The Consequence of Refreshing for Access to Non-Selected Items in Young and Older Adults

Abstract

We examined the effect of competition on briefly thinking of just-seen items. In Experiment 1, participants saw a set of three related or unrelated words, and then read one of the words again (repeat) or thought briefly of one of the words (refresh). Participants read the set a second time, after which they refreshed a second word from the set or read a new word. Compared to reading a new word, response times were slower to refresh the second item having just refreshed vs. just repeated the first item. This increase was larger for related vs. unrelated words and for older vs. younger adults. In Experiment 2, a negative impact of refreshing was observed when repeating a different word from the set. The pattern of findings suggest that the negative impact of refreshing comes from increased competition from the refreshed item rather than inhibition of the non-refreshed items.

According to the Multiple-Entry, Modular memory (MEM; Johnson, 1992; Johnson & Hirst, 1993) model of cognition, complex reflection such as planning, deciding, and organizing is made up of component processes such as retrieving information and noting relations between representations. For example, planning ahead may involve noting the relevance of current information to future goals and retrieving relevant memories of past events.

One of the simplest reflective processes is immediately thinking of information that was just experienced and whose representation is still active--refreshing (Johnson, 1992; Johnson & Hirst, 1993). Refreshing can be thought of as an executive function of working memory (e.g., Baddeley & Hitch, 1974; Baddeley, 1986) in that it is recruited to briefly maintain or manipulate information that is currently active (e.g., D'Esposito et al., 1998; Blumenfeld & Ranganath, 2006). Refreshing has been proposed as a mechanism that foregrounds the representation of a just-experienced event (either perceptual or reflective) with respect to other representations that are currently active (Johnson et al., 2005; Raye, Johnson, Mitchell, Greene, & Johnson, 2007). Thus, refreshing would be a mechanism by which items can be brought into the focus of attention (e.g., Cowan, 1999; Oberauer & Kliegl, 2006). For example, in a search of working memory (e.g., Sternberg, 1966), refreshing might be a mechanism by which individual representations from the set are foregrounded for the purpose of comparison with the probe item.

Refreshing can be distinguished operationally from other component processes of reflection (e.g., Johnson, Raye, Mitchell, Greene, & Anderson, 2003; Raye et al., 2007). For example, while refreshing typically operates on individual items over very brief intervals, rehearsing typically maintains multiple items over longer intervals. As noted above, some process such as refreshing is logically needed to select an item from a set of items being actively rehearsed. Refreshing may also be involved in updating (e.g., Bjork, 1978; Morris & Jones, 1990; Roth & Courtney, 2007) items in working memory, in that refreshing would help foreground the representation of the “new” target, privileging it relative to the representations of older targets that are still active. Refreshing is also distinct from processes that revive information from long-term memory (e.g., reactivating, retrieving). At the same time, during recall, a cue may be refreshed, and this foregrounding of the recall cue may help to reactivate representations in long-term memory or facilitate the initiation of more strategic retrieval of additional cues and the target. For example, during a category generation task (e.g., Loftus, 1973; Blaxton & Neely, 1983), the letter cue may be refreshed which helps reactivation and evaluation of an appropriate exemplar representation.

Johnson and colleagues (e.g., Johnson et al., 2005) have investigated refreshing by cuing participants to think immediately of an item they just experienced perceptually. In one study (Johnson, Reeder, Raye, & Mitchell, 2002), participants saw a word, followed by the same word again, a new word, or a dot cue. Participants read each word aloud, and when they saw the dot, participants thought of (i.e., refreshed) the word they had just seen and said that word aloud. Response times (RTs) to refresh a word were significantly longer than to read the word again or to read a new word, reflecting additional time necessary to refresh an item that is no longer perceptually present. Refreshed items were more memorable on a later recognition test than items read once or items read twice.

Refreshing may play an especially important role in selection among active representations. That is, information (e.g., a word) rarely occurs in isolation but instead is part of a more complex experience. A study investigating selective refreshing compared participants' RTs to refresh one of three potential items to RTs to refresh one item that was shown alone (Raye, Mitchell, Reeder, Greene, & Johnson, 2008). Relative to repeat and read conditions, participants were slower to refresh when selection was required than when it was not. One hypothesis is that selective refreshing involves resolving competition from active distractor items in order to select the target item. This idea is consistent with neuroimaging results showing that, compared to refreshing one item shown alone, selectively refreshing one item from a set of three is associated with greater activation of the anterior cingulate cortex (Johnson et al., 2005, Experiment 5), an area thought to be involved in conflict detection (e.g., Botvinick, Nystrom, Fissell, Carter, & Cohen, 1999).

Previous studies of selective refreshing have not examined the “fate” of the non-selected items. Does refreshing an item from a set of active items affect the accessibility of the other items in the set? That is, does refreshing from currently active representations produce negative effects similar to output interference (e.g., McGeoch, 1936; Tulving & Arbuckle, 1963) or retrieval-induced forgetting (e.g., Anderson, Bjork, & Bjork, 1994) seen in accessing information from long-term memory? In the current studies we investigated the effect of selectively refreshing on subsequent reflective (Experiment 1) and perceptual (Experiment 2) processing of the non-refreshed items. The degree of competition was manipulated using sets containing related or unrelated words. Because related sets contain both semantically similar and episodically similar items they should result in more competition than do unrelated sets in which only episodic competition is present. To foreshadow, in Experiment 1, we observed that selective refreshing had the negative impact of reducing accessibility of the non-selected items during subsequent reflective processing especially under conditions of high competition (i.e., related trials). Older adults showed a larger negative impact than young adults, which, given age-related deficits in inhibitory mechanisms (Hasher, Lustig, & Zacks, 2007; Hasher & Zacks, 1988), suggests that this reduced accessibility does not arise from inhibition of the non-refreshed items. In Experiment 2, we observed a negative impact of refreshing on subsequent perceptual processing which was equal for related and unrelated items.

