Methodology

Cape Town Ward Safety Index v2.0

Every one of Cape Town's 116 electoral wards carries a safety score from 0 to 10 (10 = safest). The score is not a single opinion: it is the weighted sum of five named pillars, each scored on the same scale, each published with its evidence, its basis and a confidence grade. This page is generated from the same code that computes the scores, so what you read here is what the map does.

Open the ward map District crime reports

1. The five pillars and their weights

Violent crime carries the largest weight because it is the risk that most separates Cape Town's wards from one another and the one visitors and residents weigh most heavily. Organised crime is scored separately from violence because gang control can suppress street violence while making an area more dangerous, not less (Hangberg in Hout Bay is the textbook case).

2. Score bands

• Very safe 8.0–10.0: Normal city awareness.
• Safe 6.0–7.9: Normal urban precautions; keep valuables discreet.
• Moderate 4.0–5.9: Exercise caution, especially after dark.
• Caution 2.0–3.9: Avoid at night; visit only with local knowledge.
• Avoid 0.0–1.9: Not recommended for visitors.

3. How each pillar is scored

1. Organised crime & gangs (absolute). Each ward's gang-activity grade comes from the curated GI-TOC Western Cape Gang Monitor table and maps directly to a score: war → 0high → 2active → 4low → 6none → 10
2. Headline score (researched). Every ward has an individually researched score, refreshed after each quarterly SAPS release, with its reasoning, sources and a list of tagged crime factors published in the ward record. That score is the calibrated headline.
3. Crime pillars (decomposed). The headline's crime component is the baseline for the four crime pillars, and each tagged crime factor moves the pillar it belongs to by a fixed amount. The offsets are then re-centred so the weighted composite reproduces the headline exactly. The pillars therefore describe the ward's crime profile (what kind of crime is the problem here) without ever contradicting its headline.
   baseline m = (headline − 0.20 × gangs) / 0.80
   pillar_i = m + offset_i − weighted_mean(offset)

   Factor rules (first match wins; tags qualified 'rare' or 'occasional' count +0.5 instead):

   • murder|killing|shooting|gang violence|gang activity|gang spillover|vigilante|taxi-conflict → violent -1.0
   • armed robbery|armed house robbery|violent house robbery|assault|extortion → violent -0.75
   • carjack|vehicle|car break|car theft|theft from cars|theft from parked → vehicle -0.75
   • house robbery|burglar|house break|stock theft|industrial theft → property -0.75
   • robber|snatch|pickpocket|mugging|street crime|station crime|theft|catwalk → street -0.75
   • Accumulated offsets are capped at −1.5 (concern) and +1.0 (relief) per pillar. Drug-related tags are deliberately unmapped: SAPS drug counts measure police activity, not victim risk.
4. SAPS station evidence. Where a ward's police precinct has a district crime report, the latest official SAPS category counts and year-on-year trends are attached to the matching pillar and shown on the map. Evidence is displayed, not re-scored: the researched headline already reflects those releases, and scoring them a second time would double count.
5. News-sentiment modifier. City-wide ±0.3 on the overall score from the week's news sentiment; pillars unchanged.

4. Basis and confidence

5. Worked examples

6. Limitations

• SAPS publishes statistics per police station, not per electoral ward. A ward inherits its precinct's picture; sub-areas inside a ward can differ sharply (Hangberg versus the Hout Bay harbour, Joe Slovo Park versus coastal Milnerton). Read the ward reasoning for the sub-area split.
• Station counts are not population-normalised: the City Bowl records the most theft in the country partly because it hosts the most people and businesses. Counts are shown as evidence, not converted to rates.
• SAPS figures are recorded crime. Under-reporting is highest for the crimes that most affect residents of gang-affected wards.
• Crime pillars are analyst decompositions of the researched score, calibrated by the crime-factor rules; they are not independent measurements and should be read as relative profiles within a ward.
• The news-sentiment modifier is city-wide (±0.3) and reflects the tone of the week's reporting, not a measured change in crime.

7. Changelog