Calendar effects in the analysis of seasonal data

A variety of statistical methods exist for analyzing seasonal patterns in epidemiologic data. As a simplification in the calculations, these methods often do not explicitly take into account certain calendar effects, such as the variation in month length, the irregular number of weekend days in each month, and the occurrence of holidays. This paper evaluates the bias caused by failing to recognize these effects. It is found that with the sample sizes commonly encountered in this type of analysis of epidemiologic data, calendar effects have a high probability of producing a spuriously significant seasonal effect, the amplitude of which may be of the same order of magnitude as the true underlying seasonal trend. Therefore, it is recommended that calendar effects be routinely taken into account, and some methods for doing so are proposed.