Professional Documents
Culture Documents
Working Memory Training On Near and Far Transfer
Working Memory Training On Near and Far Transfer
Working Memory Training On Near and Far Transfer
To cite this article: H. Lee Swanson & Meaghan McMurran (2017): The impact of working memory
training on near and far transfer measures: Is it all about fluid intelligence?, Child Neuropsychology,
DOI: 10.1080/09297049.2017.1280142
Article views: 46
of sharing a common process is consistent with studies that have attempted to directly
intervene on WM performance and influence achievement (Holmes et al., 2009;
Klingberg et al., 2005).
In general, it is assumed that self-initial strategies are beneficial to near transfer
measures (same task format), whereas conditions that maximize performance and
thereby enhance focusing attention on previously forgotten items may be better suited
to far transfer measures.
To investigate whether individual differences in fluid intelligence influence treatment
outcomes, a mixed regression model was implemented by adjusting the covariate (fluid
intelligence in this case) at pretest and estimating post-test performance. In contrast to
an analysis of covariance (ANCOVA) procedure, this multiple linear regression (MLR)
increases power in treatment estimates, provides greater precision to treatment out-
comes, and avoids a dichotomization of the covariate (Hernández, Steyerberg, &
Habbema, 2004; Pocock, Assmann, Enos, & Kasten, 2002; Safarkhani & Moerbeek,
2013). The model assumes a treatment by covariate interaction (Judd, McClelland, &
Smith, 1996; Leon, Portera, Lowell, & Rheinheimer, 1998; Littell, Milliken, Stroup,
Wolfinger, & Schabenberger, 2006), but also has the added benefit of detailing the
role of individual differences as a function of the experimental conditions at pretest.
This procedure is consistent with a number of studies which have indicated that
individual differences need to be taken into consideration when interpreting WM
training effects (Jaeggi, Buschkuehl, Shah, & Jonides, 2014; Shah, Buschkuehl, Jaeggi,
& Jonides, 2012).
Method
Participants
A total of 81 second, third, and fourth graders drawn from elementary schools in
southern California participated in this study. After parent permission letters were
received, random assignment placed 36 children into the control group, 22 children
into the repeated practice condition, and 23 children into the cued recall condition. The
oversampling in the control group is based on the assumption that finding a sufficient
match in propensity scores occurs when the samples in the control groups are larger
than those in the treatment conditions (Rubin, 1979). Although children were ran-
domly assigned to conditions, propensity scores were computed to control for potential
unobserved biases emerging in the random assignment.1 A logistic regression was used
to predict treatment assignment. Consistent with Luellen, Shadish, and Clark (2005),
several covariates were entered into a logistic regression model to estimate propensity
scores, including forward digit span on the Wechsler Intelligence Scale for Children –
Third Edition (WISC-III; Wechsler, 1991), fluid intelligence on the Raven Colored
Progressive Matrices Test (Raven, 1976), pretest math problem-solving accuracy on
1
As a result of the use of random assignment to treatment, this study has high internal validity. However, as found in
several randomized treatment control studies, participants vary in WM span as well as a host of individual differences
(Jaeggi et al., 2014). As indicated in the literature, participants are very rarely randomly selected from a well-defined
population of interest into an experiment, which in turn results in low external validity (Tipton, 2013). To control for
any unintended pretest biases in the present study, propensity scores were computed in order to improve the
generalization of the findings.
CHILD NEUROPSYCHOLOGY 5
KeyMath (Connolly, 1998), and math calculation on the Wide Range Achievement Test
3 (WRAT3; Wilkinson, 1993), along with grade level, gender, pretest listening span and
pretest operation span. Only children with propensity scores within a 0.25 SD band
were analyzed (participants < 0.25 and > 0.75 were deleted for the analysis; see
Rosenbaum & Rubin, 1985).
From the initial sample of 81 children, the analysis was limited to participants who
were within the propensity score range of 0.25 to 0.75 SDs; thus, the final sample
includes 54 participants, with 19 in the control condition, 16 in the practice condition,
and 19 in the cued recall condition. A mixed analysis of variance (ANOVA) conducted
on propensity scores as a function of treatment was not found to be significant, F(2,
51) = 0.94, p = .40.
The mean age, gender representation and performance on normed reference mea-
sures—as a function of treatment conditions—are shown in Table 1. The mean age of
the sample was 8.57 years (SD = 1.01, range = 7.00–10.00) and the mean fluid
intelligence (Raven Test) was 108.03 (SD = 13.20, range = 77.00–133.00).
Training
The control condition received no training or practice on WM tasks. For the training
conditions, one designated WM task was administered for each session until an error
occurred. Errors are defined as recalling information out of order, inserting a non-
targeted item, and/or providing an incorrect answer related to the process question.
Training occurred for 15 minutes for six separate sessions that occurred on six separate
consecutive weekdays (e.g., Friday, Monday–Friday). The designated WM tasks admi-
nistered include WM tasks that were administered for both the repeated practice and
the cued recall conditions. The presentation of items and the duration of presenting the
WM measures are comparable between the two treatment conditions. The WM training
tasks require the participant to hold increasingly complex information in memory while
responding to a question about the task. The questions serve as distracters to item recall
because they reflect the recognition of targeted and closely related non-targeted items.
