After the data collection, all the questionnaires were edited to obtain clean and reliable data.
The questionnaires were screened to identify and eliminate incomplete, poorly recorded and inconsistent or ambiguous responses to ensure completeness and accuracy. Since all the questions were pre-coded, the collation of responses for data analysis was made easier.
The data entry and final analysis were carried out by using IBM SPSS Amos (Analysis of Moment Structures) version 22. Covariance-based structural equation model, (in this case Amos) was adopted because it is recognized as an efficient method of analysing not only reflectively but also formatively-measured constructs (Diamantopoulos, 2011; Jarvis, Mackenzie & Podsakoff, 2003; MacCallum & Browne, 1993).
4.11.1 Data Analysis Procedure
Descriptive and inferential statistics were the main statistical tools used in this study.
4.11.2 Descriptive Statistics
It has been established that the purpose of descriptive statistics is not to make predictions but rather to describe and make comparison of the variables in a study. Data generated through the use of descriptive statistics are usually displayed by using graphs and tables, and computation of measures of central tendency and variability (Sheskin, 2004).
In the present study, descriptive statistics were utilized to explain and compare the demographic characteristics, perceptions of packaging and brands of plant medicines purchased by the participants through the use of frequency distribution tables. In addition, cross-tabulation was used to measure the interdependence between levels of participants` education, income and patronage of herbal medicines.
Chi-square test (Х2) was also employed to examine the likelihood that associations among education, income and frequency of patronage of herbal medicines could occur by chance factors. More importantly, it was used to test the statistical significance of the observed interdependence among participants` education, income and patronage of herbal medicines (Malhotra & Birks, 2007). According to Saunders et al. (2007), a probability value of 0.05 or less (p < 0.05) shows that the relationship that exists between the variables is significant. This implies that there is at least 95 percent certainty that the association between the two variables could not have occurred by chance alone.
4.11.3 Inferential Statistics
Sheskin (2004) emphasised that the employment of inferential statistics allows predictions or conclusions to be drawn from the larger population from which the sample was obtained. As a result, inferential statistics was employed through the use of factor analysis and structural equation modelling to test the significance of the hypotheses stated in the research.
4.11.3.1 Structural Equation Modelling
Structural equation modelling (SEM) has been increasingly used across many disciplines by researchers as a vital statistical technique and ingredient for testing theory as well as its building (Babin & Svensson, 2012). In their view, the structural equation modelling is a multivariate analytical tool used for specifying and assessing linear and /or “causal” relationships between multiple independent and dependent variables through a simultaneous multiple equation process. Byrne (2016) pointed out that SEM is a statistical tool that relies on a confirmatory procedure to measure structural theory relating to some phenomena. Hoyle (1995) described structural equation modelling as a statistical technique which offers a comprehensive approach of testing hypothesised relationships among indicators and latent factors.
Weston and Gore (2006) also highlighted that structural equation modelling is made up of two analytical procedures; the measurement model and structural model. The authors are of the view that, in the SEM procedure, the measurement model is the precursor of the structural model. Confirmatory factor analysis (CFA) is often used to analyse the measurement model. It defines the relationship between the indicator items and the respective latent factors that they are purported to measure. On the other hand, the structural model spells out the “causal”
relations among the constructs as proposed in the theory. Essentially, the structural model defines the degree at which specific unobserved factors directly or indirectly influence changes in the estimates of other constructs in the model (Byrne, 2016).
The author further noted that the structural relations are indicated by a set of regression equations and are depicted pictorially to provide better conceptualization of the theory under the study. Thus, structural equation modelling is usually shown by a path diagram. The path diagram provides the summary of the hypothesized relationships among common factors and observed variables as well as the non-directional (correlational) and directional (regression) relationship among the unobserved factors (Bowen & Guo, 2012). In the path diagram, the indicator items are shown by rectangles while common factors are indicated by circles. In addition, the single-headed arrows illustrate the effect of one variable on another, whilst the double-headed arrows explain the correlations among the latent constructs.
Structural equation model analysis presents path coefficients to examine the statistical significance of individual structural paths which indicate the impact of a latent variable on another or a latent variable on an observed variable. The test of statistical significance of path
coefficient is determined by either t-statistic or z-statistic (Schreiber et al., 2006). In this study, the test of statistical significance of the parameter estimates was determined at critical ratio greater than 1.96 at probability value of 0.05 or less, or at critical ratio greater than 2.58 at probability value of 0.001 or less (Timothy & Moore, 2012).
Furthermore, the maximum likelihood estimation method was used when conducting the structural equation modelling in the current study. Hu and Bentler (1999) highlighted that the maximum likelihood estimation procedure is widely used as it produces the best model-fit comparable to other estimation methods. It has also been emphasized that the maximum likelihood method has proven to offer reliable parameter estimates and is robust even under the violation of the assumption of normality (Hair et al., 2010).
The present study relies on structural equation modelling to test the hypothesized relations posited in this research. It is most appropriate for the current research because the focus of the research is to analyse the relationship among multiple latent variables, each of which is represented by several indicators. These multiple constructs could be independent variable in one relationship and at the same time act as dependent in another relationship (Hair et al., 2010). In the current research, packaging, brand equity and its dimensions are latent variables, each is represented by multiple indicator variables. The relationship between packaging and brand equity is mediated by the dimensions of brand equity.
In addition, structural equation modelling was employed because of the following advantages it has over the other conventional methods of analysing multivariate data. Long (1983) emphasized that structural equation modelling is a flexible technique that can deal with a variety of essential statistical applications. Byrne (2016) highlighted that while the conventional multivariate methodologies are unable to either assess or eliminate measurement errors, the SEM procedures offer clear estimates of these error variances in the parameters.
Secondly, structural equation modelling permits the entire model to be tested statistically through a simultaneous analysis to examine the degree at which it is consistent with the data.
The author further argued that the model is deemed plausible if the goodness-of-fit is adequate while the model is rejected if the goodness-of-fit indices are inadequate.
Besides, whereas traditional methods of analysing data is based on the measurement of indicator variables only, SEM models can combine both latent variables and indicator
variables. Moreover, Bowen and Guo (2012) established that SEM permits the simultaneous regression of equations in which the relationship between the exogenous variable and the endogenous variable is hypothesized to be partially or fully explained by mediating variables.
The authors further stated that, unlike traditional regression models, SEM can estimate regression relationships among common factor and between indicator variables and unobserved factors simultaneously. Moreover, as a confirmatory method, SEM provides a comprehensive procedure to evaluate and modify theoretical models which has the potential to further theory development (Anderson & Gerbing, 1988).