Experiment 1

In Experiment 1, a trial consisted of two tasks. In Task 1, participants saw and read aloud three related or unrelated words, after which they saw and read aloud one of the three words presented again (repeat) or thought of and said aloud the word that had appeared in a cue's location (refresh). In Task 2, participants saw and read aloud the same word set a second time, after which they either refreshed a different item from the set or read a new word presented on the screen. Thus, we tested whether selective refreshing during Task 1 influences subsequent refreshing of non-selected items (i.e., during Task 2). Specifically, we predicted that selective refreshing might have the negative consequence of reduced accessibility of the non-refreshed items, particularly for related trials. This inaccessibility would be reflected in longer RTs to refresh an item on Task 2 having just refreshed vs. just repeated from the same set on Task 1. In contrast, refreshing should have less or no influence on subsequent processing of a new item that was not present at the time of refreshing. Hence, RTs to read a new item on Task 2 should be less influenced by the nature of processing on Task 1.

In addition, we tested whether the negative impact of selective refreshing results from inhibition of the non-refreshed items during Task 1, which reduces their accessibility during Task 2. To this end, in Experiment 1C, we investigated the effect of refreshing on subsequent refreshing in older adults. Aging is associated with inhibitory deficits (Hasher et al., 2007; Hasher & Zacks, 1988). If the negative impact of prior selective refreshing is due to inhibition, older adults should show less of an effect than young adults because they should have less efficiently inhibited the non-target items while refreshing on Task 1.

Experiments 1A and 1B tested young adults and had similar designs (as described above) with one exception. In Experiment 1A, filler trials were also included. Filler trials consisted of one (single) presentation of a three word set after which participants repeated, refreshed, or read a new word. Experiment 1C tested older adults using the same methods and procedure as Experiment 1B and was run concurrently with Experiment 1B.

Methods

Participants

Young adults (YA) in Experiments 1A and 1B were recruited from the Yale community and received payment or course credit for their participation. Older adults (OA) in Experiment 1C were recruited from the New Haven community and received payment for their participation.

Experiment 1A

Participants were 30 YA (15 men; mean age = 19.03 yrs, SD = 1.19 yrs, range = 18-22 yrs). Participants had an average of 13.83 yrs (SD = .99 yrs) of education and scored an average of 23.55 (SD = 3.78) out of a possible 30 on an abbreviated version of the verbal subscale of the Wechsler Adult Intelligence Scale—Revised (WAIS-R; Wechsler, 1987).

Experiment 1B

Participants were 24 YA (9 men; mean age = 20.21 yrs, SD = 1.93 yrs, range = 18-25 yrs). Participants had an average of 14.21 yrs (SD = 1.35 yrs) of education and an average score of 22.92 (SD = 3.54) on the WAIS-R (Wechsler, 1987).

Experiment 1C

Participants were 24 OA (6 men; mean age = 72.60 yrs, SD = 6.22 yrs, range = 60-85 yrs). Participants had an average of 17.38 yrs (SD = 1.47 yrs) of education and scored an average of 22.08 (SD = 4.62) on the WAIS-R (Wechsler, 1987). OA demonstrated high levels of cognitive function as evidenced by an average Mini-Mental State Examination (Folstein, Folstein, & McHugh, 1975) score of 28.54 (SD = 1.14) out of a possible 30. Compared to YA (collapsed across Experiments 1A and 1B), OA had significantly more years of education (YA = 14.00 yrs) [t(76) = 10.88, Cohen's d = 2.67], but scored comparably on the WAIS-R (Wechsler, 1987; YA = 23.27) [t(76) = 1.22, p = .23].

Apparatus

Participants were seated with their heads centered vertically and horizontally with respect to a computer monitor, at a distance of 30-40 cm. Verbal responses were recorded using a head mounted microphone interfaced with a Psyscope button box. Psyscope software was used to control stimulus presentation and the recording of RTs.

Materials

Word stimuli for all experiments were low, medium, and high ranking exemplars of semantic categories (Battig & Montague, 1969; Shapiro & Palermo, 1970). Four exemplars were chosen from 96 different semantic categories for a total stimulus set of 384 words. Each category appeared once on each stimulus list. For Experiment 1A, the stimulus set had a word frequency (Kucera & Francis, 1967), syllable length, and letter length of 29.73, 2.04, and 6.29, respectively. Minor substitutions to the stimulus set for Experiments 1B and 1C resulted in an overall frequency, syllable, and letter length of 30.92, 1.99, and 6.18, respectively.

Three of the four exemplars were chosen such that they included one high (e.g., soda), one medium (e.g., coffee), and one low (e.g., lemonade) ranking exemplar from the category. Two of these items were designated the critical items for Task 1 (e.g., lemonade) and Task 2 (e.g., soda). The Task 1 and Task 2 targets were never the same word (and hence, were never the same rank). The fourth exemplar (e.g., water) functioned as a filler item for the Task 2 Read task. As illustrated in Table 1, the read filler replaced the critical item in the word set (therefore, rendering the critical item a new word during Task 2). This was done to equate the words being spoken across all task types when critical RTs to say the word aloud were taken. Read filler items were chosen to be of the same rank as the items they replaced.

Table 1.

Sample trials in Experiment 1. The trial started with the presentation of the three-word set. Participants read aloud three words starting with the top word. After the set disappeared, participants either refreshed or repeated one of the words (Task 1). Participants saw and read aloud the word set a second time, after which they either refreshed an item from the set or read a new word (Task 2). The Task 1 target and Task 2 target were never the same item.

