"Trend design" data are gathered at different points in time from the "same population" but, in contrast to the "panel design", not from the same people. Sensory impairment data obtained by pooling the annual National Health Interview Surveys from 1986 to 1994 cannot be regarded, in a strict statistical sense, as consisting of nine representative samples from the "same" population. This is due to the fact that some people from the original population died during the period of investigation and new people were "born into it". However, in this study the aggregate data will be considered as trend data.
Use of log-linear models will permit to ascertain overall and subgroup-specific net changes or trends in a particular characteristic (e.g. visual impairment) and, also, to investigate the simultaneous net changes in two or more characteristics (e.g. multiple sensory impairment). Log-linear modeling will also be used for determining the influence of variables such as age, income, race/ethnicity, etc on these trends. All analyses will start with a model which assumes that the log linear effects of time on a given characteristic are linear and, therefore, that the log-odds follow a linear pattern over time. This type of model implies that the odds themselves increase (decrease) over time by a constant and, consequently, lie on a logistic curve. In these cases logistic regression will be used for further analyses. If linear models do not fit the data, higher order models in which the log odds follow a curvilinear pattern must be explored.
Learning Objectives: At the conclusion of the session, the participant will be familiar with important issues related to trend analysis of NHIS data
Keywords: Statistics, Methodology
Presenting author's disclosure statement:
Organization/institution whose products or services will be discussed: None
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.