Analyzing Results

Sources of Error

Random Error: Measures would be distorted in either direction because of stuff you cannot help (“The World is a Random Place”). This will affect precision. Remember that Precision is PPV (Positive Predictive Value). Given the outcome, how well did your model/approach do with identifying the truth? $P(Y = 1 | \hat{Y} = 1)$

One way to get around this: You could just increase the sample size 🤷‍♂️. So things (confounders, noise) cancel out. But this is not always feasible.

Systematic Error: This affects accuracy.