Language: English Français

# Why do I need to adjust for age and sex of children again if z score calculation already involved gender and sex in its calculation?

This question was posted the Assessment and Surveillance forum area and has 5 replies.

### Mark Myatt

Frequent user

14 Apr 2019, 10:44

I think you may be confusing data-management and data-analysis.

Data management : When we calculate a z-score for a child with a particular age and sex (e.gender. weight-for-age z-score) we find the child's distance from an expectation of ideal (WGS) ot normal (NCHS, CDC) growth for children of the same age and sex. We typically use this to identify nutritional deficits so that (e.g.) a child with WAZ < -3 may be defined as severely underweight compared to the average for a child of similar age and sex in the reference population. We take into account age and sex since children grow over time and boys and girls grow a bit differently. It we only used a single weight threshold such as the reference average for children 6 to 59 months we would likely end up defining young children as overweight and older children as underweight. What we have when we calculate WAZ using age and sex is a method of ensuring that (e.g.) WAZ < -3 has the same statistical meaning for all children. (it is important to realise that this may not translate into functional meaning). No analysis is taking place here we are simple creating a distance measure for (e.g.) weight-for-age (or weight-for-age-for-sex) that we will use in data-analysis.

Data-analysis : Often we want to look at factors associated with a negative (or positive) outcome. We might find (e.g.) that sex (e.g. being male) is associated with low WAZ. We might also find that this relationship is strongest in younger children. We need to include sex and age in models to discover this. You use the term "control". This suggests a concern to find effects that are independent of age and sex. In this case you would also want to include age and sex in models.

I hope this is of some help.

### Mark Myatt

Frequent user

16 Apr 2019, 13:50

I am not sure what variables and models are being used here. I assume that you have a linear model with something:

```    HAZ ~ constant + age + ...
```

If you see a negative coefficient such as 0.22 we have:

```    HAZ ~ constant -0.22 * age + ...
```

You'd interprent this as a drop of -0.22 H/A z-scores per month. This looks a bit high to me unless you have a narrow range of age in your data.

WRT sex (gender is a confusing term and usually refers to cultural and social context) ... you need to be careful how you code sex. It is usually best to include sex as a binary (0, 1) variable (AKA dummy variable or constrast) for a single sex so that you would have a variable 'female' coded 1 (female) and 0 (not-female == male). If you had:

```    HAZ ~ constant -0.22 * female + ...
```

You'd indterpet this as a difference of -0.22 z-scores associated with being female.

I hope this of some use.