A question was asked for each set of items and the tasks were discontinued if the
question was answered incorrectly or if all items within a set could not be remembered.
A full description of each WM task, as well as the construct validity and reliability of
each measure, is reported in Swanson (1992, 1996). A brief description of each WM
task, administered in order for each session, is provided below.
the words lip, slip, and clip, and asked whether ship or lip was presented in the word set.
The participant is then asked to recall the previously presented words (lip, slip, and clip)
in order. The dependent measure is the number of sets recalled correctly (range = 0–9).
within each category. The dependent measure is the number of sets recalled correctly
(range = 0–8).
Probing Procedures
Participants in the cued recall condition were administered tasks in which initial,
probe, and gain scores were determined. The order of probing (cueing) is based on
the assumption that the first hint provides information about the final items
because these items are least susceptible to interference (Swanson, 1992). The
second hint provides information about the primacy (first) items because it is
assumed that they are most reliant on long-term memory processes. The third
hint provides additional information about the middle-presented items because
these items are the most susceptible to interference and storage limitations.
Finally, if the participant failed to benefit from the first three hints, all items
were repeated and retested.
Sample instructions related to the probing procedures are provided for Task 1 (the
rhyming task). If the participant omits or incorrectly inserts or orders the words, a
series of probe responses is presented. For example, consider the sample item “car, star,
bar, far” (Item 3) and the process question “Which word did I say—jar or star?”. The
probe sequence for this item is as follows:
(1) The last word in the sequence was far; now can you tell me all the words in
order?
(2) The first word in the sequence was car; now can you tell me all the words in
order?
CHILD NEUROPSYCHOLOGY 9
(3) The middle words in the sequence are star and bar; now can you tell me all the
words in order?
(4) All the words in order are car, star, bar, far; now can you tell me all the words?
For each set of items not recalled in the correct order, and for omitted or substituted
items, the experimenter provides a series of hints based on the error that most closely
relates to the probe. That is, the probes proceed from the least obvious hint (Probe 1) to
the next explicit hint that facilitates the recall of the answer. Once the appropriate hint
has been identified, on the basis of the location of the error, probes are presented in
order until the correct sequence is given. For example, suppose the participant responds
“car, bar, far” to Item 3. A word in the middle has been left out, so the experimenter
provides a hint related to the middle words (Probe 3 in this case). If Probe 3 does not
provide the correct response, the experimenter would then move on to Probe 4. In
contrast, if the participant initially responds by saying only “car” then the sequence
begins with Probe 1 and proceeds through all probes until the correct response is given.
If a correct response does not occur after the probing, the task is discontinued and the
next task is administered. If a correct response is given then the next set of items of
increased difficulty is presented.
and contain two assignment propositions, one relation, one question, and one extra-
neous proposition (irrelevant sentence) related to the solution. To control for reading
problems, the examiner orally read (a) each problem and (b) all multiple-choice
response options as the participant followed along. The total score possible for proposi-
tions relating to question, number, goal, operations, algorithms, and true/false questions
is 12.
To ensure that the participants comprehend the sentences (i.e., processed their
meaning and did not merely try to remember the target word or treat the task as one
of short-term memory), they are required to answer a question after each group of
sentences that was presented. Questions are related to a randomly selected sentence in
the set other than the last sentence. For example, for the 3-sentence example set from
above, the participants were asked “Where did we wait?”. After each set of unrelated
CHILD NEUROPSYCHOLOGY 11
sentences is presented, the participant answers a question about one of the sentences
and is then asked to recall the last word of each sentence. The number of sentences in
the set gradually increases from 1 to 8. Two identical forms were created and counter-
balanced for presentation order.
Moderator Measure
Fluid Intelligence
Fluid intelligence was assessed by the Raven Colored Progressive Matrices Test (Raven,
1976), which is considered a prime example of a fluid intelligence test in the WM
literature (Engel de Abreu, Conway, & Gathercole, 2010; Engle et al., 1999). For this
study, participants were given a booklet with patterns displayed on each page. There
was a piece missing from each pattern and six possible replacement pattern pieces were
displayed. Participants are required to circle the replacement piece that best completes
the patterns. After the introduction of the first matrix, the participants completed their
booklets at their own pace. The patterns progressively increase in difficulty. The
dependent measure (range 0 to 36) is the number of problems solved correctly,
which yields a standardized score (M = 100, SD = 15).
Procedures
Two post-doctoral fellows and six doctoral students administered the treatment condi-
tions and/or the pre- and post-tests. The WM training was conducted individually, with
one task delivered each day for 10 to 15 minutes for the total of 6 consecutive weekdays
of training. Two doctoral students who did not administer treatment conditions
implemented fidelity checks related to the administration of the treatment conditions.
The mean implementation fidelity across all sessions is 96%.
