Lesson 8 of 8
Unit · What the numbers mean
The section that lists what you refused
A protection report needs a block naming the analyses that were requested and declined, and what was offered instead. It belongs in the document rather than in an email, because the request will be made again by someone who has not read the email.
The report, whole
Protection and GBV data, 2024
Three admin1 areas, six admin2 areas
DATA HANDLING read this first
Analysis extract: 14 fields, 1,850 cases. No name, contact detail, free
text, incident date, location below admin2, exact age, incident type or
perpetrator detail. Age in five bands, time in months.
Suppression rule: no cell below 5, with secondary suppression so rows
cannot be differenced.
Incident-level analyses remain with the case management agency and are
reported as findings, not shared as data.
REFERRAL PATHWAY 1,638 consenting cases
Consented to referral 88.5% 1,638 of 1,850
Declined 11.5% 212 a decision, not a failure
Referral made 69.7%
Accepted, of those made 66.3%
Reached service, of accepted 94.7%
End to end, of consenting 43.8%
The pathway fails upstream. Once accepted, 94.7% reach a service.
Livelihood support completes at 20.7% against health at 57.6%.
EQUITY
Cases reporting a disability 26.7% complete, against 46.2%
Consent is identical (89.0% and 88.5%). The gap is entirely at
referral-making (54.5% against 71.9%) and acceptance (56.4% against 67.3%).
Disability is self-reported by 12.3% of cases, so the gap is a lower bound.
CASE MANAGEMENT 1,108 cases, 17 caseworkers
Mean caseload, highest area 33.2 peak 51; guidance is about 25
Mean caseload, lowest area 14.8
Case plan reviews per case, highest-caseload area 1.78 against 2.45
Closed for lost contact, highest-caseload area 38.9% against 20.2%
Closed by the cut-off 636 57.4%
Still open 472 42.6% censored, not missing
Median months to closure 4 of closed cases only;
140 open cases already
exceed it
Case plan objectives met 29.1% of closures
WHAT THESE NUMBERS ARE NOT
Case counts measure access and reporting, not incidence. Saint-Louis-du-Nord
opened a service in July and its monthly intake rose 74%; no conclusion
about violence in that area follows.
Prevalence is not estimable from service data at any level of
sophistication. It requires a population-based survey.
REQUESTED AND DECLINED
Case counts by commune, month and incident category — declined; cells of
0 to 3 identify individuals. Offered: district by quarter, and commune-level
completion rates on annual denominators.
Full case list with age and area for a partner's targeting exercise —
declined; the extract is not a beneficiary list and consent does not cover
it. Offered: aggregate caseload by area to inform their staffing.
DATA QUALITY
62 cases (3.4%) have no age band; age-disaggregated figures use 1,788.
11 cases record a service time on an unaccepted referral; excluded, and
the completion rate is computed without them.
7 cases close before they open; excluded from duration statistics.
One caseworker identifier spans two areas after a transfer; caseload is
computed per worker.
One area records disability as Yes/No rather than true/false; normalised.Why the declined section is in the document
Because the request will be made again. By a new information manager, by a donor’s consultant, by a colleague who was not copied on the email. A refusal that lives in a mailbox has to be re-argued every time, usually by someone junior, under time pressure, to someone senior.
Because it shows the refusal was reasoned rather than reflexive. A report that lists two declined requests alongside what was offered instead reads as professional judgement. A report with no such section, from an analyst who says no in meetings, reads as obstruction.
Because it is a design brief. Two years of declined requests is a specification for what the information system should be able to produce safely, and nobody collects that unless it is written down somewhere durable.
declined = [
{"request": "counts by commune, month, incident category",
"reason": "cells of 0-3 identify individuals",
"offered": "district by quarter; commune completion rates"},
{"request": "full case list with age and area",
"reason": "not a beneficiary list; consent does not cover it",
"offered": "aggregate caseload by area"},
]# Keep the log as data. It is the only way it survives a staff change.The four sentences that carry the report
Everything above is evidence. A reader takes away four things, and they should be written before the tables rather than extracted from them.
The pathway fails before the service door. 94.7% of accepted referrals reach a service; the losses are at referral-making and acceptance, which are two conversations with two named parties.
People with disabilities are failed by the system, not by their own choices. Identical consent, nineteen points of completion gap, entirely at the two gates the service system controls.
One area is carrying a third more cases than the guideline and it shows in two other tables. Caseload is the cause, and reviews per case and lost contact are the symptoms.
No number here measures how much violence occurred. Every denominator in this report is people who reached a service.
The habit this course was for
Three questions, in the platform’s usual order, with one addition specific to this sector.
- What did I count, over what? Consenting cases, accepted referrals, open cases, closures — four denominators and none of them is the population.
- What else could explain it? A service opening, a coding habit, a caseworker’s judgement about what “administratively closed” means.
- What should someone do differently because of this? Two conversations about referral-making, one about establishment in Port-de-Paix, one about closure coding in Hinche.
- Who could be harmed if this were published as it stands? The question the other three do not ask, and the one that governs whether the answer to them ever leaves the building.
The fourth question is the whole of this course, and it is the only one on this platform where the right answer is sometimes to produce nothing.
What comes next
Module 4 has one course left. Education attendance and learning outcomes bring a new sector and, for the first time in this module, a dataset the platform does not yet have — which is a content problem before it is an analytical one.