A Comparison of Four Factor Analytical Methods Used with Ordinal Data

The Royal Shakespeare Company created the Stand Up for Shakespeare (SUFS) program to change the way students encounter Shakespeare in school. The program prepares teachers to help students engage with Shakespeare the way actors would—interacting with the plays as scripts to be acted rather than texts to be read. Through this pedagogy, SUFS aims to increase students' interest in Shakespeare and both their interest and ability in reading. The most thorough evaluation of the SUFS program (Strand, 2009) used factor analysis to examine the structure of attitudes toward Shakespeare and found it to be unidimensional and reliable (Cronbach's α = 0.85). The resulting factor scores correlated somewhat with academic self concept (r = 0.22) and school engagement (r = 0.37), but not with attainment in Language Arts. However, despite measuring student attitudes on an ordinal scale, many of the analyses utilized methods that assume continuous and normally distributed data, consistent with findings that ordinal data are often treated inappropriately in analyses in applied research (Kampen & Swyngedouw, 2000). Because ordinal data are the norm in education research but are also frequently analyzed incorrectly, this study explored the internal structure of the SUFS data, including the stability of the factor structure, to illustrate how more and less appropriate analytic decisions manifest in real data characteristic of the field. More importantly, it compared four factor analytic methods head-to-head to determine which produced the most stable factor structure, validated by a CFA on data gathered at two time points. The four methods included a traditional exploratory factor analysis (EFA), a full-information or ordinal EFA (Jöreskog & Moustaki, 2006), and two exploratory factor analyses within the confirmatory factor analysis framework (E/CFA); one according to the