Full text of DOI:10.1093/geronb/57.2.P173

Journal of Gerontology: PSYCHOLOGICAL SCIENCES
Copyright 2002 by The Gerontological Society of America
2002, Vol. 57B, No. 2, P173–P186
Differential Age Effects in Semantic and
Episodic Memory
Philip A. Allen,1 Martin Sliwinski,2 Tanara Bowie,3 and David J. Madden4
1The University of Akron, Ohio.
2Syracuse University, New York.
3Cleveland State University, Ohio.
4Duke University Medical Center, Durham, North Carolina.
Results from 4 experimental tasks and 8 data sets (the 4 tasks involved either multiple sessions or different stimuli) as
well as a vocabulary test conducted on the same 80 participants (40 younger and 40 older adults) are reported. The au-
thors employed 2 semantic memory tasks (lexical decision and multiplication verification) using data from 2 sessions
(for a total of 4 semantic data sets) and 2 episodic memory tasks (hybrid visual search and memory search with digits
and with words as stimuli). Factor analyses using slope and intercept data from the 8 experimental data sets indicated
the presence of 3 latent factors: a single intercept factor for both episodic and semantic tasks and separate slope factors
for episodic and semantic tasks. A structural equation model with paths from age to 3 different 1st-order latent factors
(episodic central processes, semantic central processes, and combined episodic and semantic peripheral processes) fit
better than general factor models. These data are consistent with a theoretical framework in which there are
age-related dissociations between peripheral and central processes across semantic and episodic memory.
UR goal in the present study was to examine whether
cognitive aging is generalized or process/stage-specific
tic, although two different stimulus types were used for the
episodic tasks and participants were tested twice on each se-
mantic task so that there were four episodic and four seman-
tic data sets on which to examine intercepts and slopes) and
a standardized vocabulary test (which is a semantic task).
The research question was whether age differences are gen-
eralized or process-stage-specific. If age differences are
generalized, then age differences for semantic and episodic
memory should be comparable across slope and intercept
data. If age differences are process-specific, then differen-
tial age effects should be present across semantic and epi-
sodic slope and intercept data.
We used these particular tasks because all (except the
vocabulary test) involved RT as the primary dependent
variable, all are relatively simple tasks involving both pe-
ripheral- and central-process components, and all have diffi-
culty manipulations that have been studied extensively (e.g.,
Allen, Madden, Weber, & Groth, 1993; Fisk & Rogers,
1991; Geary & Wiley, 1991). Hybrid visual search and
memory search involve difficulty manipulations of probe
set size and memory set size, respectively (Fisk & Rogers,
1991). Lexical decision and multiplication verification tasks
involve difficulty manipulations of word frequency (Allen,
Madden, & Crozier, 1991) and problem size (Geary &
Wiley, 1991), respectively. We know of no other memory
type manipulation of tasks, using RT as the dependent vari-
able, for which the difficulty manipulations are more com-
parable. The present tasks also allow us to use words and
digits as stimuli for both episodic and semantic tasks. Finally,
all four tasks require that participants compare presented
stimuli to information stored in memory—this suggests that
all these tasks involve a similar memory comparison pro-
cess. This comparison process is crucial because it is at the
core of our central-process manipulation.
O
using a psychometric approach. To examine this issue, we
replicated and extended Mitchell’s (1989) study, which ob-
served differential age effects in episodic and semantic
memory. We were particularly interested in examining age
differences in memory type across processing stage. The
tasks in the present investigation used reaction time (RT) as
the primary dependent variable so that we could use Stern-
berg’s (1967) method of measuring peripheral and central
processes. (Mitchell, 1989, did not use RT as a dependent
variable in several of his tasks, and he did not separate slope
and intercept data.) Sternberg’s (1967; Roberts & Sternberg,
1993) method of indexing peripheral processing stages (en-
coding and response execution) by using RT intercepts and
indexing central processing stages (storage, retrieval, re-
sponse selection, and decisions) by using RT slopes (for ag-
ing applications, see Allen, Smith, Jerge, & Vires-Collins,
1997; Bashore & Smulders, 1995; Cerella, 1985, 1991) has
been applied widely in experimental psychology. Sternberg
(1967, 1969) proposed that slopes measure the comparison
time per item (for matching tasks such as memory search,
hybrid visual search, multiplication verification, and lexical
decision), but that the intercepts measure the duration of all
other processes not involved in the comparison process. RT
slopes and intercepts were computed in the present study as a
function of task difficulty. For example, in the present series
of experiments, we used problem size (multiplication verifi-
cation), word frequency (lexical decision), memory set size
(memory search using digits or words), and visual search set
size effects (hybrid visual search using digits or words) as
measures of task difficulty.
In the present study, we tested 40 younger adults and 40
older adults on four RT tasks (two episodic and two seman-
P173

P174
ALLEN ET AL.
The Episodic Versus Semantic Memory Distinction
Mitchell (1989) reported evidence for the existence of
three memory systems in an aging study. The first system is
semantic, or context-independent, memory. Tulving (1985,
p. 386) defined semantic memory as an “organism’s knowl-
edge of its world.” The second system is episodic, or con-
text-dependent, memory, which involves the conscious re-
membrance of “personally experienced events and their
temporal relations” (Tulving, 1985, p. 387). The third sys-
tem is procedural memory (knowledge and memory for how
to do procedures; Tulving, 1985). In the present study we
concentrated on episodic and semantic memory because fre-
quently there are no age differences in procedural memory
(e.g., Light & Singh, 1987).
There is disagreement about the semantic/episodic distinc-
tion. McKoon, Ratcliff, and Dell (1986) proposed that there
is no theory that predicts the differential manner in which
episodic and semantic information is encoded, stored, and
retrieved. Also, Roediger (1984) claimed that many of Tul-
ving’s functional dissociations between episodic and se-
mantic memory involved only a single measure of each
memory type. To show true converging operations, investi-
gators must use at least two measures of each memory type
so that they can test whether both of the “episodic” mea-
sures index the episodic memory system but not the seman-
tic system, and vice versa for the “semantic” tasks.
Mitchell (1989) used the multiple dissociation method sug-
gested by Roediger (1984) and found that (a) older adults
showed a performance decline in episodic memory tasks but
not in semantic memory tasks, and (b) exploratory factor
analysis revealed separate latent factors for episodic and se-
mantic memory. Consequently, Mitchell provided empirical
evidence of a functional dissociation between episodic and
semantic memory using two or more measures of each mem-
ory type, and the exploratory factor analysis of Mitchell’s
data showed separate constructs for episodic and semantic
memory. These separate constructs for episodic and semantic
memory laid the foundation for a theory of the relation be-
tween age (younger adults vs older adults) and memory type
(episodic vs semantic memory) that McKoon and colleagues
(1986) found to be lacking in Tulving’s (1972, 1985) concept
of differential episodic and semantic memory systems.
Conceptual Underpinnings of the Present
Task Manipulations
As noted earlier, Mitchell (1989) used Tulving’s (1985)
classification scheme for memory: episodic, semantic, and
procedural memory. Two of these memory types—episodic
and semantic—are pertinent to the present study. In the
present study, we used both memory search (one, three, or
five targets and one probe) and hybrid visual search (two
targets and one, three, or five probes, i.e., a two-alternative,
forced-choice task) separately employing words and digits
as stimuli (see Table 1 for a description of the tasks used in
this study). Memory search is used widely as an episodic
memory task. Furthermore, we classified the hybrid visual
search task as being an episodic task because of the autobio-
graphic nature of the task. That is, participants were re-
quired to hold two target items in memory on each trial (and
the targets changed across trials), and these two items held
in memory were compared to a subsequent set of one, three,
or five probe items. Participants were instructed to decide
which of the two target items they had seen earlier as a
probe item (see Fisk & Rogers, 1991).
Episodic tasks are linked in time to autobiographical con-
texts and, thus, are different than semantic tasks that do not
require an autobiographical context (Mitchell, 1989). In the
present study, we used lexical decision and multiplication
verification tasks as semantic memory tasks. Word fre-
quency effects obtained from a lexical decision task are as-
sumed to index semantic memory (lexical) access (e.g.,
Allen, McNeal, & Kvak, 1992; Allen, Smith, Lien, Weber,
& Madden, 1997; Forster, 1976; Monsell, Doyle, & Hag-
gard, 1989; cf. Balota & Chumbley, 1984). In a lexical in-
stance framework, it is assumed that each word is coded
separately in the “mental dictionary” or lexicon (Carr &
Pollatsek, 1985). It is also assumed that words can be pro-
cessed more rapidly and/or more accurately as they become
more familiar (i.e., as word frequency increases). Logogen-
type models (e.g., Allen & Emerson, 1991; Allen, Wallace,
& Weber, 1995; McClelland & Rumelhart, 1981; Morton,
1969) assume that the “firing threshold” of the logogen rep-
resenting a higher frequency word is lower than the firing
threshold of the logogen representing a lower frequency
word.
