using Irish Data Anne Nolan (YE)

Transcription

1 The Determnants of Urban Households Transport Decsons: A Mcroeconometrc Stud usng Irsh Data Anne Nolan (YE) Abstract Ths paper uses Irsh mcro-data to analse the determnants of urban households transport decsons b estmatng elastctes of demand for car ownershp, car use and publc transport wth respect to ncome and varous household soco-demographc characterstcs. Ths paper uses expendture data to examne car and publc transport use and analses the latter decson for separate samples of households, namel, those ownng one car and those ownng no car. A bnar probt model s estmated for the car ownershp decson, whle for the car use and publc transport expendture decsons, Tobt models adusted for heteroscedastc and non-normal errors are estmated. JEL Classfcatons: D1, R41 Kewords: household, car ownershp, transport expendture, probt, Tobt. The author s a Ph.D. student n the Department of Economcs, Trnt College, Dubln, Ireland. In undertakng ths research the author receved assstance and advce from Professor Frances Ruane of Trnt College Dubln (TCD) and Professor Dens Connffe of the Economc and Socal Research Insttute, Dubln. The anonmsed mcro-data used n ths analss were obtaned from the Irsh Central Statstcs Offce (CSO) and access to these data s gratefull acknowledged. Helpful suggestons on an earler draft of the paper were receved at the TCD Graduate Semnar and Irsh Economcs Assocaton Annual Conference 001. Ths research s funded b a Government of Ireland Research Scholarshp from the Irsh Research Councl for the Humantes and Socal Scences. 1

2 1 Introducton Ths paper addresses a number of aspects of urban households transport decsons b usng Irsh cross-sectonal mcro-data to analse the soco-economc determnants of the demand for car ownershp, car use and publc transport n the captal ct, Dubln. The mportance of ncome and soco-demographc factors n nfluencng household transport choces has been well documented and studed usng cross-sectonal data from a varet of sources. Whle there are numerous studes examnng the ssue of car ownershp at both an aggregated and dsaggregated level, few studes have examned the related ssues of car and publc transport use usng expendture mcrodata. In ths paper, the analss of the car ownershp decson complements, and adds to, exstng lterature n the area. A new approach to the analss of urban car and publc transport use s presented through the use of expendture data. In the case of the publc transport decson, the dvson of the sample nto households ownng no car and one car helps to hghlght the dfferences between these two groups n ther publc transport decsons. In addton, the examnaton of the determnants of tax fare expendture as well as bus fare expendture emphasses the mportance of analsng the dfferent components of publc transport expendture separatel. Fnall, the estmaton of these relatonshps ndcates whether the factors nfluencng the transport decsons of households n the Dubln area are an dfferent to those found for other countres. The data emploed n ths stud are mcro-data from the 1994/1995 Irsh Household Budget Surve (HBS), whch are the latest data avalable. A bnar probt model s estmated for the car ownershp decson. For those households whch report ownershp of one car, we use a Tobt model to examne car use, whch s proxed b petrol expendture. In analsng publc transport

3 expendture decsons, Tobt models of expendture on bus and tax fares are estmated for those households ownng no car and those households ownng one car. Where relevant, adustments are made to the models to account for heteroscedastc and non-normal errors. Secton summarses related lterature and detals the econometrc methodologes emploed, whle Secton 3 descrbes the data used n ths analss. Secton 4 presents econometrc results for all models and outlnes practcal aspects of the estmaton ncludng the nterpretaton of the estmated coeffcents. Secton 5 contans some concludng remarks. Relevant Lterature and Econometrc Methodologes.1 Car Ownershp Earler studes of the demand for car ownershp usng cross-sectonal data at an aggregated level nclude those b Button et al (1993), Sad (199), Stanovnk (1990), McCarth (1977), Buxton and Rhs (197) and Farhurst (1965). Whle provdng some nsght nto the general relatonshps between car ownershp and varables such as populaton denst and average ncomes at regonal or countr level, the nature of the data lmts the number and tpe of ndependent varables that can be consdered. 1 In addton, man of the above studes are now outdated and the studes of Button et al (1993) and Stanovnk (1990) relate to low-ncome countres, thus lmtng the applcablt of the results to Ireland, and n partcular to Dubln. The ncreasng avalablt of mcro-data n recent ears has enabled researchers to overcome man of the problems nherent n aggregated data. It has allowed the formulaton of more 1 A partcular problem assocated wth aggregated data s that f a varable vares more wthn a regon than t does between regons, the true effect of the varable wll be dffcult to determne (Farhurst, 1965). 3

4 accurate models of car ownershp at an ndvdual or household level emplong a wder range of soco-economc characterstcs as ndependent varables [see Alperovch et al (1999), Darga and Vthoulkas (1999), Cragg and Uhler (1970) and Bennett (1967)]. The dscrete nature of the car ownershp decson means that dscrete choce econometrc methodologes, such as bnar and multnomal probt and logt, are often emploed n modellng demand [see Alperovch et al (1999), Stanovnk (1990) and Cragg and Uhler (1970)]. Recentl, the demand for car ownershp at the mcro-level has also been analsed n the context of other transport decsons such as car use and modal choce usng the nested multnomal logt technque [see De Palma and Rochat (000), Borner (1999), Thoban (1984), Tran (1980) and Ben-Akva and Lerman (1975)]. Income s the most consstentl mportant household soco-economc factor found to have a postve relatonshp wth car ownershp. However, some studes fnd that the relatonshp between ncome and car ownershp s non-lnear [Cragg and Uhler (1970)] whle others fnd that ncome elastctes of car ownershp declne n magntude as ncome ncreases [Darga and Vthoulkas (1999)]. These results suggest that the effect of ncome on household car ownershp s not constant, wth the effect beng more pronounced at lower ncome levels. Other varables found to have a sgnfcant effect on car ownershp nclude general household characterstcs such as the number of adults, chldren and workers n the household and household locaton. Head of household (HOH) characterstcs such as gender, age, educaton and occupaton are also commonl emploed. The results for age of HOH often conflct. Alperovch et al (1999) and Bennett (1967) fnd that the probablt of car ownershp, whle ntall ncreasng, declnes once the HOH reaches 40 ears and retrement age respectvel, whle Cragg and Uhler (1970) fnd that as the age of the HOH ncreases, the probablt of car ownershp decreases. 4

