ACEI working paper series LOCAL GOVERNMENT ALLOCATION OF CULTURAL SERVICES

Transcription

1 ACEI workng paper seres LOCAL GOVERNMENT ALLOCATION OF CULTURAL SERVICES Lars Håkonsen Knut Løyland AWP Date: October 2013

2 Paper presented at the 6 th EWACE Workshop n Ljubljana, 5-7 September 2013 Local government allocaton of cultural servces by Lars Håkonsen, Telemark Research Insttute and Telemark College, Norway * and Knut Løyland, Telemark Research Insttute, Norway ** Keywords: Allocaton of cultural servces; Local government budget; Demand system JEL-codes: Z11; Z18; H72 Abstract In the present paper we analyse the allocaton process of cultural servces n Norwegan muncpaltes. The cultural sector on ths admnstratve level s decomposed nto the followng eght subcategores: chldren and youth actvtes, lbrares, cnemas, museums, arts dssemnaton, sports, cultural schools, and other cultural servces. By means of budget shares for these eght cultural servces and a resdual sector consstng of all other muncpal servces, we estmate a system of demand relatons whch are nterdependently lnked to each other by a budget restrcton. Our analyses are based on data from 406 out of 429 Norwegan muncpaltes durng the perod 2002 to In the emprcal analyses we manly focus on the effects of ncome varaton for the cultural servces. We estmate effects of free ncome, matchng grants to each sector, and user fees and other sector-specfc ncome for each sector. We also estmate crowdng-out effects for the cultural sectors of demographc varables ndcatng hgher demand for servces lke educaton, chld care, and health servces. Our results confrm prevous results. There are nterestng dfferences wthn the group of cultural servces, and these are partly related to dfferent natonal standardzaton and regulaton among the cultural servces. In the concludng secton we dscuss some cultural polcy mplcatons of the results obtaned. * e-mal: lars.hakonsen@ht.no ** e-mal: loyland@tmforsk.no Vdar Rngstad, Lus César Herrero and partcpants at the IV Workshop on Cultural Economcs and Management n Blbao, 2012, are gratefully acknowledged for comments and suggestons for mprovements on an earler verson of ths paper. Any remanng errors are our own.

3 1. Introducton In most countres the provson of cultural servces s dvded by the publc and the prvate sector. In addton, the publc provson s dvded between dfferent admnstratve levels central, county level and local muncpaltes. The emprcal lterature on local or regonal government provson of cultural servces s somewhat lmted. However, there exst several nterestng studes of varous aspects of local cultural polcy and local cultural spendng, c.f. Depalo and Fdel (2011) and Bento et al (2013) for two recent examples. Most studes of local cultural polces analyse local cultural expendtures as a whole,.e., an aggregate of all cultural expendtures n the muncpalty. Ths apples to the two studes mentoned above, as well as for Werck et al (2008), and Stastna (2009). In addton, some studes analyse a specfc local or regonal cultural nsttuton lke an opera house, cf. Schultze and Ursprung (2000) or a theatre, cf. Getzner (2004). The contrbuton of the present study s that local cultural expenses are dsaggregated nto eght dfferent cultural purposes lke e.g. cnemas, lbrares, museums or cultural schools. In addton to these eght cultural servces, we aggregate all other muncpal servces nto a resdual spendng category. By means of budget shares for the eght cultural servces and the aggregate of all other muncpal servces, we estmate a demand system of relatons whch are nterdependently lnked to each other by a budget restrcton. Thus we analyse how local governments allocate recourses to the nne servces n queston gven ther needs, preferences, ncome and other factors of nterest. The set of control varables accountng for needs, preferences and poltcal factors are closely related to the control varables used n Bento et al (2013) and Werck et al (2008). Our analyses are based on data for 406 Norwegan muncpaltes whch are qute homogenous n ther purposes, but very heterogeneous n other respects. For example, the rchest 10 % of the muncpaltes had n 2011 a per capta ncome of about 2.3 tmes hgher than the 10 % poorest. Such large ncome dfferences wll of course strongly affect the provson of local servces n general. It s well establshed n the cultural economcs lterature that most cultural goods and servces are luxury goods wth an ncome elastcty well above unty, see for nstance Rngstad and Løyland (2012). The general concepton of cultural servces as ncome elastc goods s also confrmed n several prevous emprcal studes of Norwegan muncpaltes, cf. Rattsø (1989), Borge and Rattsø (1995), Aaberge and Langørgen (2003), and Håkonsen and Løyland (2011). All these studes estmate ncome elastctes for varous muncpal servces, among whch an aggregated cultural sector represents one of these sectors. The cultural sector belongs to the most ncome elastc servce sectors n all these studes. Aaberge and Langørgen fnd a relatvely suppressed structure for the ncome elastctes,.e., levels close to 1.0 for most sectors, and the ncome elastcty for culture s 1,09. Although ths s only modestly elastc, the cultural sector nevertheless s the second most elastc among the eght sectors ncluded n ther study (the only sector beng more elastc s nfrastructure). Borge and Rattsø also dentfy culture as an ncome elastc sector, wth short- and long run ncome elastctes of 1.32 and 1.23 respectvely. In a more recent analyss, Håkonsen and Løyland (2011) (n Norwegan) fnd an ncome elastcty for culture as hgh as There are two man reasons why culture servces typcally belong to the most ncome elastc servce sectors. The frst s that the character of the cultural servces themselves suggests that these servces are less necessary than servces that deal wth more mmedate and fundamental needs lke e.g. health ssues or prmary educaton. Ths explanaton relates to the genune ncome elastcty,.e., the preference structure per se mples that cultural servces are less strctly necessary than e.g. schools or health care. The second reason s that many other local 2