Table 1. Tasks Used in the Present Study and Examples of Target and Probe Stimuli
Task Name
Target
Probe
Question Participant Is Answering in Responding to the Probe
Hybrid visual search for words
Hybrid visual search for digits
When, then
Then, what, have
Which one of the two target words appeared later as a probe word? (probe set size ꢀ 3)
Which one of the two target digits appeared later as a probe digit? (probe set size ꢀ 3)
3
5
3
7
6
Have
8
Memory search for words
Memory search for digits
Then, what, have
Does the probe word match any of the target words?
Does the probe digit match any of the target digits?
1
7
2
4
9
Lexical decision 1
Lexical decision 2
Multiplication verification 1
Multiplication verification 2
Then
Then
6 ꢁ 5 ꢀ 32
6 ꢁ 5 ꢀ 30
None
None
None
None
Does the target letter string form a real word, or not?
Does the target letter string form a real word, or not?
Is the target answer the correct answer to the problem presented?
Is the target answer the correct answer to the problem presented?

AGE DIFFERENCES IN MEMORY TYPE
P175
The present multiplication verification task (which used
0–9 as operands) is also assumed to be a semantic memory
task. There is evidence that simple multiplication and addi-
tion tasks involve the retrieval of general knowledge infor-
mation from long-term memory (Allen, Ashcraft, & Weber,
1992; Allen, Smith, Jerge, et al., 1997; Ashcraft, 1992;
Geary & Wiley, 1991; but cf. Sliwinski, Buschke, Kuslan-
sky, Senior, & Scarisbrick, 1994). For example, problem
size effects (problems with products of 20 or less were clas-
sified as being of a small problem size, and problems with a
product of greater than 20 were classified as being of a large
problem size) in arithmetic verification tasks suggest that
retrieval from long-term semantic memory occurs (see Ash-
craft, 1992, 1995).
The network model (Ashcraft, 1992) hypothesizes the ex-
istence of associations between problem operands and prob-
lem answers that are stored in long-term semantic memory.
The degree of learning is represented by the strength of the
connections among operands and answers (Allen, Smith,
Jerge, et al., 1997; Ashcraft, 1995). This network model
predicts that fact retrieval speed from long-term memory is
a function of the absolute value of operands, the number of
competing associations between operands and potential an-
swers, and the frequency of exposure (Allen, Smith, Jerge,
et al., 1997; Ashcraft, 1995; Gallistel & Gelman, 1992).
Consequently, the network model predicts the problem size
effect: Retrieval duration increases as problem size in-
creases for addition (Geary & Wiley, 1991), subtraction
(Geary, Frensch, & Wiley, 1993), and multiplication (Allen,
Ashcraft, et al., 1992; Allen, Smith, Jerge, et al., 1997).
Thus, for the small operands (0–9) used in the present multi-
plication verification task, there is evidence that individuals
retrieve answers from semantic memory.
were primarily the result of a central-process decrement on
the part of older adults.
Age Differences in Semantic Memory?
After concluding that there are age differences in epi-
sodic memory (e.g., Craik, 1977; Craik & Jennings, 1992;
Light, 1991) and that these age differences show a pro-
nounced central-process locus (Cerella, 1985), we now turn
to the influence of increased adult age on semantic memory.
Mitchell (1989) did not find significant age differences in
semantic memory performance on four tasks. Other reviews
of the lexical/semantic memory aging literature have come
to the same conclusion (e.g., Allen, Madden, & Slane, 1995;
Amrhein, 1995; Laver & Burke, 1993; Light, 1992).
Namely, although older adults do show evidence of periph-
eral-process decrements in their intercept data (Allen,
Smith, Jerge, et al., 1997), they show no appreciable evi-
dence of central-process age differences in their slope data
for semantic priming (Laver & Burke, 1993), lexical deci-
sion/naming (Allen, Madden, et al., 1995, 12 data sets, 184
data points), picture-word production (Amrhein, 1995; Am-
rhein & Theios, 1993), and multiplication verification
(Allen, Smith, Jerge, et al., 1997). The only evidence of a
slope difference for semantic memory tasks was found for
verbal-pictorial comparison tasks (Amrhein, 1995), and this
was probably the result of age differences in recoding (from
words to pictures or vice versa) rather than the result of age
differences in semantic retrieval. Indeed, there is no evi-
dence of an age decrement in semantic retrieval for lexical
decisions (Allen et al., 1991, 1993), lexical naming (Allen,
Cerella, Madden, Smith, & Lien, 1999; Balota & Ferraro,
1993), or basic multiplication or addition processing (Allen,
Ashcraft, et al., 1992; Geary & Wiley, 1991; Verhaeghen,
Kliegl, & Mayr, 1997). Consequently, the literature on ag-
ing and semantic memory indicates that there are periph-
eral-process age differences, but no appreciable age differ-
ences in central processes. These results are the opposite of
those found by Cerella (1985) for nonsemantic memory tasks.
Age Differences in Episodic Memory: Meta-Analyses
and Processing Stage
Most research on episodic memory and aging has indi-
cated that older adults show a decrement relative to younger
adults (e.g., Allen, 1990, 1991; Allen, Kaufman, Smith, &
Propper, 1998a, 1998b; Coyne, Allen, & Wickens, 1986;
Craik, 1977; Craik & Jennings, 1992; Kausler, 1994; Light,
1991, 1992). Light (1991) noted that there presently is no
single acceptable explanation of age differences in memory.
However, meta-analyses examining age differences in timed
tasks have provided insight into how increased adult age af-
fects stages of information processing. In particular, a meta-
analysis conducted by Cerella (1985) has been quite influ-
ential. Cerella used meta-analysis techniques to analyze the
data (189 data points) from 35 tasks that required a speeded
response (nonsemantic memory tasks that included prima-
rily episodic memory tasks). Cerella, using Brinley plots,
found large age differences in slopes and much smaller age
differences in intercepts. Cerella hypothesized that the Brin-
ley plot slopes indexed central processes but that the Brin-
ley plot intercepts indexed peripheral processes. A Brinley
plot examines the RTs of older adults (plotted on the y-axis)
relative to younger adults (plotted on the x-axis) across pro-
cessing complexity within a task(s) (Bashore & Smulders,
1995; Brinley, 1965). The key result in Cerella’s Brinley
plot study was that age differences in nonsemantic memory
Differential Age-Related Effects in Episodic and
Semantic Memory
Lima, Hale, and Myerson (1991) suggested that differen-
tial age-related effects exist across “processing domains.”
Lima and colleagues (1991) reported three Brinley plot
meta-analyses. Analysis 1 involved 10 lexical decision stud-
ies, Analysis 2 involved nine lexical tasks that were not lex-
ical decision tasks (e.g., naming), and Analysis 3 involved
seven nonlexical tasks (e.g., visual search and memory
search). The two “lexical domain” analyses (semantic mem-
ory tasks) revealed slopes of approximately 1.5, but the
nonlexical domain analysis (primarily episodic memory
tasks) revealed a slope of approximately 2.0. Given that
slopes are assumed to measure central processing speed
(Cerella, 1985; Allen, Smith, Jerge, et al., 1997), Lima and
colleagues suggested that older adults showed smaller cen-
tral-process decrements for lexical domain tasks than for
nonlexical domain tasks.
It may appear that the Brinley slope of 1.5 found by Lima
and colleagues (1991) for lexical domain tasks is inconsis-

P176
ALLEN ET AL.
tent with our earlier claim that most meta-analyses of se-
mantic memory tasks showed no age differences in slopes
(a Brinley plot slope of approximately 1.0; e.g., Allen, Mad-
den, et al., 1995; Amrhein, 1995; Laver & Burke, 1993).
These results indicate that older adults’ central-process per-
formances were slowing at the same rate as were younger
adults’ central-process performances. However, as Amrhein
noted, Lima and colleagues mixed production and nonpro-
duction lexical domain tasks. When production (e.g., nam-
ing) and nonproduction (e.g., categorization) tasks are sepa-
rated, they show different slowing functions. The slope for
the production tasks was .93, but the slope for the nonpro-
duction tasks was 1.47 (Amrhein, 1995). Thus, the categori-
zation-type lexical memory tasks show evidence of age dif-
ferences in central processing speed only. Amrhein’s finding
of differential Brinley slopes across different lexical domain
tasks suggests that process-specific age differences continue
to persist even within the lexical domain.
