Knowledge Engine

Do you remember what you produce better than what you read?

·no read data

The idea

In 1978 the psychologists Norman Slamecka and Peter Graf published five experiments in the Journal of Experimental Psychology: Human Learning and Memory comparing two ways of meeting the same word. In the read condition a participant saw a pair printed in full, such as “rapid-fast”. In the generate condition they saw the cue and only the first letter of the answer, “rapid-f”, and had to produce a word that fitted the rule, in this case a synonym. Later, people remembered the words they had produced better than the words they had only read.

The authors tested this with recognition, free recallfree recall A memory test in which people write down everything they can remember from a list, in any order, with no hints. and cued recallcued recall A memory test in which people are given a hint, such as the word that was paired with the target, and asked to supply the item that went with it., and with five generation rules: associate, category, opposite, synonym and rhyme. In every case the generate condition came out ahead, and the size of the advantage did not differ significantly between rules. In the cued recall experiment, for example, participants recalled 68 percent of generated responses against 61 percent of read ones. The experiments were small, with 12 to 24 participants each. The authors concluded the effect was real and that it posed an interpretive problem; they could not settle on why it happened.

This is now called the generation effect: information you produce from your own memory tends to be better retained than the same information presented to you.

Why it might change a decision

The effect makes a checkable prediction about any learning material that can be presented either complete or partly withheld. If two groups study identical content, one reading full statements and one filling in a missing word or answer they can work out, the second group should remember more of those words at a later test. The meta-analysesmeta-analysis A study that pools the results of many earlier studies on the same question to estimate an overall effect and what makes it vary. below add sharper predictions: the gap should be larger for short lists than long ones, and should disappear if the thing to be produced is meaningless, such as a nonword.

How solid is this?

The generation effect is one of the better replicatedreplication Running a study again, independently, to see whether the original result holds up. findings in memory research, and it has been pooled in at least two large meta-analyses. What they show is a reliable effect of moderate size that depends a good deal on the conditions.

Sharon Bertsch, Bryan Pesta, Richard Wiscott and Michael McDaniel, writing in Memory & Cognition in 2007, combined 445 effect sizes from 86 studies with 17,711 participants. The overall effect sizeeffect size How large an effect is, as opposed to whether it exists at all. A result can be statistically reliable and still be too small to matter. was d = 0.40. In plain terms, generated items were remembered at an average rate of .559 against .471 for read items, a gap of about nine percentage pointspercentage point The plain difference between two percentages, so a move from 40 to 45 percent is a rise of 5 percentage points.. The variation around that figure was large. The effect was bigger in cued recall (d = 0.55) and recognition (0.46) than in free recall (0.32). It was larger in within-subjects designswithin-subjects A study design in which the same people take part in every condition being compared, rather than different people in each. (0.50), where the same person both reads and generates, than between subjects (0.28). Lists of 25 items or fewer gave d = 0.60; lists of more than 50 gave 0.09. Tests given more than a day later showed a larger effect (0.64), though that estimate rests on only 30 effect sizes. For nonwords the effect was almost absent, at d = 0.05.

Matthew McCurdy and colleagues revisited the literature in Psychonomic Bulletin & Review in 2020, covering 126 articles and 310 experiments. They estimated that generated items were remembered about 10 percentage points better, or about 8 points after adjusting for possible publication biaspublication bias The tendency for studies with striking or positive results to be published more often than studies that found nothing, which can inflate pooled estimates., which still left a significant effect. They again found the advantage for words but not for nonwords, and again found it smallest in free recall. They also found it in between-subjectsbetween-subjects A study design in which different groups of people each take part in only one of the conditions being compared. designs, which counts against the idea that it is purely an artefact of people rehearsing generated items at the expense of read ones in a mixed list. A further moderatormoderator A condition that changes how strong an effect is, or whether it appears at all. they emphasised was generation constraint: how tightly the cue limits what can be produced. Less constrained tasks gave better recall, although this difference did not appear for recognition.

The nonword result is the most telling boundary. It fits the view that generation helps by activating what a person already knows about a word’s meaning; when there is nothing to activate, there is nothing to gain. Competing accounts, such as greater mental effort or extra attention, have mixed support, and McCurdy and colleagues found that harder and easier generation tasks gave similar effects, which counts against effort alone.

The popular version of the idea, that producing answers is a powerful general study technique, outruns these data. In the 2007 review, most effect sizes came from lists of words, almost all from younger adults, and tests within a day far outnumbered tests after longer delays. The effect is real in that setting; how far it carries to complex material and real courses is a separate question these reviews do not answer.

Sources

  • Slamecka, N. J., & Graf, P. (1978). The generation effect: Delineation of a phenomenon. Journal of Experimental Psychology: Human Learning and Memory, 4(6), 592-604. doi:10.1037/0278-7393.4.6.592
  • Bertsch, S., Pesta, B. J., Wiscott, R., & McDaniel, M. A. (2007). The generation effect: A meta-analytic review. Memory & Cognition, 35(2), 201-210. doi:10.3758/BF03193441
  • McCurdy, M. P., Viechtbauer, W., Sklenar, A. M., Frankenstein, A. N., & Leshikar, E. D. (2020). Theories of the generation effect and the impact of generation constraint: A meta-analytic review. Psychonomic Bulletin & Review, 27, 1139-1165. doi:10.3758/s13423-020-01762-3

Check yourself

Answers are checked in your browser. Nothing is recorded or sent anywhere.

  1. In Slamecka and Graf's 1978 experiments, what did participants in the 'generate' condition do?

  2. According to Bertsch and colleagues' 2007 meta-analysis, roughly how big is the average generation effect?

  3. Which condition did both meta-analyses find gave little or no generation effect?

  4. Which is the fairest summary of the evidence described in the post?

Questions without answers

  1. Most generation experiments use word lists studied for minutes. What would you need to see before believing that generating answers helps someone learn a textbook chapter over a term?
  2. Lists of more than 50 items showed almost no effect in the 2007 review. What might explain that, and how could you test your explanation?

Back to the archive