Researchers got it right in estimating numbers of doctors lost from NHSBMJ 1999; 319 doi: https://doi.org/10.1136/bmj.319.7209.579 (Published 28 August 1999) Cite this as: BMJ 1999;319:579
- Michael Goldacre, unit director,
- Trevor Lambert, statistician ()
EDITOR—Hall questions the independence of the two methods we used to identify doctors working in the NHS and therefore questions the results obtained by our use of capture-recapture analysis.1 2 He assumes that both methods depend on doctors' propensity to respond to inquiries. They do not.
We identified doctors working in the NHS by using two fundamentally different approaches. In both cases, the starting point was the nominal list of all doctors who qualified in Great Britain in 1988.
The first approach used the questionnaires of the Medical Careers Research Group sent to all qualifiers regardless of where they were by 1995 (the year of the survey). The information obtained about doctors' employment in the NHS depended on the doctors' responses The second approach used the Department of Health's records, analysed by the department for the same doctors at the same point in time These records are based on information supplied to the department by all NHS trusts, generally from their payroll and personnel records; the information is not based on any inquiry to doctors but on whether the trust held an NHS contract for the doctor at the relevant time It is the combination of results from these two independent methods that gives strength to our figure of an 83% participation rate in the NHS.
Even if the two methods were not statistically independent, Hall's point is flawed. He assumes that doctors identified as working in the NHS by one method would be more likely to be identified as working in the NHS by the other method. If true this would reduce the estimated total number working in the NHS rather than increase it. In calculating the capture-recapture estimates, the higher the overlap between those identified by both methods, the smaller the estimated additional number not identified by either method. Thus the real participation rate would be a little lower, not higher, than that with which Hall takes issue.