PMT score methodology
The PMT score (Proxy Means Test) estimates a household's standard of living from observable characteristics, without measuring its income directly. It is the method of choice when income is hard to declare and to verify, as in a largely informal economy.
This page describes the computation as implemented in the software. The coefficients and thresholds are taken from the code.
How to read a score
The score is the logarithm of the household's estimated per-capita consumption. In practice it lies between 11 and 14.
- A low score means a poorer household.
- A high score means a better-off household.
This is the convention of World Bank surveys, and it is the convention of the whole application: lists, detail pages, filters and eligibility rules always compare the score below a threshold to identify poor households.
The score belongs to the project
The computation relies on a PMT method: a set of coefficients and thresholds identified by a code and a name. All methods are managed in Settings → PMT methods, and nowhere else.
A method is assigned to a project. A household is scored with the method of the project it is selected in; a household that is in no project, or whose project has no method, has no score. There is no "national" score and no default method.
| Level | Role | Where to set it |
|---|---|---|
| Project | Score of every household of the project, with rank and percentile | Settings → PMT methods, action "+ project" |
| Programme | Recomputes the score for that programme's evaluation only | Programmes → programme form, field "PMT method" |
Resolution order is programme → project. A programme's detail page shows which method applies and where it comes from ("own to the programme" or "inherited from the project").
The method shipped is GABU_BASELINE, "PCH PMT method (Gabu Baseline)", calibrated on the Gabú household survey (2023, World Bank). It is assigned to the PCH project, and to the PRSSB project for its historical scores.
The computation
The score is the result of a linear regression:
score = constant + Σ (coefficient × value observed for the household)
Each variable is derived from the household questionnaire (composition, housing, assets). A negative coefficient pulls the score down, that is towards poverty; a positive coefficient pulls it up.
The 25 variables and their coefficients (GABU_BASELINE)
| Variable | Coefficient | What it measures | How it is derived |
|---|---|---|---|
| constant | +13.410 | starting point of the model | — |
log_taille_menage | −0.570 | log of household size | declared size |
head_female | −0.129 | female household head | head's sex |
head_dailyjob | +0.057 | head is a day labourer | main occupation = day labourer |
share_under5_f | −0.390 | share of young girls | children aged 0–10 ÷ size, half girls |
share_under5_m | −0.269 | share of young boys | children aged 0–10 ÷ size, half boys |
share_over65_f | −0.678 | share of women aged 65+ | persons 65+ ÷ size, half women |
share_over65_m | −0.412 | share of men aged 65+ | persons 65+ ÷ size, half men |
roof_natural_material | −0.056 | roof of natural material | straw, cibe palm, tarpaulin |
floor_natural_material | −0.078 | floor of natural material | packed earth, clay |
water_well | +0.041 | well water | protected or unprotected well |
light_flashlight | −0.074 | precarious lighting | oil lamp, torch, candle |
cooking_fuel_charcoal | +0.358 | cooking with charcoal | fuel = charcoal |
moto | +0.127 | owns a motorbike | assets |
car | +0.333 | owns a car | assets |
bicycle | +0.026 | owns a bicycle | assets |
bed_wood | +0.056 | wooden bed | assets |
foam_mattress_percap | +0.252 | foam mattresses per person | mattresses without springs ÷ size |
spring_mattress_percap | +0.992 | spring mattresses per person | spring mattresses ÷ size |
iron_traditional | +0.101 | charcoal iron | assets |
sewing_machine | +0.248 | sewing machine | assets |
computer | +0.119 | computer | assets |
flashlight | +0.035 | torch | assets |
land | +0.010 | land (hectares) | declared area |
poultry | +0.003 | poultry (count) | declared headcount |
By far the most decisive variable is household size: with the same equipment, a large household consumes less per person.
The questionnaire does not record the sex of children or of elderly members: the shares by sex are split equally. Children aged 0–10 stand in for the under-fives of the original model.
Thresholds and categories
Each method carries its own thresholds, on the same scale as the score. Those of GABU_BASELINE:
| Category | Condition | Threshold |
|---|---|---|
| Extreme poverty | score < extreme threshold | 11.5 |
| Severe poverty | score < severe threshold | 12.0 |
| Moderate poverty | score < moderate threshold | 12.5 |
| Vulnerable | score < vulnerable threshold | 13.0 |
| Not poor | above | — |
The method's eligibility threshold is 12.5: a household whose score is strictly below it is eligible in the PMT sense. It coincides with the moderate poverty line.
Thresholds are edited on the method, in Settings → PMT methods, and the project is then re-evaluated. A programme can also set its own bounds (minimum and maximum PMT threshold in its form): they then take precedence over the method's.
Reading a household's breakdown
A household's detail page shows, under the score, how it is composed: each line carries the observed value, its coefficient and the resulting contribution. The sum of the contributions and the constant gives back the recorded score, so any computation can be checked.
Variables are sorted by decreasing contribution and the eight most decisive are displayed: the others, often zero-valued, explain nothing of the result.
When the score is computed
- When a household enters the registry (data entry, import, Kobo collection): it receives the score of its project's method.
- On integration into a project's registry: the score is established with the project's method and stored with its rank and percentile in that project.
- On programme evaluation: if the programme has its own method, the score is recomputed with it for that evaluation only; the project score is left unchanged.
- After a change of coefficients or thresholds: Settings → PMT methods, "Re-evaluate" button on the project concerned. Every household of the project is rescored and the evaluation date is shown next to the project.
The score is never recomputed just by opening a detail page. An archived project is read-only: its scores are never re-evaluated.
Other method types
A method can be of type regression (the case described here), hybrid (same coefficients, computed the same way) or grid, a legacy scoring mode by categories on a 0–100 scale. The grid is no longer used by any project and its thresholds are not comparable with the regression's; it is kept only to read old scores.
What the score does not do
The PMT score does not decide eligibility on its own. It feeds each programme's evaluation, which may add its own rules: geographic criteria, household composition, target groups, specific vulnerabilities. A household can be eligible for one programme and not for another with the same score.