Demand for Environmental Quality: An Empirical Analysis of Consumer Behavior in Sweden

Transcription

1 : An Empirical Analysis of Consumer Beavior in Sweden Tarek Galwas Department of Economics, Umeå Universy SE Umeå, Sweden Abstract In tis paper we estimate te income elasticy of demand for recreational services and oter tradional groups of goods in Sweden and test for potential canges in suc estimates over te twentiet century. Due to te difficulty of directly observing te demand for recreational services, we employ an indirect metodology by using te demand for some outdoor goods as a proxy for te demand for recreational services. In line w most prior researc, our results confirm te expectation tat recreational services, as a public good, is a luxury good in Sweden. Our results also sow tat te income elasticies for tradional goods are stable over time, indicating tat consumer preferences for expendure on tese specific commodies do not cange over time. Keywords: Houseold demand; environmental services; income elasticies; Engel curves. JEL Classification: D12; H41; Q26. Te autor acknowledges financial support from te Bank of Sweden Tercentenary Foundation (Riksbankens Jubileumsfond, Kulturdonationen). Te autor also wises to tank Runar Brännlund, Jonas Nordström, Peter Berck, and Karl-Gustaf Löfgren for valuable comments.

2

3 1. Introduction One of te main goals in studying individual consumption beavior is to analyze te relationsips between commody expendure and income or total expendure (i.e. te well-known Engel curves). Tere are several reasons wy tese relationsips are of interest. Firstly, may be useful to see ow demand for various goods develops under different growt scenarios. A second reason is to determine weter consumer preferences regarding various commodies ave canged over time. Te main objective of tis paper is to compare ow demand for recreational services and oter major private goods in Sweden reacts to canges in income and, in particular, ow tese reactions ave canged over time. Specifically, we investigate ow te income elasticy in Sweden as canged over time w respect to some specific goods tat are closely related (complementary) to environmental services. 1 Because consumption of recreational services is not directly observed in te market, te study uses te demand for complementary market products as a proxy for te demand for environmental services. Furter, te objective of tis paper is closely related to te notion of an environmental Kuznets curve (EKC), wic describes a specific relation between environmental qualy and growt. Several studies ave examined te income elasticy for different commodies using data from different countries in different time frames. For example, Segal (2001) reports tat te budget sare for food as fallen dramatically in te Uned States over te twentiet century, from 50% for poor ouseolds and 30% for affluent ouseolds in 1901 to 10-15% in Segal s (2001) finding reaffirms Engel s law of food from On te oter and, Segal (2001) found tat te budget sare for transportation increased from about 2% to more tan 20%. Suc results indicate a remarkable instabily of te budget sare for food and transportation expendure in te U.S. over te century. 1 2 For convenience, we use te label environmental services for all goods and services provided by te environmental and te ecological system, including environmental qualy, see for example, Mäler (1974). Formulated by German-born statistician, Ernst Engel ( ), Engel's Law states tat as incomes increase, te proportion of income spent on food falls. 1

4 Larsen (2001) used Norwegian survey data on purcasing beavior for equipment and lodging over te period He found tat te income elasticy was fairly stable over time, and tat bot equipment and lodging were luxury goods over te wole period. Furter, Miles et al. (2002) used several models, bot parametric and nonparametric, to estimate Engel curves using survey data from Uruguay. Tey found tat te results differ substantially depending on model specification and estimation metod. However, teir results support te ypotesis tat te environment is a luxury good in Uruguay. Kriström and Riera (1996) using estimates of te willingness to pay for environmental goods for different European countries (Finland, France, Norway, Holland, Spain and Sweden), found tat te ypotesis tat environmental goods are necessary goods cannot be rejected in most cases (income elasticy is less tan one). Te estimation of income elasticies over several years as strict requirements for data. Te data sets must be comparable, span a substantial period, be of ig qualy, exaustively cover expendure opportunies, and preferably be random samples. Te Swedis Family Expendure Survey (FES) ave some attractive features for te question at and: respondents are randomly selected, data contain information of actual market beavior were budget constraints are observed and obeyed, te classifications of goods are retained over time and te coice set is saturated and exausts purcase possibilies. Time trends are detectable since data span a considerable time period and are comparable over time. Tis paper uses FES data for 1913, 1984, 1988, and Te rest of te paper is organized as follows: In te next two sections, we elaborate furter on te existing lerature, as well as on te teoretical framework for our empirical investigation. Our econometric model is presented in section 4. Section 5 describes te data used in te study. Te results from te model are presented in section 6. Finally, a sort summary and some concluding remarks are given in section Previous Studies Muc discussion exists in te economic lerature of te possible effects of income and economic growt on te environment, including speculation on te possible existence of an environmental Kuznets curve. Tis curve sows an inverted U-saped relation between pollution and per capa income, indicating tat in te early stages of a country s economic development pollution increases up to a turning point and ten 2