4 servces are strongly regulated by the central government. These regulatons nvolve both quantty and qualty, and are partcularly strong for sectors lke chld care and prmary educaton. Also wthn care for the elderly, there has gradually been ntroduced more and more regulatons gvng qute strong ndvdual rghts. Such rghts and regulatons restrct the room for manoeuvre for the muncpaltes, and enforce the muncpaltes to have a more standardsed supply of chld care, educaton and care for the elderly than would otherwse be the case. Ths naturally leads to relatvely low ncome elastctes for the most regulated sectors. Ths s confrmed n the results found n Håkonsen and Løyland (2011), where the lowest ncome elastctes are found for kndergartens and prmary educaton at 0.6, followed by health care at 0.8. However, sectors lke cultural servces, nfrastructure and general admnstraton are to a far lesser extent subject to restrctons and regulatons from the central government. These sectors are therefore typcally more ncome elastc, wth elastctes well above 1.0, and as mentoned an elastcty as hgh as 2.04 for the cultural sector. Stronger regulatons and more standardsaton of some servces naturally lead to more varaton wthn the least regulated sectors. Ths follows drectly from the budget constrant and the derved result that the budget share-weghted sum of ncome elastctes equals one. An nterestng contrbuton from ths paper, s to analyse the structure of ncome elastctes wthn the group of locally provded cultural servces: Are all local cultural servces luxures, or do we fnd both necesstes and luxures among the cultural servces? And are all local cultural servces equally free from central government regulatons? These are some of the man questons we wll address n the present paper. Wth ths background, we wll n secton two take a closer look at the Norwegan local government system n order to gve the reader an dea of the dfferent admnstratve levels and how and by whom publc goods and servces are provded n Norway. In secton three we descrbe the theoretcal and econometrc approach. Secton four descrbes the data and secton fve presents the results of the analyss. Fnally, n secton sx, we dscuss the results and conclude. 2. The Norwegan settng Norwegan muncpaltes are organsed as mult-purpose provders of core goods of the Norwegan welfare state. Thus they are, ndependent of sze and locaton, expected to provde ther nhabtants wth a numbers of goods and servces lke care for the elderly, prmary educaton, chld care, prmary health care, socal benefts and nfrastructure lke roads, water and sewage systems. In addton, the muncpaltes shall provde cultural servces. Even f all muncpaltes are expected to provde an dentcal set of the core goods of the Norwegan welfare state, they are strongly heterogeneous many other respects. Whle the smallest have less than 1000 nhabtants the largest have more than The large dfferences n congeston between for nstance northern and southern parts of the country also make dfferences n the need for travel tme, n order to provde nhabtants wth dfferent servces and nfrastructural nvestments. The cultural servces provded by the muncpaltes can be dvded nto eght dfferent servces. Ths s the maxmum dsaggregaton avalable from the natonal nformaton system for muncpal servces ( Kostra ) operated by Statstcs Norway. These cultural subcategores are: provson of cultural related actvtes for chldren and youth, lbrares, cnemas, museums, arts dssemnaton, sports, cultural schools, and other cultural servces. In addton, muncpaltes are partly responsble for ssues relatng to cultural hertage, but the share of expendtures related to cultural hertage s on average very margnal compared to other cultural servces, and t s also mxed up wth servces related to nfrastructure. Thus we have omtted cultural hertage from the analyss. 3

5 In Table 1 we present some fgures of the local government s provson of cultural servces. Budget shares, number of muncpaltes provdng the servces n queston and how the fundng of the servces are dstrbuted between grants from the central government and the countes, user fees, and own fundng by the local government, are presented. The table shows that cultural servces make a rather small share of the total budget. Except for sports, none of the eght servces consttutes on average more than 1 % of the total budget, and the total cultural budget represents less than four per cent of the total budget. Only three out of eght servces have faced an ncrease n the budget share n the perod analysed. In the same perod the muncpaltes have faced a real ncrease n (free) ncome of about 4 per cent. Table 1. Budget shares, number of muncpaltes wth zero-budgets and the fundng of local government provson of cultural servces (nhabtant weghted averages) and N=406. Chldren and youth actvtes Budget shares (per cent) Number of muncpaltes wth no expenses on cultural servces Central and county government grants Sources of fundng (per cent) User fees and box offce ncome Local government fundng Lbrares Cnemas Museums Arts dssemnaton Sports Cultural schools Other cultural Other servces Total culture The fact that the expendture share of the aggregated cultural goods falls from 4 % n 2002 to 3.86 % n 2010 at frst sght seems to contradct the relatvely hgh ncome elastctes found n the emprcal studes cted n the ntroducton. As wll be shown from the results n secton 5, however, one cannot nfer nformaton about the genune ncome elastctes of cultural servces only by observng the development of the expendture share over a certan perod of tme. From Table 1 we also notce that some of the cultural servces are not provded at all by many muncpaltes. Especally, ths goes for cnema, arts dssemnaton and museums. About 50 per cent of the muncpaltes have no provson of cnema. But ths s only partly due to the lack of cnemas n the muncpaltes n queston. The reason s also due to a complete prvatzaton of cnemas n some muncpaltes. Arts dssemnaton ncludes dssemnaton of the performng arts as well as vsual arts and some nfrastructural nvestments necessary to provde such servces. Smaller muncpaltes are not n the poston of explotng the economes of scale due to bg fxed costs related to such nvestments, and are thus less lkely to provde arts dssemnaton 4

