Production between and within: distinctiveness and the relative magnitude of the production effect

ABSTRACT

The production effect is the memory advantage for items studied aloud over items studied silently. Three experiments examined the influence of (1) the distinctiveness heuristic in a pure-list paradigm and (2) statistical distinctiveness during study. Aloud versus silent processing was manipulated within-subject in a mixed-list procedure and additional pure-list items were alternated with the to-be-remembered words. This arrangement permitted the first examination of the production effect using both within-subject and between-subjects manipulations in the same experiment. The quite large between-subjects production effect observed for the pure-list words is attributed to the distinctiveness of the aloud words being enhanced by the co-occurring within-subject manipulation. In addition, when the pure-list words were all read aloud, they effectively increased the overall proportion of aloud words, thereby decreasing the distinctiveness of the to-be-remembered aloud words in the mixed list. Correspondingly, there was a decrease in the magnitude of the production effect. However, when the pure-list words were all read silently, the magnitude of the production effect was unchanged relative to baseline. These results provide partial support for the influence of statistical distinctiveness on the magnitude of the production effect.

KEYWORDS:

• Production effect
• distinctiveness account
• recognition
• memory

Acknowledgment

We thank David McLean, Junwen Liu, Zoey Hu, and Zelin Chen for their assistance with data collection.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Data availability

The datasets and programme code for this study are available from the authors upon request.

Notes

1 Studying non-visually presented items may, however, lead to a different pattern of results. For example, Mama and Icht (2016) reported that words presented auditorily were better remembered when written down relative to spoken aloud. In general, the form and the magnitude of the production effect may be related to the number of encoding modalities during learning (see also Forrin & MacLeod, 2016, 2018).

2 Estimates of Hedges’ g were computed according to the formulae provided in Cumming (2012). For confidence intervals of the effect sizes, we first computed the boundaries for the noncentrality parameter using the MBESS package in R Version 3.6.3 (Kelley, 2007), then (except for paired t-tests) computed the corresponding confidence intervals for the effect sizes according to the formulae provided in Smithson (2003), chapters 4 and 5. For paired comparisons, confidence intervals for the effect sizes were computed according to Algina and Keselman (2003).

3 Effect size confidence intervals for t-tests were calculated based on d (see Cumming, 2012, chapter 11).

4 For effect sizes represented by the partial eta squared statistic, 90% confidence intervals were computed instead of 95% confidence intervals since partial eta squared cannot be negative (see Steiger, 2004).

5 We only examined the aloud items here because in referencing the results from Icht et al. (2014), we did not expect to find differences across groups for memory of the silent items.

6 A supporting Bayesian repeated measures ANOVA, conducted in JASP using uniform priors, revealed anecdotal evidence for the null hypothesis (Lee & Wagenmakers, 2013), BF₀₁=2.273. This may be a consequence of the blue aloud and white silent items having been tested second in Experiment 2, yielding somewhat lower performance numerically compared to Experiment 3. Because there is agreement between the individual analyses in Experiments 2 and 3, we see this as a reasonable basis to conclude that the results of Experiments 2 and 3 were similar. (Note that the error percentage from this analysis was 6.886%, indicating that the Bayes Factor will change slightly when the analysis is repeated; however, this is within the acceptable error percentage limit: See van Doorn et al., 2020.)

7 A supporting Bayesian one-way ANOVA examining just the white silent items across all three experiments, conducted using uniform priors in JASP Version 0.14 (JASP Team, 2020), revealed a Bayes factor indicating moderate evidence for the null hypothesis (Lee & Wagenmakers, 2013), BF₀₁=3.069 (with error percentage < 0.001%).

8 We thank Reviewer 1 for suggesting this idea.

Additional information

Funding

This research was supported by the Natural Sciences and Engineering Research Council of Canada Discovery [grant number A7459].