Allen and colleagues (1991, 1993) and Balota and Fer-
raro (1993, 1996) also found evidence of process-specific
age differences in the lexical domain. They found evidence
of peripheral-process age differences but no evidence of
central-process age differences for both lexical decision and
naming tasks. These data indicate that differential age-related
effects across processing stage for lexical/semantic memory
indicate the presence of process-specific slowing within the
same processing domain.
we refer to process-specific or domain-specific latent fac-
tors (e.g., central semantic memory factors). Type B models
have been referred to as hierarchical, common factor mod-
els (Salthouse & Czaja, 2000). This type of model contains
exclusively shared (generalized) age effects. That is, age is
assumed to have a common cause on component first-order
latent factors such as episodic and semantic memory, and
there are no nonshared, or independent, age effects on first-
order latent factors. Thus, in Type B models, all age effects
are carried through a single common factor. Finally, Type C
models are referred to as hierarchical, common factor,
mixed models (Allen et al., 2001). Type C models contain
both shared and nonshared age effects. We modeled this
by including both indirect paths (shared age effects) and
direct paths (nonshared age effects) in the structural equa-
tion model.
Predictions
For factor analysis using the eight task-difficulty-based
slopes (central processes) and eight task-difficulty-based in-
tercepts (peripheral processes) as indicators, we predicted
that there would be two separate latent factors for central
processes across memory type (semantic and episodic mem-
ory) but that just a single latent factor for peripheral pro-
cesses would be observed.
For structural equation modeling, we predicted that age
would have significant direct paths to both the central, epi-
sodic memory factor and to the episodic/semantic periph-
eral factor, but not to the central, semantic factor (because
there are no appreciable age differences in semantic central
processes, Allen et al., 1991, 1993). A particular prediction
of the present factor-specific model was that a Type A
model would fit better than a Type B model (see Figures 1–3).
This is because we predicted that there would be nonshared
age effects for the two of the three predicted first-order la-
tent factors—peripheral semantic/episodic and central epi-
sodic—but not for central semantic (because this factor was
predicted to show no age effects). Our factor-specific theo-
retical model was also consistent with a Type C model.
Structural Equation Modeling
We used structural equation modeling in the present
study to test three theories of aging. Structural equation
modeling was used because it provides a formal quantitative
method of testing theoretical models. The three model types
are illustrated in Figures 1–3 as Type A, Type B, and Type
C. Type A models have been referred to as independent fac-
tors models (Salthouse & Czaja, 2000). This type of model
contains exclusively nonshared (process-stage-specific) age
effects. That is, age is assumed to have independent effects
on different latent first-order factors. By first-order factors,
Figure 1. Type A, an independent factors model of cognitive aging in which all age effects on first-order latent factors are nonshared. Indica-
tor variables: V2 ꢀ hybrid visual search for digits; V5 ꢀ hybrid visual search for words; V6 ꢀ lexical decision 1; V13 ꢀ WAIS-R Vocabulary
scores; V16 ꢀ lexical decision 2; V7 ꢀ memory search for digits; V8 ꢀ hybrid visual search for digits; V9 ꢀ multiplication verification 1; V10 ꢀ
memory search for words; V17 ꢀ lexical decision 2.

AGE DIFFERENCES IN MEMORY TYPE
P177
Figure 2. Type B, a hierarchical, common factor model of cognitive aging in which all age effects on first-order latent factors are shared. Indi-
cator variables: V2 ꢀ hybrid visual search for digits; V5 ꢀ hybrid visual search for words; V6 ꢀ lexical decision 1; V13 ꢀ WAIS-R Vocabulary
scores; V16 ꢀ lexical decision 2; V7 ꢀ memory search for digits; V8 ꢀ hybrid visual search for digits; V9 ꢀ multiplication verification 1; V10 ꢀ
memory search for words; V17 ꢀ lexical decision 2.
independent-living individuals from the Cleveland, OH,
community. Project 60 is an Ohio program in which resi-
dents aged older than 60 can audit college courses for free.
The older adults were paid $20.00 each for participating.
All participants were screened for visual acuity of at least
20/40. Participants completed the Vocabulary and Digit
Symbol Substitution subscales of the Wechsler Adult Intel-
ligence Scale-Revised (WAIS-R; Wechsler, 1981).
METHOD
Participants
We tested the same 40 younger adults (mean age ꢀ 23.4
years, range ꢀ 18–44 years) and the same 40 older adults
(mean age ꢀ 71.7 years, range ꢀ 63–90 years) on all six
tasks. Each individual participated in two 90–120-min ses-
sions. Younger adults were Cleveland State University
(CSU) undergraduates who participated for course credit.
Older adults were participants in CSU’s Project 60 or were
There were no age differences in mean years of education
(younger ꢀ 14.5 years, older ꢀ 14.83 years; p ꢂ .50). How-
Figure 3. Type C, a hierarchical, common factor, mixed model of cognitive aging in which age effects on first-order latent factors are both
shared and nonshared; that is, there are both direct and indirect paths from age to the first-order latent factors. Indicator variables: V2 ꢀ hybrid
visual search for digits; V5 ꢀ hybrid visual search for words; V6 ꢀ lexical decision 1; V13 ꢀ WAIS-R Vocabulary scores; V16 ꢀ lexical decision 2;
V7 ꢀ memory search for digits; V8 ꢀ hybrid visual search for digits; V9 ꢀ multiplication verification 1; V10 ꢀ memory search for words; V17 ꢀ
lexical decision 2.

P178
ALLEN ET AL.
ever, older adults (59.95) scored higher than younger adults
(51.8) on the Vocabulary subscale, t(78) ꢀ ꢃ4.38, p ꢄ
.001, and younger adults (61.15) scored higher than older
adults (45.53) on the Digit Symbol Substitution subscale,
t(38) ꢀ 4.98, p ꢄ .001.
search and hybrid visual search were presented first equally
often.
There were 6 practice trials and 40 experimental trials (20
yes or left trials and 20 no or right trials) for each set size
(i.e., one, three, and five items) for all four episodic memory
tasks: memory search and hybrid visual search with digits
as stimuli as well as memory search and hybrid visual
search with words as stimuli. For the multiplication verifi-
cation task, there were 100 basic multiplication problems,
although only the data from problems with multiplicands
between 2 and 9 were analyzed (i.e., 64 problems) because
participants use rules rather than retrieval from semantic
long-term memory to process problems with multiplicands
of 0 or 1 (Allen, Ashcraft, et al., 1992; Allen, Smith, Jerge,
et al., 1997). Half of these 64 problems were true (e.g., 3 ꢁ
3 ꢀ 9), and half were false (e.g., 3 ꢁ 3 ꢀ 11). Also, prob-
lems with products of 20 or less were classified as being of
a small problem size, and problems with a product of
greater than 20 were classified as being of a large problem
size (see Allen, Ashcraft, et al., 1992; Allen, Smith, Jerge, et
al., 1997). This resulted in 25 small problems (13 true and
12 false) and 39 large problems (19 true and 20 false). Fi-
nally, for the lexical decision task, there were 432 trials
(216 word and 216 nonword). These 432 stimuli (nonwords
were formed by misspelling words from the same frequency
category; see Allen, Wallace, et al., 1995) were split into
four different word frequency categories. Because we wanted
to examine slopes and intercepts, and we derived slopes and
intercepts for the lexical decision tasks (Sessions 1 and 2)
using the four word frequency categories to predict RT, we
reported just the analyses for the 216 word trials.
Apparatus
We tested all participants individually using a 286 micro-
computer with a video graphics adaptor (VGA) card. We
used Micro Experimental Laboratory (MEL) software
(Schneider, 1988) to present the stimuli and collect RT and
accuracy data. For all tasks, participants responded by using
the index and middle fingers of the right hand to tap the left
and right arrow keys located in the lower right corner of the
computer keyboard. Participants pressed a true or false but-
ton for multiplication verification. For the lexical decision
task, a word or nonword button was used. For hybrid visual
search using digits or words as stimuli, a left or right button
was used (i.e., a two-alternative, forced-choice task). For
memory search using digits or words as stimuli, a yes or no
button was used. For all six tasks, the two response buttons
always corresponded to the left and right arrow keys located
in the lower right corner of the keyboard. Yes or no was
counterbalanced with the left or right buttons across partici-
pants. Each letter or digit in the display subtended a visual
angle of approximately 0.28 degrees horizontally and 0.56
degrees vertically.
Materials and Procedure
For the semantic tasks, we used a lexical decision task on
two sessions and a multiplication verification task on two
sessions. For the episodic tasks, we used hybrid visual
search and memory search tasks involving both digits (the
stimuli used for the multiplication verification task) and
words (the stimuli used for the lexical decision task). For
the memory search and hybrid visual search tasks using
words, if a word was used as a target on one trial it was not
used as a distractor on any subsequent trials (on 48 hybrid
visual search trials, a word was used as a distractor on two
trials; all 48 of the trials occurred for set sizes of three and
five words). The words were selected from the corpus pre-
sented in Allen, Wallace, and Weber (1995), based on Ku-
cera and Francis (1967). A total of 432 words were used
with equal numbers of four-, five-, and six-letter words as
well as equal numbers of words from four different word
frequency categories (108 words in each): very high fre-
quency (range ꢀ 240–1,016 occurrences in the Kucera &
Francis, 1967, corpus), medium high frequency (range ꢀ
151–235 occurrences), low frequency (range ꢀ 40–54 oc-
currences), and very low frequency (range ꢀ 1–5 occur-
rences). Target words and probe words were of the same
length and word frequency category.
