Episodic Memory and Age-Related Deficits in Inhibitory Effectiveness

Background/Study Context: Age-related deficits in inhibitory control are well established in some areas of cognition, but evidence remains inconclusive in episodic memory. Two studies examined the extent to which a loss in inhibitory effectiveness—as measured by the extent of retrieval-induced forgetting (RIF)—is only detectable in (1) the very old, and (2) that a failure to control for noninhibitory mechanisms can lead to the misinterpretation of intact inhibition in episodic memory in the very old. Methods: In Study 1, the authors employed a modified independent cue test—to provide as clean a measure of inhibitory functioning as possible—and examined whether there were significant differences in inhibitory effectiveness between younger-old (60–64 years), old (65–69 years), and older-old (70–74 years) adults. In Study 2, the authors directly manipulated the contribution of output interference (a noninhibitory mechanism) to RIF in a group of young adults (18–34 years), old (61–69 years), and older-old (70–85 years) adults. Results: In Study 1, both younger-old (60–64 years) and old (65–69 years) adults demonstrated RIF on the modified independent cue test but older-old (70–74 years) adults did not. In Study 2, all age groups demonstrated RIF in conditions where output interference was promoted. However, when output interference was controlled, only the young (18–34 years) and old (61–69 years) age groups demonstrated RIF; the older-old (70–85 years) age group did not. Conclusions: The findings suggest that inhibitory functioning remains intact in older adults under the age of 70 years. However, a misleading impression can be formed of inhibitory effectiveness in adults over the age of 70 when memory tests do not exclude the use of noninhibitory processes, such as output interference. These two issues may partly explain the previous inconclusive findings regarding inhibitory deficits in normal aging.

The notion that a deficit in inhibitory effectiveness accompanies old age continues to gain acceptance, particularly for its ability to explain age-related deficits in working memory performance (e.g., Collette, Schmidt, Scherrer, & Salmon, 2009;Hasher, 2007;Hasher, Stoltzfus, Zacks, & Rypma, 1991;Hasher & Zacks, 1988;Healey, Campbell, & Hasher, 2008;Lustig, Hasher, & Tonev, 2001;Lustig, Hasher, & Zacks, 2007;Persad, Abeles, Zacks, & Denburg, 2002;Verhaeghen & Salthouse, 1997;Wilson & Kipp, 1998; although see McDowd, 1997, for an alternative view). One of the attractions of such an inhibitory deficit model is that it provides an intuitive means of explaining why as we grow older, we seem less able to keep distracting stimuli, thoughts, and memories from coming to mind. What remains unclear, however, is whether such an inhibitory deficit is inevitable, and whether such a deficit extends to spheres of cognitive functioning beyond working memory.
From a theoretical point of view, inhibitory resources are also considered to play a pivotal role in resolving unwelcome competition from related but irrelevant information accessed by common cues during the retrieval of episodic memories (see Anderson, 2003;Anderson, Bjork, & Bjork, 1994;Bjork, 1989;MacLeod & Macrae, 2001;Macrae & MacLeod, 1999;Storm & Levy, 2012). By resolving unwanted competition, the retrieval of target memories can be promoted. By the same token, the inability to deal with unwanted competition during retrieval should result in less successful remembering of target material. If we assume this to be the case, any agerelated deficit in inhibitory effectiveness in episodic memory should mean that older adults would be less able to deal with interference from competing information during retrieval.
Despite the research effort in this field, an ambivalent picture has emerged regarding the role of inhibition in age-related decrements in episodic memory. Evidence consistent with an age-related inhibitory deficit in episodic memory comes from a range of paradigms, including list-method directed forgetting (e.g., Aslan & Bäuml, 2013;Titz & Verhaeghen, 2010;Zellner & Bäuml, 2006), and Think/No-Think tasks (e.g., Anderson, Reinholz, Kuhl, & Mayr, 2011)-both of which provide participants with explicit instructions to forget or suppress a subset of previously learned material. This act of inhibition conveys benefits for one's ability to retrieve other related material; that is, the greater one's ability to inhibit interfering information, the greater one's ability to retrieve target material on demand because of the consequent reduction in interference. In general, older adults have been found to be less able to inhibit or suppress information effectively following explicit instructions to do so, which means that recall performance for target material tends to be poorer.