		TASK 1		TASK 2
Related Trials
		Refresh (or Repeat)		Refresh
	Coffee		Coffee
	Lemonade	● (or Lemonade)	Lemonade
	Soda		Soda	●
		Refresh (or Repeat)		Read
	Coffee		Coffee
	Lemonade	● or Lemonade	Lemonade
	Water		Water	Soda
Unrelated Trials
		Refresh (or Repeat)		Refresh
	Hood		Hood
	Forgery	● or Forgery	Forgery
	Soda		Soda	●
		Refresh (or Repeat)		Read
	Hood		Hood
	Forgery	● (or Forgery)	Forgery
	Film		Film	Soda

Note. To equate target items across the tasks, a filler word was substituted for one word in the three-word set for trials in which Task 2 was a Read.

Design

Each experiment used a 2 (Task 2: Refresh, Read) × 2 (Task 1: Refresh, Repeat) × 2 (Semantic Relatedness: Unrelated, Related) within-subject design. A trial proceeded in the following manner: first presentation of the word set, Task 1, second presentation of the word set, Task 2. There were four possible Task 1-Task 2 combinations: Refresh-Refresh, Refresh-Read, Repeat-Refresh, and Repeat-Read. Half of all trials contained related items, while the other half contained unrelated. Within a stimulus list, trial order was pseudo randomized such that no more than three instances of any task or trial type could appear consecutively (e.g., no more than three trials in a row in which Task 1 was a Refresh, or in which the items were related, etc.). A practice session containing one trial of each type was conducted prior to performing the main task.

Experiment 1A

There were 16 (half unrelated, half related) instances of each Task 1-Task 2 combination for a total of 64 critical trials. In addition to critical trials, 32 filler trials (half unrelated, half related) were included. On a filler trial, the word set was presented one time, followed approximately one-third of the time by a Repeat, Refresh or Read task. Filler trials were included to prevent an expectation on the part of the participants that they would always be required to process the set a second time. Such an expectation may have engaged additional processes in anticipation of later task demands that may have obscured our main effect of interest (i.e., the effect of selectively refreshing on subsequent processing of the non-refreshed items). RTs from filler trials were not analyzed. Across different lists, critical Task 2 items were rotated through each of the four possible critical trials, and appeared twice as filler items, resulting in a total of six different stimulus lists that were rotated across participants. For Task 1, one of the two items not designated as a critical item on Task 2 was randomly assigned to be either a refresh or repeat item on Task 1. Location and rank of the Task 1 and Task 2 targets were equated such that across all trials one-third of the time the target appeared in the top, middle, or bottom location and was of high, medium, or low rank.

Experiments 1B and 1C

There were 24 (half unrelated, half related) trials of each Task 1-Task 2 combination for a total of 96 critical trials. No filler trials were included which allowed us to increase the number of critical trials. Critical items on Task 1 were rotated across both types of tasks in Task 1 (i.e., Refresh and Repeat). As Task 2 target items were also rotated across all types of Task 2 (i.e., Refresh-Unrelated, Refresh-Related, Read-Unrelated, Read-Related), this resulted in a total of eight different stimulus lists that were rotated across participants. The location and ranks of the Task 1 and Task 2 targets, as well as Task 1-Task 2 location and rank combinations (e.g., Top-Bottom combinations, High-Low combinations) were roughly equated within each condition such that across the stimulus list each combination appeared an equal number of times.

Trial Presentation

A modified version of the selective refresh paradigm (e.g., Johnson et al., 2005, Experiment 5) was used in which the three-word set contained either all related or all unrelated words. Each critical trial began with three boxes presented in a single column in the middle of the screen, with one word inside each box. These words were either semantically related or unrelated to each other and remained on the screen for 2250 ms. Participants were instructed to read aloud the words as quickly but as accurately as possible. Following a 500 ms interval, one of the two types of task occurred (Task 1). For a Task 1 Refresh task, a dot cue appeared in one of the locations previously occupied by a word. Participants were instructed to think of the word that had just appeared in the dot's location and to say that word aloud. For a Task 1 Repeat task, one of the original three words was presented again in the same location it had appeared in as part of the set. Participants were instructed to read the re-presented word aloud. The dot or word remained on the screen for 1500 ms. Following a 500 ms inter-stimulus interval, the entire word set reappeared on the screen for 2250 ms. Participants were instructed to read the word set aloud. Following a 500 ms interval, one of two types of task occurred (Task 2). For a Task 2 Refresh task, a dot cue appeared in one of the boxes. Participants thought of and said aloud the word that had just appeared in the dot's location. For a Task 2 Read task, a new word appeared in one of the boxes, which the participants read aloud. During a 1500 ms inter-trial-interval, only the three boxes remained on the screen. Participants were instructed to respond as quickly but as accurately as possible at all times.

Results

RTs to refresh or read aloud the target words on Task 2 were recorded and analyzed.¹ In all experiments, unless otherwise noted, the alpha level was set to .05 for all statistical tests and all t-tests were two-tailed.

Errors

Errors were classified as being of two types: “response” errors and “other” errors. Response errors included trials in which participants failed to respond, stuttered, responded with an incorrect word, or in which another vocalization (e.g., “um”) preceded the correct response (and thus triggered the microphone inaccurately). Other errors included trials in which an extraneous sound (e.g., a sneeze) occurred, in which an RT shorter than a minimum cut-off value (i.e., more than three standard deviations below the group mean) occurred, or in which the microphone failed to trigger due to technical problems.

The average proportions of Task 2 response (and other) errors in Experiments 1A, 1B, and 1C, respectively, were: Refresh-Refresh-Unrelated = .02 (.00), .03 (.01), .08 (.08); Refresh-Refresh-Related = .03 (.01), .03 (.02), .06 (.07); Refresh-Read-Unrelated = .02 (.01), .02 (.01), .02 (.05); Refresh-Read-Related = .00 (.03), .01 (.00), .00 (.04); Repeat-Refresh-Unrelated = .01 (.02), .02 (.00), .03 (.04); Repeat-Refresh-Related = .02 (.02), .02 (.01), .02 (.04); Repeat-Read-Unrelated = .01 (.00), .01 (.02), .01 (.02); Repeat-Read-Related = .01 (.01), .00 (.01), .01 (.01).

