Cultural Economics
Participation through the media and through practice as complements to live attendance.
Abstract
Cultural Economics defines participation by the different ways in which it takes place: attendance at live performances, consumption of cultural goods through the media, and practice of artistic activities. Using data from the Survey of Public Participation in the Arts for the United States (SPPA 2002), this paper estimates attendance at different live arts and at museums. It builds upon contributions on Cultural Economics by Borgonovi (2004) and by Ateca Amestoy (2008), as well as on Count Data regression models, mostly applied previously to the demand for health services.The purpose is to determine empirically if participation in any of those activities corresponds to sequential decision or whether the low attendance rate is due to the heterogeneity across subpopulations. . We estimate 7 participation equations, one for attendance to each activity: jazz, classical music, opera, musicals, theatre, ballet and dance, and museums and art galleries. For each activity, we include as explanatory variables participation through the media and through practice. We find evidence supportive for the latent models and for the complement nature of the alternative ways of participation to live attendance.
Introduction
There are a number of reasons for an economist to study participation in the arts. Economics and individual decision making models could be challenged at some instances by issues concerning the demand for cultural goods. For instance, when trying to explain heterogeneity in observed behavior while holding on taste stability hypotheses, one may observe that the more someone has consumed in the past, the more he/she would like to consume in the present. Rational addition theory [Stigler Becker1977], for instance, has tried to overcome the puzzling fact of consumers deriving more and more utility out of the consumption of cultural goods, without establishing exceptions to diminishing marginal utility. By modeling the personal cultural capital accumulation process, Stigler and Becker were able to explain positive addictive phenomena. However, there are also pragmatic arguments that call for a deep study of the determinants of participation in the arts, accounting for policy implications. For some countries, mostly in the European Union, the equity in the access to culture (broadly defined) acquires the status of subjective right. This right becomes even constitutionally protected in several countries, such as Spain. We have evidence that cultural participation rate diverges a lot among different racial, social, sex and economic groups. If worried about equity in access, then we should be concerned not only about identifying which groups are less likely to participate, but also about disentangling and understanding which are the determinants of participation itself.
Until very recently, research on participation in the arts had focused mainly on who participates and who does not, without paying special attention to the determinants of participation [McCarthy2001b] and [NEA2004]. Possibly, the main contribution of Cultural Economics has been the introduction of individual decision making models that enable to contrast empirically the effect of constraints and tastes over observed behavior.
Economics makes the distinction between underlying preferences (structure of the individual's utility function) and revealed preferences (mainly through choice) [Frey2003]. This distinction turns out to be particularly relevant when trying to conclude on the determinants of participation in the arts through live attendance. What we observe through the survey that we will explore is revealed preference. Individuals report how many times they attended to some cultural activities during the last year. Typically, we like to infer underlying preferences from observed choice. However, we have to take into account that the individual faces a constrained maximization process of utility. Therefore, choices are not only explained by the underlying preference structure, but mainly, by differences in constraints. A person may have a high preference for an artistic activity but may choose not to participate given his/her lack of time or given the fact that there are no facilities nearby.
Participation factors involve personal, household and institutional characteristics that accommodate the personal resources that an agent has in order to satisfy a general need called "cultural appreciation", as well as some measure of personal tastes. In this paper, we will adopt a Beckerian approach, and will consider that tastes are stable. Particularly, we will focus on the relevance of educational factors, of family conditions and of the effect of participation in that disciple under the two other alternative definitions of participation: through the media and by actively performing. Could it be the case that attendance at live performances may be challenged by participation through the media, a less time intensive activity? We can understand that the intangible cultural good present in each of these three definitions is the same. It is just the material substrate which differs from one another, leading for different technologies for the production of the "cultural appreciation" commodity.
We take benefit of the rich dataset available for the study of arts attendance in the USA. The Survey of Public Participation in the Arts, in its 2002 release, provides us with a large dataset that contains detailed information on participation in the arts through the three types of participation, specific arts education, and a bundle of socio economic factors regarding the individual and the household.
