Multitasking on an e-commerce webpage
---
The effects of different types of information on
cognitive load moderated by product knowledge
2
Multitasking on an e-commerce webpage
---
The effects of different types of information on
cognitive load moderated by product knowledge
University of Groningen
Faculty of Economics and Business
MSc Marketing Management & Marketing Intelligence
Author:
Marten van der Zee
Date:
January 11, 2016
Address:
Turfstraat 5, 9712JK Groningen
Telephone:
+31 624774924
Email address:
M.van.der.zee.6@student.rug.nl
Student number: s2587386
3
MANAGEMENT SUMMARY
Current research underlines the importance of information presentation in e-commerce environments. By presenting information to consumers, online retailers and webshop developers try to understand the best ways of presenting information to consumers. However, presenting too much information to consumers can result in an information overload that leads to a decrease in task performance and information recall. Cognitive load theories suggest that sometimes combinations of information can be processed nearly effortlessly, but sometimes combinations of information cannot. The reason for this is that multiple tasks need mental resources, and if tasks need the same resources concurrently, task performance can decrease. This study focuses on what the effects are of presenting combinations of information via on screen text, pictures and sound on cognitive load in an e-commerce environment.
200 participants were randomly assigned to four conditions where the same 25 product specifications of a laptop were presented. Information was presented using only text (visual), text and pictures (visual- visual), text and sound (visual- audio) and pictures and sound (visual- audio). After presentation of the information, participants were asked to fill in a multiple-choice questionnaire to measure how many product specifications they remembered.
This research finds evidence that when product information is presented only via text, people remember the most product specifications compared to other conditions. In addition, comparing the visual- audio conditions to the visual- visual condition, more information is remembered when a person is subjected to the visual- audio conditions. In addition, a moderating role of product knowledge was found. As an increase in product knowledge in both the visual- visual condition and the visual- audio condition, led to a lower cognitive load in the visual- visual condition compared to the visual- audio conditions.
4
ACKNOWLEDGEMENTS
This master thesis marks the end of my Master Marketing Intelligence and Marketing Management at the University of Groningen. Writing this thesis has been very instructive and at the same time challenging as well. I want to thank several people that have helped me directly but also in an indirect way with writing this master thesis.
To start, I want to thank dr. Hans Risselada for his supervision and guidance in the process during the previous months. Furthermore, I want to thank my fellow students in my thesis group for their active participation and feedback during the group meetings.
Furthermore, I would like to thank my fellow student and dear friend Koen Wegstapel for his help and support this period. I also want to thank my girlfriend, who supported me continuously during my study at the University of Groningen.
Lastly, I want to dedicate this Master thesis to my parents Harold van der Zee and Teunie van der Zee- Lievestro. By means of my thesis I want to express my greatest gratitude for supporting me unconditionally and giving me the opportunity to study in Groningen.
Marten van der Zee
5
TABLE OF CONTENTS
MANAGEMENT SUMMARY ... 3 ACKNOWLEDGEMENTS ... 4 1. INTRODUCTION ... 6 2. THEORETICAL FRAMEWORK ... 10 2.1 Multitasking Defined ... 10 2.2 Multitasking Theories ... 11 2.3 Cognitive Load ... 122.4 Cognitive Load Theories ... 13
2.5 Product Knowledge ... 17
2.6 Moderating Role of Product Knowledge ... 18
2.7 Conceptual Model ... 19 3. RESEARCH DESIGN ... 20 3.1 Method ... 20 3.2 Procedure ... 20 3.3 Measurements ... 21 3.4 Method of Analysis ... 23 4. RESULTS ... 26 4.1 Descriptive Analysis ... 26 4.2 Scale Validation ... 26 4.3 Mean Scores ... 28
4.4 Kruskal- Wallis H test ... 28
4.5 Negative Binomial Regression ... 30
4.6 Hypothesis Testing ... 35
5. DISCUSSION ... 37
5.1 Theoretical Implications ... 37
5.2 Managerial Implications ... 38
6. LIMITATIONS AND FURTHER RESEARCH ... 40
7. REFERENCES ... 41
6
1. INTRODUCTION
Forrester Research has estimated that online retail in the United States will reach €340 billion by 2017 with an annual growth rate of 10%(Mulpuru, 2013). For this reason, Internet retailers present customers with a lot of information and strive to present proper product information on their e-commerce websites. According to Google (2013), 78% of luxury shoppers use the Internet to search for product information prior to making a purchase decision.
