36 th Cordoba Guitar Festival : Spectator Analysis using Structural Equation Modelling ( SEM )

The Cordoba Guitar Festival is one of the most important cultural events in Spain. This article analyses the musical preferences, satisfaction, attitudinal loyalty and behavioural loyalty of spectators who attended the 36th festival held in July 2016, as well as the festival’s economic impact on the city. To achieve this aim, a structural equation model (SEM) was used. The results show the goodness-of-fit of the model and indicate that the observed data fit the expected dataset.


Overview
The Cordoba Guitar Festival is one of the most important cultural events in Spain.Its title alone combines two awe-inspiring concepts: the guitar-the instrument par excellence of Spanish music-and Cordoba, a World Heritage site and host to the event.
This paper aims to analyse spectators' responses to the different performances [1] they attended over the 10-day festival.The data were collected through a questionnaire administered to spectators attending the 36 th festival held in July 2016 and analysed using latent and observed variables.The latent variables include musical preferences, satisfaction, attitudinal loyalty, behavioural loyalty, and economic impact on the city.The relationship between the latent and observed variables, which were subdivided into exogenous and endogenous variables, was also analysed.Therefore, the following four hypotheses (H) were formulated: H1: Individual musical preferences influence spectator festival experience.

H2:
Spectator festival experience is measured by level of satisfaction.

H3:
Level of satisfaction affects loyalty.Two variables related to loyalty were analysed: attitudinal loyalty, which measures the probability of recommending the festival, and behavioural loyalty, which measures the number of festivals and concerts attended.

H4:
The level of economic impact of the festival is influenced by spectators' musical preferences.
To test the hypotheses, structural equation modelling (SEM) was used, a statistical tool which has become increasingly popular among social science researchers.The model enables investigating a set of relationships among one or more continuous or discrete independent variables and developing a way of breaking down the observed correlations into a system of equations that describe the hypotheses concerning causal relationships.Therefore, the aim of SEM is to study the causal relationships among directly observable data under the assumption that the relationships are linear [2].

Literature review
Music festivals are unique, special events that attract spectators for a variety of reasons [3].
According to [4], musical events, in the same way as art festivals, are able to offer spectators a musical genre, for example, classical music, or a combination of several genres.Many authors have studied music festivals, among them [5], [6], [3], [7], [8], [9], [10] and [11].[12] highlighted that the consumption of different musical categories is influenced by sociodemographic factors and motivations and that different musical genres have different target audiences.[11] concluded that music festivals have the potential to create benefits beyond the festival itself or the host destination, in particular by encouraging the appreciation of a musical genre, increasing the purchase of musical products, and promoting musical tourism.Likewise, [13], [14] and [15] studied the sustainable management of cultural events, a relatively new field in the economy of culture.
This section presents a review of the SEM literature that has supported this research.[16] proposed a model in the framework of SEM to assess preferences among quality of life dimensions for the elderly.Likewise, the recent work of [17] provides clear examples of SEM using AMOS.As regards the classical literature on this topic, [18] proposed a framework for representing personality constructs at four levels of abstraction using latent variables.[19] considered two types of error involved in fitting a model: the error of approximation and overall error.According to [20], normedthe coefficient yield new normed (CFI)-and nonnormed fit indexes (FI) are frequently used as adjuncts to chi-square statistics for evaluating the fit of a structural model.Moreover, [21] provided a comprehensive and more advanced treatment of structural equations with latent variables.
[27] used a SEM approach and found that motivation and participation are linked to the perceived value of the tourist experience, and that motivation affects level of participation.This supports the hypothesis that motivation has a positive effect on the perceived value of the destination experience.This same hypothesis is also supported by the study of [26] using SEM.
[28] analysed the importance of festivals in determining emotions, satisfaction, and future behaviour of participants in gastronomic events.The authors used SEM with latent variables to survey the responses of attendees to the Italian festival Friuli DOC.
Several studies have analysed loyalty using SEM [29].[30] suggested that future research could use SEM to determine the relationships between scales, including festival experience, festival loyalty, and festival benefits.[29] used SEM to examine tourists' experience of heritage tourism.The authors showed that the quality of the experience has a direct effect on perceived value and satisfaction, which supports the hypothesis that an increase in tourists' perceived value increases behavioural intentions (increased loyalty).This hypothesis is also supported by [31], who found that perceived value affects behavioural intentions both directly and indirectly.[32] examined the relationship between hedonic and utilitarian values and tourists' overall shopping experience satisfaction and destination loyalty.The authors tested hypotheses on the hedonic value of word-ofmouth purchasing and the utilitarian value of word-of-mouth purchasing, and found that the first hypothesis is supported but that the second is not significant.
Other studies using SEM, such as [22], [23] and [26], do not directly relate perceived value to loyalty but through satisfaction, finding support for both hypotheses (perceived value-satisfaction and satisfaction-loyalty).Although these studies do not report the indirect effects, satisfaction is very likely a mediating variable between perceived value and loyalty.Lastly, some studies using SEM do not support the hypothesis that higher perceived value leads to increased spectator loyalty ([25]).

