Effects of age on the neural correlates of familiarity as indexed by event related potentials

Abstract

Event related potentials (ERPs) were recorded from samples of young (18–29yrs) and older (63–77yrs) subjects while they performed a modified `Remember/Know' recognition memory test. ERP correlates of familiarity-driven recognition were obtained by contrasting the waveforms elicited by unrecollected test items accorded `confident old' and `confident new' judgments. Correlates of recollection were identified by contrasting the ERPs elicited by items accorded `Remember' and confident old judgments. Behavioral analyses revealed lower estimates of both recollection and familiarity in older than in young subjects. The putative ERP correlate of recollection – the `left parietal old/new effect' – was evident in both age groups, although it was slightly but significantly smaller in the older sample. By contrast, the putative ERP correlate of familiarity – the `mid-frontal old/new effect' – could be identified in the young subjects only. This age-related difference in the sensitivity of ERPs to familiarity was also evident in sub-groups of young and older subjects in whom familiarity-based recognition performance was equivalent. Thus, the inability to detect a reliable mid-frontal old/new effect in older subjects was not a consequence of an age-related decline in the strength of familiarity. These findings raise the possibility that familiarity-based recognition memory depends upon qualitatively different memory signals in older and young adults.

Introduction

There is substantial evidence that recognition memory can be supported by two different processes, usually termed familiarity and recollection. In what is probably the most widely accepted `dual-process' account of recognition memory (Yonelinas, 1994; Yonelinas, 2001; see Diana et al., 2006; Yonelinas & Parks, 2007 for reviews of this and several other models), these processes are held to be computationally distinct and to act independently. According to this account, familiarity reflects an acontextual, continuous memory signal that does not provide qualitative details about the study episode. By contrast, recollection has a threshold-like character, and involves retrieval of consciously accessible details of the study event (see Wixted & Mickes, 2010, for an alternative account in which recollection, like familiarity, is assumed to depend on a continuous signal).

Numerous studies have investigated the effects of advancing age on recognition memory performance from the dual-process perspective (see Yonelinas et al., 2002 for review; for examples of more recent studies see Bastin & Van der Linden, 2003; Howard et al., 2006; McCabe et al., 2009; Prull et al., 2006). Although there is a strong consensus that recollection declines with increasing age (e.g., Bastin & Van der Linden 2003; Bugaiska et al., 2007; Healy et al., 2005; Howard, et al., 2006; Parkin & Walter 1992; Prull et al., 2006; for review see Yonelinas 2002), there is less consensus about familiarity. Whereas some studies have reported that familiarity estimates do not vary with age (e.g., Howard et al., 2006; Jennings & Jacoby, 1997; Parkin & Walter, 1992; Perfect et al., 1995¹; Mark and Rugg, 1998), other studies reported age-related decrements (e.g., Duarte et al., 2006; Perfect & Dasgupta 1997; Parks 2007; Toth & Parks, 2006). In one recent study, the effects of age on familiarity estimates depended on the test procedure used to operationalize and estimate familiarity (Prull et al., 2006). Although the reasons for these divergent findings are currently unclear, the findings raise the possibility that that familiarity-based recognition depends upon multiple processes or sources of information, some of which are more vulnerable to advancing age than others.

The foregoing behavioral studies addressed the question of whether recollection- and familiarity-driven recognition memory are differentially impaired as a function of age, as evidenced by age-related differences in the accuracy of different classes of memory judgment assumed to largely depend on one or the other memory process (e.g., `Remember' vs. `Know' judgments, see below). A complementary question asks not whether memory performance differs quantitatively according to age, but whether it differs qualitatively. Thus, this question focuses on whether there are age-related changes in the computational or informational bases of recollection or familiarity. Unlike the prior question, which is necessarily addressed at the behavioral level, it is arguable that this second question is most amenable to investigation with methods that identify and contrast the neural correlates of recollection and familiarity. According to this logic, if the neural correlates of successful recognition memory do not differ with age, or differ only quantitatively, this would suggest that the processes supporting recognition judgments are age-invariant. By contrast, if the neural correlates of recognition differ qualitatively with age (that is, in a manner suggesting the engagement of different neural populations) this raises the possibility that the nature of the processes supporting recognition changes with age, potentially shedding light on any corresponding differences in recognition performance. In the present study, we addressed this question using scalp-recorded event-related potentials (ERPs) to identify and contrast the neural correlates of recollection and familiarity in groups of young and older subjects.

The present study builds on extensive evidence that ERP effects elicited in young subjects by test items recognized on the basis of familiarity or recollection differ in their temporal properties and scalp distributions (for reviews see Friedman & Johnson, 2000; Mecklinger, 2006; Rugg & Curran, 2007). Both classes of effect take the form of a positive-going ERP modulation relative to ERPs elicited by correctly rejected new items. However, whereas recognition on the basis of familiarity is associated with a relatively early (ca. 300–500 ms), frontally distributed `old/new' effect (the `mid-frontal' old/new effect, also termed the `FN400'), recognition on the basis of recollection is reflected by a later effect (ca. 400–800 ms) that is often maximal over the left parietal scalp (the `left parietal' old/new effect). In recent studies in which non-recollected test items were segregated according to their familiarity strength (operationalized by recognition confidence; Woodruff et al., 2006; Yu & Rugg, 2010), the mid-frontal effect was reported to vary monotonically with familiarity strength, but to be insensitive to whether or not an item was endorsed as recollected. The left parietal old/new effect, by contrast, was insensitive to familiarity strength, and varied instead according to whether or not the test item was endorsed as recollected. These ERP effects arguably constitute a double dissociation between the neural correlates of familiarity and recollection.