As people search for product information, they often perform multiple tasks at the same time. Multitasking is the ability of humans to concurrently handle the demands of multiple tasks between task switching (Rubinstein, Meyer, & Evans, 2001). According to Borst, Taatgen and Rijn (2010), exerting multiple tasks at the same time can sometimes be done nearly effortlessly (e.g. walking and talking); for other situations it can be difficult (e.g. driving and reading). Due to the emergence of the Internet, multitasking is shifting to online situations. Whilst browsing, people often perform different tasks during the search for product information. An example of an online product information search is reading a product review and simultaneously looking at product pictures. Evidence from multitasking and instructional learning literature has shown that information can be presented through different mediums (e.g. text, pictures, sound, movies), and that they are processed differently in the minds of consumers (Mayer & Moreno, 2003). Furthermore, performing multiple tasks has been shown to have an impact on the performance of tasks (Srivastava, 2013).
Having more information available does not always lead to better decisions. From a psychological perspective, combinations of multiple tasks have shown to increase cognitive load. Cognitive load occurs when tasks require too much capacity in the working memory, resulting in a restraint in learning (de Jong, 2010). Too much information can cause information overload, in which a person’s intended cognitive processing exceeds the available cognitive capacity (Wickens, 1984). From a marketing perspective, research has shown that information overload can produce uncertainty and can even lead to depression (Aljukhadar, Senecal, & Daoust 2012). Too much information can result in people remembering less of the provided information and hence, lead to more uncertainty, as consumers make purchase decisions.
7 Previous research has focused on examining what tasks can be performed concurrently and what tasks cannot. Multitasking theories have shown that each task requires certain resources and these resources have to be divided over the different tasks (Wickens, 1884; 2002). Furthermore, Salvucci and Taatgen (2008) found that tasks could have the same or different goals. If tasks have the same goal but do not require the same resources, it is likely that these tasks can be performed simultaneously. However, if multiple tasks require the same resources simultaneously, the mind can get overloaded. The reason that a person can become overloaded is that the used resources during tasks are limited. If this happens, information- or cognitive- load can occur (Salvucci & Taatgen, 2008).
From an information processing perspective, it is known that different types of information are processed differently in the mind of a consumer (Wickens, 1984; 2002). Cognitive load theories have concluded that information is processed via different channels and each channel has limited capacity to process this information (Kirschner, Ayres, & Chandler, 2011). The basic understanding of these channels is that information can be processed via the eyes, ears, or both simultaneously. When information is presented via eyes and ears simultaneously, this is called dual channel information processing.
Nowadays, multitasking research is also focusing on doing multiple tasks via personal computers or second screens (Wood et al, 2012). For example, searching for information while concurrently watching TV (Cauwenbergen, Schaap, Roy, 2014), or multitasking with laptops during a lecture (Sana, Weston & Sepeda, 2012). Most of these studies conclude that multitasking affects cognitive load, which results in a reduction of memory performance when recalling information.
8
‘’What are the effects of combinations of different types of product
information on cognitive load?’’
Furthermore, the moderating role of product knowledge is examined. According to Kim and Gupta (2009), the amount of product knowledge changes the way consumers’ process information. Therefore, it should be evident that product knowledge has an important role on the relationship between presentation of information and cognitive load. Based on the expertise reversal effect (Kalyuga & Sweller, 2004), it is expected that an increase in product knowledge leads to a higher cognitive load. This is expected when information is presented via text and pictures, compared to when information is presented via the dual channel approach.