In this paper, we run an extension of the methodology applied in a previous paper [Ateca Amestoy2008], where we explored the most suitable way to estimate theatre attendance in the United States. Here, we focus on the seven benchmark cultural activities, as defined by the USA National Endowment for the Arts. We estimate up to seven participation equations, one for each of the following artistic activities: jazz concerts, classical music concerts, opera, musicals, theater, dance and ballet and museums and art galleries. We do not account for any possible effect of participation in one discipline on the other disciplines. Therefore, potential complementarities across different disciplines are not considered, and the interest concentrates the effect of different formats of participation on attendance.
We will also try to answer the following question: are "non goers" really "never goers"? "Non goer" being defined as a person who reported in the survey non participation at a particular activity, and "never goer" as a person who would belong to a different subpopulation, potentially having a nil taste for those cultural activities. We design an empirical strategy that allows us to explore heterogeneity in behavior. We also account for the possibility of different decision making processes. Two families of models are estimated. First, latent models (zero inflated count models) and, then, results are confronted with those derived from a two part model (namely, a hurdle model that combines a logit and a zero truncated negative binomial model).
The plan of the paper goes as follows. Section 2 overviews the literature on participation in the arts, methods and findings. Section 3 briefly presents the data and the empirical specification. Section 4 presents the results, and section 5 concludes.
Previous findings and estimation approaches
Only very recently, did the literature focus on the study of the determinants of cultural participation (for a broad review, see [Seaman2005]). Before, studies described who did participate, instead of why he/she did [Gray1995], [McCarthy2001b] and [NEA2004]. Contributions from Cultural Economics started testing the validity of individual decision models for cultural goods.
Closely related papers explain the influence of past consumption capital on present demand, and the relevance of general human capital and specific human cultural capital in determining the profile of people who participate more in the arts. Some of there departure from theoretical models ([Ateca Amestoy2007], [Levy Garboua1996] and [Stigler Becker1977]), and some others test empirically the propositions derived from the former contributions ([Borgonovi2004], [DiMaggio Mukhtar2004], [Fernandez Blanco2004], [Gray2003], [Upright2004]). Most of those empirical works accounts for phenomena such as rational addiction [Stigler Becker1977], or learning by consuming processes [Levy Garboua1996].
The empirical strategy depends on the nature of the variables available. The most frequent approach has been the consideration of participation as a dichotomous variable ([Gray2003] and [Seaman2005]). Thus, logistic or probit regressions where run to model the probability of attending at some particular manifestation of life arts.
If the way in which variables have been elicited and recorded allows it, some other interesting questions can be addressed by using different estimation methods. For instance, the one employed in an study of the demand for cinema in Spain can model very heterogeneous behavior [Fernandez Blanco2004]. These authors estimate a finite mixture model, so different behavior for different subsamples can be estimated. If we believe that we cannot distinguish between never goers and attendees, but can distinguish between groups with high average participation ("enthusiasts") and with low average participation (plain "goers"), then this kind of estimation provides a good approach.
Borgonovi [Borgonovi2004] uses the same dataset that we do, and treats attendance and frequency as two different entities. She justifies the estimation of a binary model for participation and an ordered model for frequency of participation on the grounds of the following assumption: there may be major differences characterizing the behavior of occasional attendees and frequent participants in the live arts. For this author, this is a call to study the 2 groups separately.
Count data models have been estimated before for theater participation using this dataset. In a previous article, we estimate participation and compare the estimated results from an ordered probit, from a zero inflated negative binomial and from a hurdle model [Ateca Amestoy2008]. All those methods are broadly explained in that article. In this one, we proceed in exactly the same way.
Data and empirical specification
The Survey of Public Participation in the Arts for the year 2002 collected information on participation in the arts by US citizens between August 1st 2001 and August 1st 2002. It was the fifth study in a series conducted by the Bureau of the Census for the National Endowment for the Arts since 1982. It was conducted as a supplement to the Current Population Survey [BureauoftheCensus2003]. Households were sampled and, for each member over 18, the dataset contains information on attendance at the following artistic activities: jazz, classical music, opera, musicals, theatre plays, ballet, dance, art museums, art crafts, and visits to historical parks and monuments. It also covers participation by means of other types of cultural practice such as the consumption of cultural goods through the media, through recorded or printed matter, and through active practice of artistic activities. Further, it provides information on education in arts, preferences for different types of cultural activities and school socialization. The survey collected 17,135 questionnaires for a representative sample of households in the USA. In each of the selected households, all individuals over 18 were interviewed.