In principle, there is no reason to expect any different pattern of effects when older adults are tested under conditions that promote interference and where there is no explicit instruction to forget. The reality, however, has produced a more complicated picture. This particular kind of nondirected forgetting has been extensively explored using the retrieval practice paradigm (Anderson et al., 1994). In young healthy adults, this paradigm produces a retrieval-induced forgetting effect (RIF)the magnitude of which is generally taken as a proxy for the effectiveness of inhibition; basically, the bigger the RIF effect observed, the more effective inhibition is assumed to be (although see Anderson &Levy, 2011, andNoreen &MacLeod, 2015, for some important caveats to this principle). In the current article, we set out to demonstrate why we need to exercise further caution about how such RIF effects are interpreted-especially when the retrieval practice paradigm is used with elderly populations.
In the initial stage of the retrieval practice paradigm, participants are typically presented with a series of category-exemplar pairs to study (e.g., loud-jackhammer, siren, traffic, during the final test phase of the retrieval practice paradigm; therefore, we cannot eliminate the possibility that the RIF effects observed may have been a function output interference. Participants may have tended to output stronger Rp+ items first, which may have subsequently interfered with the retrieval of weaker Rp− items. Storm and White (2010) had previously drawn attention to this potential problem when they demonstrated that inhibitory deficits in people suffering from attention-deficit/hyperactivity disorder (ADHD) only became apparent where output interference had been adequately controlled during the final test phase. Specifically, they showed that where ADHD participants were presented with category-only cues, RIF effects emerged. In contrast, where category plus stem cues were employed-and therefore the order in which items were retrieved could be controlled-no RIF effect emerged. Thus, where output interference is not adequately controlled for at final test, confounds can arise, which, in turn, prevent an adequate test of the inhibitory deficit hypothesis.
Recognizing the need to disambiguate between inhibitory and noninhibitory accounts of RIF in older adults, Aslan, Bäuml, and Pastötter (2007) reported the results of two studies, the first of which established that the magnitude of RIF in older adults was comparable to that of young adults. In their study, Aslan and colleagues not only employed the independent probe procedure (cf. Anderson & Spellman, 1995) but also took the additional precaution of not cueing participants for Rp+ items during final recall. The rationale for this novel modification was that if participants were not required to recall Rp+ items at final test, any observed forgetting effects could not be a function of stronger practiced items interfering with the retrieval of weaker items (Rundus, 1973). Having adopted these careful controls, Aslan and colleagues confirmed that there was no evidence of an age-related inhibitory deficit in episodic memory retrieval. In fact, the extent of RIF in older adults was found to be roughly equivalent to that in younger adults (−7% vs. −8%, respectively).
More recently, however, Aslan and Bäuml (2012) sought to provide a more nuanced approach by looking at older-old adults in addition to old adults. Retrieval-induced forgetting effects emerge in old adults (aged 60-75 years, M = 70 years), thereby confirming their earlier findings, but failed to emerge in very old adults (above 75 years of age, M = 84 years). This latter finding is also broadly consistent with Ortega and colleagues (2012, Study 2) who found that, on using a divided attention manipulation during the selective retrieval practice phase, young adults required a relatively demanding task before RIF effects could be eliminated. Older adults, in contrast, required only a moderately demanding task to eliminate RIF. Taken together, these studies would indicate that there is a decrease in the effectiveness of inhibition in later life, but that this loss of inhibition may only become apparent in the very old-at least, as measured by this paradigm.
Despite the additional clarity provided by such research, the majority of retrieval practice studies published to date continue to challenge the notion of an inhibitory deficit in episodic memory in older adults (see Barber & Mather, 2012;Gómez-Ariza, Pelegrina, Lechuga, Suárez, & Bajo, 2009;Hogge, Adam, & Collette, 2008;Koutstall, Schacter, Johnson, & Galluccio, 1999;Ortega et al., 2012, Study 1). Given this to be the case, it would seem particularly important to clarify whether the apparent disparity in the literature is simply a function of the age ranges of older adults sampled. Specifically, the mean age of older adults in the latter studies where no inhibitory deficits were reported was 69.3 years, whereas the mean age of older-old adults in Aslan and Bäuml's (Study 2) was considerably higher at 84 years.
In the current article, we report two studies that attempt to address these various issues. In our first study, we sought to establish the robustness of RIF in groups of younger-old, old, and older-old adults. This also allowed us to establish whether any marked differences in inhibitory effectiveness existed between these three older age groups. Following Aslan and colleagues (2007), we used independent cues. Also, Rp+ item retrieval cues were not presented at final test in order to minimize the possibility that stronger practiced items could interfere with the retrieval of related but unpracticed Rp− items. Based on Aslan and Bäuml's (2012) and Ortega and colleagues' (2012) work, we could expect RIF effects to emerge for younger-old and old adults, but to be absent for older-old adults-consistent with an inhibitory deficit account of episodic memory in the very old.
