Another update on the Performance of the UK Chasers: September 2026
/Back in 2020 I wrote a piece here about the comparative performance of UK and Australian Chasers, and I would refer you back to that piece if you want to know more about the data set I’m using for this new article. In November of 2025 I provided an update on the relative performances of the UK Chasers and in this blog I’ll be providing a fresh update that spans data from the first ever UK Chase show back on 29 June 2009, all the way through to 11 September 2026.
SUMMARY PERFORMANCE DATA
Let’s firstly take a look at some views of the raw data by Chaser.
On the right we have a table showing the number of episodes in which a particular Chaser took part in the Final Chase, along with a number of performance metrics.
The first thing to note is that the win rate of all Chasers has increased relative to last year - by a couple of percentage points for Darragh.
Anne Hegerty still has the highest proportion of wins, Jenny Ryan has now offered fewest pushbacks per episode and remains narrowly most accurate in Final Chases, and Mark Labbett still answers quickest in Final Chase.
Shaun Wallace has the lowest win rate, offers the most pushbacks per episode, and has the lowest Final Chase accuracy.
Darragh Ennis, however, remains narrowly the slowest to answer questions in Final Chase.
As we noted last year, a moment’s thought will flag that these performance metrics are likely to be influenced by a number of factors, perhaps foremost of which is the target set by the team for the Final Chase.
The table on the left breaks down each Chaser’s performance based on the target set, and we can see that all Chaser’s win rates except Jenny’s decline as the target increases, most notably for Darragh and Mark.
Darragh, however, still provides few pushbacks and is relatively accurate when the target is low, but falls off in accuracy and increases the tempo of his answering at a slower rate than most fellow Chasers as the target increases.
Shaun offers most pushbacks and is least accurate of all the Chasers across all target ranges, but he is also second quickest in providing answers when the target is high.
Anne’s response to an increase in target from low to high is still to offer about 80% more pushbacks but maintain a similar accuracy and increase the rate at which she answers questions by about 9% (from 4.9s per answer to 4.4s per answer).
Jenny remains particularly impressive in that she offers comfortably the fewest pushbacks of all Chasers for large targets and actually increases her accuracy, Her speed also increases by about 9% but only to 4.6s per answer.
Mark still offers over twice as many pushbacks when the target is large compared to when it is small and sees a 3% point decline in accuracy. He does, however, answer questions fastest at about 4.3s per answer.
Paul sees a similar increase to Anne in terms of pushbacks, but shows a slightly larger decline in accuracy, albeit to end at the same level as Anne. He is also the third fastest in terms of response time.
Another dimension that we can investigate is the number of contestants who make the Final Chase - a variable that is likely to be highly but not perfectly correlated with the target set (the correlation is +0.51).
We see a similar pattern of decline in performance here as we did in the previous table.
Anne continues to dominate in terms of win rate in episodes with a single contestant in Final Chase, Darragh when there are two contestants, Anne when there are three, and Jenny when there are four.
Shaun shows the greatest decline in performance as the number of contestants increases, and Jenny the least.
Paul shows the greatest increase in response rate as the number of contestants in Final Chase increases answering an additional question per minute when there are four compared to one player in that Final Chase.
(I should note that, in both this table and the previous one, Darragh still has relatively few episodes in some of the categories, which means that the various estimates of his performance are subject to relatively large standard errors.
The same is true here for Jenny, but to a lesser extent, with contestant counts of 1 and 4.)
STATISTICS FOR DIFFERENT-SIZED FINAL CHASE TEAMS
Before moving on to building a predictive model of Chaser performance, it’s interesting to look at performance metrics for Final Chase teams of different sizes. These appear in the table at right
You can see there that Chase teams’:
pushback success rate increases by about 50% (from 40% to 60%) for a team of four versus a solo team
average target increases by almost six questions for a team of four compared to a solo team
average prize fund more than doubles for a team of four compared to a solo team
win rate increases from about 10% to 40% for a team of four compared to a solo team
Interestingly, neither pushback success rates or average targets increase by much when moving from a team of three to a team of four, but the win rate increases by over one third from 31% to 42%.
