelling to work is a regular and bounded trip. According to33V˚agane et al. (2011), travelling to work constitutes 18 per cent of all travels in Norway.34During workdays, approximately 25 per cent of the travels are journey to work. Most35commuters travel at the same point in time every day, which also creates congestion. Road36transportation infrastructure is often given a capacity to deal with such traffic peaks. At37the same time, investments in the road network affect labour market accessibility, which38is in turn capitalized into house prices. In order to reduce many of the transportation39problems related to commuting, planners may seek to locate houses in areas where job40accessibility is assumed to be high.41According to Handy, Niemeier (1997), there is no consensus in the literature on a42good measure of accessibility. In explaining housing prices, gravity-based accessibility43measures have been suggested as a generalization of modern polycentric labour market44structures. Although more recent research has shown that gravity-based accessibility45measures explain significant spatial variation in housing prices, Handy, Niemeier (1997)46show that the gravity-based measures are not the only weighted measures that can be47used to capture the job opportunity density of a given area.48Gravity-based measures of labour market accessibility reflect the potential for commut-49ing from a specific residential area. In this paper we introduce three other measures, based50on actual commuting patterns. One measure is origin specific: the percentage of the total51working population living in the zone and working in a different zone. The second measure52∗We are grateful to two anonymous reviewers for very helpful comments.5349545556===== PAGE 13 (score 100, numeric-paren 67, table-words 4, var-words 2) =====5758A. Gjestland, L. Osland, I. Thorsen596160Table 2: Estimated results from alternative hedonic house price models61Variable name62M063M164M265M366M467M568Constant6912.9707012.4177112.9427212.0857312.5437411.43075(71.89)76(63.52)77(71.75)78(59.50)79(49.53)80(38.87)81LotSize82-0.17383-0.20484-0.16785-0.20386-0.17787-0.22188(-3.00)89(-3.62)90(-2.91)91(-3.55)92(-3.1)93(-3.92)94LotSize2950.022960.025970.022980.025990.0231000.026101(4.62)102(5.27)103(4.55)104(5.22)105( 4.72)106(5.59)107RurLotSize108-0.025109-0.021110-0.027111-0.023112-0.025113-0.019114(-10.28)115(-8.83)116(-10.15)117(-9.64)118(-10.11)119( -8.36)120Age121-0.2017122-0.192123-0.205124-0.199125-0.204126-0.196127(-11.75)128(-11.38)129(-11.84)130(-11.72)131(-11.75)132( -11.47)133Age21340.0171350.0161360.0181370.0171380.0181390.016140(5.75)141(5.23)142(5.92)143(5.65)144(5.85)145( 5.38)146Garage1470.0421480.0431490.0401500.0421510.0421520.040153(7.21)154(7.44)155(6.93)156(7.16)157(7.19)158(7.06)159LivingArea1600.5101610.5051620.5101630.5031640.5111650.504166(42.77)167(42.69)168(42.80)169(42.66)170( 42.99)171(42.88)172YearDum041730.0971740.0981750.0971760.0971770.0971780.097179(8.89)180(9.03)181(8.90)182(9.05)183(8.89)184(8.98)185YearDum051860.2061870.2071880.2051890.2051900.2051910.205192(18.75)193(18.92)194(18.68)195(18.90)196(18.68)197(18.84)198YearDum061990.3702000.3712010.3702020.3702030.3702040.370205(33.55)206(33.67)207(33.49)208(33.82)209(33.51)210(33.82)211YearDum072120.5582130.5582140.5582150.5572160.5582170.555218(53.95)219(54.17)220(53.85)221(54.44222(53.93)223(54.39)224DistCBD225-0.056226-0.104227-0.043228-0.114229-0.076230-0.171231(-3.14)232(-5.36)233(-2.28)234(-5.57)235(-3.90)236(-7.40)237DistCBD2238-0.046239-0.027240-0.051241-0.024242-0.043243-0.0150244(-13.18)245(-5.84)246(-12.22)247(-4.53)248(-10.86)249(-2.85)250ACCESS2510.0642520.0962530.114254(6.48)255(7.79)256(7.06)257IN-COM258-0.030259-0.083260(-2.06)261(-4.90)262OUT-COM263-0.0432640.001265(0.03)266(2.31)267MCT2680.3932690.399270(2.45)271(2.29)272MCT2273-0.081274-0.054275(-2.41)276(-1.45)277n278439227943922804392281439228243922834392284R22850.8222860.8242870.8232880.8262890.8232900.826291R2(adj)2920.8222930.8242940.8222950.8252960.8222970.825298Log-likelihood2991184.023001207.053011187.483021226.903031187.603041228.14305VIF30616.3630717.5630816.0630917.6031045.8731146.20312Ramsey reset3130.7673140.7603150.4643160.3163170.80033180.5364319APE320510405321508469322510366323507978324510486325508223326SRMSE3270.2893280.2883290.2903300.2883310.2893320.288333RLM-lag3341.703350.863361.383370.173381.703390.39340RLM-error341323.59342313.09343324.15344307.78345320.36346313.01347REGION : Volume 7, Number 1, 2020348349350===== PAGE 14 (score 7, numeric-paren 0, table-words 0, var-words 