In our second study, we sought to provide a direct test of the possible contribution of noninhibitory interference to the production of RIF effects in young adults, old adults, and older-old adults. In order to accomplish this, we systematically manipulated whether practiced items were cued early (i.e., early Rp+ condition), or late (i.e., early Rp− condition), during final recall. In other words, we wished to consider not only whether RIF effects could be eliminated in the early Rp− condition but also whether they could be facilitated in the early Rp+ condition. We also looked at the pattern of second-order inhibition effects; that is, the retrieval performance for those Nrp items that are semantically related to Rp− items in comparison with Nrp items that are not semantically linked to Rp− items. The rationale here is that if inhibition is responsible for forgetting in the retrieval practice paradigm, then systematic forgetting effects should be evident for those Nrp items related to Rp− items (as well as Rp− items) by virtue of their semantic relatedness to Rp+ items (Anderson & Spellman, 1995;MacLeod & Saunders, 2005Saunders & MacLeod, 2006). Thus, if an inhibitory deficit exists only in very old adults, we could expect first-order (RIF) and second-order forgetting effects to emerge in young adults and old adults, irrespective of whether Rp+ items are cued early or late. Older-old adults, in contrast, could be expected to show evidence of first-and second-order forgetting effects only when Rp+ items are cued early. In other words, any forgetting effects in very old adults would be a consequence of noninhibitory interference rather than a function of inhibition per se.

Participants and Design
Our study had a 3 (age: younger-old, old, older-old) × 2 (item type: Rp− and Nrp) mixed design with repeated measures on the latter factor. In this study, we considered episodic memory performance in three age groups: younger-old (60-64 years), old (65-69 years), and older-old (70-74 years) adults. Sixty participants volunteered to participate in this study, with 20 participants assigned to each condition. All participants were recruited from the southern region of Wales. In order to determine comparability in intellectual functioning between groups, we used participants' mean error score on the National Adult Reading Test (NART) in order to calculate the Wechsler Adult Intelligence Scale-Revised (WAIS-R)-predicted full scale intelligence quotient (FSIQ) (Davis, Bradshaw, & Szabadi, 1999). This provided an estimate of premorbid IQ. We also used the Mini-Mental State Exam (MMSE) as a means of screening for dementia with a score below 27 as the cutoff (see Table 1 for details). No differences were detected between the age groups on the MMSE, F(2, 57) = 0.28, MSE = 0.01,ns, or FSIQ, F(2, 57) = 1.43, MSE = 1.89, ns. No participant with a past history of head trauma was recruited. We did not collect any other subjective measures of health, as previous research has indicated that they are not predictive of cognitive performance (see, e.g., Earles & Salthouse, 1995;Salthouse & Babcock, 1991;Salthouse, Kausler, & Saults, 1988;Salthouse & Mitchell, 1990).

Materials and Procedure
Given that we wished to replicate and extend Aslan and colleagues' (2007, Study 2) original finding, we employed the same procedures and experimental materials. The only exception to this was in respect of the cue-exemplar pairing "green-police," which was replaced with "green-emerald," as police officers in the UK do not wear green uniforms. In the study phase, participants were presented with a booklet containing 32 category cueexemplar word pairs (see Appendix A). Each word pair was presented on a separate sheet of paper and presented for 5 s. The order of word pairs in the booklet was randomized for each participant. Participants were paced through the booklet by the experimenter.
After studying the last word pair, participants engaged in selective retrieval practice where participants were cued to retrieve half of the exemplars (i.e., two items) from half of the categories (i.e., four categories). Participants were presented with the category cue plus the first two letters of the exemplar and given 5 s in which to respond. Participants practiced each item twice. Following the retrieval practice phase, participants worked on sets of unrelated anagrams for a period of 60 s.
In the final phase, recall performance for Rp− and Nrp items was tested. Participants were presented with a series of independent probes, each of which consisted of a new (i.e., never-before-seen) cue plus the first two letters of the target word (see Appendix A). Participants were given the following instructions: "In this task we would like you to remember the words you had studied at the beginning of the experiment. To help you, we have given you some cues or category labels that describe the item. These cues differ from the ones you had originally studied at the beginning of the experiment. Please consider the cue and then think of the words that you originally studied and whether any of those items belong to that category. For Age-Related Inhibitory Deficits example, if you studied 'strawberry' under the cue 'FRUIT' and are given the cue 'RED' at final test, then a 'strawberry' is also a 'red thing' and should be recalled in response to 'RED'." Following Aslan and colleagues (2007, Study 2), we chose not to cue Rp+ items at final test in order to eliminate stronger practiced items interfering with the retrieval of weaker unpracticed items. Participants were given 4 s in which to respond to each cue. This final recall task was paper-and-pencil based. On completion, participants were paid £3.00 (~$4.73) for their participation, thanked, and debriefed.