PREDICTIVE MODEL
Ideally, we’d like to arrive at a measure of Chaser performance that takes into account the quality of the team that he or she has tended to face, and if we consider the average team size for Final Chase, percentage of successful pushbacks, average prize money fund, and average target set in the Final Chase, we can see that this quality has varied quite markedly across Chasers.
Mark and Paul appear to have faced the strongest teams in terms of pushback success, targets and prize funds.
The average size of the teams that Jenny has faced in Final Chase have been very slightly higher, but those teams have set lower targets and successfully pushed back at a lower rate. The same can be said for the teams that Darragh has faced, but they have set even lower average targets.
Anne has faced targets roughly as large, on average, as Paul’s, but with a lower rate of successful pushbacks.
Finally, Shaun has faced the weakest teams in terms of getting to Final Chase, and the second-weakest in terms of pushback success and target setting.
So, to account for the main sources of variation we’ll build a predictive model that incorporates the following variables:
The target faced
The number of players in the Final Chase (which will reflect an additional level of pressure and ability to secure pushbacks)
We’ll interact the target faced with Chaser to allow for the possibility that each Chaser might respond differently as the target increases.
The fitted predictive model, which is a binary logit designed to provide an estimate that a Chaser will be victorious given the target and the team size he or she faces in Final, is summarised in the table at left.
Speaking firstly at a high level, the negative coefficient on the PlayersInFinal variable tells us that Chasers are less likely to prevail the more players there are in the Final Chase, even after we adjust for the Target size. This is presumably because, as conjectured, the pushback success rate increases as the number of players in Final increases.
The EDF coefficients on each of the Chasers tell an interesting story about how they respond to changes in Target size and that is that the win probabilities for Darragh, Anne, Jenny, and Paul respond linearly to increases in Target, but Shaun’s and Mark’s respond very non-linearly, as we saw in the earlier table.
Using this model we can estimate each Chaser’s expected win rate for any sized target and any number of players in the Final Chase.
Since the coefficient on PlayersinFinal is quite small and does not vary by Chaser, we can choose to create a plot simply assuming that there are three players in the Final Chase.
That plot appears at right.
We can see that:
Mark slightly dominates for small to moderate targets but falls away for targets 19 and higher
Darragh starts to fall below his colleagues from targets sized about 14 onwards and has the lowest performance for targets above 19
Shaun has the lowest performance up until targets sized 19 but tapers off much more slowly than his colleagues for larger targets
Anne dominates for targets 19 and higher
Paul and Jenny have quite similar trajectories and are consistently in the top three across all target sizes
QUALITY OF MODEL FIT
Before moving on, we should check to see how well the model fits the actual data in terms of individual Chaser win rates as the target varies.
The table at right reveals that the model consistently produces win rate estimated within a couple of percentage points of the actual observed values except for Darragh in games with large targets.
For those games, the model underestimates Darragh’s win rate by about 2.6% points (Note that Darragh has only had 18 such episodes so far).
The only other, non-trivial differences are for
Mark and large targets (1.8% point overestimate)
Jenny and small targets (2% point underestimate)
Overall, the model fits all Chasers and target ranges exceptionally well.
PERFORMANCE OVER TIME
The model we fitted for the previous sections includes all of the episode data for every Chaser. We can enhance that model by including a term that allows us to estimate a separate smooth career trajectory for each Chaser.
We then use the model to estimate win probabilities for each Chaser, basing every estimate on the same overall mix of targets and player counts.
The results are shown in the plot at right and suggest that:
Anne started with an extremely high win probability which tapered off over the first 150 episodes or so at which point it reached a level she has since maintained.
Darragh started at about the level Jenny is now at but has declined fairly uniformly since. Note, however, that his small number of episodes means the confidence bands around his estimates are large.
Jenny started off relatively poorly before reaching her peak level around her Episode 100 before tapering off slightly since then.
Mark started off with high estimated win rates, gradually declining until about his Episode 300 after which his estimate has been fairly stable.
Paul also started off with high estimated win rates and saw a gradual decline until about his Episode 200 then saw a gradual increase until about his Episode 350 after which there has been a gradual, small decline.
Shaun started off with what would turn out to be relatively high estimated win rates that fell away until about his Episode 125 after which he attained his steady state about 20% points below his colleagues. Note that there have been signs of a slight increase over recent episodes.
