See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/330201504

The Cognitive Basis for The Split-attention Effect

Article in Journal of Experimental Psychology General · January 2019

DOI: 10.1037/xge0000578

CITATIONS READS
21 1,744

4 authors:

Wim Pouw Gertjan Rop

Radboud University Erasmus University Rotterdam
71 PUBLICATIONS 893 CITATIONS 15 PUBLICATIONS 111 CITATIONS

SEE PROFILE SEE PROFILE

Björn de Koning Fred Paas

Erasmus University Rotterdam Erasmus University Rotterdam and University of Wollongong
70 PUBLICATIONS 2,117 CITATIONS 317 PUBLICATIONS 33,726 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Role of Gesture as Simulated Action in Reinterpretation of Mental Imagery View project

Better together? The Effect of Collaborative Learning and Shared Regulation on Learning Outcomes in Higher Education View project

All content following this page was uploaded by Wim Pouw on 07 February 2019.

The user has requested enhancement of the downloaded file.

 1

This paper is not the copy of record and may not exactly replicate the final, authoritative
version of the article. Please do not copy or cite without authors' permission. The final
article will be available, upon publication, via its DOI: 10.1037/xge0000578

The Cognitive Basis for The Split-attention Effect

Wim Pouw1, 2**, Gertjan Rop1**, Bjorn de Koning1, and Fred Paas1,3

**Shared first authorship

1. Erasmus University Rotterdam, Department of Psychology, Education & Child Studies, The

Netherlands

2. University of Connecticut, Department of Psychological Sciences, USA

3. University of Wollongong, School of Education/Early Start, Australia

Author Note:

Correspondence: Gertjan Rop, Erasmus School of Social and Behavioural Sciences, Erasmus

University Rotterdam, 3000 DR Rotterdam, E-mail: rop@essb.eur.nl, Phone: +31 10 4089041.

Open Data:

Raw data, analyses scripts, pre-registration experiment 2, and Python experiment code and

stimuli supporting this research report can be retrieved from The Open Science Framework

(https://osf.io/ruqfk/).

Acknowledgments:

We would like to express our gratitude to Alex van Straaten, Sven Cammeraat, and Tudor

Cristea for their help during the data collection. We would also like to thank prof. dr. Rolf

Plötzner for providing the materials used in Experiment 3. This research was funded by the

Excellence Initiative grant from the Erasmus University Rotterdam awarded to the Educational

Psychology section. We also thank the Erasmus Behavioral Lab, especially Marcel Boom, for

their continued help in the lab.

 2

Abstract (249/250 words)

The split-attention effect entails that learning from spatially separated, but mutually referring

information sources (e.g., text and picture) is less effective than learning from the equivalent

spatially integrated sources. According to cognitive load theory, impaired learning is caused by

the working memory load imposed by the need to distribute attention between the information

sources and mentally integrate them. In this study, we directly tested whether the split-attention

effect is caused by spatial separation per se. Spatial distance was varied in basic cognitive tasks

involving pictures (Experiment 1) and text-picture combinations (Experiment 2; pre-registered

study), and in more ecologically valid learning materials (Experiment 3). Experiment 1 showed

that having to integrate two pictorial stimuli at greater distances diminished performance on a

secondary visual working memory task, but did not lead to slower integration. When participants

had to integrate a picture and written text in Experiment 2, a greater distance led to slower

integration of the stimuli, but not to diminished performance on the secondary task. Experiment

3 showed that presenting spatially separated (compared to integrated) textual and pictorial

information yielded fewer integrative eye movements, but this was not further exacerbated when

increasing spatial distance even further. This effect on learning processes did not lead to

differences in learning outcomes between conditions. In conclusion, we provide evidence that

larger distances between spatially separated information sources influence learning processes,

but that spatial separation on its own is not likely to be the only, nor a sufficient, condition for

impacting learning outcomes.

Keywords: split-attention effect, cognitive load theory, working memory, educational

psychology, cognitive psychology

Word count: 10.646

 3

The combination of instructional text (written or spoken) and pictorial information (static

or dynamic) is ubiquitous nowadays in textbooks and e-learning resources. Research on this so-

called multimedia learning, which is typically based on Cognitive Load Theory (CLT; Sweller,

Ayres, & Kalyuga, 2011) and the Cognitive Theory of Multimedia Learning (CTML; Mayer,

2014), has shown that learning generally improves when learning materials combine pictures and

text (i.e., the multimedia principle; Butcher, 2014). However, simply combining text and pictures

without further deliberation about how to present them to learners likely leads to suboptimal

learning (Ayres & Sweller, 2014; Mayer & Fiorella, 2014). A well-known finding in this respect

is the split-attention effect (e.g., Ayres & Sweller, 2014; Chandler & Sweller, 1991, 1992; Florax

& Ploetzner, 2010; Ginns, 2006; Mayer & Fiorella, 2014)1. The effect entails that when textual

and pictorial information that need to be integrated for learning (e.g., text and pictures) are

spatially separated, learning is hindered as compared to spatially integrated sources. This general

finding has led instructional designers to promote spatial integration of multimedia sources

(Ayres & Cierniak, 2012).

Often when the split-attention effect is obtained it is merely assumed that the effect is

produced by “splitting attention” over spatially separated information; however the underlying

mechanism of the split-attention effect is rarely directly tested. The general explanation for the

split-attention effect is provided by CLT (Sweller et al., 2011), and dictates that diminished

learning is caused by increased cognitive load imposed by spatial separation (Paas & Sweller,

1
Note that the “split-attention effect” can have a different meaning outside of educational

psychology. This then concerns the degree to which humans can visually track or detect two or

more (moving) objects at once (e.g., Awh & Pashler, 2000). In this study, we focus on the split-

attention effect as it is referred to in educational psychology.

 4

2014). The need to search for related elements in the textual and pictorial information sources,

while keeping relevant information active in working memory to mentally connect

corresponding information has been argued to impose an extraneous cognitive load (e.g.,

Sweller, Van Merrienboer, & Paas, 1998). Given that working memory is limited in capacity and

duration (Baddeley, 2000; Barrouillet & Camos, 2007; Cowan, 2001; Miller 1956; Puma et al.,

2018), this reduces working memory resources available for processes that are relevant for

learning, such as schema construction, elaboration, and automation in long-term memory

(Sweller, 1994). Consequently, learning is hampered. For integrated sources though, the burden

on working memory is limited as information sources can be directly visually compared. A

myriad of studies have shown that, in line with this explanation, spatially integrated learning

materials impose a lower cognitive load and lead to higher learning outcomes than spatially

separated learning materials (Bodemer, Ploetzner, Feuerlein, & Spada, 2004; Chandler &

Sweller, 1991, 1992; Mayer, Steinhoff, Bower, & Mars, 1995; Tarmizi & Sweller, 1988; see

Ginns, 2006 for a meta-analysis).

However, the split-attention effect may not be caused by spatial separation of related

information per se. Several techniques exist that can be used to resolve the split-attention effect

that are not resolving spatial separation per se. For example, the direction of learners‟ attention

by signaling the corresponding parts of the text and picture is frequently employed (De Koning

& Jarodzka, 2017; Jamet, 2014; Van Gog, 2014). Research has shown that mental integration of

textual and pictorial information is improved when corresponding text and parts of the picture

are presented in the same colour (De Koning, Tabbers, Rikers, & Paas, 2009; Ozcelik, Arslan-

Ari, & Cagiltay, 2010; Ozcelik, Karaku.s, Kursun, & Cagiltay, 2009). Importantly, when a

student actively integrates instructional materials consisting of two mutually referring

 5

information sources, comprehension is improved (Mayer, 2014; Schnotz, 2014). Moreover, using

co-referring labels, such as when dividing spatially separated information into smaller segments

and labelling the corresponding text-picture parts with numbers, also reduces the split-attention

effect (Florax & Ploetzner, 2010). It might therefore be that the searching costs that are imposed

by non-integrated information sources are not caused by spatial distance, but because the learner

does not know which pieces of information belong together and has to perform an effortful

search to semantically relate them. In integrated learning materials, spatial contiguity

automatically signals which information sources belong together. Thus, it is currently unclear

whether split attention is caused by spatial separation, or due to obscurities about which

informational sources should be related.

In the present study, we therefore investigated to what extent spatial distance of

information sources which impose working memory load on the learner can account for the split-

attention effect. That spatial distance plays a role in the split-attention effect, as originally

conceived, is consistent with classic cognitive psychological research on embedded cognition

(Ballard, Hayhoe, & Pelz, 1995; Gray & Fu, 2004; for a review see Pouw, van Gog, & Paas,

2014). It has been shown that when information that needs to be integrated is spatio-temporally

separated (Ballard et al., 1995; Gray & Fu, 2004), problem solvers change from a perceptually

intensive strategy (less prone to mistakes; higher saccade counts) to what seems to be a memory

intensive strategy (leading to more mistakes; lower saccade counts). More precisely, Ballard et

al. (1995) used a task in which participants had to copy a pattern of colored blocks. When the

distance between the model and the workspace in which participants had to copy the pattern of

blocks was small (model and workspace were separated by 15o), and the cost of direct

acquisition of information was small, participants made more saccades, implying less use of
 6

working memory resources. When this distance between information sources was increased

(70o), thereby elevating the cost of direct visual comparison, participants made fewer saccades

implying more use of working memory resources (see also Haselen, 2000 for a replication). In

the study by Gray and Fu (2004), it was found that participants who memorized task-relevant

information before the main task were more likely to use this memorized information when it

would take more time to attain this information from the computer display given a number of

mouse-clicks. When the number of mouse-clicks that were needed to attain the relevant

information was reduced, participants were less likely to rely on their own memory and would

attain this information in the digital environment. Interestingly, this had the effect that more

mistakes were made when information was less easily attainable as participants were more likely

to rely on imperfect “information-in-the-head”, as opposed to participants that could rely on

perfect “information-in-the-world” in the condition where there was low cost of retrieving the

information (i.e., time and effort to get the relevant information from the display). Together,

these studies (Ballard et al., 1995; Gray & Fu, 2004) indicate that there seems to be a trade-off

between spatio-temporal separation and the use of memory resources.

In more applied settings, comparable findings have been obtained. Johnson and Mayer

(2012), for example, recorded participants‟ eye movements while they learned how car brakes

work using a single-slide multimedia lesson consisting of a diagram and text. When the text was

integrated in the relevant parts of the diagram, participants made more saccades between these

two sources of information than when the text was presented separated from the diagram.

