Hierarchical Linear Modeling versus Visual Analysis of Single Subject Design Data

• Main Author: Nelson, Elizabeth Godbold
• Other Authors: Jones, Glenn
• Format: Others
• Language: en
• Published: LSU 2012
• Subjects: Psychology
• Online Access: http://etd.lsu.edu/docs/available/etd-04252012-152015/

Visual analysis is the gold standard for single-subject design data because of a presumed low Type I error rate and consistency across raters. However, research has found it less accurate and reliable than typically assumed. Many statistics have been proposed as aids for visual analysis, but most suffer from limitations either due to methods of investigation or problems inherent to the statistics themselves. Several researchers have proposed the use of Hierarchical Linear Modeling to analyze single-subject data because it can withstand violations of assumptions often present in single-subject data that other statistics cannot. In addition, HLM is similar to the actual data structure of single-subject designs as it allows predictors to be nested within different levels of analysis. Godbold (2008) tested the accuracy of HLM against visual analysis ratings of the same data and found HLM to be a potentially useful statistical aid. The current study rectified the limitations of the 2008 study and extended the applicability of HLM to more types of single-subject designs. HLM was again shown to be a viable statistic across a wide variety of design types including single and multiple baseline designs. Comparisons between two HLM models indicated a longitudinal HLM model was more accurate as compared to visual analysis than a simpler non-longitudinal 2-level model, however, more research is warranted.