Statistical Analysis
A mixed ANCOVA was computed (SAS v9.4, PROC MIXED) as a function of treat-
ment on two near transfer measures (operation span, listening span) and two far
transfer measures (KeyMath problem-solving accuracy and identification of word
problem-solving components). The covariates in each mixed ANCOVA computation
are the continuous variables of pretest and fluid intelligence. A mixed ANCOVA was
computed because of the nested structure of the data. The nested structure includes the
children who participated in this study who were drawn from 12 separate classrooms
within 3 different schools. The primary purpose of the mixed ANCOVA is to determine
if a treatment × fluid intelligence interaction would emerge.
As a followed up to the interaction, a mixed regression model was computed that
adjusted fluid intelligence at pretest. This regression procedure is commonly used in
clinical trials when the sample size is small because the procedure allows for use of the
complete sample without dichotomizing the data (e.g., high and low fluid intelligence).
In addition, the procedure increases power and provides more accurate estimates of
treatment outcomes (Hernández et al., 2004; Pocock et al., 2002; Safarkhani &
Moerbeek, 2013). Thus, because a major focus of the present study is the influence of
12 H. L. SWANSON AND M. MCMURRAN
Results
Descriptive Analysis
Table 1 presents the demographic data and mean performance of the participants in
different treatment conditions at pretest. As shown in Table 1, the majority of partici-
pants performed within the normal range on measures of fluid intelligence (Raven Test)
and reading (WRAT3 and Test of Reading Comprehension). The Appendix presents
the means, SDs, kurtosis and skewness of raw score at pre- and post-test as a function of
each treatment condition.
For comparative purposes across the criterion measures, post-test criterion mea-
sures were converted to z-scores based on pretest performance (M = 0, SD = 1).
Table 2 shows the adjusted post-test z-scores (adjusted for pretest, fluid intelligence,
and random effects) for accuracy based on the means and SDs at pretest for the far
transfer measures (KeyMath, mathematical word problem-solving components).
Table 3 shows the same information as Table 2, but reports the results related to
Table 3. Adjusted Near Transfer Post-Test z-scores as a Function of Treatment and Measure.
Operation Span Gain Listening Span Gain
Group M SE M SE M SE M SE
Sample
Control 0.09 0.01 −0.21 0.08 0.02 0.17 −0.22 0.24
Repeated Practice 0.63 0.03 0.62 0.11 0.60 0.01 0.34 0.04
Cued Recall 0.20 0.12 −0.11 0.24 0.12 0.12 −0.11 0.08
Tukey RP>CR =C RP>CR = C RP>CR = C RP>CR = C
High Fluid Intelligence
Control 0.09 0.01 −0.21 0.09 0.05 0.19 −0.30 0.29
Repeated Practice 0.55 0.08 0.45 0.21 0.84 0.24 0.50 0.12
Cued Recall 0.22 0.15 0.04 0.32 −0.23 0.12 −0.39 0.08
Tukey RP>CR =C RP>CR = C RP>CR = C RP>CR = C
Average Fluid Intelligence
Control 0.09 0.01 −0.21 0.08 0.03 0.17 −0.23 0.24
Repeated Practice 0.62 0.04 0.59 0.13 0.63 0.03 0.36 0.03
Cued Recall 0.21 0.13 −0.10 0.25 0.08 0.12 −0.15 0.08
Tukey RP>CR =C RP>CR = C RP>CR = C RP>CR = C
Low Fluid Intelligence
Control 0.08 0.02 −0.21 0.06 −0.02 0.16 −0.11 0.17
Repeated Practice 0.74 0.04 0.83 0.08 0.28 0.32 0.13 0.22
Cued Recall 0.18 0.08 −0.21 0.14 0.59 0.11 0.26 0.09
Tukey RP>CR =C RP>CR = C CR>C; NS
RP = C; CR = RP
Note. Average Fluid Intelligence = 50th percentile; C = control group; CR = cued recall group; Gain = post-test minus
pretest; High Fluid Intelligence = 75th percentile; Low Fluid Intelligence = 25th percentile; RP = repeated practice
group; Sample = scores without setting fluid intelligence level.
the near transfer measures (operation span, listening span). Also shown in Tables 2
and 3 are the estimated adjusted post-test scores for the sample means (post-test
scores are not adjusted for pretest fluid intelligence levels). A comparison of the
treatment conditions when the fluid intelligence was not set to pretest levels shows
that the cued recall condition favors the far transfer measures and the repeated
practice condition favors the near transfer measures when compared to the control
conditions.
For all the analyses below, a mixed ANCOVA was first computed comparing the
three treatment conditions on post-test performance. The covariates in the analysis are
pretest and fluid intelligence. As a follow-up to the treatment × covariate interaction,
Tables 2 and 3 show the estimated post-test scores as a function of the mixed regression
setting fluid intelligence scores at pretest to high (75th percentile), middle (50th
percentile) and low (25th percentile) levels. As a further follow-up analysis, gain scores
(post-test to pretest) were computed to identify those conditions that yielded the largest
changes in performance.
Far Transfer
KeyMath Story Problem Accuracy
A mixed ANCOVA was computed on post-test KeyMath accuracy scores as a function
of treatment conditions. A significant effect was found for treatment, F(2, 50) = 12.04,
p < .0001, pretest, F(1, 50) = 68.17, p < .0001, and the treatment × fluid intelligence
interaction, F(2, 50) = 55.90, p < .0001. No significant effect was found for fluid
14 H. L. SWANSON AND M. MCMURRAN
intelligence as a main effect, F(1, 50) = 0.96, p = .33. Table 2 shows the adjusted post-
test scores as a function of treatment conditions.
