Subjective judgments (e.g., confidence ratings) have been widely used as a tool for measuring metacognition in a variety of cognitive domains, including learning and memory (Rhodes and Tauber, 2011, Yang et al., 2021), decision-making (Fleming et al., 2010, Hu et al., 2023), and deductive reasoning (Block, 2008, Shynkaruk and Thompson, 2006). Numerous behavioral and neuroimaging studies have measured individuals’ metacognitive ability by asking them to make item-by-item judgments, such as a confidence rating after each perceptual (Hu et al., 2023) or memory (Hu, Yang, & Luo, 2022) decision. Most of these studies implicitly assumed that such metacognitive judgments provide neutral assessments of the cognitive processes they measure and have no impact on those processes or accompanying task performance (Spellman & Bjork, 1992).

Research going back several decades, however, shows that in many situations target processes and behaviors are reactively affected by these metacognitive judgments (Arbuckle and Cuddy, 1969, King et al., 1980, Zechmeister and Shaughnessy, 1980, Li et al., 2023, Shi et al., 2023, Li et al., 2022, Mitchum et al., 2016, Shi et al., 2023, Li et al., 2023, Soderstrom et al., 2015, Zechmeister and Shaughnessy, 1980, King et al., 1980; Zhao et al., 2022, Zhao et al., 2023; W. L. Zhao et al., 2022), a phenomenon known as the reactivity effect and which suggests that metacognitive judgments are not passive measures but can themselves alter “reality” (for reviews, see Double and Birney, 2019, Double et al., 2018, Double and Birney, 2018). For instance, many studies found that soliciting confidence ratings reactively enhances perceptual decision accuracy (Bonder & Gopher, 2019), slows down decision speed (Lei et al., 2020, Li et al., 2023), improves reasoning performance in high-confidence individuals (Double & Birney, 2017), facilitates recognition memory (W. L. Zhao et al., 2022), and so on (Double and Birney, 2018, Double et al., 2018). Besides soliciting explicit metacognitive judgments, classic research established that asking individuals to explain what they are doing or thinking (Nisbett and Ross, 1980, Nisbett and Wilson, 1977) can trigger meta-awareness (Winkielman & Schooler, 2011), which in turn reactively alters their task performance (i.e., the reactivity effect of concurrent verbalization). The recent work on reactivity induced by metacognitive judgments extends earlier assessments of reactivity induced by concurrent verbalization (for a review, see Fox, Ericsson, & Best, 2011).

One form of reactivity induced by metacognitive judgments, which has attracted substantial research interest in recent years, is the effect of judgments of learning (JOLs; metacognitive estimates about the likelihood of remembering a studied item on a later memory test). Research has observed that many learning outcomes are reactively changed as a consequence of making a JOL while or after studying each item (e.g., Ariel et al., 2021, Chang and Brainerd, 2023, Double et al., 2018, Double and Birney, 2018, Janes et al., 2018, Li et al., 2022, Mitchum et al., 2016, Myers et al., 2020, Rivers et al., 2021, Senkova and Otani, 2021, Soderstrom et al., 2015, Tauber and Witherby, 2019, Tekin and Roediger, 2020, Witherby and Tauber, 2017b; W. B. Zhao et al., 2022; W. L. Zhao et al., 2022). In addition, many studies have documented the reactivity of memory to JOLs with different types of materials, such as word lists (Li et al., 2022, Maxwell and Huff, 2022), pure lists of related word pairs (Li et al., 2022, Rivers et al., 2021, Witherby and Tauber, 2017b; W. L. Zhao et al., 2022), visual images (Shi et al., 2023, Li et al., 2023), and inter-item relations (W. B. Zhao et al., 2022).

Why does soliciting metacognitive judgments reactively temper the entity being monitored? A possible explanation, proposed by Mitchum et al. (2016), is that monitoring ongoing mental processes during a cognitive task changes individuals’ task goals, in turn leading to a reactive influence on performance. Below, we term this explanation the changed-goal hypothesis, following Mitchum et al. (2016). It is well-known that overtly soliciting metacognitive judgments enhances individuals’ awareness of any discrepancy between their current level of mastery and their desired goals, which then guides subsequent metacognitive control process (e.g., adjusting task goals, changing task strategies; Finn, 2008, Thiede et al., 2003, Yang et al., 2017). This might explain why metacognitive judgments can affect the very processes being judged.