All trials began with a 500-ms fixation presentation—a ꢅ
in the center of the display screen. For hybrid visual search
trials, the two target digits or words were presented until the
participants pressed the space bar to continue the trial (self-
paced presentation for up to a maximum of 32 s, i.e., partic-
ipants were allowed to look at the targets for up to 32 s).
Next, a 500-ms mask consisting of uppercase Xs covered
the display location previously used by the target stimuli.
After the offset of the mask, the probe set of one, three, or
five words or digits appeared in a circular display extending
around the central fixation point location (the probe stimu-
lus that matched the target stimulus could occur at any of
five different positions around the imaginary circle).
For the memory search trials, the fixation presentation
was followed by the presentation of the target (or memory)
set consisting of one, three, or five words or digits. The tar-
get set presentation was for 200 ms per target-set item (i.e.,
either 200 ms, 600 ms, or 1,000 ms) in one, three, or five
lines extending above and below the central fixation point.
After the target set was offset from the computer screen, a
500-ms mask consisting of one, three, or five mask stimuli
of six uppercase Xs covered the display location previously
used by the target set. After the offset of the mask stimuli, a
single probe word or digit was presented in the center of the
screen (i.e., in the location of the original fixation point). On
half of the trials, this probe stimulus matched one of the tar-
get items (i.e., a yes trial), and on the remaining half of the
trials this probe stimulus did not match any of the target
items (i.e., a no trial).
For half of the participants, memory search with word
stimuli, lexical decisions, memory search with digits, and
multiplication verification were presented on the first ses-
sion, and for the other half of the participants, hybrid visual
search with words, lexical decisions, hybrid visual search
with digits, and multiplication verification were presented
first. A counterbalancing design was used so that memory

AGE DIFFERENCES IN MEMORY TYPE
P179
Table 2. Intercept and Slope Reliabilities as a Function of Task
Table 4. Results of Exploratory Factor (Promax Rotation) on
Variables From Table 3, Age Not Partialled
Reliability
Mult Ver
Lex Dec
VS-D
MS-D
VS-W
MS-W
Episodic
Central
Processes
Semantic
Central
Processes
Intercept
Slope
.90
.62
.83
.67
.92
.77
.92
.80
.94
.86
.96
.78
Peripheral
Processes
Variable
Slopes
Note: Mult Ver ꢀ multiplication verification; Lex Dec ꢀ lexical decision;
VS-D ꢀ hybrid visual search for digits; MS-D ꢀ memory search for digits;
VS-W ꢀ hybrid visual search for words; MS-W ꢀ memory search for words.
Memory search for digits
.090
.257
ꢃ.217
ꢃ.044
.078
.145
.036
.229
.229
.411
.702a
.379
.823a
.205
.188
.098
ꢃ.305
.098
ꢃ.369
.208
ꢃ.128
.194
.738a
.624a
.243
Hybrid visual search for digits
Memory search for words
Hybrid visual search for words
Lexical decision 1
For both the lexical decision and multiplication verifica-
tion tasks, each trial began with a 500-ms fixation point pre-
sentation (again, a ꢅ) in the center of the screen. Next, a
word or nonword (for the lexical decision task) or a true or
false math problem (for the multiplication verification task)
was presented in the center of the screen until the partici-
pant responded.
Lexical decision 2
Multiplication verification 1
Multiplication verification 2
WAIS-R Vocabulary scores
.060
ꢃ.637a
Intercepts
Memory search for digits
Hybrid visual search for digits
Memory search for words
Hybrid visual search for words
Lexical decision 1
Lexical decision 2
Multiplication verification 1
Multiplication verification 2
.749a
.757a
.885a
.876a
.866a
.853a
.837a
.649a
ꢃ.015
.056
.168
.139
.037
.105
ꢃ.225
.114
ꢃ.068
ꢃ.135
ꢃ.091
ꢃ.080
.168
RESULTS
Task Reliability
ꢃ.034
ꢃ.126
.404
The intercept and slope reliabilities (using the Spearman-
Brown method) for these eight data sets are presented in
Table 2. Reliabilities ranged from .62 to .96. We derived
reliability indices for intercepts and slopes for the eight ex-
perimental data sets using the Spearman-Brown method. Re-
liabilities for the multiplication verification and lexical deci-
sion tasks were computed across sessions. We computed
split-half reliabilities for the other four tasks using odd ver-
sus even trials across set size.
aRefers to retained indicators. Values are standardized regression coefficients.
partialled and age-partialled correlation matrices. However,
we also extended Mitchell’s work by developing and testing
structural equation models using age as the manifest, causal
variable. We developed these models using the correlation
matrix including all eight slopes and eight intercepts as well
as age (entered as a continuous variable) and WAIS-R Vo-
cabulary score (see Table 3). We first conducted an explor-
atory factor analysis (using a promax rotation) in which the
effect of age was not partialled from the remaining factors.
The rotated factor pattern of this analysis (in standardized
Exploratory Factor Analysis
Following Mitchell (1989), who reported an exploratory
factor analysis using data similar to the present slopes, we
reported exploratory factor analyses using both non-age-
Table 3. Correlation Matrix for All 16 Experimental Variables, Vocabulary, and Age Correlation Matrix
V1
V2
V3
V4
V5
V6
V7
V8
V9
V10
V11
V12
V13
V14
V15
V16
V17
V18
V1
V2
V3
V4
V5
(48)
.273 (72)
.219 .100
.439 .094
.281 .653
(4.6)
.445
.318
(49)
.240
(81)
V6 ꢃ.111 .325
V7 ꢃ.188 .385
.068 ꢃ.109
.047 ꢃ.015
.119 ꢃ.076
.143
(28)
.168
.344
.236
.150
.052
.192 (248)
.189 .626
.217 .621
.243 .654
.253 .569
V8
V9
.055 .349
(320)
.671
.669
.677
.609
.113
.008
.572
.285
.656
.531
.281 .505 ꢃ.080 ꢃ.144
.072 .431 ꢃ.084 ꢃ.467
.100 .461 ꢃ.056 ꢃ.217
.344 .328
.221 .094 ꢃ.078
(369)
V10
V11
V12
V13
V14
V15
.798 (335)
.726 .717
.713 .660
.212 .175
ꢃ.036 .038
.822 .626
.252 .162
.715 .663
.624 .553
(431)
.610
.058
.149
.548
.209
.756
.568
.144
.063
.106
.145 ꢃ.108 .586
.086 ꢃ.430 .141
(178)
.249
.159
.683
.096
.708
.562
(9.2)
.008
.289 ꢃ.085 (307)
ꢃ.270
.032
.501
.087 .008
.402 .566
.452
.087
.095 ꢃ.075 ꢃ.009 .022
.128
(2.8)
.375
.257
.186
.104 .652
.502 .303
.274 .537
V16 ꢃ.144 .260
.230 ꢃ.025
.115 ꢃ.011
.028 .167
.147 .544
.132 .566
(22)
.216 (227)
.023 .525
V17
V18
.238 .362
.256 .479
.030
.061
.305 ꢃ.014 .418
(25)
Notes: N ꢀ 80; Slopes: V1 ꢀ memory search for digits; V2 ꢀ hybrid visual search for digits; V3 ꢀ multiplication verification 1; V4 ꢀ memory search for words;
V5 ꢀ hybrid visual search for words; V6 ꢀ lexical decision 1; V14 ꢀ multiplication verification 2; V16 ꢀ lexical decision 2. Intercepts: V7 ꢀ memory search for dig-
its; V8 ꢀ hybrid visual search for digits; V9 ꢀ multiplication verification 1; V10 ꢀ memory search for words; V11 ꢀ hybrid visual search for words; V12 ꢀ lexical
decision 1; V15 ꢀ multiplication verification 2; V17 ꢀ lexical decision 2. Other variables: V13 ꢀ WAIS-R Vocabulary scores; V18 ꢀ age. Standard deviations are on
the negative diagonal.

P180
ALLEN ET AL.