Results and Discussion
Retrieval practice success rates were 71.88%, 70.63%, and 71.25% for the younger-old, old, and older-old age groups, respectively. Note that facilitation effects (i.e., Rp+ vs. Nrp baseline) are not reported here, as Rp+ items were not cued at final test. Holm's sequential Bonferroni approach was employed for all multiple comparisons.
Given that we had employed independent cues at final test (Anderson & Spellman, 1995) and that no Rp+ items had been cued, we can be reasonably confident (according to inhibitory account) that when RIF effects emerged for younger-old and old adults, they were likely to have been a function of inhibition. Following this logic, we could also expect that when no RIF emerged in the older-old group, this is likely to have been due to a deficit in inhibitory functioning-at least, as measured via this particular paradigm.
One of the complexities with this inhibitory deficit interpretation in the older-old group is that it involves the interpretation of a null effect. In our second study, therefore, we sought to provide a further test of the age-related inhibitory deficit hypothesis (see Levy & Anderson, 2008;Storm & White, 2010) by assessing directly the contribution of noninhibitory interference to the production of RIF. As indicated previously, one of the difficulties in establishing the action of inhibition is the need to eliminate the possibility that the observed forgetting effects may be a function of noninhibitory mechanisms. To date, this has been addressed by employing a number of modifications, including the independent cue method (Anderson & Spellman, 1995), recognition test procedures at final test instead of recall (Aslan & Bäuml, 2012;Ortega et al., 2012; but see Verde & Perfect, 2011, for an alternative noninhibitory account), asking participants to engage in retrieval practice for "impossible" exemplars (Storm, Bjork, Bjork, & Nestojko, 2006), or not cueing Rp+ items at final test (Aslan et al., 2007; Study 1 of the current article). In our second study, we decided to take a novel approach whereby we directly manipulated the likelihood of stronger practiced items interfering with the retrieval of weaker unpracticed items. In other words, could we directly manipulate the magnitude of RIF by manipulating the prevailing recall conditions and, thereby, demonstrate the importance of controlling recall order when looking at older populations?
We set out to accomplish this by assigning participants to conditions in which they were required to retrieve Rp+ items early or where they had to retrieve Rp− items early during the final test phase of the retrieval practice paradigm. Our rationale was that if there is no inhibitory deficit in older-old adults, we could expect RIF to emerge irrespective of whether Rp+ items are cued early or late. If Rp− items are cued early and a RIF effect still occurs, it is unlikely to be due to stronger practiced items interfering with the retrieval of weaker unpracticed Rp− items. Thus, where RIF effects emerge under these retrieval conditions, we can be reasonably confident that they are not due to noninhibitory means and that we can infer that inhibition remains intact. In contrast, if there is a deficit in inhibitory functioning in older-old adults, we could expect RIF to emerge only where stronger Rp+ items are cued early at final test-based on the assumption that the early retrieval of Rp+ items can contribute to RIF by interfering with the subsequent retrieval of weaker Rp− items (i.e., output interference). It is worth noting in this regard that Aslan et al. (2007) reported a drop of 5% in the magnitude of RIF in older adults between their first study and second study where they had initially not controlled for output interference.
Given that, in Study 1, we found no significant difference in inhibitory functioning between young-old and old adults, we could expect that young adults and old adults (up to age 69 years) could be expected to produce RIF effects, irrespective of whether Rp+ items are cued early or not. In other words, RIF should be apparent in these two age groups in both retrieval conditions, consistent with there being no inhibitory deficit. If, in contrast, older-old adults (70+ years) have a deficit in inhibition, we could expect RIF to emerge only when Rp+ items are cued early during the final retrieval phase (i.e., where the retrieval of stronger items can interfere with the retrieval of subsequent weaker Rp− items).