As a follow-up to the interaction, a regression analysis estimated post-test scores
when setting fluid intelligence to high, average and low levels at pretest. The Tukey test
comparing the estimated means indicates a significant post-test advantage for the
treatment conditions relative to the control that occurred when fluid intelligence is
set to a low level (25th percentile; ps < .05). However, an inspection of the estimated
post-test scores reported in Table 2 suggests that performance for the control condition
is more likely to decrease as the fluid intelligence score decreases relative to the other
treatment conditions. Thus, it appears that the post-test advantage related to treatment
conditions is related to the decreasing performance on the control condition as fluid
intelligence scores decrease. A comparison of estimated mean post-test scores, via a
Tukey test, indicates that a significant difference occurs between the treatment condi-
tions and the control condition when fluid intelligence is set to a low level (ps < .05). No
other significant advantages relating to the treatment group when compared to the
control group emerged.
As a follow-up to this analysis, gain scores were computed. A significant effect was found
for treatment, F(2, 50) = 69.47, p < .001, and the treatment × fluid intelligence interaction, F
(2, 50) = 156.42, p < .0001. No significant effect was found for fluid intelligence as a main
effect, F(1, 50) = 0.01, p = .94. The estimated mean gain scores as a function of levels of fluid
intelligence are shown in Table 2. A Tukey test of the estimated mean gain scores yielded a
significant advantage for both WM conditions when compared to the control condition
when fluid intelligence is set to a low level (25th percentile; ps < .05).
(p < .05). In addition, the results suggest that the adjusted post-test mean scores
increase in the control and cued recall conditions as the fluid intelligence scores
decrease.
Summary
In general, the above results indicate that the post-test scores for the far transfer
measures are generally higher under the cued recall condition than the control condi-
tion. Figure 1 demonstrates a clear advantage for the cued recall condition relative to
the other conditions at post-test when fluid intelligence is set to a high level compared
to a low level. In addition, the general pattern in Figure 1 is that the estimated post-test
scores when fluid intelligence is set to a low level (25th percentile) do not supersede
estimates of the post-test scores when set to a high level (75th percentile). Thus, weak
support was found for the assumption that WM training compensates for low fluid
intelligence on far transfer measures.
Near Transfer
Operation Span
A significant main effect was found for treatment, F(2, 50) = 113.71, p < .0001, pretest, F
(1, 50) = 91.09, p < .0001, and the treatment × fluid intelligence interaction, F(2,
50) = 507.86, p < .0001. No significant effect was found for fluid intelligence, F(1,
50) = 0.38, p = .54. As shown in Table 3, the Tukey test indicates a significant advantage
for the repeated practice condition when compared to the other conditions (ps < .05).
1
Far Transfer
0.8
0.6
0.4
Adjusted
Post-Test 0.2 High GF
Z-score Low GF
0
-0.2
-0.4
-0.6
K-Control K-Practice K-Cued P-Control P-Practice P-Cued
Figure 1. Comparison of estimated adjusted post-test scores on far transfer (KeyMath Story
Problems and Identification of Problem-solving Components) as a function of setting fluid intelli-
gence to high (75th percentile) and low (25th percentile) levels.
Note. Control = control group; Cued = cued recall group; K = KeyMath Story Problems;
P = Identification of Problem-solving Components; Practice = repeated practice group; GF =
Group Fluid Intelligence.
16 H. L. SWANSON AND M. MCMURRAN
Summary
For the near transfer measures, the results in Table 3 show that the gain scores are
larger for the repeated practice condition relative to the control condition and this
advantage occurs across high and low levels of fluid intelligence. The general pattern is
that gain scores increase for the operation span measures as the fluid intelligence
measures decrease, whereas the reverse effect occurs for the listening span measure.
The clear advantage at post-test for the repeated practice condition relative to the
CHILD NEUROPSYCHOLOGY 17
1.2
Near Transfer
1
0.8
0.6
Adjusted 0.4
Post-Test High GF
Z-score 0.2 Low GF
0
-0.2
-0.4
-0.6
O-Control O-Practice O-Cued L-Control L-Practice L-Cued
Figure 2. Comparison of estimated adjusted post-test scores on near transfer (Operation Span,
Listening Sentence Span) as a function of setting fluid intelligence to high (75th percentile) and low
(25th percentile) levels.
Note. Control = control group; Cued = cued recall group; L = Listening Sentence Span;
O = Operation Span; Practice = repeated practice group.
for tasks that primarily require the processing of sentences (problem-solving compo-
nents, listening span). In addition, the ESs are generally higher for the cued recall
condition on the far transfer measures, whereas the ESs are higher for the repeated
practice condition on the near transfer measures.