7) =====35135262353A. Gjestland, L. Osland, I. Thorsen354Note: The dashed lines refer to M3, while the solid line refers to a corresponding model specification355without a gravity-based accessibility measure (M2). Predicted housing prices for variation in OUT-COM356based on M2 is not shown in the Figure to the right, given that this variable does not contribute to357explain the variation in housing prices in these two models.358Figure 3: Predicted house prices of a standard house359Note: The solid line refers to M1 where the commuting variables are excluded. The dotted line refers to360M3, and the dashed line refers to M5. The values of the accessibility variables is mean-normalized.361Figure 4: Predicted house prices of a standard house362Figure 6, however, there is a substantial difference between the model M5, and the363more parsimonious model M4. Technically, this is due to the different parameter values364estimated for the quadratic term, but it can also be argued that the results reflect changes365in the characteristics of the urban structure.366As pointed out in Section 3, a high MCT may reflect either a peripheral location, or367a centrally located area with heterogenous agents and considerable excess commuting.368For the model M5, the effects of labour market heterogeneity and a high level of spatial369interaction in the central parts of the urban area seem to dominate, since MCT is370estimated to have a positive impact on housing prices. Over time, the job growth in the371Stavanger urban area has come in areas that used to be the outskirt of the city, mainly in372suburban industrial parks. As a consequence, workers living in residential areas close to373the city centre, see Figure 1, no longer have on average shorter commuting times than374workers living in some of the suburban areas. Still, some of the residential areas close375to the city centre have very high housing prices. These areas have traditionally been376fashionable residential locations, with neighbourhoods that are popular, beyond the pure377urban attraction effect. This is one possible explanation why housing prices are predicted378to be an increasing function of MCT.379The first, rising part of the curve resulting from model M4 in Figure 6 can be explained380from the same line of reasoning as above. In this model, however, the labour market381accessibility measure is not incorporated. Since MCT is closely and negatively correlated382to labour market accessibility (Pearson’s correlation coefficient is -0.8677), the labour383REGION : Volume 7, Number 1, 2020384385386===== PAGE 17 (score 84, numeric-paren 62, table-words 1, var-words 2) =====387388A. Gjestland, L. Osland, I. Thorsen38965390Table 3: Estimated direct, indirect, and total impact from variants of M3 and M5 using391the spatial Durbin Estimator392M3393M5394Variable Name395Direct396Indirect397Total398Direct399Indirect400Total401Lotsize402-0.2014030.150404-0.051405-0.1734060.012407-0.161408(-6.47)409(2.21)410(-0.70)411(-4.92)412(0.15)413(-1.84)414LotSize24150.025416-0.0134170.0134180.024419-0.0054200.020421(9.69)422(-2.20)423(2.01)424(8.73)425(-0.67)426(2.64)427RurLotsize428-0.0284290.007430-0.021431-0.0414320.024433-0.017434(-8.89)435(1.76)436(-8.10)437(-3.92)438(2.21)439(-5.24)440Age441-0.1964420.038443-0.158444-0.161445-0.097446-0.285447(-11.93)448(1.05)449(-4.13)450(-9.39)451(-2.51)452(-6.22)453Age24540.016455-0.0044560.0124570.0094580.0204590.028460(5.89)461(-0.71)462(1.84)463(2.95)464(3.09)465(4.13)466Garage4670.0404680.0264690.0664700.0394710.0094720.048473(6.48)474(2.06)475(4.70)476(6.78)477(0.63)478(2.86)479LivingArea4800.503481-0.0374820.4664830.4874840.0944850.581486(52.91)487(-1.84)488(21.11)489(53.64)490(3.72)491(20.94)492YearDum044930.099494-0.0294950.0704960.097497-0.0014980.096499(9.58)500(-1.24)501(2.86)502(9.05)503(-0.04)504(2.22)505YearDum055060.209507-0.0275080.1825090.2075100.0385110.245512(19.93)513(-1.16)514(7.48)515(19.40)516(1.03)517(5.72)518YearDum065190.373520-0.0145210.3595220.371523-0.0195240.353525(35.41)526(-0.57)527(14.40)528(35.31)529(-0.53)530(8.50)531YearDum075320.560533-0.0225340.5385350.5575360.0015370.558538(56.92)539(-0.92)540(21.51)541(54.24)542(0.04)543(14.02)544DistCBD545-0.081546-0.073547-0.153548-0.036549-0.103550-0.139551(-3.45)552(-2.09)553(-4.86)554(0.40)555(-0.97)556(-4.23)557DistCBD2558-0.0245590.008560-0.0165610.014562-0.033563-0.019564(-4.20)565(0.95)566(-1.94)567(0.37)568(-0.87)569(-2.66)570ACCESS5710.0845720.0355730.1205740.161575-0.0515760.109577(7.23)578(1.87)579(6.99)580(2.00)581(-0.63)582(5.22)583IN-COM584-0.073585-0.064586-0.136587(-4.67)588(-2.62)589(-6.18)590OUT-COM5910.0405920.0335930.073594(2.36)595(1.87)596(2.72)597MCT5980.247599-0.1226000.125601(1.80)602(-0.88)603(3.43)604Note: Z-values in parentheses.605The spatial Durbin model yields the same sign on all the estimated parameters, as606for the ordinary least squares estimator. The direct impacts yield results that are within607the 95% confidence region of the ordinary least squares regression results for the model608variant based on M3. For the specification based on M5, this is not always the case. For609most of these variables the total impact is within the 95% confidence region. The impacts610related to age variables is not within the mentioned 95% confidence region. For M1, not611reported, the total impact of ACCESS is not within the 95% confidence region. The612indirect impact is negative, but not significant.613Indirect impacts are significant at the 5% level for the variables LivingArea and Age2614for the model based on M5. The same is the case for the spatial Durbin variant of615M1. In the model based on M3, only IN-COM has a significant indirect impact at the6165% significance level. The indirect impact is negative. This means that an increase in617IN-COM of the houses neighbouring a house i, will on average have a negative impact618on the price of house i. Hence, there are negative spillover effects of having a house619located in an area with a high relative level of in-commuting. This result is in line with620our interpretations of regression results found in previous sections. The average indirect621REGION : Volume 7, Number 1, 2020622623624===== PAGE 18 (score 12, numeric-paren 0, table-words 1, var-words 10) =====62562666627A. Gjestland, L. Osland, I. Thorsen628impacts of OUT-COM and ACCESS are positive, but not significant at the 5% level.629The results of this robustness check also show that all the direct impacts related to the630studied variables have the same sign as the ones documented in Table 2. Given the fact631that the correlation between the variables are relatively high, it is to be expected that632the inclusion and exclusion of variables will have an impact on the value of the estimated633coefficients. The direction of any existing bias is not possible to determine based on any634of the robustness checks.6357.2636Reverse Causality637The second relevant type of endogeneity problem could be reverse causality. By way of638example, neighbourhoods with higher housing prices could attract high-income earners,639perhaps also with higher educational attainments. Locations close to these places could be640relevant for specific firms in order to get access to an attractive pool of labour. This type641of reasoning is based on the hypothesis that “jobs follow people”, which subsequently may642improve labour market accessibility in places with higher housing prices. Hoogstra et al.643(2017) provide a useful meta-analysis of the related literature regarding this question.644Their main finding is that the evidence is inconclusive. The causality between jobs and645people could run in different directions (pages 371-372). If this holds true, for our data,646the resulting bias could be minor, and unpredictable.647If there exists reverse causality, identification of the impact of accessibility would648necessitate an instrument.649One possible instrument could be an exogenous change650in measures of accessibility. However, we do not have access to any such instrument.651Moreover, we focus on several variables, which are potentially endogenous. Testing for652exogeneity is, hence, not straightforward according to e.g. Baum et al. (2007). We have653to use several instruments, and in these cases, the traditional IV-estimators could be654biased and inconsistent (see also Nordvik et al. 2019, for further discussions of this issue).655Finally, it is also possible to argue that the need for an instrument is less important656in our case, given the results from the spatial Durbin model. This model accounts for657indirect spill over impacts and the spatial Durbin model, in general, is robust to omitted658systematic spatial variation of characteristics (LeSage, Pace 2009).6598660Conclusions661In explaining spatial variation in housing prices, gravity-based accessibility measures have662been suggested as a generalization of modern polycentric labour market structures. From663a computational and data collection perspective, it would be convenient if easily available664information on actual commuting patterns could replace a more complex measure of665commuting potential. According to our results, two