1.13 Percentages and Absolutes
Data comparisons can be made using absolute differences, or percentage-based differences; neither method is necessarily right or wrong. For most comparisons involving groups with different absolute sizes, however, percentage differences are often the most meaningful.
When calculating percentage differences, always remember the expression “Change over Base.” The “change” is found by subtracting the old value from the new value, and the “base” is the original value.
If Lobster Land sold 5000 season passes last year, and sold 7200 season passes this year, by what percentage did season pass sales increase?
By applying the change/base formula, we can see that the percentage change is (7200-5000)/5000. We can express this as 0.44, or as 44%.
A common mistake that people make when describing model performance is to conflate percentage improvement with percentage point improvement.
If you built a model whose original version had a 58% accuracy rate, but then brought that rate up to 72% after some adjustments to the settings and the choice of input variables, how much did you improve from Version A to Version B? Do not sell yourself short by saying “14%” here. Yes, your accuracy increased by 14 percentage points, but in terms of overall percentage improvement, yours was actually (.72-.58) / .58 = 24.14%.