Moreover, participants‟ understanding of how car brakes work was better in the integrated

condition. In a study by Bauhoff, Huff, and Schwan (2012) participants judged whether or not

two depictions of a mechanical pendulum clock were identical. The spatial distance between
 7

these two depictions was varied, and Bauhoff et al. (2012) observed that participants made fewer

saccades between the two depictions of these clocks when the spatial distance between the

pictures was increased, suggesting higher working memory constraints. Together, these studies

suggest that non-integrated information (Johnson & Mayer, 2012), or increased spatial distance

between information sources (Bauhoff et al., 2012) leads to more memory-intensive strategies,

which provides indirect evidence for the CLT-based explanation of the split-attention effect

described above (Paas & Sweller, 2014; Sweller et al., 2011). So far, however, it has not yet been

investigated within more basic cognitive tasks whether spatial separation affects cognitive load

and task performance directly and to what extent varying the spatial distance between two

unintegrated information sources influences learners‟ perceptual and cognitive processing in

more applied settings. If spatial distance is the key factor in producing the split-attention effect, it

could be argued that given a linear relationship between the spatial distance between two

information sources and working memory load (Hardiess, Gillner, & Mallot, 2008; Inamdar &

Pomplun, 2003), the further apart two information sources are, the more likely it will be that

learners‟ working memory is overloaded and that learners experience the negative consequences

of split-attention.

The Present Study

The aim of this study was to investigate whether the split-attention effect can be

explained by the spatial distance between information sources, and whether and how this basic

cognitive phenomenon affects learning processes and outcomes in more ecologically valid

learning materials. We therefore conducted three experiments in which the distance between two

information sources was varied. Experiment 1 intended to assess distance effects using a basic

paradigm wherein participants made similarity judgments based on two pictorial information
 8

sources (cards with symbols) that were separated at different spatial distances. In Experiment 2,

for half of the cards we replaced the symbols on the card with a written description of the

information presented on the card. This enabled us to examine whether the results from

Experiment 1 would replicate when participants had to actively integrate pictorial and textual

information and laid the foundation for the next experiment in which distance effects were

investigated with more educationally relevant material. In Experiment 3, participants learned

about human brain processes from a multimedia presentation consisting of a picture with

accompanying text. Both information sources were unintelligible in isolation, so participants had

to mentally integrate the pictorial and textual information to understand the process. The picture

and text were either presented in a spatially integrated way (i.e., integrated condition) or spatially

separated in such a way that the picture and text were presented in close proximity to each other

(i.e., small-separation condition) or were separated at a larger spatial distance (i.e., large-

separation condition). Additionally, Experiment 3 applied eye-tracking methodology to

investigate participants‟ information gathering strategies. Across experiments, the general

prediction was that with greater distance between two information sources learners would show

decreased performance.

Experiment 1

Drawing on fundamental cognitive science research (e.g., Ballard et al., 1995), this first

experiment aimed to establish an effect of spatial distance when processing two pictorial

information sources. Participants judged the similarity of two cards each containing three

symbols and the spatial distance between the cards was varied. In half of the trials participants

maintained a visual pattern in working memory, leading to additional cognitive load during

information integration. We predicted that greater distance of to-be-integrated information would

 9

lead to diminished performance (Hypothesis 1), and that such diminished performance would be

more pronounced under higher a cognitive load condition (Hypothesis 2). Furthermore, we

predicted that a larger distance between information sources would lead to more demanding

working memory strategies, which would negatively affect retrieval performance of the visual

pattern that was concurrently held in working memory (Hypothesis 3).

Method

Participants and Design

Fifty-two (Mage = 21.00 years, SD = 3.57 years; 39 female) undergraduate students from

Erasmus University Rotterdam participated for course credits or a 5 euros reward. This study

was designed and conducted in accordance with the guidelines of the ethical committee of

Erasmus University Rotterdam, Department of Psychology, Education, and Child Studies. Note,

for the mixed effects regression analyses for repeated measures reaction time data (which usually

generates small effect sizes d = 0.1) Bruysbaert and Stevens (2018) recommend to use at least

1600 observations per condition for 80% power. In the current experiment we have 3120

observations per condition (1560 for hypothesis 3). A within-subjects design was employed with

the factors Cognitive Load (2 levels: load absent vs. load present) and Card Similarity (3 levels:

no similarity vs. one similarity vs. two similarities), and Distance as a continuous covariate (see

below).

Apparatus and Stimuli

The experiment was presented on an Eizo FlexScan S2243W 22 inch monitor of 47 cm x

30 cm, and resolution was set at 1920 x 1080. The task was programmed in Python (Toolbox

Pygaze; Dalmaijer, Mathôt, & van der Stigchel, 2014).

Stimuli for the card integration task consisted of a full card deck of the Wisconsin Card

Sorting Task (WCST; retrieved from Stoet, 2016). Each card had three feature dimensions
 10

(SHAPE + NUMBER + COLOR) with 4 possible levels (SHAPE: star, cross, triangle, circle;

NUMBER: 1, 2, 3, 4; COLOR: blue, yellow, green, red). The total card deck of 64 cards (4*4*4

levels) was randomly placed on an 8x8 matrix (see Figure 1; matrix = 928 x 928, pixels = 25.77

x 25.77 cm). For each unique card integration trial (60 trials) card selections were

pseudorandomly generated for each participant, such that 20 trials consisted of two cards that

were dissimilar on all dimensions (card similarity = 0), 20 trials contained selections of cards that

were similar on one dimension (e.g., similar in COLOR; card similarity = 1), and 20 trials

contained selections of cards that were similar on two dimensions (e.g., similar in COLOR and

SHAPE; card similarity = 2). Note that a similarity of three dimensions was not possible because

there were only unique cards in a deck. The selection of the cards to be compared for similarity

was signaled by two bright yellow rectangles around the selected cards (see Figure 1).

Participants responded for similarity per dimension using the response buttons „c‟(SHAPE

match), „v‟ (NUMBER match), „b‟ (COLOR match). If there was more than 1 match,

participants had to push two buttons in a row (order was irrelevant). SPACE needed to be

pressed to continue to the next trial. It is important to note that depending on card similarity (0,

1, or 2), more buttons needed to be pressed, as card similarity = 0 required only a SPACE press

(1 key press), while card similarity = 1 required a match button + a SPACE press (2 keypresses),

and card similarity = 2 required two match buttons + a SPACE press (3 keypresses).

The unique set of 60 card integration trials was presented twice, once with and once

without a secondary cognitive load task (order fully randomized). As such, card integration trials

were identical in nature (i.e., matched on card similarity type and distance) across cognitive load

conditions. Figure 3 shows a trial flow with secondary cognitive load task. The final list of 120

trials was randomized in order of presentation. The Euclidean distance (measured in pixels)
 11

between the random selections of cards was the main variable of interest. The distances could

vary between 116 pixels (3.22 cm) for directly adjacent card selections, and 1148 pixels for card

selections in opposite corners (31.88 cm).

 12

Figure 1. Example of a card integration task trial with cards selected (card similarity = 0). In the

current example participants should respond with the continue button („SPACE‟) as the selected

cards (signaled by yellow rectangles) were not similar on any of the dimensions SHAPE,

NUMBER, COLOR. In the current example the Distance was 478 pixels (12.91 cm).

In half of the trials (60 trials), a secondary visual cognitive load task was performed (see

procedure). This task is an adapted visual patterns test (Della Sala, Gray, Baddeley, & Wilson,

1997) and has been used to measure visual working memory capacity. For each trial, a random

pattern was generated of 8 black squares filling a 8x8 matrix. This pattern was presented for

memorization for 3000 ms preceding the card integration trial. For the response phase (which

was preceded by a card integration trial) participants recalled the pattern. For each trial the

response buttons were randomly chosen for each matrix cell from a list from a to z (excluding

the response buttons of the integration task, „c‟, „v‟, „b‟), such that letters were not associated

with particular locations across trials. Participants typed in the letters that corresponded with the

pattern of black squares (order irrelevant) and could proceed to the next trial by pressing SPACE.

Figure 2. Visual cognitive load task: presentation phase (left) and response phase (right).
 13

For the instruction phase, 50 practice trials were randomly created per participant. In the

first 3x10 trials participants learned to correctly respond on the integration task to single features,

namely SHAPE, NUMBER, and COLOR. For the subsequent 10 practice trials, participants

needed to respond for similarities to all features (i.e., SHAPE, NUMBER, and COLOR) at once.

In the final 10 practice trials, participants learned to also perform the visual cognitive load task

concurrently.
 14

Figure 3. Example of a single trial with a secondary cognitive load task. A trial without cognitive

load would not have the card integration task preceded/followed by a visual pattern

presentation/response (i.e., only slides 1, 3, and 4 as counted from above).

Procedure

Participants were seated in a well-lit cubicle at about 50 cm from the computer screen. To

remind participants of the response buttons for indicating similarity between selected cards, the

experimenter had labeled the response keyboard buttons „c‟, „v‟, „b‟, with stickers indicating

“SHAPE”, ”NUMBER” “COLOR”, respectively.

First, participants were instructed about the nature of the task. During this instruction

phase, participants were repeatedly prompted to ask questions to the experimenter if they did not

understand the task. Participants learned to press SPACE when cards were dissimilar on all
 15

dimensions (card similarity = 0), and press two buttons when there was a similarity on one

dimension (card similarity = 1, e.g., pressing „c‟ then “SPACE” to continue), and press three

buttons when there were two similarities (card similarity = 2, e.g., pressing „v‟ and „b‟, and then

“SPACE” to continue). After the practice phase, participants performed the 120 experimental

trials. The experiment was administered without breaks, and took about 40 minutes.

Performance and Scoring

For the card integration task, the main measures of performance were accuracy

(integration accuracy) and integration reaction time (integration RT). Integration accuracy was a

dichotomous measure of performance per dimension per trial (e.g., correct [mis]match for

SHAPE, NUMBER, and COLOR; max = 3 points). Integration reaction time was a continuous

measure of performance which entailed the time between card selection onset and participants

finalizing card integration by pressing SPACE. Note, that in the analyses we only focused on

Integration reaction time as we found that accuracy was very high (> 95%), leaving little

meaningful variance to analyze.