The left side of Table 4 shows the magnitude of the ESs at post-test for the far
transfer measures. None of the ESs for problem-solving on the KeyMath measures
exceed a moderate magnitude (ESs > .50). Of interest, however, is the magnitude of the
ES when the treatment conditions are compared to the control condition. As shown in
Table 4, the magnitude of the ES for KeyMath performance for the cued recall
condition compared to the control condition is higher when fluid intelligence is set
to a low level (ES = 0.33 for accuracy, ES = 0.45 for gain score) compared to a high level
(ES = 0.09 for accuracy, ES = 0.02 for gain score). However, as mentioned earlier, these
findings may reflect a substantial decrease in post-test performance for the control
condition. In contrast to the KeyMath measure, the magnitude of the ES for the
problem-solving component measure for the cued recall condition is highest when
fluid intelligence is set to a high level (ES = 0.69 for accuracy, ES = 0.37 for gain score)
compared to a low level (ES = 0.47 for accuracy, ES = 0.29 for gain score). A moderate
ES emerges when comparing the cued recall condition to the control condition
(ES = 0.69 for accuracy).
The right side of Table 4 shows the magnitude of ESs at post-test for the near
transfer measures. The magnitude of the ES shows a clear advantage for the repeated
practice condition when compared to the other conditions. When compared to the
control condition, the highest ES for the operation span measure occurs for the
repeated practice condition when the fluid intelligence pretest level is low (ES = 0.57
for accuracy, ES = 1.07 for gain scores). The reverse pattern occurs for performance on
CHILD NEUROPSYCHOLOGY 19
the listening span measure. When compared to the control condition, the highest ES for
the listening span measure occurs for the repeated practice condition when the fluid
intelligence pretest level is high (ES = 0.57 for accuracy, ES = 0.73 for gain scores)
rather than low (ES = 0.22 for accuracy, ES = 0.22 for gain scores).
Discussion
This study investigates the role of fluid intelligence on WM training outcomes. The
criterion measures that are used to assess training outcomes include both near and far
transfer measures. A significant treatment × fluid intelligence interaction emerged
across all criterion measures. For the far transfer measure (tasks not directly related
to treatment), improvements on orally solving math problems or identifying compo-
nents of sentences in word problems were assessed. The performance analysis shows a
general advantage for the cued recall condition. In general, as shown in Figure 1, few
advantages in performance for the cued recall condition occur in the far transfer
performance when the fluid intelligence at pretest is set to a low level compared to a
high level. For the near transfer measures, improvements on the operation and listening
span measures are assessed. Because the participants in each training session received
practice on the WM tasks, and because this skill is closely aligned with the intervention,
increases in the near transfer measures at post-test were expected. The performance
analysis shows a general advantage for the repeated practice condition. As shown in
Figure 2, an advantage is found for the repeated practice condition on the operation
span measure when the fluid intelligence at pretest is set to a low level. The reverse
pattern occurs for the listening span measure. The results are now discussed in terms of
the two questions that directed the study.
The first question is: Do individual levels of fluid intelligence play an important role in
accounting for WM training outcomes? Although the answer to this question is in the
affirmative, the results must be placed in the context of the models discussed in the
introduction. One model argues that individual differences in fluid intelligence would
20 H. L. SWANSON AND M. MCMURRAN
that the training advantages in performance were generally better for the repeated
practice condition on the near transfer measures and better for the cued recall condi-
tion on the far transfer measures. The best explanation for this is that the two
conditions draw upon different strategies. The repeated practice condition reflects the
participant’s self-initiated strategies to access information. In contrast, the cued recall
condition may have eliminated idiosyncratic retrieval processes by requiring partici-
pants to reinstate memory traces or retrieve forgotten items in a serial fashion. Thus, it
appears that advantages related to practice conditions may reflect procedures related to
self-initiated strategies for accessing information whereas the cued recall condition
directs attention toward retrieving forgotten items. Thus, it is inferred that one of the
reasons why the performance on the near transfer measure was improved is that similar
strategies related to practice could be used. In contrast, it is inferred that the cued recall
condition may have facilitated far transfer because the training emphasized maximizing
performance, fine-tuning procedural knowledge, and more specifically directing parti-
cipants to focus their attention on retrieving forgotten items.
There are at least three limitations to this study. First, the regression estimates made
of post-test outcomes are limited to the range of the fluid intelligence data. Thus,
although the study uses regression modeling with an adequate sample to task ratio as
well as taking into consideration the complete sample and ranges in fluid intelligence,
further testing is necessary with a larger sample and larger ranges in measures of fluid
intelligence.
Second, depending on where the pretest values of fluid intelligence are set, some
participants did not appear to benefit from WM training, suggesting that such training
may need to be adapted to the fluid intelligence of the individual at pretest. Although
such procedures make random assignment to conditions difficult, accounting for
individual differences prior to treatment may yield different treatment outcomes (on
adaptive procedures, see Jaeggi, Buschkuehl, Jonides, & Shah, 2012; Karbach et al.,
2015).
Finally, the mechanism that WM training compensates for when fluid intelligence is
set to a relatively low level has not been identified. That is, fluid intelligence and WM
comprise much more than controlled attention or increases in procedural knowledge.
Although controlled attention is a mechanism associated with both fluid intelligence
and WM, there may be other areas of overlap not captured in this study. No doubt,
measures at several levels of WM (primary memory, secondary memory, and attention
control) are required before suggesting that WM training has facilitated both near and
far transfer (Shipstead, Redick, & Engle, 2010). The training in this study is assumed to
have tapped into one process (attention control), but others (primary memory, second-
ary memory) may also have played an important role.