Sample
The sample was drawn from a survey of 631 spectators who attended the festival, of which 612 were validated.The survey is divided into four large blocks: (i) spectator characteristics and attendance to previous festivals; (ii) spectator preferences or tastes in relation to the musical genre on offer in the programmed concerts and motivation for attending the festival; (iii) perceived satisfaction and expectations for the future; and (iv) sociodemographic characteristics of the spectators.

Types of variables in SEM
Different types of variables are defined in SEM according to their measurability and function in the model ( [17]) as follows: • Latent variables, also known as unobserved or hidden variables, are usually the key variables of interest in an analysis and capture abstract concepts that can be observed indirectly through their effect on observed variables.They are represented in Figure 1 by ovals.For this study, the latent variables have been defined as musical preferences, satisfaction, attitudinal loyalty, behavioural loyalty, and economic impact.
• Observed variables are variables that can be measured.They are represented by rectangles (Figure 1).There are three types of observed variables: -

Exogenous variables (X).
In multiple regression, exogenous variables are known as independent variables, that is, they influence but are not influenced by other variables.On the other hand, exogenous variables have no arrows pointing to them, only arrows originating from them.Five musical genres related to spectators' musical preferences at the festival were defined as exogenous variables: Blues, Classical, Flamenco, Jazz, and Rock.Seven factors related to spectator satisfaction, spectator attitudes and economic impact have been defined as endogenous variables.
-Error variables.These variables capture the variance not considered in the model.All exogenous and endogenous variables have associated error terms, which are represented by small circles.The full set of variables can be seen below in Figure 1.
Under these rules, all causal and measurement theories are usually represented in a similar way to systems of equations, providing they fulfil the following criteria: (i) all causal relationships should be represented in the diagram; (ii) all variables that cause endogenous variables should be included in the diagram; and (iii) the diagram should be as simple as possible and only contain relationships that can be justified theoretically.
As a result, SEM diagrams follow particular conventions to derive the corresponding equations: (i) observed variables are represented by rectangles and latent variables are represented by ellipses; (ii) errors are represented by small circles, and (iii) bidirectional relationships (covariances) are represented by curved, dual-headed arrows and unidirectional relationships (correlations) by a single-headed arrow.1, it should be noted that:

Amos diagram
-The first latent variable, Musical preferences, is related to the observed variables that represent the five musical genres at the festival: Blues, Classical, Flamenco, Jazz and Rock.
-The second latent variable, Satisfaction, is related to the observed variables Rock, Performance of latest album tracks, Variety of tracks performed, and I will return to future festivals.
-The third latent variable, Attitudinal loyalty, is related to the observed variables Performance of latest album tracks, I made the right choice, and I will return to future festivals.
-The fourth latent variable, Behavioural loyalty is related to the observed variables Attendance at previous editions, Number of concerts attended, and Average planned expenditure.
-The fifth latent variable, Economic impact, is related to the observed variables I made the right choice, I will return to future festivals, Number of concerts attended, and Average planned expenditure.
As regards the error variables (which capture the variance not considered in the model), it can be seen that the observed variables X1 Blues and Y7 Average planned expenditure do not have covariance relationships with the other error variables.This is due to the need for simplicity as mentioned in section 2.2.That is, if they were related to another musical style error variable or to another endogenous variable, the fit would improve slightly but would overload the design of the model and make it more difficult to interpret.

Goodness-of-fit
All the variables were estimated using maximum likelihood estimation (MLE), which is the most common method used to fit SEM as it provides consistent, efficient, and unbiased estimates for smaller samples.MLE requires normally distributed variables, although failure to meet the multivariate normality condition does not affect the method's ability to estimate unbiased model parameters.MLE is able to facilitate the convergence of estimates even in the absence of normality.
The results obtained are correct; therefore, the model conforms to the baseline assumptions.
Following [33], the main results of the study are presented below: chi-square (CMIN), probability (P), the minimum fit measure (FMIN), comparative fit index (CFI), and root mean squared error of approximation (RMSEA).