For the 60 trials where a secondary cognitive load task was performed the main measure

of interest was Visual Pattern Test (VPT) score (hereinafter VPT score), which was determined

by the number of correctly pressed buttons minus the number of incorrectly pressed buttons with

a maximum of 8 incorrect keypresses (maximum score of 8, and minimum score of -8). If

participants pressed a button more than once this was only scored (in)correctly once. VPT

reaction time was not of main interest because the measure of reaction time is less meaningful

when the number of buttons pressed can vary.

 16

Results

Descriptives
Table 1 shows the main descriptives of performance on the Card Integration task as well

as the performance on the VPT.

Table 1

Mean (and Standard Deviation) Reaction Time (in Milliseconds) and Integration Accuracy with

95% Percent Confidence Intervals Around the Mean (in Square Brackets) on the Card

Integration Task and the Visual Pattern Task

Integration RT Integration VPT VPT RT

Accuracy Accuracy
%correct

Load 3320 (1960) 2.95 (0.22) 4.99 (1.50) 11080 (5541)

[3251, 3388] [2.94, 2.96] [4.89, 5.09] [10896, 11264]
98.33%

No Load 3057 (2293) 2.95 (0.24)

[2977, 3138] [2.94, 2.95]
98.33%

Card similarity = 0 2076 (1689) 3.00 (0) 5.55 (2.57) 10792 (4995)
(one keypress) [2004, 2149] [3.00, 3.00] [5.39, 5.70] [10488, 11096]
100%

Card similarity = 1 3405 (2130) 2.95 (0.23) 5.01 (2.76) 11135 (5012)
(two keypresses) [3314, 3497] [2.94, 2.96] [4.85, 5.18] [10830, 11439]
94.81%

Card similarity = 2 4083 (2058) 2.89 (0.32) 4.42 (3.10) 11314 (5671)
(three keypresses) [3995, 4172] [2.88, 2.91] [4.23, 4.60] [10969, 11659]
96.33%

Overall 3189 (2137) 2.95 (0.23) 4.99 (1.50) 11080 (5541)

[3136, 3241] [2.943, 2.954] [4.89, 5.09] [10896, 11264]
98.33%

Distance
 17

r 0.022 0.015 -0.033 0.015

Note. Data before trimming.

Pre-analysis and Outliers

In total 6240 trials were run (52 participants x 60 trials x 2 conditions). Following

common practice, we excluded trials further than 3 standard deviations from the mean of the

Integration RT, 64/6240 trials (0.1%). We further restricted our main performance analyses to

only correct trials for the card integration task, and excluded all incorrect trials (5.02% of the

remaining trials).

Hypothesis 1 and 2

To assess Hypotheses 1 and 2 (i.e., greater distance of to-be-integrated information would

lead to diminished performance, which would be more pronounced when cognitive load was

present), we performed a linear mixed effects model (R version 3.4.0, nlme version 3.1-131).

Throughout, we used maximum likelihood estimation with random intercepts for participants.

See Figure 4 for a graphical overview of the Integration RT data.

In building our model, we first entered Cognitive Load as predictor of Integration RT.

This added predictive value compared to a model predicting the overall mean (BIC = 101484.20,

Chi-square change [1] = 12.19, p < .001). We further entered Card Similarity in the model, and

this improved the model as compared to the model with Cognitive Load only (BIC = 98694.37,

Chi-square change [1] = 2807.18, p < .001). Additionally, we entered Distance which did not

improve the model further (BIC = 98702.57, Chi-square change [1] = 0.478, p = .490). Finally,

we entered the interaction between Cognitive Load and Distance, which also did not improve the

model as compared to previous models (BIC = 98709.28, Chi-square change [1] = 1.97, p =

.161).
 18

The resulting model with Cognitive Load and Card similarity showed that there was a

main effect of Cognitive Load on Integration RT, b = 265.55, 95% CI = [118.23, 412.85], t(51) =

3.617, p < .001. This indicates that participants were slower to successfully integrate card stimuli

on trials with concurrent cognitive load. Cardmatch Type was a statistically significant predictor

showing an increase in Integration RT when cards were more similar (and more buttons needed

to be pressed). Going from zero to one similarity increased RT by b = 1331.70, 95% CI =

[1265.87, 1396.78], t(5760) = 39.859, p < .001, from zero similarity to two, b = 1987.514, 95%

CI = [1921.07, 2053.95], t(5760) = 58.621, p < .001. Note that in the model with Distance added

there was a positive overall relation with RT, but this was not significant.
 19

Figure 4. Effect of Distance and Cognitive Load on subsequent Integration RT. Each point

represents the mean score for all participants on that particular card distance (only successful

card integrations). Error bars represent 95%CI‟s. Note that card positions at maximum distance

concern fewer observations, and therefore CI‟s are wider and also less influential in the model.
 20

Hypothesis 3
Hypothesis 3 predicted that greater distance between information sources would

negatively affect retrieval performance on the VPT. Figure 5 shows the relation between

Distance of the cards to be integrated and the subsequent performance on the VPT (thus only for

Cognitive Load trials). We further performed a linear mixed effects model similar to the previous

analyses for Hypothesis 1 and 2, with random intercepts for participants and cognitive load

condition. Adding Distance to the model predicting VPT score resulted in a significant increase

in predictive value compared to a model predicting the overall mean, BIC = 14044.85, Chi-

square change [1] = 5.41, p = .020. Adding Card Similarity further improved the model, BIC =

14044.85, Chi-square change [1] = 93.48, p < .001. Adding an interaction between Card

Similarity and Distance did not benefit the model, BIC = 13981.51, Chi-square change [1] =

1.99, p = .370.

As predicted by Hypothesis 3, the model shows that greater Distance resulted in lower

VPT scores, b = -0.000492, 95% CI = [-0.00091, -0.0008], t(2767) = -2.33, p = .020. This means

that for every 100 pixels (ca. 2.77 cm) in distance the model predicts a decrease in performance

of 0.05. Furthermore, card similarity again affected performance such that higher similarity

(higher keypresses) resulted in lower VPT scores. Going from zero to one similarity decreased

performance by b = -0.546, 95% CI = [-0.77, -0.32], t(102) = 3.617, p < .001, from zero

similarity to two, b = -1.13, 95% CI = [-1.36, -0.90], t(102) = -9.624, p < .001.

2
Note that the b-value is so small because it expresses a relationship between 1 pixel change in

distance relative to one point change in the VPT score. The effect of 100 pixels change in

distance can be calculated by multiplying the current b-value with a 100 (i.e., 0.05 VPT point

change per 100 pixel increased distance)

 21

Figure 5. Effect of card integration distance on subsequent performance on the visual pattern test

score (VPT score). Each point represents the mean score for all participants on that particular

card distance (only successful card integrations). Error bars represent 95% CI‟s. Note, that card

positions at maximum distance concern fewer observations, and therefore CI‟s are wider and

data points are also less influential.

 22

Discussion

Based on CLT, we predicted that when information that needs to be integrated is spatially

separated, problem solvers will have to mentally carry over information to integrate it with the

spatially distant information source. In the current Experiment, we did not find that spatial

distance between information affected information integration time (Hypothesis 1 and 2).

However, we did obtain that integration of information at higher spatial distances resulted in

lower performance on a secondary visual working memory task (Hypothesis 3). This fits an

explanation assuming that integrating information sources that are more distant from each other

invites a more memory-intensive strategy that in turn leads to interference of information already

maintained in working memory, leading to lower retrieval performance (Gray & Fu, 2004). That

spatial distance affected performance on the visual cognitive load task and not the card

integration task suggests that unintegrated information can be successfully dealt with in terms of

reaction time losses through a more memory intensive strategy, but this comes at the cost of

other processes that also make use of the working memory system. The current finding that

working memory can effectively step in to solve the task in time aligns with the finding of Gray

and Fu (2004) who suggested that participants adopt a strategy that allows for the quickest

problem-solving solution at the cost of accuracy.

It is important to note that in Experiment 1, participants had to compare and contrast two

pictorial stimuli, while the split-attention effect in multimedia is generally studied with materials

consisting of a combination of text and pictures (e.g., Ayres & Sweller, 2014; Chandler &

Sweller, 1991, 1992; Florax & Ploetzner, 2010; Ginns, 2006; Mayer & Fiorella, 2014).

Therefore, we were interested whether the results of Experiment 1 would replicate when
 23

participants needed to integrate pictorial and textual information using the present paradigm.

Experiment 2 was conducted to address this.

Experiment 2

Experiment 2 was a direct replication of Experiment 1, with one small adjustment. Half

of the Wisconsin cards were substituted with a written description of the information presented

on the card it replaced. For example, instead of the picture with one red star, the three

dimensions (i.e., number, color, and object) were written on a card. In this experiment,

participants had to compare an original WCST card with a containing a written description on

the three dimensions, allowing us to test whether the results of Experiment 1 would replicate

when participants have to integrate textual and pictorial information. This experiment, and all

planned analyses were pre-registered, and all analyses, data, and materials are retrievable

(https://osf.io/ruqfk/).

Method

Participants and Design

Fifty (Mage = 20.34 years, SD = 3.00 years; 46 female) undergraduate students from

Erasmus University Rotterdam participated for course credits. The same within-subjects design

as in Experiment 1 was used.

Apparatus and Stimuli

The apparatus and stimuli were identical to Experiment 1, with two small exceptions.

First, the stimuli for the card integration task were expanded with a textual variant of each

WCST card. As a result, 128 cards were used, which where again randomly placed on an 8x8

matrix, with half of the cards pictorial, and the other half textual (see Figure 6). Second, the

experiment now comprised 45 trials, as participants in Experiment 1 indicated that the

experiment was quite taxing and boring to complete. Fifteen trials consisted of two cards that
 24

were dissimilar on all dimensions, 15 trials with consisted of cards that were similar on one

dimension, and 15 trials that consisted of cards that were similar on two dimensions. The

procedure, outlier detection, and analyses were identical to Experiment 1.

Figure 6. Example of the card-integration task used in Experiment 2, in which participants had to

compare a pictorial and textual version of the card. In the current trial participants should

indicate a match for NUMBER (1 item) and for COLOR (red).

