Following Sekaran (2003), research design involves a series of rational decision making choices regarding: the purpose of the study (exploratory, descriptive, causal), the type of investigation, the extent of researcher interference, the time horizon and the level to which the data will be analysed (unit of analysis). In addition to the above, the researcher has to make certain decisions regarding the research setting, sampling design, the data collection method, the necessary sample size required and the data analysis
procedures that will be followed. The above topics will be analysed in the following sub-sections.
4.3.1. Purpose of the Study
According to Churchill (1999) and Aaker (1997) research design is categorised as exploratory, descriptive or causal. The main importance in exploratory research is to discover ideas and seek potential alternatives and related variables that should be considered. The role of descriptive research is to provide a true picture of some feature of the market environment. Finally, causal research is used when the researcher is attempting to show that a variable instigates or influences other variables. Even though these three design types seem as different processes, the differences between them are not easily distinct (Churchill, 1999).
The purpose of the present study is to describe consumers’ behaviour towards music acquisition from multiple channels. This study employs the deductive approach in the sense that it uses an existing theory (that of TPB) by testing it in a new context (that of music acquisition in a multi-channel framework). In addition, the theory will be tested by accepting or rejecting hypotheses, while empirical data that can be measured will be collected and analysed.
4.3.2. Research Setting and Unit of Analysis
For the research setting it is sensible to choose a place where music acquisition is observed and measured. The researcher following the approach used by previous studies (Chang, 1998; Cronan and Al-Rafee, 2008; Gopal and Sanders, 2000, 2003;
Wang et al., 2009; Cockrill and Goode, 2012; among many others) believes that universities provide a sensible research setting for the current study. Since one of the
aims is to examine the music acquisition from digital channels, universities provide particular facilities (i.e. high speed Internet connections in the halls and the university premises) that enable users to engage in this act. Also, according to Cheng et al.
(1997), university students are considered the most appropriate target population to test digital piracy. Based on the above, universities will provide the most appropriate sampling setting whereas the unit of analysis will be the undergraduate university student.
Since the empirical analysis will be carried out on student population, the data collection for the quantitative study will be conducted on the university premises (classes, lecture rooms, the refectory etc.). Furthermore, the data collection will be completed at the individual level. Further details about the data collection procedure are given later on, in Chapter 5 (for the calibration study) and in Chapter 6 (for the main empirical study).
4.3.3. Sample Design and Data Collection
Following Churchill (1999), the sampling procedure involves 6 different steps. It starts by defining the population and identifying the relevant sampling frame. Then the sampling procedure and the appropriate sample size must be determined. Finally, the sampling unit must be specified and the collection of the data from the designated elements follows.
The definition of a population includes a complete collection of the people that are being studied in accordance to the research objectives (Aaker, 1997). Identifying the population accurately and precisely is a significant starting point of every research since sampling is supposed to obtain information and make inferences about the population.
In the current research the target population includes consumers of music who reside in two European countries, namely UK and Greece. According to IFPI (2013) Greece is one of the highest EU countries in terms of physical piracy, while the UK has high figures of digital music piracy. Based on the above, the population of this study consists of undergraduate university students in UK and Greek institutions5.
Because of the nature of this study, a population frame cannot be achieved since it is necessary to ensure participants anonymity and confidentiality. Therefore, the application of probability sampling techniques that eliminates biases cannot be obtained. Thus, the study in an attempt to eliminate sampling error adopts a convenience sampling technique asking all participants to fill-in the questionnaire a few minutes before or after a certain lecture/class. Also, particular effort is made to ensure that the sample selection will not be formed on the basis of any judgement.
The data for present study will be collected using a cross sectional questionnaire survey. The survey approach is considered most appropriate technique because is faster, inexpensive, efficient, and can be administered to a relatively large sample (Churchill, 1995, Sekaran, 2003). The cross-sectional study is performed when data are gathered only once from a specific population. Malhotra and Birks (2003) suggest that sectional design can be either single or multiple. This research will use a multiple cross-sectional design by collecting the data from various samples of the population only once.