5 begins to decrease as per capa income increases furter. Te EKC idea as triggered a good deal of researc, teoretical as well as empirical. Te teoretical lerature as focused mostly on assumptions regarding te relationsip between tecnology/preferences and emissions (Lopez, 1994, Selden and Song, 1995, McConell, 1997, Cicilnisky 1998, de Groot 1999). In general, empirical models are of a reduced form type using cross country data (Grossman and Krueger, 1995, Stern and Common, 2001). An obvious drawback w most of te empirical models is tat tey can only describe te relation, not explain. To understand te mecanisms at work, we need furter knowledge about tecnological progress and ow consumer preferences are formed. Clearly related to tis issue is te question of ow consumer demand for recreational services and environmental goods reacts to income canges. If te income elasticy is greater tan one, tis would be consistent w te EKC ypotesis. However, income elasticy is also important from a distributional perspective, since will tell us wic groups in society will reap te benefs of projects tat improve environmental services. Terefore, suc estimates of demand and income elasticy of recreational services and environmental goods may provide significant information to any cost-benef analysis or ex-post project evaluation; see Kanninen and Kriström (1992), Kriström and Riera (1996), and Hökby and Söderqvist (2001). Te basic problem in te estimation of income elasticy for environmental goods is tat we cannot directly observe individual demand for recreational services due to s public good and/or non-market priced nature. Terefore, we cannot directly estimate te income elasticy for suc goods. To overcome tis problem, two different approaces are suggested in te lerature. Te first approac is to use stated preference data, 3 and te second is to employ an indirect estimation tecnique derived from te fact tat ouseolds ave to purcase complementary goods. Te first approac is a direct approac based on contingent valuation surveys (Kriström and Riera, 1996, Hökby and Söderqvist, 2001). Under tis approac, willingness to pay data is regressed on income and oter individual caracteristics. In Kriström and Riera 3 Tis approac mainly relies on individuals ypotetical beavior on markets set up for environmental service in some survey setting. Te contingent valuation metod (CVM) is widely used in tis approac (Mcell and Carson, 1989 and Batemen and Willis, 1999). 3

6 (1996) willingness to pay data for various environmental goods in a number of European countries are regressed on income. Contrary to te conventional wisdom, Kriström and Riera found tat willingness to pay for environmental improvements decreases w income, wic indicates tat te income elasticy is lower tan one. Using te same metodology, Hökby and Söderqvist (2001) found similar results. A problem w tis approac is tat te magnude of te willingness to pay elasticy w respect to income may not give complete information concerning te demand elasticy w respect to income. 4 To address tis problem, Hökby and Söderqvist also merge data from several willingness to pay studies for te same environmental good, reduced marine eutropication in te Baltic Sea, and again found tat te income elasticy for reduced eutropication is less tan one. Te second approac is an indirect estimation approac based on te fact tat individuals, in order to generate utily from te environment, need private goods tat are bougt and sold in te market and, terefore, can be observed. For example, to enjoy te excement of a salmon river is necessary to ave some fising gear, or at least some outdoor gear. Tus, if demand for fising gear and oter goods tat are closely related to te consumption of environmental amenies increase more tan proportionally w te increase in income, te interpretation may be tat te environment is a luxury good (Mäler, 1974). Tere are a number of previous studies tat ave used te indirect approac, suc as Costa (1997), Pereyra and Rossi (1998), Miles et al. (2002), and Larsen (2001). Costa (1997), using U.S. data, reported elasticies greater tan one for recreation goods. However, se found tat tese elasticies decreased significantly over te last undred years. Pereyra and Rossi (1998) applied a parametric metod using data from Uruguay and found corroborative evidence tat environmental goods constute a luxury good. Miles et al. (2002) used parametric estimates to confirm te ypotesis tat te outdoor recreational services constute a luxury good in Uruguay. Larsen (2001) used Norwegian survey data on purcasing beavior for equipment and lodging over te period to estimate Engel curves. He found tat te income elasticy was 4 Te income elasticy of willingness to pay and te ordinary income elasticy of demand are related. However, knowledge of one is insufficient to determine te magnude or even te sign of te oter. Te income elasticy of willingness to pay is influenced by addional factors tat are generally unobservable. For more details, see Flores and Carson (1997). 4

7 fairly stable over time, and tat bot equipment and lodging were luxury goods over te wole period. In tis study, we follow te second approac, using ouseold survey data for Sweden. Using tis data, we estimate Engel curves for private goods tat are used in te production of environmental services. As well as estimating te income elasticy for proxy goods used for recreational services (outdoor recreation), we also estimate te income elasticy of demand for oter tradional market goods in Sweden over te same period in order to compare relative canges in consumer preferences. To acieve our objectives, we formulate and estimate an econometric model for purely private goods and for private goods tat are complementary to public goods. Te model employed is based on te assumption of a two-stage budgeting process. It is assumed tat in te first stage, te ouseold allocates s total expendure for purely private goods and goods complementary to recreation on different commody aggregates, or groups. Tere are five groups: one group of goods complementary to outdoor recreation, and four purely private groups: food, transportation, energy goods, and oter goods. Given te allocation to eac commody group, ouseolds in te second stage allocate teir group expendures on te various goods win te group. Our econometric model is based on Deaton & Muellbauer s (1980) almost ideal model (AIDS). Te inclusion of data from a budget survey for 1913 enables us to compare te results over a longer time span. 3. Teory Te teory beind our approac can be outlined as follows. Assume tat individuals ave preferences over a vector of private goods x = [x 1,, x K ] and a vector of environmental commodies (experiences), e = [e 1,, e P ], tat can be translated into a utily function tat is weakly separable in x and e: U(x, e) = U(x, u e (e)) (1) 5

