|
Abstract:
|
In a longitudinal data analysis, researcher is often confronted with the task of explaining the trend in the response variable. A simple regression model with time as the independent variable may be affected by errors caused by variables not represented in the model. The autocorrelated residuals violate the basic assumptions of linear regression analysis. If the error variances are not properly estimated, very important aspects of the regression analysis will be incorrect. Although the estimates of the parameters would not be biased, assessments regarding their variances would be incorrect, and the significance tests and confidence intervals of the parameters would be in error. In this paper, several alternative methods will be discussed to reduce the autocorrelation and test their effect on trend evaluation. Other methods of estimating trend will also be compared.
|