In parallel with research on neural correlates of recognition memory in young subjects, a substantial literature has developed which compares these ERP correlates across groups of young and older subjects (for reviews see Cansino, 2009; Friedman, 2000, 2003; Friedman et al., 2007). A large proportion of these studies employed test procedures that did not permit ERPs to be segregated according to whether the eliciting test items were recognized on the basis of recollection or familiarity (e.g., Ally et al., 2008; Fabiani & Friedman, 1997; Friedman et al., 2010; Guillaume et al., 2009; Wegesin et al., 2002; Wolk et al., 2009). Hence, these studies cannot speak directly to the question of whether the ERP correlates of the two memory processes differ according to age. Of those studies that did segregate ERPs according to whether trials were associated with recollection or familiarity, some focused solely on ERP correlates of recollection (e.g., Li et al., 2004; Mark & Rugg, 1998; but see also Duarte et al., 2006; Trott et al.,1999, Wegesin et al., 2002). It has been reported that the recollection-related left parietal old/new effect, while often delayed in onset (e.g., Duarte et al., 2006; Mark & Rugg, 1998; Trott et al., 1999; Wegesin et al., 2002) and, in one study, smaller in magnitude in older relative to young subjects (Trott et al., 1999), does not qualitatively differ with age (that is, its scalp distribution is age-invariant). Such findings support the proposal (e.g., Rugg & Mark, 1998) that age-related decrements in recollection are largely the result of factors that impact the probability of successful recollection, and that the processes (or more correctly, those processes indexed by ERPs) engaged when recollection is successful are little affected by age.

Few studies have investigated the effects of age on the putative ERP correlate of familiarity – the mid-frontal effect. To our knowledge, to date there are only two published studies in which the effects of age on the mid-frontal effect were investigated in the context of a retrieval test that segregated familiarity- and recollection-driven recognition judgments (Duarte et al., 2006; Trott et al., 1999). Trott et al. recorded ERPs at test to sequentially presented noun pairs, each word within the pair requiring a speeded `old/new' judgment. Subsequent to the two judgments, words judged old were re-presented for both a source memory judgment (which of two study lists did the word belong to?) and a `Remember'/'Know' judgment (when subjects are instructed to endorse items as `Remembered' if recognition is accompanied by recollection of one or more study details, and `Known' if recognition is based solely on a sense of familiarity). In the young subject group, a reliable mid-frontal effect was elicited both by items endorsed as `Known', and items attracting an incorrect source. By contrast, these classes of test item both failed to elicit a reliable effect in older subjects. In the study by Duarte et al. (2006), ERPs were elicited by studied and unstudied items (grayscale pictures) during a `Remember/Know' test. The authors reported that although a robust mid-frontal old/new effect – operationalized by the contrast between ERPs elicited by test items accorded a `know' judgment versus items incorrectly judged new (misses) – was reliable in the young subject group, the effect was absent in the older subjects.

The findings of Trott et al. (1999) and Duarte et al. (2006) raise the possibility that the neural (and, presumably, the cognitive) bases of familiarity-driven recognition memory change with increasing age. These findings are subject to an important caveat, however. Although familiarity strength was not estimated by Trott et al. (1999), overall recognition memory was lower in their older than in their young subjects. Duarte et al. (2006) computed familiarity estimates for the two age groups, and reported that these estimates were significantly lower in the older subjects. Thus, in both studies, age-related differences in the magnitude of the mid-frontal old/new effect were confounded with differences in recognition memory performance. Since the magnitude of the effect covaries with familiarity strength (Woodruff et al., 2006; Yu & Rugg, 2010), it is possible that the failure of Duarte et al. (2006) to detect a reliable mid-frontal effect in their older subjects merely reflects a relatively small difference in the mean familiarity strengths of studied items accorded `Know' and `New' judgments. The findings of Trott et al. (1999) might also be accounted for along similar lines.

In light of these possibilities, the present study re-visited the question of whether ERP correlates of familiarity differ as a function of age. We employed a modified Remember/Know procedure identical to that used to segregate unrecollected test items according to their familiarity strength in prior studies (Woodruff et al., 2006; Yonelinas et al., 2005; Yu & Rugg, 2010). In this procedure, subjects respond `remember' when a test item elicits a subjective sense of recollection, but in the absence of recollection they provide a rating of their confidence that the item is studied or unstudied, rather than merely making a `know' or a `new' judgment. By forming ERPs only from items at the extremes of the distribution of familiarity strength (items accorded `confident old' and `confident new' judgments), contrasts of the ERP correlates of familiarity across age groups can largely (but not entirely; see results section below) be unconfounded with age-related differences in the strength of the underlying familiarity signal.

Experimental Methods

Subjects

Twenty-five older subjects (14 female) aged between 63 and 76 years (mean 70 yrs) and twenty-seven young subjects (19 female) aged between 18 and 28 years (mean 21 yrs) participated in the experiment at a compensation rate of $15/hr. Three young subjects were excluded because of poor EEG quality. One young subject was excluded because the subject had too few trials in a critical response category. One older subject was excluded because of poor performance during the practice of the experimental task prior to the EEG session while another fell asleep intermittently during the duration of the test. Two additional older subjects were excluded due to the contribution of too few trials in a critical response category. The final group included in the data analysis comprised 21 older subjects (13 female; mean = 70 yrs) and 23 young subjects (16 female; mean = 21yrs). Young subjects were recruited from the University of California (UCI) undergraduate community. Most of the older subjects were recruited from the local community through newspaper advertisements and flyers. Additional older subjects were recruited from the control cohort of UCI Alzheimer's Disease Research Center. All subjects were right-handed and had normal or corrected-to-normal vision and scored 26 or more on the Mini Mental State Examination (MMSE). Exclusion criteria included a history of cardiovascular disease (other than treated hypertension), a history of diabetes, psychiatric disorder, illness or trauma affecting the CNS, substance/alcohol abuse, and current or recent use of psychotropic medication. Additional exclusion criteria included a score on a standardized memory test >1.5 SD below the age-appropriate norm, or low performance (>1.5 SD below norm) on two or more of the non-memory tests on the neuropsychological test battery described below. Informed consent was obtained in accordance with UCI Institutional Review Board guidelines.