The descriptive statistics on cultural participation for the USA (see Table participation related variables) provides us with a first approximation to the heterogeneity on participation patterns among individuals. Let us take jazz participation in its three different types to describe the pattern. While there is a 89,2% of people that did not attend any jazz concert during the sampled year (standard error 0.003), the average number of concerts among goers was 3,104 (standard error 0.1390). Given this heterogeneity, due to the presence of a high proportion of non goers, average attendance is reduced to 0,331 times (0.1793). Participation through the media is captured by the proportion of people who listened to jazz records (17,21% with standard error of 0.003), and by those who did it on the media (16,43%, with 0.003). Finally, 1,31% performed jazz themselves (0.001). A similar pattern can be observed for each of the other 6 activities.
Overall, we can consider that, even if some of these activities more popular than others (26,5% of the adult population visited some museum or art gallery in the previous year, whereas 8,7% went to see ballet or dance), there is a very big fraction of the population that did not attend to any of these activities. Actually, 59,56% of the adult population did not attend at any of the benchmark activities during the pervious year. We would like, therefore, to set a suitable empirical specification that is able to capture the possibility of unobserved heterogeneity in our population. Our dependent variable is, for every participation equation, the number of times that the individual reported having attended at that particular activity during the previous year. All of them are count variables that include a high proportion of zeros (ranging from 73,08% of the observations for museums to up to 97,04% of the observations for opera).
We will rely on both theoretical models of demand for cultural goods and on empirical regularities reported in the literature. Theoretical models departure from beckerian models and consider a basic commodity called "cultural appreciation" that each individual has to produce and consume in order to fulfill that common, stable and basic need. In order to produce cultural appreciation, the agent has to allocate a number of resources: mainly income and time. Typically, participation by means of attendance has always been considered a time intensive commodity with respect to participation through the media. The former implies some travel cost and some previous planning that can be avoided by the consumption of that type of cultural goods in the media matter.
Income and time get combined with other personal resources in order to produce pleasurable and meaningful experiences. One of the most important variables in order to determine how productive a person might be combining those resources is cultural capital. In our work, cultural capital is modeled as a personal resource that is accumulated by previous consumption and by formal and informal education. Given the social dimension of participation in the live arts, we will also consider some family related variables. Unfortunately, from our sample, we cannot identify joint participation for individuals in the same households. However, we can test simple hypotheses of the effect of the participation of other member of the family over owns attendance, in the spirit of the work by Upwright [Upright2004].
We must however account for some of the constraints over the resources of the individual. Income will be the first one, but also some variables that determine time availability. Since we do not find in the survey a direct measure of the amount of free time, we try to accommodate time availability by introducing labour status and household composition related variables. A variable that informs about the preference for that activity is included.
We will also control for several personal variables that influence cultural participation. Those are what Seaman calls "mixed factors", such as sex, age and ethnicity [Seaman2005]. Given that there is broad evidence of attendance being a mostly urban phenomenon, we include the size of the habitat. Explanations for this regularity take into account that live arts are mostly provided in urban areas. Thus, attendance is living in non urban areas is subject to higher implicit costs (if only, because of travel costs and the amount of time required to reach the venue). Given the importance of location and supply availability, we control for residence by putting some structure in the variance covariance matrix. In our estimations, we will allow for clustering among observations that belong to the same consolidated metropolitan statistical areas (CMSA), as defined by the Bureau of the Census.
This set of explanatory variables, would determine the amount of resources allocated and the technology being used. But, as mentioned before, participation by means of attendance is not the only way in which a particular category of cultural goods can be consumed. To test the influence of participation through the media and through active practice, we introduce for each of the participation equations, the reported participation by those two alternative means.
Participation equation for a particular activity is determined by the following explanatory variables:
• demographic variables: age (we introduce dummies for up to 8 age intervals, defining people on their forties as the baseline), sex (female is the base), ethnicity (black, indian and islander; white is the baseline), and labor status (dummies for employed full time baseline , employed part time, unemployed, retired, not in the labor force).
• household variables: income (4 quartiles, the baseline is the first one), household size (as a count variable), and familiar participation (a dummy that accounts for the fact that someone else in the household reported having attended to that discipline).
• geographic variables: size of habitat (metropolitan baseline , central, balance).
