Use of calibration constraints and linear moments for variance estimation under stratified adaptive cluster sampling

Analyzing sample data is a difficult task made more difficult whenever the data contains extreme values that impair the precision of variance estimation under traditional moments because such moments assign equal weights to all observations, including extreme observations. Calibration is a method of adjusting sample weights to improve estimation. In this article, under a stratified adaptive cluster sampling, we propose new variance estimators with the appearance of extreme values through the use of calibration constraints along with linear and trimmed linear moments based on variance and coefficient of variation of the auxiliary variable. The percentage relative efficiency of the proposed estimators in comparison with the traditional ones is calculated. The proposed estimators' performance is assessed using real-life and artificial data. Based on numerical comparisons, the proposed estimators outperform the traditional variance estimator. Thus, the proposed estimators can be considered very resistant estimators and they surely boost the chances of obtaining more accurate estimates of population variance with the presence of extreme values.