Neuropsychological testing

A standardized neuropsychological battery was administered to all participants on a separate day from the ERP session. The battery was intended to assess functioning across a broad range of cognitive domains (refer to Table 1 for a complete list).

Table 1.

Subject characteristics by age group and performance on standardized neuropsychological tests.

	Young Group		Older Group
	Mean (SD)	Range	Mean (SD)	Range	p
Age	21.0 (2.5)	18–28	70.3(3.3)	63–76
Years of education	15.3 (2.1)	13–20	17.1 (2.5)	12–21	< .01
Mini Mental State Examination	29.7 (0.6)	28–30	29.1 (1.1)	27–30	n.s.
CVLT1 immediate free recall	12.8 (1.7)	10–15	11.0 (3.0)	5–16	<.05
CVLT1 immediate cued recall	13.3 (1.8)	9–16	12.1 (2.3)	6–16	n.s
CVLT1 delayed free recall	13.7 (1.7)	10–16	11.7 (2.7)	5–16	<.01
CVLT1 delayed cued recall	14.0 (1.9)	10–16	12.4 (2.5)	6–16	<.05
CVLT1 delayed recognition	15.5 (0.7)	14–16	15.1 (1.1)	13–16	n.s.
NYU2 paragraph immediate recall	7.8 (3.5)	2–14.5	7.3 (2.2)	3.5–11.5	n.s.
NYU2 paragraph delayed recall	11.2 (2.9)	5–16	9.5 (2.8)	5–13	n.s.
Forward/backward Digit Span	18.3 (3.5)	15–27	18.3 (4.6)	12–27	n.s.
Digit/Symbol substitution test	65.7 (10.9)	52–89	48.6 (6.7)	36–59	<.001
Trail Making test A (seconds)	25.1 (7.2)	16–38	29.9 (8.5)	21–57	n.s.
Trail Making test B (seconds)	49.2 (13.0)	27–71	68.5 (18.5)	38–104	<.001
Letter Fluency	41.2 (10.0)	27–58	42.3 (9.7)	23–58	n.s.
Category fluency	24.2 (6.3)	16–42	18.9 (3.9)	12–27	<.005
WTAR3 Raw	41.8 (4.6)	34–50	43.2 (4.5)	33–49	n.s.

¹California Verbal Learning Test

²New York University

³Wechsler Test of Adult Reading Full Scale Intellectual Quotient

Materials

Stimulus lists of 300 critical items were generated from a pool of 526 words that had been selected from the Medical Research Council Psycholinguistics Database (http://www.psy.uwa.edu.au/mrcdatabase/uwa_mrc.htm; Wilson, 1987). The words were between three and nine letters in length (mean = 5.6, SD = 1.5) and had a frequency range of 0 – 50 (mean = 14.0, SD = 12.9) occurrences per million (Kucera and Francis, 1967). The study phase employed 150 words, along with two filler words at the beginning and two filler words at the end of the study list. For the test phase, the 150 words from the study list were intermixed with 150 new words. Two filler words were presented at the beginning of the test list and immediately following the mid-list break (see below). There were an equal number of words that represented living and non-living items for both study and test phases. Stimuli were displayed on a black background on an LCD monitor. All words were presented in black uppercase 30-point Helvetica font (subtending a minimum/maximum horizontal visual angle of 1.5°/3.2° and a vertical visual angle of 0.5°) within a solid grey rectangle (3.5° × 3.8°). A fixation cross (+; 0.5° × 0.5°) was presented centrally within the rectangle during the inter-stimulus intervals. Words were presented in the same manner across study and test phases.

Experimental Procedure

Practice of both study and test phases was administered prior to the experiment proper (see below). Following the practice phase, subjects were fitted with an electrode cap and seated in a sound attenuated room facing the computer monitor.

Each study trial began with the presentation of a red fixation cross for 500 ms. A word replaced the fixation cross and remained on the screen for 1000 ms. The word was replaced by a black fixation cross for 1650 ms, followed by the onset of the next trial, giving a stimulus onset asynchrony (SOA) of 3150 ms. The study session lasted approximately 8 minutes and EEG was not recorded.

The study task required subjects to make an animacy judgment to each word. Subjects indicated their response with their right and left index fingers, with response assignment counterbalanced across subjects.

The test phase began immediately after the completion of the study phase. Test trials were structured in the same manner as the study trials (see above). The test list was administered in two 9 minute sessions, with a short break every 3 minutes to reduce subject fatigue. Test instructions were to make one of five different responses depending on the quality of the memory elicited by each test word. If they remembered the word and specific details about the first time they saw the word (i.e., during the animacy study task), they were to press `R' with the thumb. In the absence of recollected details, subjects were instructed to rate their confidence about the study status of the test item using buttons on the opposite hand [`confident old' (thumb), `unconfident old' (index finger), `unconfident new' (middle finger) and `confident new' (ring finger)]. The response hand assignments were counterbalanced across all subjects so that approximately half used the left thumb for the R response and the other half used the right thumb. Subjects were instructed to respond in a timely manner but to emphasize accuracy over speed.