Results

Descriptives
Main descriptives for the Card Integration task and the VPT task are provided in Table 2.
 25

Table 2
Mean (and Standard Deviation) Reaction Time (in Milliseconds) and Integration Accuracy with

95% Percent Confidence Intervals Around the Mean (in Square Brackets) on the Card

Integration Task and the Visual Pattern Task

Integration RT Integration VPT VPT RT

Accuracy Accuracy (milliseconds)
%correct

Load 5938 (3917) 2.84 (0.42) 5.11 (2.99) 13924 (6808)

[5777, 6100] [2.82, 2.85] [4.99, 5.24] [13643, 14205]
94.67%

No Load 5443 (7340) 2.84 (0.41)

[5140, 5747] [2.83, 2.86]
94.67%

Card similarity = 0 3628 (2261) 3.00 5.51 13430 (6744)

(one keypress) [3514, 3743] [2.99, 3.00] [5.31, 5.71] [12946, 13913]
100%

Card similarity = 1 6499 (8824) 2.83 4.96 14074 (6702)

(two keypresses) [6053, 6947] [2.81, 2.85] [4.74, 5.18] [13593, 14553]
94.33%

Card similarity = 2 6945 (3817) 2.69 4.88 14270 (6955)

(three keypresses) [6751, 7138] [2.67, 2.72] [4.66, 5.10] [13771, 14768]
89.67%

Overall 5691 (5887) 2.84 (0.411) 5.11 (2.99) 13924 (6808)

[5519, 5862] [283, 2.85] [4.99, 5.24] [13643, 14205]
94.67%

Distance
r -0.015 -0.013 -0.015 .020
Note. Data before trimming.
 26

Pre-analysis and Outliers

For this experiment, 4500 trials were run (50 participants x 45 trials x 2 conditions). RT

values higher or lower than 3 standard deviations from the mean of the Integration RT, 35/4500

trials (0.008%) were excluded from analyses. Similar to Experiment 1, our main performance

analyses were executed with data for correct trials, excluding RT‟s for all incorrect trials (85.6%

trials remaining).

Hypothesis 1 and 2

We performed a similar mixed linear analyses as in Experiment 1 (with participant as

random intercept). Cognitive Load was entered as predictor of Integration RT, which added

predictive value compared to a model predicting the overall mean (BIC = 71756, Chi-square

change [1] = 17.75, p < 001). Next, we entered Card Similarity to the model, improving the

model as compared to Cognitive Load only (BIC = 70642, Chi-square change [1] = 1130.61, p <

.001). Furthermore, entering Distance improved the model even further (BIC = 70593, Chi-

square change [1] = 7.72, p = .005). We also looked for possible interactions between Distance

and Card Similarity, but these models were not reliable.

The resulting model with Cognitive Load, Card similarity, and Distance showed an effect

of Cognitive Load on Integration RT, b = 795.42, 95% CI = [459.44, 1131.40], t(49) = 4.76, p <

.001, indicating slowed responses on trials with concurrent cognitive load. Cardmatch Type led

to increased Integration RT when cards were more similar, going from zero to one similarity

increased RT by b = 2404.93, 95% CI = [2221.19, 2588.65] , t(3719) = 25.65, p < .001, from

zero similarity to two, b = 3383.40, 95% CI = [3191.74, 3575.06], t(3719) = 34.59, p < .001.

Finally, and most importantly, higher distance between cards lead to higher RT‟s, b = 0.492,
 27

95% CI = [0.145, 0.838], t(3719) = 2.78, p = .006. In conclusion, higher distance between picture

and text reliably slowed down Integration RT‟s, confirming our main hypotheses.

Hypothesis 3
We predicted that increased distance between information sources would negatively

affect retrieval performance on the secondary task (VPT). We again performed a linear mixed

model, with random intercepts for participants. Adding Distance to the model did however not

add predictive value as compared to the model predicting the overall mean, BIC = 9283, Chi-

square change [1] = 0.919, p = 0.34. Adding Card Similarity to the model did improve predictive

value, BIC = 9281, Chi-square change [1] = 17.78, p < .001. Adding an interaction between Card

Similarity and Distance did not benefit the model, BIC = 9294, Chi-square change [1] = 2.03, p =

.363. In conclusion, card distance did not lead to reduced accuracy on the secondary VPT task.

Discussion
In the current experiment, distance between to-be-compared text-versus-picture cards led

to slower responses on the main task, even after controlling for the amount of keypresses

participants had to make (i.e., Card Similarity). This confirms our hypothesis that physical

distance between information is a genuine source of interference for integrating information

sources, as working-memory processes are likely to be taxed. However, in contrast to

Experiment 1 performance on the secondary visual working memory task did not reveal an effect

of distance on visual working memory. One possible explanation for this is that the integration

task in Experiment 1 was unimodal in nature (visual comparison) while in the current task it was

cross-modal (text and visual comparison). Since the the VPT task is a visual working memory

task, it is likely to especially be affected when the concurrent primary task requires a visual

comparison alone, rather than cross-modal comparison which is likely to involve more than

visual working memory capacity. That the cross-modal integration is a different process than
 28

unimodal integration is further signaled by the longer integration time for the cross-modal vs.

unimodal task. Surprisingly though, we observed no differences across experiments in the

relationship between VPT performance on cognitive load trials, while Integration performance

for those trials was relatively higher for Experiment 2, r = - 0.113, 95% CI = [-0.029, -0.102], t

(1906) = -4.956, as compared to the same relationship for Experiment 1, r = - 0.066, 95% CI =

[-0.068, -0.157], t (2922) = -3.578. Thus, as the large overlap in confidence intervals of the

correlation estimates indicate, we cannot draw any conclusions on differences in correlation

strength as to support our proposed explanation that Experiment 1 was more taxing for visual

working memory capacity than Experiment 2.

All in all, Experiment 1 and 2 have confirmed that an increase in spatial distance between

two stimuli has an effect on cognitive load and integration speed. A next step is to study whether

these effects would scale-up, and also influence learning from more complex multimedia

materials. This was investigated in Experiment 3.

Experiment 3

The aim of this experiment was to investigate whether increasing the distance between

spatially separated textual and pictorial information yields a stronger split-attention effect when

using a learning task. To this end, participants learned about human brain processes, with

materials adapted from Florax and Ploetzner (2010) consisting of a picture with accompanying

text. The text described the relevant processes also portrayed in the picture, and both sources of

information were needed to fully grasp the process of information transmission (Florax &

Ploetzner, 2010). We created three conditions: the integrated condition (i.e., the text and picture

are spatially integrated), the small separation condition (i.e., the text is segmented and the picture

is labelled, and they are separated by a small spatial distance), and the large separation condition
 29

(i.e., the text is segmented and the picture is labelled, and they are separated by a large spatial

distance). To enable investigating whether text segmenting and picture labelling could

effectively reduce the split-attention effect (cf. Florax & Ploetzner, 2010), in the spatially

separated conditions the text was segmented and the picture was labelled. Eye-tracking

methodology was applied to examine whether an increase in spatial distance leads to a more

memory-intensive strategy, as indicated by fewer transitions between the text and picture (e.g.

Ballard et al., 1995; Gray & Fu, 2004; also see Experiment 1).

We expected that learning (i.e., retention and comprehension) and processing demands

(i.e., cognitive load) in the integrated and small-separation conditions would not differ, because

the segmenting and labelling would alleviate any negative effect of split-attention (Hypothesis

1). This would replicate the results of Florax and Ploetzner (2010). Based on the literature

discussed above, we expected that learning would be more cognitively demanding (i.e., an

increased cognitive load) and learning outcomes (i.e., retention and comprehension) would suffer

in the large-separation condition compared to the small separation and integrated conditions

(Hypothesis 2). To test whether an increase in spatial distance would indeed make learning more

cognitively demanding, we asked participants to rate how much mental effort they invested in

learning the materials (as an indicator of how much cognitive load participants experienced:

Paas, 1992; Paas, Tuovinen, Tabbers, & Van Gerven, 2003). We also asked participants to rate

how much mental effort they invested during the posttest, as participants who gained more

knowledge during the learning phase should be able to attain higher test performance with less

investment of mental effort (Paas & Van Merrienboer, 1993; Van Gog & Paas, 2008).

An increase in spatial distance should lead to a more memory-intensive learning strategy,

leading to fewer transitions between the text and picture (e.g. Ballard et al., 1995; Gray & Fu,
 30

2004), and spatially integrating two mutually referencing information sources should lead to

more transitions than spatially separated information sources (cf. Holsanova, Holmberg, &

Holmqvist, 2009; Johnson & Mayer, 2012). Therefore, we expected that participants in the

integrated condition would make the most transitions between the text and the picture, followed

by participants in the small-separation condition, who in turn make more transitions than

participants in the large-separation condition (Hypothesis 3). Fewer transitions are indicative of

fewer integration of the text and picture, which can explain why an increase in spatial distance

would hamper learning (cf. Mason, Pluchino, & Tornatora, 2015, 2016). We also measured the

total fixation duration on the text and picture, as it seems that learning from text and pictures is

mostly text driven, with little to no attention to the picture (cf. Cromley et al., 2010; Hannus &

Hyönä, 1999; Schmidt-Weigand, Kohnert, & Glowalla, 2010). As an increase in distance

between the text and picture should aggravate this effect, we expected that the fixation duration

would be longest on the text and shortest on the picture in the large-separation condition,

followed by the small separation and integrated conditions (Hypothesis 4).

Method

Participants and Design

Participants were 75 undergraduate university students who participated for course credit

or a 10 a euros reward. We arrived at this sample size as we collected as many participants as

possible before the lab facilities closed down for the summer. Given that these sample sizes are

within common sample size ranges in applied educational psychology and given the research

resource constraints we decided to terminate the study for these 75 participants as well as add

additional Bayesian analyses to provide extra indications for the reliability of our data. All

participants had normal or corrected-to-normal vision. For three participants, study times
 31

indicated that they had skipped a part of the learning phase3. The data of these participants were

excluded for further analyses, resulting in a sample of 72 students (Mage = 21.68 years, SD = 2.86

years; 44 female), who were randomly assigned to one of the three conditions: integrated (n =

24), small separation, (n = 25), and large separation (n = 23).

Materials

All materials were adapted from Florax and Ploetzner (2010). They were translated from

German to English, and the distance manipulation was administered by moving the text closer to

(small separation) or further from (large separation) the picture.

Background information. Participants were presented with a short expository text on

the subject, to provide them with enough background knowledge to understand the learning

materials. This background information was presented on paper, and participants could spend as

much time reading the information as they wished. On average, it took around ten minutes.