In this study, data was collected by means of an online experiment. Product information about a fictive laptop was presented to 200 respondents. Each respondent was randomly assigned to one of four conditions. In these conditions, the same information was presented in different ways through the senses of sight and hearing (visual and auditory channels). In each condition, the same twenty-five product specifications were presented via exclusively text, pictures and text, text and sound or pictures and sound. Then, respondents were asked to answer twenty multiple-choice questions to recall the presented product specifications. Subsequently, respondents were asked to rate their general product knowledge about laptops and their multitasking frequency.
The results of the multiple-choice test were analysed by means of a Kruskal- Wallis H test and a Negative Binomial Poisson Regression. The results show that presenting information via different channels has a different impact on cognitive load. First, presenting information via solely text resulted in the lowest cognitive load. Second, data showed that, comparing information presentation between text and pictures and the dual channel approach, lower cognitive load was measured in the dual channel condition. Furthermore, this study provides evidence for a moderating role of product knowledge. However, data was found to infer the contrary to what was expected as an increase in product knowledge led to a lower cognitive load when information was presented via text and pictures simultaneously, compared to when information was presented via the dual channel approach.
9 presenting information exclusively via text resulted in the lowest cognitive load compared to other forms of presenting information. Furthermore, presenting information including sound leads to less cognitive load compared to presenting information through the use of text and pictures. Next to that, building on previous research, product knowledge has shown to positively influence recall when text and pictures are presented simultaneously compared to when visual and audio information is presented.
The managerial relevance of this paper is that managers should be aware of the importance of information presentation to optimize their e-commerce environments in order to facilitate customer information search tasks. Managers should explore using more text instead of combinations of other types of information presentations. However, if managers or practitioners make the choice to use different types of information presentation, visual information should be shown concurrent with auditory information. An increased ease of remembering information on an e-commerce website can lead to revisiting and therefore increase sales. Furthermore, managers should adapt their information presentation to the amount of product knowledge one has.
10
2. THEORETICAL FRAMEWORK
This study researches the effect of multitasking on cognitive load on an e-commerce webpage. In this chapter, existing literature is given to provide a deeper understanding of the topic: multitasking. To illustrate this, clarification is needed to show how multitasking is used in this study by providing definitions. Furthermore, multitasking theories and recent studies are explained. Next to that, it is explained how the mind processes different types of information. In time, theories are developed to understand how information processes work in the mind of a consumer. Finally, the moderating role of product knowledge is explained.
2.1
Multitasking Defined
It is relevant to introduce different multitasking definitions because in literature, multitasking has been researched in different ways. Rubinstein et al. (2001) describe multitasking as the ability of humans to simultaneously handle multiple tasks when people have to switch between more than one task. Multiple tasks often require using different mental resources, and these resources are limited.
Studies examining multitasking between different media and online multitasking use different definitions. For example, media multitasking is explained as simultaneously processing different forms of media (e.g. TV, social media, newspapers) (Pilotta, Schultz & Rist, 2004). In addition, Bardhi, Rohm, and Sultan (2010) explain media multitasking as the involvement in two or more commercial media between social media and traditional media. They argue that multitasking can occur when one is listening to music and chatting with friends, but also shopping online while reading the news and sending messages to friends. In addition, Benbunan-Fich, Adler and Mavlanova (2011) studied multitasking in an online environment. They argue that multitasking occurs when a user shifts attention when he tries to perform more than one independent, but concurrent, task on a single computer.
11 ‘’The ability to integrate, interleave, and perform multiple tasks and/or component subtasks of a larger complex task.’’
2.2
Multitasking Theories
Listening to music whilst doing homework can be controlled better than watching a television program while doing homework (Xu, Wang & David, 2016). In time, different theories are developed explaining why some tasks can be performed better concurrently compared to others.
Bottleneck theories
Early multitasking theories suggest that the mind act as a bottleneck and only one task can process at a time (Pashler, 1984)
.