There are important implications from the findings of this study. Namely, depending
on the type of training and the level of the child’s fluid intelligence, WM training can
have a positive impact on both near and far transfer measures. Moreover, assuming that
the effect of WM training on near and far transfer measures depends on the individual’s
degree of fluid intelligence, it may be possible to identify which individuals, based on
their measure of fluid intelligence, will benefit from WM training and which will not.
Thus, future work aimed toward identifying precisely at what threshold of fluid
intelligence training begins and/or ceases to improve performance would be very
22 H. L. SWANSON AND M. MCMURRAN
Acknowledgements
The authors are indebted to Lindsay Flynn, Xinhua Zheng, Dana Bronwell, Loren Alberg,
Catherine Tung, Dennis Sisco-Taylor, Kenisha Williams, Garett Briney, Kristi Bryant, and
Orheta Rice for their work on the data collection and/or treatment administration.
Appreciation is given to Bonsall School and the Academy for Academic Excellence/Lewis
Center for Educational Research (Corwin and Norton Campus).
Disclosure statement
No potential conflict of interest was reported by the authors.
Funding
This work was supported by the US Department of Education’s Institute of Education Sciences
[grant number USDE R324A090002].
References
Ackerman, P. L., Beier, M. E., & Boyle, M. O. (2005). Working memory and intelligence: The
same or different constructs? Psychological Bulletin, 131(1), 30–60. doi:10.1037/0033-
2909.131.1.30
Alloway, T. P., & Alloway, R. G. (2010). Investigating the predictive roles of working memory
and IQ in academic attainment. Journal of Experimental Child Psychology, 106, 20–29.
doi:10.1016/j.jecp.2009.11.003
Anderson, J. R. (1987). Skill acquisition: Compilation of weak-method problem situations.
Psychological Review, 94, 192–210. doi:10.1037/0033-295X.94.2.192
Au, J., Buschkuehl, M., Duncan, G. J., & Jaeggi, S. M. (2016). There is no convincing evidence
that working memory training is NOT effective: A reply to Melby-Lervåg and Hulme (2015).
Psychonomic Bulletin & Review, 23(1), 331–337. doi:10.3758/s13423-015-0967-4
Cattell, R. B. (1971). Abilities: Their structure, growth, and action. Boston, MA: Houghton Mifflin.
Cohen, J. (1988). Statistical power analysis for the behavioral science. (2nd ed.) Hillsdale, NJ:
Lawrence Erlbaum Associates
Colom, R., Abad, F. J., Quiroga, M. Á., Shih, P. C., & Flores-Mendoza, C. (2008). Working
memory and intelligence are highly related constructs, but why? Intelligence, 36(6), 584–606.
doi:10.1016/j.intell.2008.01.002
Connolly, A. J. (1998). KeyMath (revised/normative update). Circle Pines, MN: American
Guidance.
CHILD NEUROPSYCHOLOGY 23
Conway, A. R., Cowan, N., Bunting, M. F., Therriault, D. J., & Minkoff, S. R. (2002). A latent
variable analysis of working memory capacity, short-term memory capacity, processing
speed, and general fluid intelligence. Intelligence, 30(2), 163–183. doi:10.1016/S0160-2896
(01)00096-4
Daneman, M., & Carpenter, P. (1980). Individual differences in working memory and reading.
Journal of Verbal Learning and Verbal Behavior, 19, 450–466. doi:10.1016/S0022-5371(80)
90312-6
Dunning, D. L., Holmes, J., & Gathercole, S. E. (2013). Does working memory training lead to
generalized improvements in children with low working memory? A randomized controlled
trial. Developmental Science, 16(6), 915–925. Retrieved from http://search.proquest.com/doc
view/1499095407?accountid=14521
Engel de Abreu, P. M. J., Conway, A. R. A., & Gathercole, S. E. (2010). Working memory and
fluid intelligence in young children. Intelligence. doi:10.1016/j.intel.2010.07.003
Engle, R. W., Cantor, J., & Carullo, J. J. (1992). Individual differences in working memory and
comprehension: A test of four hypotheses. Journal of Experimental Psychology: Learning,
Memory and Cognition, 18, 972–992. doi:10.1037//0278-7393.18.5.972
Engle, R. W., Tuholski, S. W., Laughlin, J. E., & Conway, A. R. (1999). Working memory, short-
term memory, and general fluid intelligence: A latent variable approach. Journal of
Experimental Psychology: General, 128, 309–331. doi:10.1037//0096-3445.128.3.309
Fukuda, K., Vogel, E., Mayr, U., & Awh, E. (2010). Quantity, not quality: The relationship
between fluid intelligence and working memory capacity. Psychonomic Bulletin & Review, 17
(5), 673–679. doi:10.3758/17.5.673
Harrison, T. L., Shipstead, Z., Hicks, K. L., Hambrick, D. Z., Redick, T. S., & Engle, R. W. (2013).
Working memory training may increase working memory capacity but not fluid intelligence.