Prior knowledge test. Participants‟ prior knowledge and understanding of the

background information was assessed with a paper-based multiple-choice test consisting of

twelve questions about neural-chemical transmissions and communication in the human nervous

system (e.g. what is a synapse?). These questions had five possible answer alternatives; four of

these alternatives could possibly be correct (e.g., the correct alternative: „The connection of two

nerve cells, which do not physically touch’), while the fifth alternative was always „I don’t

know’. Participants were encouraged not to guess, but to pick the fifth answer alternative when

they were unsure which answer alternative was correct. Participants were awarded one point

when they gave the correct answer and no points when they gave an incorrect answer, or when

3
These study times were logged by the eye tracker, from which it appeared that the recordings

did not contain the full 18 minutes that the learning phase was programmed to last.
 32

they picked „I don’t know’. Thus, they could score a maximum of twelve points on the prior

knowledge test, which took approximately five minutes.

Learning materials. The learning materials consisted of one page with text and pictures

concerning information transmission in the human nervous system, presented on the computer

screen. The information transmission process showed how different neurotransmitters are

released into the synaptic cleft, which either activate or inhibit information transfer. The text

consisted of 261 words, divided over 21 numbered segments. In the integrated condition, the text

segments were presented in close proximity of the relevant part of the picture (see Figure 7). In

the two separated conditions, the text segments were presented above the picture, while the

relevant parts of the picture were numbered in the same manner as the text segments (see Figures

8 and 9). We calculated the distance in pixels between the centre of each text segment and the

centre of the associated picture segment (see Figures 7, 8, and 9; e.g., text box 1 and picture box

1). The average text-picture distance was 701 pixels (SD = 72.91) or 19.76 cm for the large-

separation condition, 475.29 pixels (SD = 73.54) or 13.40 cm for the small-separation condition,

and 150.48 pixels (SD = 49.35) or 4.24 cm for the integrated condition.The difference between

the small-separation and large-separation condition was the largest difference possible on the

computer screen used. As in the study of Florax and Ploetzner (2010), the learning materials

were presented in a system-paced fashion and in all conditions participants had 18 minutes to

study the materials.

Figure 7. Learning material in the integrated condition with the AoI‟s as an overlay.
33
 34

Figure 8. Learning material in the small-separation condition with the AoI‟s as an overlay.
 35

Figure 9. Learning material in the large-separation condition with the AoI‟s as an overlay.

Retention and comprehension tests. Knowledge was tested directly after the learning

phase, by means of a paper-and-pencil multiple-choice test consisting of 30 questions. Twenty-

two of these questions measured retention (i.e., what potential exists over the cell membrane of a

cell that is not activated?) while eight questions required comprehension of the materials to be

answered correctly (i.e., how would the potential ratio over the membrane be if a non-activated

cell would be permeable to potassium instead of sodium?). The retention questions required

recall of the textual and pictorial information presented in the learning phase, while the

comprehension questions required participants to make inferences based on this information.

Both retention and comprehension questions had five possible answer alternatives; four of these

alternatives could possibly be correct while the fifth alternative was always „I don’t know’.

Participants were encouraged not to guess, but to pick the fifth answer alternative when they
 36

were unsure which answer was correct. Participants were awarded one point when they gave the

correct answer and no points when they gave the wrong answer, or when they picked the „I don’t

know’ answer. Thus, they could score a maximum of 22 points on the retention questions, and

eight points on the comprehension questions. Generally, participants took about 20 minutes to

answer the retention and comprehension questions.

Invested mental effort. Participants were asked to indicate how much effort they

invested in learning on a nine-point rating scale (Paas, 1992), ranging from one (extremely low

effort) to nine (extremely high effort). Moreover, participants were asked to indicate how much

effort they invested in answering the complete posttest (i.e., the retention and comprehension

questions), using the same nine-point scale.

Apparatus

The materials were presented in SMI Experiment Center (Version 3.6; SensoMotoric

Instruments), on a 22 inch monitor with a resolution of 1680 x 1050 pixels. Participants‟ eye

movements were recorded using a SMI RED 250 Mobile eye tracker (SensoMotoric Instruments)

that records binocularly at 250 Hz using SMI iView software (Version 2.8; SensoMotoric

Instruments). The eye tracking data were analyzed using BeGaze software (Version 3.7;

SensoMotoric Instruments).

Procedure

Participants were tested individually in a dedicated eye-tracking lab. First, they read the

background information, after which the prior knowledge test was administered and participants

were asked to provide their age and gender. Next, participants were seated in front of the

computer monitor with their head positioned in a chin- and forehead rest. The distance to the

monitor was approximately 60 cm. After a short introduction about the experiment, the eye
 37

tracker was calibrated using a thirteen-point calibration plus four-point validation procedure, and

participants were instructed to move as little as possible. Then, the learning phase started, for

which participants were instructed to study the materials to the best of their abilities, because

afterwards they would be tested on what they had just learnt. After the learning phase,

participants indicated how much mental effort they invested during learning, and then completed

the posttest. Finally, participants indicated how much mental effort they invested in answering

the posttest. In total, the experiment took approximately 60 minutes.

Eye-tracking Measures

For the eye tracking analyses, we first checked the accuracy of calibration. Based on this,

five participants were excluded because of inaccurate calibration (i.e., deviation from the four

validation points exceeded 1o visual angle), and three participants because the tracking ratio (i.e.,

the percentage of time for which the eye tracker actually measured the eye movements) was

below 70%. This threshold was chosen a-priori, as it leads to a high average tracking ratio,

without much data loss and has been used before (e.g., Rop, Schüler, Verkoeijen, Scheiter, &

Van Gog, 2018. For the remaining 64 participants, mean calibration accuracy was 0.48o visual

angle (SD = 0.13o), while the average tracking ratio was 95.30% (SD = 4.62%). The participants

were distributed over the conditions as follows: integrated (n = 21), small separation (n = 21),

and large separation (n = 22).

For the eye tracking analyses, we defined fixations using a 40o/s velocity threshold and a

minimal duration of 100 ms (cf. Holmqvist, Nyström, Andersson, Dewhurst, Jarodzka, & Van de

Weijer, 2011). We created an area of interest (AoI) for each segment of text (leading to 21 text

AoIs), and for each corresponding relevant part of the picture (leading to 21 picture AoIs). The

part of the screen not covered by an AoI was labelled „white space‟. The AoIs had exactly the
 38

same area-size across conditions; distance between text and picture AoIs was systematically

varied according to our experimental conditions. To measure the amount of attention the text and

picture attracted, we calculated the total fixation duration on the picture and the total fixation

duration on the text by summing the fixation duration on each individual AoI (i.e., the fixation

duration on the picture as reported in the results section is the grand total of the fixation duration

on each of the 21 picture AoIs). To measure the text-picture integration attempts (i.e., the

saccades between and within information sources), we used the number of transitions between

the different AoIs. We defined three types of transitions: text-picture transitions, which are

transitions between the text and the picture and vice versa; text-text transitions, which are

transitions between two text blocks; and picture-picture transitions, which are transitions

between two parts of the picture. We only counted the transitions between corresponding parts of

the text and the picture (i.e., a transition from text block 1 into picture part 1, or vice versa),

between consecutive text blocks (i.e., a transitions from text block 1 into text block 2, or vice

versa), and between consecutive picture parts (i.e., a transitions from picture part 1 into picture

part 2, or vice versa).

Results

All data were analyzed with one-way ANOVA‟s with Condition (small separation, large

separation, or integrated) as between-subjects factor. We used partial eta-squared and Cohen‟s d

as measures of effect size; both can be interpreted in terms of small (ηp2 ~ .01, d ~ 0.2), medium

(ηp2 ~ .06, d ~ 0.5), and large (ηp2 ~ .14, d ~ 0.8) effect sizes (Cohen, 1988). When post-hoc

follow-up tests were performed, we used a Bonferroni correction (i.e., multiplying the p-value

with the number of tests performed).

Prior knowledge
 39

Performance on the prior knowledge test (see Table 3) did not differ significantly

between conditions, F(2, 69) = 0.48, p = .619, ηp2 = .01. Hence, conditions were considered

similar in their knowledge about the topic before the learning phase.

Table 3

Mean (and Standard Deviation) Performance with 95 Percent Confidence Interval Around the

Mean (in Square Brackets) on the Pretest (max. = 8), Retention Test (max. = 22) and

Comprehension Test (max. = 8) as a Function of Condition

Large separation Small separation Integrated

Pretest 4.74 (1.63) 4.20 (1.76) 4.42 (2.26)

[4.03, 5.44] [3.48, 4.92] [3.46, 5.37]

Retention 11.48 (3.49) 12.48 (3.73) 12.04 (4.50)

[9.97, 12.99] [10.94, 14.02] [10.14, 13.94]

Comprehension 4.00 (1.98) 3.88 (1.83) 3.71 (2.05)

[3.14, 4.86] [3.12, 4.64] [2.84, 4.58]

Retention and comprehension performance

The means and standard deviations on the retention and comprehension questions for

each of the three conditions are presented in Table 3. As can be seen in this table, for both the

retention questions, F(2, 69) = 0.39, p = .679, ηp2 = .01, and the comprehension questions, F(2,

69) = 0.13, p = .876, ηp2 < .01, no significant difference on test performance was found between
 40

conditions. To provide an estimate of the evidence for this null-finding, we performed an

additional Bayesian analysis with JASP (JASP Team 2016, Version 0.8.4). Bayes Factors (BF)

were computed while operating with non-informative default priors p(M) = 0.5 (Cauchy prior of

h = .75; Rouder, Morey, Verhagen, Swagman, & Wagenmakers, 2016). Jeffrey (1961) classifies

Bayes Factors as follows: no evidence BF = 1, anecdotal evidence BF = 1-3, substantial evidence

BF = 3-10, strong BF = 10-30, very strong BF = 30-100, decisive BF >100. Note that Bayes

values will be reported hereinafter next to standard statistical measures as an extra measure of

effect strength as well as an measure of certainty to interpreting null-findings.

We obtained that there was substantial evidence for the absence of an effect for retention

(BF null-model = 6.30), such that the data were 6.301 times more likely under the null-model as

compared to the alternative model predicting an effect. We similarly obtained substantial

evidence for the null-model for comprehension (BF null-model = 7.66), such that the observed

data were 7.66 times more likely under the null-model as compared to the alternative model. To

provide another estimate of how much evidence there is for a likely absence of an effect of

Condition, we have performed another JASP Robustness analysis which provides an estimate of

the Bayes Factor‟s sensitivity to changing prior estimates. For this analysis, we compared the

effect of the integrated condition versus the large-separation condition with a Bayesian t-test and

a concomitant robustness analysis. We contrast the integrated condition and the large-separation

condition as this is the most likely contrast to detect the presence of an effect of split attention.

