Article summary
Separate Lesotho +266 from South Africa +27 and control gender/age observations with minimum cells, missingness reports and purpose limits.
Lesotho is surrounded by South Africa, but a Lesotho phone is not a subtype of a South African phone. A WhatsApp Gender and Age project must separate +266 from +27 before checking. Customer market, commuting or cross-border orders belong to another evidence layer; geography cannot replace numbering rules.
Four evidence layers in one record
| Layer | Example | Purpose |
|---|---|---|
| Numbering plan | +266 or +27 | Format and task batching |
| Business market | Order, address, self-report | Service context |
| Profile observation | Age, gender, avatar | Controlled aggregation |
| Permission | Sender, channel, purpose | Whether contact is allowed |
A small market needs larger reporting cells
Crossing district, age band, gender and order may leave only a few people, allowing an aggregate to point back to an individual. Use broad age bands, combine sparse groups and enforce a minimum cell. Suppress a cell below threshold rather than hiding sample size behind a percentage.
Report missingness before distribution
Compare age, gender and avatar unknown rates by +266 source, list age and customer stage. An analysis of populated rows alone hides systematic absence. When missingness bias cannot be explained, do not generalize the finding to Lesotho customers.
What Gender and Age fields are not
WS Gender and Age may return phone, age, gender, avatar and mapped WhatsApp phone. These are limited observations, not date of birth, legal gender, self-identification, citizenship or confirmation of the account holder.
Keep the source list inside the organization
Upload TXT with one phone per line. member_id, address, order and permission remain in a private crosswalk. Excel joins a research staging area through batch_row, while analysts receive an aggregate table without direct access to phones and avatars.
How to detect a cross-border mismatch
If market=Lesotho and number_plan=+27, do not rewrite the phone automatically. It may be a legitimate cross-border relationship or a provenance problem. Mark country_conflict and let business evidence decide instead of changing digits to make two columns agree.
Exclude high-impact decisions
Inferred profiles must not decide employment, credit, insurance, price, identity or service eligibility and cannot reliably identify minors. Language and form of address come from user choice. Opt-out always outranks a profile observation.
When the profiling study should end
Record the research question, sample scope, checked_at and delete_after. Once the question is answered, delete person-level inference and retain only aggregates resistant to re-identification. This Lesotho page earns its place by explaining small-sample risk, not promising finer labels.