of the hypotheses formulated in666Section 3 have to be rejected, however. The labour market accessibility effect is not667adequately represented by the proposed characteristics of observed commuting patterns.668We used a wide range of different methods to obtain robust conclusions. Labour market669accessibility in relation to housing prices is best captured by the gravity based potential670variable.671Our results, on the other hand, provide support for the hypotheses HB6720 and HD6730 .674Observed measures of commuting patterns are found to contribute with information that675adds to the effect of a potential measure in explaining spatial variation in housing prices.676In particular, the results from the ordinary least squares and the spatial Durbin estimator677support the hypothesis that a relatively high level of commuting into a zone corresponds678to negative externalities, such as noise, pollution, or other negative effects of heavy679traffic and/or industrial activities. There is only weak support for the hypothesis that a680relatively high level of out-commuting from a zone corresponds to attractive neighbourhood681characteristics, which are positively related to housing prices. Average commuting time,682a priori, reflect something about the actual commuting costs for households. However,683our interpretation of the positive impact of this variable is that a high level of spatial684interaction in the central parts of the urban area seem to dominate. Over time, the685job growth in the Stavanger urban area has come in areas that used to be the outskirt686of the city. Still, some of the residential areas close to the city centre have very high687REGION : Volume 7, Number 1, 2020688689690===== PAGE 22 (score 101, numeric-paren 75, table-words 1, var-words 4) =====69169270693A. Gjestland, L. Osland, I. Thorsen694A695Appendix: Results of spatial error model estimations696Table A.1: Estimated results for the hedonic house price models based on the spatial697error model formulation as specified in equation (9)698Variable Name699M0700M1701M2702M3703M4704M5705Constant70613.06070712.50670813.03170912.18071012.71071111.596712(120.88)713(75.10)714(119.33)715(66.87)716(48.96)717(37.09)718Lotsize719-0.187720-0.204721-0.183722-0.202723-0.189724-0.214725(-6.11)726(-6.61)727(-5.94))728(-6.56)729(-6.15)730(-6.93)731LotSize27320.0247330.0267340.0247350.0267360.0247370.026738(9.55)739(9.98)740(9.39)741(9.99)742(9.58)743(10.30)744RurLotsize745-0.025746-0.021747-0.026748-0.023749-0.025750-0.019751(-8.76)752(-7.26)753(-8.86)754(-7.86)755(-8.73)756(6.47)757Age758-0.1809759-0.178760-0.183761-0.182762-0.182763-0.180764(-10.71)765(-10.51)766(-10.78)767(-10.82)768(-10.78)769(-10.67)770Age27710.0137720.0127730.0137740.0137750.0137760.013777(4.64)778(4.43)779(4.75)780(4.77)781(4.75)782(4.60)783Garage7840.0397850.0407860.0387870.0397880.0397890.039790(6.71)791(6.82)792(6.59)793(6.70)794( 6.68)795(6.67)796Size of house7970.4937980.4917990.4938000.4908010.4938020.491803(52.31)804(52.19)805(52.31)806(52.27)807(52.47)808(52.36)809YearDum048100.09558110.0968120.0968130.0968140.0968150.096816(9.87)817(9.91)818(9.87)819(9.93)820(9.88)821(9.90)822YearDum058230.1998240.1998250.1998260.1998270.1998280.199829(20.41)830(20.45)831(20.39)832(20.46)833(20.41)834(20.43)835YearDum068360.3708370.3708380.3708390.3708400.3708410.370842(38.33)843(38.34)844(38.34)845(38.36)846(38.32)847(38.35)848YearDum078490.55738500.5578510.5578520.5578530.5578540.556855(58.91)856(58.88)857(58.92)858(58.93)859(58.93)860(58.82)861DistCBD862-0.0574863-0.101864-0.040865-0.105866-0.074867-0.163868(-2.58)869(-4.14)870(-1.58)871(-3.93)872(-2.96)873(-5.71)874DistCBD2875-0.047876-0.029877-0.053878-0.028879-0.045880-0.018881(-11.06)882(-5.07)883(-9.17)884(-3.96)885(-9.48)886(-2.82)887ACCESS8880.0598890.0888900.109891(4.36)892(5.81)893(6.07)894IN-COM895-0.033896-0.082897(-1.78)898(-4.39)899OUT-COM900-0.0069010.034902(-0.22)903(1.65)904MCT9050.3159060.306907(1.48)908(1.47)909MCT2910-0.063911-0.035912(-1.44)913(-0.80)914λ9150.3749160.3649170.3739180.3599190.3749200.358921(23.67)922(22.85)923(23.60)924(22.30)925(23.60)926(22.24)927Log-likelihood9281420.269291429.589301422.569311440.239321422.699331440.57934Note: Z-values in parentheses.935c⃝2020 by the authors. Licensee: REGION – The Journal of ERSA, European936Regional Science Association, Louvain-la-Neuve, Belgium. This article is distri-937buted under the terms and conditions of the Creative Commons Attribution, Non-Commercial938(CC BY NC) license (http://creativecommons.org/licenses/by-nc/4.0/).939REGION : Volume 7, Number 1, 2020940