The sample size decision is made in accordance to the statistical technique that will be used later in the analysis. Although in general large sample sizes tend to produce
5The UK Universities are City University (London) and the University of Lincoln (Lincoln). The Greek Universities are the University of Macedonia (Thessaloniki) and the National and Kapodistrian University of Athens (Athens).
more reliable results, the sample size decision must be based on a number of factors related to the complexity of the model, the expected rate of missing data and the estimation procedure that will be used (Hair et al., 2010). A detailed description of each sample employed and its size will be presented in the following chapters (prior to each of the two empirical studies).
According to Malhotra (2009) there are different methods for data collection in marketing research. Apart from focus groups and in-depth interviews surveys are also popular and widely used. Surveys can be distinguished in two broad categories: non- Internet survey forms and Internet survey methods. The non-Internet surveys can be administered by a variety of techniques, such as door-to-door interviews, mall intercept interviews, telephone interviews, self-administered questionnaires and mail surveys.
Since the population of this study consists of students in UK and Greece (there is a wide geographical distribution of the samples), it would not be economical neither time efficient to conduct face-to-face or telephone interviews. The closest alternative is to use email questionnaires but this choice is not possible for the current research. This is because confidentiality and anonymity are important due to the nature of the research (illegal behaviour). Therefore, since the base of music consumers is widely spread and in the absence of a sampling frame due to the lack of census, the current study adopts the convenience sampling approach as mentioned before.
4.3.4. Necessary Sample Size
Determining the required sample size for an empirical analysis is really important in order to obtain accurate results. The role of sample size is crucial in all statistical analysis (Hair et al., 2010). Obviously the more sophisticated the statistical
analysis the larger the necessary sample size is. Therefore, the sample size requirement in this study was based on the selected statistical analysis technique used in every phase of the quantitative study.
In the first phase of the study, Exploratory Factor Analysis (EFA) is being used through the estimation of the Principal Components Analysis (PCA). For PCA, Kass and Tinsley (1979) suggest that a researcher should use between 5-10 participants per variable, up to a total of 300 participants. This is because they argue that after the threshold of 300 the test parameters tend to become stable, regardless the ratio of participants to variables. Tabachnick and Fidell (2007) suggest as well 300 cases while Comrey and Lee (1992) conclude that a sample of 100 is poor, of 300 is good and of 1000 is excellent. However, more recent studies utilized Monte Carlo simulations and found that in various cases parameter stability can be obtained with less than 300 participants depending on the design of the study. Field (2009) concludes that the best
“rule” is to use a 5:1 ratio of participants for every item. The current research will adopt this guideline in order to determine the necessary sample size for the PCA.
During the second phase of the quantitative analysis, the research uses Confirmatory Factor Analysis (CFA) followed by Structural Equation Modelling (SEM) and multi-group analysis. CFA, SEM and multi-group analysis, like any other statistical technique, require an appropriate sample size in order to obtain reliable estimates (Hair et al., 2010). Hair et al. (2010) suggest that SEM in general requires larger samples compared to other multivariate techniques. Given that larger samples are usually more difficult to obtain, both in terms of effort and cost, the critical question is to identify a sample that will definitely provide trustworthy results. Researchers’ opinions regarding minimum sample sizes are varied (MacCallum et al., 2001). However, there are some
general guidelines that can be applied considering mainly the complexity of the estimated model. In general, Hair et al. (2010) suggest that larger samples are required in cases where: (a) models have a large number of constructs, (b) some constructs have fewer than three measured/indicator variables and (c) multi-group analysis will be later on conducted requiring an adequate sample for each group. The current research falls in cases (a) and (c) mentioned above. Namely, it involves a relatively large number of constructs (nine) and multi-group analysis will be later on conducted in order to check the possible effect of demographics. In cases like this, Hair et al. (2010) suggest a minimum sample size of 500 participants. Thus, since this research takes place in two different countries, the researcher will try to obtain a minimum of at least 250 observations for each country. This sample is deemed adequate both for SEM and for the corresponding multi-group analysis.