Although 12 variables were used to build the version of Philadelphia's forecasting model in place in 2012, not all made an equal contribution to the model's accuracy. The graph shows how much accuracy was lost when researchers replaced the actual values for that predictor with random numbers.