Marketing Analysis Project

Regression

· Run regression with X11 to X21 as Ind. Var. and X22 as Dep. Var.

· Problem: X14, X17, X18 and X19 have negative coefficients. Of these, X17 and X18 are significant. This is a problem.

· This problem is due to multicollinearity (P 332).

How to minimize or overcome the effects of multicollinearity?

Method 1

· Use stepwise regression: In the estimation use stepwise and then forward procedure.

· See where the negative coefficient appears. This happens at step 4. So you stop at stop 3 before the model is contaminated.

· But this has drawbacks. You are using only three of 11 variables. The information from eight variables is excluded.

Method 2 (use in class)

The other solution is to use Constructs (not raw variables) as independent variables. For example, x15, x18, x20 relate to food quality . So create a construct called food quality.

Use procedure Correlation to determine variables that are highly related.

How to develop a construct called food quality ?

Use the following sequence in MYSTAT: Data > Transform > LET (complete the expression)

Foodquality = (x15+x18+x20)/3.

This will create a new variable called Foodquality (no space between food and quality).

Similarly develop other constructs. For example variables 12, 19 and 21 refer to service quality. So create this construct. To identify constructs use procedure correlation and combine highly correlated variables.

Save the data file in your Q drive or a flash drive (not in classes folder).

Now run the regression again with X22 as DV and newly developed constructs as independent variables. Variables that cannot be combined can be used as raw variables.

You will see that the negative coefficients have disappeared.

Now you repeat the same for life style section.