The practice session of the test phase was split into two sections. In the first section, subjects were required to explain the basis of each of their R responses in a self-paced version of the retrieval task. This was done to ensure there was a full understanding of the difference between an R and a `confident old' judgment. The second half of the practice session was presented in the manner of the experiment proper.

EEG Recording and Analysis

EEG was recorded continuously during the test phase from 29 silver/silver-chloride electrodes embedded in an elastic cap (EASYCAP; Herrsching-Breitbrunn, Germany; www.easycap.de). Electrode sites corresponded to the International 10–20 system (American Electroencephalographic Society, 1994) and included three midline sites (Fz, Cz, Pz) and 13 homotopic (left/right) pairs of sites (Fp1/Fp2, AF7/AF8, F3/F4, F5/F6, F7/F8, C3/C4, C5/C6, T7/T8, P3/P4, P5/P6, P7/P8, PO7/PO8, and O1/O2). Additional electrodes were affixed to the left and right mastoid processes respectively and a ground electrode was embedded in the cap at FCz. Vertical and horizontal EOG were recorded from electrode pairs situated above and below the left eye and on each outer canthus, respectively. Data were acquired with respect to a common Cz reference using a Contact Precision Instruments (London, UK; www.psylab.com) system at a 256 Hz sampling rate and an amplifier bandwidth of 0.01 to 40 Hz (−3dB). Prior to data acquisition, electrode impedances were adjusted to be under 5 kΩ and re-checked during each break period. EEG epochs (2048 ms in duration including a 102 ms pre-stimulus baseline) were extracted off-line. The epoched data were downsampled to a 125 Hz sampling rate and re-referenced to averaged mastoids. Trials were excluded if they contained artifact associated with baseline shift > 40 μV or eye movement artifact other than a blink. ERPs were averaged and smoothed with a five-point moving-window filter at a cut-off of 19.4 Hz (−3 dB). Blink artifacts were corrected using a previously described linear regression method (see Henson et al., 2004).

The principal analyses compared a group of 23 young subjects to a group of 21 older subjects. In all ANOVAs reported below, degrees of freedom associated with repeated measures factors were corrected for non-sphericity (Greenhouse & Geisser, 1959).

Results

Neuropsychological Test Scores

Performance on the neuropsychological test battery is reported in Table 1. Older subjects scored less well than young subjects on immediate and delayed free recall and delayed cued recall. Differences between the groups on immediate cued recall, recognition, and the NYU, did not reach significance. The scores of the older subjects also did not differ significantly from those of the young subjects on digit-span, letter fluency, Trail making A or the WTAR. The older subjects were impaired, however, on Trail making B, digit symbol substitution and category fluency.

Behavioral performance

Behavioral performance on the memory task is shown in Table 2 as the proportion of old and new items assigned an R response or one of the four different confidence ratings. ANOVA of these data (factors of group, study status and response category) revealed an age × study status × response category interaction (F_3.4, 141.3 = 6.46, p < .001). Follow up t-tests revealed this interaction to be driven by a greater proportion of R responses to old items in the young subjects relative to the older subjects (t(42) = 3.87, p < .001). Conversely, older subjects were more likely than the young group to respond confident new (t(42) = 4.32, p < .001) and unconfident new (t(42) = 2.94, p < .01) to old items.

Table 2.

Experimental task performance by age group

Young Group
	R	Conf Old	Unconf Old	Unconf New	Conf New
Old	0.36 (.12)	0.34 (.08)	0.20(.07)	0.09(.04)	0.02 (.02)
New	0.01 (.01)	0.03 (.03)	0.14 (.06)	0.41 (.07)	0.41 (.11)
*RT	1428 (347)	1643 (393)	1956 (439)	1934 (454)	1698 (448)
Older Group
Old	0.24 (.07)	0.34 (.11)	0.21 (.06)	0.15 (.06)	0.06 (.06)
New	0.02 (.02)	0.07(.03)	0.16(.07)	0.37(.09)	0.38(.13)
RT	1477 (336)	1647 (347)	1835 (392)	1855 (400)	1655 (418)
Standard deviations are in parenthesis

^*RTs are presented collapsed across old/new status for confidence response conditions. R response RTs are presented for old items only (see text).

Reaction times (RTs), shown in Table 2, were collapsed across study status except in the case of R responses (very few new items were endorsed `R'). An ANOVA with factors of age and response category revealed a main effect of response category (F_2.9,119.7 = 84.6, p < .001) and an age by response category interaction (F_2.9,119.7 = 4.31, p < .01). Despite the presence of this interaction, follow-up pair-wise t-tests failed to identify any response category in which RTs differed significantly according to age.

Estimates of familiarity were calculated by collapsing items in the confident old (CO) and unconfident old (UCO) response categories and calculating the proportion of items judged old on the basis of familiarity according to the assumption that recollection and familiarity are independent of one another [(p(F)=((p(CO+UCO|old)/(1-p(R|old))−((pCO+UCO|new)/(1- p(R|new)); Yonelinas & Jacoby, 1995)]. Familiarity estimates were found to be lower in the older than in the young subject group (means of .65 and .49 for young and old groups, respectively; t(42) = 4.44, p < .001). Corrected recollection estimates [p(R)=(p(R|old)- p(R|new))] were also significantly lower in the older group than the young (means of .35 and .22 for young and old groups, respectively; t(42) = 4.17, p < .001). Additionally, we calculated the accuracy associated with highly confident `old' and `new' judgments (cf. Wixted et al., 2010), the response categories employed for the key ERP analyses reported below. Mean accuracy of the old judgments (Confident Hit/(Confident Hit + Confident False Alarm)) was 0.92 for the young subjects and 0.82 for the older group. These means differed significantly (t(42) = 4.87, p < .001). Accuracy of new judgments (Confident Correct Rejection/Confident Correct Rejection + Confident Miss) also differed significantly, with means of 0.94 for the young subjects and 0.88 for the older group (t(42) = 3.67, p < .001). Thus, the strength of the memory signal supporting highly confident familiarity-driven recognition judgments was lower in the older subjects.