Thus, as expected from previous refresh studies (e.g., Johnson et al., 2002), response accuracy was high and statistical analyses of the response error rates showed no evidence of a speed-accuracy trade-off in any of the experiments. Error RTs were removed from further analysis.

RTs

To investigate whether the presence of filler trials affected performance, RTs to refresh or read on Task 2 for young adults in Experiments 1A and 1B were submitted to a 2 (Experiment: 1A, 1B) × 2 (Task 2: Refresh, Read) × 2 (Task 1: Refresh, Repeat) × 2 (Semantic Relatedness: Unrelated, Related) analysis of variance (ANOVA). Experiment was a between-subject factor and Task 2, Task 1, and Semantic Relatedness were within-subject factors. There was a main effect of Experiment with slower RTs in Experiment 1A (686 ms) than 1B (649 ms) [F(1, 52) = 4.19, MS_e = 35,137.63, η_p² = .074]. Because Experiment did not interact significantly with any of the other factors, Task 2 RTs for young adults were collapsed across Experiment in the subsequent analyses.

RTs to refresh or read a word on Task 2 (see Table 2a) were submitted to a 2 (Age: YA, OA) × 2 (Task 2: Refresh, Read) × 2 (Task 1: Refresh, Repeat) × 2 (Semantic Relatedness: Unrelated, Related) ANOVA. Age was a between-subject factor and Task 2, Task 1, and Semantic Relatedness were within-subject factors. There was a significant main effect of Age with slower RTs for OA (837 ms) than YA (670 ms) [F(1, 76) = 90.19, MS_e = 41,366.27, η_p² = .54]. There were also significant main effects of Task 2 with slower RTs to refresh (788 ms) a word than to read a new word (720 ms) [F(1, 76) = 38.05, MS_e = 16,050.49, η_p² = .33], and of Semantic Relatedness with faster RTs on related (745ms) than unrelated (762 ms) trials [F(1, 76) = 17.48, MS_e = 2,255.15, η_p² = .19]. Finally, there was a significant main effect of Task 1 with longer RTs to respond on Task 2 having just refreshed (765 ms) vs. just repeated (742 ms) on Task 1 [F(1, 76) = 24.11, MS_e = 2,948.33, η_p² = .24].

Table 2.

Task 2 RTs in milliseconds in Experiments 1 and 2. a) RTs to refresh on Task 2 were slower having just refreshed vs. just repeated on Task 1. This mean increase was larger for related vs. unrelated trials and for older vs. young adults. b) RTs to repeat on Task 2 were slower having just refreshed vs. just repeated on Task 1. This mean increase was equal for related and unrelated trials.

a) Experiment 1
		Ref-Ref	Rep-Ref	Diff	(95% CI)	Ref-Rd	Rep-Rd	Diff	(95% CI)
Young Adults
	Unrelated	691	691	0	(−23, 24)	665	657	8	(−4, 20)
	Related	700	672	28	(7, 50)	642	642	0	(−13, 12)
Older Adults
	Unrelated	927	861	66	(22, 111)	805	802	3	(−30, 36)
	Related	927	831	96	(46, 147)	764	781	−17	(−36, 1)
b) Experiment 2
		Ref-Rep	Rep-Rep	Diff	(95% CI)	Ref-Rd	Rep-Rd	Diff	(95% CI)
Young Adults
	Unrelated	608	593	15	(3, 28)	690	693	−3	(−23, 16)
	Related	590	575	15	(3, 27)	664	656	8	(−8, 23)

Note. Ref-Ref = RTs to refresh on Task 2 having just refreshed on Task 1. Rep-Ref = RTs to refresh on Task 2 having just repeated on Task 1. Ref-Rd = RTs to read on Task 2 having just refreshed on Task 1. Rep-Rd = RTs to read on Task 2 having just repeated on Task 1. Ref-Rep = RTs to repeat on Task 2 having just refreshed on Task 1. Rep-Rep = RTs to repeat on Task 2 having just repeated on Task 1. Diff = Mean increase in RTs on Task 2 having just refreshed vs. just repeated on Task 1. CI = confidence interval of the mean increase.

The following 2-way interactions were significant: Age × Task 1 [F(1, 76) = 8.82, MS_e = 2,948.33, η_p² = .10]; Age × Task 2 [F(1, 76) = 7.63, MS_e = 16,050.49, η_p² = .091]; Task 1 × Task 2 [F(1, 76) = 23.16, MS_e = 3,499.74, η_p² = .23]; Task 2 × Semantic Relatedness [F(1, 76) = 4.20, MS_e = 1,837.75, η_p² = .052]. These interactions were qualified by the presence of two 3-way interactions.

There was a significant Task 2 × Task 1 × Semantic Relatedness interaction [F(1, 76) = 7.79, MS_e = 2,010.60, η_p² = .093]. Compared to having just repeated on Task 1, having just refreshed significantly increased RTs to refresh on Task 2 for related [t(77) = 4.44, Cohen's d = .50] and unrelated [t(77) = 1.91, p = .06, Cohen's d = .22] trials. However, this increase was larger for related (Mean increase = 49 ms) than unrelated trials (Mean increase = 21 ms) [t(77) = 2.15, Cohen's d = .24]. Task 2 RTs to read both related and unrelated words did not differ as a function of Task 1 (Mean increase: Related = −6 ms; Unrelated = 7 ms) [Related: t(77) = −1.09, p = .28; Unrelated: t(77) = 1.05, p = .30].

