Data analysis

4.3.1 Descriptive statistics and K-Means Cluster Analysis

Collected data was analyzed by using IBM SPSS (Statistical Package for Social Sciences) Sta-tistics 25 -program. At first, all the responses were analyzed by descriptive staSta-tistics for getting an overall image of the respondents. However, as the second research question reveals, seg-mentation of the respondents was one of the points of interest in this study to get a clue who those tourists are interested in silence tourism, in other words, what distinguishes them from the other tourists. Hence, the data was analyzed, and segmentation made by K-means clustering.

“Market segmentation in tourism research has been defined as a process of dividing a market into market segments, which are defined as groups of consumers who are expected to exhibit similar purchasing responses” (Smith in Rid, Ezeuduji & Pröbstl-Haider, 2014, 104). Obvi-ously, the interest in this study is on the tourists who would like to buy similar purchase, in this case, a silence tourism product. Segmentation used in this study is called activity-based seg-mentation.

Cluster analysis aims to group similar observations or variables to the same groups and is used especially in the situations where these can be grouped without knowing a classification crite-rion. Cluster analysis is hence an explorative way to make an analysis. (Metsämuuronen 2005, 812.) This was the case with the data used in this study, as the number of segments were possible and reasonable to determine to be three without knowing the reason why certain respondents will belong to the certain group, for example to the “Silence tourists”. The idea behind three segments was to divide tourists roughly enough to the groups which were seen as reasonable;

to the ones who are very interested in silence tourism, to the others who are somewhat interested and to a third segment which members are not interested in silence tourism, which relates to Beane’s & Ennis’ idea about using market segmentation for evaluating new opportunities to tourism products (in Rid et. al. 2014, 104). With the help of these three segments it can be evaluated if there are opportunities for silence tourism products. Due to the relatively big data

set (N=471), K-means Cluster Analysis was chosen instead of Hierarchical Cluster Analysis.

So, following two questions from the survey of this study were used for K-means clustering:

1. How interested would you be at this moment to travel to a tourism destination providing the described silent tourism services?

2. If you had a chance, how likely would you purchase a holiday in a destination where you could participate in silence tourism experiences?

Table 2. Forming of clusters using two selected questions.

Using K-means clustering, mentioned three segments were formed. However, SPSS presents the clusters as numbers, so it was needed to rename them. Therefore, clusters 1, 2 and 3 were renamed accordingly: 1 = Silence tourists, representing the ones who are very interested in silence tourism, 2 = Potential silence tourists, representing those who are somewhat interested in silence tourism, and 3 = Not interested in silence tourism (Table 3). Later on, profiles of mentioned segments are compared by crosstabulation and mean value.

4.3.2 Crosstabulation

After forming three clusters by K-means cluster analysis, the responses of each cluster were analyzed by using different analysis methods. At first, responses to the questions regarding clusters’ sociodemographic variables, background information and travel behaviour were com-pared by crosstabulation. In addition, responses regarding question three about silence tourism expectations, question four regarding interest in buying nature tourism services based on silence

experiences, and question seven regarding type of destination a respondent is interested in tra-velling at the moment were analyzed by crosstabulation as well. With the help of crosstabulation, a dependence between two or more variables can be illustrated (Metsämuuronen, 2005, p. 531).

4.3.3 Comparison of means

Responses to question six about holiday travel behaviour were analyzed by comparing means among the clusters. Comparison of means analyses in SPSS program can produce even 12 sta-tistics of which simple means compare is only one (Griffith, 2010, p. 229), and the one chosen to be used in this study. Chosen comparison of means analysis includes mean and standard deviation.

4.3.4 Multinomial logistic regression analysis

Multinomial logistic regression analysis was used to analyze responses to question five about travel motivations, in more detailed, how motivations affect to an interest toward silence tour-ism. Regression analysis, in general, is about predicting the unknown future based on collected data from the past (Griffith 2010, 240), which describes this study, as the data has been collected from respondents’ previous travel behavior. In logistic regression analysis, Dependent may be Binary Logistic which gets two values, or Multinomial Logistic, which gets more than two val-ues (Metsämuuronen, 2005, p. 701). Multinomial logistic regression analysis is used, according to its name, for predicting a membership of several categories (Field 2014). As there are several motivation factors in the survey of this study, a multinomial logistic regression analysis was chosen to analyze how the motivations affect to an interest toward silence tourism. In other words, how the motivations explain to which cluster (“Silence tourists”, “Potential silence tour-ists, or “Not interested in silence tourism”) is the one a respondent represents. Logistic regres-sion analysis aims to find best variables among several variables, in this case, motivations, that explain the phenomenon best. With the help of logistic regression analysis, it can be tested if some of the variables explain the results better than the others. Logistic regression analysis can be used when variables chosen to the analysis are relevant. Choosing irrelevant variables makes the results unreliable. (Metsämuuronen, 2005, p. 687-688.) In this study, motivation factors were chosen based on previous literature for ensuring they are relevant. However, as logistic

regression analysis tells just the association between variables, it must be kept in a mind that any of the motivation factors is not necessarily a reason for a respondent belonging to a certain cluster (Metsämuuronen, 2005, p. 688-689). In other word, there might be an association be-tween a single motivation factor and some of the clusters, but it does not say that the motivation factor is a reason why a respondent with the certain motivation belongs to the certain cluster.

When using logistic regression analysis, amount of responses must be sufficient to make sure there are differences between the clusters, and that there are enough responses compared to the variables, in this case, to motivation factors. Otherwise there is a danger of getting Independent variable for each of the responses, which is not a purpose. (Metsämuuronen, 2005, p. 688-689.) Mentioned problem did not occur in this study.