Table 5. Results of Exploratory Factor Analysis (Promax Rotation)
on Variables From Table 4, Age Partialled
served as indicators for a single latent factor. Alternatively,
for slopes we used separate latent factors for semantic and
episodic tasks. Our Model 1 included 11 variables (10 indi-
cator variables and the manifest variable of age, entered as a
continuous variable). In this conceptualization, age was the
cause of performance differences on three different latent
factors. Model 1 used five indicator variables to measure the
latent factor for Peripheral Processes (memory search with
words and with digits, hybrid visual search with words,
multiplication verification 1, and lexical decision 2), two in-
dicator variables to measure the latent factor for Episodic
Central Processes (visual search with words and with dig-
its), and three indicator variables to measure the latent fac-
tor for Semantic Central Processes (the two sessions of the
lexical decision task, and WAIS-R Vocabulary). Model 2
was identical to Model 1 except that we included a multipli-
cation verification slope indicator for the Semantic Central
Processes factor. We tested Model 2 even though it was not
justified by exploratory factor analysis because our a priori
theory predicted that both lexical decision and multiplica-
tion verification tasks were indices of semantic memory.
Note that both Models 1 and 2 were Type A models that in-
volved just nonshared age effects. That is, there were just
direct paths from age to the first-order latent factors. Models
1 and 2 are illustrated in Figure 4.
Multiplication
Semantic Episodic Verification
Peripheral Central
Central
Central
Variable
Slopes
Processes Processes Processes
Processes
Memory search for
digits
.103
.076
.384
.426
.241
.608a
.641a
.575
.127
ꢃ.171
.451
Hybrid visual search for
digits
Memory search for
words
ꢃ.187
Hybrid visual search for
words
ꢃ.137
.087
.163
ꢃ.144
.778a
.689a
.777a
.080
.140
ꢃ.034
.043
.213
Lexical decision 1
Lexical decision 2
Multiplication
verification 1
Multiplication
verification 2
WAIS-R Vocabulary
scores
.051
.112
.086
.241
.130
.294
ꢃ.275
.033
.806a
.816a
ꢃ.753a
ꢃ.119
Intercepts
Memory search for
digits
Hybrid visual search for
digits
Memory search for
words
.672a
.660a
.770a
.176
.229
ꢃ.025
.012
.021
.086
Models 3 and 4 used the same indicator variables as
Models 1 and 2, except that Models 3 and 4 were Type B
models (see Figures 1–3). That is, first-order latent factors
were linked to age through a common factor. Thus, all age
effects in these models were shared. The only difference be-
tween Models 3 and 4 was that Model 4 added a fourth indi-
cator variable (multiplication verification 2 slopes, as used
in Model 2) for the first-order semantic slopes factor. Mod-
els 3 and 4 are illustrated in Figure 5.
ꢃ.112
ꢃ.184
ꢃ.209
Hybrid visual search for
words
.732a
.846a
.854a
.226
ꢃ.289
ꢃ.105
ꢃ.233
.058
ꢃ.020
.062
.223
.228
Lexical decision 1
Lexical decision 2
Multiplication
verification 1
Multiplication
verification 2
.801a
.638a
.020
.154
.516
ꢃ.212
ꢃ.177
ꢃ.116
Models 5 and 6 were analogous to Models 3 and 4, except
that Models 5 and 6 included all significant direct paths
from age to first-order factors or to indicator variables.
Models 5 and 6 included a common factor but also included
direct paths from age to the first-order semantic slope factor
and the semantic and episodic intercept factor. A direct path
from age to the Vocabulary indicator variable was also in-
cluded in Models 5 and 6 (see Type C, Figure 3). Models 5
and 6 are illustrated in Figure 6.
We describe the fits of Models 1 through 6 at the bottom
of Table 6 using criteria suggested by Hatcher (1994). In
particular, a good fit is characterized by a ratio of chi-square
and degrees of freedom of less than 2, CFI (comparative fit
index) of .90 or greater, and NNFI (nonnormed fit index) of
.90 or greater. Models 1, 2, 5, and 6 met these criteria, but
Models 3 and 4 did not. The key aspect that differentiated
the better fitting models from the poorer fitting models was
that the better fitting models all included direct paths from
age to first-order factors (Types A and C), but the poorer fit-
ting models contained just indirect paths from age to the
first-order factors (Type B; refer to Figures 1–3). Models 5
and 6 were the best fitting models. These hierarchical, com-
mon factor, mixed Type C models contained both direct
(nonshared) and indirect (shared) age effects. However, as
Figure 6 illustrates, the present Models 5 and 6 contained
standardized path loadings greater than 1 and the direct path
aRefers to retained indicators. Values are standardized regression coefficients.
regression coefficients) is reported in Table 4. We retained
three separate factors. The scree plot also suggested the
presence of three separate factors. This three-factor solution
exhibited simple structure. The three factors were Periph-
eral Processes (episodic and semantic intercepts), Episodic
Central Processes (episodic slopes), and Semantic Central
Processes (semantic slopes).
We conducted a separate exploratory factor analysis us-
ing age-partialled variables in an attempt to confirm that our
results were similar for both age groups. This analysis showed
similar results to the first analysis except that the order of
the second and third factors reversed relative to the non-
age-partialled analysis, and a fourth latent factor emerged: a
Multiplication Verification Central Processes (i.e., slope-
based) factor (see Table 5). The age-partialled results did
provide evidence, then, that both age groups showed similar
patterns of latent factors.
Structural Equation Modeling
From the results shown in Table 4, the beginning mea-
surement model included the 17 aforementioned variables.
Using the working hypothesis developed in the introductory
section, we developed a model in which all intercepts

AGE DIFFERENCES IN MEMORY TYPE
P181
Figure 4. Model 1, an independent factors model (all direct age effects). Indicator variables: V2 ꢀ hybrid visual search for digits; V5 ꢀ hybrid
visual search for words; V6 ꢀ lexical decision 1; V13 ꢀ WAIS-R Vocabulary scores; V16 ꢀ lexical decision 2; V7 ꢀ memory search for digits; V8
ꢀ hybrid visual search for digits; V9 ꢀ multiplication verification 1; V10 ꢀ memory search for words; V17 ꢀ lexical decision 2.
from age to the first-order Semantic Central Processes fac-
tor had a negative path loading. Although standardized path
loadings greater than 1 are typically interpreted as evidence
for a mis-specified model, this outcome in aging research
can actually provide evidence for correlated first-order fac-
tors that result in model overfitting (Hall & Allen, 2000). In
the present Models 5 and 6, even though we could explain
the seemingly aberrant path loadings by alluding to cor-
related indicator variables (Pedhazur, 1982) resulting in
model overfitting, we cannot so readily explain why there
was a significant negative path from age to the first-order
central episodic factor. This loading suggested that episodic
memory slopes became less steep with increasing adult
age—a result that was clearly at odds with the observed
positive path loadings from age to the Episodic Central Pro-
cesses factor in Models 1 and 2 (see Figure 4) as well as the
correlation matrix (see Table 3; there was a positive correla-
tion between age and episodic slopes). These results suggest
that although Models 5 and 6 showed quite good fits, they
were most likely mis-specified as the result of model over-
fitting. Consequently, the present structural equation Mod-
els 1 and 2 appeared to be the best fitting proper models for
the present data (see Table 6). Models 1 and 2 suggest that
all age effects were nonshared in these data. Models 5 and 6
did fit the best, but these models resulted in unacceptable
standardized path loadings that appeared to be the result of a
mis-specified model because of model overfitting. Finally,
Models 3 and 4 did not fit acceptably well. Thus, the present
data could not be accounted for with a model in which all
age effects were shared.
The addition of the multiplication verification indicator
variable on the Semantic Central Processes factor (Models
Figure 5. Model 3, a hierarchical, common factor model (all indirect age effects). Indicator variables: V2 ꢀ hybrid visual search for digits; V5 ꢀ
hybrid visual search for words; V6 ꢀ lexical decision 1; V13 ꢀ WAIS-R Vocabulary scores; V16 ꢀ lexical decision 2; V7 ꢀ memory search for
digits; V8 ꢀ hybrid visual search for digits; V9 ꢀ multiplication verification 1; V10 ꢀ memory search for words; V17 ꢀ lexical decision 2.

P182
ALLEN ET AL.
Figure 6. Model 5, a hierarchical, common factor, mixed model with both direct (nonshared) and indirect (shared) age effects. Indicator vari-
ables: V2 ꢀ hybrid visual search for digits; V5 ꢀ hybrid visual search for words; V6 ꢀ lexical decision 1; V13 ꢀ WAIS-R Vocabulary scores; V16 ꢀ
lexical decision 2; V7 ꢀ memory search for digits; V8 ꢀ hybrid visual search for digits; V9 ꢀ multiplication verification 1; V10 ꢀ memory search
for words; V17 ꢀ lexical decision 2.