5 In modellng the demand for car ownershp, we use a bnar probt model as appled b Alperovch et al (1999) and Stanovnk (1990). The bnar probt model s emploed n stuatons where the dchotomous dependent varable ndcates the choce between two alternatves (e.g. to own a car or not). It s charactersed b a contnuous latent varable *, dfferent values of whch determne the observed value of the dependent varable,.e., * = + 0, 0 ~ NID (0,1) and = 0 f * 0 (1) = 1 f * > 0 The model s estmated b means of the maxmum lkelhood method of estmaton where the followng log-lkelhood functon s maxmsed wth respect to each of the estmated coeffcents: In L = In [1 = 0 - ( )] + In - ( ) = 1 where Φ s the cumulatve standard normal dstrbuton functon. A common feature of man bnar choce models s that the error terms are heteroscedastc, n whch case the estmated coeffcents are nconsstent. B allowng the error terms to var across observatons, ths problem can be overcome. Heteroscedastct of the followng form s assumed [Greene (1997)]: ( z ' h ) = exp (3) where z s a vector of contnuous ndependent varables assumed to cause the heteroscedastct. A lkelhood rato test s undertaken to test the null hpothess of homoscedastc errors (h=0). 3 The ssue of multple car ownershp (for example, usng a multnomal logt model) s not consdered here due to the small number of observatons ownng two or more cars (see footnote twelve below). 3 The test statstc s dstrbuted as $ wth the number of degrees of freedom equal to the number of varables ncluded n z. The lkelhood rato test of homoscedastc errors s not reected (see Secton 4 and Table 4.1) and so the log-lkelhood functon adusted to take account of heteroscedastct s not presented 5

6 . Car Use and Publc Transport Expendture In contrast to the large number of studes analsng car ownershp at both an aggregated and dsaggregated level, fewer studes have examned the related ssues of car and publc transport use. In terms of car use, a common approach s to analse the decson smultaneousl wth the car ownershp decson. For example, both Borner (1999) and Mannerng and Wnston (1985) use data at the household level to estmate nested multnomal logt models of the ont demand for car ownershp and use. 4 Button et al (1993) and Mannerng (1983) examne car use ndependentl b usng mleage data and whle the latter uses household level data, the former stud uses countr level data for a sample of low-ncome countres. Studes examnng petrol expendtures nclude those b Kaser (000) and Labeaga and Lopez (1997) who both utlse household mcro-data, albet applng dfferng econometrc methodologes. 5 The maort of studes analsng publc transport demand use specall constructed transport surves that seek to determne the factors nfluencng modal choce decsons, n partcular for the ourne to work. There are two man econometrc methodologes emploed to model these decsons. Frstl, bnar and multnomal probt and logt models analse the determnants of an ndvdual s decson to travel b a number of alternatve modes of transport [see Ben-Akva and Lerman (1975) and De Donnea (1971) for applcatons to the ourne to work decson]. Secondl, nested multnomal logt models smultaneousl examne the two decsons of car ownershp and mode of transport. De Palma and Rochat (000), Thoban (1984) and Tran (1980) all use ether ndvdual or household level data to determne the factors nfluencng car 4 The use of the nested multnomal logt econometrc methodolog requres the exstence of alternatve specfc characterstcs that do not var across observatons, such as fuel costs n the context of car use or ourne tme n the context of modal choce. Irsh Household Budget Surve data provde no such varables. 5 Kaser (000) uses the Heckman Sample Selecton methodolog whle Labeaga and Lopez (1997) use an AIDS model. 6

7 ownershp level and mode of transport to work. To date, lttle research has been undertaken usng expendture on publc transport as a prox for publc transport demand. Exceptons are Bergantno (1997) and Mng-Chu (1994), who use household expendture mcro-data to estmate transport Engel curves and the demand for recreatonal travel respectvel. Smlar ndependent varables to those used n explanng varatons n car ownershp levels are emploed n determnng the soco-economc nfluences on car and publc transport use. A sgnfcant fndng n man of the car use studes above s the low postve ncome elastct of car use; one stud [Mannerng and Wnston (1985)] also shows that the ncome elastct s smaller n the later tme perods. Kaser (000) and Labeaga and Lopez (1997) smlarl report that petrol ma be classfed as a necesst due to ts ncome elastct of demand beng less than unt. Both Borner (1999) and Mannerng (1983) fnd that the age of the HOH has a consstentl negatve effect on car use, a fndng that s dfferent to the postve, but declnng effect of the age of the HOH on car ownershp found b Alperovch et al (1999) and Bennett (1967). In terms of publc transport use, De Palma and Rochat (000) fnd that households n whch the household sze s large and the HOH s older than 50 ears are sgnfcantl more lkel to use publc transport rather than the car for the ourne to work. The econometrc methodolog emploed n ths paper to examne household car and publc transport use follows most closel the approach of Mng-Chu (1994) and other studes whch use Tobt models to estmate the demand for varous commodtes based on mcro-data [see Ca (1999), Gould and Km (1998), Yen et al (1996), Hamlton Lankford and Wckoff (1991) and 7