ERP Results

Grand average ERPs from lateral frontal and parietal electrode sites are illustrated in Figures 1 (young subjects) and 2 (older subjects). In the case of the young subjects' ERPs, an old/new effect (the mid-frontal effect), elicited by items endorsed either as recollected or highly familiar, is evident at frontal sites between around 300–500 ms. A later-onsetting parietal effect (the left parietal effect), elicited predominantly by recollected items, is also evident. Whereas the latter effect is also evident in the ERPs from the older group, the earlier mid-frontal effect is not discernable in these subjects.

Figure 1.

Grand average ERP waveforms elicited by items accorded confident old, confident new and R responses in young subjects. See insert for electrode locations.

Figure 2.

Grand average ERP waveforms elicited by items accorded confident old, confident new and R responses in older subjects. See insert for electrode locations.

ERPs were initially quantified by computing the mean amplitude (with respect to the pre-stimulus baseline) of two consecutive latency regions: 300–500 ms and 500–800 ms post-stimulus onset. These regions were selected a priori on the basis of prior studies in order to capture familiarity- and recollection-related effects respectively (e.g., Duarte et al., 2006; Woodruff et al., 2006; Yu & Rugg, 2010). In the case of the familiarity effects, we focused on the contrast between ERPs elicited by items accorded confident old vs. confident new responses, since these two response categories represent the extremes of familiarity strength. For the analysis of recollection effects, we contrasted items receiving R responses (recollected items) with items endorsed as confident old and confident new. As in the case of the RT data, ERPs elicited in association with each confidence rating were formed from items collapsed across study status (cf. Woodruff et al., 2006; Yonelinas et al., 2005; Yu & Rugg, 2010)². Mean trial numbers in the young subjects (range in parentheses) after pre-processing to eliminate trials contaminated by artifact were 40 (16–70), 41 (2268), and 44 (30–70) for the Remember, Confident Old, and Confident New categories, respectively. The corresponding numbers for the older group were 28 (18–50), 49 (18–82), and 51 (17–101).

Initial ANOVAs were conducted on data from both age groups across 12 electrodes, arranged in chains of 3 electrodes over each scalp quadrant (left anterior, right anterior, left posterior, right posterior; see Figure 1). Each electrode chain extended from an inferior to a superior site. ANOVAs were conducted separately for the two latency regions described above. Results from the initial ANOVAs are reported in Tables 3 and 4 for all effects involving the factor of response category and its interaction with the factors of age, hemisphere, anterior/posterior location, or electrode site. Subsidiary ANOVAs, reported below, were performed to elucidate the interactions revealed by the initial ANOVAs.

Table 3.

Outcome of initial ANOVA for mean amplitudes of ERPs associated with confident old and confident new responses over the 300–500 ms latency region. Only effects that involve the factor of response category are reported.

Effect	Df	F	p
Rc	1,42	17.90	<.001
Group × Rc	1,42	10.64	<.005
Rc × Hm × Fp	1,42	7.76	<.01

Rc= Response category Hm = Hemisphere Fp-Frontal/Parietal

Table 4.

Outcome of initial ANOVA, for mean amplitudes of ERPs associated with confident old and recollected responses over 500–800 ms. Only effects that involve the factor of response category are reported.

Effect	Df	F	p
Rc	1,42	108.42	<.001
Group × Rc	1,42	8.01	<.01
Rc × Fp	1,42	4.56	<.05
Rc × Hm	1,42	4.54	<.05
Rc × Site	1.1,45.0	39.78	<.001
Group × Rc × Hm	1,42	11.73	<.001
Group × Rc × Site	1.1,45.0	4.25	<.05
Group × Rc × Hm × Site	1.4,57.1	5.74	<.05

Rc= Response category Hm = Hemisphere Fp = Frontal/Parietal

Familiarity-related ERP effects

The initial ANOVA conducted on the 300–500 ms data associated with confident old and confident new responses revealed a main effect of response category and a significant group by response category interaction (see Table 3). To directly assess the effect of age on the mid-frontal old/new effect, a follow-up ANOVA was conducted on the data from the six frontal electrode sites. This revealed a reliable age by response category interaction (F_1,42 = 8.16, p < .01). Further ANOVAs on the data from each age group separately revealed a main effect of response category for the young (F_1,22 = 40.56, p < .001), but not for the old group (F<1).

The foregoing analysis of the 300–500 ms latency region indicates the presence of a mid-frontal effect in the young group only. It is possible however that the failure to find the effect in the older group merely reflects an age-related delay in its onset. To assess this possibility we performed ANOVAs on the older subjects' frontal ERPs in two additional latency regions, 400–600 ms and 500–700 ms. In each case, the ANOVA gave rise to an interaction between response category and hemisphere (400–600 ms: F_1,20 = 7.57, p < .05; 500–700 ms, F_1,20 = 17.34, p < .001). These interactions reflect a cross-over effect between the two hemispheres, such that the waveforms elicited by confident old items are somewhat more negative than those for confident new items at left frontal sites but are more positive at sites over the right hemisphere.³ Effects of response category in follow-up ANOVAs on the data for each hemisphere were however uniformly non-significant (maximum F_1,20 = 2.71, p > .1). Thus the findings from these later time-regions offer little or no support for the possibility of a delayed mid-frontal old/new effect in the older group.

