How to get better at pub quizzes

Pub quiz performance depends on two capacities: the size of a person's store of retrievable general knowledge, and the reliability with which that store can be accessed, unprompted and under time pressure. Both respond to training. The methods that work are not specific to quizzing — they come from the experimental literature on human memory, where they have been replicated for over a century. Set out below are the findings that bear on preparation, the evidence for each, and what each implies in practice.

1. Retrieval practice outperforms review

Attempting to recall a fact strengthens later memory for it more than re-encountering the same fact does. This is the testing effect. Roediger and Karpicke (Psychological Science, 2006, 17(3), 249–255) had participants either re-read a prose passage or test themselves on it for equal time; the tested group recalled substantially more a week later, and the advantage widened as the retention interval lengthened. The same experiments found that participants predicted the opposite outcome. Re-reading felt more productive at the time and was not.

The distinction that matters for quizzing is between recognition and recall. Selecting an answer from four options is recognition. A pub quiz requires production from an unprompted question, which is recall. Practice consisting of multiple-choice trivia games or televised quiz shows trains the former, and transfer to the latter is weak. Practice should reproduce the retrieval conditions of the test: question presented alone, answer produced from memory, feedback afterwards.

2. Distributed practice outperforms massed practice

Reviews separated by intervals produce better long-term retention than the same number of reviews in a single session. Ebbinghaus described the underlying forgetting curve in 1885. The modern quantitative case is the meta-analysis by Cepeda, Pashler, Vul, Wixted and Rohrer (Psychological Bulletin, 2006, 132(3), 354–380), covering 254 studies, which found distributed practice reliably superior and the optimal gap between reviews to scale with the interval over which the material must be retained.

The consequence is that the efficient moment to review an item is shortly before it would otherwise be forgotten, and that this moment recedes after each successful recall: a day, then several days, then weeks. Software scheduling reviews on this principle is called spaced-repetition software; the widely used SM-2 family of algorithms implements exactly this expanding interval. Massed revision produces a temporary availability that decays within weeks, which is why it feels effective and measures poorly.

3. Item selection dominates scheduling

The two findings above concern when material is reviewed. Neither addresses which material, and in practice that is the larger term: a perfectly scheduled review of items a quiz will never ask returns nothing.

Users of spaced-repetition software frequently report decks of several thousand to twenty-five thousand self-written cards accompanied by limited improvement in quiz performance. The scheduler is not at fault in these cases. Items written without reference to what is actually asked tend to reflect what was convenient to write, what was recently encountered, or whichever reference work was being transcribed.

Pub quiz content is not uniformly distributed across knowledge. It concentrates in a stable and countable set — capital cities, chemical symbols, monarchs and regnal dates, Academy Award winners, flags, currencies, number-one singles, football club nicknames and similar closed or semi-closed sets — and these recur across venues in measurable proportions. Ordering revision by observed frequency therefore returns more marks per hour than covering the same domain evenly.

This is the step QuizMaxx exists to do for you: its questions are ordered by how often their subject matter actually comes up in archives of real quiz questions, so the first twenty cards you see are the twenty worth knowing first. Open it and play a round — it needs no account, and the first session takes about ten minutes.

4. The documented case

The best-documented application of all three findings together is that of Roger Craig, who won $77,000 on Jeopardy! on 14 September 2010, then the highest single-day total in the programme's history and since surpassed. Craig retrieved the archive of previously broadcast clues, analysed it to establish which categories and items recurred, and drilled the result through Anki, a free spaced-repetition program. He has described the method publicly.

The order of that work is the informative part. The scheduling software is the detail usually reported, but it was applied to a selection derived from frequency analysis of the actual question distribution. Both steps were necessary. The second is the one generally omitted from accounts of the method.

Two further effects worth using

Feedback from real quizzes. Attending a weekly quiz supplies retrieval practice under realistic conditions and a sample of the local question distribution. Questions answered incorrectly are the highest-value revision items available: they are demonstrably within the distribution being asked and demonstrably not yet known. Recording them at the time is necessary, since recall of one's own errors decays quickly.

Narrative encoding. Isolated facts are harder to retain than facts embedded in a causal or narrative structure, which supplies additional retrieval routes. A date learned as a date is one cue; the same date attached to an account of the events around it is several. Background reading is therefore a complement to drilling rather than an inefficient substitute for it — the origin of the Ashes urn, the Decca audition, the work at Bletchley Park, or the widow who shaped modern champagne. QuizMaxx publishes 364 short essays of this kind.

Summary

Practise recall rather than recognition. Distribute reviews rather than massing them. Weight revision by the observed frequency of material rather than covering the domain evenly. Attend a quiz regularly and record every incorrect answer. The four are independent and additive. The third is the one most often omitted.

What QuizMaxx does

QuizMaxx implements the above. It supplies a curated body of general-knowledge questions ordered by measured frequency of occurrence in archives of real quiz questions, schedules reviews with an SM-2 spaced-repetition algorithm, and requires typed answers rather than multiple-choice selection, so that practice reproduces the retrieval conditions of a quiz. It runs in the browser, needs no account and is free. Open QuizMaxx.

Assembling an equivalent by hand is entirely possible and has been done. The comparison with Anki sets out where each is the better tool, and the general-knowledge case applies the same findings outside quizzing.