8 Bennett (1967) 6 ]. The Tobt model s used n stuatons where the dependent varable s censored,.e., values n a certan range are all reported as a sngle value, usuall zero. It s wdel emploed n modellng cross-sectonal expendture decsons n whch a large proporton of respondents report zero expendture. In common wth the bnar probt model consdered above, t also assumes the exstence of a contnuous latent varable *, the values of whch determne the actual value of the observed lmted dependent varable,.e., * = + 0, 0 ~ NID (0, 1 ) and = 0 f * 0, (4) > 0 f * > 0 The contnuous latent varable s assumed to represent desred expendture, thus allowng for negatve desred expendture. All negatve and zero values of desred expendture are transformed to a sngle value of zero for observed expendture. The Tobt model therefore assumes that zero observatons are due to corner solutons 7,.e. f relatve prces or ncome changes, expendture occurs. The model s estmated b maxmum lkelhood estmaton whereb the followng log-lkelhood functon s maxmsed wth respect to each of the estmated coeffcents: 6 Ca (1999), Gould and Km (1998), Hamlton Lankford and Wckoff (1991) and Bennett (1967) all use the Tobt methodolog to estmate the demand for food, dar products, chartable donatons and cars respectvel. Whle Ca (1999), Hamlton Lankford and Wkcoff (1991) and Bennett (1967) use expendture data, Gould and Km (1998) use quantt data. Yen et al (1996) analses the concentraton of ntrate n Amercan water supples usng quantt data. 7 The valdt of assumng that ths s the correct process determnng zero values of expendture s often questoned. Alternatve sources of zero expendtures ncludng nfrequenc of purchase and non-partcpaton n the market are gnored n the Tobt model. In addton, the Tobt model does not consder that there ma be a dfference between the effect of a varable on the partcpaton decson and the effect of that varable on the consumpton decson, e.g. the effect of chldren on the probablt of gong on holda s negatve but once expendture takes place, the effect of chldren on expendture s postve [Verbeek (000)]. An alternatve to the Tobt model s the Double Hurdle model, orgnall formulated b Cragg (1971), whch allows for corner solutons as well as non-partcpaton n the market. The double hurdle model s not consdered here. 8

9 In L = In = 0 > 0 In 1 + In φ 1 (5) where - and φ are the cumulatve standard normal dstrbuton and standard normal probablt denst functons respectvel. The estmated Tobt coeffcents are also senstve to the dstrbutonal assumptons that are made about the error term. If the error terms are heteroscedastc and/or non-normal, the coeffcent estmates are nconsstent. 8 Heteroscedastct of the same form as n the bnar probt model s assumed. To adust for non-normalt, an nverse hperbolc sne (IHS) transformaton s appled to the dependent varable as follows: 1 ( ) log[ ( 1) + + ] / = (6) where s a parameter estmated b the model [Yen et al (1996) and Renolds and Shonkwler (1991)]. Ths transformaton overcomes the problem of non-normalt caused b the presence of outlers b behavng logarthmcall for large values of the dependent varable [Renolds and Shonkwler (1991)]. Lkelhood rato tests are undertaken to test the null hpotheses of a homoscedastc error structure, a normal error structure and the ont null hpothess of a homoscedastc and normal error structure respectvel. 9 The log-lkelhood functon ncorporatng the adustments for heteroscedastc and non-normal errors s: 8 See papers b Renolds and Shonkwler (1991), Arabmazar and Schmdt (198) and Nelson (1981). 9 7KH WHVW VWDWLVWLFV DUH GLVWULEXWHG DV $ wth the number of degrees of freedom equal to the number of varables ncluded n the heteroscedastct functon and/or the IHS parameter. The lkelhood rato test of a heteroscedastc and non-normal error structure s not reected for all fve Tobt regressons and performs the best of all the lkelhood rato tests. Therefore, the log-lkelhood functons adusted for heteroscedastct and for non-normalt separatel are not presented. 9

10 In L = 1 ( ) 1 + In 1 + In φ x ' x' In In (7) 1 > 1 = 0 0 When = 0 and = the log-lkelhood reduces to that of the standard Tobt log-lkelhood (5) presented above. 3 Data The data emploed n ths analss are mcro-data from the 1994/1995 Irsh Household Budget Surve (HBS). 10 The surve conssts of 7,877 urban and rural (farm and non-farm) households. Ths stud focuses on the,148 households n the Dubln area. Each household was asked to complete a questonnare contanng nformaton on varous household ncome and socodemographc characterstcs and ownershp of durable goods as well as an expendture dar recordng ever tem of expendture b each member of the household over a two-week surve perod. Snce the data are cross-sectonal at a pont n tme, we cannot estmate prce effects and thus we concentrate on the ncome and soco-demographc determnants of transport demand. Varable defntons are presented n Table 3.1 whle summar statstcs for the transport varables extracted from the surve are detaled n Table 3. and those for the household ncome and soco-demographc varables are presented n Tables 3.3 and 3.4 below. In examnng car ownershp, the dependent varable s a bnar varable ndcatng whether or not the household owns a car. For the car use decson, the dependent varable s petrol expendture, whch s adusted for seasonalt 11. Publc transport use s proxed b bus and tax fare 10 The 1994/1995 Irsh HBS was conducted b the Central Statstcs Offce (CSO) between Ma 1994 and Jul 1995 and s the most current data set avalable. 11 All expendture varables are adusted for seasonalt because the HBS was conducted over a ffteen month perod between Ma 1994 and Jul