Measures based on covariances: CMIN, FMIN, RMSEA and CFI
The chi-square ( 2 ) goodness-of-fit test is one of the most widely used nonparametric tests to determine the discrepancy between an observed dataset and an expected dataset.This test and its limiting probability (P) is the only goodness-of-fit test associated with significance testing; the other measures and indices are merely descriptive.The result obtained with  2 is very good: 27.790 with 23 degrees of freedom (DF) and a probability level, P, of 0.224.Therefore, it can be confirmed that the  2 obtained enables us to test the four hypotheses, and that the model fits the observed data well (612 observations).

FMIN
The FMIN index is used to measure a model's fit.FMIN is the minimum value of the discrepancy function F, which is obtained by fitting the model to population moments rather than sample moments.The minimum value of the discrepancy (0.045) indicates a good fit.

RMSEA
RMSEA is an index based on the model covariance that can also be interpreted as the mean approximation error per degree of freedom.According to [19], values below 0.05 indicate a good fit of the model.Therefore, our result of 0.018 also indicates a good fit.

CFI
CFI compares the discrepancy between the covariance matrix predicted by the model and the observed covariance matrix to the discrepancy between the zero model covariance matrix and the observed covariance matrix in order to assess the degree of loss that occurs when the model is changed to the zero model.The values of this index range from 0 to 1.According to [20], the CFI value should be greater than 0.90, which indicates that at least 90% of the data covariance can be reproduced by the model.Therefore, our result of 0.996 is excellent.It should be mentioned that, in general, the results for the regressions (Table 1), the coefficients of variances (Table 2), and the covariances (Table 3) also indicate the goodness-of-fit of the model to the observed data.

Discussions and Conclusions
After 36 editions, the Cordoba Guitar Festival, which is held at the beginning of summer for around 10 days, has become a national benchmark in the world of culture due to the fact that (i) it revolves around the guitar, the Spanish musical instrument par excellence, (ii) it offers a diversity of musical styles, and (iii) it creates a very special atmosphere which envelops the city and is particularly suited to catering to spectators.
The many musical styles on offer at the festival are gradually becoming a reference point for research; not only in its facet as a music or cultural festival, but also in the many facets that characterise the spectators who attend the concerts.
In this paper we formulated four hypotheses concerning the spectators who attend the festival that are supported by the fit of the SEM statistical tool.The goodness-of-fit of the model was primarily determined by the chi-square index ( 2 ), which, as mentioned above, is the most commonly used nonparametric test for evaluating how well an observed dataset fits an expected dataset.In this respect,  2 is equal to 27.790 with 23 degrees of freedom and a probability of 0.224, thus indicating a very good fit.
As regards hypotheses H1 Individual musical preferences influence spectator festival experience, H2 Spectator festival experience is measured by level of satisfaction, and H3 Level of satisfaction influences loyalty, the table of frequencies presented in Table 4, 5, 6, 7 and 8, confirm that the most preferred musical genre was Rock (40.8%), followed by Classical (35.6%),Jazz (33%), Flamenco (32.4%), and Blues (29.1%).These findings are corroborated in Figure 1 and Table 4, where it can be seen that the error variable e65 Rock has five covariances.Specifically, covariance is found for e62 Classical with a value of -0.411; e63 Flamenco with a value of -0.204; d112 Performance of latest album tracks with a value of 0.218; d123 Attitudinal loyalty with a value of 0.106, and d7 Number of concerts attended with a value of -0.124.The first three values show significant dependence and minor standard error (between 0.074 and 0.071).
The fourth hypothesis, H4 The level of economic impact of the festival is influenced by spectators' musical preferences, is the only hypothesis that is not fully supported.Thus, for example, although the result of the covariance between the latent variables Economic impact and Musical preferences indicates a direct dependence of 0.082, this value is not sufficiently consistent with the expected dependence.However, the covariance of 2.617 between Economic impact and Attitudinal loyalty mitigates the result of the previous covariance.
Lastly, we would like to highlight two aspects.First, the results of this study indicate a good fit of the model.Second, given that the Cordoba Guitar Festival is a major cultural event, it requires greater dissemination.In this regard, as we have already mentioned, we believe it is important to continue research on the festival as a cultural benchmark, as well on the spectators who attend the event.In this sense, the issues currently studied by cultural management [14], in which institutions must be involved, both public and private sector, are issues that must be considered.Moreover, a study on the festival's visibility, especially at the international level, would make an important contribution.

Figure 1 .--
Figure 1.Path diagram of the SEM model