Psychological Science, 24, 2409–2419. doi:10.1177/0956797613492984
Hayes, T. R., Petrov, A. A., & Sederberg, P. B. (2015). Do we really become smarter when our
fluid-intelligence test scores improve? Intelligence, 48, 1–14. doi:10.1016/j.intell.2014.10.005
Hernández, A. V., Steyerberg, E. W., & Habbema, J. D. (2004). Covariate adjustment in
randomized controlled trials with dichotomous outcomes increases statistical power and
reduces sample size requirements. Journal of Clinical Epidemiology, 57(5), 454–460.
doi:10.1016/j.jclinepi.2003.09.014
Holmes, J., Gathercole, S. E., & Dunning, D. L. (2009). Adaptive training leads to sustained
enhancement of poor working memory in children. Developmental Science, 12, F9–F15.
doi:10.1111/j.1467-7687.2009.00848.x
Jaeggi, S. M., Buschkuehl, M., Jonides, J., & Shah, P. (2012). Cogmed and working memory
training—Current challenges and the search for underlying mechanisms. Journal of Applied
Research in Memory and Cognition, 1(3), 211–213. doi:10.1016/j.jarmac.2012.07.002
Jaeggi, S. M., Buschkuehl, M., Shah, P., & Jonides, J. (2014). The role of individual differences in
cognitive training and transfer. Memory & Cognition, 42(3), 464–480. doi:10.3758/s13421-013-
0364-z
Jaeggi, S. M., & Buschkuehl, M. (2014). Working memory training and transfer: Theoretical and
practical considerations. In B. Toni (Ed.), New frontiers of multidisciplinary research in
STEAM-H. Switzerland: Springer International Publishing.
Judd, C. M., McClelland, G. H., & Smith, E. R. (1996). Testing treatment by covariate interac-
tions when treatment varies within subjects. Psychological Methods, 1, 366–378. doi:10.1037/
1082-989X.1.4.366
Kane, M. J., Bleckley, M. K., Conway, A. R. A., & Engle, R. W. (2001). A controlled-attention
view of working-memory capacity. Journal of Experimental Psychology: General, 130(2), 169–
183. doi:10.1037/0096-3445.130.2.169
Karbach, J., Strobach, T., & Schubert, T. (2015). Adaptive working-memory training benefits
reading, but not mathematics in middle childhood. Child Neuropsychology, 21(3), 285–301.
doi:10.1080/09297049.2014.899336
Klingberg, T., Fernell, E., Olesen, P. J., Johnson, M., Gustafsson, P., Dahlström, K., & Westerberg,
H. (2005). Computerized training of working memory in children with ADHD–A
24 H. L. SWANSON AND M. MCMURRAN
randomized, controlled trial. Journal of the American Academy of Child & Adolescent
Psychiatry, 44(2), 177–186. doi:10.1097/00004583-200502000-00010
Kyllonen, P. C., & Christal, R. E. (1990). Reasoning ability is (little more than) working-memory
capacity? Intelligence, 14, 389–433. doi:10.1016/S0160-2896(05)80012-1
Leon, A. C., Portera, L., Lowell, K., & Rheinheimer, D. (1998). A strategy to evaluate a covariate
by group interaction in an analysis of covariance. Psychopharmacology Bulletin, 34(4), 805–
809.
Littell, R. C., Milliken, G. A., Stroup, W. W., Wolfinger, R. D., & Schabenberger, O. (2006). SAS
for mixed models (2nd ed.). Cary, NC: SAS Publishing.
Loosli, S. V., Buschkuehl, M., Perrig, W. J., & Jaeggi, S. M. (2012). Working memory training
improves reading processes in typically developing children. Child Neuropsychology, 18(1), 62–
78. doi:10.1080/09297049.2011.575772
Luellen, J. K., Shadish, W. R., & Clark, M. H. (2005). Propensity scores: An introduction and
experimental test. Evaluation Review, 29(6), 530–558. doi:10.1177/0193841X05275596
MacCallum, R. C., Zhang, S., Preacher, K. J., & Rucker, D. D. (2002). On the practice of
dichotomization of quantitative variables. Psychological Methods, 7(1), 19–40. doi:10.1037/
1082-989X.7.1.19
Melby-Lervåg, M., & Hulme, C. (2013). Is working memory training effective? A meta-analytic
review. Developmental Psychology, 49(2), 270–291. doi:10.1037/a0028228
Peng, P., Namkung, J., Barnes, M., & Sun, C. (2016). A meta-analysis of mathematics and
working memory: Moderating effects of working memory domain, type of mathematics
skill, and sample characteristics. Journal of Educational Psychology, 108(4), 455–473.
doi:10.1037/edu0000079
Pocock, S. J., Assmann, S. E., Enos, L. E., & Kasten, L. E. (2002). Subgroup analysis, covariate
adjustment and baseline comparisons in clinical trial reporting: Current practiceand problems.
Statistics in Medicine, 21(19), 2917–2930. doi:10.1002/sim.1296
Preacher, K. J., Rucker, D. D., MacCallum, R. C., & Nicewander, W. A. (2005). Use of the
extreme groups approach: A critical reexamination and new recommendations. Psychological
Methods, 10(2), 178–192. doi:10.1037/1082-989X.10.2.178
Raven, J. C. (1976). Colored Progressive Matrices Test. London, England: H. K. Lewis.
