Abstract
Although some research has already focused on the analysis of sports and recreational (SAR) service expenditures, some shortcomings exist regarding the study design, data reliability, and applied analysis techniques. To provide consistent information on German households' consumption patterns, a regular demand system with a demographic translation extension is derived and applied to the aggregated category of SAR services and the subcategories of sport events, swimming pools, music lessons, dancing lessons, fitness studios, ski lift fees, and sports club membership fees. It is shown that consumers have different consumption patterns for different subcategories of SAR services. Furthermore, goodness of fit measures indicate that the models derived from the (extended) neoclassic demand theory seem to be appropriate. This means that deriving a theoretical demand model from utility maximization behavior is sui (or even necessary) for a comprehensive (and generally accepted) demand analysis also in the field of sports and recreation.
Notes
1. It has to be discussed whether a multivariate Tobit model might be necessary. Such models are required if the qualitative and/or quantitative decisions of the analyzed services (EVENT, POOL, MUSIC, DANCE, FITNESS, SKI, and CLUB) depend on each other. From a statistical point of view, this is the case if the error terms of two services are correlated. While this does not seem unrealistic (e.g., a general preference factor for or against sport might exist that is not part of the set of available independent variables), the development of adequate multivariate models is not satisfying: an approach developed by Heien and Wessells (Citation1990) is not consistent while the model developed by Shonkwiler and Yen (1999) generates inefficient estimates (Tauchmann, Citation2005). However, since Halvorsen and Nesbakken (Citation2004) could detect that the consideration of stochastic interdependencies (e.g., a seemingly unrelated regression in the second stage of the Tobit model type II) does not yield appreciable different estimates, this analysis is focused on a separate estimation equation-by-equation.
2. STATA did not compute the F-Test for the SKI models (Heckit II, LM) based on the LWR data of 2006. This might indicate a possible specification problem. However, in a major research project, we calculated the same models based on the LWR data of 2005. Note that the estimates of both years are very similar, while the F-Tests for the second stage estimation (Heckit II: F(18, 334)=4.23) as well as the linear model without correction of the sample selection (LM: F(17, 345)=2.21) are highly significant based on the LWR data of 2005.