Selection bias in GP fundholding

This paper uses a logistic regression model based on 1993 data for general practices in a single Family Health Services Authority (Lincolnshire) to analyse the differences in characteristics between existing fundholding (up to and including wave three) and non‐fundholding practices. A high degree of classification accuracy is obtained. Fundholders are revealed to be more likely than non‐fundholders to meet a number of the various quality criteria laid down by central government following the 1990 National Health Services Act, for example, with respect to prescribing cost control, minor surgery and cervical screening uptake. The model is employed to forecast the fourth wave of fundholding and poor predictions suggest the existence of a structural break in the characteristics of fundholders between those in the first three waves and those of wave four. The evidence presented also supports the existence of selection bias in the first three waves of fundholding, although further logistic regression analysis reveals a form of such bias in the fourth wave also.