6 servces. Ths explanaton apples to museums as well. There are also a number of small muncpaltes that are located close to larger towns and ctes, and thereby may choose to free rde on the supply of such cultural servces n ther larger neghbourng muncpaltes. Table 1 also ncludes nformaton about the fundng of the cultural servces. As can be seen from the table cnemas are manly funded by user fees, but the share of user fees have decreased from 2002 to 2010, probably due to a declne n attendance. There has also been a decrease n the share of user fees for museums from about 10 per cent n 2002 to less than 4 per cent n 2010, also most lkely due to a declne n attendance n combnaton wth fallng tcket prces n real terms. For museums we also notce a large decrease n central and county government grants. Ths s most lkely due to a consoldaton of the owner structure of museums durng the tme perod analysed. Many museums have merged nto bgger unts and been organsed n separate companes. They are thus not under drect muncpal control and receve ther central and county government grants drectly. A decrease n grants also appears n the cultural school sector. Ths s due to a substtuton of earmarked fundng by general grants durng the studed perod. On the other hand, arts dssemnaton has faced an ncrease n the share of central and county government grants. Ths s manly because a central government program amng at ncreasng arts dssemnaton n contemporary schools. The program s denoted The Cultural Rucksack, and was ntroduced n from 2001, and then gradually expanded n sze and scope. We may also note an ncrease n the central and county government grant share of the fundng of the aggregate of all other servces. Ths s manly due to central government ncreased ambtons n reachng full coverage of chld care servces for chldren n the age from 1 to 5 years. Untl 2011 ths polcy program was funded by steadly ncreasng matchng grants each year. 3. Theoretcal framework and emprcal model specfcaton The prmary goal of ths paper s to study how local governments allocate resources to the range of cultural servces or sectors outlned n the prevous secton. The natural approach s then to model each muncpalty as a utlty maxmzng agent on behalf of the nhabtants. Ths approach s qute standard n the lterature on local government behavour, and rests on the tradton from medan voter models, cf. Inman (1979) and Rubnfeld (1987). A key dfference between the lterature based on Amercan local government decsons and our Norwegan settng, s that Norwegan muncpaltes have very lttle local freedom wth respect to tax rates and tax bases. The man decson problem for a Norwegan muncpalty s therefore to choose the budget shares to be allocated to the varous servce sectors wthn a (more or less) gven total budget. The approach n ths paper dffers from the Norwegan emprcal studes cted n secton 1, n at least two mportant respects. Frst of all, we adopt a talor-made decomposton of muncpal servces for studyng cultural goods, wth maxmum detal (eght sectors) for cultural servces, whle we aggregate all other muncpal servces nto one resdual sector. Furthermore, we do not nclude prce data (unt costs) n our model. Ths omsson s done for several reasons. Frst of all, there s a well known problem of dentfyng unt costs of adequate qualty for muncpal servces n general. Ths problem arses prmarly because of the nature of most muncpal servces, whch typcally are rather mult-faceted servces for whch there lacks good outputmeasures. The soluton to ths problem s typcally to replace unt costs wth proxy varables based on wage data, the so-called publc employment approach, cf. Ehrenberg (1973) and Bahl et al (1980). Another approach s taken by Aaberge and Langørgen (2003), who nfer nfor- 5

7 maton about unt costs from varaton n local cost factors n a modfed extended lnear expendture system (ELES). The problem of dentfyng relevant unt cost data s exacerbated further wth our hghly dsaggregated cultural servces. Furthermore, we are prmarly nterested n the effects of general ncome, matchng grants and fee ncomes on the varous culture servces. The effects of removng prce proxes from the demand system have been tested n the eght-sector model adopted n Håkonsen and Løyland (2011). The results (elastctes) wth respect to general ncome and sector-specfc matchng grants and fee ncomes are only margnally dfferent n the models wth and wthout prce proxes. Instead of havng nomnal prce or unt cost-proxes n the model, we deflate all economc data (ncome varables) to 2010-prces by a specal cost ndex for Norwegan muncpal servces mantaned by the Mnstry of Local Government and Regonal Development and the Mnstry of Fnance. 1 Our emprcal demand system thus conssts of the budget shares S for nne servce sectors, ndexed = 1,..,9, where sectors 1 to 8 are the eght cultural servces and sector 9 s the resdual sector comprsng all other servces, cf. Table 1 above. The demand system s qute standard, except for prces beng removed and total ncome Y splt nto general ncome I (tax ncome and general purpose grant from the central government), sector-specfc matchng grants G from the central government and the 18 countes, and user fees and other sales ncome for each sector UF,.e., Y = I + ΣG + ΣUF. 2 The dependent varables are the expendture shares of each servce sector, S. In addton to general ncome, matchng grants and fee ncomes, we nclude a vector of altogether 26 control varables Zk. Ths vector comprses muncpalty-specfc varables that potentally reflect the preferences, demand, or cost level of muncpal servces, and nclude soco-demographc characterstcs, poltcal varables and geographc characterstcs. The lst of control varables are descrbed n more detal n the next secton. In addton we nclude a trend varable, τ, runnng from 0 to 8 for the years 2002 to The stochastc error term s u and s assumed to be ndependently and dentcally, normally dstrbuted. Snce many muncpaltes have zero expendture on some of the cultural servces, cf. Table 1, we use a Tobt model to avod potental bases due to ths partcular feature of the data. Snce observatons wth zero expendture are due to a corner soluton for the muncpaltes n queston, and not data censorng as such, we follow the corner soluton nterpretaton of the Tobt model, cf. Wooldrdge (2003). Thus the demand system becomes the followng: S max I jg j jufj k Zk u,0 where 1,...,9 j1 j1 k1 (1) The budget constrant of course mples that the sum of gross expendtures for each sector must equal total ncome Y, or alternatvely that ΣS = 1. 3 The followng set of restrctons follows from the budget constrant: 1 The cost ndex for muncpaltes s ncreasng consderably faster than the consumer prce ndex, snce the man cost component for the muncpaltes s wage cost and there s rather lmted room for productvty mprovements for muncpal servces, cf. Baumol (1967). 2 The total number of countes n Norway s 19 ncludng Oslo, but Oslo has been omtted from the dataset due to some mssng varables, thereby reducng the number of countes n the dataset to For smplcty we have assumed that the local governments cannot make any ntertemporal adjustments. Thus savngs are not allowed wthn ths model. 6

