Dataset
Protection referral pathway monitoring — 2024
1,850 synthetic protection cases tracked from intake to first service, modelled to teach referral pathway performance under the data protection standards that govern this sector.
syntheticProtectionProtection monitoringCase managementReferral trackingGender-based violence (GBV)Child protectionPersons with disabilities
- Rows
- 1,850
- Variables
- 14
- Period
- 2024-01 to 2024-12
- Licence
- CC BY 4.0
- Completeness
- 94%
Standards and methodologies
Core Humanitarian Standard (CHS)Sphere Standards
Files
Files are versioned by filename. A corrected release ships as .v2.csv rather than replacing the file in place, so an analysis pinned to v1 keeps reproducing.
Data dictionary
| Variable | Type | Description | Allowed values |
|---|---|---|---|
case_id | string | Pseudonymous case reference, sequential and unlinked to anything outside this file. | — |
admin1 | categorical | First-level administrative area. No location below admin2 is recorded anywhere in this dataset. | — |
admin2 | categorical | Second-level administrative area, the finest geography this dataset carries. | — |
referral_month | string | Month the case entered the referral pathway, as YYYY-MM. No exact date is recorded. | — |
case_category | categorical | Deliberately coarse case category. No incident type, perpetrator detail or narrative is recorded. | gbv, child-protection, general-protection |
age_band | categorical | Age band of the person concerned. Exact age is never recorded. | 0-11, 12-17, 18-24, 25-59, 60+ |
sex | categorical | Sex of the person concerned. | f, m |
disability_reported | boolean | Whether a disability was reported at intake. Recorded with a different code in one area. | true, false |
service_requested | categorical | Service the case was referred toward. | health, psychosocial, legal, safety-and-security, livelihood-support |
consent_to_refer | boolean | Whether informed consent to refer was given. A case without consent is not a pathway failure and must be excluded from the completion denominator. | true, false |
referral_made | boolean | Whether a referral was actually issued to a service provider. | true, false |
referral_accepted | boolean | Whether the receiving provider accepted the referral and the person reached the service. | true, false |
days_to_first_servicedays | integer | Days from referral to first service received. Blank where the referral was not accepted, or where the date was never recorded. | — |
case_status | categorical | Status of the case at the end of the reporting year. | open, closed-resolved, closed-lost-contact, transferred |
Provenance
Data quality
- Every record is generated. No real person is described, and this file must never be used as a template for storing real case data — the safe version of that is a consent-governed case management system, not a CSV.
- What this dataset deliberately omits is the lesson. There are no names, no contact details, no free text, no incident date, no location below admin2, no exact age, no incident type and no perpetrator detail. None of those are required to measure whether a referral pathway works, and under GBV information management principles the incident-level fields are never shared outside the case management agency at all.
- Collecting less than you could is the discipline being modelled. `case_category` is coarser than any caseworker would use, because a pathway performance analysis does not need the detail and holding it creates risk with no analytical return.
- Consent gates the pathway. The completion denominator is cases that consented to referral, not all cases. Counting a non-consenting case as a pathway failure both misstates performance and misrepresents a person's decision.
- Reference figures: 88.5% of cases consented, and 46% of consenting cases reached a service. Completion ranges from about 23% for livelihood support to 62% for health, and from 31% to 57% across the six admin2 areas.
Known issues
- Cases where a disability was reported complete at about 31% against 48% where none was reported. This gap is deliberate and is the equity finding the dataset exists to surface.
- Sixty-two cases have a blank age_band. It is the field most often missing in real intake data and the one most disaggregations depend on.
- One area recorded disability as Yes and No rather than true and false, which fragments the disability disaggregation into four categories, two of them too small to interpret.
- Twenty-one cases record a time to first service even though the referral was never accepted, a logical contradiction that must be resolved before any completion rate is trusted.
- Forty accepted referrals have no time to first service recorded. The timeliness denominator is therefore smaller than the completion denominator, and using one for both misstates both.
Worked examples
Python
Referral completion and where the pathway breaks
Builds the consent-gated completion rate, decomposes it by service and area to locate the failing node, and tests the disability gap for significance.
Described here; the notebook is still being written.
Timeliness against the clinical standard
Measures the share of GBV health referrals reaching care within 72 hours, and shows how the figure changes depending on how missing service dates are treated.
Described here; the notebook is still being written.
R
Referral completion and where the pathway breaks
The same decomposition in dplyr, with the contradictory and missing records handled explicitly rather than dropped silently.
Described here; the notebook is still being written.