This can cause interference when multiple tasks have to be performed at the same time. Central bottleneck theories suggest that, because only one task can be done at a specific time, simultaneously requesting different mental resources will result in a detrimental effect on cognitive processes (Wang et al., 2015).Multiple resource theories
12 compete for the same resource, and the performance of doing multiple tasks is not likely to decrease. On the other hand, if two tasks do ask for the same resource, it is likely that a decrease in task performance occurs (Wang et al. 2012).
Threaded Cognition Theory
A more recent theory is the threaded cognition theory (Salvucci & Taatgen, 2008; 2010). This theory combines bottleneck and multiple resource theories into one theory. Compared to the multiple resource theory, it describes a more advanced insight into the interrelatedness of multiple task resources and bottlenecks. The key assumption in the Threaded Cognition Theory is that, although several tasks can be activated concurrently, a particular resource can only be used at a single task at the time. The theory suggests that also resources in memory affect the performance of tasks. An example is that if one task needs a visual resource and another task needs a fact from memory, these tasks can be done simultaneously, since the two tasks do not compete for these resources. However, if two tasks both need a fact from memory, these tasks compete for the same resources. Hence, as long as the same particular resources have no time overlap, the theory does not predict interference. However, if two tasks need the same resources in the same time, a bottleneck is created that delays multitasking (Borst et al., 2010).
2.3
Cognitive Load
13 According to Srivastava (2013), in general, performing multiple tasks is more difficult than completing a single task since more cognitive resources are necessary. Comparisons between single task and multitasking performance found that under most conditions multitasking has a negative influence on task performance. Borst and Taatgen (2010) studied that response time increases when intermediate information has to be stored between multiple tasks compared to a single task. David, Srivastava and Kim (2013) examined that conversations during tasks negatively influence task performance. Also Wang et al. (2012) found that under time pressure, task performance on a visual-pattern task is lower in a multitasking condition compared to a single-task condition. In addition, Cauweberge et al. (2014) examined multitasking when using a second screen. The study concludes that when multitasking is done including a second screen, this leads to lower factual recall compared to a non- multitasking situation. Furthermore, multitasking via a second- screen resulted in a higher cognitive load and high loaded people remembered and recalled less news content compared to low loaded people. Most of these studies found that multitasking reduces memory performance (Srivastava, 2013).
Based on the above discussion and in line with the multiple resource theory and the threaded cognition theory, it is expected that engaging in multitasking will increase cognitive load compared to performing a single task. The reason for this is that multitasking requires multiple resource pools, which increases cognitive load (Wickens, 2002; Wang et al., 2012). For this reason, as this research is focusing on performing online information search tasks, it is expected that performing one online information search subtask leads to less cognitive load compared to performing more than one information search subtask. Therefore, following hypothesis is developed:
Hypothesis 1: A person’s cognitive load in the single-subtask condition will be lower than a person’s cognitive load in the multitasking condition.
2.4
Cognitive Load Theories
Since this study is focusing on the effect of presenting different types of information on cognitive load on an e-commerce website, it is necessary to understand how resources are allocated when information needs to be processed.
Working memory
14 working memory of a person, it is necessary to understand how the working memory works (Sweller et al., 1998). Different working memory theories and cognitive load theories are developed over the years and most cognitive load theories suggest that multiple processing channels exist to process information (Baddeley & Hitch, 1974; Mayer, 2014).
Working memory model
One example of a cognitive load theory is the working memory model of Baddeley and Hitch (1974). In their working memory model, it is assumed that people have different information processing channels for different types of information. According to the authors, the memory has three subsidiary systems: the phonological loop, the visuospatial sketchpad and the episodic buffer. The phonological loop holds speech- based information whereas the visuospatial sketchpad holds visual- based information. Both of these systems function as short-term episodic memory. If a task requires two separate domains (phonological loop as well as visuospatial sketchpad) task performance decreases less than when two task require the same domain (two times phonological loop or two times the visuospatial sketchpad) (Baddeley 2003).
Information processing model
15 memory box represents the fact that it has limited capacity. In figure 1 the cognitive theory of multimedia learning is presented (Mayer, 2014).
