Forming the data sets

The chosen method for gathering empirical data is a web-based survey. Since there are plenty of variables affecting media choices, a lengthy questionnaire is needed in order to find out interdependencies between the variables. There were 52 questions in the questionnaire (appendix 2) proving 190 variables. There were questions related to media usage and preferences in general and questions about demographics, but the main part of the questionnaire was dedicated to the last media usage event. The respondents were asked to ponder which media products they considered, media usage motives, situations, habits, and the way they reached their decision. Due to the vast number of variables and the length of the survey, the questionnaire was split. Figure 10 describes how the data set was formed.

Figure 10. Description of empirical design of this study

.',*

.+-Splitting the questionnaire is not unheard of in quantitative research; see more about Split Questionnaire Design in Chipperfield et al. (2015) or Gonzalez (2012).

The questionnaire was divided into two parts, namely “The choice of media to read”

and “The choice of media to watch” (see Figure 10). After the split, both questionnaires had 48 questions. The split was necessary in order to get the questionnaires even remotely respondent-friendly, but it was problematic, too, for two reasons. Firstly, it meant that the interrelations on questions that were only in one questionnaire but not the other one could not be analyzed. For some variables, this was not a problem, since they were only descriptive, and no correlations were needed. Secondly, the number of respondents per question were also split, of course.

Again, for some variables, this was not a problem due to the nature of the question.

For example, in the question about habitual media usage, it was assumed that many people have media habits, therefore, there would be enough answers to habit-related questions in only one of the questionnaires. In order to overcome these problems, some questions were the same in both questionnaires, some varied. Some variables were left out due to the length of the questionnaire. For example, it was decided that the forming of the expectations would not be tested empirically, due to a great number of questions it would have required. Some variables were excluded because they were difficult to measure in a survey—for example, how to measure the amount of information one has. The questionnaires were combined afterwards making one pretty large database. There were 336 acceptably completed questionnaires in total.

Some answers were very incomplete and therefore they were not included. Three people used email, and even though it is media, it is so different that those respondents were excluded.

The data was gathered during summer 2014. To provide an incentive for participation, a 50€ prize was drawn among the participants. The questionnaires and sweepstakes were marketed on the Internet and on Facebook. Because of the inadequate number of responses (probably due to the length of the questionnaire), the questionnaire was further marketed in the Tampere University’s doctoral student email list and an invitation was sent to Miratio’s (a marketing research company) mailing list. Later an additional data set was gathered, themed” Facebook usage”, containing many of the same questions but also some new ones. This new data set was needed in order to get more reliable results (more data per question). The additional data set was gathered in fall 2016. Even though the sample is collected from several sources, it represents Finnish people rather well by age, living area, and education (see Appendix 1, Table 17). The respondents were more educated than

people on average, according to Statistics Finland data. Younger adults were also overrepresented in the study. The clear difference is that there are many more women among the respondents than in the population on average. This bias has been solved by analyzing all the results by sex and reporting if there are significant differences. Sex mattered only in very few questions.

In addition to the split, the questionnaires were customized for each respondent according to their answers in order to provide only relevant questions for each respondent. Therefore, the number of respondents for each question varies a lot.

This feature creates the phenomenon that significance levels for each question vary according to the number of respondents. Significance levels are marked by * for one sided ANOVA and ** for two sided (p<0.05). The results were considered as statistically significant using the 5% probability level. Many questions are formed with 5-step Likert scales and analyzed with Sperman correlation, which is applicable for ordinal scales. Even though there is some missing information (people did not answer all questions), it has not been patched with averages as Metsämuuronen (2002) suggests. The analysis has been done with those who answered the question, and in the case of comparing two or more questions, the analysis has been done with those who have answered both questions.

The respondents were advised to choose one recent media usage situation which they remembered well: “Choose one of the listed media usage situations you have had quite recently, and you remember well. There will be more questions about that specific media usage situation later”. Respondents chose the media from a list of media groups (see figure 11) and provided details of the chosen media in an open question (channel and program, name of the newspaper, etc.). The open questions were used to control the classification of media groups. Since many people mentioned watching YouTube or reading Facebook in the open questions, they were coded as alternatives in the study even though they were not in the original list. Figure 11 illustrates the alternatives on their own in the study. After the data was collected, the questionnaires were combined into a one data set.

Figure 11. The media “groups” used in this study

The alternatives are called brands later on in this study, even though there were only two actual brand names in the research (Facebook, YouTube). Calling newspapers, netpapers, or television a brand might feel confusing for some readers.

However, there are several reasons for this practice. Firstly, it is believed that everything has a brand. That is, every person, product, task, or idea has a brand (see section 2.4.6). Since brand is an idea of product qualities in every person’s mind, even the media groups have brands. Secondly, treating media groups as brands allows us to test the bond people have with the media groups by using concepts familiar from brand research.