A statistical model is presented for the analysis of infectious disease data from family studies in the community. The model partitions the sources of infection into those from within the household and those from the community at large. The parameters reflecting these sources of infection are estimated as functions of the risk factors. This new model is used to overcome problems associated with the lack of independence of observations in infectious disease data and negative confounding due to the association of unmeasured exposures and immunity. An example of how this new statistical model is used to provide a clearer and less confounded description of risk factor effects is presented for data from influenza A(H3N2) epidemic seasons in the Tecumseh Respiratory Illness Study. The risk factors examined are age and pre-epidemic season antibody level as measured by the hemagglutination-inhibition test, while the outcome is the infection rate. A standard analysis of the data indicates that the efficacy of protective antibodies is 70% in children and only 47% in adults. However, such an efficacy measurement is negatively confounded by past exposure which is age dependent. By means of the model, the true, unconfounded, efficacy of protective antibodies is shown to be 90% in both adults and children.