11 expendtures separatel, adusted for seasonalt and household sze. Household ncome s proxed b total weekl household expendture, also adusted for household sze and seasonalt. Table 3.1 Varable Name Transport Varables CAR CYCLE PETROLEXP BUSEXP TAXIEXP FREETRAV EXPENSES Varable Defntons Defnton =1 f the household owns one or more cars =0 otherwse Number of Motorccles Petrol Expendture Bus Fare Expendture Tax Fare Expendture Number enttled to free penson/school travel =1 f the household receves motor expenses =0 otherwse Contnuous Household Income and Soco-Demographc Varables HHEXP Total Weekl Household Expendture (dvded b 100) ADULTS Number of Adults 18+ ears CHILDREN Number of Chldren aged 17 ears and ounger Dscrete Household Soco-Demographc Varables Accommodaton Tpe APART, SEMI APART=1 f the household lves n an apartment or bedst, =0 otherwse SEMI=1 f the household lves n a sem-detached house, =0 otherese (Base Categor = household lves n a detached house) Household Workng Status WORKING =1 f at least one household member 15+ ears at work =0 otherwse Gender of HOH FEMALE Age of HOH SIXTY, FIFTY, FORTY, THIRTY =1 f household s headed b a female who s the onl adult n the household =0 otherwse SIXTY =1 f the HOH s 60+ ears, =0 otherwse FIFTY =1 f the HOH s ears, 0 otherwse FORTY=1 f the HOH s ears, 0 otherwse THIRTY=1 f the HOH s ears, 0 otherwse (Base Categor = HOH s aged 0-9 ears) Hghest Educaton Level of HOH PRIMARY, PRIMARY=1 f the HOH has a prmar school educaton onl, =0 otherwse SECONDARY SECONDARY=1 f the HOH has a secondar school educaton onl, =0 otherwse (Base Categor = HOH has a thrd level educaton) 11

12 The HBS does not record nformaton relatng to dstance travelled to work, dstance from ct centre, publc transport avalablt etc. and so the tpe of household accommodaton s used, albet mperfectl, to prox these factors. Ths assumes that those lvng n detached houses are more lkel to lve n outlng areas of Dubln and thus have longer dstances to travel to the ct centre and/or poorer publc transport lnks than those lvng n sem-detached houses or apartments. A complcatng ssue for the bus fare expendture models s the fact that all pensoners are enttled to free publc transport at off-peak tmes n Ireland. Ths means that n the case of pensoners one of the most mportant assumptons of the analss s volated,.e., that publc transport use s reflected n publc transport expendture n the surve. To overcome ths problem, a dumm ndependent varable ndcatng that there s at least one person n the household enttled to free publc transport s ncluded. Smlarl, for the car and publc transport use regressons for car-ownng households, a dumm varable ndcatng whether at least one person n the household receves remuneraton for motor expenses such as petrol s ncluded. An examnaton of the summar statstcs n Tables 3., 3.3 and 3.4 below reveals large dfferences between non-car and car-ownng households, n terms of ther transport expendtures and ther ncome and soco-demographc characterstcs. For those households whch do not own a car, expendtures per capta on bus and tax fares are consderabl hgher than n car-ownng households, whle the proportons recordng these expendtures are also hgher n non carownng households. In terms of household ncome and soco-demographc characterstcs, sgnfcant dfferences between non car- and car-ownng households are evdent for man of the varables, especall household ncome, the gender and educaton level of the HOH and the tpe of accommodaton and workng status of the household. On the bass of these dfferences, and n 1

13 an attempt to avod multcollneart, the regressons for bus and tax fare expendtures are estmated separatel for (a) households not ownng a car (950 households or 44. per cent of the total sample), and (b) households ownng one car (935 households or 43.5 per cent of the total sample). 13

14 Table 3. Transport Varables Full Sample No Car One Car 1 Varable Mean St.Dev. Max. % pos Mean St.Dev. Max. % pos Mean St.Dev. Max. % pos CAR CYCLE PETROLEXP BUSEXP TAXIEXP FREETRAV EXPENSES Table 3.3 Household Income and Soco-Demographc Varables (Contnuous) Full Sample No Car One Car Varable Mean St.Dev. Max. % pos Mean St.Dev. Max. % pos Mean St.Dev. Max. % pos HHEXP ADULTS CHILDREN Table 3.4 Household Soco-Demographc Varables (Dscrete): Percentage of Sample n each Categor Varable Full Sample No Car One Car APART SEMI (Base=Detached) WORKING (Base= No household member workng) FEMALE (Base=All other households) SIXTY FIFTY FORTY THIRTY (Base =0-9 ears) PRIMARY SECONDARY (Base=Thrd Level Educaton) Due to the small number of households ownng two or more cars (63 households or 1.3 per cent of the total sample) and the consequent poor ft of the regressons, onl the regressons for samples ownng no car and one car are reported and dscussed. 14

15 4 Estmaton Results For all sx regressons, the same set of ndependent varables s emploed for comparson purposes. For all the contnuous household ncome and soco-demographc ndependent varables, non-lnear terms are ncluded n the specfcaton where the are sgnfcant 13 to account for the fact that the effect of these ndependent varables ma dffer over the range of the varables [see Alperovch et al (1999) and Cragg and Uhler (1970)]. All models are estmated usng the LIMDEP econometrc package. Lkelhood rato tests were used to decde on the most approprate model specfcaton. In allowng for heteroscedastct, onl sgnfcant contnuous ndependent varables were ncluded n the heteroscedastct functon [Yen et al (1996)]. In the bnar probt model, all of the ndependent varables n the heteroscedastct term were nsgnfcant,.e., the lkelhood rato test of a homoscedastc error structure was not reected at the one per cent level. However n all of the Tobt models, the sgnfcance of at least one contnuous ndependent varable n the heteroscedastct adustment and of the IHS parameters resulted n a reecton of the null hpothess of a homoscedastc and normal error structure at the one per cent level of sgnfcance. 14 In both the bnar probt and Tobt models, the estmated coeffcents cannot be nterpreted n the same wa as n a lnear regresson model. Margnal effects for the contnuous ndependent varables n the model are calculated b dfferentatng the expected value of the dependent 13 Non-lnear terms sgnfcant n the standard Tobt models were also ncluded n the specfcaton of the adusted Tobt models. 14 In all cases the models adusted for heteroscedastct and non-normalt performed the best n terms of the estmated log-lkelhood values. See Tables B1-B3 of Appendx B for the results of the three lkelhood rato tests of heteroscedastc errors, non-normal errors and heteroscedastc and non-normal errors. 15