2, 4, and 6) did improve the fit somewhat (compared with
Models 1, 3, and 5). We believe that this was an important
conceptual addition to the model because it resulted in con-
verging operations of indicator variables. This issue is cov-
ered in more detail in the Discussion.
sets. There were also two semantic memory tasks—lexical
decision and multiplication verification—using words and
digits, respectively. There were two sessions each for the
semantic memory tasks for a total of four semantic memory
data sets (five tasks when WAIS-R Vocabulary scores were
included). We used the correlation matrix resulting from
the 80 within-subject data sets to examine age differences
across memory type and processing stage using factor anal-
ysis and structural equation modeling. Our goal in the present
investigation was to test different versions of generalized
and process-specific models of cognitive aging (see Figures
DISCUSSION
In the present study we tested 40 younger adults and 40
older adults on the same four tasks. There were two episodic
tasks—hybrid visual search and memory search—using both
words and digits as stimuli, for a total of four episodic data
Table 6. Loadings of Variables on Constructs and Fit Statistics for Structural Equation Models
Construct and Variable
Model 1
Model 2
Model 3
Model 4
Model 5
Model 6
Peripheral Processes (intercepts)
Memory search for digits
Hybrid visual search for digits
Memory search for words
Multiplication verification 1
Lexical decision 2
Semantic Central Processes
Lexical decision 1
Lexical decision 2
WAIS-R Vocabulary scores
Multiplication verification 1
Episodic Central Processes
Hybrid visual search for digits
Hybrid visual search for words
Fit Statistics
.72
.78
.87
.90
.79
.72
.78
.87
.90
.79
.72
.78
.87
.90
.78
.72
.79
.87
.90
.79
.72
.78
.87
.90
.79
.72
.78
.87
.90
.79
.91
.59
ꢃ.47
.91
.59
ꢃ.47
.01
1.00
.46
ꢃ.40
1.00
.45
ꢃ.40
.008
.87
.62
ꢃ.48
.87
.62
ꢃ.48
.009
1.00
.65
1.00
.65
1.00
.65
1.00
.65
1.00
.65
1.00
.65
Chi square
N
74
80
85
80
93
80
103
80
49
80
59
80
df
41
52
40
51
39
49
CFI
NNFI
RMSEA
.93
.90
.10
.93
.91
.09
.88
.84
.13
.88
.85
.11
.98
.97
.06
.98
.97
.05
Note: CFI ꢀ comparative fit index; NNFI ꢀ nonnormed fit index; RMSEA ꢀ root mean square error of approximation.

AGE DIFFERENCES IN MEMORY TYPE
P183
1–3) by replicating and extending the Mitchell (1989) study
on age differences in semantic and episodic memory tasks.
Using exploratory factor analysis, we tested the hypothe-
sis that there would be a single peripheral factor for both ep-
isodic and semantic memory tasks but that there would be
separate latent factors for episodic memory task slopes and
semantic memory task slopes. Also, we were especially in-
terested in examining age differences in episodic and se-
mantic memory across processing stage using causal (struc-
tural equation) modeling, because Mitchell (1989) did not
present such models and such models are critical in deter-
mining whether a general factor/hierarchical, common fac-
tor model (Salthouse & Czaja, 2000; Salthouse, Hambrick,
& McGuthry, 1998) or a factor specific/hierarchical, com-
mon factor, mixed model (Allen et al., 2000) is optimal.
ables (eight slopes and eight intercepts) that include RT.
WAIS-R Vocabulary scores are also used as a measured
variable for semantic memory because this factor is a tradi-
tional index of semantic memory. Surprisingly, WAIS-R
Digit Symbol Substitution task scores do not load on a sin-
gle latent factor in either the non-age-partialled or age-par-
tialled exploratory factor analyses (so they are not included
in Table 3).
In the non-age-partialled analysis (see Table 4), we find
evidence for three separate latent factors: a Peripheral Pro-
cesses factor (with intercept indicators from both semantic
and episodic tasks), a Semantic Central Processes factor
(lexical decision slopes and WAIS-R Vocabulary scores),
and an Episodic Central Processes factor (slopes from hy-
brid visual search using digits and words). These results are
consistent with the a priori hypotheses concerning memory
type and processing stage except for the finding that multi-
plication verification slopes do not load on the Semantic
Central Processes factor. A similar factor pattern emerges
when the effect of age is partialled from the variables prior
to factor analysis—except for one difference. Namely, a
fourth factor emerges when age is partialled: a Multiplica-
tion Verification Central Processes factor (the other three
factors remain; see Table 5). Both rotated factor patterns ex-
hibit simple structure.
These results suggest that either three (non-age-par-
tialled) or four (age-partialled) underlying constructs ex-
plain the present data. It is clear that both semantic task and
episodic task intercepts load on the same factor, but that se-
mantic and episodic slopes load on different factors. This
pattern of results is consistent with our a priori hypothesis
that peripheral processing is shared across memory types
but that central processes differ across memory type. In-
deed, the present results indicate that semantic memory is
probably not a unitary construct—the automatic retrieval of
mathematical facts is qualitatively different from the auto-
matic retrieval of words from the mental lexicon. Conse-
quently, we do find evidence of separate latent factors
across memory type and processing stage. Again, these re-
sults replicate and extend Mitchell’s (1989) earlier study.
Task Reliability of Slopes and Intercepts
We computed intercepts (peripheral processes) and slopes
(central processes) using task difficulty to predict response
latency for all 80 participants separately as a function of age
and memory type (episodic vs semantic memory). Other re-
searchers have cautioned that slopes and intercepts cannot
be readily compared because of uninterpretable intercepts
and spuriously high correlations between intercepts and
slopes (e.g., Cerella, 1991). In particular, this can happen
when the distance between 0 and the minimum value of X is
very large (as is frequently the case when regressing older
adults’ RTs on younger adults’ RTs in Brinley plot analy-
ses). However, such is not the case in the present data set,
because the x-axis reflects the number of specific pro-
cessing steps, and the intercept provides a plausible estimate
of the speed of executing “control” (peripheral) processes.
The observed correlations between slopes and intercepts
for the same task in the present study are quite modest
(for the eight experimental tasks involving RT, ꢃ.188,
.349, ꢃ.080, ꢃ.467, .052, ꢃ.108, ꢃ.085, and .216; see Ta-
ble 2). The ꢃ.467 correlation is not an issue for the present
structural equation modeling, because this slope factor is
not retained. This results in the use of only one slope-inter-
cept pair with a statistically significant correlation (and this
is quite modest—r ꢀ .349), so we do not deem it necessary
to center the present data as was done by Sliwinski and Hall
(1998) when analyzing Brinley plots. Consequently, al-
though Cerella’s (1991) concern of spurious correlations
between slopes and intercepts on a given task can poten-
tially result in a confound, there is no evidence that it re-
sulted in a confound in this study.
Are There Shared or Nonshared Age Differences Across
Memory Type and Processing Stage?
We have two reasons for developing structural equation
models of the present intercept and slope data. First, struc-
tural equation modeling provides us with the opportunity to
test whether a hierarchical, common factor model (see Fig-
ure 2, Type B) for age differences in episodic memory and
semantic memory across processing stage (e.g., Salthouse &
Czaja, 2000; Salthouse et al., 1998) or whether a multiple
processing speed, latent factors (independent factors) model
is needed to account for age differences in memory type and
processing stage (i.e., the presently hypothesized model).
The hierarchical, common factor model predicts shared (or
generalized) age effects, that is, that age will show “com-
mon” effects on first-order latent factors (i.e., Semantic
Central Processes, Episodic Central Processes, and seman-
tic and episodic Peripheral Processes). Alternatively, the in-
dependent factors model predicts nonshared age effects,
That is, that age will show age effects on first-order latent
Are There Different Latent Constructs of Memory Type
and Processing Stage?
If there are both semantic and episodic memory stores
that have separate peripheral and central processing stages,
then memory type and processing stage should be function-
ally differentiated when data are analyzed with exploratory
factor analysis. In factor analysis terminology, there should
be separate latent factors for semantic and episodic memory
and for slopes and intercepts.
The present data set of eight RT variables (four semantic
and four episodic variables) results in 16 measured vari-

P184
ALLEN ET AL.
factors that are independent of a common factor. Thus,
structural equation modeling allows us to determine how
age is related to the three latent factors that emerge in the
exploratory factor analysis.
Why Are There Differential Age Differences Across
Memory Type?
There are several potential explanations for why there are
differential age differences across memory type. First, se-
mantic memory retrieval (e.g., retrieving the code for ice
cream) tends to be more practiced than episodic memory re-
trieval (e.g., “What did I have for breakfast today?”). Thus,
it could be that age differences are not are prevalent when a
task is highly overlearned. However, even when an episodic
task is overlearned, there still appear to be age differences.
