| Re: Age in months or years? [message #2888 is a reply to message #2865] |
Sat, 06 September 2014 17:30   |
Reduced-For(u)m
Messages: 292
Registered: March 2013
|
Senior Member |
|
|
This is a statistical point that may or may not be relevant to your work, but which I think is important to keep in mind when doing these "determinants of stunting" type regressions.
If you collapse HAZ/WAZ by age you will see a very strong age profile, where in most countries children start out with z-scores around 0 to -1, and then those z-scores fall off very quickly over the first few years of life. So most children are not stunted at birth or 6 months old, but the median child may be stunted by age 2 or 3, depending on the country.
I mention this just because it makes interpreting regressions coefficients of the "determinants of stunting" a bit difficult when looking at all children aged 0-5. Including a linear trend in age (which will have a significant and negative estimated coefficient) doesn't totally help with this interpretive problem (and does not fully "control" for the highly non-linear HAZ/WAZ-age profile). If you really wanted to completely control for the age effect (say because you are interested in the effects of some other covariates that may in some way be correlated with child age - like parental age or seasonality or any covariates that, within a household, change over time systematically) you would probably want to include either dummy variables for age in months or some very flexible polynomial in age.
Whatever specification for age you put in your regression, you will get a more refined control by using age in months instead of age in rounded years (or course, age in years with decimal places will give you the same result).
|
|
|
|
The DHS Program User Forum
Members
Search
Help
Register
Login
Home
