A reinforcement schedule in which the reinforcer arrives on an unpredictable proportion of correct responses rather than every time.

Variable reinforcement is a specific form of reinforcement schedule and is the workhorse of maintained trained behaviour across species. The animal cannot predict which response will be reinforced, only that reinforcement will arrive on some proportion of responses, and this unpredictability has powerful effects on behavioural persistence.

Behaviours maintained on variable reinforcement are highly resistant to extinction, sometimes called the partial reinforcement extinction effect. This is one of the most reliably replicated findings in the behavioural science literature and applies across the range of species studied.

The same principle underpins both useful and problematic phenomena. Slot machines, designed to produce highly persistent gambling behaviour, deliver reinforcement on a variable schedule. Good clicker training, designed to produce highly persistent trained behaviour, also moves to a variable schedule once a behaviour is established. The mechanism is identical; the application is what differs.

In animal training, the transition from continuous to variable reinforcement is one of the most useful tools for producing reliable performance. A horse who has been positively reinforced every time for a particular behaviour during the learning phase will perform reliably when reinforcement becomes intermittent in the maintenance phase, often even more reliably than when reinforcement was continuous. The same applies to dogs, parrots, zoo animals, and the rest of the trained species range.

The transition needs to be gradual. Moving too quickly from continuous to variable reinforcement can cause the animal to abandon the behaviour. Skilled trainers thin the reinforcement schedule incrementally over many sessions.