Normalizing criteria scores for a decision comparison
How to normalize criteria scores so ratings in different units become comparable, with min-max scaling, z-scores and vector normalization explained plainly.
Contents
- Why different units block comparison
- Three ways to rescale scores
- Handling cost and benefit criteria
- Questions about normalizing criteria scores
- What does normalizing criteria scores mean in a decision comparison?
- How do I normalize mixed rating scales?
- Is min-max scaling or z-score standardization better for multi-criteria data?
- Can normalization change which option wins?
Normalizing criteria scores turns ratings that use different units into numbers you can compare side by side in one decision comparison. A price in dollars, a weight in kilograms and a rating out of five share no scale, so each criterion is rescaled to a common range before the scores are combined. The page explains the main methods, how to treat benefit and cost criteria, and where normalization can mislead you.
Why different units block comparison
Criteria measured in different units cannot be added or ranked against each other until they share a scale. A score of 40 on one criterion may mean excellence, while 40 on another may mean failure. Normalizing multi-criteria data converts every score into a unit-free number, so each criterion speaks the same language. Without that step a weighted sum model silently favors whichever criterion carries the largest raw numbers.
Three ways to rescale scores
Min-max normalization stretches each criterion so its lowest value becomes 0 and its highest becomes 1. Z-score standardization subtracts the average and divides by the spread, which centers scores at 0 and keeps their shape. Vector normalization divides each score by the length of the whole column, which preserves relative distances between options. Each method answers a different question, so pick the one that matches how you intend to combine the scores.
Handling cost and benefit criteria
Benefit criteria reward higher values, while cost criteria such as price or effort reward lower ones. Min-max scaling handles this by reversing the formula for cost criteria, so the cheapest option receives the top normalized score. Z-score standardization needs the sign flipped after standardizing, or the direction of the criterion will be lost. Mixing the two directions without correction produces a shortlist that ranks the wrong options first.
Questions about normalizing criteria scores
What does normalizing criteria scores mean in a decision comparison?
Normalizing criteria scores means converting each criterion's raw values onto a shared scale before they are compared or weighted. The conversion removes the original units, so a rating out of ten and a distance in kilometers can sit in the same table. The result is a set of comparable criteria scores that reflect position within each criterion. The method you choose determines what that position means.
How do I normalize mixed rating scales?
Mixed rating scales are normalized by applying one rescaling rule to every criterion separately, never to the raw numbers combined. Min-max scaling to a 0-1 range is the simplest route when you know each criterion's minimum and maximum. Scale conversion to 0-10 works the same way if a ten-point range suits your audience better. Keep the direction of each criterion in mind, because cost criteria need reversing.
Is min-max scaling or z-score standardization better for multi-criteria data?
Neither method suits every case, so the choice depends on your data and your aggregation step. Min-max scaling gives bounded scores between 0 and 1 and keeps the ranking of options within a criterion. Z-score standardization keeps the spread of each criterion and copes with extreme values more gently. Vector normalization preserves geometric distance and pairs with some scoring models.
Can normalization change which option wins?
Yes, normalization can change the ranking of options when criteria use different scales or contain outliers. Rescaling to a 0-1 range compresses everything between the smallest and largest value, so one extreme score can flatten the rest. Z-score standardization reacts to the average and spread, which shifts scores when an unusual value appears. Check the shortlist under a second method before you commit to a choice.