**Two Way Anova
Revisited**

The GLM can be run for each pair of independent variables with the outcome as WAZ score, but should be run for only the suspicious or documented interactions, such as measles and age. All possible combinations are shown below:

There are 10 separate combinations that __could
__be tested, but not all are reasonable options:

1 | Age and low education | not likely |

2 | Age and Measles immunization | YES, test this |

3 | Age and respondent’s height | not likely |

4 | Age and sex | not likely |

5 | Low education and immunization | POSSIBLY- test this |

6 | Low education and respondent’s height | POSSIBLY- test this |

7 | Low education and sex | not likely |

8 | Measles immunization and respondent’s height | POSSIBLY- test this |

9 | Measles immunization and sex | not likely |

10 | Respondent’s height and sex | not likely |

The steps to using GLM are as follows:

- Open keast4j.sav
- Click on Statistics, General Linear Model, GLM General Factorial (first choice)
- Place waz in the Dependent Variable box and age and dlowed (low education yes/no) in the Fixed Factor box.
- Click on the Model button and select Full Factorial and click Continue and OK.

The full factorial model automatically computes the interaction term. Each of the following models displays pairs of variables marked for testing with the GLM full factorial method.

**AGE and MEASLES IMMUNIZATIONS**

**LOW EDUCATION and MEASLES
IMMUNIZATION**

**LOW EDUCATION and RESPONDENTS
HEIGHT**

* *

**MEASLES IMMUNIZATION and
RESPONDENT’S HEIGHT**

Those ANOVA’s of interest include are those with significant interaction variables AND strong suspicion or previous documentation from the literature. If these criteria are met, the interaction variables will need to be entered in the regression model to control for the effect. An easy way to determine if the interaction is necessary is to choose a cut-off for significance such as 0.05. The significance can be more or less stringent depending on how much evidence supports that the variables DO have interaction. For instance, measles and age would be included even if the interaction variable had only 0.09 because there is a strong belief that these do effect the model. Return to the main document to see the selected interaction terms.