8 , 0, 0, 0, µ 0, 0, (2) j j k j 1,...,9, k 1,...,25. The matchng grants Gj and the fee ncomes UFj represent a potental endogenety problem, snce the levels of Gj and UFj n part are the results of prortes made by the muncpaltes themselves. Thus there mght be a problem wth reversed causalty n the model whch n turn mples a postve bas n the estmated s and s. In order to elmnate the reversed relatonshp, we use a smple nstrumental varable method by ntroducng a one year lag n the G s and UF s. At least the drect (reversed) relatonshp between the local government prortes n the budget and the level of the G s and UF s s reduced by ths procedure. Some muncpaltes have no expendtures on sx out of nne purposes. Thus the dependent varable s equal to S for muncpaltes wth strctly postve expendtures and 0 for the corner soluton outcome. Followng Wooldrdge (2003), when employng the corner soluton model, we are both nterested n the condtonal probablty of havng strctly postve expendtures, 0 E S x, S 0. 4 If u s ndependent PS x, and the condtonal expected budget share, of x and u follows a normal dstrbuton, we can fnd and explct expresson for can be wrtten as: where ES x βx ES x PS 0 x E S x, S 0 βx (3). Ths [(0 βx) / ] ( βx) / ) 1 [(0 β x) / ] ( β x) / ) (4) ( ) and ( ) are the cumulatve and densty dstrbutons, respectvely, for the standard normal dstrbuton functon and s the standard error. Ths expresson s often denoted the nverse Mlls rato, see Wooldrdge (2003) for more detals. The margnal effect of the corner soluton model s gven by: ES [ x] βx j j 1,..., K, (5) x j where K s the number of muncpaltes. McDonald and Mofft (1980) decompose (5) as follows: E[ S β ] E[ S, S 0] P[ S 0] P[ S 0] E[ S, S 0] x x x (6) x x x j j j Ths means that a change n the ndependent varables have two effects: () t affects the condtonal mean of S for S > 0 and, () t affects the probablty that an observaton wll fall n ths 4 To avod notatonal clutter we nclude the I, G, UF, Z and τ varables n the vector x n the rest of ths secton. The correspondng vector of parameters for equaton s denoted β. 7

9 part of the dstrbuton,.e. where S > 0. In our applcaton of the model where muncpaltes chose quanttes of cultural servces, the dervatve n (6) accounts for the fact that some muncpaltes that have chosen the corner soluton at S=0, may swtch to S>0 when xj changes. In order to calculate the dfferent ncome elastctes from the ncome elements of (1) we need the slope coeffcents or the margnal effects as gven by (5). Instead of calculatng by predctng the average probablty of havng strctly postve budget share from the output of the maxmum lkelhood procedure used to estmate the Tobt coeffcents, we use an approxmaton whch smply s the observed proporton of those muncpaltes wth a strctly postve budget share, see for nstance Greene (2003 p. 766). 4. The data The data used n the present analyss s based on two man sources. Frst, we use Kostra, whch s a natonal nformaton system for all 429 Norwegan muncpaltes plus the 19 countes. Kostra provdes a wde varety of nformaton talored to carry out economc analyses of the muncpaltes, for nstance nformaton on economc, socal, demographc, and poltcal factors. The data system s mantaned and admnstered by Statstcs Norway. Second, we also use crtera data from the cost redstrbuton element of the General Purpose Grant Scheme (GPGS) whch, n addton to taxes, are the man fundng system of Norwegan muncpaltes. The crtera from GPGS used n the present analyss s mantaned and admnstered by The Mnstry of Local Government and Regonal Development. The data have a panel structure that runs over the perod 2002 to It comprses the same 406 muncpaltes n all years and we thus have a balanced panel data set consstng of 3654 observatons. In the analyses we dvde the muncpal budget as shown n Table 1. Thus the cultural sector s dvded nto eght dfferent purposes whch all together represent about four per cent of the total budget. The remanng local government servces are aggregated nto one sngle purpose representng about 96 per cent of the budget. The calculaton of the budget shares for the nne purposes s based on the gross operatng expenses. In addton, for each of the nne sectors, we collect separate nformaton about matchng grants from central and county government as well as any grants from other muncpaltes. User fees for varous purposes, ncludng box offce ncome, are also added. Gross operatng expenses mnus user fees and matchng grants are defned as net operatng expenses. In the followng we present summary statstcs of the data used n the analyss. Requrng a balanced data set mples that muncpaltes wth mssng data n at least one year wll be omtted for the whole perod. Among other muncpaltes, the largest ctes Oslo, Bergen and Trondhem have been omtted for ths reason. We have dvded the presentaton of the varables n two tables: () Budget shares, matchng grants, user fees and general ncome, and () other explanatory varables. Table 2 shows the nne purposes' budget shares, as well as matchng grants, user fees and muncpal free ncome. As mentoned n secton 2, culture consttutes a very small proporton of the total muncpal budget. In contrast to Table 1 the averages n Table 2 are not weghted by the number of nhabtants. In Table 2 we also present matchng grants, user fees and free ncome per capta measured n NOK deflated to 2010-prces. 5 As can be seen from the table the grants and user fees per capta 5 NOK 1000 = EURO 123,7 n