There was a significant Age × Task 2 × Task 1 interaction [F(1, 76) = 14.50, MS_e = 3,499.74, η_p² = .16]. Compared to having just repeated on Task 1, having just refreshed increased RTs to subsequently refresh [Mean increase = 35 ms, t(77) = 3.98, Cohen's d = .45], but not to read a new item [Mean increase = 1 ms, t(77) = .14, p = .89], on Task 2 for both age groups. However, this increase in RTs to refresh on Task 2 was disproportionately larger in OA (Mean increase = 81 ms) than in YA (Mean increase = 14 ms) [t(76) = 3.83, Cohen's d = .94].

No other interactions were significant: Age × Semantic Relatedness [F(1, 76) = 1.74, MS_e = 2,255.15, p = .19]; Task 1 × Semantic Relatedness [F(1, 76) = .63, MS_e = 2,673.72, p = .43]; Age × Task 1 × Semantic Relatedness [F(1, 76) = .07, MS_e = 2,673.72, p = .79]; Age × Task 2 × Semantic Relatedness [F(1, 76) = .034, MS_e = 1,837.75, p = .85]; Age × Task 2 × Task 1 × Semantic Relatedness [F(1, 76) = .22, MS_e = 2,010.60, p = .64]².

Discussion

In Experiment 1, young and older adult participants read aloud a set of three words, and then either refreshed or repeated one of the words (Task 1). Participants then saw the word set a second time after which they either refreshed a second word from the set or read a new word (Task 2). Both age groups were slower to refresh on Task 2 having just refreshed vs. just repeated on Task 1 suggesting that briefly thinking of one item from a set has the negative consequence of reducing accessibility of the non-refreshed items. This increase in response times was greater when the items were related. While not semantically similar to each other, unrelated items, by virtue of being presented in a set, can be considered episodically similar and thus also be sources of competition. Related sets would be expected to produce a greater degree of competition (both episodic and semantic). Hence, the negative consequence of selective refreshing depends upon the amount and/or type of competition present. The finding that the nature of prior processing (refreshing vs. repeating) did not influence RTs to read a new item on Task 2 indicates that the negative impact of refreshing is greater for (or restricted to) items that were present at the time of refreshing.

Given that aging is associated with inhibitory deficits (Hasher et al., 2007; Hasher & Zacks, 1988), if refreshing on Task 1 results in inhibition of the non-refreshed items, older adults might be expected to show less of a negative impact of selective refreshing on subsequent refreshing of non-selected items. However, they showed the opposite pattern. Older adults showed a larger negative effect of prior selective refreshing relative to young adults. This finding does not support the idea that selective refreshing produces inhibition of the non-selected items. If not inhibition, what causes the reduced accessibility of the non-selected items? An alternative possibility is that reduced accessibility of non-selected items is a consequence of enhanced activation of the refreshed item. According to this “enhanced-activation” account, refreshing on Task 1 enhances the activation of the target item and does not inhibit the activation of the non-target items. This highly activated item (i.e., the Task 1 target) is now a source of strong competition on Task 2 when it becomes a distractor item. Given that aging is associated with increased vulnerability to interference (e.g., Hedden & Park, 2001), the current finding of a larger negative impact of prior refreshing in older adults is consistent with an age-related vulnerability to the presence of a highly active competitor (i.e., the prior refresh target). It is unlikely that older adults are enhancing targets to a greater degree during refreshing relative to young adults, but rather that they are more sensitive to interference from the enhanced target, even if this enhancement is relatively less than in young adults. This enhanced-activation account is also consistent with our finding that a larger negative impact was observed for related compared to unrelated trials. Because related items were presumably more active than unrelated items initially (due to shared semantic features), an enhanced Task 1 related target item will provide even greater competition when it becomes a distractor on Task 2 than an unrelated Task 1 target item. For additional analyses relevant to an inhibition vs. target enhancement account of the negative impact of refreshing on the non-selected items see the Appendix.

Experiment 2

In Experiment 2, we tested the specificity of the negative impact of refreshing. Does prior refreshing from a set of items only impair subsequent reflective access to the non-refreshed items, or would subsequent perceptual processing of the non-refreshed items be similarly impaired? The first outcome would be consistent with a segregation of reflective and perceptual representations. On the other hand, Yi, Turk-Browne, Chun and Johnson (2008) have shown that refreshing an item affects the subsequent perceptual processing of that item. Here we ask whether refreshing affects subsequent perceptual processing of the non-selected items as well. If so, that would lend further evidence for an interaction between the reflective and perceptual systems. To investigate these possibilities we ran a modified version of Experiment 1B in which participants refreshed or repeated after the first presentation of the word set and then repeated or read a new word after the second presentation of the set. If prior refreshing only impairs subsequent reflective processing, then we would not expect a negative impact of prior refreshing on subsequent repeating. If, however, prior refreshing does impair subsequent perceptual processing of the non-refreshed items, then RTs to repeat on Task 2 should be slower having just refreshed vs. just repeated on Task 1.

Methods

Participants

Participants were 16 YA (4 men; mean age = 19.56 yrs, SD = 1.59, range = 18-23) recruited as described in Experiment 1. Participants had an average of 13.19 yrs (SD = 1.38 yrs) of education and scored an average of 24.50 (SD = 2.63) out of a possible 30 on the vocabulary subscale of the WAIS-R (Wechsler, 1987).

Apparatus, materials, and design

Apparatus, materials, and design were similar to Experiment 1B, except that Task 2 Refresh trials were replaced with Task 2 Repeat trials. Hence, the design of Experiment 2 was a 2 (Task 2: Repeat, Read) × 2 (Task 1: Refresh, Repeat) × 2 (Semantic Relatedness: Unrelated, Related) within-subjects design.

Results

RTs to say aloud the target words (see Table 2b) on Task 2 were recorded and analyzed.³

Errors

The average proportions of Task 2 response (and other) errors by trial type were: Refresh-Repeat-Unrelated = .00 (.03); Refresh-Repeat-Related = .01 (.01); Refresh-Read-Unrelated = .02 (.01); Refresh-Read-Related = .02 (.01); Repeat-Repeat-Unrelated = .01 (.01); Repeat-Repeat-Related = .00 (.01); Repeat-Read-Unrelated = .01 (.01); Repeat-Read-Related = .03 (.00). Statistical analyses of the response error rates showed no evidence of a speed-accuracy trade-off. Error RTs were removed from further analysis.