Figure 10 shows that a prior width change will not likely render a different conclusion for the

current dataset; even at a maximally constrained prior predicting an effect (H1) our data are not

supportive of H1, although our evidence for the H0 does become less pronounced (from

moderate evidence to anecdotal).

 41

Figure 10. Prior width effects on Bayes Factor estimation for retention and comprehension

Note. Bayesian T-tests for the integrated condition versus the large separation condition. The

figures (left panel DV = retention, right panel DV = comprehension), provide an estimate of the

sensitivity of the Bayes factors as a function of cauchy prior width changes. Higher (lower)

widths indicate higher uncertainty (or higher certainty) of the effect size assuming the alternative

hypothesis is true (H1). The gray dot indicates the default Cauchy prior width of .707, the red

dot is the prior width where there is very high certainty that there is an presence of an effect.

Figures were produced by JASP.

Invested mental effort.

In Table 4, the means and standard deviations for learners‟ self-reported invested mental

effort during the learning phase and the test phase are presented. The analyses on these scores

revealed no significant differences in invested mental effort during the learning phase, F(2, 69) =

1.75, p = .181, BF null-model = 2.23 (anecdotal evidence), ηp2 = .05, or the test phase, F(2, 69) =

0.35, p = .709, BF null-model = 6.51 (substantial evidence), ηp2 = .01.

Table 4
 42

Mean (and Standard Deviation) Invested Mental Effort (max. = 9) with 95 Percent Confidence

Interval Around the Mean (in Square Brackets) during the Learning Phase and During the Test

Phase as a Function of Condition

Large separation Small separation Integrated

Learning 6.52 (0.85) 6.76 (1.20) 6.17 (1.24)

[6.16, 6.89] [6.89, 7.00] [5.64, 6.69]

Test 6.74 (1.42) 6.68 (1.18) 6.46 (1.06)

[6.12, 7.35] [6.19, 7.17] [6.01, 6.91]

Transitions

The means and standard deviations for the number of transitions between the different

AoIs are presented in Table 5. On the picture-text transitions (which measure the integration

attempts between the text and the picture), the analysis revealed a large significant main effect of

Condition, F(2, 61) = 60.55, p < .001, BF alternative-model = 1.546 * 1012 (decisive evidence)4,

ηp2 = .67. Follow-up tests showed that participants in the integrated condition made more

picture-text transitions than participants in the small-separation condition, p < .001, d = 2.24,

95% CI = [1.43, 2.96], and the large-separation condition, p < .001, d = 3.10, 95% CI = [2.16,

3.92]. Participants in the two separated conditions did not significantly differ in their number of

picture-text transitions, p = .594, d = 0.54, 95% CI = [-0.08, 1.14]. On the text-text transitions

4
Here, we report the evidence in favor of the alternative model, which includes Condition as a

factor as the initial ANOVA revealed a large effect of condition.

 43

(which measure the integration attempts within the text), we again found a large significant main

effect of Condition, F(2, 61) = 32.09 p < .001, BF alternative-model = 2.913 * 107 (decisive

evidence)4, ηp2 = .51. Follow-up analyses showed that participants in the integrated condition

made fewer text-text transitions than participants in the small-separation condition, p < .001, d =

1.57, 95% CI = [0.85, 1.25], and the large-separation condition, p < .001, d = 2.94, 95% CI =

[2.03, 3.74]. Moreover, participants in the small-separation condition made fewer text-text

transitions than participants in the large separation condition, p = .040, d = 0.69, 95% CI = [0.06,

1.29]. Finally, on the picture-picture transitions (which measure the integration attempts within

the picture), the analysis revealed no significant differences between conditions, F(2, 64) = 0.66,

p = .522, BF null-model = 4.770 (substantial evidence), ηp2 = .02.

 44

Table 5

Mean (and Standard Deviation) Number of Transitions with 95 Percent Confidence Interval

Around the Mean (in Square Brackets) Between the Text and Picture, Text and Text, and Picture

and Picture as a Function of Condition

Large separation Small separation Integrated

Text-Picture 10.27 (4.84) 15.91 (14.12) 54.29 (19.73)

[8.13, 12.42] [9.48, 22.33] [45.30, 63.27]

Text-Text 153.41 (37.79) 121.90 (52.66) 56.00 (27.32)

[136.65, 170.16] [97.94, 145.87] [43.56, 68.44]

Picture-Picture 34.91 (20.13) 38.19 (17.00) 31.81 (16.64)

[25.98, 43.83] [30.45, 45.93] [25.98, 43.83]

Fixation Duration

The total fixation duration on the text and picture AoIs for each of the three conditions is

presented in Table 6. Please note that the fixation duration on the text and picture does not equal

the 18 minutes that participants studied the materials. The remaining was time either spend

fixating white space (which covered a considerable part of the learning materials, as we only

labeled the most relevant parts of the picture and text as an AoI; see Figure 7, 8, and 9), was not

fixated at one AoI long enough to be labelled a fixation, or was used to make saccades.

Participants in all conditions allocated an equal amount of attention towards the text, F(2, 61) =

0.60, p = .554, BF null-model = 4.987 (substantial evidence), ηp2 = .02. The amount of attention
 45

allocated towards the picture did not differ significantly between the three conditions, F(2, 64) =

2.97, p = .059, BF null-model = 0.865 (no evidence), ηp2 = .09.

Table 6

Mean (and Standard Deviation) Fixation Duration (in Seconds) with 95 Percent Confidence

Interval Around the Mean (in Square Brackets) on the Text and the Picture AoI’s as a Function

of Condition

Large separation Small separation Integrated

Text 397.31 (104.06) 398.90 (92.35) 428.22 (113.66)

[351.17, 443.45] [356.86, 440.94] [376.48, 479.95]

Picture 118.26 (40.59) 152.52 (56.37) 125.57 (47.00)

[100.03, 136.25] [126.87, 178.18] [104.17, 146.96]

Discussion Experiment 3
The present experiment examined whether an increase in spatial distance between text

and picture leads to a stronger split-attention effect in a learning task. Moreover, we aimed to

provide corroborating evidence for the finding by Florax and Ploetzner (2010) that spatial

integration of a text and picture is not necessary to counteract the split-attention effect when the

spatial distance is increased. The results show that spatially integrating text and picture is not a

prerequisite to reduce split attention: We found no differences between the integrated and the

small-separation condition on learning outcomes or cognitive load (Hypothesis 1). This

replicates the result reported by Florax and Ploetzner (2010). Moreover, an increase in the spatial

distance between text and picture did not seem to influence learning outcomes and cognitive load

as we found no differences between the two separated (i.e., small vs. large) conditions
 46

(Hypothesis 2). Therefore, it seems that the results presented in Experiment 1 and 2 do not

capitulate into cognitive benefits when learning from text and pictures.

So, we must conclude that the present results indicate that spatial distance does not

influence the occurrence of the split-attention effect during multimedia learning in the present

context. Importantly, one of the reviewers of the current paper suggested that it is still possible

that there is an indirect effect of condition on learning outcomes via an indirect effect of

condition on transitions. As such we have performed additional exploratory mediation analyses

(output retrievable at https://osf.io/vx98u/) as to ascertain whether the extent to which condition

affected picture-text transitions was related to performance. We found no indication that there

was a reliable indirect effect of condition on performance (retention or comprehension) via the

number of picture-text transitions.

However, despite concluding an absence of an effect of condition on cognitive

performance, spatially integrating the text and picture does promote text-picture integration at a

behavioral level, as participants in the integrated condition made more text-picture transitions

than participants in the two separated conditions. Unexpectedly, an increase in spatial distance

between the two spatially separated information sources did not lead to fewer text-picture

integration attempts (Hypothesis 3). This suggests that given a certain separation participants

change information gathering strategies (i.e. behavioral level), possibly indicating a non-linear

relationship between spatial separation and learning processes. That there is a drastic strategy

shift is indicated by the large effect size of d = 2.24 for the small-separation condition and d =

3.06 for the large-separation condition, meaning that, on average, participants in the integrated

condition undertake about 4 or 5 times as many integration attempts between the text and picture

than participants in the separated conditions. Participants in the separated conditions primarily
 47

made transitions between different parts of the text, undertaking about 2 or 3 times as many

integration attempts between the different parts of the text than participants in the integrated

condition. These results also align with previous studies showing that learners mostly focus on

the text in a split-attention format (e.g., Cromley, Snyder-Hogan, & Luciw-Dubas, 2010; Hannus

& Hyönä, 1999; Schmidt-Weigand et al., 2010). Regardless of this large effect of spatial distance

on participant's processing behavior as reflected in their eye-movements, this observed increase

in integrative transitions did not translate into better cognitive performance (i.e., learning

outcomes) for these learning materials.

A further finding is that although participants in the separated conditions made more

transitions within the text, they did not allocate more attention towards the text than participants

in the integrated condition, as measured by the fixation duration (Hypothesis 4). It seems that all

participants read all the text, and inspected all relevant parts of the picture. The only major

difference elicited by the spatial integration of the two sources is more integration of the text and

picture, and fewer integrations within the text. Possibly, this did not lead to differences in

learning outcomes as participants in the separated conditions already made a reasonable amount

of text-picture transitions, and further integration of text and picture was redundant for learning.

Therefore, it seems that eliminating the visual search that is often required in a split format by

signalling the corresponding parts of the text and picture is a robust way to avoid split-attention

irrespective of the distance between the text and picture.

 48

General discussion

With the current experiments we probed the grounding of the split-attention effect in a

more basic cognitive mechanism as predicted by Cognitive Load Theory (CLT). We predicted

that when information needs to be integrated but is spatially separated, participants will have to

visually decouple for longer periods (depending on distance) from one information source so as

to integrate it with the spatially distant second information source. Subsequently, an increase in

spatial distance between the two sources was expected to impose higher demands on working

memory as longer visual decoupling is required, which will impair learning processes. With

three experiments we examined 1) whether an increase in spatial distance between two to-be-

compared pictorial stimuli would increase working memory load and impair integration

performance, 2) whether such an effect would be present, and perhaps be stronger with picture-

text stimuli, and 3) whether these results would generalize to more complex multimedia learning

materials. Results show that increasing the distance between two pictorial stimuli (i.e., a

unimodal integration), hampers performance on a secondary visual-working memory task, while

leaving integration speed of the visual integration task unaffected. However, when increasing the

distance between picture-text stimuli (i.e., a cross-modal comparison) integration speed is

reduced, but spatial distance does not affect performance on the secondary visual-working

memory task. Finally, increasing the distance between text and pictures in a multimedia learning

task influences learning processes as operationalized by eye-movements that reflect integration

of information, but has no effect on perceived mental effort or learning outcomes.

