These are very high values for WHZ. Anything > +2 would be classed as overweight and anything > +3 would be classed as obese.

WHZ > +5 should have been flagged when using WHO flagging criteria (see page 41 in this WHO manual) as "biologically implausible". The child with WHZ = 5.56 should have been censored "after not including the implausible z score". Something did not work as expected. NCHS flagging criteria accept WHZ values up to +6,

Date-of-birth / age has no bearing on WHZ which is calculated using weight, height, and sex. I think that erroneous data (e.g. low height or high weight) might be to blame.

Note that flagging criteria usually apply to survey data. It is common to see (e.g.) a few true HAZ values more extreme than the lower WHO flagging criteria in clinical (CMAM/OTP) samples.

I hope this is of some use.

Dear Mark Myatt:

I checked my data; the maximum WHZ in my dataset was +4.

I was looking at the HAZ line.

If WHZ is still +4 (within plausible range by WHO), are children being considered overweight or obese since they are outside +2 WHZ.

I would very much appreciate for your kind expert clarification.

Sincerley

Dear Mark Myatt:

If anything above +2 overweight (for WHZ), how does inclusion of those children with WHZ > +2 affect the multiple regression predication, where WHZ is used as an outcome after controlling for covariates in the model?

Thanks so much again.

Humbly

Sorry for getting to this late ... Two questions ...

When we used indices like WHZ we are comparing our data to data from a reference population. WHZ > +2 is seen in about 2.28% of the reference population and WHZ > +3 is seen in about 0.13% of the reference populations. This is a statistical (not functional) defintion of overweight and obese. We can get ourselves in terminological confusions. We usually define underweight and overweight using weight-for-age but we can use weight-for-height as this matches with BMI (another W/H measure) commonly used in adults.

I am not sure what you are asking WRT multiple regression. The main issue is probably one of outliers having excessive leverage on regression estimates. You can identify outliers using methods such as Mahalanobis distance and censor. An alternative is to use robust regression methods.

I hope this is of sume use.

Dear Mark Myatt:

Thanks so much. Yes, I have presented standard regression results and robust regression results.

I very much appreciate for your technical guidance.

With best regards