For example, even after considerable practice, older adults
continue to show age differences in performance (e.g., Fisk
& Rogers, 1991). Another possibility is that semantic and
episodic memory are mediated by different neural systems
and that brain changes with increased adult age affect epi-
sodic memory more than semantic memory (Langley &
Madden, 2000; Nyberg & Cabeza, 2000). Finally, differen-
tial age effects in semantic and episodic memory for central
processing may result because of the different nature of
coding and retrieval in semantic and episodic memory. It
may be that the retrieval of context-dependent (episodic) in-
formation involves more processing steps and thus requires
more time than the retrieval of context-independent (seman-
tic) information. Alternatively, this issue can be viewed
from a neural network perspective. Allen and colleagues
(1998a, 1998b), for example, found that older adults had
increased entropy relative to younger adults on a spatial
memory scanning task (an episodic memory task). The age-
related increase in entropy resulted in increased “computa-
tional temperature” in older adults’ molar neural networks.
By entropy we mean the level of disorder in a system, and
by computational temperature we refer to the temperature
of a statistical mechanics system. Performance on a task be-
comes poorer when the computational temperature of a neu-
ral network increases. In the experiments of Allen and
colleagues (1998a, 1998b), older adults’ increased entropy/
computational temperature resulted in an age-related de-
crease in memory performance. If we assume that computa-
tional temperature has a relatively less adverse impact on
semantic memory than on episodic memory (perhaps as the
result of additional interconnections among nodes in seman-
tic memory, relative to episodic memory), then we can ac-
count for the memory type effects observed in the present
study. Although all three of these hypotheses remain viable,
we believe that latter two are particularly important in ac-
counting for this age difference across memory type.
As noted earlier, Models 5 and 6 are the best-fitting mod-
els (see Table 6), but these models result in standardized
path coefficients that contradict the path coefficients ob-
served in Models 1 and 2 (and correlations between age and
episodic task slopes presented in Table 3). Namely, older
adults exhibit significantly steeper slopes for episodic mem-
ory tasks, yet Models 5 and 6 exhibit negative path loadings
for the path from age to the Episodic Central Processes fac-
tor. This predicts that as age increases, episodic slopes de-
crease—a prediction that is wrong. This model mis-specifi-
cation was probably the result of model overfitting (Hall &
Allen, 2000).
The independent factors Models 1 and 2 actually fit better
than the less parsimonious hierarchical, common factors
Models 3 and 4. (Parsimony refers to the number of degrees
of freedom in a structural equation model—the more de-
grees of freedom, the more parsimony; see Allen et al.,
2000.) Hence, the present data do require direct paths from
age to first-order latent factors to achieve an acceptable
model fit (see Table 6). However, the present data do not re-
quire the presence of a common factor between age and the
first-order latent factors to achieve an acceptable fit (see Ta-
ble 6). The CFI and NNFI indices for the hierarchical, com-
mon factor Models 3 and 4 are both below .90, whereas
analogous indices for independent factors Models 1 and 2
are both above .90. These results suggest that age effects in
the present study are primarily nonshared (direct) in nature.
Models 5 and 6 suggest the existence of both shared and
nonshared age effects, but, again, these models resulted in
contradictory standardized path coefficients.
Conceptual Rather Than Empirical Inclusion
The only difference between Models 1 and 2, Models 3
and 4, and Models 5 and 6 is that the odd models contain
three indicator variables for the Semantic Central Processes
factor, whereas the even models contain four indicator vari-
ables for this factor (i.e., the multiplication verification 2
variable is added; see Table 6). Although the addition of this
fourth indicator variable for the Semantic Central Processes
factor does slightly improve model fit, without additional
conceptual justification it would not be included because
this variable did not load on the Semantic Central Processes
factor in the empirical, exploratory factor analysis. How-
ever, recall that we predicted that both lexical decision and
multiplication verification slopes would load on a semantic
memory factor. With Models 2, 4, and 6 we are able to de-
velop models with acceptable fits and with converging op-
erations for memory type. That is, in Models 2, 4, and 6, we
do have converging operations in the structural equation
modeling data for a differentiation across memory type and
processing stage (see Roediger, 1984), because there are
multiple measures for slopes and for intercepts across age.
Consequently, we believe that the even-numbered Model 2
provides initial evidence of a theoretical framework for the
relationship among adult age, memory type, and processing
stage.
Conclusion
In the present study we find that age directly mediates
performance on two of three process-specific and memory-
specific latent factors. To our knowledge, whereas investi-
gators have used structural equation modeling to examine
the potential contribution of domain-specific factors to age
differences (e.g., Salthouse & Czaja, 2000; Salthouse et al.,
1998), ours is the first study that actually develops latent
factors based on specific processes (episodic and semantic
memory) and processing stages (peripheral vs central pro-
cessing stages; although for a factor-specific theory of age
differences in intelligence, see McArdle & Prescott, 1992).
Therefore, we conclude that there are observable process-

AGE DIFFERENCES IN MEMORY TYPE
P185
Ashcraft, M. H. (1992). Cognitive arithmetic: A review of data and theory.
Cognition, 44, 75–106.
specific and stage-specific age differences when a structural
equation modeling methodology is used.
Ashcraft, M. H. (1995). Cognitive psychology and simple arithmetic: A re-
view and summary of new directions. Mathematical Cognition, 1, 3–34.
Balota, D. A., & Chumbley, J. I. (1984). Are lexical decisions a good index
of lexical access? The role of word frequency in the neglected decision
stage. Journal of Experimental Psychology: Human Perception and
Performance, 10, 340–357.
Acknowledgments
This research was supported by Research Grant AG09282 from NIA/
NIH. We are grateful to Beth Goldstein and Kathleen Jerge for technical as-
sistance.
Balota, D. A., & Ferraro, F. R. (1993). A dissociation of frequency and reg-
ularity effects in pronunciation performance across young adults, older
adults, and individuals with senile dementia of the Alzheimer’s type.
Journal of Memory and Language, 32, 573–592.
Address correspondence to Philip Allen, Department of Psychology,
325D Polsky Building, The University of Akron, Akron, OH 44325-4301.
E-mail: paallen@uakron.edu
Balota, D. A., & Ferraro, F. R. (1996). Lexical, sub-lexical, and implicit
memory processes in healthy young and healthy older adults and in in-
dividuals with dementia of the Alzheimer’s type. Neuropsychology, 10,
82–95.
Bashore, T. R., & Smulders, F. (1995). Do general slowing functions mask
local slowing effects? A chronopsychophysiological perspective. In P.
Allen & T. Bashore (Eds.), Age differences in word and language pro-
cessing (pp. 390–426). Amsterdam: North-Holland.
References
Allen, P. A. (1990). Influence of processing variability on adult age differ-
ences in memory distribution of order information. Cognitive
Development, 5, 177–192.
Allen, P. A. (1991). On age differences in processing variability and scan-
ning speed. Journal of Gerontology: Psychological Sciences, 46,
P191–P201.
Brinley, J. F. (1965). Cognitive sets, speed and accuracy of performance in
the elderly. In A. T. Welford & J. E. Birren (Eds.), Behavior, aging,
and the nervous system (pp. 114–149). Springfield, IL: Thomas.
Carr, T. H., & Pollatsek, A. (1985). Recognizing printed words: A look at
current models. In D. Besner, T. G. Waller, & G. E. MacKinnon (Eds.),
Reading research: Advances in theory and practice (Vol. 5, pp. 1–82).
New York: Academic Press.
Allen, P. A., Ashcraft, M., & Weber, T. A. (1992). On mental multiplica-
tion and age. Psychology and Aging, 7, 536–545.
Allen, P. A., Cerella, J., Madden, D. J., Smith, A. F., & Lien, M. (1999).
Adult age differences in visual word recognition: General or process-
specific deficits? Unpublished manuscript.
Allen, P. A., & Emerson, P. L. (1991). Holism revisited: Evidence for par-
allel independent word-level and letter-level processors during word
recognition. Journal of Experimental Psychology: Human Perception
and Performance, 17, 489–511.
Cerella, J. (1985). Information processing rates in the elderly. Psychologi-
cal Bulletin, 98, 67–83.
Cerella, J. (1991). Age effects may be local, not global: Comment on Fisk
and Rogers (1991). Journal of Experimental Psychology: General,
120, 235–243.
Coyne, A. C., Allen, P. A., & Wickens, D. D. (1986). Influence of adult
age on primary and secondary memory search. Psychology and Aging,
1, 187–194.
Allen, P. A., Hall, R. J., Druley, J., Smith, A. F., Sanders, R. E., & Murphy,
M. D. (2001). How shared are age-related influences on cognitive and
non-cognitive variables? Psychology and Aging, 16, 532–549.
Allen, P. A., Kaufman, M., Smith, A. F., & Propper, R. (1998a). A molar
entropy model of age differences in spatial memory. Psychology and
Aging, 13, 501–518.
