**2. Method**

### *2.1. Design and Analytic Strategy*

The present investigation consisted of secondary analyses of a corpus of well-characterized stimuli. The analyses involved a by-item approach followed by a confirmatory linear mixed logistic regression model.

In the by-item analysis, we applied a mixed ANOVA model on the means of recall probability and incidental noun recall and image latency times (RTs) for the stimulus words collapsed across all participants. This procedure, which is ordinarily the most widely used for by-item analysis [42], permits to avoid violating the assumption of independence needed to perform statistical hypothesis testing. By averaging all observations for each stimulus word, it was ensured that only one instance of a participants' data was used per stimulus word. Thus, the stimuli were the units of analysis, as they were treated as random variables (as if they were "subjects"). The generalizability of results, therefore, referred to both subjects and items populations, washing out individual di fference e ffects (see [42]).

To confirm the ANOVA model, the linear mixed logistic regression approach consisted of analyzing each individual observation nested within participants and stimuli, instead of comparing the averaged responses by stimulus word. This supplementary method was adopted because of its ability to account for within-subjects e ffects, thereby enabling statistical testing within and between subjects without violating the assumption of independence, further allowing for stronger statistical power than the ANOVA model. (For details on the particular use of linear mixed regression models followed here see [43]).

The database used in this study is available on the archived website [44].
