CHAPTER 3 METHODOLOGY
3.6 Data Analysis
3.6.3 Scale Measurement
Scale of measurement use to measure the variables in statistic or data in the qualitative research. Based on the research, measurement in the questionnaire is distributed in three urban areas of Malaysia which are located at Kuala Lumpur, Johor and Penang. It is included four different types or levels which is nominal, ordinal, interval and ratio scale of measurement (Sawamura, Morishita and Ishigooka, 2014). For nominal scale of measurement, it measure the categorical variables which is cannot be ranked. Gender, religion, states and age are variables that under nominal scale of measurement. Ordinary scale of measurement is to measure the variables that can be order or rank, each answer in ordinary scale has their own unique meaning. Interval and ratio scale of measurement are quite similar, interval scale of measurement are used to compare the interval in the same state but the zero point in ratio scale are meaningful but not in interval scale of measurement.
Normality is an important model that allows us to determine whether the random variables is normally distributed or non-normally distributed and to calculate the probability of a normal distribution of the underlying random variables at the data set (Ghasemi and Zahediasl, 2012). It is because many processing in observation data are from normality test to make the statistical analysis much easier so the reality presented accurate and reliable conclusions. The small sample sizes mostly can pass the normality test due to normality test have little power to reject null hypothesis when the sample
Undergraduate Research Project Page 45 of 135 Faculty of Business and Finance
sizes is small (Ghasemi and Zahediasl, 2012). Also, the significant result will have small deviation when the sample sizes is large but the result of the parametric test will not affected by this small deviation.
22.214.171.124 Pilot test
The pilot test is considered as a pre study on the research data. It is to measure of internal consistency in a survey or questionnaire form with a scale to determine whether it is reliable or not. In simple way this test is to understand whether the questions in the questionnaire or survey are all reliably measure the same latent variables. Furthermore, pilot testing is also considered as a session or two before the real test which helps fine-tune usability studies, leads to more reliable results. It provides an opportunity to validate the wording of the tasks, understand the time necessary for the session, and, if all goes well, may even supply an additional data point for the study conducted (Schade, 2015). According to Kasunic (2004), there is a structured approach to lead to the efficient on pilot study such as planning, training, monitoring, evaluating and lastly recommendation. In conducting pilot study, it may become unforeseen difficulties for researchers as omitting step and constraining time often happen, (Hassan, Schattner and Mazza, 2006).
Multicollinearity problem is one of the major problem and common used in the regression model, it will occur when the independent variable (X) are correlated with another independent variables in the same model (Yoo, Mayberry, Bae, Singh, He and Lillard Jr, 2015). It will make the estimation become more sensitive if having a small change in the model. This problem may difficult to show which independent variables are affecting the
Undergraduate Research Project Page 46 of 135 Faculty of Business and Finance
dependent variable due to independent are highly correlated with one another. However, multicollinearity only affects the calculations associated with the individual predictions and will not reduce the predictive power or reliability of the overall model because it misleads inflate the large amount of standard error (Duzan and Shariff, 2016). There are three different ways to detect the multicollinearity problem. Firstly, the higher R-square with the few significant ratios will brings this model to multicollinearity problem because R-square is to detect how many independent variables are correlated with another variables. Secondary, this problem will occur when the two independent variables on that model are high pair-wise correlated coefficient with each other. Lastly, when the model having high amount of R-square the amount of variance inflation factor more than 10 and the tolerance amount is to zero.
Autocorrelation is the problem that when the observation’s error term are correlated with the error term from other observation between two different time series which is one is from the original form and another one is lagged one or more than one periods from the linear regression model (Chen, 2016).
The amount of variables will influenced by its own historical data, a positive correlation when increase the value in one time series it will leads to increase the value in another time series of the same variables. There are two types of autocorrelation problem, the pure autocorrelation and impure autocorrelation. Impure autocorrelation happen when the specification error that can be omit or correlated with other variables, pure autocorrelation is the error term are cannot be change by the researchers, it is the true specification from the equation. Durbin-Watson test use to detect the pure autocorrelation problem because it is a method that easy to calculate and understand the problem (Chen, 2016). Besides, Breusch-Godfrey LM test is
Undergraduate Research Project Page 47 of 135 Faculty of Business and Finance
use to further confirm the autocorrelation problem after reject the null hypothesis in the Durbin-Watson test.
126.96.36.199 Specific Bias
Specific bias also known as test bias which caused by cultural bias, construct bias or method bias. If test bias occurred, the results conducted will not be accurate and have some sort of bias.
In the nutshell, this chapter has provided clear explanation for the research design, data collection, sampling design, research instrument, data processing and data analysis. However, several tests use to conduct the test the relationship in between the dependent and independent variables which included Normality Test (Jarque-Bera Test), Pilot Test (SPSS), Multicollinearity correlation Analysis, Autocorrelation (Breusch Godfrey LM Test), Model Specification Error (Ramsey-Reset Test), Individual T-test, and Overall Significant F-Test. Furthermore, the following of the chapter will reveal on the empirical results for this study.
Undergraduate Research Project Page 48 of 135 Faculty of Business and Finance