**Mean Outcome
by Independent Variables**

**Mean WAZ by Roofing Type and Education
Mean AC prevalence by Latrine and Water Type**

**Individually: Roofing**

First, examine each independent variable in relation with the outcome to determine the associations. At this point allowances for any further cleaning of the independent variable or determining if it should be further categorized (to eliminate cells under a count of 5 for further analysis) can be made.

1. Open

keast4j.sav2. Click on

Statistics, Compare Means,andMeans.3. In the

Dependentlist, enter the variableswaz(waz SD)andwaprev(prev waz <-2 SD)from the left hand variable list4. In the

Independentlist, enter the variable for roofing type,roofand level of education of the respondent,edn.5. Click on the

Optionsbox and check thebox under the area labeledANOVA tableStatistics for the first layer.6. Click on

ContinueandOK.

First, notice the numbers (N) in each cell in the table. The group with tiles as roofing is very small, insignificant for consideration. Although the group with tiled roofs is better off nutritionally indicative of the higher SES, the two individuals that make up this category will be deselected for further analysis here.

To deselect the group with tiles:

1. In the Data Editor, click on

Utilities, Variables,and then select the variable namedrooffrom the list. Once you click on roof, it will give all of the details of that variable including those that are coded as missing (99).2. Click on

Go Toand it will take the cursor to highlight this variable in the data set.3. Double click on the word roof at the top of the column in the dataset and a

Define Variablebox will appear. Then click on the box labeledMissing Valuesand in the second Discrete value box, add the number31(so now both 99 and 31 are missing discrete values).4. Click on

Continue, OK.5. Now, rerun the exercise to calculate the mean above, step 1-5 (now without tiles for roofing).

The mean z-score for the grass /thatch (lower income indicator) is much lower than that of the corrugated iron roof group (higher income indicator). The mean z-score for low SES is -1.46 and high SES is -1.20, which corresponds in the right hand column to a prevalence (<-2 SD WAZ) of 38% and 26%, respectively. The ANOVA results show that this difference is significant, as the p= 0.01.

The ANOVA presentation here has two areas
of interest for practical application which are the **'F-STATISTIC'
AND 'SIGNIFICANCE'**.

**Individually: Education**

Running the analysis table by
educational attainment, the results are as shown here, also in the expected direction.
There is a decreasing prevalence of malnutrition in each higher educational category, and
an increasing mean weight for age z-score. The largest difference in the results is
seen between the group that did and the group that did not complete primary level
education (-1.1 to -1.5 SD WAZ and 37% to 20% prevalence of underweight). The ANOVA
results support that the difference between malnutrition levels in different educational
groups is significantly different (p=0.000).

**Together: Roofing and
Education**

We are now looking for the combined
effect of education *within category of housing quality*-- in other words removing
the effect of housing, as far as we can by using the two categories. When running an
analysis with two independent variables together, the results will be presented in layers
for each category. To run the tables, either the *Compare means* function can be
used like we have done for the one independent variable. There is one problem with
this option though, it will not give the ANOVA with both education and roofing in the
model. Instead, it is necessary to run the analysis using General Linear Model
(GLM), which will give both the tabular presentation and the ANOVA table.

1. Open

keast4j.sav2. Select

…Statistics,General Linear Model, GLM –General Factorial3. Select

wazand move it to thefield.Dependent Variable4. Select

edn(educational attainment of respondent) androof(good/ bad roof), and move them into thefield.Fixed Factor(s):5. Click the

Options… button.6. Click on

Display, Descriptive StatisticsandObserved Power.7. Click

Continue.8. Click

OK.

Within each educational group, there is a pattern in the expected direction of mean z-scores for roofing type, where individuals with grass/ thatch have lower mean z-scores than those with iron roofs. Also, malnutrition seems to decrease within increasing categories of education (seen by totals). Both education and roof type have the association with nutrition status of the child that follows the expected direction, but it appears that education has the stronger association.

In the ANOVA, an additional factor called an**
INTERACTION** variable is included. It is labeled ROOF*EDN in the 'Tests
Between Subjects Effects' output. Interaction determines if there is an association
between the two variables (ROOF and EDN) to a degree that would change the effect of
education or roofing individually. In this case, the interaction of education and roofing
is not significant at the 0.05 level. Since the interaction variable is __not__
significant, it is good to run the same exercise without the interaction term and
reinterpret the results (Interaction will be covered in far more detail on page 2-
Interaction of chapter 4). To do so:

1. Open

keast4j.sav2. Select

…Statistics,General Linear Model, GLM –General FactorialO.K.3. Select.

wazandmove it to thefield.Dependent Variable4. Select

edn(educational attainment) androof(good/ bad roof), and move them into thefield.Fixed Factor(s):5. Click the

button and select the dot markedModelCustom, then moveednandrooffrom the box calledFactors and Covariatesand enter them into the box labeledModelusing the arrow button.6. Click on the down arrow under the

Build terms boxand selectMain Effects.7. Click

Continue.8. Click

OK.

From the ANOVA __without__ the interaction term, it appears that
education and nutritional status has a more significant result ( p= 0.000) than for roof
and nutritional status (p=0.133). When both education and roof are in the model, the
association with roof is no longer significant above the effect of education. This
does not indicate that roofing (an estimate of SES) is not an important determinant of
nutrition status, but the size of the effect is much smaller when education of the
respondent is considered in the model. The results show that the level of education
is a strong determinant of the nutritional status of the child and roofing is not as
strong once education is accounted for. The next task is to present this data,
return to the two-way page 1 to see this.