RTs

RTs to repeat or read on Task 2 were submitted to a 2 (Task 2: Repeat, Read) × 2 (Task 1: Refresh, Repeat) × 2 (Semantic Relatedness: Unrelated, Related) ANOVA. There was a significant main effect of Task 2 with slower RTs to read a new word (676 ms) than to repeat a word (591 ms) [F(1, 15) = 124.03, MS_e = 1,837.48, η_p² = .89]. There was also a main effect of Semantic Relatedness with faster RTs on related (621 ms) than on unrelated (646 ms) trials [F(1, 15) = 19.29, MS_e = 1,009.78, η_p² = .56]. Finally, there was a significant main effect of Task 1 with slower RTs on Task 2 having just refreshed (638 ms) vs. just repeated (629 ms) on Task 1 [F(1, 15) = 8.80, MS_e = 275.59, η_p² = .37].

There was a marginally significant Task 2 × Semantic Relatedness interaction [F(1, 15) = 4.40, MS_e = 302.57, p < .06, η_p² = .23]. While RTs were faster on related than on unrelated trials for both reading a new word (Unrelated = 691 ms; Related = 660 ms) and repeating a word (Unrelated = 600 ms; Related = 582 ms), this relatedness benefit was larger for reading (Mean difference = 31 ms) than repeating (Mean difference = 18 ms) [t(15) = 2.10, p = .053, Cohen's d = .52].

The Task 2 × Task1 [F(1, 15) = 2.54, MS_e = 534.87, p = .13] and the Task 2 × Task 1 × Semantic Relatedness [F(1, 15) = .54, MS_e = 476.51, p = .47] interactions did not approach significance. However, given that the effects of prior processing were of a prori interest, planned contrasts were performed to examine the effect of Task 1 on Task 2 for unrelated and related trials. RTs to repeat were significantly slower having just refreshed vs. just repeated for both unrelated (Mean difference = 15 ms) [t(15) = 2.66, Cohen's d = .66] and related trials (Mean difference = 15 ms) [t(15) = 2.69, Cohen's d = .67). In contrast, RTs to read on Task 2 were not significantly influenced by the nature of Task 1 (Unrelated = −3 ms; Related = 8 ms) [Unrelated: t(15) = −.36, p = .73; Related: t(15) = 1.07, p = .30].

Discussion

In Experiment 2, participants read aloud a set of three words, and then either refreshed or repeated one of the words (Task 1). Participants then saw the word set a second time after which they either repeated a second word from the set or read a new word (Task 2). Participants were slower to repeat an item on Task 2 having just refreshed vs. just repeated from the set. This suggests that the negative impact of refreshing on refreshing found in Experiment 1 is not simply due to the same type of task being performed on the same set of items. Instead, reduced accessibility of the non-selected items resulted specifically from a prior instance of selective refreshing. While the source of the negative impact is specific to prior refreshing, the negative consequence of refreshing is general to both subsequent reflective (Experiment 1) and subsequent perceptual (Experiment 2) processing of the non-refreshed items. Additionally, this negative impact was observed even when the task was relatively simple (i.e., repeating) and RTs were relatively fast (as compared to refreshing or reading a new word). Again, this negative impact was not observed when reading new items, suggesting that selective refreshing only impairs subsequent access to the items present at the time of refreshing and does not impair a subsequent cognitive event involving a new item.

The finding of a negative impact of refreshing on repeating provides evidence that refreshing may be a mechanism by which reflective and perceptual processes interact. For example, Yi et al. (2008) demonstrated that refreshing a visual scene affects subsequent perception of the scene. After viewing a scene, participants refreshed the scene (refresh), were presented with the scene a second time (repeat), or were presented with a new scene. Later in the session, Yi et al. measured neural activity when participants were presented with the original scene again. They found repetition attenuation (i.e., reduced activity for repeated vs. novel scenes) for refreshed as well as repeated scenes in the parahippocampal place area, a region know to be activated when scenes are perceived (Epstein & Kanwisher, 1998). Hence, briefly thinking of a visual stimulus had a facilatory effect similar to perceiving the stimulus a second time. In contrast to the facilatory effect of refreshing on subsequent perception of the selected item in the Yi et al. study, the current data demonstrate a negative impact of refreshing on the subsequent perception of the non-selected items. It would be interesting to further explore whether the nature of this interaction (i.e., positive vs. negative) depends upon whether the selected or non-selected items are the focus of subsequent perception.

Despite the pattern of data in Experiment 2 and the outcome of the planned contrasts, the failure of the Task 2 × Task 1 interaction to reach significance suggests caution in interpreting these results as evidence that selective refreshing can negatively influence subsequent perceptual processing of the non-selected items. Instead, we might conclude from the main effect of Task 1 that prior refreshing generally slows subsequent perceptual processing whether or not this processing involves a new or previously seen item. If so, this negative impact on subsequent repeating might reflect a response time cost in switching between process “type” (i.e., reflective to perceptual) from Task1 to Task 2. In other words, response times may be slower to perform a perceptual process on Task 2 (i.e., repeating an item, reading a new item) if one has just performed a reflective task (i.e., refreshing) on Task 1 compared to if one has just performed a perceptual task (i.e., repeating an item). Further work is needed to determine the relative contributions of a general reflective-to-perceptual switch cost vs. a specific cost from refreshing that disadvantages perceptual processing of non-refreshed items.