10 are qute small as compared to general ncomes measured by the free ncome varable and matchng grants and user fees for other servces, reflectng the small proporton the cultural servces makes of the total budget. Table 2. Summary statstcs for budget shares and ncome varables. N=3654. All ncome terms n 2010-prces. Varable Label Mean Standard Mn Max devaton Budget share, actvtes for chldren and youth (%) S Budget share, lbrares (%) S Budget share, cnemas (%) S Budget share, museums (%) S Budget share, arts dssemnaton (%) S Budget share, sports (%) S Budget share, cultural schools (%) S Budget share, other cultural servces (%) S Budget share, other welfare servces (%) S Grants per capta, actvtes for chldren and youth (%) G Grants per capta, lbrares (NOK) G Grants per capta, cnemas (NOK) G Grants per capta, museums (NOK) G Grants per capta, arts dssemnaton (NOK) G Grants per capta, sports (NOK) G Grants per capta, cultural schools (NOK) G Grants per capta, other cultural servces (NOK) G Grants per capta, other welfare servces (NOK) G User fees per capta, actvtes for chldren and youth (NOK) UF User fees per capta, lbrares (NOK) UF User fees per capta, cnemas (NOK) UF User fees per capta, museums (NOK) UF User fees per capta, arts dssemnaton (NOK) UF User fees per capta, sports (NOK) UF User fees per capta, cultural schools (NOK) UF User fees per capta, other cultural servces (NOK) UF User fees per capta, other welfare servces (NOK) UF Muncpal free ncome per capta (NOK) I In Table 3 we present descrptve statstcs for the other varables used n the analyses. We dstngush between polcy varables, varables of socal condtons, populaton, personal ncome, local cost crtera and educatonal background of the nhabtants. We wll return to hypotheses concernng the dfferent varables below. Here we lmt ourselves to a summary descrpton of the varables we thnk deserves some extra attenton. The frst poltcal varable s the share of representatves of lberal/conservatve partes n the muncpal councl (R). Lberal/conservatve partes are defned from the poltcal centre and towards the rght (manly Lberals, Chrstan Democrats, Conservatves, etc). Moreover, we also have the share of representatves on ndependent lsts (F) and the share of representatves of left wng partes (L). The left wng partes nclude representatves from The Labor Party and 9

11 further left,.e. socalst and communsts. Share of representatves from lberal/conservatve partes s the reference category n the calculatons below. Fnally, we have calculated a Herfndahl ndex of party concentraton. It s defned as follows: Herf x x 2 where x s the number of representatves from party. Thus we have 0 < Herf 1. Wth equal numbers of representatves from all poltcal partes, the ndex wll lmt zero and we have a fragmented and less quorum councl. The opposte s the case when one of the partes has all the representatves. When the ndex lmts unty we have a strong party concentraton and thus a more quorum councl. From Table 3 we see that the average Norwegan muncpal s qute fragmented wth a Herfndahl ndex of The varables accountng for socal factors nclude share of non-marred, mmgrants, unemployed, dvorced and share of mentally retarded nhabtants both above and below 16 years old. In addton, muncpal mortalty rate s ncluded. The reason why we nclude nformaton on such matters s that these varables are supposed to affect the need for muncpal provded core welfare servces of varous knds. On average, these shares are very low, but f we look at the maxmum values we fnd that some are very much hgher than n the average muncpalty. If a muncpalty s facng hgh scores on these varables, they are also subject to latent costs that mght crowd out servces of less mportance, for nstance cultural servces. Populaton varables are dvded by age groups because dfferent age groups are supposed to have dfferent needs regardng welfare servces. For example, the proporton n the age group 0-5 year olds express needs for chld care servces, the sze of the age group of 6-15 year olds express requrements for contemporary school servces, whle the proporton aged 80 years and older wll express the need for elderly care servces. Regardng the latter, these are very resource-ntensve servces and from Table 3 we can see that there are lkely to be bg dfferences between muncpaltes regardng care for elderly: Whle the average share of nhabtants aged 80 years and older s about 5% for Norway, some muncpaltes have a share of ths age group between 10 and 15%. In the calculatons below, nhabtants years s the reference category. In Table 3 we also present fgures for average and medan prvate ncome n the muncpaltes. Earnngs are calculated for nhabtants 17 years and older. Medan ncome s the ncome of the person n the mddle of the dstrbuton (wthn a muncpalty) when the populaton s sorted from lowest to hghest ncome. We have a symmetrc ncome dstrbuton when medan ncome equals average ncome. Because some nhabtants have very hgh ncome and thereby pullng the average ncome upwards, the average and medan ncome wll dffer. We have, n other words, a skewed ncome dstrbuton. Medan ncome would n such cases be more representatve because t represents the ncome of a "typcal" person n the populaton. To take care of potental effects of these ncome condtons, we ntroduce a rough measure of ncome nequalty, more precsely the relatonshp between average and medan ncome (varable name AINC/MINC ). If ths relatonshp equals unty we have a symmetrcal dstrbuton of ncome. Is the rato greater than 1, we have an upward skewed ncome dstrbuton and the opposte f t 10