16 varable wth respect to the ndependent varable of nterest, evaluated at the sample mean of ths ndependent varable. Ths enables the calculaton of elastctes of demand wth respect to these contnuous ndependent varables. The margnal effects for dscrete ndependent varables are calculated as the dfference n expected values when the varable takes the value one and when t takes the value zero. In order to ascertan the relablt of all margnal effects, standard errors for the margnal effects must be calculated. These are approxmated usng the delta method as presented n Yen et al (1996). The formulae for calculatng margnal effects n both the probt and Tobt models are presented n Appendx A whle those for the standard errors are presented n Appendx B. As the calculaton of standard errors for the margnal effects n the non-normal and heteroscedastc Tobt models s work n progress, the standard errors and sgnfcance levels presented are those of the estmated coeffcents. Tables 4.1 to 4.3. below present the estmated margnal effects for the correctl specfed car ownershp, car use, bus fare expendture and tax fare expendture models respectvel. For comparson purposes, the margnal effects for the unadusted Tobt models are presented n Tables C1-C3 of Appendx C. Table D1 of Appendx D reports the estmated elastctes of demand wth respect to the three contnuous ndependent varables for all models Car Ownershp The car ownershp model performs partcularl well wth all varables sgnfcant at the fve per cent level or less. Most of these results are n lne wth those of other studes wth the excepton of the effect of the age of the HOH, whch exerts a postve effect on the probablt of car 16

17 ownershp. The ncome elastct of car ownershp demand of 1.15 supports the results of man studes that have classfed the demand for the prvate car as a luxur good [see Blundell et al (1993), Button et al (1993), Stanovnk (1990), Thoban (1984), McCarth (1977), Buxton and Rhs (197), Bennett (1967) and Farhurst (1965)]. 15 Table 4.1 Bnar Probt Model of Household Car Ownershp Varable Margnal Effects Constant (0.196)*** HHEXP 0.33 (0.089)*** ADULTS (0.169)** CHILDREN (0.03)*** APART (0.070)*** SEMI (0.059)*** WORKING (0.037)*** FEMALE (0.051)*** THIRTY (0.077)*** FORTY (0.079)*** FIFTY (0.080)*** SIXTY 0.66 (0.077)*** PRIMARY (0.049)*** SECONDARY (0.044)** Number of Observatons, Non-Lnear Terms HHEXP, HHEXP 3, ADULTS, ADULTS 3, CHILDREN Log-Lkelhood Heteroscedastc Log-Lkelhood (3) Elastctes: 15 However man studes show that the ncome elastct of car ownershp demand s declnng over tme wth the prvate car now ncreasngl regarded as a necesst rather than a luxur [see Borner (1999), Darga and Vthoulkas (1999) and McCarth (1977)]. In addton man of the former studes that found ncome elastctes n excess of unt emploed data on low ncome countres or are now out of date. 16 The car ownershp regresson s estmated usng the full sample of,148 households (.e., those households ownng two or more cars are also ncluded). An examnaton of Table E1 n Appendx E reveals lttle dfference between these results and those based on the sample of households one car or less. 17

18 HHEXP 1.145*** ADULTS 0.197** CHIDLREN 0.178*** Notes: * sgnfcant at 10% level, ** sgnfcant at 5% level, *** sgnfcant at 1% level Standard errors for elastctes are those of the margnal effects. For all the contnuous ndependent varables (ncome and the number of adults and chldren n the household), non-lnear terms are found to be sgnfcant. Ths s consstent wth Darga and Vthoulkas (1999), Cragg and Uhler (1970) and Bennett (1967) who also found that the postve effect of ncome on the probablt of car ownershp decreases n magntude as ncome ncreases whle Alperovch et al (1999) found a smlar result for the number of adults n the household. In addton, the sgnfcance of the cubed terms suggests that once household ncome and the number of adults ncreases above a second threshold, the probablt of car ownershp ncreases at an ncreasng rate once more. For the number of adults and chldren n the household, the sgnfcance of the squared term ma ndcate a scale economes effect whle the sgnfcance of the cubed term for the number of adults ma suggest ntal ncreasng returns to scale that dmnsh as the sze of the household becomes larger. The effect of accommodaton tpe supports the ncluson of ths varable as a prox for dstance and the qualt and quantt of publc transport lnks wth those households lvng n apartments beng least lkel to own cars n comparson wth the base categor of those lvng n detached houses. Numerous studes also fnd that as dstance from the ct centre ncreases, populaton denst declnes and publc transport provson deterorates, the demand for car ownershp ncreases [see Alperovch et al (1999), Darga and Vthoulkas (1999), Tran (1980), McCarth (1977), Buxton and Rhs (197), Cragg and Uhler (1970), Bennett (1967) and Farhurst (1965)]. The postve effect of havng at least one person n emploment n the household s n agreement 18

19 wth the results of Cragg and Uhler (1970) and Bennett (1967) and ndcates the effect that the presence of ndvduals n the household wth regular moblt needs has on car ownershp probablt. Whle there s no a pror reason wh the gender of the HOH per se should nfluence the probablt of car ownershp, sngle female-headed households are sgnfcantl less lkel to own cars than other household tpes. The results for age of HOH show that the probablt of car ownershp ncreases wth the age of the HOH. These results are n conflct wth those of Alperovch et al (1999), Darga and Vthoulkas (1999) and Bennett (1967) where the effect of the age of HOH, whle ntall postve, decreases n magntude as the age of the HOH ncreases 17. Explanatons for ths dvergence ma le n dfferent costs, wth the costs of car nsurance beng partcularl hgh for oung people n Ireland. Fnall, the hghest level of educaton varable suggests that those wth a prmar educaton are least lkel to own a car relatve to the base categor of those wth a thrd level qualfcaton. Alperovch et al (1999) also fnd that those wth the hghest levels of educaton are most lkel to own cars although the admt that there s no obvous reason wh ths should be the case, gven that household ncome has been taken nto account. 17 Cragg and Uhler (1970) even fnd that ncreasng the age of the HOH lnearl reduces the probablt of car ownershp. 19

