G*Power: importancia del tamaño muestral en análisis de series temporales

Abstract

Introduction: Time series analysis is particularly useful, especially in disciplines that require close longitudinal monitoring. Despite their great usefulness, its use is not common in fields such as psychology. For this reason, the authors of this work used time series to carry out a study on tobacco addiction. The results showed that tobacco behaviour followed an AR (2)(7) 8 model, that is, the sample had a 56-day memory. The objective of the present work is to verify the statistical power and the effect size of the model that they found. Method: Given the absence of information in the previous references, an analysis of time series was performed a posteriori imitating the models founded in the previous studies. It was calculated using G*Power software if our sample size is large enough to obtain a model with statistical power and a good effect size. Results: The output indicates that a minimum of 17 subjects are needed, with 63 data each day (a total of 1071 data) to obtain a model with a good statistical power and effect size. Conclusion: To sum up, we conclude with the affirmation that time series analysis has a poor statistical power, so samples for this type of analysis should be quite large. Furthermore, the ideal number of subjects to obtain an adequate statistical power and an effect size should be checked by a previous study or, if that is not possible, a posteriori analysis.