In addition to the foregoing analysis of the mid-frontal effect, we also contrasted the effect in balanced subsets of subjects from each age group for whom familiarity strength was matched. We matched for familiarity by taking the maximum number of subjects (15) in whom it was possible to statistically equate the mean accuracy of confident old responses (means of 0.88 and 0.85 for the young and older groups, respectively, p > .1). Accuracy of confident new judgments (means of 0.92 and 0.89 for young and older groups, respectively, p > .1)) and overall familiarity estimates (0.59 and 0.52 for young and old groups, respectively, p > .1) were also equated. Recollection estimates however remained higher for the young subgroup (0.34 vs. 0.22, t(28) = 3.72, p < .01).

The left and right frontal waveforms elicited by items endorsed as confident old and confident new are shown for the accuracy-matched older and young subjects in Figure 3. As can be seen in the figure, whereas old/new effects are evident in the 300–500 ms latency interval at both left and right frontal sites in the young subjects, no such effects can be discerned in the older group. Although an ANOVA on the 300–500 ms data for the six frontal electrode sites across both age groups did not yield a reliable response category by age group interaction in these smaller samples, within-group analyses of these same electrode sites revealed a significant effect of response category for the young (F_1,14 = 18.94, p < .005) but not for the older subjects (F<1). Thus, the finding of an undetectable mid-frontal effect in the full sample of older subjects cannot be attributed to these subjects' reliance on a familiarity signal weaker than the signal available to young subjects.

Figure 3.

Grand average ERP waveforms, from left and right frontal electrodes, elicited by items accorded confident old and confident new responses in young and older sub-groups in whom familiarity strength was matched. Box indicates 300–500 ms time interval.

In a final analysis conducted on the data from the young subjects only, we determined whether the magnitude of the mid-frontal old/new effect varied monotonically with familiarity strength, as we have reported previously (Woodruff et al., 2006; Yu & Rugg, 2010). As can be seen from Figure 4, this was the case. As in the prior studies, we assessed whether this monotonic trend was reliable by regressing the amplitude of the ERPs in the 300–500 ms latency region at each frontal electrode against a dummy variable that coded for familiarity strength. Collapsed over the six frontal sites, the regression coefficient significantly differed from zero (mean = 0.39; t(22) = 6.05, p < .001), replicating our prior findings.

Figure 4.

A) Grand average ERP waveforms from the F3 electrode according to the associated recognition judgment in young subjects. B) Mean across-subject ERP amplitudes over 300–500 ms at three left frontal electrode sites. These ERPs were elicited by test items belonging to the four different categories of confidence judgment (from left to right: confident new, unconfident new, unconfident old, confident old, respectively).

Recollection-related ERP effects

An initial ANOVA on ERPs associated with recollected and confident old responses within the 500–800 ms latency region across all 12 electrode sites revealed a main effect of condition, along with condition by hemisphere and condition by age group by hemisphere interactions (see Table 4).

To directly compare recollection-related effects across age groups, a follow-up ANOVA on data from the six parietal electrode sites was performed. The ANOVA revealed a significant main effect of response category (F_1,42 = 120.72, p < .001) and an age group by response category interaction (F_1,42 = 9.50, p < .005). The ANOVA also revealed an age group by response category by hemisphere interaction (F_1,42 = 5.81, p < .05), which was driven by a significant age group by response category interaction over the left (F_1,42 = 17.64, p < .001) but not the right parietal sites (F_1,42 = 2.73, p > .1). These results indicate that the magnitude of the parietal effect was smaller in the older than in the younger subject group at left parietal sites only.

Additionally, ERPs associated with confident old responses were compared with those elicited by items given confident new responses. An ANOVA (factors of age group, response category, hemisphere, and site) was conducted on the data from the six parietal electrode sites. The ANOVA revealed an interaction between response category and hemisphere (F_1,42 = 11.40, p < .005), but no interaction with age group (F<1) Follow up ANOVAs revealed a significant response category effect for left (F_1,42 = 7.86, p < .01) but not right hemisphere electrodes. Thus, this analysis indicates that the small left parietal difference between confident old and confident new items evident in Figures 1 and 2 is reliable, possibly indicating that some items endorsed as confident old carried a small recollection signal (cf. Wixted and Mickes, 2010). Importantly, there was no evidence that this effect differed according to age group. The equivalent analysis conducted on the data from the familiarity-matched sub-groups (see above) revealed the same pattern of effects (response condition by hemisphere interaction, F_1,28 = 8.91, P<.01; response condition by hemisphere by age group interaction, F<1).

To assess whether the between-group differences in the magnitude of the left parietal effect reflected the lower recollection estimates in the older group (see `behavioral performance' above), we used a matching procedure analogous to that employed for familiarity strength to form two balanced sub-groups (Ns of 16 in each group)⁴ in which recollection was statistically equated (pRecollection was equal to 0.28 and 0.25 for young and older sub-groups respectively, p > .1. Familiarity estimates remained higher in the young sub-group, however: 0.63 vs. 0.49; t(30) = 3.25, p < .005). The corresponding grand average waveforms are illustrated in Figure 5, where it can be seen that the left parietal effect appears still to be smaller in the older subjects. This impression was confirmed by ANOVA of the 500–800 ms latency data from the three left parietal electrode sites, which gave rise to an age group by response category interaction (F_1,30 = 14.67, p < .005).⁵

Figure 5.