8 Following Freeman (2003), we assume tat environmental commodies, e, are produced using environmental attributes, 5 A = A,..., A ), and market goods, z = ( 1 m z,..., z ), according to: ( 1 k e r = e( z, A) r = 1,..., l (2) Te production function (2) as te properties tat e is increasing in A and z, and tat all inputs are essential in te production of e. In oter words, bot environmental attributes and market goods contribute to production of outdoor recreational experiences. By substuting (2) into (1), we obtain te following optimization problem: max x, z e { U ( x, u ( e( z, A)) } s. t. p x + p z y, x z (3) were p x and p z are te price vectors corresponding to x and z respectively, and y is te expendure on private and complementary goods. Te first order condions to tis problem implicly define te demand functions for te instrumental goods, z, as a function of prices, income, attributes, preferences, and production tecnology, i.e.: z = z(p, p, A,y) i =,...,m (4) i x z 1 were p x is te price index for private goods, and y is expendure on goods tat are complementary to environmental goods. 6 According to equation (4), canges over time in expendure on z i may result from canges in prices, income, or environmental attributes. However, we do not attempt to account ere for canges in environmental attributes, and terefore consider tem to be constant over time. Tus equation (4) constutes te basis for our analysis, and will serve as a starting point in te specification of te econometric model in te following section. 5 6 Freeman gives examples of environmental attributes, suc as number of fis per volume of water and water qualy. Here, we may add air qualy, sounds, wild-life, ski tracks and number of sunny days. Te reason for using p x instead of a vector of private good prices is te assumption of weak separabily between private and public goods. 6

9 4. Te Modeling Framework In tis section we formulate a demand system for public and private goods. In te first stage, te ouseold determines, given s total budget, ow muc to spend on food, energy goods, transportation, and oter goods as private groups, and outdoor recreation services as a public group. In te second stage, te ouseold allocates resources win eac of tese groups. For example, given a specific amount of money to be spent on transportation, te ouseold determines ow muc of tat sould be allocated to gasoline, car maintenance, and public transport. In te same manner, te ouseold determines in te second stage ow to use s budget for outdoor recreation. In tis case, te ouseold can coose between equipment for sporting, fising and camping, and oter recreational goods. Our main objective is to model and estimate ouseold coices in te first and second stage. Te Linear Almost Ideal Demand system (LAIDS) is one of te most popular demand models for estimation of Engel curves. In te empirical estimation of Engel curves, nonlineary as been found to be important for some goods. For instance, Banks et al. (1997) found tat te Engel curves for some specific goods in te UK are non-linear in te logarm of expendure. To overcome te problem of non-lineary, Banks et al. (1997) developed te Quadratic Almost Ideal Demand system (QUAIDS). In tis paper we take te quadratic AIDS (QUAIDS) model as our basic specification. Given te structure of two-stage budgeting, we can express demand for te complementary goods, z, and pure private aggregates, x, in budget sare form for ouseold as: 7 w w = α + γ ( ln p + γ ln p + β ln R ln P + λ ln R ln P + ε (5) ( z ) t ( z ) zz) t ( zx) t ( z )( t t ) ( t t ) ( z = α + γ ( ln p + γ ln p + β ln R ln P + λ ln R ln P + ε t xz) t ( xx) t ( t t ) ( t t ) ( x Equation (5) describes te budget sare for te public commody group for ouseold w ) = 1,..., H, were ( z t denotes te budget sare for group z in period t, ( z t and ( x t are group price indices for public and private goods, respectively R t is total expendure p ) 2 2 ) t ) t (6) p ) 7 In te estimation, we ave tree main private groups (foodstuff, energy goods, and transportation). Tus x in equation (6) can be viewed as a vector of private goods, and te group price index for private goods as a vector of group price indices. 7

10 on public and private goods, P t is te overall consumer price index, andε t is te error term. In te same manner, equation (6) gives te budget sare for private commodies, were w x ) t ( denotes te budget sare for private goods x in period t, p ( x ) t is a group price index for private goods and ε t is te error term. Te parameters to be estimated are α, γ, β, and λ. Te demand functions for ouseold in goods win te sub-groups ave te same functional form as te demand equations for te main groups. Te demand function for goods win te z t, and te x t, groups can tus be wrten as w 2 ( ln R ln p ) + λ ( ln R p ) ε m = i + γ ij ln p( z) jt + β( z) i t t i t ln j = 1 α + t, (7) were i = 1,..., m denotes te number of goods win z and = 1,..., H denotes ouseolds and w n 2 ( ln R ln p ) + λ ( ln R ln p ) ε, = α i + γ ij ln p( x) jt + β( x) i t t i t t + j = 1 (8) were i = 1,..., n denotes te number of goods win x. Equations (7) and (8) give te allocation win te public and private groups, were w and w ( x ) are te budget sares for te individual goods win eac group, p ( z )j ( z ) and p x ) j ( are te commody prices win respective group, R (z)t is te total expendure on goods complementary to public goods, and R (x)t is te total expendure on te pure private goods. In any time period t, we assume tat te prices of goods are equal across all ouseolds. Tis means tat since we will estimate te model for eac cross-section separately, te prices can be included directly into te intercept term for any time period. Tus, for eac cross-section we can wre te budget sares to be estimated as: 8 8 Tis means tat we can exclude te price in te estimation of eac cross-section. 8

