• Advanced
  • About 17 min
  • 5 practice questions

Building a composite inclusion index

Normalise inclusion indicators with the min-max method, turn 'bad' indicators the right way round, combine them with weights into one index, and test how rankings change when the weights change.

If something looks wrong, please tell us.

The Practice course measured access, usage and quality one indicator at a time. Policymakers and examiners often want a single number that ranks districts or states. That number is a composite index, and building one involves three choices: how to put indicators on a common scale, which way round to count 'bad' indicators, and how much weight each dimension gets. This lesson works through each choice and shows why the weights matter so much. All data are invented.

Step 1: put indicators on a common scale

Outlets per 10,000 adults might range from 2 to 8, while the share of active adults ranges from 30% to 60%. Adding them directly would let the bigger numbers dominate. The min-max method rescales each indicator to run from 0 to 1 across the units being compared: the worst unit scores 0, the best scores 1, and the rest fall in between in proportion.

Normalised score = (value − minimum) ÷ (maximum − minimum). For an indicator where lower is better, use (maximum − value) ÷ (maximum − minimum). Index = sum of weight × normalised score, with weights adding to 1.

Step 2: turn 'bad' indicators round

For complaints per 10,000 accounts, a lower value is better. Using the ordinary formula would reward the district with the most complaints. The fix is to measure the distance from the maximum instead, so the district with the fewest complaints scores 1.

Step 3: weight and add

Each dimension gets a weight, and the weights add up to 1. The index is the weighted sum of the normalised scores, so it too runs from 0 to 1.

Worked example 1: three districts

Invented indicators
DistrictOutlets per 10,000 adultsAdults actively using accounts (%)Complaints resolved on time (%)
D126070
D253090
D384550

Using weights of 0.3 for access, 0.5 for usage and 0.2 for quality, rank the districts. Working: access, range 2 to 8: D1 = 0, D2 = 3 ÷ 6 = 0.5, D3 = 1. Usage, range 30 to 60: D1 = 1, D2 = 0, D3 = 15 ÷ 30 = 0.5. Quality, range 50 to 90: D1 = 20 ÷ 40 = 0.5, D2 = 1, D3 = 0. Index: D1 = 0.3 × 0 + 0.5 × 1 + 0.2 × 0.5 = 0.60; D2 = 0.15 + 0 + 0.2 = 0.35; D3 = 0.3 + 0.25 + 0 = 0.55. Answer: D1 (0.60), D3 (0.55), D2 (0.35).

Worked example 2: change the weights

Now use weights of 0.5 for access, 0.2 for usage and 0.3 for quality. Working: D1 = 0 + 0.2 × 1 + 0.3 × 0.5 = 0.35; D2 = 0.5 × 0.5 + 0 + 0.3 × 1 = 0.55; D3 = 0.5 × 1 + 0.2 × 0.5 + 0 = 0.60. Answer: D3 (0.60), D2 (0.55), D1 (0.35). The ranking has turned completely round although no data changed. Weights are a judgement about what matters, so a good index states them openly and is tested for sensitivity.

Two further cautions. An index hides trade-offs: D1 tops the first ranking with the worst access of the three, so a single score should always be read alongside its components. And min-max scaling exaggerates small differences when all units are close together, because even a tiny gap becomes the full range from 0 to 1.

Shortcuts and traps

  • The best unit on an indicator always scores exactly 1 and the worst exactly 0; write these down first and compute only the middle values.
  • Min-max scores are relative: adding a new district with a more extreme value changes everyone else's scores.
  • With two units and two indicators, find the weight at which they tie by setting the two index expressions equal; above it one leads, below it the other.
  • Check direction: any 'lower is better' indicator must be turned round before weighting.

Key takeaways: normalise with min-max so indicators share a 0-1 scale; invert indicators where lower is better; weight and add; and remember that rankings can depend heavily on the weights and on which units are included.

Check your understanding

  1. 1. Across several districts an indicator ranges from 20 to 70. Using min-max normalisation, what is the score of a district with a value of 45?

    1. 0.64
    2. 0.5
    3. 0.25
    4. 0.9
    Show answer

    Answer: B. (45 − 20) ÷ (70 − 20) = 25 ÷ 50 = 0.5.

  2. 2. Complaints per 10,000 accounts range from 4 to 12 across districts. What is the normalised score of a district with 6 complaints, if lower is better?

    1. 0.25
    2. 0.5
    3. 0.75
    4. 0.6
    Show answer

    Answer: C. For a 'lower is better' indicator: (12 − 6) ÷ (12 − 4) = 6 ÷ 8 = 0.75.

  3. 3. A district's normalised scores are access 0.4, usage 0.8 and quality 0.5. With weights 0.25, 0.5 and 0.25, what is its index?

    1. 0.567
    2. 0.575
    3. 1.7
    4. 0.625
    Show answer

    Answer: D. 0.25 × 0.4 + 0.5 × 0.8 + 0.25 × 0.5 = 0.1 + 0.4 + 0.125 = 0.625.

  4. 4. A new district with far more outlets than any other is added to a min-max comparison. What happens to the existing districts' normalised access scores?

    1. They stay exactly the same
    2. They can change, because the maximum used in the formula has changed
    3. They all become 1
    4. They all become 0
    Show answer

    Answer: B. Min-max scores are relative to the minimum and maximum of the group; a new maximum lowers the scores of the others.

  5. 5. District X has normalised access 0.9 and usage 0.2; district Y has access 0.3 and usage 0.8. The index uses weight w on access and (1 − w) on usage. Above what value of w does X rank higher than Y?

    1. 0.4
    2. 0.6
    3. 0.75
    4. 0.5
    Show answer

    Answer: D. X = 0.9w + 0.2(1 − w) = 0.2 + 0.7w. Y = 0.3w + 0.8(1 − w) = 0.8 − 0.5w. Equal when 1.2w = 0.6, so w = 0.5; above it X leads.

Sign in to keep track of the lessons you finish.

Next lesson → Gaps, weighted averages and targets

Privacy · Terms · Refunds · Delivery