The classical assumption test is a statistical requirement that must be met in both simple and multiple linear regression analysis based on ordinary least squares (OLS). Regression analysis not based on OLS does not require the classical assumption requirement. An example of regression analysis that is not based on OLS is logistic regression. Simple linear regression analysis is a regression analysis where there is only one independent variable/explanatory, while the explanatory variable in multiple linear regression analysis is more than one. Not all classical tests need to be performed in linear regression analysis, for example, multicollinearity tests do not need to be performed in simple linear regression analysis and autocorrelation tests do not need to be applied to cross-sectional data.

Types of classical assumption test:

1. Normality test

The normality test is used to determine whether the residual value is normally distributed or not. A good regression model will have normally distributed residuals. Therefore, the normality test was not carried out for each of the variables studied, but was carried out for the residual value. The error that often occurs is that the normality test is carried out on each variable studied, not on the residual value.

Normality test can be done by histogram test, P-Plot normal test, Chi-square test, Skewness and Kurtosis test or Kolmogorov-Smirnov test.

If the results of the residual test show that the data is not normally distributed, there are several ways that can be done, namely data transformation, trimming outlier data, or adding observation data. Data transformation is changing data into the form of natural logarithms, square roots, inverses, or other forms, depending on the normal shape of the curve, whether left, right, center, or right and left.

2. Multicollinearity test

The multicollinearity test was used to determine whether there was a high correlation between the independent variables in the multiple linear regression model. If there is a high correlation between the independent variables, the relationship between the independent variable and the dependent variable is disturbed.

Statistical tools that are usually used to test multicollinearity are variance inflation factor (VIF) or Pearson correlation between independent variables.

3. Heteroscedasticity Test

The heteroscedasticity test will check whether there is a difference between one residue and another observation. One regression model that satisfies the requirement is that there is a similarity in the variance between the residues of one observation and another which is called homoscedasticity

Proof of heteroscedasticity can be made using the scatterplot method by plotting the predictive value with the residual value. A good model is when the chart does not contain a certain pattern, such as converging in the middle, narrowing and enlarging or vice versa

The statistical test used is the Glejser test, Park test or Wei test.

Some solutions if the model violates the assumption of heteroscedasticity is to transform the data into logarithmic form. But the transformation can only be done if all the data is positive. Another solution is that all variables can be divided by variables that have heteroscedasticity disorders.

4. Autocorrelation test

Autocorrelation test is used to see whether there is a correlation between the data in period t and the previous period (t -1).

An example is data on the inflation rate for a particular month, for example inflation in February is affected by the inflation rate in January. In other words, the model has an autocorrelation problem.

The autocorrelation test is only carried out on time series data and does not need to be carried out on cross section data such as in a questionnaire where the measurement of all variables is carried out simultaneously at the same time.

Several statistical tests that are often used to test autocorrelation are the Durbin-Watson test, the run test. If the observation data is more than 100 observations, it is better to use the Lagrange Multiplier test.

Some ways to solve the autocorrelation problem are to change the data or convert the regression model into a general difference equation. In addition, it can also be done by entering the lag variable from the dependent variable into one of the independent variables, so that the observation data is reduced by 1.

5. Linearity Test

Linearity test is used to see whether the model built has a linear relationship or not. This test is rarely used in some studies because the model is usually built on the basis of theoretical studies that the relationship between the independent variable and the dependent variable is linear. The relationship between variables that are theoretically not linearly related cannot be analyzed by linear regression, such as the problem of elasticity.

Linearity tests can be performed using the Durbin-Watson test, the Ramsey test or the Lagrange Multiplier test.