11 2 w = α z t + β z t Rt + λ z t ln Rt + ln ε (9) ( z ) t ( ) ( )( ) ( )( ) w t α ( ) + β ( )( ln R ) + λ( )( ln R ) + ε ( ) w w x t x t t x t t 2 (z)t = (10) = ( z ) ( z ) ( z ) ( z ) t = x t 2 ( R ) + λ( z ) ( ln R( z ) t ) ε ( z ) α + β + t ln, i=1,,m (11) 2 ( R ) + λ ( ln R t ) ε α + β + ln, i=1,,n (12) were t = 1,..., T is te number of cross-sections, and were α (z)t, now include te (constant) price. α (x)t, α, and α According to tis system, equations (9) and (10) describe ow ouseold allocate s total expendure between public and private groups respectively, wile equation (11) and (12) describe te allocation of ouseold expendure to goods win te public and private groups. 5. Data and Econometric Consideration Tis study uses cross-sectional data from four Swedis Family Expendure Surveys (FES) 1913, 1984, 1988, and Te first ouseold expendure survey in Sweden was performed in 1913, covering approximately 900 ouseolds in eigt towns. Te 1984 survey included 4354 ouseolds, te 1988 survey 3764 ouseolds, and te 1996 survey 1104 ouseolds. Te surveys contain expendure data on a rater disaggregated level. Here, owever, we will focus on four main aggregates: 9 Outdoor recreation: Expendure on sporting, fising, and camping equipment, and oter recreation goods. Transport: Expendure on petrol, car maintenance, and public and oter transport. Energy goods: Expendure on electricy, and oter energy goods. Foodstuffs: Expendure on food and beverages. Table 1 presents te budget sares for te four main groups and teir sub-groups. As sown, ouseold expendure for outdoor recreation, transportation, and energy goods 9 Te 1913 survey uses only an aggregate general tle for recreational goods, wic includes outdoor and indoor recreation and gives no data win te groups except for foodstuffs. 9

12 ave been fairly stable over time, altoug tere is a weak negative trend in te energy goods sare. However, te sare of foodstuffs as decreased significantly, wile approximately 45% of ouseold expendure went to foodstuff in 1913, tis sare declined to less tan 30% in 1984, and to less tan 20% in Table 1 also sows tat tere was no dramatic cange in ouseold expendure for most of te goods win te main groups, except for spending on sporting, fising and camping equipment, wic decreased substantially. Approximately 60% of ouseold expendure on recreation went to equipment in 1984, but by 1996 tis sare ad decreased to 35%. Houseold caracteristics may affect consumer beavior w respect to tese four groups of goods. Tere are basically two different ways to consider different ouseold caracteristics in te model estimation (Pollak and Wales, 1992). Te first tecnique is to consider te sample as a wole and use different dummy variables to capture different ouseold caracteristics. Te second is to divide te sample into omogenous sub-samples depending on ouseold caracteristics. In tis paper, we follow te first approac. Te variables relating to ouseold caracteristics were: (a) a continuous variable tat represents te number of adults, and (b) tree dummy variables for te cases wen te ouseold as one cild, two cildren, or more tan two cildren less tan 18 years of age. 10 We also create regional dummy variables (seven for te eigt census regions in te 1913 survey, and five for te six census regions in te 1984, 1988, and 1996 surveys). In te estimation, we use total expendure rater tan income because expendure better reflects permanent income. 10 We also estimated te model by using dummy variables for te number of adults, but tis did not cange te results. 10

13 Table 1. Descriptive statistics of budget sares of various commody goods Budget sare % s.e Budget sare % s.e Budget sare % s.e Budget sare % s.e MAIN GROUPS Foodstuff 0.47 (0.07) 0.30 (0.09) 0.27 (0.09) 0.18 (0.07) Energy goods 0.05 (0.01) 0.14 (0.11) 0.14 (0.10) 0.13 (0.12) Transport 0.02 (0.01) 0.08 (0.06) 0.07 (0.06) 0.08 (0.07) Outdoor recreation 0.01 (0.009) (0.01) (0.02) (0.02) Sum FOODSTUFFS Food 0.97 (0.03) 0.90 (0.09) 0.89 (0.10) 0.86 (0.10) Beverages 0.03 (0.03) 0.10 (0.09) 0.11 (0.10) 0.14 (0.10) Sum ENERGY GOODS Electricy 0.39 (0.40) 0.37 (0.32) 0.34 (0.21) Oter energy goods 0.61 (0.40) 0.63 (0.31) 0.66 (0.28) Sum TRANSPORT Petrol 0.51 (0.33) 0.51 (0.12) 0.60 (0.33) Car maintenance 0.40 (0.31) 0.41 (0.33) 0.28 (0.31) Public and oter transport 0.09 (0.12) 0.08 (0.19) 0.12 (0.21) OUTDOOR RECREATION Equipment for sporting, 0.61 (0.45) 0.52 (0.42) 0.37 (0.44) fising, and camping Oter recreational goods 0.39 (0.42) 0.48 (0.44) 0.63 (0.46) Sum Number of observation Note: Standard errors win parenteses. In tis study, we estimate te quadratic almost ideal demand system in expendure form. To capture te effect of ouseold size and composion on te consumer preferences, we follow Blundell et al. (1993) in allowing te parameters in te model ( α, β, and λ ) to vary over time and over different types of ouseold caracteristics (e.g. ouseold size and composion, as well as place of residence): q ( r ) t = α ( r)0 + α ( r) k k = 1 α D r = z, x (13) q ( r ) t = β ( r)0 + β ( r ) k k = 1 kt β D r = z, x (14) kt 11

