Despite the unequal variance signal-detection (UVSD) model’s prominence as a model of recognition memory, a psychological explanation for the unequal variance assumption has yet to be verified. According to the encoding variability hypothesis, old item memory strength variance (σo) is greater than that of new items because items are incremented by variable, rather than fixed, amounts of strength at encoding. Conditions that increase encoding variability should therefore result in greater estimates of σo. We conducted three experiments to test this prediction. In Experiment 1, encoding variability was manipulated by presenting items for a fixed or variable (normally distributed) duration at study. In Experiment 2, we used an attentional manipulation whereby participants studied items while performing an auditory one-back task in which distractors were presented at fixed or variable intervals. In Experiment 3, participants studied stimuli with either high or low variance in word frequency. Across experiments, estimates of σo were unaffected by our attempts to manipulate encoding variability, even though the manipulations weakly affected subsequent recognition. Instead, estimates of σo tended to be positively correlated with estimates of the mean difference in strength between new and studied items ( d), as might be expected if σo generally scales with d. Our results show that it is surprisingly hard to successfully manipulate encoding variability, and they provide a signpost for others seeking to test the encoding variability hypothesis.

We thank Iris van Rooij, Sam Forbes and Brad Wyble for helpful comments on a previous draft of this manuscript.

The unequal variance signal detection (UVSD) model of recognition memory assumes that the variance of old item memory strength (σ_o) is typically greater than that of new items. It has been suggested that this <i>old item variance effect</i> can be explained by the encoding variability hypothesis. However, Spanton and Berry (2020) failed to find evidence for this account, suggesting that σ_o may simply scale with mean memory strength (<i>d</i>) in the UVSD model. Experiments 1 and 2 examined the effects of encoding variability and strength scaling on old item variance by creating conditions in which mean memory strength and variability in item characteristics was either low or high in 2 × 2 factorial designs. In Experiment 1, overall strength determined estimates of σ_o, with no effect of item characteristic variability. The same effect of overall strength was found in Experiment 2; there was also a significant effect of item characteristic variability, although this manipulation also had some effect on <i>d</i> and was therefore partially confounded. Experiment 3 similarly found a simultaneous increase in old item variance and memory strength in a design using mixed item characteristic variability conditions in a single study/test block. We conclude that old item variance increases with mean memory strength in the UVSD model, with uncertainty about the effects of encoding variability, and that future explanations of the old item variance effect should bear this in mind.

The unequal variance signal detection (UVSD) model of recognition memory assumes that the variance of memory strength for studied items (σo) is typically greater than that of non-studied items. It has been proposed that this old item variance effect is caused by many factors that affect memory strength at encoding, which add variable amounts of strength to a baseline amount. However, Spanton and Berry (2020) failed to find evidence for this encoding variability hypothesis and instead found that estimates of σo tended to be strongly positively correlated with estimates of mean memory strength (d) across participants, suggesting that σo may simply scale with d. The present experiments examined the effects of encoding variability and scaling on old item variance by creating conditions in which mean memory strength and encoding variability was either low or high in 2 × 2 factorial designs. In Experiment 1, overall strength determined estimates of σo, with no effect of encoding variability. The same effect of overall strength was found in Experiment 2; there was also a significant effect of encoding variability, although this manipulation also affected overall memory strength and was therefore partially confounded. We conclude that there is strong evidence that mean memory strength at least partially determines estimates of old item variance, and that satisfactory explanations of the old item variance effect should account for this.