Factor analysis is addressed in our Chapter 4. Factor analysis graphs may look strange; it certainly takes some time to learn how to read them (to that effect, you should re-read our chapter and look into our suggestions for further reading). Still, many historians want to use factor analysis, once they have understood that it is an exploratory method that allows them to get a general idea of “everything that is in their dataset,” and to create typologies. (Perhaps this is also true in other disciplines, but the words “exploratory”, “having a look at everything at once,” and “typology” seem to have a special appeal for historians – ourselves included)
This is all good; but if you go straight to the tutorials below, without first thinking hard on how to categorize your data, and which variables to include (in fact, it’s never “everything”; you especially have to choose what to include as an active variable, a supplementary variable, or not at all), you will be severely disappointed. So go back to our chapter 4 for advice on these topics! And do not hesitate to ask us for more practical suggestions in your specific case.
Once you know what exactly you want to analyze, you need software: a spreadsheet editor won’t perform factor analysis. We favor R, which you’ll need to install; but you can perform factor analysis without using a command line (if this is a difficulty for you), thanks to RCommander. Here you can learn how to install R and RCommander.
Once you have done this, you can then install FactoMineR, a package for factor analysis, and its RCommander extension (“RcmdrPlugin” in the jargon). FactoMineR offers all the variants of factor analysis; in our experience, the most useful one for the humanities (where most variables are “qualitative” rather then “quantitative”) is multiple correspondence analysis (MCA). The authors of FactoMineR provide three types of documentation – on all methods and MCA specifically:
- a book, Exploratory Multivariate Data Analysis by Example Using R, that is very accessible;
- a MOOC and video tutorials;
- written tutorials in the “Facto’s method” category, for example this one on MCA.
The tutorials above mostly refer to the command line version of FactoMineR: you don’t really need them to run the RCommander version, which is straightforward if you already know the vocabulary of MCA, but they will help you to interpret the results. The same is true for the more detailed written tutorial on MCA with FactoMineR (by Alboukadel Kassambara) that you can find here.
If you want to read only one tutorial, you should choose the one that Ryan Deschamps wrote specifically for historians. It uses data on the membership in recent parlimentary committees in Canada. The data is simple enough that the author does not elaborate on difficult choices at the stages of data collection and selection for factor analysis. But the tutorial is very commendable because it not only lists command-lines, but explains how to read graphs (both in lay terms and with mathematical explanations in an appendix), and demonstrates that interpretation involves drawing successive graphs experimenting with different parameters (evolving from a substantive initial question).
FactoMineR’s default graphs are however not beautiful and, in some cases, unreadable. You might want to use an additional package to get better graphs. The easiest way is probably to install, then load the explor package in R and type explor(res) as a command line (if you have just performed an MCA with FactoMineR. As is explained here, it will open an interactive graph in your browser. (Ryan Deschamps’ tutorial suggests other ways to improve graphs in the command-line version of FactorMineR)
FactoMineR is a good package to explore your data, especially as the RCommander version includes a clustering option, allowing you to create typologies (see our book and the documentation for the principles of clustering). In the command-line version, it also offers the option of drawing ellipses around categories of individuals (more options here), addressed in our Chapter 4 (this one is not included in the RCommander menus: you will have to copy the command line).
Finally, if you want to perform specific correspondence analysis (a technique that we briefly mention in our Chapter 4; it allows you to treat specific categories as supplementary and is especially useful if you have a lot of missing data), you will have to use a specific package, gdatools, which requires the use of the command line. There is no genuine tutorial, but the examples given here are quite clear.