Redick, T. S., Shipstead, Z., Harrison, T. L., Hicks, K. L., Fried, D. E., Hambrick, D. Z., . . . Engle,
R. W. (2013). No evidence of intelligence improvement after working memory training: A
randomized, placebo-controlled study. Journal of Experimental Psychology: General, 142(2),
359–379. doi:10.1037/a0029082
Rosenbaum, P. R., & Rubin, D. B. (1985). Constructing a control group using multivariate
matching sampling methods that incorporate the propensity score. American Statistican, 39,
33–38.
Rubin, D. B. (1979). Using multivariate matched sampling and regression adjustments to control
bias in observational studies. Journal of the American Statistical Association, 74, 318–328.
Safarkhani, M., & Moerbeek, M. (2013). Covariate adjustment strategy increases power in the
randomized controlled trial with discrete-time survival endpoints. Journal of Educational and
Behavioral Statistics, 38(4), 355–380. doi:10.3102/1076998612461832
SAS Institute, Inc. (2010). SAS program version 9.3. Cary, NC: Author.
Schwaighofer, M., Fischer, F., & Bühner, M. (2015). Does working memory training transfer? A
meta-analysis including training conditions as moderators. Educational Psychologist, 50(2),
138–166. doi:10.1080/00461520.2015.1036274
Shah, P., Buschkuehl, M., Jaeggi, S., & Jonides, J. (2012). Cognitive training for ADHD: The
importance of individual differences. Journal of Applied Research in Memory and Cognition, 1
(3), 204–205. doi:10.1016/j.jarmac.2012.07.001
Shipstead, Z., Harrison, T. L., & Engle, R. W. (2015). Working memory capacity and the scope
and control of attention. Attention, Perception, & Psychophysics, 77(6), 1863–1880.
doi:10.3758/s13414-015-0899-0
Shipstead, Z., Redick, T. S., & Engle, R. W. (2010). Does working memory training generalize?
Psychologica Belgica, 50(3–4), 245–276. doi:10.5334/pb-50-3-4-245
CHILD NEUROPSYCHOLOGY 25
Shipstead, Z., Redick, T. S., & Engle, R. W. (2012). Is working memory training effective?
Psychological Bulletin, 138(4), 628–654. doi:10.1037/a0027473
Swanson, H. L. (1992). Generality and modifiability of working memory among skilled and less
skilled readers. Journal of Educational Psychology, 84, 473–488. doi:10.1037/0022-
0663.84.4.473
Swanson, H. L. (1996). Individual and age-related differences in children’s working memory.
Memory & Cognition, 24(1), 70–82. doi:10.3758/BF03197273
Swanson, H. L. (2014). Does cognitive strategy training on word problems compensate for
working memory capacity in children with math difficulties? Journal of Educational
Psychology, 106(3), 831–848. doi:10.1037/a0035838
Swanson, H. L. (2017). Word problem solving, working memory and serious math difficulties:
Do cognitive strategies really make a difference? Journal of Applied Memory and Cognition.
doi:10.1016/j.jarmac.2016.04.012
Swanson, H. L., & Alloway, T. P. (2012). Working memory, learning, and academic achievement.
In K. R. Harris, S. Graham., T. Urdan, C. B. McCormick, G. M. Sinatra, & J. Sweller (Eds.),
APA educational psychology handbook, vol 1: Theories, constructs, and critical issues (pp. 327–
366). Washington, DC: American Psychological Association.
Swanson, H. L., & Beebe-Frankenberger, M. B. (2004). The relationship between working
memory and mathematical problem solving in children at risk and not at risk for serious
math difficulties. Journal of Educational Psychology, 96(3), 471–491. doi:10.1037/0022-
0663.96.3.471
Swanson, H. L., Kehler, P., & Jerman, O. (2010). Working memory, strategy knowledge, and
strategy instruction in children with reading disabilities. Journal of Learning Disabilities, 43(1),
24–47. doi:10.1177/0022219409338743
Tipton, E. (2013). Improving generalizations from experiments using propensity score subclas-
sification: Assumptions, properties, and contexts. Journal of Educational and Behavioral
Statistics, 38(3), 239–266.
Titz, C., & Karbach, J. (2014). Working memory and executive functions: Effects of training on
academic achievement. Psychological Research, 78(6), 852–868. doi:10.1007/s00426-013-0537-1
Turley-Ames, K. J., & Whitefield, M. M. (2003). Strategy training and working memory task
performance. Journal of Memory and Language, 49, 446–468. doi:10.1016/S0749-596X(03)
00095-0
Wechsler, D. (1991) Wechsler intelligence scale for children (3rd ed.). San Antonio, TX:
Psychological Corporation.
What Works Clearing House. (2006). Technical details of WWC-conducted comparisons (9-12-
2006). U.S. Department of Education.
Wilkinson, G.S. (1993). The wide range achievement test. Wilmington, DE: Wide Range, Inc.
26 H. L. SWANSON AND M. MCMURRAN
Appendix