12 s less than unty. From Table 3 we see that the relatonshp between average and medan ncome s 1.13 for Norwegan muncpaltes on average. Ths means that n Norway we have a moderately upward skewed ncome dstrbuton. Table 3. Summary statstc for other factors. N=3654 Varable Label Mean Standard Mn Max devaton Share of representatves from rght wng partes R Share of representatves wth no party connecton F Share of representatves from left wng partes L Party fragmentaton (Herfndahl s ndex) HERF Non-marred nhabtants per capta NM Dvorced per capta DIV Immgrants per capta IMM Unemployed per capta U Mortalty rate MORT Mentally retarded above 16 years old per capta PU_16a Mentally retarded less than 16 years old per capta PU_16u Inhabtants, 80 years and older per capta POP Inhabtants, years per capta POP67_ Inhabtants, 6-15 years per capta POP6_ Inhabtants, 0-5 years per capta POP0_ Number of nhabtants POP 8,980 13, ,850 Average ncome for nhabtants 17 years 734 AINC and older (NOK) (Deflated by CPI) 977 Medan ncome for nhabtants 17 years 377 MINC and older (NOK) (Deflated by CPI) 506 Measure of dsparty (AINC/MINC) ULIK Average travel tme to closest school dstrct zone (mnutes) Average travel tme to muncpalty centre (mnutes) Share of nhabtants n areas wth hgh populaton densty Share of nhabtants years old wth prmary school educaton Share of nhabtants years old wth secondary school educaton Share of nhabtants years old wth degree from college and/or unversty (1-3 years) Share of nhabtants years old wth degree from college and/or unversty (more than 3 years) Share of nhabtants years old wth unknown educaton ZONE TRAVEL TETT EDU EDU EDU EDU EDU Dfferent crtera are used n GPGS n order to capture dfferences n urban structure between muncpaltes and n the muncpal revenue system to capture addtonal costs due to the settlement pattern. Here we use two such crtera: Travel Dstance to the muncpal centre and the 11

13 zone crteron. The travellng crteron s a measure of average travel tme of nhabtants travellng from the populaton centre n ts own due-crcut to the admnstratve centre. The zone crteron s the nhabtant aggregated dstances from the centre of each populaton centre n the muncpalty to the centre of the zone. Each zone ncludes about 2000 nhabtants and s desgned to capture the average school dstrct. We also have ncluded a varable descrbng the populaton densty n the muncpalty. Muncpaltes wth an urban populaton are often assocated wth a more actve cultural lfe, but we have no justfcaton beyond ths to nclude the varable. From Table 3 we see that about half the populaton lves n urban areas n the 406 muncpaltes ncluded n the data. However, we must have n mnd that ths fgure s not weghted for the populaton of the muncpaltes. Thus small and sprawled muncpaltes n terms of populaton, have the same weght as large muncpaltes wth a more dense populaton pattern. Fnally, we also nclude varables descrbng the local populaton's average level of educaton. More precsely, we have dvded the nhabtants between 16 and 74 years n varous educatonal groups. Table 3 shows that on average, people wth prmary and secondary educaton domnate. The motvaton for these varables s that we assume that a more educated populaton tend to demand more cultural servces, especally relatng to the fner arts. Share of nhabtants, years old wth prmary school educaton, s the reference category n the calculatons carred out n the next secton. 5. Results 5.1 Elastctes wth respect to free ncome The elastctes wth respect to free ncome I s found from A.3 n Appendx 2. The expendture share weghted sum of ncome elastctes wth respect to I does not sum to 1.0 snce I does not consttute the full ncome Y. The weghted sum nstead becomes the share of free ncome of total ncome I/Y. Thus the defnton of ncome neutralty s shfted from 1.0 to I/Y = 0.71 n our dataset. For most readers we beleve t s more convenent to mantan the famlar scale of ncome elastctes, such that servces wth an ncome elastcty above one s defned as luxures and below one as necesstes. We have therefore chosen to scale up all elastctes εi wth respect to free ncome by the factor Y/I. Table 4. Free ncome elastctes, Tobt-estmaton. Lnear model. 1. Chldren and youth actvtes 2. Lbrares 3. Cnemas 4. Museums 5. Arts dssemnaton 6. Sports 7. Cultural schools 8. Other cultural 9. Other servces εi As dscussed n secton 1, Håkonsen and Løyland (2011) found the ncome elastcty for an aggregate of all cultural servces to be Ths made the cultural sector the most ncome elastc servce sector n ther study. The level of the ncome elastctes n Table 4 confrms the fndng n several earler studes (cf. references n secton 1) that cultural servces ndeed are clearly ncome elastc on average. However, we also see substantal varaton wthn the cultural subcategores. The least elastc categores are cultural schools, lbrares and cnemas, and cultural schools s the only culture sector wth an ncome elastcty below one. In the other end of the spectrum we fnd qute hgh ncome elastctes for museums (2.72), other cultural servces (2.52), sports (2.14) and arts dssemnaton (1.92). These results ft well nto the general dea descrbed n secton 1 and 2 that the level of the ncome elastctes n part reflects the degree of central government regulatons and standards wthn the varous sectors. To our knowledge 12