Craik, F. I. M. (1977). Age differences in human memory. In J. E. Birren &
K. W. Schaie (Eds.), Handbook of the psychology of aging (pp. 384–
420). New York: Van Nostrand.
Allen, P. A., Kaufman, M., Smith, A. F., & Propper, R. (1998b). Age dif-
ferences in entropy: Primary versus secondary memory. Experimental
Aging Research, 24, 307–336.
Craik, F. I. M., & Jennings, J. M. (1992). Human memory. In F. I. M. Craik
& T. A. Salthouse (Eds.), The handbook of aging and cognition (pp.
51–110). Hillsdale, NJ: Erlbaum.
Allen, P. A., Madden, D. J., & Crozier, L. C. (1991). Adult age differences
in letter-level and word-level processing. Psychology and Aging, 6,
261–271.
Fisk, A. D., & Rogers, W. A. (1991). Toward an understanding of age-
related memory and visual search effects. Journal of Experimental
Psychology: General, 120, 131–149.
Forster, K. I. (1976). Accessing the mental lexicon. In R. J. Wales & E.
Walker (Eds.), New approaches to language mechanisms (pp. 89–109).
Amsterdam: North-Holland.
Allen, P. A., Madden, D. J., & Slane, S. (1995). Visual word encoding and
the effect of adult age and word frequency. In P. Allen & T. Bashore
(Eds.), Age differences in word and language processing (pp. 30–72).
Amsterdam: North-Holland.
Allen, P. A., Madden, D. J., Weber, T. A., & Groth, K. E. (1993). Influence
of age and processing stage on visual word recognition. Psychology
and Aging, 8, 274–282.
Allen, P. A., McNeal, M., & Kvak, D. (1992). Perhaps the lexicon is coded
as a function of word frequency. Journal of Memory and Language, 31,
826–844.
Allen, P. A., Smith, A. F., Jerge, K. A., & Vires-Collins, H. (1997). Age
differences in mental multiplication: Evidence for peripheral but not
central decrements. Journal of Gerontology: Psychological Sciences,
52B, P81–P90.
Gallistel, C. R., & Gelman, R. (1992). Preverbal and verbal counting and
computation. Cognition, 44, 43–74.
Geary, D. C., Frensch, P. A., & Wiley, J. G. (1993). Simple and complex
mental subtraction: Strategy choice and speed-of-processing differ-
ences in younger and older adults. Psychology and Aging, 8, 242–256.
Geary, D. C., & Wiley, J. G. (1991). Cognitive addition: Strategy choice
and speed-of-processing differences in young and elderly adults. Psy-
chology and Aging, 6, 474–483.
Hall, R. J., & Allen, P. A. (2000). Methodological issues in the use of
structural equation modeling in cognitive aging research. Manuscript
in preparation.
Hatcher, L. (1994). A step-by-step approach to using the SAS system for
factor analysis and structural equations modeling. Cary, NC: SAS
Institute.
Allen, P. A., Smith, A. F., Lien, M., Weber, T. A., & Madden, D. J. (1997).
Word frequency effects at brief exposure durations: Comment on Paap
and Johansen (1994). Journal of Experimental Psychology: Human
Perception and Performance, 23, 1792–1797.
Allen, P. A., Stadtlander, L. M., Groth, K. E., Pickle, J. L., & Madden, D. J.
(2000). Adult age invariance in sentence unitization. Aging, Neuropsy-
chology, and Cognition, 7, 54–67.
Allen, P. A., Wallace, B., & Weber, T. A. (1995). Influence of case type,
word frequency, and exposure duration on visual word recognition.
Journal of Experimental Psychology: Human Perception and Perfor-
mance, 21, 914–934.
Amrhein, P. C. (1995). Evidence for task specificity in age-related slow-
ing: A review of speeded picture-word processing studies. In P. Allen
& T. Bashore (Eds.), Age differences in word and language processing
(pp. 143–170). Amsterdam: North-Holland.
Kausler, D. H. (1994). Learning and memory in normal aging. New York:
Academic Press.
Kucera, H., & Francis, W. N. (1967). Computational analysis of present-
day American English. Providence, RI: Brown University Press.
Langley, L. K., & Madden, D. J. (2000). Functional neuroimaging of mem-
ory: Implications for cognitive aging. Microscopy Research and
Technique, 51, 75–84.
Laver, G. D., & Burke, D. M. (1993). Why do semantic priming effects in-
crease in old age? Psychology and Aging, 8, 34–43.
Light, L. (1991). Memory and aging: Four hypotheses in search of data.
Annual Review of Psychology, 43, 333–376.
Light, L. (1992). The organization of memory in old age. In F. I. M. Craik
Amrhein, P. C., & Theios, J. (1993). The time it takes elderly and young
individuals to draw pictures and to write words. Psychology and Aging,
8, 197–206.

P186
ALLEN ET AL.
& T. A. Salthouse (Eds.), The handbook of aging and cognition (pp.
111–167). Hillsdale, NJ: Erlbaum.
Light, L., & Singh, A. (1987). Implicit and explicit memory in young and
old adults. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 13, 531–541.
Salthouse, T. A., & Czaja, S. J. (2000). Structural constraints on process
explanations in cognitive aging. Psychology and Aging, 15, 44–55.
Salthouse, T. A., Hambrick, D. Z., & McGuthry, K. E. (1998). Shared age-
related influences on cognitive and noncognitive variables. Psychology
and Aging, 13, 486–500.
Lima, S. D., Hale, S., & Myerson, J. (1991). How general is general slowing?
Evidence from the lexical domain. Psychology and Aging, 6, 416–425.
McArdle, J. J., & Prescott, C. A. (1992). Age-based construct validation using
structural equation modeling. Experimental Aging Research, 18, 87–115.
McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation
model of context effects in letter perception: Part I. An account of basic
findings. Psychological Review, 88, 375–407.
McKoon, G., Ratcliff, R., & Dell, G. (1986). A critical evaluation of the se-
mantic-episodic distinction. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 12, 295–306.
Schneider, W. (1988). Micro Experimental Laboratory: An integrated sys-
tem for IBM PC compatibles. Behavior Research Methods, Instruments,
and Computers, 20, 206–217.
Sliwinski, M., Buschke, H., Kuslansky, G., Senior, G., & Scarisbrick, D.
(1994). Proportional slowing and addition speed in old and young
adults. Psychology and Aging, 9, 72–80.
Sliwinski, M., & Hall, C. B. (1998). Constraints on general slowing: A
meta-analysis using hierarchical linear models with random coeffi-
cients. Psychology and Aging, 13, 164–175.
Sternberg, S. (1967). Two operations in character recognition: Some evidence
for reaction-time experiments. Perception & Psychophysics, 2, 45–53.
Sternberg, S. (1969). The discovery of processing stages: Extensions of
Donder’s method. Acta Psychologica, 30, 276–315.
Mitchell, D. B. (1989). How many memory systems? Evidence from aging.
Journal of Experimental Psychology: Learning, Memory, and Cogni-
tion, 15, 31–49.
Monsell, S., Doyle, M. C., & Haggard, P. N. (1989). Effects of frequency
on visual word recognition tasks: Where are they? Journal of Experi-
mental Psychology: General, 118, 43–71.
Tulving, E. (1972). Episodic and semantic memory. In E. Tulving & W.
Donaldson (Eds.), Organization and memory (pp. 381–403). New
York: Academic Press.
Morton, J. (1969). Interaction of information in word recognition. Psycho-
logical Review, 76, 165–178.
Tulving, E. (1985). How many memory systems are there? American Psy-
chologist, 40, 385–398.
Nyberg, L., & Cabeza, R. (2000). Brain imaging of memory. In E. Tulving
& F. I. M. Craik (Eds.), The Oxford handbook of memory (pp. 501–
519). Oxford, England: Oxford University Press.
Verhaeghen, P., Kliegl, R., & Mayr, U. (1997). Sequential and coordina-
tive complexity in time-accuracy functions for mental arithmetic.
Psychology and Aging, 12, 555–564.
Pedhazur, E. J. (1982). Multiple regression in behavioral research. New
York: CBS College Publishing.
Wechsler, D. (1981). Wechsler Adult Intelligence Scale-Revised. New
York: Psychological Corporation.
Roberts, S., & Sternberg, S. (1993). The meaning of additive reaction time
effects: Tests of three alternatives. In D. E. Meyers & S. Kornblum
(Eds.), Attention and performance XIV (pp. 611–653). Cambridge, MA:
MIT Press.
Roediger, H. L. (1984). Does current evidence from dissociation experi-
ments favor the episodic/semantic distinction? Behavioral and Brain
Sciences, 7, 252–254.
Received December 7, 1999
Accepted December 4, 2000
Decision Editor: Toni C. Antonucci, PhD