14 q ( r ) t = λ( r)0 + λ( r ) k k = 1 λ D r = z, x (15) q = α i0 + α ik k = 1 kt α D i = 1..., m, = 1,..., H (16) q = α i0 + α ik k = 1 kt α D i = 1..., n, = 1,..., H (17) q = β i0 + β ik k = 1 kt β D i = 1..., m, = 1,..., H (18) q = β i0 + β ik k = 1 kt β D i = 1..., n, = 1,..., H (19) q = λ( z) i0 + λ( z) ik k = 1 kt λ D i = 1..., m, = 1,..., H (20) q = λ( x) i0 + λ( x) ik k = 1 kt λ D i = 1..., n, = 1,..., H (21) kt D kt represents dummy variables for demograpic caracteristics, including number of cildren and a dummy variable for different regions. Te number of dummy variables, q, is equal to 11 in te 1913 survey, and 9 in te1984, 1986, and 1996 surveys. Given te estimates of te parameters in te demand model, we can now calculate te income elasticies as 11 ξ i βˆ 2λˆ lnrt = (22) wˆ wˆ Were ξ i denotes ouseold s income elasticy for good i. A good w an income elasticy larger tan one is a luxury, wile a good w an income elasticy lower tan one is a necessy. However, equation (16) implies tat eac good can be eer a necessy or a luxury for different ouseolds, depending upon te distribution of total expendure and te ouseold specific parameters. Finally, te most callenging problem is ow to deal w observed zero expendure, since te parameter estimation tends to be biased in a regression model were a large 11 Te income elasticy for a specific good denotes te percentage cange in te consumption of te good as a result of te percentage cange in total consumption. See Calfant (1987) for a derivation of income elasticiy in an AIDS model. 12

15 proportion of te dependent variable is zero (Deaton 1986, Greene 2000). 12 Tere are at least two possible reasons for an observation of zero. One is tat te ouseold is not interested in te good. Anoter is tat even if a ouseold does ave a preference for a good, expendures may be infrequent and lie outside te observation period. To be sure tat our data is consistent w te estimation results, we estimate te demand equation by an alternative estimator (Tob estimator) wic assumes tat any observation for wic te dependent variable takes a zero value is truncated. 13 However, tis does not cange te results concerning income elasticy to any great extent Results Tis section presents te results of applying te parametric approac to estimate te income elasticy for outdoor recreation, energy goods, transportation, and foodstuffs for Swedis ouseolds in four different years during te twentiet century. We start te analysis by testing te functional form for te expendure equations in order to decide weter te non-linear expendure term sould be included in te model or not. Table 2 sows te results of tis test. According to te F-tests in table 2, we cannot reject lineary for any of te main groups in any of te surveys except for te energy goods group in 1984 and Among te sub-groups, lineary cannot be rejected for any of te goods win te foodstuffs group in all surveys, for public and oter transport in te 1984, 1988, and 1996 surveys, or for any goods win te recreation group in te 1996 survey Te proportion of zero expendure on outdoor recreation in our surveys is approximately 35%. Te standard Tob model was originally formulated by Tobin (1958). Te results from te Tob estimator was compared w OLS estimator wic we used in tis estimation. 13

16 Table 2. Test for Lineary (F-test) MAIN GROUPS Foodstuff Energy goods * 5.82* 0.89 Transport Outdoor recreation FOODSTUFFS Food Beverages ENERGY GOODS Electricy 8.19* 12.55* 1.93 Oter energy goods 8.19* 12.56* 1.93 TRANSPORT Petrol 31.54* 22.36* 13.60* Car maintenance 31.89* 23.22* 11.80* Public and oter transport OUTDOOR RECREATION Equipment for sporting, 3.68* 2.43* 1.71 fising, and camping Oter recreational goods 3.68* 2.43* 1.72 * Lineary rejected at te 5% level. Following te specifications in equations (9), (10), (11), and (12), te demand equations for te goods in te main groups and win te main groups are estimated by ordinary least squares (OLS) and te functional form is determined according to te lineary tests above. 15 For example, in te main group, te linear form sould be used in te estimation of every group except energy goods for te 1913 and 1988 surveys. Table 3 presents estimates w standard errors of te income elasticies of te various goods. Tese elasticies are computed from te coefficient estimates, te estimated budget sares, and te mean total expendures for all ouseolds in every survey, following equation (22). Standard errors are computed w te delta metod (see Greene, 2000). 15 Estimates w standard errors of te parameters of te demand equations are available from te autor upon request. 14

