Multivariate Lab Exercises

                                          Getting Started

n   Remember again to put on the filter

n   Think about which independent variables you want to use

n   For today we will only look at one dependent variable, so choose the one that you think does
the best job in measuring the issue you are interested in.

n   For example, if you are interested in gender equality, choose the one question that you think
most clearly shows gender attitudes. If you are choosing market-liberalism, think whether less
regulation really is the best question or should you choose a different question.

                         After adding the filter, go to linear regression

                                 Choose your independent variables

   Ask for collinearity diagnostics under ”test” (you can also choose Durbin-Watson and casewise
   diagnostics if you want to test for heteroscedasticity, but consult the textbooks for advise)

The F-statistic shows that the model as a whole is significant, but this is almost ALWAYS the case,
                  so it tells you when the model is bad, but not when it is good

                               Which variable(s) is/are significant?
                    Which Variable explainces the greatest amount of variance?

                               Which variable(s) is/are significant?

                               Is there a problem with Collinearity?

  Nevertheless, SEX and AGE are rather equally distributed on dimensions 3 and 4 which indicates
                               there could be a problem between them

                                             Next step

n  Probably only educational level is a good predictor, so it would be best to use a bivariate
regression or replace SEX and AGE with some other variables

n  But since there might have been a problem of collinearity between AGE and SEX and since AGE had
a much lower std. Coefficient, we will eliminate AGE now and see if SEX become significant

                                           Elminate AGE

                                    Sex is still insignificant

                                         Adjusted R-square

n  Increased now from .005 to .007

n  (Not shown here)

n  So eliminating a variable made the model better

n  But still it cannot explain even 1% of the change in LESSREG

With only EDUCATIONAL LEVEL: it is significant, but the adjusted r-square is only .06, so again it
                           shows we should look for different variables

                                        Make table in Word

                                             Your Task

n  Run a multiple regession with the dependent variable you have chosen and at least 5 independent
variables

n  Eliminate all the variables that are not significant

n  Make sure there is no-collinearity

n  Try to make the best model

n  Make a table in Word