• cultural capital variables: own level of formal education (5 levels), parental education (both parents', the same dummies) and specific education in the arts controlling for starting age (the baseline is no specific education, one of the dummies controls for having started before 18, and another if started after 18).
• preferences: would go more if no restrictions (a dummy variable).
• alternative participation: by means of media and by practice (by records, by books, by media and performing or acting).
In what follows, we will concentrate on determining the most suitable estimation method to get some estimates that allow us to better understand the determinants of participation. Thus, when trying to model attendance at live arts, we will allow distinguishing between what determines that a person is more/less likely to attend at performing arts, and what determines that a person is more/less likely to attend more, or with higher intensity. When doing so, we will be able to test empirically if there is some evidence supporting that some of these types of participation correspond to sequential decision, or if the distinct behavior is due to the heterogeneity of different subpopulations.
The estimation procedure goes as follows. We estimate up to 4 count data regression models: Poisson, Negative Binomial (II); Zero Inflated Poisson, Zero Inflated Negative Binomial (II). We do the hypotheses testing suitable to determine the best among them. As we will discuss below, for all our estimations it turns out that best fit is provided by Zero Inflated Negative Binomial (II) models. Particularly, a logit rules the inflation part of the model, i.e., the probability of belonging to one subpopulation or another; and a Negative Binomial II distribution rules the counts of the model, including zero, each of which comes with a positive probability conditioned on the fact of belonging to one precise subpopulation. That last model belongs to the family of latent class models. We find evidence of underlying heterogeneity in the observed behavior. Actually, we find that observed behavior is the mixing of two distinct subpopulations: the "never goers" and the "goers". The type of the individual is not elicited in the survey, but inferred from behavior. Never goers find their optimum in zero. Among the goers, we take into account not only the positive values of the dependent variable, but also some of the zeros that result from corner solutions. Should some of the restrictions be relaxed, the individual would attend a positive amount of times.
We compare the results derived from the selected model with those results derived from the estimation of a hurdle model. We estimate a two part model formed by a logit for the inflation part and a zero truncated negative binomial for the positive counts. The behavioral models underlying hurdle models determine that decision making process consists on two parts. First, any person is a non goer until there is some change that determines that the individual starts going. All the zeros come from the same distribution, and some variables make the individual cross the hurdle and become a goer. Once that decision is taken, the zero truncated process rules the positive counts. By using the goodness of fit and the Bayesian information criteria, we compare how well each of the models performs. As we will discuss below, we find evidence in our estimations that support that there are no two part processes for arts attendance. For every activity, we find that latent class models are the ones providing a best fit.
Results
We start by discussing the selected estimation procedure, since we find evidence supportive for the latent class model with respect to the two parts model. Therefore, we gain some insights on the nature of the participation decision making process. We do not find evidence that indicates that participation is a two part decision model, in the sense that a person first decides whether to be an attendant or not and, latter, decides how many times he/she wants to attend. What we find, for all the considered activities, is that heterogeneity in observed behavior is determined by the underlying heterogeneity. There are two subpopulations: a first one of never goers, characterized by some particular traits, some of which are common to any type of activity; there is a second subpopulation formed by those who may attend a positive number of times. Therefore, we find that all zero outcomes of our dependent variable do not come from the same distribution.
Recall that by using the Zero Inflated Negative Binomial to estimate the participation equations, we have found evidence of the underlying existence of two different sub populations. Therefore, we have estimated for each of those participation equations a part of the model that rules the probability of being a never goer, i.e., the inflation part; and another part that rules the intensity of participation, i.e., the probability of each of the counts. For the inflation part of the model, a positive sign on the estimated coefficient informs that the variable has a positive effect over the probability of being a never goer. For the count part of it, a positive coefficient determines the higher probability of attending more times. Discussion of the estimation results starts by presenting the effect of each of the bundles of explanatory variables and relating our empirical findings with the prediction of traditional models and with previous results.
Regarding the age variable, even if it is statistically significant, the effect is not very big. We find monotone and positive effects for the count part of the model of classical music concerts. The same for the count part of the ballet and dance attendance. However, there is a somehow U inverse shape for museum attendance.
