Classical Test Assumptions Multiple Regression

• Share
• Tweet

Classical assumptions necessary testing to determine if the results of the regression estimates made completely free of the symptoms of heteroscedasticity, multicollinearity symptoms, and symptoms of autocorrelation. Regression model will be used as a tool that does not bias estimates if compliant BLUE (Best Linear Unbiased Estimator) that there is no heteroskedastistas, there are no multicollinearity, and there is no autocorrelation (Sudrajat 1988: 164). If there is heteroscedasticity, the variance is not constant so that it can lead to biased standard errors. If there is multicollinearity, it will be difficult to isolate the individual effects of variables, so the level of significance of regression coefficients to be low. With the resulting autocorrelation estimator is still biased and still remain consistent only be inefficient. Therefore, the assumption of classical test needs to be done. Tests performed are as follows:

Multicollinearity 1.Uji Classical Assumptions

Heteroscedasticity test aims to test whether the regression model and residual variance inequality occurred one observation to another observation. if the residual variance from one observation to another observation remains, it is called and if different homoskedastisitas called heteroscedasticity. A good regression model is not the case that homoskedastisitas or heteroscedasticity.

Heteroscedasticity test performed using Glejser test, conducted by meregresikan absolute value of residuals obtained from regression models as dependent variables against all independent variables in the regression model. If the value of the regression coefficients of each independent variable in the regression model was not statistically significant, it can be concluded not happen heteroscedasticity (Sumodiningrat. 2001: 271).

2.Uji Heteroskedasitisitas Classical Assumptions

Multicollinearity test aims to test whether the regression model found no correlation between the independent variables (independent). In the regression model should not be a good correlation between the independent variables. Multicollinearity test done by looking at the value of tolerance and variance inflation factor (VIF) of the results of the analysis using SPSS. If the tolerance value higher than 0.10 or VIF is less than 10 then we can conclude there is not multicollinearity (Santoso. 2002: 206).

Autocorrelation 3.Uji Classical Assumptions

Autocorrelation test aimed at testing whether a linear regression model is no correlation between bullies error in period t with an error in period t-1 (before). If there is no problem of correlation is called autocorrelation. A good regression model is free from autocorrelation regression autocorrelation tests performed using the Durbin-Watson test (DW), with a confidence level  = 5%. If DW lies between -2 to +2 then there is no autocorrelation (Santoso. 2002: 219)