Page 145 - Contributed Paper Session (CPS) - Volume 6

CPS1851 Hee Young Chung et al.

A method of bias correction when response rate
follows linear function
Hee Young Chung, Key-Il Shin
Hankuk University of Foreign Studies, Yongin, Rep. of Korea

Abstract
In recent sample surveys, the accuracy and precision of estimates are
decreasing due to non-responses. In particular, there are cases where
non-response is affected by the variables of interest and if we apply some
commonly used non-response treatment methods to those cases, then
we may have bias in estimation. Recently, a method has been proposed
to improve the accuracy of estimation by appropriately reducing the bias
occurred in the case where the response rate is an exponential function
of the variable of interest. In this study, we propose a method to increase
the accuracy of estimation when the response rate is a linear function of
variable of interest and the distribution of errors included in the super
population model follows normal distribution. Simulation results show
the superiority of the proposed method. We also suggest the optimal
number of substrata that can be used in practice based on the simulation
results.

Keywords
linear inclusion probability, sample distribution, regression model, sample
weight

1. Introduction
In recent sample surveys, the importance of proper treatment of non-
response is increasing. The non-response rate becomes significantly higher,
resulting in insufficient number of final survey data, which increases sampling
error. Of course, this problem is already well known and several treatments are
developed. However, there are some cases where the rate of non-response or
response depends on the value of the variable of interest and we need to apply
a proper method to those cases. Especially if we have a super population
model and a corresponding response rate model like the informative sampling
technique, we can calculate the magnitude of bias and so we can correct the
bias caused by non-response.
Chung and Shin (2017) studied the case that the super population model
is a simple regression model and the response rate model is exponential. They
showed that the suggested method improved the accuracy of estimation by
correcting the bias. In this paper, we study the case where the response rate
is a linear function and the super population model is a simple regression

134 | I S I W S C 2 0 1 9