In this study we propose that some common scientific misconceptions are caused by associative learning biases that create undesired associations, and we accordingly design and study methods to overcome them. In Phase 1, we trained learners on specific, novel associations essentially to mimic the creation of an undesired association or “misconception”, and established that low salience though fully predictive cues are learned to be ignored, making them difficult to learn in subsequent tasks. This “learned inattention” not only explains why learners often attend to salient yet spurious, non-causal cues and ignore less salient yet fully predictive, causal cues, but it also explains why, once learned, these associations are hard to “correct”.

For example, students commonly associate force with velocity (a spurious cue) while ignoring acceleration (which is fully predictive), and correcting this association is very resistant to instruction. In Phase 2 we study the effectiveness of several different counter-training methods aimed at shifting attention to low salience cues both in an artificial domain and with actual physics misconceptions. Results in both domains indicate that that once a low salience cue has been overshadowed, multiple kinds of positive examples are not effective in learning the low salience cue, whereas negative examples attacking the validity of competing higher salience cue are effective in shifting attention to the low salience cue. These results suggest a general design principle for instructional materials that address misconceptions involving a low salience variable. Careful attention must be paid to the choice of variation of parameters among examples.