As an additional test of this inhibitory account of forgetting, we also constructed our materials in such a way to permit second-order inhibition effects to be explored. Specifically, our materials were constructed so that some Nrp items were similar (i.e., semantically related) to Rp− items, or dissimilar to Rp− items (i.e., semantically unrelated). According to inhibitory theory, all items that are related to the practiced target item should be subject to inhibition; that is, inhibition should affect not only Rp− items by virtue of their relatedness to Rp+ items but also any Nrp items that could compete for retrieval with either Rp + items or Rp− items. Thus, as well as the inhibition of Rp− items (first-order forgetting effects), we could expect any Nrp-similar items to be inhibited relative to an Nrp-dissimilar baseline (second-order forgetting effects). Given this to be the case, we could expect such second-order effects to emerge for young adults and old adults, irrespective of whether Rp+ items are cued early or not as inhibition should remain effective. For older-old adults, however, second-order effects should emerge only when Rp+ items are cued early, consistent with an inhibitory deficit account.

Participants and Design
One hundred and eighty participants volunteered to participate in this study (60 young adults, 60 old adults, and 60 older-old adults). Young adults were included in order to provide an appropriate baseline (i.e., intact inhibition) against which performance by old and older-old adults could be compared. As these experimental materials had not previously been tested on these populations, it was important to establish a baseline performance in young adults. Young adults were recruited from the Universities of Swansea and Strathclyde, and old and older-old adults were recruited from the local community in South Wales. Half of each of these groups was randomly assigned to either an early Rp+ condition or an early Rp− condition.
As in the previous study, the mean NART (National Adult Reading Test) error score was used to calculate the WAIS-R-predicted FSIQ (Davis et al., 1999), which, in turn, was used to estimate premorbid IQ. The Mini-Mental State Examination (MMSE) was used to screen for dementia with a score of below 27 as the cutoff. We also applied these tests to our young adults sample for completeness and comparability (see Table 1 for further details). A good level of comparability in intellectual functioning was found across all three age groups in both the early Rp+ and early Rp− conditions. There were no interactions between age and output order on either the MMSE, F(2, 174) = 0.88, MSE = 0.74, ns, or the WAIS-R-predicted FSIQ, F(2, 174) = 0.27, MSE = 14.12, ns. As in Study 1, participants were not included if they had reported a previous history of head trauma. No one was excluded on this basis.
Items in this study were classified in the following way: Rp+ items were practiced items from practiced categories; Rp− items were unpracticed items from practiced categories; and NsimRp− items were Nrp items drawn from nonpracticed categories that shared an implicit category with Rp− items; and NsimC items were unpracticed items from unpracticed categories drawn from the semantically similar set but which did not share an implicit category with Rp− items. Finally, due to the potential for list-order effects to occur during the final recall phase, we split the Nrp dissimilar group into two sets: those Nrp items that were reported in the first half of the recall procedure (NdissC (A)) and those Nrp items that were reported in the latter half of the recall procedure (NdissC(B)). NdissC(A) and (B) were drawn from the semantically dissimilar set, which did not share an explicit or implicit category with any of the items that formed either the Rp+ or Rp− sets. See Appendix B for illustration.

Second-Order Similarity and Implicit Categories
The materials used in this study were identical to those previously employed by Anderson and Spellman (1995, Experiments 2, 3a, and 4) and comprised six categories (cotton, leather, green, soups, loud, and sharp), each containing six exemplars (see Appendix B for details). Three of the exemplars in each category were semantically dissimilar to exemplars in other categories but were semantically similar to other members of their own category. Appendix B provides a detailed breakdown of the associations between exemplars for different categories. The remaining three exemplars in each category were semantically similar to three members of a second implicit category (e.g., pajamas, robe, and slacks are originally studied under the cue "cotton" but are also semantically related to belt, boots, and skirt, as they are all exemplars of the category "clothing"). However, the three remaining items (i.e., curtain, napkin, and sheet) are unrelated to any other items outside of their initially studied category.