20 4. Car Use Table 4. IHS Heteroscedastc Tobt Model of Household Car Use Varable Margnal Effects Heteroscedastc Terms Constant.430 (3.169) HHEXP (0.959)*** ADULTS (0.37)*** CHILDREN (0.308)*** 0.13 (0.0)*** (0.01)*** APART (1.57)** SEMI (0.84)*** WORKING.805 (0.801)*** FEMALE -.63 (1.00)** THIRTY (.534) FORTY (.535) FIFTY (.593) SIXTY (.556) PRIMARY (0.93) SECONDARY (0.760) MOTOR (1.81)* EXPENSES (0.776)*** 5.70 (0.376)*** (0.006)*** Number of Observatons 935 Non-Lnear Terms HHEXP Log-Lkelhood IHS Het Log-Lkelhood (3) Elastctes: HHEXP 0.514*** ADULTS 0.5*** CHILDREN 0.115*** Notes: *sgnfcant at 10% level, ** sgnfcant at 5% level, *** sgnfcant at 1% level Standard errors (n parentheses) and sgnfcance levels are those of the estmated coeffcents 0

21 In explanng varatons n petrol expendture for those households ownng one car, household ncome, the number of adults and chldren, household locaton, the presence of workers n the household and the gender of the HOH are all sgnfcant explanator factors. For those households ownng one car, household ncome enters as a postve and sgnfcant varable n explanng household car use, n part reflectng the costs such as petrol, tax and nsurance that are ncurred n runnng a car. It ma also ndcate that households wth hgher ncomes place a greater value on tme savngs and comfort relatve to poorer households, thus choosng the car over more tme consumng and less comfortable methods of transport such as cclng or usng the bus. The low but postve ncome elastct of 0.51 suggests that car use demand ma be classfed as a necesst, a result consstent wth those of Borner (1999), Labeaga and Lopez (1997), Blundell et al (1993) and Mannerng (1983). The sgnfcance of the squared term ndcates that a non-lnear relatonshp exsts, a result also found b Kaser (000). The effects of the number of adults and chldren 18 n the household on petrol expendture are both postve and lnear, ndcatng the effects of the number of elgble drvers and dverse household actvt patterns on car use. The accommodaton tpe varables are of the expected sgn, showng how car use ncreases wth dstance and/or the non-avalablt of publc transport. The results are ver smlar to those of Borner (1999) who fnds that relatve to those lvng n the ct of Copenhagen, those lvng n rural areas of Denmark use ther cars the most n terms of annual mleage. In common wth the result of Kaser (000), the presence of one or more persons n emploment ncreases car use, suggestng that the effect of regular commutng patterns on car use s postve. The effect of a sngle female HOH s negatve and sgnfcant, consstent wth the results of Kaser (000), 18 Kaser (000) fnds no effect for the number of chldren n the household on petrol expendture. 1

22 Borner (1999) and Mannerng (1983). Ths ma mean that such sngle women are engaged n actvtes that requre less travellng such as part-tme local work and/or are more wllng to walk and use publc transport than men. 19 Even wth access to a car, sngle female households use ther cars less than all other household tpes. The age of the HOH s not sgnfcant although the sgns of the effects are, wth one excepton (50-59 ears), consstent wth the results of Borner (1999) and Mannerng (1983) who fnd that car use declnes wth the age of the HOH. 19 Ths s assumng that the HOH s the prncpal drver n the household.

23 4 Publc Transport Expendture Bus Fare Expendture Table IHS Heteroscedastc Tobt Models of Household Bus Fare Expendture 1994/1995 Varable No Car One Car Margnal Effects Het Terms Margnal Effects Het Terms Constant (1.147)*** -3.3 (0.893)*** HHEXP (0.578)*** 0.18 (0.056)*** (0.076) ADULTS (0.473)*** (0.034)** (0.397)*** CHILDREN (0.08) (0.0)** (0.055) APART (0.736) (0.384) SEMI (0.711) (0.10) WORKING (0.1)*** (0.3)** FEMALE (0.34)*** (0.347)*** THIRTY (0.385) (0.544) FORTY (0.44)* (0.534) FIFTY (0.465) (0.541) SIXTY (0.465)*** (0.54) PRIMARY (0.405)*** (0.6) SECOND (0.399)** (0.179) FREETRAV (0.10)*** (0.19)*** CYCLE (0.308)*** (0.51)*** (0.405) EXPENSES (0.194)**.39 (0.69)*** (0.041)** (0.03)*** (0.084)*.67 (0.77)*** 0.49 (0.070)*** 0.11 (0.033)*** Number of Observatons Non-Lnear Terms HHEXP, ADULTS ADULTS Log-Lkelhood IHS Het Log-Lkelhood (5) (4) Elastctes: HHEXP 0.686*** ADULTS 0.65*** 1.70*** CHILDREN Notes: sgnfcant at 10% level, ** sgnfcant at 5% level, *** sgnfcant at 1% level Standard Errors (n parentheses) and sgnfcance levels are those of the estmated coeffcents 3

24 Sgnfcant dfferences exst between car- and non car-ownng households n terms of the factors nfluencng per capta bus fare expendtures. Whle household ncome has a postve and sgnfcant, but dmnshng, effect on the demand for bus travel n non car-ownng households, t s nsgnfcant n explanng varatons n per capta bus fare expendtures n car-ownng households. In non car-ownng households therefore, the demand for urban bus travel ma be classfed as a necesst wth an ncome elastct of Ths result s smlar to that found for publc transport b Bergantno (1997) on the bass of UK mcro-data. The number of adults n the household mpacts postvel on per capta bus fare expendture n both samples wth the sgnfcance of the squared terms ndcatng that a non-lnear relatonshp exsts. The hgh postve elastct of 1.7 n car-ownng households suggests that household members ma compete for the use of the household car. The effect of the number of adults on expendture n both non car- and car-ownng households s partcularl sgnfcant gven that the dependent expendture varable s alread adusted for household sze. Whle the effect of the number of chldren under the age of 17 ears s nsgnfcant n explanng expendture on bus fares n both sets of households, the negatve sgn of the elastct n car-ownng households s consstent wth the results of De Palma and Rochat (000) and Bergantno (1997) who fnd that the effect of chldren on publc transport demand s negatve, reflectng perhaps the returns to scale nvolved n drvng chldren to school. A postve and sgnfcant relatonshp exsts between a sngle female HOH and per capta bus fare expendture. Even when sngle female households have access to a car, the spend more on bus fares per capta than other car-ownng households thus renforcng the pont of Mannerng (1983) that females select frequences and tpes of actvtes that requre less vehcular travel 4