Together, these results show that an effect of distance between two sources of

information (either in a visual integration task, or in a learning task) exists, although this (1) is a

small effect (Experiment 1 and 2), (2) mostly affects learning processes (Experiment 3), and (3)
 49

not always affects primary learning and problem solving outcomes (Experiment 1, 2, 3). Whether

the increase in distance interferes with the learning process seems to depend on the type of

integration that has to be performed, as witnessed by the differences between Experiment 1 (a

pictorial-pictorial integration), and Experiment 2 (a pictorial-textual integration). The current

results conceptually replicate prior research, showing that increasing the spatial distance between

two information sources leads to different information gathering strategies (i.e., making more use

of working memory; Ballard et al., 1995; Gray & Fu, 2004). We further extend these findings by

showing that such a change in processing strategy also seems to occur in a learning context,

although it does not directly influence performance on the primary learning task.

Another noteworthy finding is that in Experiment 3, fewer integration attempts did not

translate into diminished learning. While previous research shows that a higher number of

integrative transitions is indicative for better learning (e.g., Hannus & Hyönä, 1999; Johnson &

Mayer, 2012; Mason et al., 2015; 2016), such a positive relation between transitions and learning

outcomes is not always observed (Arndt, Schüler, & Scheiter, 2015; Scheiter & Eitel, 2015).

Schüler (2017) surmises that, while studying a picture or reading the text, learners are able to

retrieve previously seen information from memory as to mentally integrate the two sources

without shifting one‟s gaze. As such, learners can successfully use “knowledge-in-the-head” as

to replace “knowledge-in-the-world” (Gray & Fu, 2004). Of course, it is possible that with

increased complexity of the learning materials this “knowledge-in-the-head” strategy will

become less successful due to higher working memory demands. Future research should

therefore probe whether an effect of spatial separation does translate into diminished learning

outcomes when complexity of the learning task is increased. In the current study, complexity
 50

between Experiments 2 and 3 differed, but given that the nature of the tasks differed as well, it is

difficult to draw any conclusions regarding the role of task complexity from these experiments.

Besides the visual search for referents in the text and picture, it has been argued that

learners have to keep information active when studying spatially separated learning materials,

which imposes working memory constraints. According to time-based resource sharing models

of working memory, which have recently been introduced to cognitive load theory (e.g.,

Barrouillet, Bernardin, & Camos, 2004; Puma, Matton, Paubel, & Tricot, 2018), reduced

performance in split-attention materials reflects a time-related decay of the memory traces when

attention is switched away from information elements. Therefore, increasing the spatial distance

between information elements that need to be integrated in working memory is expected to

increase the duration that the elements need to be activated in working memory and consequently

to lead to more time-related decay of memory traces.

Together, the resources needed for visual search and information maintenance in working

memory are assumed to lead to a high extraneous working memory load, and hamper learning

(Paas & Sweller, 2014). The results of the present study show that at least spatial distance is

important for the split-attention effect, but also that both spatial distance and visual search

related processes are likely to underlie the occurrence of the split-attention effect, and these

processes are not mutually exclusive. Indeed, an increase in distance (meaning that information

has to be kept active for longer periods, while not manipulating searching processes), did not

elicit a split-attention effect in learning performance in Experiment 3. However, in Experiment 1

and 2, when participants‟ working memory was taxed by a secondary visual-working memory

task, an increase in spatial distance did lead to a split-attention effect, even though no visual

search was required (because the to-be-compared stimuli were signaled by a yellow rectangle).
 51

Therefore, both searching related information in the text and picture, as well as keeping

information active in working memory is likely to contribute to the split-attention effect.

In sum, current results indicate that increased cognitive load demands due to spatial

separation of information is a viable underlying mechanism for the split-attention effect,

supporting CLT (Sweller et al., 2011). As such, this study provides a more cognitively basic

grounding of the split-attention effect which could help to counteract the negative effects on

learning in the future. Yet, it is also clear from the results that spatial separation is likely not the

only, nor a sufficient, condition for the “split-attention effect” to occur. Finally, with this study

we hope to inspire further research that integrates basic cognitive research with more applied

instructional design effects.

Context

This study was conceived and designed when Wim Pouw and Gertjan Rop discussed how

their research backgrounds could be combined to strengthen the scientific basis for Educational

Psychological assumptions. Pouw‟s earlier work mostly concerns problem solving as informed

by embedded/embodied approaches to cognition, while Rop mostly works on instructional

design principles based on Cognitive Load Theory. Bjorn De Koning has an extensive

background in instructional design and signaling effects, and Fred Paas is an authority on all

these subjects. With this study, we wanted to combine our strengths and approach the split-

attention effect from a more fundamental viewpoint, which is underrepresented in the current

literature. This research fits well into all authors‟ respective research programs, and expands

these programs by tying the fields of embedded/embodied cognition and instructional design

together. At the moment, De Koning, Paas, and Rop are continuing this line of research in their

applied research, trying to shed more light on the learning detriments of spatial distance and
 52

cognitive integration on learning from text and pictures, while Pouw is pursuing more

fundamental topics in embodied cognitive science.

 53

Author contributions

WP & GR were main contributors of writing the introduction and discussion, BdK co-wrote the

Introduction and Discussion, and FP provided critical revisions. GR and WP designed

Experiments 1, 2, and 3, with critical guidance by FP. WP programmed and analyzed

Experiments 1 and 2, and wrote the results sections. WP wrote the method section of Experiment

1 and GR wrote the method section for Experiment 2. GR programmed and analyzed Experiment

3, and wrote the method and results sections with WP contributing to the JASP analyses. WP and

GR supervised data collection of Experiment 1, GR supervised data collection of Experiment 2

and 3.
 54

Compliance with ethical standards

Conflicts of interest: The authors have no conflicts of interest to report.

Funding: This research was funded by the Excellence Initiative grant from the Erasmus

University Rotterdam awarded to the Educational Psychology section.

Ethical Approval: This experiment was designed and conducted in accordance with the

guidelines of the ethical committee of the Department of Psychology, Education, and Child

Studies, at the Erasmus University Rotterdam. All procedures performed in studies involving

human participants were in accordance with the ethical standards of the institutional and/or

national research committee and with the 1964 Helsinki declaration and its later amendments or

comparable ethical standards.

Informed consent: Informed consent was obtained from all individual participants included in

the study.
 55

References

Arndt, J., Schüler, A., & Scheiter, K. (2015). Text-picture integration: How delayed testing

moderates recognition of pictorial information in multimedia learning. Applied Cognitive

Psychology, 29, 702-712. doi: 10.1002/acp.3154

Awh, E., & Pashler, H. (2000). Evidence for split attentional foci. Journal of Experimental

Psychology-Human Perception and Performance, 26, 834-846. doi: 10.1037/0096-

1523.26.2.834

Ayres, P., & Cierniak, G. (2012). Split-Attention Effect. In Encyclopedia of the Sciences of

Learning (pp. 3172-3175). Springer US.

Ayres, P., & Sweller, J. (2014). The split-attention principle in multimedia learning. In R. E.

Mayer (Ed.), The Cambridge handbook of multimedia learning (2nd, rev. ed.) (pp.206-

226). New York: Cambridge University Press. doi: 10.1017/CBO9781139547369.011

Baddeley, A. D. (2000). The episodic buffer: a new component of working memory? Trends in

Cognitive Sciences, 4, 417-423. doi: 10.1016/S1364-6613(00)01538-2

Ballard, D. H., Hayhoe, M. M., & Pelz, J. B. (1995). Memory representations in natural tasks.

Journal of Cognitive Neuroscience, 7, 66-80. doi: 10.1162/jocn.1995.7.1.66

Barlow. H. B. (1958). Temporal and spatial summation in human vision at different background

intensities. The Journal of Physiology, 141, 337-350. doi:

10.1113/jphysiol.1958.sp005978.

Barrouillet, P., Bernardin, S., & Camos, V. (2004). Time constraints and resource sharing in

adults‟ working memory spans. Journal of Experimental Psychology: General, 133, 83–

100. doi: 10.1037/0096-3445.133.1.83

 56

Barrouillet, P., & Camos, V. (2007). The time-based resource-sharing model of working

memory. In N. Osaka, R. H. Logie, & M. D‟Esposito (Eds.), The cognitive neuroscience

of working memory (pp. 59-80). Oxford: Oxford University Press

Bauhoff, V., Huff, M., & Schwan, S. (2012). Distance matters: Spatial contiguity effects as trade-

off between gaze-switches and memory load. Applied Cognitive Psychology, 26, 863-871.

doi: 10.1002/acp.2887

Bodemer, D., Ploetzner, R., Feuerlein, I., & Spada, H. (2004). The active integration of

information during learning with dynamic and interactive visualizations. Learning and

Instruction, 14, 325-341. doi: 10.1016/j.learninstruc.2004.06.006

Butcher, K. R. (2014). The multimedia principle. In R. E. Mayer (Ed.), The Cambridge

handbook of multimedia learning (2nd, rev. ed.) (pp.174-206). New York: Cambridge

University Press. doi: 10.1017/CBO9781139547369.010

Brysbaert, M. & Stevens, M., (2018). Power analysis and effect size in mixed effects models: A

tutorial. Journal of Cognition, 1, 1-20. doi: http://doi.org/10.5334/joc.10

Chandler, P., & Sweller, J. (1991). Cognitive load theory and the format of instruction. Cognition

and Instruction, 8, 293-332. doi: 10.1207/s1532690xci0804_2

Chandler, P., & Sweller, J. (1992). The split-attention effect as a factor in the design of

instruction. British Journal of Educational Psychology, 62, 233-246. doi: j.2044-

8279.1992.tb01017.x

Cohen, J. (1988). Statistical power analysis for the behavioral sciences (2nd ed.), New Jersey:

Lawrence Erlbaum Associates

 57

Cowan, N. (2001). The magical number 4 in short-term memory: A reconsideration of mental

storage capacity. Behavioral and Brain Sciences, 24, 87-114. doi:

10.1017/S0140525X01003922

Cromley, J. G., Snyder-Hogan, L. E., & Luciw-Dubas, U. A. (2010). Cognitive activities in

complex science text and diagram. Contemporary Educational Psychology, 35, 59-74.

doi: 10.1016/j.cedpsych.2009.10.002

Dalmaijer, E. S., Mathôt, S., & Van der Stigchel, S. (2014). PyGaze: An open-source, cross-

platform toolbox for minimal-effort programming of eyetracking experiments. Behavior

Research Methods, 46, 913-921. doi: 10.3758/s13428-013-0422-2

De Koning, B. B., & Jarodzka, H. (2017). Attention guidance strategies for supporting learning

from dynamic visualizations. In R. Lowe and R. Ploetzner (Eds.), Learning from dynamic

visualizations: Innovations in research and practice (pp. 255-278). Cham: Springer

[ISBN: 978-3-319-56202-5]

De Koning, B. B., Tabbers, H. K., Rikers, R. M. J. P., & Paas, F (2009). Towards a framework

for attention cueing in instructional animations: Guidelines for research and design.

