Use of logistic regression in comparing multiple limiting dilution analyses of antigen-reactive T cells

Traditionally, statistical analyses of multiple limiting dilution analyses have used linear regression in comparing the fit of each analysis to a single-hit Poisson model. Multiple analyses can, however, result in an inflated type I error leading to spurious rejection of the null hypothesis. Logistic regression was used as an umbrella statistic to analyze eight limiting dilution analyses for antigen-reactive T cells from two mouse strains (SJL/J; B10.S), at two priming doses of antigen, for each of two antigens (porcine myelin basic protein, PMBP; purified protein derivative of M. tuberculosis, PPD) used to induce experimental allergic encephalomyelitis (EAE). This proved superior to traditional methods: six of the eight strain/dose/antigen combinations were consistent with a viable model, whereas only three of eight were consistent with the Poisson single-hit model. Moreover, only three of the resultant 32 standardized residual values (four per group times eight groups) fell outside the 95% confidence interval. Logistic regression, therefore, is recommended when nominal dependent measures can be structured as a function of more than one independent variable.