25 than males. The age of HOH varable s nsgnfcant n car-ownng households but explans some varaton n bus fare expendtures n non car-ownng households. The results ndcate that onl those households wth a HOH aged and 60 ears and over spend sgnfcantl less amounts to those households wth a HOH n the base categor. The latter result s all the more sgnfcant gven that the presence of free publc transport for pensoners s also controlled for and exerts a negatve and sgnfcant effect on per capta bus expendtures. These results, whle onl sgnfcant for non car-ownng households, are n drect contrast to those of De Palma and Rochat (000) and Bergantno (1997) who fnd that older people are more lkel to use publc transport than ounger people. The level of educaton of the HOH s onl sgnfcant n the non car-ownng sample and ndcates that those wth a thrd level educaton spend the most per capta on bus fares. Ths result ma be explaned b households consstng solel of thrd level students who use the bus to travel to college. De Palma and Rochat (000) smlarl fnd that those wth a thrd level educaton are more lkel to use publc transport to travel to work, although ths effect was found for ndvduals ownng cars. Fnall, for bus fare expendture n car-ownng households, the presence of at least one person enttled to remuneraton for motor expenses reduces per capta expendture on bus fares, ndcatng that f the opton of cheaper prvate transport s avalable t wll be chosen. 5

26 4.3. Tax Fare Expendture Table 4.3. IHS Heteroscedastc Tobt Models of Household Tax Fare Expendture Varable No Car One Car Margnal Effects Het Terms Margnal Effects Het Terms Constant -.93 (.5)*** (.914)*** HHEXP (0.51)*** (0.063)*** (1.08)*** (0.066)*** ADULTS (0.70)*** (0.040)*** 0.80 (0.34)*** (0.044)*** CHILDREN (0.1)** (0.04)** (0.69) APART (1.410) (1.9) SEMI 0.30 (1.314) (0.89)*** WORKING (0.56)*** 0.3 (0.819) FEMALE 0.10 (0.708) (1.057) THIRTY (0.83) (1.707)*** FORTY (0.898)** (1.75)*** FIFTY (0.964)** (1.693)** SIXTY (1.040)*** -1.1 (1.71)*** PRIMARY (0.993) (0.789) SECOND (0.913) 0.00 (0.659) CYCLE (1.69) (1.011) EXPENSES (0.618) (0.89)*** (1.034)*** (0.09)*** (0.040)*** Number of Observatons Non-Lnear Terms HHEXP Log-Lkelhood IHS Het Log-Lkelhood (3) Elastctes: HHEXP 0.934*** 1.178*** ADULTS 0.069*** 0.641*** CHILDREN -0.05** 0.00 Notes: sgnfcant at 10% level, ** sgnfcant at 5% level, *** sgnfcant at 1% level Standard Errors (n parentheses) and sgnfcance levels are those of the estmated coeffcents 6

27 The dfferences between non car- and car-ownng households tax fare expendtures are less obvous than for the bus fare expendture case. Household ncome has a postve and sgnfcant effect n both non car- and car-ownng households wth the effect also found to be non-lnear n car-ownng households. The former result s n contrast to that found for the bus fare expendture case where household ncome was onl sgnfcant n explanng bus fare expendtures n non car-ownng households. However, the elastct of demand s slghtl hgher n non car-ownng households than n car-ownng households meanng that households wthout cars are more responsve to changes n ncome than those wth cars. The margnal effect of ncreasng number of adults n the household s postve and lnear n both samples, wth the large postve elastct of 0.64 n car-ownng households suggestng that there s competton for the household car. The effect of chldren s nsgnfcant, a result consstent wth expectatons 0. The effect of accommodaton tpe shows that there are no sgnfcant dfferences between households lvng n apartments and households lvng n detached houses n terms of ther tax fare expendture but those n sem-detached houses spend sgnfcantl more. Ths result ma mean that dstance and cost nfluences tax fare expendture wth those lvng furthest awa from the ct centre (detached homes) spendng least because of cost whle those lvng near the ct centre (apartments) can use alternatve, cheaper forms of transport such as walkng and takng the bus. The presence of at least one workng member n the household has a postve and sgnfcant effect on tax fare expendture onl n non car-ownng households, whch ma ndcate the use of taxs for occasonal commutng. The nsgnfcant effect n car-ownng households suggests that taxs are used manl for lesure travel rather than commutng. 0 It s dffcult to explan wh the presence of chldren should be sgnfcant n explanng varatons n tax fare expendture except where t ma prox household tpe. 7