Educational Psychology Review, 21, 113-140. doi: 10.1007%2Fs10648-009-9098-7

Della Sala S., Gray C., Baddeley, A., & Wilson, J. T. L. (1997) Visual patterns test: a test of

short-term visual recall. Thames Valley Test Company.

Florax, M., & Ploetzner, R. (2010). What contributes to the split-attention effect? The role of text

segmentation, picture labelling, and spatial proximity. Learning and Instruction, 20, 216-

224. doi: 10.1016/j.learninstruc.2009.02.021

 58

Ginns, P. (2006). Integrating information: a meta-analysis of the spatial contiguity and temporal

contiguity effects. Learning and Instruction, 16, 511-525. doi:

10.1016/j.learninstruc.2006.10.001

Gray, W. D., & Fu, W. T. (2004). Soft constraints in interactive behavior: The case of ignoring

perfect knowledge in-the-world for imperfect knowledge in-the-head. Cognitive Science,

28, 359-382. doi: 10.1016/j.cogsci.2003.12.001

Hannus, M., & Hyönä, J. (1999). Utilization of illustrations during learning of science textbook

passages among low- and high-ability children. Contemporary Educational Psychology,

24, 95-123. doi: https://doi.org/10.1006/ceps.1998.0987

Hardiess, G., Gillner, S., & Mallot, H. A. (2008). Head and eye movements and the role of

memory limitations in a visual search paradigm. Journal of Vision, 8, 1-13. doi:

10.1167/8.1.7

Haselen, G. L. V., van der Steen, J., & Frens, M. A. (2000). Copying strategies for patterns by

children and adults. Perceptual and Motor Skills, 91(2), 603 – 615. doi:

10.2466/pms.2000.91.2.603

Holmqvist, K., Nyström, M., Andersson, R., Dewhurst, R., Jarodzka, H., & Van de Weijer, J.

(2011). Eye tracking: A comprehensive guide to methods and measures. Oxford

University Press.

Holsanova, J., Holmberg, N., & Holmqvist, K. (2009). Reading information graphics: The role of

spatial contiguity and dual attentional guidance. Applied Cognitive Psychology, 23, 1215-

1226. doi: 10.1002/acp.1525

 59

Inamdar, S., & Pomplun, M. (2003). Comparative search reveals the tradeoff between eye

movements and working memory use in visual tasks. Proceedings of the Twenty-Fifth

Annual Meeting of the Cognitive Science Society, 599-604.

Jamet, E. (2014). An eye-tracking study of cueing effects in multimedia learning. Computers in

Human Behavior, 32, 47-53. doi: 10.1016/j.chb.2013.11.013

Johnson, C. I., & Mayer, R. E. (2012). An eye movement analysis of the spatial contiguity effect

in multimedia learning. Journal of Experimental Psychology: Applied, 18, 179-191. doi:

10.1037/a0026923

Mason, L., Pluchino, P., & Tornatora, M. C. (2015). Eye-movement modeling of text and picture

integration during reading: effects on processing and learning. Contemporary

Educational Psychology, 14, 172-187. doi: 10.1016/j.cedpsych.2015.01.004

Mason, L., Pluchino, P., & Tornatora, M. C. (2016). Using eye-tracking technology as an indirect

instruction tool to improve text and picture processing and learning. British Journal of

Educational Technology, 47, 1083-1095. doi: 10.1111/bjet.12271

Mayer, R. E. (Ed.) (2014). The Cambridge handbook of multimedia learning (2nd, rev. ed.). New

York: Cambridge University Press. doi: 10.1017/CBO9781139547369

Mayer, R. E., & Fiorella, L. (2014). Principles for reducing extraneous processing in multimedia

learning: Coherence, signaling, redundancy, spatial contiguity, and temporal contiguity

principles. In R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (2nd,

rev. ed.) (pp. 279-315). New York: Cambridge University Press. doi:

10.1017/CBO9781139547369.015

Mayer, R. E., Steinhoff, K., Bower, G., & Mars, R. (1995). A generative theory of textbook

design: Using annotated illustrations to foster meaningful learning of science text.

 60

Educational Technology Research and Development, 43, 31-43. doi:

10.1007/BF02300480

Miller, G. (1956). The magic number seven, plus or minus two: Some limits to our capacity for

processing information. Psychological Review, 63, 81-97. doi: 10.1037/h0043158

Ozcelik, E., Arslan-Arib, I., & Cagiltay, K. ( 2010 ). Why does signaling enhance multimedia

learning? Evidence from eye movements. Computers in Human Behavior, 26, 110-117.

doi: 10.1016/j.chb.2009.09.001

Ozcelik, E., Karakus, T., Kursun, E., & Cagiltay, K. ( 2009). An eye-tracking study of how color

coding affects multimedia learning. Computers & Education, 53, 445-453. doi:

10.1016/j.compedu.2009.03.002

Paas, F. (1992). Training strategies for attaining transfer of problem-solving skill in statistics: A

cognitive load approach. Journal of Educational Psychology, 84, 429-434. doi:

10.1037/0022-0663.84.4.429

Paas, F., & Sweller, J. (2014). Implications of cognitive load theory for multimedia learning. In

R. E. Mayer (Ed.), The Cambridge handbook of multimedia learning (2nd, rev. ed.) (pp.

27-425). New York: Cambridge University Press. doi: 10.1017/CBO9781139547369.004

Paas, F., Tuovinen, J. E., Tabbers, H., & Van Gerven, P. W. M. (2003). Cognitive load

measurement as a means to advance cognitive load theory. Educational Psychologist, 38,

63-71. doi: 10.1207/S15326985EP3801_8

Paas, F., & Van Merriënboer, J. J. G. (1993). The efficiency of instructional conditions: An

approach to combine mental effort and performance measures. Human Factors, 35, 737-

743. doi: 10.1177/001872089303500412

 61

Pouw, W. T. J. L., Van Gog, T., & Paas, F. (2014). An embedded and embodied cognition

review of instructional manipulatives. Educational Psychology Review, 26, 51-72. doi:

10.1007/s10648-014-9255-5

Puma, S., Matton, N., Paubel, P. V., & Tricot, A. (2018). Cognitive load theory and time
considerations: Using the time-based resource sharing model. Educational Psychology
Review, 30, 1199-1214. 10.1007/s10648-018-9438-6
Rop, G., Schüler, A., Verkoeijen, P. P. J. L., Scheiter, K., & Van Gog, T. (2018). Effects of task

experience and layout on learning from text and pictures with or without unnecessary

picture descriptions. Journal of Computer Assisted Learning. Advance online publication.

doi: 10.1111/jcal.12287

Rouder, J. N., Morey, R. D., Verhagen, J., Swagman, A. R., & Wagenmakers, E. J. (2016).

Bayesian analysis of factorial designs. Psychological Methods, 22, 304-321. doi:

10.1037/met0000057

Scheiter, K., & Eitel, A. (2015). Signals foster multimedia learning by supporting integration of

highlighted text and diagram elements. Learning and Instruction, 36, 11-26. doi:

10.1016/j.learninstruc.2014.11.002.

Schmidt-Weigand, F., Kohnert, A., & Glowalla, U. (2010). A closer look at split visual attention

in system- and self-paced instruction in multimedia learning. Learning and Instruction,

20, 100-110. doi: 10.1016/j.learninstruc.2009.02.011

Schnotz, W. (2014). Integrated model of text and picture comprehension. In R. E. Mayer (Ed.),

The Cambridge handbook of multimedia learning (2nd, rev. ed.) (pp.73-103). New York:

Cambridge University Press. doi: 10.1017/CBO9781139547369.006

 62

Schüler, A. (2017). Investigating gaze behaviour during processing of inconsistent text-picture

information: Evidence for text-picture integration. Learning and Instruction, 49, 218-231.

doi: 10.1016/j.learninstruc.2017.03.001

Stoet, G. (2016, August 18th). Wisconsin Card Sorting Task (WCST). Retrieved from

http://www.psytoolkit.org/experiment-library/wcst.html

Sweller, J. (1994). Cognitive load theory, learning difficulty, and instructional design. Learning

and Instruction, 4, 295-312. doi: 10.1016/0959-4752(94)90003-5

Sweller, J., Ayres, P., & Kalyuga, S. (2011). Cognitive load theory. Springer: New York. doi:

10.1007/978-1-4419-8126-4

Sweller, J., Van Merrienboer, J. J., & Paas, F. (1998). Cognitive architecture and instructional

design. Educational Psychology Review, 10(3), 251-296.

Tarmizi, R., & Sweller, J. (1988). Guidance during mathematical problem-solving. Journal of

Educational Psychology, 80, 424-436. doi: 10.1037/0022-0663.80.4.424

Van Gog, T. (2014). The signaling (or cueing) principle in multimedia learning. In R. E. Mayer

(Ed.), The Cambridge handbook of multimedia learning (2nd, rev. ed.) (pp.263-278). New

York: Cambridge University Press. doi: 10.1017/CBO9781139547369.014

Van Gog, T., & Paas, F. (2008). Instructional efficiency: Revisiting the original construct in

educational research. Educational Psychologist, 43, 16-26. doi:

10.1080/00461520701756248

Wurtz, R. H. (2008). Neuronal mechanisms of visual stability. Vision Research, 48, 2070-2089.

doi: 10.1016/j.visres.2008.03.021.

View publication stats