Regarding memory reactivity to JOLs, Mitchum et al. (2016) speculated that when the difference in learning difficulty among list items is obvious (e.g., when studying a mixed list of related and unrelated word pairs), soliciting JOLs encourages participants to consider that some items are more memorable than others. To prevent “labor in vain” (i.e., exerting effort toward remembering difficult items produces little improvement; Nelson & Leonesio, 1988), participants change their study goal from mastering all items to prioritizing easy ones, with a sacrifice in learning difficult ones. This leads to positive reactivity for easy items and negative reactivity for difficult ones. In essence, the use of a JOL scale with extreme values at 0 and 100 communicates to participants that some items are not memorable at all whereas others are highly memorable. This encourages participants to be more selective in their learning goals, switching away from their normal “mastery” mindset.

Mitchum et al. (2016) provided a clear demonstration of the memory reactivity effect of JOLs and of the possible role of goal-change in this effect. In their Experiments 1–3, Mitchum et al. instructed two groups (JOL vs. no-JOL) of participants to study a mixed list of related (e.g., computer – keyboard) and unrelated (e.g., book – shoe) word pairs, and they were allowed to spend as much time as they wanted to study each pair. The only difference between the two groups was that the JOL group was required to make a JOL after studying each pair, whereas the no-JOL group was not. Mitchum et al. found that, although both the JOL and no-JOL groups spent longer studying unrelated than related pairs, the JOL group did so to a lesser extent than the no-JOL group. Stated differently, the correlation between study time and cue-target relatedness (an index of learning difficulty) was less negative in the JOL than in the no-JOL group. This pattern is consistent with a shift in the JOL group towards prioritizing encoding of related (easy) pairs and sacrificing unrelated (difficult) ones.

What about actual recall? Mitchum et al. found that recall of related pairs was numerically enhanced by making JOLs while recall of unrelated ones was significantly impaired. In other words, they found that the difference in recall between related and unrelated pairs (i.e., the relatedness effect) was significantly larger in the JOL than in the no-JOL group, a phenomenon we term the enhanced relatedness effect (Janes et al., 2018). Clearly, the enhanced relatedness effect is also consistent with the changed-goal hypothesis.

It is worth noting that enhanced relatedness effects were not only observed in self-paced (Mitchum et al., 2016, Experiments 1–4) but also in experimenter-paced study conditions (Mitchum et al., 2016, Experiment 5; Janes et al., 2018). For instance, in Mitchum et al.’s Experiment 5, participants were instructed to study a mixed list of related and unrelated pairs in an experimenter-paced study procedure (i.e., with each pair presented for 5 s for participants to study). Participants in the JOL group made item-by-item JOLs after studying each word pair, whereas those in the no-JOL group did not. Strikingly, the results again showed an enhanced relatedness effect. Mitchum et al. proposed that, in experimenter-paced study conditions, the limited study time creates pressure on word pair learning, leading to a shift from a mastery orientation (i.e., mastering all items) toward a concentration on easy pairs (Metcalfe & Kornell, 2003). Indeed, Mitchum et al. noted that they heard participants spontaneously exclaim “I will never remember that” during the study phase, suggesting that they might simply wait out the item presentation screen without attempting to encode the study items when they perceived them to be too difficult to remember. Hence, Mitchum et al. claimed that the enhanced relatedness effect in experimenter-paced conditions is again consistent with the changed-goal hypothesis.

Janes et al. conducted two experiments with an experimenter-paced procedure to further test the changed-goal hypothesis (Janes et al., 2018, Experiments 2 and 3). Participants studied either a mixed list or two pure lists of related and unrelated word pairs. In the mixed list condition, participants studied 60 word pairs in total, composed of 30 related and 30 unrelated pairs. By contrast, in the pure list condition, they only studied a pure list of 30 related or a pure list of 30 unrelated pairs.

