Perhaps the intervention has been more successful in older children? Do you have any age data?

Before and after what? An intervention, a harvest, a year in which something or nothing happened? I'll assume an intervention. The first thing that comes to mind is that that the intervention may acted to improve the health of older children (or children that have aged during the intervention) but has done little to improve the status of younger children who may not have been born at the start of the intervention. Another thing that comes to mind is that the two surveys did not have comparable samples. I have seen these sorts of odd findings when (e.g.) children with height <= 110 cm but age > 59 months are in one sample but not others. This is due to survey methodology changes. I have also seen "improvements" after high mortality in the younger age groups. You should have age data (it will probably not be of great quality) to check this. Sample differences can be tricky since you will probably have sampled different villages in the two surveys. If, by accident, one of your samples favoured (e.g.) different ethnic groups or food-economy zones or agro-ecological zones then you may see a before and after difference when none existed. Can you map the sampling locations. Also, MUAC is more sensitive to infection. Did the surveys take place at the same time of year? Let us know more and we may be able to help more.