Scores and levels

The score

Every state and LGA has one score for each month, from 0 to 1.

The score is the expected share of malaria rapid diagnostic tests that would come back positive in that area, in that month. A score of 0.53 means the model expects about 53 in every 100 tests to be positive.

It is a model’s estimate. It is not a count of reported cases, and it is not drawn from health-facility records.

Note

The model learned from malaria test results collected in national household surveys, in which children aged 6 to 59 months were tested. Read the score as a measure of how common malaria infection is likely to be, not as a prediction of clinic attendance.

The likely range

Each score comes with score_low and score_high: the range the true value is likely to sit in.

{ "state": "Zamfara", "score": 0.55, "score_low": 0.423, "score_high": 0.677 }

The range is the model’s typical error, measured by training it on earlier national surveys and testing it against a later one, which it had no information about. Eight in ten results fell inside the range.

Area	Typical range
A state	The score, plus or minus 0.13
An LGA	The score, plus or minus 0.24

LGA ranges are wider because less survey data stands behind each one. The range is the same width for every month: it describes how accurate the model is in general, not how uncertain one particular month is.

Risk levels

level places a score in one of five bands:

Level	Score
very_low	0.00 to 0.20
low	0.20 to 0.30
moderate	0.30 to 0.40
high	0.40 to 0.50
very_high	0.50 to 1.00

The bands describe where an area sits among Nigeria’s LGAs on this model’s scale. Scores are estimates that have been pulled towards the middle so that they match what surveys measure on average, so very few areas score above 0.6.

A score exactly on a boundary belongs to the higher band.

The API always uses these bands. Dashboard users can set their own thresholds, so a level shown in someone’s dashboard may differ from the level the API returns for the same score.

Caution

Most bands are 0.1 wide and a state’s likely range is about 0.26 wide, so an area could reasonably sit in a neighbouring band. Use score_low and score_high when the difference matters.

State and national figures

A state’s score is the average of its LGAs’ scores, weighted by population. population is the number of people living in the area, from WorldPop.

Using scores well

• Compare areas and watch trends. The model is better at ranking places than at pinning down an exact value.
• Treat a score as one input alongside local knowledge and routine surveillance, not as a decision on its own.
• Read Limitations before using LGA-level figures for anything important.