 | The 131st Annual Meeting (November 15-19, 2003) of APHA |
|
| The 131st Annual Meeting (November 15-19, 2003) of APHA |
Jeffrey S. Hoch, PhD, Department of Epidemiology and Biostatistics, University of Western Ontario, UWO School of Medicine and Dentistry, Kresge Building, London, ON N6A 5C1, Canada, (519) 661-2111 x86270, jeffhoch@biostats.uwo.ca and Peter Austin, PhD, Institute for Clinical Evaluative Sciences, Room G1 06, 2075 Bayview Avenue, Toronto, ON M4N 3M5, Canada.
The effect of a censored dependent variable has been widely studied in the econometrics and biostatistical literature. There has been less attention paid to the effect of a censored independent variable. This seems peculiar given the potential frequency with which public health researchers may encounter them. Censored independent variables are especially common when the information measured by the variable is considered sensitive. Examples include variables measuring income or risky behaviours. By censoring unusual values, survey respondents are given some assurance of confidentiality. In some cases, even if respondents are willing to report true values, privacy laws or standard practice mandates that data from groups with small membership (e.g., five or fewer members) must be either suppressed or aggregated into larger groups. As a result, censored independent variables are generated. In this paper we formally consider the effect of a censored independent variable in the context of an "errors in variables" framework. The framework provides a useful way of characterizing the limitations of using ordinary least squares (OLS) to estimate coefficients when an independent variable has been censored. After proving results about the bias (both small sample and asymptotic) associated with OLS, we illustrate our theoretical findings with an empirical example. Alternatives to the commonly employed OLS method for estimation and inference in the presence of a censored independent variable are considered.
Learning Objectives:
Keywords: Methodology, Health Care
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: N/A
I do not have any significant financial interest/arrangement or affiliation with any organization/institution whose products or services are being discussed in this session.