Adam C. Carle, MA, PhD, Statistical Research Division, U.S. Census Bureau, Center for Survey Methods Research, Washington, DC 20233-9100, 301-763-1863, adam.c.carle@census.gov

Measurement bias, a form of non-sampling error, occurs when individuals equivalent on a construct being measured, but from different groups, do not have equal probabilities of observed responses. Confirmatory factor analysis (CFA) is a commonly used quantitative model to examine bias. In the model, equations specify the relation between observed responses and the latent variable of interest. Measurement bias exists when the relevant parameters in these equations differ significantly across the groups. However, given the number of parameters included in the model, it is reasonable to expect that some parameters will differ significantly. Likewise, in large scale epidemiological research, it will often be the case that sufficient power will be present to identify even small differences as statistically significant. Partial measurement invariance refers to the case when some parameters are equivalent, while others are not. Under partial invariance, the possibility exists that statistically significant differences are not large enough to impact observed scores in a meaningful way. Unfortunately, no standard method exists and few guidelines are available to evaluate empirically the impact of partial measurement invariance. Using data from the National Longitudinal Alcohol Epidemiological Survey (NLAES), a nationally representative household survey of 42,692 adults, the current paper extends a recently proposed method to evaluate the impact of measurement bias. By studying group differences in selection accuracy as a function of measurement bias, the technique quantifies the impact of measurement bias. The general mathematical model is developed and the analytical approach is discussed with real data.

Learning Objectives:

• At the end of the session the learner will be able to define impact, as it applies to statistically significant measurement bias, and discuss a new quantitative method to assess the impact of observed measurement bias.

Keywords: Statistics, Behavioral Research

Presenting author's disclosure statement:
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.