Interaction run in the Regression
BANGLADESH


Return to Previous Page

First, create the interaction term between water and latrine:

1.  Open bdeshc.sav

2.  Click on Transform, Compute in the data editor menu. 

3.  Type in the name wat_lat (this will be the name of your interaction between water and latrine use) in the Target Variable box.

4.  From the variable list on the right, select wathcat and move it to the Numeric Expression box using the arrow button.

5.  Type the symbol * next to the term wathcat, then select latcat from the variable list and move it into the Numeric Expression box.  The Numeric Expression box should now read wathcat*latcat.

6.  Click on OK.

7.  You should check for the new variable at the end of your variable columns, where a variable called wat_lat should now be.

 

Now, an exercise to show how to get the regression results with an interaction term (Bangladesh data):

1.  Open bdeshc.sav

2.  Click on Statistics, Regression,Linear and enter acprvfm (prevalence of low arm circumference males/females) in the Dependent variable box.

3.  Enter the wathcat (high or low prevalence of safe water use) variable into the Independent variable box using the arrow key, and then enter both latcat (high or low prevalence of safe latrine use) and wat_lat (interaction term for water and latrine) as well.

4.  Click on OK

Results of the regression:

ex3_1.jpg (23232 bytes)

ex3_2.jpg (30114 bytes)

ex3_3.jpg (35092 bytes)

INTERPRETATION:  Here, the B coefficient for each of the variables, water source, latrine usage, and the interaction of the two is listed. In this case, the largest change in nutrition status is seen for water source , which has 6.28 points increase for each unit increase in the prevalence of low arm circumference (p=0.000). For safe latrine use, the change is not significant (p=0.939) now that the interaction variable is introduced. The interaction term for water and latrine is significant and accounts for a large coefficient (or change) associated with low AC prevalence (B=6.121 and p=0.010), so a lot of the change with water is collinear with latrine improvement. The message is that it might be more useful to consider improving only water or latrine use when developing a program in order to see the most results for the input, and it might be more effective to select water improvement over latrines.  See if the graph of a two-by-two output confirms these results.

Below the graph shows the interactive effect of water and latrine with low arm circumference as the outcome. This should look familiar from the previous section.

wpe1.jpg (14305 bytes)

INTERPRETATION:  Again, this graph reinforces that an interactive effect was detected in the results of the two-by-two table.   Improving either water source or latrine shows a significantly large effect, but the additional improvement of improving both is negligible. It is an interaction of the two that causes this, and should be included as a term when testing further to determine what to include in an intervention program.  We now know the slopes of these lines are significantly different because the interaction term was significant in the regression equation we just performed (p= 0.010).

Return to Top of Page