Based on the changed-goal hypothesis, Janes et al. predicted that the enhanced relatedness effect should only occur in the mixed list but not in the pure list condition, because pure lists of related and unrelated word pairs lack variation in item difficulty. That is, the essential driver of goal-change – the presence of items of widely varying difficulty – is absent in the pure list condition. Consistent with this prediction, Janes et al. observed an enhanced relatedness effect in the mixed list but not in the pure list condition. Accordingly, Janes et al. concluded that their study provides “support for the changed-goal hypothesis” (p. 2361).

Before moving forward, it should be noted that, besides the changed-goal hypothesis, there is another available theoretical account for the enhanced relatedness effect, that is, the dual-mechanism hypothesis (Mitchum et al., 2016, Janes et al., 2018). The dual-mechanism hypothesis asserts that the enhanced relatedness effect is caused by two separate mechanisms: 1) cue-strengthening (inducing positive reactivity for related pairs; Soderstrom et al., 2015), and 2) dual-task costs (inducing negative reactivity for unrelated pairs; Mitchum et al., 2016). Specifically, Soderstrom et al. (2015) developed a cue-strengthening explanation to explain positive reactivity for related pairs. They assumed that when a participant attempts to make a reasonable JOL for a word pair, they search for “diagnostic” cues (e.g., relatedness strength between the cue and target, mediators between the cue and target) to guide JOL formation. The cues activated by the requirement of making JOLs in turn strengthen the cue-target relation for related pairs, leading to positive reactivity. By contrast, because there is no pre-existing relation between the cue and target for unrelated pairs, making JOLs fails to enhance their memory.

Mitchum et al. (2016) proposed the dual-task costs hypothesis to explain why the reactivity effect is negative in some situations. They assumed that the additional requirement of making JOLs may borrow limited resources (e.g., study time, working memory capacity) from the primary learning task (Delaney and Verkoeijen, 2009, Griffin et al., 2008, Rundus, 1971). In addition, frequent task mode switching between encoding (i.e., studying word pairs) and monitoring (i.e., making JOLs) may bring further dual-task costs (Davis et al., 2017, Doherty et al., 2019, Griffin et al., 2008, Karaca et al., 2020, Yang et al., 2018, Yang et al., 2018). Hence, making JOLs may reactively impair memory, especially when the learning task itself is highly challenging.

In summary, the dual-mechanism hypothesis asserts that generating JOLs reactively enhances memory of related pairs through inducing cue-strengthening, and impairs memory of unrelated pairs through inducing dual-task costs, in combination leading to the enhanced relatedness effect (Janes et al., 2018). The current study primarily focuses on assessing the validity of the changed-goal hypothesis. We further elaborate on the dual-mechanism hypothesis in the General Discussion.

Previous findings about the role of goal-change in JOL reactivity are conflicting and inconclusive. For instance, although the findings provided by Mitchum et al. (2016) support the changed-goal hypothesis, Janes et al. (2018) failed to replicate Mitchum et al.’s self-paced learning findings. Specifically, Janes et al. (2018, Experiment 1) observed no difference in correlation between study time and relatedness strength between the JOL and no-JOL groups in self-paced study conditions. Furthermore, Janes et al. observed that, when the study procedure was self-paced, there was no statistically detectable difference in the relatedness effect between JOL and no-JOL groups. The inconsistent findings provided by Janes et al. (2018) and Mitchum et al. (2016) suggest that further tests on the changed-goal hypothesis are needed.

Additionally, Halamish and Undorf (2023) recently provided further evidence challenging the changed-goal hypothesis. In this study, two (JOL and no-JOL) groups of participants studied a mixed list of unrelated (e.g., raft-kiss), related (e.g., lips-kiss), and identical (e.g., kiss-kiss) word pairs. Participants in the JOL group gave highest JOLs for identical pairs, medium JOLs for related pairs, and lowest JOLs for unrelated pairs, reflecting that they believed that identical pairs were easiest and unrelated ones were most difficult to memorize. According to the changed-goal hypothesis, making JOLs should induce a larger positive reactivity effect for identical than for related pairs, because the former should be prioritized on the basis of their perceived ease of learning. However, in contrast to this prediction, Halamish and Undorf observed that the effect size of positive reactivity for related pairs (Cohen’s d = 0.57) was over double that for identical pairs (d = 0.25).