MODEL CAVEATS
The key assumptions we are making in using this model (and the earlier model) are that:
1. The outcome is correctly modelled as binary - each Final Chase is treated as a win or less with the model estimating its probability. This ignores the margin of victory, time remaining, questions answered, pushbacks, and other information about how convincingly the Chaser performed.
2. Episodes are independent conditional on the model - the model treats episode results as independent after accounting for the included variables. In reality, adjacent episodes may be correlated because of filming blocks, temporary form, illness, question-set difficulty, production changes, or other short-term effects. If such clustering exists, the confidence intervals could be too narrow.
3. Episode count is an adequate representation of time - ability is modelled against the number of episodes completed by that Chaser, not calendar time. This assumes experience accumulates primarily through appearances, a gap of several years has no effect beyond the episode number, Episode 100 has a comparable interpretation across Chasers, even if one reached it much faster, and age, career breaks and changes in filming frequency do not need separate treatment.
Thus, the trajectory is best interpreted as ability over career experience, not necessarily ability over chronological time.
4. Ability changes smoothly - the GAM assumes each Chaser’s trajectory is reasonably smooth. It discourages sudden jumps unless the data provide very strong evidence for them. This is appropriate for gradual learning or decline, but it may obscure real discontinuities caused by rule changes, illness or career breaks, changes in question writers or production, changes in recording conditions. or a sudden improvement in preparation.
The amount of smoothness is selected from the data using penalization.
5. Player-count effects are constant across Chasers and time - the Players in Final variable is treated categorically, so it does not assume a linear effect of adding players. However, it does assume the effect associated with each finalist count is the same for every Chaser and throughout the programme’s history.
6. Effects are additive on the log-odds scale - the Chaser, target, player count and career trajectory effects are added together on the logistic scale. Apart from the separate time trajectory for each Chaser, there are no interactions.
7. Target and player count adequately capture episode difficulty - calling the remaining trajectory “ability” assumes other determinants of difficulty either do not change systematically over a Chaser’s career, affect all Chasers similarly, or amount only to random noise.
Potential omitted factors include question difficulty, strength of the contestants beyond their target, pushback opportunities and conversion, rule or format changes, series and production era, filming blocks, prize money or tactical behaviour.
If these factors change over time, part of their effect may be attributed to changing Chaser ability.
8. Target and player count are valid adjustment variables - the adjustment assumes it is meaningful to compare Chasers while holding the distribution of target and player count constant.
That produces estimates that answer the question: “How likely would this Chaser have been to win at this point in their career if they faced the programme’s overall historical mix of targets and finalist counts?”
This is not the same as their actual win rate at that time.
9. The common standardization population is appropriate - every Chaser is evaluated against the overall dataset’s target/player distribution. This improves comparability but assumes that this synthetic common schedule is relevant to all Chasers. Some combinations may be rare or absent during a particular Chaser’s era. The model may therefore make predictions for situations that Chaser seldom encountered, relying partly on extrapolation from other episodes.
10. Career selection is ignored - Chasers are not observed for a random number of episodes. Continued appearances may depend on performance, availability, health, popularity or production decisions.
The model describes observed careers; it does not correct for survivorship or selection.
11. The recorded variables are accurate and consistently defined - the model assumes that episodes for a Chaser are aired at least roughly in the same order as they were recorded, target is comparable throughout the show’s history.
12. The confidence intervals reflect model uncertainty, not every source of uncertainty - the ribbons use an approximate delta-method calculation based on the fitted GAM’s coefficient covariance matrix. They reflect uncertainty under the chosen model, conditional on the model specification being correct, the selected smoothing structure, the observed target/player distribution.
They do not fully represent uncertainty from model selection, omitted variables, data errors, or alternative reasonable definitions of ability.
SO, WHO’S BEST?
Using the output from the model we fitted earlier on each Chaser’s entire career, we can assess the best Chasers as being:
For Small Targets (12 or less): any Chaser except Shaun
For Medium Targets (13-17): Mark
For Large Targets (18+): Anne
If we fit the same model but constrain it to use only the most recent 100 episodes for each Chaser, we get the plot at right and the choice for bests becomes:
For Small Targets (13 or less): Mark or Paul
For Medium Targets (14-17): Mark
For Large Targets (18+): Anne and Jenny (and Mark for exactly 18)
