Essentials in the Design and Analysis of Time-Series Experiments
Regina H. Rushe and John M. Gottman University of Washington
The purpose of this chapter is to provide a very broad overview of the design and analysis of time-series experiments. Social scientists find time-series methodology a highly sensitive and specific tool to detect subtle, but important, changes over time. Psychologists find it particularly appropriate for research in education, psychotherapy, psychophysiology, and other areas where interventions are assessed in imperfectly controlled settings.
The emphasis is on concepts and design, as well as a limited amount of statistical methodology. We believe social scientists must be well grounded in both statistical and substantive issues in order to make valid decisions about specific issues in time-series analysis.
Time-series data require statistical analyses that will maximize the information available in many, dependent observations that change in important ways over time. Most statistical tests taught in introductory courses have assumptions that are violated or have limited ability to retrieve important information in such circumstances. Fortunately, statisticians have been challenged by time-series data and continue to develop statistical theory and applied techniques to make time- series analysis a fruitful and reliable endeavor.
It is useful to think of time-series methodology as an outgrowth of the observation process used in the physical sciences. Campbell and Stanley ( 1963) gave a "pre-experimental" example from 19th-century science where a bar of iron that has remained unchanged for many months is dipped in a bath of nitric acid. Should the bar then lose weight, the change would be attributed to the nitric acid bath. The physical sciences have had much success with experimental designs