Regression Outcome

Return to Previous Page

First, the results for measles alone.

1.  Open keast4j.sav

2. Click on Statistics, Regression, Linear.

3.  Place waz in the Dependent Variable box and hmeas and age in the Independent(s) box and choose the Method, Enter.

4.  Click on OK.


wpeD.jpg (7797 bytes)


wpeF.jpg (12502 bytes)

INTERPRETATION: The results of the first regression are surprising, at least initially. The variable for measles immunization was entered (yes or no) and the outcome variable is WAZ score. Measles is significant, but in the direction opposite of what is expected. It appears here that those with measles immunization have significantly lower nutrition status (0.513 point decrease in percent with measles for each unit decrease in WAZ score).

What could cause this? Well, as previously discussed, it likely has to do with the effect of age on nutrition status as well as the effect of age on measles immunization. Look below at the model with both age and measles. (Run the same model as above, but include the continuous   independent variable age in the model also.)

                                    wpe10.jpg (9109 bytes)

                                wpe13.jpg (14886 bytes)

What considerations should be made for the child's age? Does a child's age not affect both the likelihood of immunization and the outcome of nutrition status? It has been seen that as children grow older and are weaned from breast milk and exposed to more pathogens, they falter in their growth. Now that age is in the regression equation, see if it changes the size of the effect for measles immunization. If it is both significant in the model for nutrition status and it changes the effect for measles, it could be confounding the results. It does appear to decrease the size of the effect for measles, by decreasing the coefficient. So, in a sense this has controlled for the effect of age and measured the effect of measles on WAZ separately…or has it?

The problem here is that the INTERACTION has not been tested between age and measles immunization. One RULE to remember is that before Confounding can be tested, interaction must first be detected. Here is how to add the interaction variable.

1.  Open keast4j.sav

2.  Select Transform, Compute and name the new Target Variable as meas_age

3.  In the Numeric Expression box type enter the variables hmeasyn and age with the arrow key and place an asterisk between the two to make the equation read:     hmeasyn*age

4. Click on OK.

This will add the interaction variable for measles and age to test for interaction in the model. Now continue by creating an equation with measles, age and the measles/age interaction variable:

1.  Open keast4j.sav

2.  Click on Statistics, Regression, Linear.

3.  Enter waz as the Dependent Variable

4.  Enter hmeasyn, age, and meas_age as the Independent Variables

5.  Select the Enter Method and click on OK.

See the results here:

                                                            wpe14.jpg (8884 bytes)

                                           wpe16.jpg (8383 bytes)

                          wpe19.jpg (13738 bytes)



As suspected, there is interaction between measles and age, and now that we have a look there is probably confounding of age as well. We suspect this because even now that the interaction variable has been accounted for, there is still a negative effect of measles immuniztion on nutrition status (age is suspiciosly looking like a confounder).The procedure now is to leave the significant interaction variable in the model and continue to look for a ways to control for confounding. In our next section, we will do just that.

Return to Top of Page