One of the regularities recurrently reported in the literature on participation determines the high feminization of participation in the arts. By means of our analysis, we have been able to distinguish that gender has only influence for the inflation part of the model. For nearly every activity, with the exception of jazz and musicals, men have higher probability of being never goers. However, when assessing the impact of the sex variable over frequency of participation, we only find statistically significant effects for theater (a negative impact over the count), and for dance (a positive one).
Another type of results that come out of participation studies is that participants are wealthier than non participants. We find out that income has indeed an important role in the inflation part of the model. There is a nearly monotonic effect of income in the inflation part (in the sense that higher income reduces the probability of never attending). No statistically significant effects are found for opera and musicals. Surprisingly, we also found negative effect in the count part for classical music concerts and for museums.
Regarding the labor status and household composition variables, which accommodate the influence of some variables over time availability, we do not find a clear pattern for labor market related variables. However, if of a small magnitude, household size has a negative impact. Overall, if marital status has statistically significant effect, it has a positive one with respect to the baseline (being married). Familiar participation has a very big impact in the inflation part of the model, reducing dramatically the probability of never going. In some cases, we find evidence for its effect over intensity (greatest for opera, smaller for classical music concerts).
With respect to the size of the habitat, in our estimations we find scarce evidence supportive for the urban character of participation in the arts. Not living in a metropolitan area, but living in a central one reduces the probability of ever going for opera, ballet and museums. No effect is found for any of the count parts of the participation equations.
Overall, the effects of a greatest magnitude come from owns formal and informal education. We find that if parental education has any effect, it is by means of fathers' education and mostly in the inflation part (so we conjecture that it could be capturing more of a wealth effect, and less of an early exposure or transmission effect). Regarding own formal education, our findings describe the following patterns: Monotone effect for jazz in both parts, even more pronounced for theater and for museums; mostly effects over the inflation part for opera, musicals and dance; for classical music, the effect over the inflation is monotone, but U shaped over the count part. We find out that the magnitude of the coefficients for own formal education variables is, on average, twice as big as the ones for income, therefore supporting the formerly stated fact that educational variables have greater explanatory power than income variables. For arts specific arts training, we find a big effect on inflation for music related activities (jazz, classic, opera, musicals). Also intensity is determined by those variables for museums and classical music. Recall that we distinguish training by the fact that it started when adult or before. We find that if both are significant, there is a bigger effect if started when adult.
One of the interests of this paper is to determine the relationship between the alternatives to live participation (i.e. to attendance): namely, participation through the media and through arts practice. For other matter of participation, we find out that it enhances participation for every disciple (when comparable, there is a higher effect of records or books with respect to some passive media usage). It also increases intensity for every discipline. When considering acting or performing, there is no common evidence: for jazz, it determines higher intensity; for classical music, both; acting has only effect over the inflation. If statistically significant, it always has a positive sign for probability of goer as well as for intensity, as expected. We can remark the big magnitude of the estimated coefficients. However, this latter is the most infrequent type of participation.
Finally, the effect of the go more variable is very big in the inflation part (and for all participation). It highly reduces probability of being "never goer", and determines intensity for theater, ballet and museums.
Conclusions and final discussion
In this paper, we have provided evidence that, for these data, all the generating processes seem to be a finite mixture. We can, therefore, explain the heterogeneity in the observed behavior by considering that there is a latent process by which a person can be of either one of these types: never goer or goer. However, some of the consumers in the goer group can decide not to attend, given the constrained maximization process, so not all zeros come from the same distribution. We have also found evidence that participation decision is not a two part decision.
Once that we have selected the suitable estimation technique, we have characterized the binary inflation process and the count one. For the binary inflation, we have estimated the unobserved latent process. For the count part of the model we have estimated the probability of a attending a zero or positive number of times conditioned on the fact of not being a never goer. By doing so, we have been able to disentangle the effect of a bundle of explanatory variables over the two distinct processes.
One of the main results of our paper has been the complementary character of the three ways of cultural participation. Participation through the media and other matters determines both higher intensity of participation and smaller probability of being a never goer. For active involvement, it determines higher intensity and, for some cases, also smaller probability of never going.
Our results may shed some light over the always controversial initiatives such as the broadcasting of the Berlin Philharmonic Digital Concert Hall, often considered threats over live arts. Even if we neglect the social aspects of participation by attendance, there might be still the sense of uniqueness and, in terms of the pianist Daniel Baremboin, "one timeness" of attendance [Battle, 2009].
