Yadav, Vinay Kumar; Bulut, Hasan; Zaman, Tolga

In survey sampling, the calibration approach is among the most widely used techniques to improve estimator efficiency. In this study, we introduced robust regression approaches into calibrated exponential ratio-type estimators for finite population mean, which helps to ensure better estimation precision when outliers are present in the data. In particular, using stratified random sampling, we introduce the first three moments of the origin of an auxiliary variable in the calibration approach. Second-and third-order moments are also considered by using exponential mean-type estimators. This technique is further modified for study involving sensitive variables, especially in research on personal relationships and intimate health where the respondent may under-report because of privacy concerns. Biased answers are often obtained when questions such as “Have you ever engaged in unprotected sex?" or “Have you ever been tested for a sexually transmitted infection?" By reducing these biases, our strong calibration estimators improve the accuracy of the study. In a simulation study, the data is randomly generated for each stratum and outliers are introduced at 5%, 10% and 25% by randomly selecting a specified proportion of the generated data and multiplying those values by a factor of 10. The findings of our suggested robust estimators perform better than those already in use, especially when it comes to managing sensitive data and outliers, which makes them more effective for real-world use.

İncele

Zaman, Tolga; Shazad, Usman; Yadav, Vinay Kumar

In sample surveys, the problem of outliers is one of the most frequent and widest, whose solution is required to be obtained using statistical techniques. To overcome this problem, various robust regression methods are being developed such as LTS, LMS, LAD, Huber-M, Hample-M, Tukey-M and Huber-MM methods. This article presents the modified Hartley–Ross type ratio estimators to estimate the population mean in sample surveys. The proposed design is taken into account under the two suppositions one is that the study variable is a non-sensitive variable, which means that measurements on it do not embarrass participants in personal interviews and other is the sensitive variable, which means that measurement errors are introduced as a result of a few dishonest responses. The use of scrambling response models helps to reduce these measurement errors to some extent. The proposed estimator is found to be more efficient than the existing classical estimators. A Numerical illustration was performed on a real data set in R-Program Software to support of our findings.

AtıfScopus15·WoS17·OpenAlex16

İncele