Weighted Cumulative Sums for Text Analysis Using Word Counts
A. F. Bissell · Journal of the Royal Statistical Society Series A (Statistics in Society) · 1995
Morton has pioneered the use of counts of 'habit words' in textual analysis. In particular his QSUM method, using overlaid cumulative sums (CUSUMS) of sentence length and habit word frequency, can display anomalies in the use of these words by separation of the two CUSUM paths. There is an appreciable subjective element in interpretation, and no obvious means of providing probabilistic criteria. This paper describes the application of weighted CUSUMS to word counts. They offer enhanced clarity and objectivity, and permit hypothesis tests or interval estimation based on conventional statistical techniques. Some methods for more detailed analysis are also outlined.