The Past Speaks

Cheryl Schonhardt-Bailey has published an important new book in which she uses quantitative methods to measure the differences in the discourses of two groups of economic policymakers in the United States (the Fed’s Federal Open Market Committee and the congressional committees that oversee the Fed). She has found that these two groups of people use radically different terminology to describe the same phenomena.

This is from the book’s blurb:

Her innovative account employs automated textual analysis software to study the verbatim transcripts of FOMC meetings and congressional hearings; these empirical data are supplemented and supported by in-depth interviews with participants in these deliberations. The automated textual analysis measures the characteristic words, phrases, and arguments of committee members; the interviews offer a way to gauge the extent to which the empirical findings accord with the participants’ personal experiences.

This book obviously deals with a pretty important substantive issue. Moreover, I feel that the book is important because the author’s approach could adapted by business historians to look at historical sources. For instance, it would be very interesting to compare transcripts of corporate board meetings with transcripts of the annual meetings of shareholders.   There are many full-text digital resources that business historians could use for textual analysis.

In an appendix to her book, the author discusses the strengths and weakness of the Alceste textual analysis software she used to research this book.

More information about Alceste can be found here. Here is a description of the program:

Alceste is a qualitative data analysis program that incorporates sophisticated  statistical processing to help you make sense of large amounts of text very  quickly. The School’s other qualitative packages, NVIVO and Atlast-ti, help you  code and retrieve text according to your own analytic framework. Alceste is,  however, completely different. It works by applying a set of statistical  clustering techniques to your text . Different forms of language related to  your research topic are identified by the program, and the text categorized  into clusters according to the distribution of the vocabulary found.