Overall, previous findings regarding the changed-goal hypothesis of JOL reactivity are conflicting, with some supporting it (Janes et al., 2018, Experiment 2; Mitchum et al., 2016), and some challenging it (Halamish and Undorf, 2023, Janes et al., 2018, Experiment 1). It remains unclear whether making JOLs reactively alters memory through changing study goals. Besides the changed-goal hypothesis, the dual-mechanism hypothesis can also readily account for the enhanced relatedness effect (Janes et al., 2018), which will be further elaborated in the General Discussion.

The current study aims to test the changed-goal hypothesis in two ways. First, it investigates whether pre-study JOLs (i.e., JOLs made before studying each item) enhance the relatedness effect as much as immediate JOLs (i.e., JOLs made after studying each item). Second, it explores whether the enhanced relatedness effect transfers to no-JOL pairs when JOL and no-JOL pairs are studied in an interleaved manner (see below for details).

Previous studies mainly explored the reactivity effect of immediate JOLs on memory (Janes et al., 2018, Mitchum et al., 2016). Another widely studied form comprises pre-study JOLs (Castel, 2008, Price and Harrison, 2017, Witherby and Tauber, 2017a, Yang et al., 2021). Unlike immediate JOLs which are formed on the basis of a combination of processing experience (i.e., online experience obtained from the learning task, such as ease of processing) and beliefs about memory (i.e., beliefs about how a given factor, such as word frequency, affects memory), pre-study JOLs are solely based on metamemory beliefs because they are made before participants receive and study each item (Yang et al., 2021). [For discussion of the difference between immediate and pre-study JOLs, see Price and Harrison (2017).]

Pre-study JOLs have been frequently employed to measure people’s beliefs about how a given factor (e.g., concreteness, relatedness, word frequency, font size, emotion, age, and so on) affects memory (Jia et al., 2015, Mueller et al., 2014, Witherby and Tauber, 2017a, Yang et al., 2018, Yang et al., 2018). Directly related to the current study are the findings from Mueller and Dunlosky (2017). In this study, participants in a pre-study JOL group were informed whether the next word pair would be a related or unrelated one and made a pre-study JOL to predict the likelihood that they would remember the next pair in a later memory test. By contrast, participants in an immediate JOL group made a JOL after studying each pair. Mueller et al. (2017) found that participants in both the immediate JOL and pre-study JOL groups provided substantially higher JOLs for related than for unrelated pairs (for connected findings, see Price & Harrison, 2017).

Mueller et al.’s (2017) findings suggest that both pre-study and immediate JOLs encourage participants to compare the relative memorability of related and unrelated pairs (Mueller and Dunlosky, 2017, Price and Harrison, 2017). Therefore, according to the changed-goal hypothesis, pre-study JOLs, similar to immediate JOLs, should change participants’ study goals and induce an enhanced relatedness effect (Mitchum et al., 2016). More specifically, the changed-goal hypothesis assumes that making immediate JOLs enhances awareness of the difference in learning difficulty between related and unrelated pairs, which then induces participants to prioritize memorizing easy related pairs and sacrifice difficult unrelated pairs. Given that making pre-study JOLs can also enhance awareness of the difference in learning difficulty (as reflected by the substantial difference in pre-study JOLs between related and unrelated pairs), pre-study JOLs should also change participants’ study goals and induce an enhanced relatedness effect (Mitchum et al., 2016). It is even reasonable to expect that pre-study JOLs would induce a larger enhanced relatedness effect than immediate JOLs, because pre-study JOLs are made before participants observe each word pair (Castel, 2008). Knowing in advance whether the next pair will be easy or difficult to remember, they can then accordingly prepare to allocate more or fewer resources toward studying it.

Overall, the changed-goal hypothesis generates two predictions: (1) Pre-study JOLs should reactively enhance the relatedness effect, and (2) the enhanced relatedness effect of pre-study JOLs should be equal to (or even larger than) the effect of immediate JOLs. Experiments 1–4 were conducted to test the changed-goal hypothesis through assessing these two theoretical predictions.

Experiment 5 tested another prediction of the changed-goal hypothesis, concerning the transfer of reactivity from JOL pairs to no-JOL ones when they are intermixed. In the introduction to Experiment 5, we explain how the changed-goal hypothesis can be tested through investigating reactivity transfer.