The generation effect suggests that people are better able to remember material which they themselves generate, compared to material which they read or which is presented to them.
For instance, Slamecka and Graf (who coined the name for this effect in 1978) showed that memory performance was better when people had to generate a target word in an association pair (rhyme: wave / c??e) than when they read an association pair (rhyme: wave / cave).
This effect is sometimes equated with the Testing effect: when students are asked to produce answers from memory (as in free recall tests), some part of the impact on recall may be due to the processing involved in generating the answers. But experiments by Karpicke and Zaromb (2010) suggest that the two effects are distinct and can be to some extent isolated. People who studied by trying to generating targets ended up with poorer recall than people who study by attempting to recall.
Karpicke and Zaromb (2010) point to some references (which I’ve not yet read) that suggest the effect may be minimal in harder free recall tasks (contra the fourth and fifth experiments from Slamecka, which did indeed show only small effects):
Thus there is often no difference in free recall of read vs. generated lists, and in fact sometimes there is an advan- tage of reading over generating (e.g. Nairne et al., 1991; Schmidt & Cherry, 1989). The enhanced item processing that occurs in a pure list of generated items is not sufficient to counteract disrupted order processing and produce an advantage in free recall relative to a pure list of read items.