CHAPTER 3 RESEARCH METHODOLOGY
3.5 Data analysis methods
information from the candidate is mishandled, it may lead to the distortion of the original meaning (Feld & Frey, 2007). In the current study, the interview was not the main method of data collection, as it was intended to supplement the quantitative analysis with additional information. This is more especially so in a case where the observed research pattern, from quantitative analysis, required explanations for more understanding (Gialdino, 2009). Interviewees were extracted among small business owners, who used Electronic Fiscal Devices in their activities. Interview questions were not necessarily uniform.
The study chose respondents in this category purposively; a maximum of ten (10) respondents were chosen based on their availability and readiness. Therefore, the study used the combination of purposive sampling and convenient sampling in obtaining the candidates for interview. Selected respondents are those who were available at the Arusha tax centre during the day scheduled for the interview. The principal researcher conducted the interview. Table 3-5 summarised the sample for interview and selection criteria.
Table 3-5: Interview selection criteria
Selection criteria Description or selection criteria
4 respondents Age below 40 years (1male, 1 female), Age above 40 years (1 male, 1 female)
4 respondents 2 graduate, 2 non- graduate
2 business experience One with 3 years and below, and one with 4 years and above
Source: (Author’s Design, 2020)
The literature emphasises the importance of not ignoring outliers, because, they may have a significant impact on the result of the analysis (Meyers, 2005; Goldkuhl, 2012).
In analysis, the current study used descriptive statistics to describe the main features of the sample. Generally, descriptive statistics are important because they assist the study to present data in a way which is easy to understand and to draw basic meanings (Wilson, 2010; Mannion, Blenkinsopp, & Powell, 2018). The current study used descriptive statistics to know how their frequencies are spread across the sample, more especially, among demographic variables. This is important to understand the perception of the sample to a given element of the study (Gray, 2014). The following descriptive statistics were used: frequency, per cent, and mean. The study presented the descriptive information through tables, and the pie chart.
On the other hand, the study used inferential statistics namely the One Way ANOVA to understand the categorical relationships among demographic variables and tax compliance.
Studies by Kelliher (2005) and Makombe (2017) suggest the use of the One Way ANOVA where there is a need to identify whether two studied groups are significantly related. The One Way ANOVA is more useful where data are on a Likert scale (Collins, 2010); this position was observed in the current study. The use of the One Way ANOVA was supported by descriptive statistics such as the use of cross tabulation or group means. In the case where binary data (or any other clear categorical information) was involved in the analysis, the study used the chi-square model to determine the significance of two studied categories of data (Makombe, 2017).
In addition, the study used inferential statistics to test the impact of independent variables to respective dependent variables. In this case, the study used models such as ordinal regression, ANOVA and the chi-square for testing hypotheses. The Ordinal Regression Model is used practically for predicting a relationship where the dependent variable is ordinal (Kelliher, 2005; Krauss, 2005). The current study used this model because all predicted variables used an ordinal (Likert) scale. Therefore, this is the main model for decision-making in the current study. Basically, the model is symbolically represented by through equation 3.1.
Equation 3-1: The general regression equation
In the equation, represents any dependent variable under study, and represents the intercept value. In addition, represent the coefficient values for respective independent variables . In this case, these symbols are simply used to describe the equation; the relationships of this study are strictly represented through symbols presented in Table 3-6. Furthermore, the is known as the stochastic error term, and it represents all variations in which cannot be explained through the values of engaged predictor variables.
Table 3-6: Variables of the study with their symbols
Variable Symbol
The perceived level of tax compliance TC The perceived fairness of tax procedure PF The perceived effectiveness of tax auditing ETA
The rate of EFD use RATE
The perceived level of punishment due to non- compliance
PP
The perceived level of transparency LT The probability of being reported by others or the fear of whistle-blowers
PR
Source: (Author’s Design, 2020)
The next part of this section explains the analysis of objectives of the study identified in section 1.3.
The first objective (section 1.3) intended to determine whether the fear of whistle-blowers affects the rate of using EFD. Hypothetically, the rate of using EFD is a dependent variable, while the fear of whistle-blowers is an independent variable. Equation 3-2 summarises this relationship.
Equation 3-2: Rate of EFD use and fear of whistle-blowers
In the second objective, the study determined whether the perceived level of punishment due to non-compliance impacts the rate of EFD use. Based on this objective, the perceived rate of EFD use was hypothetically the dependent variable, and the perceived level of punishment was the independent variable. Equation 3.3 summarises the hypothetical relationship.
Equation 3-3: Rate of EFD use and the perceived level of punishment
Moreover, in other to achieve the third objective, the study analysed the relationship between EFD use (RATE) and three other variables. In this case, the EFD use acted as an independent variable. The three dependent variables are: the perceived effectiveness of tax auditing (ETA), the perceived level of transparency (LT) and the perceived procedure fairness (PF). These variables are traditional determinants of tax compliance extracted from different models (Fagariba, 2016; Innocenti & Rablen, 2017); the interest of the study was to evaluate the impact of EFD use rate to their positions, individually. Each relationship was tested independently. Equations 3-4 to 3-6 summarise the relationships.
Equation 3-4: Impact of rate of EFD uses on perceived effectiveness of auditing
Equation 3-5: Impact of rate of EFD uses on perceived fairness of tax procedures
Equation 3-6: Impact of rate of EFD uses on the perceived level of transparency
To analyse the fourth objective of this study, the perceived level of tax compliance (TC) is hypothetically determined through the perceived effectiveness of tax auditing (ETA), the perceived fairness of tax procedures (PF), the perceived level of transparency (LT) and the
perceived rate of EFD use. This relationship is symbolically represented through equation 3.7.
Equation 3-7: The perceived level of tax compliance
3.5.1 Multicollinearity Testing
In statistics, multicollinearity occurs in a regression model when independent variables of the study are correlated. Multicollinearity is a problem for regression analysis because independent variables are expected to be independent of each other. Therefore, the problem is significant where the correlation between independent variables is very high. The study by Wilson (2010) suggests that if independent variables are highly correlated, a change in one variable impacts the other, hence affecting the whole intention of studying the impact of independent variables differently. This is the reason why the highly correlated independent variables are not suitable for regression analysis. Knowing the impact of multicollinearity to regression analysis, the study conducted a test to justify the eligibility of current data and relations to proposed analysis. The study ignored equations 3-2 to 3-6 because each had one independent variable. As for equation 3-7, independent variables were the perceived fairness, and the perceived transparency, the perceived audit effectiveness and the perceived rate of EFD uses; Table 3-7 provides details. In this case the observed highest Variance Inflation Factor (VIF) was 1.050. The observed VIF is low, therefore the independent variables are free from being affected by the multicollinearity effect (Taylor & Medina, 2013). A VIF above 5.0 is unacceptable (Sobh & Perry, 2006).
Table 3-7: Multicollinearity coefficients of equation 3-2
Model Collinearity Statistics
Tolerance VIF 1
Rate of EFD Use .985 1.015
Fairness of Tax Procedures .967 1.034
Level of Transparency .953 1.050
a. Dependent Variable: EFD Support in Tax Audit Source: (Author’s Design, 2020)