Doordan (1998) pointed out that in qualitative research, there were several types of analyses. The most widely used analysis consisted of a narrative description and clas- sification according to preestablished categories. Llanes, 2004 (recited in Hoang Trong and Chu Nguyen Mong Ngoc (2008) pointed out that in qualitative research, data anal- ysis could simply involve careful organization, interviews. In this research, quantita- tive data were processed, analyzed, organized, and presented in categories.
After collecting data, the author analysed using Statistical Package for Social Sciences (SPSS) for quantitative data analysis and category formation and tabulation. Descrip- tive and inferential statistics includes: Cronbach’s alpha, EFA, correlation and regres- sion analysis was used. The analysis process is implemented as follows:
3.5.1 Descriptive Statistics
Survey data were applied descriptive statistical analysis in accordance with demo- graphic variables including firm size, and industry. The answer results of the ques- tionnaire were simultaneously summarized to obtain the primary information of the survey samples.
3.5.2 Factor Naming and Model Modification
After processing EFA with the set of observed variables, it may form other factors that differ from the concepts in the research model proposed in theory. Basing on the content of the observed variables, the author renames the formed observed variables
so that they match the set of research data. Simultaneously, the author modifies the research model and the new research hypothesis so that they match the surveyed data.
3.5.3 Verbal Interpretation
Interpretation of data refers to the task of drawing inferences from the collected facts after an analytical and/or experimental study. Actually, it is a search for the wider meaning of research findings. Therefore, interpretation is the device through which the factors that seem to explain what has been observed by the researcher can be better understood and it also provides a theoretical conception which can serve as a guide for further researches. In this study, the perception of the CEO on E-readiness (POER and PEER) is interpreted as shown in table3.5.
TABLE3.5: Verbal Interpretation of POER anf PEER
Scale Range Choice of descrip. Verbal Interp.
5 4.20 - 5.00 Strongly Agree Excellent (E)
4 3.40 – 4.19 Agree Good (G)
3 2.60 – 3.39 Neither Agree nor Disagree Fair (F)
2 1.80 – 2.59 Disagree Poor (P)
1 1.00 – 1.79 Strongly Disagree Very Poor (VP)
3.5.4 Testing the Hypothesizes
Research hypotheses were tested through actual data from the regression function.
Standards here are based on the corresponding t-test and p-value (Sig. value). The reliability coefficient is 95%, and p-value was directly compared with 0.05 to conclude if the hypothesis to be accepted or rejected. To test the appropriation of data and of the model, R-square, t-test and F-test were used. To evaluate the importance of factors, corresponding beta coefficients in the regression function were used.
The findings are represented using frequency tables, pie charts, and mathematical ta- bles. The e-readiness was calculated and was represented using radar graphs. Then
the regression procedure was used to determine factors affecting the e-readiness level of large and medium enterprises in Thai Nguyen Province.
3.5.5 Regression Analysis
Since the regression model is built based on the least-square summation, it is necessary to verify the hypothesis of OLS method before drawing out statistical conclusions. As- suming that the model of k independent variables is written as follows:
Y = β0+β1X1+β2X2+. . .+βkXk+Ui (3.3)
In which:
• Y: dependent variable (e-readiness level)
• β0: constant
• Xi: independent variables
• βi are angle coefficients - reflecting the level of effect of Xi on the dependent variable Y
• Uiis a random portion, also called as noise, which is the variation part of depen- dent variable Y that suffers additional effect besides that of Xi
To guarantee the model is the best, equation3.3must satisfy the following hypothesis:
Phenomenon of linear relationship among independent variables: This is a situation in which a variable in the multivariate linear regression has a linear relationship with another variable or some other variables in the model via a linear combination. If this phenomenon occurs, the estimation is not appropriate and the model with k variables (k independent variables) might consist of k-1, k-2 or smaller number of variables. To detect this phenomenon, we can use the scatter plot graph. If the graph does not reflect
any trend of the prediction value and the observation value, we can conclude that there is no phenomenon of linear relationship in the model.
Phenomenon of variation in residual variance: When OLS method is used, one im- portant hypothesis is that the residual variances are constant (var (Ui = σ2 with all observation); if any observation shows that value of var (Ui) are different, the model violates the hypothesis that the error variance varies. As a result, it makes the estima- tion of the equation is not the best, and the verification is void (Jujarati, 1995, recited in Nguyen Quang Dong (2013). Methodology of variation of error variance can be the Spearman correlation.
Verification of normal distribution of residues: Estimating by OLS based on the as- sumption that variables have normal distribution. If the variables are not normally distributed, the determined linear form is not accurate and the model might be a loga- rithm function or an exponential function or quadratic function. To verify the normal distribution of the variables, we can use the histogram graph and P-plot to examine.
If the graphs have equal bell shape, we can conclude that the variables are normally distributed.
Model without multi-collinearity: Multi-collinearity is also a phenomenon that inde- pendent variables have a linear relationship leading to magnify the results (collinear- ity) and not be able to separate the individual effect of each factor on the dependent variables. To recognize the multi-collinearity, the variance inflation factor (VIF), in which if VIF is less than 10 we can conclude that multi-collinearity does not affect the conclusions drawn from the regression equation by OLS can be used (Hair et al., 2006, recited in Hoang Trong and Chu Nguyen Mong Ngoc (2008).
Model without auto-correlation: Auto-correlation is the phenomenon that noise items Ui) of different observations have correlations with each other. As a result, it makes the t and F verification in the model not reliable any more (Dagujarati 1995, e-cited in Nguyen Quang Dong (2013). In other words, conclusions drawn from model of
the research hypothesis are not reliable. To detect the auto-correlation, Durbin Watson verification can be used to compare observed valued on the research data with the limit value of dL and dU.
Chapter 4