In the presentation phase, participants were presented with all 36 category-exemplar pairs using the same procedure as in Study 1. During the retrieval practice phase, each participant was presented with three exemplars from two of the available six categories (i.e., six Rp+ items in total). Items selected for retrieval practice were always derived from the semantically dissimilar set of items (see Appendix B). Although these items did not share similarity between subjects, they did, of course, share similarity within subjects (i.e., with the Rp− items). This was to ensure that any second-order inhibition that might be observed was due to the association between Rp− and NsimRp− items and not due to an association with the practiced set (i.e., first-order inhibition; see Anderson & Spellman, 1995;Saunders & MacLeod, 2006). For example, participants might practice the exemplars curtain, napkin, sheet, dollar, emerald, and lawn from the categories "cotton" and "green." The three remaining exemplars in each of these two categories would then become the Rp− items (e.g., pajamas, robe, slacks, artichoke, lettuce, and pepper) because they shared a semantic similarity with the Rp+ items (within-subjects similarity). These Rp Age-Related Inhibitory Deficits − items, however, also shared an implicit category with items belonging to another category (i.e., between-subjects similarity). Rp− items from the category "cotton," for instance, shared an implicit category "clothing" with some of the items from the category "leather" (i.e., belt, boots and skirt). Similarly, Rp− items from the category "green" shared an implicit category "vegetables" with some of the items contained in the category "soups" (i.e., mushroom, onion, and tomato). The three exemplars in each of the other two categories formed the NsimRp− items (i.e., six items in total). There were also six NsimC items that acted as a baseline for measuring second-order inhibition. These six items were the three remaining dissimilar items from the two implicit categories; that is, although they were semantically related to the NsimRp− items, they were not related to the Rp− items (e.g., briefcase, saddle, whip, chicken, clam, and turkey).
The remaining 12 exemplars (i.e., 3 exemplars from four categories) comprised the NdissC items (e.g., cannon, grenade, gun, dagger, spear, sword, jackhammer, siren, traffic, needle, tack, and thorn). Thus, NdissC items are unpracticed items drawn from the semantically dissimilar set. Although NdissC items were semantically related within category to one another, they were not semantically related between category to any other items (i.e., within-subject similarity but between-subjects dissimilarity). See Table 2.

Output Order
As one of the critical manipulations in our study was to control for list-based output interference (i.e., where items that appear later during recall are less likely to be reported due to the retrieval of earlier items interfering with access to later items), we needed to construct two sets of NdissC items (which we refer to here as set A and set B). In the early Rp+ condition, Rp+ items were cued for recall at final test during the first half of the recall procedure alongside half of the NdissC items (i.e., NdissC(A) items). In contrast, Rp−, NsimRp−, NsimC, and the remaining half of the Nrp control items (i.e., NdissC(B) items) were cued for recall in the latter half of the final recall procedure. Conversely, in the early Rp− condition, Rp− items were reported first intermingled with NsimRp−, NsimC, and half the Nrp dissimilar control items (i.e., NdissC(A) items). Rp+ items and the remaining Nrp dissimilar controls (i.e., NdissC(B) items) were then reported in the second half of the recall procedure. The division of the Nrp dissimilar control group into two sets allowed us to control for list-based output interference. Specifically, in the early Rp+ group, Rp− recall can be compared with NdissC(B) items, as both sets were reported in the second half of the final recall procedure. NsimRp− can also be compared with NsimC items, as they were both reported in the second half of the final recall procedure, and Rp+ items could then be compared with NdissC(A) items, as both sets of these items were reported in the first half.
In the early Rp− condition, Rp− items were compared with NdissC(A) items, as both sets of items were reported during the first half of the recall procedure. NsimRp− items were compared with NsimC items, as both sets of items were reported during the first half of recall, and finally Rp+ items were compared with NdissC(B) items, as both sets of these items were reported during the latter half of recall.

Procedure
Participants arrived at the laboratory and were greeted by the experimenter. On completing the NART and MMSE, participants were presented with 36 cue-exemplar word pairs (i.e., six categories containing six exemplars as outlined in Appendix B). Each cue-exemplar word pair appeared separately on each page of a study booklet. Participants were given 5 s in which to study each word pair. The order of word pairs was randomized for each participant (note that items were not presented in blocked fashion by studied category).
Following completion of the study phase, participants performed the selective retrieval practice procedure. Specifically, participants were presented with cue plus two-letter word stems to complete (e.g., cotton-cu______). Three exemplars from two categories were chosen for each participant (i.e., six exemplars in total), and participants were cued to retrieve each item three times each (i.e., 18 prompts in total), and these items were always semantically dissimilar items. Order of presentation was randomized for each participant during the retrieval practice phase and counterbalanced across the study.
Following retrieval practice, participants worked on unrelated word search puzzles for 60 s. Finally, participants completed a cued-recall phase employing the original category cues from the study phase plus the first two letters from the to-be-remembered item. For this phase, there were two conditions: an early Rp+ condition in which Rp+ items were cued in the first half of recall along with the NdissC(A) items. Rp−, NsimRp−, NsimC, and NdissC(B) items were presented for retrieval in the second half of the recall phase (see Table 1 for an example of the assignment of items to this condition). In the early Rp− condition, recall commenced with Rp− items along with NsimRp−, NsimC, and NdissC(A) items. Participants were then cued for the retrieval of Rp+ and NdissC(B) items in the second half. Participants were given 5 s in which to report each item before moving on to the next item. Within each first half of the final recall phase, order of presentation was randomized, but care was taken to ensure that no two items from the same category followed one another. On completion, participants were paid £3 ($4.73), thanked, and debriefed.

