# Combine several measures into one score per area, to target a programme

Build a weighted score from several census measures for every area, rank and map it, write down the method, and test how much your choices matter.

Video: Combine several measures into one score per area, to target a programme (1 min 28 s): /images/docs/videos/composite-index-by-area.mp4

## Best for

Public health teams, planners, housing and community organisations who have to decide where a programme goes first: cooling centres, outreach, grants, inspections. You end up with a score, a rank and a class for every area, a map, the method written down so it can be defended, and a check of how much the ranking moves if the weights or the scaling change.

**The example in this guide** builds a heat vulnerability score for the 184 census tracts of New Orleans (Orleans Parish, Louisiana) to target cooling centres. Nothing was attached: the four measures are the US Census Bureau's American Community Survey 2020 to 2024 5-year estimates. Results vary from run to run, so treat the figures here as one example.

## What you need

- **The measures you want to combine,** named. Mapstack can fetch official statistics itself, or you can attach a table with a column for each measure.
- **The areas,** such as census tracts in a city.
- **Your weights,** if some measures should count for more.
- **A Free, Creator or Creator Max seat.**

## Before you start

- **Say how measures are put on a common scale.** A **composite index** adds measures together, so each must be on a comparable scale first. Shares (0 to 100%) are on the same scale but not the same spread: in the example poverty runs from 3 to 84% across the tracts and age 65 and over from 0 to 51%, so poverty counts for more than its weight says. **Min-max scaling** stretches each measure to run from 0 to 1 across the areas first. Left unsaid, Mapstack uses the shares as they are.
- **Survey estimates carry error.** ACS figures for small tracts are less certain. A tract of 800 people near the top of a ranking deserves a second look before money follows it.

## The prompt

> We need a heat vulnerability score for each census tract in New Orleans to target a cooling-centre programme. Combine the share of people in poverty, households with no vehicle, people aged 65 and over and people with a disability, from the latest ACS 5-year estimates. Weight poverty and no vehicle double. Map it in quintiles and give me the table with ranks.

## What to expect

1. **It fetches each measure and builds the score**

   Mapstack fetches each measure for every area as a dataset of its own, works out each share, and combines them with your weights: here twice poverty plus twice no vehicle plus age plus disability, divided by six. The result is one dataset with every measure, the score, the rank and the class. It says which areas it could not score and why. In the example that was 4 of the 184 tracts: three with almost no one living there (9800, 9801 and 9900), and tract 145, where all 1,822 residents live in an institution, which the survey's poverty, household and disability counts leave out.

   In the example the highest score was 0.527, in tract 6.20, where 62% of people were below the poverty line, 64% of households had no vehicle, 29% of people were 65 or over and 35% had a disability. The 36 highest-scoring tracts form the top quintile.

2. **It maps the score in quintiles**

   **Quintiles** split the areas into five equal-sized groups by score, so each class holds 36 of the 180 scored tracts. The map shades each tract by its class, with a search by tract and the full table.

   ![Example: heat vulnerability by census tract in New Orleans, in quintiles, darkest red highest. The long dark tract in the east is 17.51, a large area with 887 residents.](/images/docs/composite-index-by-area/new-orleans-heat-vulnerability-quintiles.jpeg)

3. **You ask for the method in writing**

   A score someone else will act on needs its method on paper:

   > Write down the method: each indicator, its year and table, how you scaled them, the weights, and how the classes were set.

   Mapstack gives each measure's ACS table and the columns used for its top and bottom line (poverty from B17020, no vehicle from B25044, age from B01001, disability from C18130), the formula with the weights, how the quintiles were cut with the score range of each, and how areas with no score were handled. On scaling, it says the shares were used as they are.

4. **You test how much the ranking depends on your choices**

   Weights and scaling are judgement calls, so test them:

   > How much does the top 20 change if all four weights are equal? And separately, if each indicator is first rescaled min-max across the tracts, so that age and disability count as much as their weights say?

   In the example, 16 of the top 20 stayed with equal weights and 18 with min-max scaling. The same 11 tracts made up the top 11 under all three versions, in a slightly different order, so they are safe choices whatever the method. The tracts that moved are where the choice matters: tract 38, where 51% of residents are 65 or over, rose from 29th to 14th under both alternatives. Mapstack names each tract that entered or left and why.

## Common questions

**Can I change the weights?**

Yes. Say them in the ask, for example "Weight poverty and no vehicle double". To see how much they matter, ask "How much does the top 20 change if all four weights are equal?": in the example, 16 of the top 20 stayed.

**Can I stop one measure outweighing the others?**

Yes. Ask for each indicator to be "rescaled min-max across the tracts" before the weights are applied. Left unsaid, Mapstack uses the shares as they are, so a measure with a wider spread counts for more than its weight says.

**Should I trust a small area near the top?**

Treat it as a lead to check. Survey estimates for small populations are less certain, so ask for each area's population beside its score. In the example, the third-ranked tract, 48, has 823 residents.

**Why does an area have no score?**

It has almost no residents, or no data for one of the measures. Mapstack says which areas it left out: in the example, three tracts with almost no one living there and one where everyone lives in an institution.

**Can I get the method in writing?**

Yes. Ask: "Write down the method: each indicator, its year and table, how you scaled them, the weights, and how the classes were set." Keep it with the map.

## Related

- [Choose where to go next](https://mapstack.io/docs/answers/where-to-go-next.md)
- [Get census and official statistics](https://mapstack.io/docs/find-data/census-and-statistics.md)
- [Map an official statistic area by area](https://mapstack.io/docs/guides/map-an-indicator-by-area.md)
- [Make a map](https://mapstack.io/docs/make/maps.md)
