Asymmetries in time-based and feature-based discriminations in humans: linking the long+ and feature-positive effects

Three experiments examined time-based and feature-based discrimination learning in a video game task with human participants. Each participant played the role of gunner on a space ship. Participants were told that a periodically presented target object might or might not be hiding an enemy alien, which would appear briefly once the target disappeared. Points were given for 'shooting' at the target that hid the enemy alien, and were taken away if the target was ultimately not hiding an alien. In Experiment 1, either a relatively long (4 s) or a relatively short (1 s) intertrial interval (ITI) predicted whether an alien would be hiding behind the next target. Participants anticipated enemy aliens more readily when the long ITI predicted the enemy alien than when the short ITI did so, establishing a long+ effect for the first time in humans. In Experiment 2, a friendly alien was added on the negative trials, so that a different event, rather than no event, occurred on the negative trials. This eliminated the long+ effect; that is, participants now found it equally easy to predict the target enemy aliens whether they uniquely followed long or short ITIs. Experiment 3 examined whether a similar treatment reduced the theoretically related feature-positive (FP) effect. In that experiment, a unique visual image, rather than the ITI duration, served as the predictor. An FP effect occurred when the discrimination was between trials with an outcome versus no outcome, but not when the discrimination was between an outcome versus a different outcome. The findings connect the long+ and the FP effects, and suggest that both can be reduced by making the negative trials more salient.