Facilitation Effects
The comparison between Rp+ items and Nrp dissimilar items in the early Rp+, and early Rp−, output groups was based on different Nrp dissimilar sets (A or B). Specifically, in the early Rp+ output condition, Rp+ items were tested in the first half of final recall, and so these items were compared with NdissC(A) items because they were also tested in the first half of the final recall session. Conversely, in the early Rp− output condition, Rp+ items were tested in the second half of final recall, and so they were compared with NdissC(B) items, which were also tested in the second half of recall.
It can also be seen from the pattern of performance that retrieval practice effects increased with age (early Rp+ groups: young M retrieval practice effect = .13, old M = .19, and older-old M = .24; early Rp− groups: young M = .21, old M = .25, and older-old M = .37), r(180) = .19, p < .01. It should be noted here, however, that this is most likely to be an artifact in that for young adults, recall of Rp+ items was already very high and therefore it would be difficult to prime these items much further. In the two older groups, recall of Nrp− control items was much lower, thereby giving greater opportunity for retrieval practice effects to emerge.
In the early Rp− condition, paired-samples t tests similarly revealed that fewer Rp− items were reported than NdissC(A) items in young adults (Ms = .45 vs. .68, respectively), t(29) = 3.36, p < .01, d = .28; and in old adults (Ms = .36 vs. .54, respectively), t (29) = 3.75, p < .001, d = .33. There was, however, no evidence of RIF in the older-old age group in the early Rp− condition. In fact, there was evidence of some facilitation for Rp− items in comparison with the NdissC(A) baseline (Ms = .57 vs. .44, respectively), t (29) = 5.64, p < .001, d = .52. It is unlikely that this can be attributed entirely to a drop in performance in the NdissC(A) baseline.
An interference account would predict that stronger practiced Rp+ items in this condition would interfere with the retrieval of related weaker Rp− items. Similarly, we could expect that the early retrieval of Rp+ items would interfere with the subsequent retrieval of NsimRp− items because of their association to Rp− items. In other words, just as first-order forgetting effects can be produced by noninhibitory means, so can second-order forgetting effects. Where retrieval conditions promote the early retrieval of Rp+ items, we could expect to find first-and second-order forgetting effects to emerge even in the very old, as they would be independent of inhibition. Indeed, correlational analyses confirmed that there were significant correlations between size of the inhibitory effect and age (RIF: r(180) = .24, p < .001; second-order Early Rp+ Early Rp- Figure 2. Mean proportion correctly recalled as a function of item type, age and output order (Study 2). Early Rp+ = output of Rp+ items before Rp−, NsimRp−, and NsimC items; Early Rp− = output of Rp−, NsimRp−, and NsimC items before Rp+ items; Rp+ = practiced items from practiced categories; Rp− = unpracticed items from practiced categories; NsimRp − = unpracticed items from unpracticed categories that are semantically related to Rp− items; NsimC = unpracticed items from unpracticed categories that were drawn from an implicit category but which did not share a relationship with the Rp− items; NdissC(A) = Nrpdissimilar items tested during first half of recall; NdissC(B) = Nrp-dissimilar items tested during second half of recall. The pattern of first-and second-order forgetting effects observed here indicate that an inhibitory deficit is present but only in the very old. Neither first-nor second-order forgetting effects emerged for older-old adults in the early Rp− condition where output interference had been minimized. If inhibition had been intact in older-old adults, we could reasonably have expected forgetting effects to have emerged in the early Rp− condition as well as the early Rp+ condition-just as they had with young and old adults. The fact that these effects failed to emerge where output interference had been minimized would suggest that where this kind of forgetting effect is observed in very old adults, it is unlikely to be a function of inhibition.

GENERAL DISCUSSION
Although it has been widely assumed that certain populations such as young children and healthy older adults, and patients with schizophrenia, Alzheimer's disease, ADHD, or frontal lobe damage can suffer from inhibitory deficits, many studies have failed to find evidence of such deficits in episodic memory (e.g., Aslan et al., 2007;Conway & Fthenaki, 2003;Ford, Keating, & Patel, 2004;Gómez-Ariza et al., 2009;Moulin et al., 2002;Nestor et al., 2005;Zellner & Bäuml, 2005).