17 Table 3. Estimated income elasticies, standard errors win parenteses. Income elasticy Foodstuff 0.59 (0.01) Energy goods 0.55 (0.04) Transport 1.22 (0.09) Outdoor 1.78 recreation (0.08) Food 0.98 (0.002) Beverages 1.04 (0.06) Total income elasticy Income elasticy Total income elasticy MAIN GROUPS 0.53 (0.01) 0.31 (0.04) 0.99 (0.03) 1.87 (0.15) Income elasticy 0.48 (0.01) 0.22 (0.02) 0.97 (0.03) 1.94 (0.16) FOODSTUFFS (0.004) (0.01) (0.03) (0.04) ENERGY GOODS Electricy 0.70 (0.02) (0.02) Oter energy goods (0.09) (0.07) TRANSPORTS Petrol (0.01) (0.01) Car maintenance (0.06) (0.06) Public and oter (0.04) (0.04) transport Equipment for sporting, fising, and camping Oter recreational OUTDOOR RECREATION 1.23 (0.01) 0.62 (0.03) (0.02) (0.05) Total income elasticy Income elasticy 0.46 (0.03) 0.26 (0.04) 1.27 (0.06) 2.05 (0.27) (0.01) (0.05) (0.05) (0.02) (0.02) (0.17) (0.07) (0.01) (0.02) Total income elasticy goods Note: total income elasticy for any good win te main group of goods is calculated by multiplying te income elasticy for te main group by te income elasticy win te main group. Elasticies can be examined in two different ways, focusing eer on te differences between different goods, or on differences over time. Considering differences over time, we can conclude tat outdoor recreation seems to be a luxury good in 1913, and as retained tat classification until Tis result is

18 consistent w previous researc suc as tat of Costa (1997), Pereyra and Rossi (1998), Miles et al. (2002), and Larsen (2001). However, contradicts to some extent te finding of Kriström and Riera (1996), wo sowed some empirical evidence tat environmental amenies are not luxury. Our results support te classification of environmental goods as luxury goods, as income elasticies are estimated above one for all te time periods. Tis, in combination w te fact tat income in Sweden as increased over te last 100 years, implies tat demand for environmental goods as been non-decreasing for tat time period. Tus, if we ave an increase in future income, we may expect a more tan proportional increase in demand for recreation goods. Tis is consistent w te assumed sape of te environmental Kuznets curve (EKC). Table 3 also sows tat te income elasticy for te main groups seems to remain constant over time. Te income elasticy fluctuates around two for outdoor recreation and around one for transportation. On te oter and, te income elasticies for food and energy goods decreased sligtly between 1913 and Tis implies tat we cannot reject te ypotesis tat te budget for tese goods is stable over time, indicating tat consumer preferences for expendures on tese specific commodies ave not canged significantly over time. 17 Concerning te differences between different goods, te results in Table 3 sow tat te income elasticies for transportation fluctuate around one over te various crosssections. For energy goods and foodstuffs, te elasticy is smaller tan one, indicating tat tese goods are considered as necessies. From Table 3, we also notice tat te total income elasticies win te main groups indicate tat equipment for sporting, swimming, and camping, car maintenance and oter recreational goods are luxuries, (i.e. tey ave income elasticies iger tan one), wile food, petrol, public and oter transport, electricy, and oter energy goods are necessies, since tey ave income elasticies tat are less tan one. Two tailed t- tests sow tat te income elasticies for all goods are significantly different from one Remember tat te expendure elasticies are estimated for independent cross-sections, were ouseolds face te same prices. Between cross-sections, tere will be price canges and qualy canges, owever tis is not considered ere. Since te difference in te income elasticies for tese goods is que small, we cannot reject stabily. 16

19 except for te transport group in te 1984 and 1988 surveys and beverages in te 1913 survey (te results from tis test are presented in Table A1 in Appendix A). Income elasticies evaluated at te mean for different categories of ouseold are presented in Tables A2-A5 in Appendix A. From tese results, we see tat tere are no large differences in income elasticies between ouseolds in different regions or of different family size, indicating tat ouseold location and family size do not ave a big impact on consumer preferences for expendure on tese specific commodies. In summary, we find tat outdoor recreation is a luxury good and tat s luxury status seems to be robust over time. Foodstuffs and energy goods, owever, serve as necessies. We may also conclude tat te demand for transportation as increased more tan te demand for foodstuffs and energy goods, but less tan te demand for outdoor recreation. 7. Concluding Remarks In tis paper, we estimated te income elasticy of demand for recreational services and oter tradional groups of goods in Sweden, and tested for potential canges in suc estimates over te twentiet century. Te data were drawn from Swedis ouseold surveys for te years 1913, 1984, 1988, and Because of te difficulty of directly observing te demand for recreational services, we employed an indirect metodology by using te demand for some outdoor goods as a proxy for te recreational services demand. In line w most prior researc, our results confirm te expectation tat recreational services, as a public good, is a luxury good in Sweden. In relation to te sape of an environmental Kuznets curve (EKC), our results support te suggested sape of te EKC, at least to some extent. We found tat outdoor recreational service is a luxury good and tat demand was non-decreasing during te wole period. Our contribution supplements tat of oter studies of te penomenon (Grossman and Krueger, 1995, Hilton and Levinson, 1998, Selden and Song, 1995). Te results also sow tat recreational services ave maintained tis luxury good attribute in Sweden over te twentiet century, indicating no significant cange in consumer preferences over time. Te income elasticy for transportation goods fluctuated around one during te period of interest, wile bot energy goods and food 17

20 maintained teir attributes as necessies during tis period, w a steady decline in teir income elasticies over time. According to our results, expendure on environmental services increases w income. Tis is true wen all oter factors remain constant. However, wen canges occur in preferences, prices, environmental attributes, and te production structure for outdoor recreational experiences, becomes difficult to predict te demand for environmental services in te future. Tis is a question of interest for future researc. Qualy canges are well-known causes of data misinterpretations, e.g., te difficulties in disentangling te relation between canges in qualy and price. Furter, te demand function is also a function of relative prices. If outdoor recreation becomes ceaper to produce, ten, all oter tings being equal, we would expect more ouseolds to consume. If prices, preferences, and mean income cange at te same time, interpretation becomes difficult. As is, tis study relegates price effects to a constant term. 18

