The basics
Elo in plain English
Every club has a rating. The average is 1500. Before each match, we compare the two ratings (plus home advantage) to get each side's expected result. After the match, the winner takes rating points from the loser. An unexpected result moves more points than an expected one, and bigger winning margins count for more. A draw moves a few points towards the weaker side.
Settings
Our settings
- Starting rating
- 1500 for every club in 2022-23, the first season in our data
- K factor
- 45 (how fast ratings move)
- Home advantage
- 100 rating points (none at a neutral ground)
- Winning margin
- 1 for a margin of 0-1, 1.5 for 2, (11 + margin) / 8 for 3 or more
- New season
- Every rating moves one fifth of the way back towards 1500. Promoted clubs start at the average rating of the clubs relegated that season.
- Awarded results
- Count in the table but do not change ratings.
Home advantage in the Ligi Kuu Bara is modest. Across our data, home sides win about 47% of matches, away sides about 27% and 26% end level. That is worth 100 rating points here (none at a neutral ground).
Predictions
Match predictions
The rating gap gives each side's expected result. We then estimate the chance of a draw from how often Ligi Kuu Bara matches with a similar rating gap ended level, and split the rest between a home and an away win.
Accuracy
How accurate is it?
We tested the model on 480 matches from 2023–24 and 2024–25, predicting each one using only the results before it (2022–23 was used to warm the ratings up). We chose the settings with the best score.
- Log loss
- 0.921 (lower is better)
- Brier score
- 0.543 (lower is better)
- Without ratings
- 1.075 log loss, using only the league's average home, draw and away rates
The ratings beat the league averages by a clear margin. That is partly because Tanzania's three strongest clubs (Simba, Young Africans and Azam) are far ahead of the rest, which makes many matches easier to call. We tuned the settings on these same matches, so the true edge on future matches will be a little smaller, and a different 2026–27 will not behave exactly like the past two seasons. Predictions are a guide to how the clubs compare, not a forecast of any one match.
Calibration
Each dot groups predictions of a similar probability. Dots near the dashed line mean the percentages can be taken at face value; bigger dots cover more predictions.
Data
The 2025–26 season
For 2025–26 our source gives every score but not the match dates, and Elo needs results in order. So we replayed that season in 500 different random orders and averaged each club's final rating. The order made little difference: each club's rating varied by about 12–36 points across the replays. On the team charts that season is shown as a single dashed step.
Simulator
The season simulator
We play out every remaining fixture 10,000 times. In each run, every club's rating is first nudged up or down at random, because no rating is exact and form changes during a season. The size of the nudge comes from how much ratings really moved in past seasons (about 108 points over a full season) and shrinks as fewer matches remain. Scorelines are drawn from real Ligi Kuu Bara results. Clubs level on points are separated as the Tanzania Premier League Board's rules require: head-to-head points, then overall goal difference, goals scored, head-to-head goal difference and head-to-head goals scored (a random draw stands in for a play-off). The bottom two clubs are relegated, the 13th and 14th-placed clubs go into a play-off, and the top two qualify for the CAF Champions League.