28 The postve effect of a sngle female HOH s a result consstent wth the ncreased safet concerns that women face n travellng alone at nght, whle the effect n car-ownng households renforces the ponts made above about females usng ther cars less n general than males. The effect of ncreasng HOH age on tax expendture s generall negatve n both samples wth those households wth a head aged 0-9 ears spendng the most per person on tax fares. Ths perhaps reflects the dfference n actvtes undertaken b households n dfferent stages of the lfe-ccle and the general retcence of older people to use taxs, for example, ounger households ma socalse more than older households and therefore requre late nght transport, of whch taxs are a popular and often necessar form due to the poor nght bus servce n the ct. 6 Concluson In ths paper Irsh Household Budget Surve mcro-data are used to estmate the ncome and soco-demographc determnants of urban households transport decsons. The most sgnfcant results relate to the effects of ncome (as proxed b total household expendture), the number of adults n the household and the gender and age of the HOH. Household ncome s, wth one excepton 1, postve and sgnfcant n explanng dfferences n households transport decsons. An examnaton of ncome elastctes of demand reveals that whle prvate car ownershp ma be classfed as a luxur, car and publc transport use are necesstes. Along wth the hgher elastctes of demand for tax fare expendture, the nsgnfcance of ncome n explanng varatons n per capta bus fares n car-ownng households ma suggest that factors other than ncome and prce are more mportant n determnng bus fare expendture, partcularl n carownng households. Whle the effects of the number of adults on household car ownershp and use are ver smlar, n terms of publc transport expendture there are sgnfcant dfferences 1 Household ncome s nsgnfcant n explanng varatons n bus fare expendture n car-ownng households. 8

29 between non car- and car-ownng households. The effect of the number of adults s hghl elastc n car-ownng households for both bus and tax fare expendture suggestng that competton for the household car nduces some members to choose alternatve forms of transport. The postve effect of the number of adults on per capta tax fares n non car-ownng households, gven that household sze has alread been accounted for, ma be explaned b the tendenc for larger households to consst of unrelated ndvduals who do not travel together. The gender of the HOH s consstentl sgnfcant n explanng varatons n car ownershp, car use and publc transport expendtures. Whle car ownershp and use are more lkel for households headed b a male, even n households that own a car, bus and tax fare expendtures are hgher for households that are headed b a sngle female. The effects of the age of the HOH on car use, bus fare expendture and tax fare expendture are consstent wth the expectaton that ounger households are more moble and are engaged n more actvtes than older households. The postve effect of age of the HOH on car ownershp s however n conflct wth man other studes and ma reflect dfferent costs of car ownershp n Ireland n comparson wth other countres, wth the costs of nsurance beng partcularl hgh for oung people. In concluson, the results hghlght the sgnfcance of household ncome and soco-demographc characterstcs n determnng dfferences n household transport behavour. The use of expendture data to prox car and publc transport use s ustfed b results that are broadl n lne wth those of prevous research n the area. The dvson of the samples for the bus and tax fare regressons emphasses the dfferences n travel behavour for households ownng cars and those not ownng cars whle the analss of bus and tax expendtures separatel shows the 9

30 usefulness of examnng transport expendtures at more dsaggregated levels. Fnall, the mportance of correct model specfcaton s hghlghted through the mprovements n estmated log-lkelhoods as a result of the ncluson of non-lnear terms for the contnuous ndependent varables and the adustments to the Tobt models for non-normalt. 30

31 Bblograph Alperovch, G., Deutsch, J. and Machnes, Y., The Demand for Car Ownershp: Evdence from Israel Data. Internatonal Journal of Transport Economcs, 6 (3), Arabmazar, A. and Schmdt, P., 198. An Investgaton of the Robustness of the Tobt Estmator to Non-Normalt. Econometrca, 50 (4), Ben-Akva., M. and Lerman, S., Dsaggregate Behavoural Model of Automoble Ownershp. Transportaton Research Record, 569, Bennett, W., Cross-Secton Studes of the Consumpton of Automobles n the Unted States. Amercan Economc Revew, 57 (4), Bergantno, A., Estmatng Engel Curves for Transport Expendtures: Evdence from UK Household Budget Data. Internatonal Journal of Transport Economcs, 4, Borner, T., Demand for Car Ownershp and Use n Denmark: A Mcro Econometrc Model. Internatonal Journal of Transport Economcs, 6 (3), Blundell, R., Pashardes, P. and Weber, G., What do we learn about Consumer Demand from Mcro-Data? Amercan Economc Revew, 83 (3), Button, K., Ngoe, N. and Hne, J., Modellng Vehcle Ownershp and Use n Low Income Countres. Journal of Transport Economcs and Polc, 7 (1), Buxton, M. and Rhs, D., 197. The Demand for Car Ownershp: A Note. Scottsh Journal of Poltcal Econom, 9, Ca, L., Analzng Household Food Expendture Patterns on Trps and Vacatons: A Tobt Model. Journal of Hosptalt and Toursm Research, (4),

32 Cragg, J. and Uhler, R., The Demand for Automobles. Canadan Journal of Economcs, 3 (3), Cragg, J., Some Statstcal Models for Lmted Dependent Varables wth Applcaton to the Demand for Durable Goods. Econometrca, 39 (5), Darga, J. and Vthoulkas, P., Estmaton of a Dnamc Car Ownershp Model: A Pseudo- Panel Approach. Journal of Transport Economcs and Polc, 33 (3), De Donnea, F., The Determnants of Transport Mode Choce n Dutch Ctes: Some Dsaggregate Stochastc Choce Models. Rotterdam, Rotterdam Unverst Press. De Palma, A. and Rochat, D., 000. Mode Choces for Trps to Work n Geneva: An Emprcal Analss. Journal of Transport Geograph, 8, Farhurst, M., The Influence of Publc Transport on Car Ownershp n London. Journal of Transport Economcs and Polc, September, Gould, B. and Km, J., Characterstcs of Canadan and Mexcan Dar Product Purchases: A Comparson usng Household Expendture Data. Unverst of Wsconsn, College of Agrcultural and Lfe Scences, Babcock Insttute Dscusson Paper No Greene, W., Econometrc Analss. London, Prentce-Hall Internatonal. Hamlton Lankford, R. and Wckoff, J., Modellng Chartable Gvng usng a Box-Cox Standard Tobt Model. The Revew of Economcs and Statstcs, 73 (3), Kaser, H., 000. Gasolne Demand and Car Choce: Estmatng Gasolne Demand usng Household Informaton. Energ Economcs,,