The Application of Stability Bias in Conceptual Learning

When memorizing mechanical materials such as words or numbers, people have shown the tendency to overestimate their future memory due to their insensitivity to memory loss. The experiments in this paper investigate whether the same bias applies to conceptual learning and, if so, how the magnitude of this bias compares to that of mechanical learning. In our experiment, participants were divided into four groups. Two groups of participants studied a word list and took free-recall tests either two minutes or one week after learning; they also made predictions of how well they would do on the test right after learning the material. The other two groups were designed the same way except that the learning material was a concept. Our results indicate that the stability bias illusions not only applied to conceptual learning but were also more significant than in mechanical learning.


Introduction
Metacognition is the "cognition of cognition:" the awareness, understanding and regulating of one's own thinking.Metamemory is a specific type of metacognition-that refers to memory; more specifically, the judgments and beliefs of memory (Kornell, Rhodes, Castel, & Tauber, 2011).Judgment is simply the prediction or evaluation of one's remembering and forgetting.In contrast, beliefs are more rational, based on one's experience and knowledge about memory.For example, "numbers are easier to memorize than words" is a belief about memory.When studying judgments of memory, researchers usually conduct experiments where the participants learn certain materials and make predictions of how much of those materials they will remember on a future date, and those predictions are what we call judgments of learning (JOLs).Various factors affect the magnitude of these judgments, and in the following experiment we will focus on one of them.

The Stability Bias
When asked to predict their future recalling of learned information, people tend to display two biases.First, their insensitivity to memory loss over the course of time leads them to make similar predictions for their performances on tests that would take place after different retention intervals (Carroll, Nelson, & Kirwan, 1997;Maki & Swett, 1987).Second, for the same reason, people usually predict better recall performance than their actual memory (Koriat & Bjork, in press).However, the latter bias becomes nearly negligible when the study-test interval is as short and only lasts a few minutes (Koriat, Sheffer, & Ma'ayan, 2002).These two errors-similar memory predictions for different time intervals and overconfidence in performance-constitute what we call stability bias.
Researchers have tested the effects of different variables on the extent of stability bias.For example, Kornell and colleagues (2011) conducted research using participants recruited through Amazon Mechanical Turk and word lists created based on the norms of Kucera and Francis (1967)-a corpus of 1 million English words' identified by frequency of occurrence-and concluded that changing the type size of words influenced judgments of memory but not actual memory performance.In addition, Rawson and Dunlosky (2002) discovered that high coherence of texts lead people to make falsely high JOLs.Results from these investigations yielded results that support the ease-of-processing heuristic.While the qualities of stimuli (type size, coherence, etc.) have been studied in terms of their effects on stability bias, our research will focus on a new direction: the influence of the type (conceptual versus mechanical) of material on the stability bias.
1.2 The Ease of Processing Heuristic Koriat and colleagues (2004) present a dual-basis view, which suggests that judgments of memory are based on either subjective experience during learning or priori knowledge and beliefs about memory.The former is referred to as "experience-based," the latter is called "theory-based."Theory-based JOLs are likely to be more educated and accurate because they are based on knowledge; however, they are usually activated only when learners become sensitive to memory decline by having retention-interval differences made salient to them (Koriat et al., 2004).Because this reminder of memory decline is not provided in most daily circumstances, experience-based JOLs constitute the majority of individuals' predictions of memory, and they are proven to be highly affected by the ease-of-processing of materials learned-in other words, how fluently the information can be encoded by the learner.The ease of processing brings stability bias because higher perceived ease leads to metacognitive judgments for better performance but does not influence actual memory.In conclusion, in most cases, JOLs are experienced-based, subjective, and affected by the ease-of-processing; as a result, they display stability bias.This derives the general theory that the easier it is to process information, the higher the stability bias we would expect to occur.Because conceptual materials are considered more difficult to process than mechanical information (numbers, words, etc.), we conducted an experiment comparing the stability biases in learning both materials to test if the ease of processing heuristic leads to lower stability bias in conceptual learning.

The Current Experiment
In this research study, we examined the magnitude of stability biases presented by participants who learned a word list and those who learned a concept that required comprehension.Although the former type of material does not require active thinking or comprehending, the latter requires analytical efforts for understanding.
Based on our theory that (1) information that requires understanding is harder to process than a list of random, individual words and (2) processing ease results in overconfidence in prediction, we hypothesize that the word list group would display greater stability bias.That is, participants who memorized words would make more similar estimates for different retention intervals, as well as generally higher JOLs for both intervals.However, we also hypothesize that word group participants display similar prediction-performance discrepancies to that of participants in the conceptual learning condition, given that words are not only easier to process but are also easier to encode and retrieve, which results in better actual recall performances.By investigating this topic, we explore whether the ease of processing heuristic applies to conceptual learning and causes it to present a smaller stability bias than that of mechanical learning.In addition, we examine whether the stability bias applies to conceptual learning at all.
In two groups, participants studied a word list and took free-recall tests on the material after either a two-minute study-test interval or a one-week interval.We recorded the JOLs for different retention intervals.The other two groups of participants studied a video explaining a concept and took free-recall tests on the concept after a two-minute or one-week interval.We again recorded the JOLs.By comparing the general magnitudes of JOLs and the differences in JOLs for different time intervals of the two material groups, we examined the first part of our hypothesis; by calculating the prediction-performance differences for both materials, we examined the second part of our hypothesis.

Participants
A sample of 169 participants of varied genders and ages was recruited via Amazon's Mechanical Turk, a website that allows workers to complete online surveys for compensation.They were selected under the requirements of being fluent English speakers and Amazon Mechanical Turk workers with greater than or equal to 95% approval rate for all the surveys they had done.They were paid $1 for completing tasks that took five minutes.
Thirty-nine participants were assigned to the two-minute retention interval word list condition (10 female, 29 male; range = 18 -72 years); fifty-one were assigned to the one-week word list condition (22 female, 29 male; range = 20 -64 years).Twenty-seven were assigned to the two-minute retention interval conceptual learning condition (14 female, 13 male; range = 22 -55 years); fifty-two were assigned to the one-week conceptual learning condition (22 female, 30 male; range = 21 -76 years).All participants lived in the United States and spoke English fluently.

Materials and Procedures
A ten-word list stimulus was generated at Paivio et al.Word List Generator.On average, the 10 words had a Kucera-Francis Frequency of 33.7, 2.4 syllables, and 7.1 letters.The words were displayed in a video that e JOL and actu echanical mate mances.In the rmance was 0 emory enduran mory declined of the word list l.We also pr before the test aining a conce xplaining the g used to evalua The video was , we also prov r how many wo condition, the s proceeded to wn as many w ank recall.Aft on for them to emembered.
hat of mechani e showing them dea units, whi of information session two, w or analysis wa r the one-wee up (Table 1).