Which metric measures how much the likelihood of a target outcome increases under a specific condition, relative to its baseline probability?

• A. Statistical relevance
• B. Information gain
• C. Lift
• D. Precision factor

Answer: (C)

This focuses on comparing what happens to the chance of an outcome when we condition on a factor versus its overall, baseline probability. The metric called lift does exactly that: it looks at how likely the outcome is when the condition is present and divides it by the baseline probability of the outcome without the condition. If this ratio is greater than 1, the condition makes the outcome more likely; if it’s about 1, the condition has little to no effect; if it’s less than 1, the condition reduces the likelihood.

In practice, lift can be thought of as P(outcome | condition) divided by P(outcome), or equivalently as P(outcome ∧ condition) divided by P(outcome)P(condition). For example, if 20% of the overall population experiences the outcome, but 40% of those with the condition experience it, lift = 0.40 / 0.20 = 2. This means the condition doubles the likelihood of the outcome.

Other metrics don’t measure this specific relative increase. Information gain relates to how much uncertainty is reduced by knowing the condition, rather than how the outcome’s probability scales with the condition. The other terms aren’t standard metrics for this purpose.