General Discussion

Refreshing from a related three-word set, compared to reading one of the items again, slowed response times to subsequently refresh a previously non-selected item from the set. Hence, simply thinking of one particular item from, or one aspect of, an event may make it more difficult to think of other aspects of the experience. This difficulty does not appear to arise from inhibition of the previously non-refreshed items since older adults, who, in other tasks, show deficits in inhibition (e.g., Hamm & Hasher, 1992; Tipper, 1991), did not show less of a negative impact of refreshing, but, in fact, showed a greater negative effect.

If, as we propose, this negative impact of refreshing results from enhanced activation of the refreshed item, this suggests that reading a word a second time does not increase its activation to the same degree as does briefly thinking of the word. This is consistent with previous findings that long-term recognition memory is greater for items that have been refreshed than those that have been repeated (Johnson et al., 2002). Presumably, refreshing increases and/or prolongs the activation of the word's representation, which in turn results in more successful encoding of the item. Possibly, increasing the representation's activation in and of itself may result in a stronger memory trace. Alternatively, or perhaps additionally, foregrounding or privileging the activation of the refresh target may make the item more accessible to other cognitive processes (e.g., noting relationships to other items; binding an item representation to contextual information, e.g., Chalfonte & Johnson, 1996) involved during encoding. If refreshing is less efficient this may affect what information is available to other cognitive processes. For example, assume that refreshing the Task 1 target strengthens the binding of this item to its location (or temporal order), and that, the more well-bound the Task 1 target is, the easier it would be to select “against” it (i.e., select the Task 2 target) based on context cues. If so, less efficient refreshing in aging could contribute to a binding deficit that could, in turn, contribute to the greater negative impact of refreshing on subsequent refreshing observed here.

If the function of refreshing is to privilege a target representation, it may be maximally necessary when the target item is not the most active representation available. For example, salient information is often better remembered on a later memory task (e.g., emotional information, Bloise & Johnson, 2007; Bradley, Greenwald, Petry, & Lang, 1992; Ochsner, 2000), perhaps because it is the most highly active representation when initially experienced. As a result, perceptual processing of salient items may result in sufficient activation of these representations for successful encoding. Less salient items may require refreshing to increase their accessibility to encoding processes. Refreshing becomes most critical when the to-be-remembered item occurs in the presence of competing items of higher activation levels, either due to salience (e.g., emotional, Johnson, Mitchell, Raye, McGuire, & Sanislow, 2006) or from converging associates (e.g., from semantic relatedness). Possibly one reason why older adults have difficulties in working memory tasks (e.g., Hasher & Zacks, 1988; Hedden & Park, 2001) is due to an impaired ability to privilege less active items.

However, while refreshing may privilege a target item, it also results in reduced accessibility of the non-selected items. This negative impact of prior refreshing has implications for the use of whole report measures to characterize the limits of working memory, especially when items are semantically related. Partial report measures suggest that more items may be active than reflected by whole report measures (e.g., Sperling, 1960). The results of the current study suggest that every selective event enhances the activation of the selected item, thereby reducing the relative accessibility of the non-selected items (a type of output interference, e.g., McGeoch, 1936; Tulving & Arbuckle, 1963). This would produce lower estimates from whole report measures that require repeated selection from an active set.

The negative impact of selective refreshing is, on the face of it, similar to retrieval-induced forgetting (RIF) demonstrated in long-term memory. For example, Anderson et al. (1994) presented participants with a list of study words from different semantic categories. After study, participants were repeatedly cued to retrieve some (but not all) of the study items from a given semantic category (i.e., the practiced category). During a later test session, participants were tested on their memory for all of the original study items. Memory for items from the practiced category that were not themselves recalled during the practice session was poorer than memory for items from an unpracticed category. Anderson and colleagues propose that during the practice trials, interference from these related items from the practice category was resolved by inhibition impairing the later retrieval of these items.

Similarly, Blaxton and Neely (1983) proposed that inhibition resulting from repeated semantic retrieval can impair subsequent semantic retrieval. They used a category generation task in which participants generated an exemplar in response to a category name and exemplar letter cue. For the “generate-generate” condition, participants generated either one or four exemplar primes from a semantic category, before generating a target exemplar. In the “read-generate” condition, participants read either one or four exemplar primes before generating the target exemplar. The target exemplar was either semantically related or unrelated to the prime exemplars. In the four prime case, RTs were faster to generate a related than an unrelated target exemplar in the “read-generate” but not the “generate-generate” condition. Blaxton & Neely suggested that during each instance of prime generation, multiple exemplars were covertly activated and subsequently inhibited as part of the generation process. Hence, this retrieval-induced semantic inhibition accounts for the elimination of facilitation in the “generate-generate” condition. Consistent with this inhibition hypothesis, RTs to generate the target exemplar were significantly slower having just generated four vs. one prime, but only for related targets.

Common to the present studies and the RIF paradigms described above (Anderson et al., 1994; Blaxton & Neely, 1983), impairment of non-selected items depends on prior reflective, but not perceptual processing, of the items. Anderson et al. found impairment following multiple recalls, but not multiple study presentations, of the study items (see also Anderson, 2003), while priming in the Blaxton and Neely task was eliminated by prior generation, but not prior reading, of exemplar primes. Similarly, in the current study, RTs to refresh on Task 2 were only slower following prior refreshing, but not following prior repeating.

In the current study, impairment was observed after a single selective refresh, suggesting that a single instance of even the simplest of reflective processes (i.e., refreshing) can result in an immediate reduced accessibility of non-selected items. This inaccessibility may not last beyond a couple of seconds once the selected item is no longer foregrounded (a testable question). That a single refresh was sufficient to induce impairment in the current task contrasts with RIF observed in long-term memory. The standard RIF paradigm involves between one and up to twenty repeated retrieval events during the practice session, with a practice of one often being insufficient to result in impairment (e.g., Shivde & Anderson, 2001). Similarly, in contrast to the four generation case, a single generation resulted in facilitation and not inhibition in the Blaxton & Neely (1983) task. Possibly for RIF to occur in long-term memory, inhibition resulting from competition resolution may need to build up over multiple retrieval acts.

