......................................................................

NJIT Applied Mathematics Colloquium

Friday, October 29th 2010, 11:30am
Cullimore Lecture Hall II
New Jersey Institute of Technology

......................................................................


Statistical Analysis of List Experiments

Kosuke Imai

Princeton (Department of Politics)


Abstract

With the introduction of inexpensive Internet surveys, a growing number of social scientists are designing their own surveys. Although the validity of any survey research relies upon the accuracy of self-reporting, eliciting truthful answers can be challenging especially when studying sensitive questions. Recently, list experiments (also known as the total block response method and the item count technique) have attracted much attention as a methodology to overcome this problem. I introduce a suite of new statistical methods for analyzing list experiments. First, I propose new nonlinear least squares and maximum likelihood estimators for a multivariate analysis. The two-step estimation procedure and the Expectation Maximization algorithm are developed to facilitate the computation. Second, we address the potential violations of the key assumptions made in the standard analysis of list experiments. In particular, we develop a statistical test for detecting the design effect where the inclusion of the sensitive item changes responses to the non-sensitive item. We also show how to adjust for floor and ceiling effects in order to address the concern that some respondents may suppress their truthful answer to the sensitive item even in list experiments. This talk is based on the paper available at http://imai.princeton.edu/research/listP.html




Last Modified: Oct 2010
Linda Cummings
L i n d a . J . C u m m i n g s @ n j i t . e d u