Grand average waveforms, from left and right parietal electrodes, elicited by items given confident old, confident new and recollected responses for sub-groups of young and older subjects matched for recollection. Box indicates 500–800 ms time interval.

Analysis of Scalp Topographies

The scalp topographies of the familiarity-related old/new effects in the 300–500 ms latency region (young subjects only) and the recollection-related effects in the 500–800 ms region (young and older subjects) are illustrated in Figures 6a and 6b. Statistical analyses of the topographies of the effects were performed on the same subtraction data illustrated in the figures after range normalization (re-scaling) to eliminate the confounding effects of global differences in the magnitude of the effects (McCarthy & Wood, 1985). The ANOVA contrasting the topographies of the early and later effects identified in the young subjects revealed a significant latency region by site interaction (F_{4.3, 95.1} = 4.9, p < .005), indicating that these two topographies were significantly different (see Woodruff et al., 2006; Yu & Rugg, 2010 for similar findings). A second ANOVA contrasting the topography of the recollection-related effects in the 500–800ms latency region according to age group revealed a significant effect of site (F_{3.7, 157.0} = 5.80, p < .001), but no significant interaction with age group (F_{3.7, 157.0} =1.34, p > .1).

Figure 6.

A) Scalp topography for young subjects of the difference waveforms from 300–500 ms poststimulus onset for items given high confidence old vs. high confidence new judgments. B) Scalp topographies for young and older subjects of the difference waveforms from 500–800 ms poststimulus onset for items given R and high confidence old judgments. Topographic maps are scaled to the maxima (red) and minima (blue) of each effect with the range displayed below each map in microvolts.

Discussion

The primary aim of the present study was to compare the electrophysiological correlates of familiarity-driven recognition memory in older and young subjects. In contrast to the young group, we failed to find evidence for a mid-frontal familiarity-related effect in the older sample. Below, we discuss the implications of these findings.

Behavioral Performance

Both recollection and familiarity estimates were lower in the older subject group. The findings for recollection are consistent with those of numerous prior studies, and add to the general consensus that recollection is highly vulnerable to advancing age. Although they conflict with some prior reports, the findings for familiarity are consistent with other evidence that, like recollection, familiarity can also be vulnerable to increasing age (see Introduction). The reasons for these disparate findings are currently unclear, but are not obviously accounted for by differences in test procedure. For example, whether familiarity has been estimated using Remember/Know, source memory (process dissociation) or receiver operating characteristic procedures, both null (e.g., Howard et al., 2006; Jennings & Jacoby 1997; Perfect et al. 1995) and significant (e.g., Parks 2007; Toth & Parks, 2006; Prull et al., 2006) effects of age have been reported. As we discuss below, one possibility is that the extent to which familiarity-driven recognition is impaired in older relative to younger subjects is determined by whether older subjects have access to, or make use of, the same sources of familiarity information that are available to young individuals.

ERP findings

Familiarity-related effects

The failure to detect a mid-frontal old-new effect in our older subject group replicates the findings of Trott et al. (1999) and Duarte et al. (2006; see introduction). Unlike in those prior studies, the test procedure employed in the current study minimized (although it did not eliminate) the confounding effects of age-related differences in familiarity strength on the ERP measures. Crucially, the pattern of findings for the mid-frontal effect was unaltered when the analysis was repeated in older and young sub-groups in whom familiarity-strength (operationalized by accuracy of confident old and confident new judgments; Wixted et al., 2010) was closely matched. Together, these findings indicate that the failure to detect a mid-frontal old/new effect in older subjects cannot solely be attributed to their reliance on a familiarity signal that is weaker than the signal available to young subjects.

Before continuing, it is important to note that our failure to detect a mid-frontal effect in our older subject group does not necessarily mean that the neural populations reflected by the effect were insensitive to differences in the familiarity strength of the test items – such an assertion would amount to affirming a null effect. Thus, we leave open the possibility that a study with more statistical power, or that employed more sensitive ERP analysis methods, might have detected a reliable effect in the older subjects. This caveat does not however detract from the conclusion that, if present at all, the mid-frontal old/new effect in our older group demonstrated a dramatic reduction in its sensitivity to the relative familiarity of recognition test items.

The implications of our findings for elucidating the effects of age on familiarity-driven recognition memory depend on the functional significance of the mid-frontal old/new effect. Whereas there is strong evidence that the effect is a neural correlate of familiarity strength in many circumstances (see Rugg & Curran, 2007 for review, and Stenberg et al., 2009; Yu & Rugg 2010, for relevant subsequent studies), this does not mean that it is a direct reflection of the neural processes that support the derivation and representation of item familiarity. Notably, Tsivilis et al. (2001; see also Ecker et al., 2007) reported that the mid-frontal effect was absent for studied test items presented in conjunction with novel as opposed to experimentally familiar background contexts, despite the fact that the familiarity of the items (as indexed by the Remember/Know procedure) did not differ from items paired with familiar contexts (i.e. the stimulus conjunction that did elicit the mid-frontal effect). Tsivilis et al. (2001; see also Rugg & Curran, 2007) proposed that the mid-frontal effect reflects a process, such as detection of a novel element in a stimulus event, that, while usually correlated with item familiarity, operates `downstream' of the computation of familiarity and does not contribute to item recognition.

