CFE 2019: Start Registration
View Submission - CMStatistics
B1866
Title: New nonparametric approach to correct response bias on ordinal categorical data using Anchoring vignette Authors:  Mariko Takagishi - Osaka university (Japan) [presenting]
Abstract: In questionnaire surveys, Likert type questions (e.g., 1-strongly agree,...,5-strongly disagree) are often observed. However, since how to interpret each category is different among respondents, direct comparison among respondents is not straightforward. We call the bias that occurs by ignoring this interpersonal incomparability problem as ``response bias''. Anchoring vignette is a tool to correct response bias in observed response. In the Anchoring vignette framework, a method for correcting the bias is applied to the questionnaire data before the statistical data analysis. Several existing correction methods are proposed, such as an ordinal regression based, and Item response theory based. The possible question would be ``which correction method should I use before the data analysis?''. So far this problem has not been discussed well, because for the existing methods, the ``corrected value'' is defined in various ways and thus cannot compare the property among them. Therefore, we introduce a new comprehensive statistical modeling for correction which includes the existing correction methods as special cases, and propose a new simple nonparametric correction method based on the new statistical model. In addition, we derive the property of the proposed corrected value which is useful for data analysis.