14 there are few or none such regulatons wthn the four most elastc subcategores. In these sectors we fnd expendtures lke grants or supports to prvate sports clubs (sector 6), performng artsts and ensembles (s. 5), and voluntary organsatons (s. 8), expendtures to sports facltes and arenas (s. 6) and cultural houses and concert venues (s. 8 and/or s. 5), the muncpal culture admnstraton and consultants (s. 8), and grants or supports to prvate museums (s. 4). None such expendtures are n any way compulsory or regulated by the central government. In the cultural subcategores wth more moderate ncome elastctes, however, we fnd the two sectors where a supply s statutory n all muncpaltes regardless of ncome. Ths s the case for cultural schools and publc lbrares. In the former case there s a paragraph n the Law for prmary educaton, statng that all muncpaltes, alone or n collaboraton wth neghbourng muncpaltes, must mantan educaton for chldren and youths wthn musc and other arts. In the latter, there s a smlar paragraph n the Law of publc lbrares, statng that all muncpaltes must have a publc lbrary. There s also a requrement that all such lbrares must have personnel wth a specalzed educatonal background. Thus, lbrares and cultural schools are more regulated and standardsed than the other cultural servces, makng the ncome elastctes more suppressed. 5.2 Effects of matchng grants The model gves a full set own- and cross-effects of matchng grants to each sector. However, the majorty of the cross-effects are not sgnfcant at 5 % level. We therefore confne the dscusson to the own-effects of the matchng grants,.e. the effects for sector k of gvng grants to the same sector. These effects may be measured as elastctes n the same manner as the elastcty wth respect to free ncome n the prevous secton, cf. A.4. However, snce these elastctes become rather small numbers due to the grants small share of total ncome, we fnd t more llustratve to scale the effects to show the ncrease n gross expendtures for each sector followng from one extra NOK as a matchng grant to the same sector, cf. formula A.6 n Appendx 2. Ths s shown n Table 5. Table 5. Effects of ncreasng matchng grants to each sector. Extra gross expendtures per NOK extra grant ncome 1.Chldren and youth actvtes 2.Lbrares 3.Cnemas 4.Museums 6.Sports 5.Arts dssemnaton 7.Cultural schools 8.Other cultural 9.Other servces An nterestng ssue regardng the effects of grant fnancng, s the degree of accuracy or leakage,.e., how large a share of the grant fnancng that s spent n the grant recevng sector and how large a share that leaks over to hgher expendtures n other sectors. In the scale chosen n Table 5, a number of 1.0 would ndcate 100 % accuracy and zero leakage,.e., each NOK receved as a grant s beng spent n the ntended (grant recevng) sector. The fgures n Table 5 show that the leakage effects are greatest for museums (61 %) and chldren and youth actvtes (55 %).The leakage per cent s smallest for the cultural schools (8%), sports (14 %), and other cultural servces (15 %). In other words, gvng 1 NOK extra as a matchng grant to the cultural schools, NOK 0.92 extra s spent on cultural schools, whle the remanng 0.08 are spent elsewhere accordng to the estmated own- and cross-effects of the matchng grants n our model. 13

15 It should be added, however, that grant fnancng s very lmted for several of the cultural sectors, and especally low for the cnemas where only 2.6 % of the expendtures are fnanced by matchng grants. Although the effects of matchng grants to cnemas s sgnfcant at 1 per cent level, the very small use of matchng grants to ths sector makes t clear that matchng grants are of lttle mportance for the development of ths sector. 5.3 Effects of user fees and other sector-specfc revenues In ths secton we show the correspondng results as for the matchng grants above, only replacng matchng grants wth user fees and other sector-specfc revenues to each sector. Agan, we focus on the own-effects for each sector only. The nterpretaton of the own effects s the same as for the matchng grants ncreased gross expendtures measured n NOK n each sector f the same sector ncreases ts own revenues by means of hgher user fees or other sector specfc revenues by one NOK, cf. A.7 n Appendx 2. Table 6. Effects of ncreased user fees and other sector-specfc revenues to each sector. Extra gross expendtures per NOK extra fee ncome 1.Chldren and youth actvtes 2.Lbrares 3.Cnemas 4.Museums 6.Sports 5.Arts dssemnaton 7.Cultural schools 8.Other cultural 9.Other servces The effects of user fees or other sector specfc ncomes are generally hgher than one, mplyng a negatve leakage effect or an addng up-effect from extra muncpal free ncome. The only exceptons are chldren and youth actvtes, where 78% of extra fee ncome s spent n that sector (.e. leakage of 22 %), and cnemas wth an own-effect of 79 % and leakage 21 %. The strongest own-effects of user fees are found n the cultural schools (2.15), other cultural actvtes (1.81) and lbrares (1.65). The user fees n the cultural schools are a rather mportant source of fnance for ths sector, snce there are substantal parental fees per chld attendng cultural schools n most muncpaltes. On average the parental fees fnance 18 % of the gross expendtures to cultural schools, cf. Table Effects for cultural expendtures from the local age pyramd In ths secton we focus on control varables accountng for the age pyramd n the muncpaltes. Borge et al (1995) fnd that a hgh percentage of young people and elderly people reduce spendng on cultural servces. Ther explanaton s that hgher expendtures on educaton and elderly care crowd out cultural spendng. The three sectors day care (for pre-school chldren), prmary educaton, and care for the elderly alone accounts for around 60 % of the gross expendtures for the average muncpalty. It therefore seems reasonable to beleve that hgher need for expendtures n such large and expensve servce sectors may crowd out expendtures to smaller sectors lke the cultural servces. However, there are also other mechansms related to the age pyramd that may affect cultural spendng. A hgh share of famles wth chldren may reduce cultural spendng snce the parents have a more scarce tme constrant and therefore demand lttle cultural servces, cf Werck et al. (2008). Lkewse, a hgh share of elderly people may lead to hgher demand for cultural servces n a muncpalty because ths age group has a low opportunty cost of tme and therefore demands much culture. Ths effect s found by both Werck et al. (2008) and Stastna (2009). As dscussed n Bento et al. (2013), however, ths effect s 14