On first inspection, this would seem to be inconsistent with an inhibitory deficit account. It is important to realize, however, that RIF effects are not always a function of inhibition (see also Anderson & Levy, 2007, for a discussion). In the current article, we demonstrated how the early retrieval of Rp+ items can interfere with the retrieval of weaker Rp− items at test and, thereby, produce RIF effects. It is also worth noting here that many of the failures to find inhibitory deficits in "at-risk" populations have used category-cued recall, which leads to the possibility that such output interference may have driven the RIF effects observed. In contrast, when such interference is adequately controlled during the final recall stage of the retrieval practice paradigm, inhibitory deficits become apparent in these populations (e.g., Aslan & Bäuml, 2010;Soriano, Jiménez, Román, & Bajo, 2009;Storm & White, 2010). The current findings provide further support for this perspective and indicate that when attempting to measure the effectiveness of inhibitory functioning in episodic memory retrieval in "at-risk" or typically developing populations, particular care needs to be taken to control for these potential confounds.
In our first study, we took particular care to ensure that any observed forgetting effects could only be attributed to inhibition. We accomplished this via the use of independent cues at final test, and by not cueing strong practiced Rp+ items at final test. Retrievalinduced forgetting emerged in younger-old and old adults but not in older-old adults. In our second study, we provided an even more stringent test of the age-related inhibitory deficit hypothesis by expressly manipulating whether Rp+ items were cued early, or late, during final test. The rationale here was that if RIF effects emerged in the early Rp− condition, we could reasonably assume it must be due to inhibition. Similarly, if the prior retrieval of Rp+ items contributed to RIF, we could expect forgetting effects to emerge in the early Rp+ condition for all three age groups. We found that RIF emerged in young adults and old adults, irrespective of output order, but for older-old adults, RIF emerged only when Rp+ items were cued early. Thus, it would seem that where inhibition remains intact, RIF emerges irrespective of the possible contribution of noninhibitory mechanisms such as interference. Indeed, one possibility is that where inhibition remains effective, such inhibition may have primacy over other potential contributory mechanisms.
More speculatively, perhaps, Study 2 raises the issue of whether the RIF observed in the early Rp+ condition may actually be a function of a different form of inhibition. Bäuml (1998), for instance, raised the possibility that output order effects could be considered a form of inhibition (see also Anderson et al., 1994). In a sense, both RIF and output interference effects reflect a memory impairment that is caused by the prior retrieval of related material. Thus, there is the possibility that two inhibitory mechanisms may be at play in the retrieval practice paradigm and that these different inhibitory mechanisms may mediate RIF in different age groups.
It is also worth noting that Study 2 showed some evidence of facilitation for Rp− items and NsimRp− items. These effects may be due to increased relational processing in older adults. Previous research using the Deese-Roediger-McDermott (DRM) paradigm has found that item-specific processing decreases false alarms in younger, but not older adults, as well as decreasing veridical recall in older adults (Tun, Wingfiled, Rosen, & Blanchard, 1998). These findings suggest that older adults rely on shared cues at encoding to improve recall (i.e., relational encoding), whereas younger adults rely on both shared and unique cues. Tun and colleagues also found that item-specific processing did not reduce the proportion of false alarms relating to weakly related lures in older adults, suggesting that activation may spread even further within the associative network than in younger adultsperhaps because of a lack of effective inhibition. Thus, if older adults are more likely to encode information relationally, they may be more likely than younger adults to identify the link between Rp+, Rp−, and NsimRp− items. Assuming this to be the case, it is possible that when inhibition is no longer functioning effectively, it could actually lead to an increase in the recall of other items because of their relatedness.
In summary, our studies have produced patterns of forgetting in young adults, youngerold, and old adults that are difficult to account for by traditional noninhibitory explanations, such as interference. We also demonstrated that we could produce or eliminate RIF effects in very old adults simply by manipulating the nature of the retrieval conditions. Our studies provide strong inference that there is an inhibitory deficit in episodic memory retrieval but that this deficit only becomes apparent once very old age is reached. Our studies also point to the possibility of a decline in inhibitory function as we grow older, but this will only be confirmed via longitudinal studies that consider memory performance and inhibitory effectiveness. Clearly, if we are to find ways in which to maintain cognitive performance in elderly adults, we also need to fully understand how retrieval conditions can influence memory performance. Only by doing so are we likely to meet the various challenges posed by an ageing population. Indeed, only then will we truly be able to determine whether an age-related decline in memory performance is inevitable.