sPlot: a statistical tool to unfold data distributions

A novel method called sPlot, painless to implement, is presented. It projects out the signal and background distributions from a data sample for a variable that is used or not in the original likelihood fit. In each bin of that variable, optimal use is made of the existing information present in the whole event sample, in contrast to the case of the usual likelihood-ratio-cut projection plots. The thus reduced uncertainties in the low statistics bins, for the variable under consideration, makes it possible to detect small size biases such as pdf/data mismatches for a given species, and/or presence of an unexpected background contamination, that was not taken into account in the fit and therefore was biasing it. After presenting pedagogical examples, a brief application to Dalitz plots and measurements of branching ratios is given. A comparison with the projection plots shows the interest of the method. Finally are given the differents steps to implement the sPlot tool in an analysis.