Addressing Autocorrelation in Time Series

• In Event: Adaptations of GLM for Advanced Data Analysis

Abstract

The sequential nature of observations in time series make them inherently prone to autocorrelation. This can be problematic because it violates a major assumption associated with many conventional statistical methods and, therefore, data analysis requires additional technical considerations. Although there are numerous procedures that can be implemented to address autocorrelation, the literature is generally devoid of information that compares the benefits and disadvantages between various methods. This paper examines autocorrelation in time series, related concepts, diagnostic testing, and several analytical strategies commonly used in social science research. Empirical data from college course evaluations are used to contrast the results of four commonly used methods for adjusting autocorrelation. Implications of results and recommendations for choosing between these strategies are discussed.

Authors

• Shenira A. Perez, Boston College

• Larry H. Ludlow, Boston College