Identification of child growth outliers in longitudinal data

[Andressa Freire]

Hello, Stata community

I need to identify outliers for a longitudinal bench with repeated measurements of weight and height over five years of study (secondary data). The literature on identifying outliers is extensive and there are several approaches. Also, I see that excluding outliers can be dangerous. In my case, when it comes to child growth using the WHO curves, there are cutoff points for the anthropometric indices height/age, weight/age, BMI/age that identify biologically implausible values. However, this is not enough in longitudinal data, since the plausibility of a given observation depends not only on its absolute value at each time point, but also on previous and subsequent measurements of the individual.

What strategies do you recommend in Stata to identify these longitudinal outliers in the case of child growth? I've been exploring commands like bacon, but I think it's not enough to think about issues in repeated measurements, such as: children who decrease in height over time, children who gain 40 kg from one appointment to another...

I appreciate any help, insight

[Carlo Lazzaro]

Andressa:
if you cabbot send out to those who made the data entry queries aimed at clarifying "weird" results, you should get rid of all the observations that are really irrealisric according to the data generating. For instance, if a child can lose weight from one visits to another, her/his loss in height sounds irrealistic.