Anderson et al. (1994; see also Anderson, 2003) proposed that the amount of RIF observed is influenced by the amount of interference present during practice. Our finding of a larger negative impact of refreshing on related than unrelated trials is consistent with the idea that refresh-induced inaccessibility is influenced by the level of competition. However, we observed a larger negative impact of refreshing on subsequent refreshing in older than young adults, which, given evidence of inhibitory deficits associated with aging (Hasher et al., 2007), argues against the idea that the negative impact reflects inhibition of the non-selected item. A reasonable alternative is that refreshing enhances the activation of the item, making it a stronger competitor when a previously non-selected item becomes the target. Hence, while inhibition may better account for RIF, an enhanced-activation account appears to better explain refresh-induced inaccessibility.

The possibility that refreshing and retrieving (whether episodic or semantic) may resolve competition in somewhat different ways could reflect interesting differences in the characteristic dynamics of refreshing and retrieving. For example, during our refresh task, the competing information had just been explicitly processed perceptually and was, presumably, currently in an active state, whereas during retrieval practice (e.g., Anderson et al., 1994) or exemplar generation (e.g., Blaxton & Neely, 1983) tasks, the competing information is presumably not currently active and is incidentally and/or implicitly activated. It seems reasonable that processes that operate on active information versus processes that operate to revive inactive information might differ in how they resolve competition. During online processing, it may become necessary to foreground one particular item from a set of currently active items in accordance with a momentary goal. However, because the non-selected items may also be task-relevant, it may be desirable for the activation of these items to remain above threshold. By enhancing the activation of the target item (via refreshing) it becomes more accessible relative to the non-target items, while allowing the activations of the non-target items to remain above threshold. During retrieval from long-term memory, items that are currently inactive are activated in response to a cue. Non-target items that are incorrectly or inadvertently activated or reactivated will compete for selection. In contrast to the online case, these distractor items are less likely to be necessary, and their continued activation may, in fact, be disruptive to consolidation of the target information. In such cases, repeated retrieval of the target item in conjunction with repeated inhibition of distracting information may help episodic memories become consolidated.

Appendix

Influence of Exemplar Rank on the Negative Impact of Refreshing in Experiment 1

Related word sets were constructed such that there was one high, one medium, and one low rank exemplar from a semantic category. Given that high ranking exemplars have more overlapping features with other exemplars, and with the category more generally, than low ranking exemplars (e.g., Collins & Loftus, 1975; Smith, Shoben, & Rips, 1974), high ranking items should be the most active and low ranking items the least active representations within the context of this paradigm. According to an inhibition account, the more competition an item provides, the more inhibition needed (e.g., Anderson et al., 1994). Because of their higher state of activation within a related set, higher ranked exemplars should provide the most competition when they are distractors on Task 1, require the most inhibition, and thus have the lowest accessibility as Task 2 targets. Hence, an inhibition account predicts that a larger increase in response times due to prior refreshing would occur when the Task 1 target is a lower rank than that of Task 2 (hereafter, Low-High trials) than when the Task 1 target is a higher rank than that of Task 2 (hereafter, High-Low trials).

Results

Mean increases to refresh on Task 2 having just refreshed vs. just repeated on Task 1 depending on rank combination (High-Low and Low-High) in Experiment 1 are presented in Table A1. Collapsed across both age groups, this mean increase was significantly greater for related High-Low trials (Mean increase = 63 ms) than unrelated High-Low trials (Mean increase = 20 ms) [t(77) = 2.25, Cohen's d = .25], but did not differ across relatedness for Low-High trials (Mean increase: Related = 31 ms, Unrelated = 18 ms) [t(77) = .70, p = .49].

Table A1.

The impact of Task 1 refreshing on RTs to refresh on Task 2 as a function of rank combination in Experiment 1. Scores represent mean increase (and confidence interval of the mean increase) in Task 2 RTs to refresh (i.e., RTs to refresh on Task 2 having just refreshed on Task 1 minus RTs to refresh on Task 2 having just repeated on Task 1). The pattern of data is inconsistent with an inhibition account which would predict that the largest mean increase would occur on Low-High trials.

		High-Low	(95% CI)	Low-High	(95% CI)
Young Adult
	Unrelated	−4	(−38, 29)	−2	(−32, 27)
	Related	43	(12, 74)	11	(−19, 41)
Older adults
	Unrelated	75	(21, 128)	63	(−18, 143)
	Related	107	(45, 170)	78	(0, 155)

Note. High-Low = trials in which the Task 1 target was a higher ranking exemplar and the Task 2 target was a lower ranking exemplar. Low-High = trials in which the Task 1 target was a lower ranking exemplar and the Task 2 target was a higher ranking exemplar.

Discussion

If the negative impact of prior refreshing on the non-selected items were the result of these items being inhibited, this effect should be most pronounced on Low-High trials. The current results are not consistent with this hypothesis. The negative impact due to prior refreshing for related compared to unrelated items was significant on High-Low but not Low-High trials. While preliminary, the results of the rank analysis, taken together with the larger negative impact of refreshing observed in older adults than in young adults in Experiment 1, provide converging evidence that the negative impact of refreshing is not due to inhibition of the non-refreshed items.

Instead, these results are more consistent with the hypothesis that reduced accessibility is due to enhanced activation of the refreshed item and not inhibition of the non-refreshed items. As a result of being refreshed on Task 1, the mental representation of the Task 1 target is more highly activated (particularly if this item was highly activated initially, i.e., a high ranking exemplar), and consequently a particularly strong competitor during selective refreshing on Task 2 (when it becomes a non-target item). This highly activated Task 1 target may “block” (McGeoch, 1942) access to the more weakly activated Task 2 target. Additional time may be needed to resolve response conflict between the new target item and the previously refreshed target, that is, to select a less active target in the presence of a more active competitor.