Acknowledgements
I would like to acknowledge the hospitality of Prof. John O`Hagan and of the Institute for International Integration Studies in Trinity College Dublin, where part of this research was conducted. Previous work was presented in the ACEI Conference in Boston, June 2008. Financial support from the Ministry of Research, Spain (SJE 2006/10827) and from the Basque Government (IT .241 07) is acknowledged.
References
Ateca Amestoy, V. (2007), Cultural Capital and Demand. Economics Bulletin, Vol. 26 no. 1 pp. 1 9.
Ateca Amestoy, V. (2008), Determining Heterogeneous Behavior for Theater Attendance. Journal of Cultural Economics, Vol. 32(2): 127 151.
Becker, G.S. (1965), A Theory of the Allocation of Time. Economic Journal, 75(September): 493 517.
Borgonovi, F. (2004), Performing Arts: an economic approach.Applied Economics 36: 1871 1885, 2004.
Bureau of the Census (2003), Current Population Survey, August 2002: Public Participation in the Arts Supplement. Technical Documentation Attachments.Realized by the Bureau of the Census for the National Endowment for the Arts, Washington.
Cameron, Colin A. and P.K. Trivedi (1998), Regression Analysis of Count Data. Cambridge University Press.
Deb, P. and P.K. Trivedi (2002), The structure of demand for health care latent class versus two part models. Journal of Health Economics 21 (4): 601 625.
DiMaggio, P. (2004) Gender, Networks, and Cultural Capital. Poetics 32: 99 103.
DiMaggio, P. and T. Mukhtar (2004). Arts Participation as Cultural Capital in the United States, 1982 2002: Signs of Decline? Poetics 32: 169 194.
DiMaggio, P.J. and F. Ostrower (1990). Participation in the arts by Black and White Americans. In Pankratz, D., & Morrie. V. (eds) The Future of the Arts: Public Policy and Arts Research: pp. 105 140. Praeger Press.
Fernández Blanco, V., Prieto Rodríguez, J. and Orea Sánchez, L. (2004). Movie enthusiasts versus cinemagoers in Spain: A Latent Class Model Approach. Presented at 13th ACEI Conference, University of Illinois at Chicago.
Frey, B.S. (2003). Arts & Economics: Analysis & Cultural Policy. Springer Verlag,New York.
Gapinski, James H. (1986). The Lively Arts as Substitutes for the Lively Arts. American Economic Review 76(2): 20 25.
Gray, Charles M. (1995). Turning On and Tuning In: Media Participation in the Arts. National Endowment for the Arts Research Division Report #33, Seven Locks Press.
Gray, Charles M. (2003). Participation. In Towse,R. (ed). Edward Elgar.
Hilbe, J.M. (2007). Negative Binomial Regression. Cambridge University Press, New York.
Lévy Garboua, L. and C. Montmarquette (1996). A Microeconometric Study of Theater Demand. Journal of Cultural Economics 20: 25 50.
Long, J.S. and J.Freese (2006). Regression Models for Categorical Dependent Variables Using Stata (2nd edition). Stata Press.
McCarthy, K.F., Ondaatje, E.H., and L. Zakaras (2001). Guide to the Literature on participation in the Arts. RAND Corporation.
Mullahy, J. (1998). Much ado about two: reconsidering retransformations and the two part model in health econometrics.Journal of Health Economics 17: 247 281.
Mullahy, J. (1986). Specification and testing of some modified count data models. Journal of Econometrics 33: 341 365.
National Endowment for the Arts (2003). Survey of Public Participation in the Arts 2002. National Endowment for the Arts.
National Endowment for the Arts (2004). 2002 Survey of Public Participation in the Arts. Research Division Report # 45. National Endowment for the Arts.
Seaman, B.A. (2005). Attendance and Public Participation in the Performing Arts: A Review of the Empirical Literature. Nonprofit Studies Program W.P 05 03, Georgia State University.
Stigler, G. and G. S. Becker (1977). De Gustibus Non est Disputandum. Journal of Political Economy 67(1): 76 90.
Upright, C.B. (2004). Social capital and spousal participation: spousal influences on attendance at arts event