21 References Banks, J., Blundell, R. and Lewbel, A. (1997) Quadratic Engel curves and consumer demand. Te Review of Economics and Statistics, 79. Batemen, I.J. and Willis, K.G. (1999) Valuing Environmental Preferences:Teory and Practice of te Contingent Valuation Metod in te US, EU and Developing Countries. Oxford Universy Press, Oxford, UK. Blundell, R., Pasardes, P. and Weber, G. (1993), Wat Do we Learn About Consumer Demand Patterns from Micro Data? Te American Economic Review, 83. Calfant, J.A. (1987) A Globally Flexible, Almost Ideal Demand System. Journal of Business and Economic Statistics 5. Cicilinisky, G. (1998) Te Knowledge Revolution. Journal of International Trade and Economic Development, 7. Costa, D. (1997) Less of a luxury: Te Rise of Recreation since National Bureau of Economics Researc, Working Paper, Deaton, A. (1986) Demand Analysis, in Handbook of Econometrics, eds. Z. Grilices and M.D. Intriligator (Nort-Holland, Amsterdam). Deaton, A. and Muellbauer, J. (1980) An Almost Ideal Demand System. Te American Economic Review, 70. Flores, N. and Carson, R. (1997) Te Relationsip between te Income Elasticies of Demand and Willingness to Pay. Journal of environmental Economics and management, 33. Freeman, A.M. (2003) Te Measurement of Environmental and Resources Value. Teory and Metod, Wasington, DC: Resources for te Future. Greene, W.H. (2000) Econometric Analysis, Fourt Edion, Prentice Hall, Englewood Cliffs, NJ. Groot, H.L.F. (1999) Structural Cange, Economic Growt and te Environmental Kuznets Curve. A Teoretical Perspective. OCFEB Researc Memorandum 9911, Working Paper Series 1, Vrije Universe, Amsterdam. Grossman, G.M. and Krueger, A.B. (1995) Economic Growt and te Environment. Te Quarterly Journal of Economics, 110. Hilton, F.G. and Levinson, A. (1998) Factoring te Environmental Kuzents Curve: Evidence from Automotive Lead Emissions. Journal of Environmental Economics and Management,

22 Hökby, S. and Söderqvist, T. (2001) Elasticies of Demand and Willingness to pay for Environmental services in Sweden. Beijer Discussion paper series No. 137, Beijer International Instute of Ecological Economics, Te Royal Swedis academy of sciences, Stockolm. Kanninen, B.J. and Kriström, B. (1992) Welfare Benef Estimation and Income Distribution. Beijer Discussion paper series No. 20, Beijer International Instute of Ecological Economics, Te Royal Swedis academy of sciences, Stockolm. Kriström, B. and Riera, P. (1996) Is te Income Elasticies of Environmental Improvements Less tan One?. Environmental and Resources Economics, 7. Larsen, R.E. (2001) Revealing Demand for Nature experiences Using Purcase Data on Equipment and Lodging. Discussion paper No.305, Statistic Norway, Researc Department. Lopez, R. (1994) Te Environment as a Factor of Production. Te Effects of Economic Growt and Trade Liberalization. Journal of Environmental Economics and Management, 27. Mäler, K.G. (1974) Environmental economics: A Teoretical Inquiry. Baltimore: Jons Hopkins Universy Press. McConnell, K. (1997) Income and te Demand for Environmental Qualy. Environment and Development Economics, 2. Miles, D., Pereyra, A. and Rossi, M. (2002) Te Consistent estimation of Income Elasticy of Environmental Amenies in Uruguay. Studios Economics, 17. Mcell, R.C. and Carson, R.T. (1989) Using Survey to value Public Goods: Te Contingent Valuation Metod. Resources for te Future, Wasington, D.C. Pereyra, A. and Rossi, M. (1998) Are Environmental Goods a Luxury?, Working paper 13, Universy of Uruguay. Pollak, R. and Wales, T. (1992) Demand System Specification and Estimation. Oxford Universy press, Oxford, UK. Segal, J.M. (2001) Wat We Work for Now, New York Times, September 3. Selden, T. and Song, G. (1995) Neoclassical Growt, te J Curve for Abatement, and te Inverted U for Pollution. Journal of Environmental Economics and Management, 29. Stern, D.I. and Common, M.S. (2001) Is tere an environmental Kuznets curve for sulfur?, Journal of Environmental Economics and Management,

23 Tobin, J. (1958) Estimation of Relationsips for Limed Dependent Variables, Econometrica,

24 Appendix A Table A1: Test for an income elasticy different from 1 (t-test) MAIN GROUPS Foodstuff * * * * Energy goods * * 39.00* * Transport 2.44* * Outdoor recreation 9.75* 5.80* 5.87* 3.88* FOODSTUFFS Food * * -9.00* -8.00* Beverages * 17.75* 8.80* ENERGY GOODS Electricy * * * Oter energy goods -1.88* * TRANSPORT Petrol -5.00* * * Car maintenance 5.33* 9.50* 5.12* Public and oter transp * -0.75* * OUTDOOR RECREATION Equipment for sporting, 23.00* 5.00* 26.00* fising, and camping Oter recreation goods * -4.20* -5.00* Note: * denotes significance at te 5% percent level. A negative t-value indicates tat te income elasticy is less tan one, and a posive value tat te income elasticy is larger tan one. 22