Thus, the present findings might signify an age-related change not in the neural bases of familiarity, but in an as-yet-unidentified downstream, familiarity-dependent process. Evidence favoring this possibility comes from the finding that the mid-frontal effect can be dissociated from familiarity-strength even in young individuals (see above), along with the present finding that the effect could not be detected even in those older individuals in whom familiarity-strength was equated with a younger subject group. Further investigation of this possibility will depend upon the development of a hypothesis about the function of this putative process, along with its experimental operationalization.

Alternatively, the present findings might be indicative of a qualitative difference between the age groups in the nature of the familiarity signals supporting their recognition judgments. By this argument, the mid-frontal effect reflects only one of what might be a multiplicity of signals capable of supporting familiarity-based recognition. Either because of age-related neural degradation or, perhaps, a shift in processing strategy, older individuals do not generate this signal (at least, not in the present study or those of Trott et al., 1999 or Duarte et al., 2006), relying instead on other sources of familiarity information that seemingly do not have a scalp ERP correlate. The finding that, in both the present study and Duarte et al. (2006), familiarity was weaker in older than in young subjects offers support for this proposal if it is assumed that familiarity information not reflected by the mid-frontal effect provides a less efficient basis for item discrimination.

It is not possible to adjudicate between the above two interpretations on the basis of the currently available evidence. There is however suggestive evidence that older subjects are capable of generating a mid-frontal old/new effect under some circumstances. Ally and Budson (2008) contrasted the ERP old/new effects elicited by words and colored pictures in young and older subjects. They reported that whereas words failed to elicit a reliable mid-frontal effect in their older group (see Wolk et al., 2009 for similar findings), a robust effect was evident for pictures (see also Eppinger et al., 2010). Because the retrieval test employed by Ally and Budson (2008) merely required an old/new discrimination, it is not possible however to unequivocally associate the effect with a familiarity signal. If it does transpire that the familiarity-driven mid-frontal effect can be `rescued' in older subjects through the use of test materials such as colored pictures, this would make it less likely that when the effect is absent this can be attributed to age-related degradation of its neural generators. Why colored pictures might elicit a mid-frontal effect in older subjects when neither words (Trott et al., 1999; Wolk et al., 2009; the present study) nor gray-scale pictures (Duarte et al., 2006) appear to do so is unclear. One possibility is that the effect becomes increasingly sensitive with age to the perceptual richness of the test items. Another, perhaps related, possibility is that colored pictures engender especially high levels of familiarity, overcoming the tendency of the effect to become less sensitive to differences in familiarity strength with age. Adjudicating between these and other possibilities would be a worthwhile focus for future research.

It is important to note that the interpretation of the mid-frontal effect as a neural correlate of familiarity has not gone unchallenged. An alternative perspective (Paller et al., 2007; Voss & Paller 2006, 2007; Voss et al., 2009, 2010) views the effect as a correlate of conceptual priming, arguing that the apparent association between the effect and familiarity strength occurs only when familiarity and priming are confounded (but see Stenberg et al., 2009; Stenberg et al., 2010; Yu & Rugg, 2010). Since conceptual priming is unaffected by normal aging (see Fleischman, 2007 for review, and Bergerbest et al., 2009; Fleischman et al., 2009 for examples of more recent studies), the present findings are no easier to interpret from this perspective than they are under the assumption that the mid-frontal effect reflects familiarity strength. Thus, even more so than for the alternate proposal (as discussed above, familiarity is by no-means invariably age-insensitive), the proposal that the mid-frontal effect reflects conceptual priming must confront the finding that the effect is absent in a subject group – older adults – in whom its putative cognitive correlate is intact.

Recollection-related effects

As reported in several prior studies (e.g., Duarte et al., 2006⁶; Duverne et al., 2009; Mark & Rugg, 1998; Trott et al., 1999; Wegesin et al., 2002), older subjects demonstrated a robust, recollection-selective, parietal old/new effect. Consistent with most of these reports, the scalp distribution of the effect did not differ reliably with age, although in the present case the effect was somewhat smaller in magnitude in the older subjects (see also Trott et al., 1999). This difference in magnitude remained when recollection effects were contrasted between young and older sub-groups equated for probability of recollection. The present findings are consistent with the proposal (e.g., Rugg & Mark, 1998) that advancing age impacts the probability of successful recollection, likely due to a decline in the effectiveness of both encoding and retrieval cue processing (Morcom & Rugg, 2004; Duverne et al., 2009), but that age has less influence on the processes engaged when a cue gives rise to successful recollection. Although the finding of a smaller recollection effect in older subjects might be indicative of some degree of age-related degradation in these processes, an alternative explanation is also possible. This stems from the finding that the magnitude of the parietal old/new effect co-varies with the amount of episodic information recollected (Vilberg et al., 2006; Vilberg & Rugg, 2009). Thus, the present finding of an age-related reduction in the parietal old/new effect might reflect a tendency for recollection to have been accompanied by the retrieval of less information in older than in young subjects. Whether this tendency (should it exist) is a consequence of age differences in the amount of information initially encoded, or whether it reflects differences in retrieval processing, is an interesting question for the future.

Concluding Comments

Consistent with prior reports (Trott et al., 1999; Duarte et al., 2006), the present findings demonstrate that the putative ERP correlate of familiarity-driven recognition memory – the mid-frontal ERP old/new effect – is markedly attenuated in healthy older individuals. Furthermore, this age-related difference cannot be attributed to the confounding effect of a weaker familiarity signal in older than in younger subjects. The functional significance of this striking age-related dissociation in the neural correlates of memory retrieval remains to be elucidated, but this does not detract from the conclusion that age exerts a profound influence on the processing of recognition memory test items.