Estimation: logit and probit

4.5Estimation: logit and probit

Next the multinomial logit model, ordered logit and ordered probit results will be presented. It is assumed in the binary logit model that the dependent variable has two classes, whose probabilities are p and (1-p). In this case, the equation to be estimated is: where X contains the explanatory variables and β is a vector of coefficients. In the multinomial logit there are more than two alternatives or classes, e.g. “regularly”, “occasionally” and “never”. The probability of the choice “often” is where “i” stands for the person’s i choice between different alternatives k (1,…,) on the condition w_i These conditions are characteristics that have an impact on the person’s choice, like gender, year of birth or education. In the multinomial logit model one alternative is the zero alternative and the other alternatives are compared in relation to zero. The variable which describes age has been recoded more roughly due to a more rational way of presenting the results: age15_24, age25_34, age35_44, age45-54, age55_64, age64_. Unfortunately, the place of residence (province) cannot be used since the number of observations on some provinces is too small.

The often visiting group has graduated from an upper secondary school (level 5), from a university of applied sciences (level 7) or from a university (levels 8 and 9). Others with different educational background do not significantly belong to the group “regularly”. However, the marginal effects (table 4-7) of the education level on the visitor density shows that only women with the highest education are the most active. The age cohorts 45-54 and 55-64 are the most active. The “occasionally” group consists of those with elementary school, vocational school or comprehensive school education or they are still at school (zero alternative, constant in the equation). The marginal effects (in table 4-7) show that women belong significantly more often than men to the groups “regularly” or “occasionally”. The results are in line with the earlier studies (Kivekäs 1991) and support the first hypothesis: the person’s education, gender and age have a significant effect on the performing arts visiting density. The province variable is not used in the estimations presented in table 4-4 since the number of provinces is too big. To a few groups there would have been too few observations and the results would not have been credible. Due to that the models in the table 4-8 are estimated so that the provinces have been regrouped into bigger entities. The entities have mainly been formed according to the NUTS2 classification.

The results in table 4-8 show that the area variable is signicant when the zero alternative is northern Finland which is also verified in table 4-9 where the marginal effects are presented.

The first hypothesis is verified by the results of the logit and probit analysis: consumption of cultural events depends on gender, person’s age and education and there are substantial differences across regions. The effect of gender is clear between all the groups: “regularly”, “occasionally” and “never”. Women are more active. As the groups “occasionally” and “never” are compared, the effect of the person’s education is significant already when the education is either upper secondary school or vocational school. As the previous groups and the “often” group are compared, the effect of education is significant when the education is upper secondary school or any university degree. The age cohort 45-54 is most active. The impact of the region is the following: the Uusimaa region, other southern Finland and western Finland are different than other areas in Finland among those that “never” go to cultural events. The null area (in constant) in the estimations is northern Finland. To conclude, it can be argued that crucial educational level is upper secondary school. All education after the upper secondary school seems to increase cultural consumption. The separating points between the groups “occasionally” and “often” seems to be university degree and regionally Uusimaa region or eastern Finland. The ordered logit analysis confirms the multinomial logit results (table 9).

The second hypothesis proposes that cultural events and sport events are substitutes. Montgomery and Robinson (2006) show with American data (USA 2004) that these events are exclusionary. They have a separate public. Descriptive statistics of the sport audiences is presented in table 4-10.

Since the number of those that visit “daily” or “several times per week” is so small, these groups and “several times per month” are combined. Hence there are three groups: “regularly”, “occasionally” and “never” as above was done in connection with the cultural events. The correlation of the three valued (“regularly”, “occasionally”, “never”) participation into cultural events and participation into sports events is 0.09.

The results also show that area variable is not significant except western Finland, i.e. there are no differences elsewhere in comparison to northern Finland which is the zero alternative.

Table 4: Marginal effects of variables: Visitor density, sport events, including area variables