Defrauding Cyril Burt: A reanalysis of the social mobility data

Abstract

In the last comprehensive review by Mackintosh et al. Cyril Burt, Fraud or Framed? (London: Oxford University Press, 1995) of the fraud charges posthumously leveled against the once eminent psychologist Sir Cyril Burt, Mackintosh and Mascie-Taylor asserted that statistical anomalies they detected in his social mobility data of 1961 provided crucial evidence of guilt. The anomalies included apparent departures from normality in some parts of the data, incommensurate cell totals, and suspicious uniformity within IQ bands across fathers and sons. It is shown here that the departures from normality were a natural consequence of unavoidable rounding when inverting the cumulative normal distribution to construct the class IQ bands used in the tables. Elementary procedures are given, known since at least the 1930s, which could have been used by Burt to simultaneously preserve both the normality of his IQ data and the desired population proportions of occupational classes. Other anomalies first noticed by the statistician Donald Rubin are explainable as artifacts produced by fixing marginal totals in the presence of rounding to IQ scores, then using the same weighting procedures to conform to margins. The grounds given by Mackintosh and Mascie-Taylor for finding fraud in Burt's social mobility data are therefore dismissed.