Accounting for response missing at random and observed heterogeneity in the estimation of mean: Practice and application to real data of childhood respiratory disease and body fat percentage