Multi-way Analysis Details: Sanitation's Effect on Nutritional Status

This should feel somewhat old hat by now.  We will develop a step-by-step regression model including the variables of interest one at a time, then additively to observe how the coefficients and significance levels change.  This time, we will also use an interaction variable instead of breaking the analysis into sub-groups.   Because we have previously looked at the regression for NOTOILET alone, and saw that it was large and significant (B= -0.255, p= 0.044) in a model with the outcome, nutrition status.  Now take a look at a model with a sanitation variable (NOTOILET), an education variable (DLOWEDN), and the interaction variable for the two:

1.  Open keast4j.sav

2.  Click on Statistics, Regression, Linear...

3.  Select the variable waz from the variable list and use the arrow key to place it in the Dependent variable box.

4.  Select notoilet, dlowedn, and edn_san from the variable list on the left and place them into the Independent(s) box, one-by-one, using the arrow key.

5.  Select the Method: Enter, and then click on OK.

INTERPRETATION:

As compared to the previous results with just toilet access in the model, this model shows that there is now actually a larger effect from sanitation when education is included in the model after controlling for the interaction between the two.   The interaction variable is highly significant (p= 0.006), and therefore is necessary to off-set the different effects at different levels of the independent variables.  With this resulting regression equation, it becomes clear that there is an effect from improving toilet access in the higher education group (DLOWEDN=0) but not in the low education group (DLOWEDN=1). The regression here just makes it clear that toilet access is significant above and beyond education, but only after interaction is considered.  The best step to take here is to graph the results to see what they really mean.  This will be shown in the main text in the next step.

Another question you might ask yourself now is, "Does this effect persist even after considering SES such as income (VINCOME)?"

Try running this same model now, but include the variable VINCOME to see how it changes the coefficient for NOTOILET.

1.  Open keast4j.sav

2.  Click on Statistics, Regression, Linear...

3.  Select the variable waz from the variable list and use the arrow key to place it in the Dependent variable box.

4.  Select notoilet, dlowedn, edn_san, and vincome from the variable list on the left and place them into the Independent(s) box, one-by-one, using the arrow key.

5.  Select the Method: Enter, and then click on OK.

INTERPRETATION:

As you can see, there was not a dramatic change in the model when income was added.   Still, there is a meaningful effect from the toilet variable (B= -0.766, p=0.011), education and the interaction variable are still significant.  There were actually larger changes in the coefficients for the interaction variable and for education than there was for toilet access.  This continues to support that sanitation does have an effect on nutrition status after controlling for interaction with education.  For a full explanation, return to the main text.