you have to recognize both the limits and the dangers of data-driven analysis, in which biases are hidden behind layers of numbers, imprecision is masked by precise-looking estimates, and decision makers let models make choices for them on what to do and when to do it.