16 The control varables are measured as populaton shares, wth a scale from 0 to 1. Snce the dependent varable s the budget share of the nne sectors, the nterpretaton of the regresson parameters shown n Appendx 1 s the ncrease n budget share f the populaton share ncreases by the number 1. We have then used the regresson parameters to compute new predcted budget shares and expendtures followng from a 10 per cent pont ncrease for each of the populaton shares. Snce the dependent varables (expendture shares) for cultural servces are very small n numbers (cf. Table 2), we scale the expendture shares to per cent shares, and show the change n per cent ponts for each sector followng from a 10 per cent pont ncrease n the populaton shares n Table 7. Table 7. Effects on cultural servces from populaton share varables. Change n per cent ponts n expendture shares per 10 per cent pont ncrease n populaton shares. Chldren and youth actvtes probably reversed as the elderly becomes older, snce most very old people smply become too old to attend cultural events (or other events n general), Thus, as ponted out by Bento et al. (2013), the age pyramd may be expected to affect cultural spendng both va demand factors, actng through the votng behavour, and va crowdng out effects from a gven budget constrant. Lbrares Cnemas Museums Arts dssemnaton Sports Cultural schools Other cultural Other servces ** ** ** ** ** ** 6.00 ** ** ** ** ** ** ** ** ** * * ** ** * 0.02 ** ** 1.05 ** * * * * 1.01 ** Inhabtants above 16 years old, mentally retarded, per capta Inhabtants less than 16 years old, mentally retarded, per capta Inhabtants, years, per capta Inhabtants, 6-15 years, per capta Non-marred nhabtants per capta Dvorced per capta 0.10 * ** ** * Immgrants per capta * ** * ** Unemployed per cap ** ** 0.38 ** ** ** ta ** Correspondng regresson coeffcent n Appendx 1 s sgnfcant at 1 per cent level of sgnfcance. * Correspondng regresson coeffcent n Appendx 1 s sgnfcant at 5 per cent level of sgnfcance. For the most part, the results show the same effect as found by Borge et al. (1995): Varables ndcatng hgher demand for welfare servces tend to ncrease the expendtures to the resdual sector ( other servces ) and reduce the expendtures to the cultural sectors. There are a few exceptons, however. A hgher share of 67 to 79 year old nhabtants ncreases the expendtures to lbrares, ceters parbus. Also, a hgher share of dvorced people ncreases the expendtures to chldren and youth actvtes and to arts dssemnaton. The frst of these seems rather natural, whle the latter result s somewhat harder to explan. One varable goes n the opposte drecton altogether: A hgher share of unemployed reduces the expendtures to other servces and ncreases the expendtures to three of the cultural servces; lbrares, sports, and cultural schools. Only other cultural servces are negatvely affected by the unemployment level. 15

17 One qualfcaton s n order regardng the effects of these varables. All the varables (apart from the trend) n Table 7 are ncluded n the cost ndces n the General Purpose Grant Scheme. Ths scheme adjusts for varaton n the demands for welfare servces among the muncpaltes, and compensates the muncpaltes for (nvoluntary) extra costs from hgher demands for welfare servces. The results n Table 7 only show the partal effects from hgher shares of the varous populaton characterstcs n the table, but not effects of hgher ncome. To see the net effect for each sector, one would n prncple need to compare the partal effects shown n Table 7 wth the effect from ncreased general ncome followng from a change n the redstrbuton through the GPGS-scheme. We have not attempted to compute such net effects, snce t would be a rather tedous exercse to compute the extra ncome followng from hgher populaton shares. 5.5 Effects of other control varables Of the all n all 26 control varables we have commented upon effects of eght of them n the prevous secton. Of the remanng 18 we present the effects of a selecton dependent on what we consder the most nterestng. The results for all these varables are shown n Appendx Table A1. There are three varables that we have categorzed as poltcal varables. These are share of representatves wthout poltcal party connecton (F) n the local government, share of left-wng representatves (L) and the thrd s a Herfndahl ndex (HERF) measurng the party concentraton n the local government. The effects of the frst two are evaluated n relaton to the share of lberals/conservatves. The lterature on the effects of the rght-left dmenson on cultural spendng s ambguous, cf. Bento et al. (2013), Getzner (2002). Left-wng poltcans tend to favour hgher publc spendng n general, and may also be nterested n spendng more on culture, thus makng culture more affordable to low-ncome people. On the other hand, people wth hgher ncome tend to apprecate culturale experences more than the average ctzen, and therefore support publc spendng on culture (Schultze and Ursprung (2000), Getzner (2004), Potrafke (2010). We fnd that muncpaltes wth a large share of representatves wthout party connecton prortze cultural schools stronger than governments run by conservatves. A large share of left-wng representatves prortzes actvtes for chldren and youth, sports and cultural schools stronger than conservatves/lberals, whle they prortze lbrares and arts dssemnaton slghtly weaker. Thus, there s not a clear effect of left or rght wng majortes n the muncpalty board for the cultural spendng n general. The results for the coeffcents of Herfndahl ndex suggests that non-fragmented muncpal governments tend to prortze actvtes for chldren and youth, lbrares, museums and sports stronger than fragmented, whle they gve less prorty to cnemas and cultural schools. Stastna (2009), Getzner (2002) and Werck et al. (2008) fnd no effect of government fragmentaton on cultural spendng, and our result wth some postve and some negatve effects seems to confrm such a lack of a general effect from party fragmentaton for cultural spendng as a whole. The populaton sze s assumed to have an mpact on budget prortes, partcularly n sectors wth economes of scale. Bg muncpaltes can of course explot scale economes to a greater extent than small muncpaltes. On the one hand ths may lead to lower spendng per capta n bgger than n smaller muncpaltes gven the same cultural servce level for the nhabtants. However, many cultural goods requres such a large scale that the small muncpaltes smply become too small to offer the full range of cultural servces. Ths effect suggests that mportant parts of the cultural lfe become concentrated n the larger ctes. The emprcal lterature suggests that the latter argument domnates, snce Wert (2006), Borge et al. (1995), and Werck et al. (2008) fnd that populaton sze has a sgnfcant postve mpact on culture. Our results show 16