25 Table A2: Estimated income elasticies in Table A3 : Estimated income elasticies in Foodstuffs Standard Energy Goods Standard Transport Standard Outdoor Recreation Standard Number of cildren C (0.01) 0.55 (0.03) 1.12 (0.08) 1.69 (0.09) C (0.01) 0.55 (0.03) 1.07 (0.08) 1.65 (0.09) C (0.01) 0.55 (0.03) 1.17 (0.08) 1.83 (0.09) Region R (0.01) 0.55 (0.03) 1.07 (0.08) 1.64 (0.09) R (0.01) 0.54 (0.03) 1.06 (0.08) 1.64 (0.09) R (0.01) 0.56 (0.03) 1.06 (0.08) 1.57 (0.09) R (0.01) 0.56 (0.03) 1.07 (0.08) 1.62 (0.08) R (0.01) 0.56 (0.03) 1.07 (0.08) 1.64 (0.09) R (0.01) 0.56 (0.03) 1.04 (0.08) 1.63 (0.09) R (0.01) 0.55 (0.03) 1.06 (0.12) 1.66 (0.09) Notes: C1= 1 cild below 18, C2= 2 cildren below 18, C3= more tan 2 cildren below 18, R1=Uppsala, R2= Eskilstuna, R3= Jönköping, R4=Malmö, R5= Hälsingborg, R6= Gotenburg, R7=Västerås. Standard errors win parenteses. Foodstuffs Standard Energy Goods Standard Transport Standard Outdoor Recreation Standard Number of cildren C (0.01) 0.32 (0.03) 0.92 (0.03) 1.89 (0.13) C (0.01) 0.35 (0.03) 0.93 (0.03) 1.85 (0.14) C (0.01) 0.33 (0.03) 0.94 (0.03) 1.87 (0.13) Region R (0.01) 0.35 (0.03) 0.92 (0.03) 1.89 (0.13) R (0.01) 0.35 (0.03) 0.92 (0.03) 1.88 (0.13) R (0.01) 0.36 (0.03) 0.92 (0.03) 1.90 (0.13) R (0.01) 0.34 (0.03) 0.92 (0.03) 1.89 (0.13) R (0.01) 0.36 (0.03) 0.93 (0.03) 1.90 (0.13) R (0.01) 0.35 (0.03) 0.93 (0.03) 1.90 (0.13) Notes: C1= 1 cild below 18, C2= 2 cildren below 18, C3= more tan 2 cildren below 18, R1 = Stockolm, R2= Gotenburg/Malmö, R3= major towns, R4=soutern areas, R5= major towns nortern areas, R6= nortern areas. Standard errors win parenteses. 23

26 Table A4 : Estimated income elasticies in Table A5 : Estimated income elasticies in Foodstuffs Standard Energy Goods Standard Transport Standard Outdoor Recreation Standard Number of cildren C (0.01) 0.16 (0.03) 0.91 (0.03) 1.86 (0.13) C (0.01) 0.13 (0.03) 0.90 (0.03) 1.86 (0.14) C (0.01) 0.16 (0.03) 0.91 (0.03) 1.85 (0.14) Region R (0.01) 0.19 (0.03) 0.89 (0.03) 1.83 (0.13) R (0.01) 0.19 (0.03) 0.89 (0.03) 1.83 (0.13) R (0.01) 0.18 (0.03) 0.89 (0.03) 1.83 (0.13) R (0.01) 0.18 (0.03) 0.89 (0.02) 1.83 (0.13) R (0.01) 0.19 (0.03) 0.90 (0.03) 1.84 (0.13) R (0.01) 0.18 (0.03) 0.90 (0.03) 1.83 (0.13) Notes: C1= 1 cild below 18, C2= 2 cildren below 18, C3= more tan 2 cildren below 18, R1 = Stockolm, R2= Gotenburg/Malmö, R3= major towns, R4=soutern areas, R5= major towns nortern areas, R6= nortern areas. Standard errors win parenteses. Foodstuffs Standard Energy Goods Standard Transport Standard Outdoor Recreation Standard Number of cildren C (0.03) 0.32 (0.03) 1.11 (0.06) 1.93 (0.24) C (0.03) 0.32 (0.04) 1.13 (0.06) 1.98 (0.25) C (0.03) 0.33 (0.04) 1.13 (0.06) 1.97 (0.26) Region R (0.03) 0.33 (0.04) 1.10 (0.06) 1.93 (0.24) R (0.02) 0.33 (0.04) 1.11 (0.06) 1.94 (0.24) R (0.04) 0.32 (0.03) 1.11 (0.06) 1.94 (0.24) R (0.04) 0.32 (0.03) 1.11 (0.06) 1.92 (024) R (0.03) 0.32 (0.03) 1.10 (0.06) 1.93 (0.24) R (0.03) 0.32 (0.03) 1.11 (0.06) 1.95 (0.24) Notes: C1= 1 cild below 18, C2= 2 cildren below 18, C3= more tan 2 cildren below 18, R1 = Stockolm, R2= Gotenburg/Malmö, R3= major towns, R4=soutern areas, R5= major towns nortern areas, R6= nortern areas. Standard errors win parenteses 24