Inferential statistics. Inferential studies were done according to results of “Stata 14” software. Before doing regression analysis we had check association, normality and multicoleniarity problems. According to the given data sets our independent variables are unemployment rate, GDP per Capita, people using at least basic drinking water and urban population percentages, and dependent varibale is mortality rate under 5 years old children per 1000 live births (in our data sets this variable given as MOR5 for convinience).
To be more clear, we took natural log from GDP per Capita, people using at least basic drinking water and urban population percentages. Therefore they were given as UNMPL, LNPERCAPGDP, lgPUD, lgURPOP.
Correlation analysis was done by us and found following outcomes which are given in the 5th graph.
Figure 5. Source: research findings
According to the figure 5, it is easy to see that correlation between controlled and control variable and also within control variables. MOR5 has adverse relationship with all explanatory variables. The correlation between MOR5 and UNMPL is equal to -0.2249, correlation of LNPERCAPGDP and MOR5 was equal to -0.3345. For lgPUD and lgURPOP correlations to the MOR5 were equal to -0.6402 and -0.2697 which showed negative impact of them on MOR5.
However, the independent variables have positive and high level of correlations. For example, correlation of UNMPL to LNPERCAPGDP, lgPUD and lgURPOP were 0.7933, 0.5547 and 0.6335 respectively. This situation clarifies that we chose proper data set for our research.
Then we checked multicollinearity problem and the following result was found:
Figure6. Source: research findings
According to the test, if VIF achieves to be equivalent to 1 or at least the outcome should be less than 5, it means that in our model there is no any multicollinearity problem. As it is clear from figure 6, the peak level of VIF was to 6.06 which is also considered as good. Also, as it was given in figure 6, the mean value of VIF was 3.92, which can be said that our model does not have any multicollinearity problem.
In the next test, we have to check is normality problem. This problem was tested with “Stata 14” software and we found following result which was given in figure 7. From the graph, it is clear that our model density follows nearly normality density and we can assume than our model relaxes normality assumption. Thus, we can continue to do regression analysis
Figure7. Source: research findings
We checked three regression models in order to know which model fitted in our model. OLS, Random effect and Fixed effect regression analyses were made and found following result which was given in Figure 8. According to OLS regression result it was found that all variables had negative impact on dependent variable, but R square and Adjusted R square which results were equal to 0.30 and 0.2758, respectively, showed very low result. Since R square and Adjusted R results were so low, we could not accept this model.
Then panel data analyses based on random and fixed effect. Overall R square of Random effect test was higher than the result of Fixed effect test, but coefficients of independent variables of fixed effect test were higher than variable coefficients of random effect. Thus, we can’t conclude which model we have to choose. In order to check which model is best fits in our model we have to test Hausman test using Stata. In this case our hypotheses become following:
H0: Random effect describes better the model
HA: Fixed effect describes better the model
Do'stlaringiz bilan baham: |