Research has examined cognitive load theories to a great extent. According to Lewandowski and Kobus (1993), participants who were presented with the same words concurrently in the audio and visual channel recalled more words than participants who were presented with words and pictures concurrently in only one visual channel. Also, Mousavi, Low and Sweller (1995) concluded that learning was better when students studied from visual presentation with spoken words compared to studying from the same visual presentation with written words. Furthermore, Mayer and Moreno (1998) found that spoken words and animations results in a better learning performance compared to text and animations and that they are indeed processed via separate channels. Mayer (2005) compared twenty-one experiments examining the dual channel approach. In each of these experiments, an effect was seen whereby students who received audio and visual information performed better on solving problems than did students who received on-screen text and graphics. The assumption modality principle is that words should be presented as sound (spoken words) rather than visual (written)(Moreno & Mayer, 1999). Furthermore, pictures should be presented with spoken text rather than pictures. The explanation for this assumption is that reading written texts and looking at pictures would compete for the same resources in the visual (eyes) part of the visual/pictorial channel since both are presented visually. However, spoken text is processed in the auditory part of the auditory/verbal channel and thus does not interfere with the visual channel.
The general reasoning in these studies is that, presenting two visual pieces of information exceeds the capacity of information processing resulting in a loss of information, as on the other hand the mixed presentation (visual and auditory/ dual channel) splits the information over the visual and the auditory channel (Moreno & mayer, 2003).
Whereas the aforementioned studies merely focused on multitasking via different devices, less literature exists regarding the processing of information on a single device. To the best of our
16 knowledge, one research has focused on the effects of message presentations on a single device. Bergen, Grimes and Potter (2005) examined the attention capacity and the factual recall of a story among people by focusing on the simultaneously displaying of multiple messages in news programs on a television. The different information types presented were graphics, scores, stock quotes, and textually delivered news on one screen. Furthermore, a news anchor presented spoken news. The results of the study revealed that visual complex formats drive people away from attention. Thereby leading to lower task scores compared to simple presenting formats. The findings were ascribed to the fact that visual complex information exceeded the viewer’s attention capacities, resulting in an inefficient allocation of resources.
The aforementioned research examining cognitive load and dual processes infers that under different presentation circumstances, information is processed differently. According to these theories, information formats can be processed via words and pictures and via verbal or nonverbal methods (Baddeley & Hitch, 1974; Moreno & mayer, 2003; Mayer, 2014). Furthermore, the capacity in both of these channels is limited (Mayer, 2014). Most studies about multitasking have found that it negatively influences the primary task at hand. This is mostly due to the fact that tasks require more cognitive load than a person can handle. However, when information is presented via the visual and audio channel (dual channel approach), cognitive load is expected to be lower compared to information presentation via a visual- visual format. Therefore, in line with the theories explaining the concepts of information processing, the following hypothesis is presented:
Hypothesis 2: processing two pieces of information via separate channels (visually- auditory) results in less cognitive load compared to processing two pieces of information in only one channel (visually- visually).
17 differences were found in effectiveness. However, when building more complex models, the pictorial instruction was more effective for students. These results indicate that when information is presented via pictures, cognitive load is reduced and allow respondents to solve more complex tasks compared to when information is presented via text (Plass et al., 2009). Hence, the following additional hypothesis is presented:
Hypothesis 3: Processing two pieces of information via pictures and sound results in less cognitive load compared to processing two pieces of information via written text and sound.
2.5
Product Knowledge
A crucial element in the information-processing model is the knowledge stored in a person’s long-term memory (See last box figure 1). Product knowledge is defined as the ability to perform product- related tasks successfully (Jacobi et al., 1986). In literature, product knowledge and expertise are often used interchangeably. The amount of product knowledge a person has affects information processing and impacts attention devoted to information (De Bont & Schoormans, 1995; Alba & Hutchinson, 1987).
18 Hypothesis 4: Product knowledge positively affects cognitive load.
2.6
Moderating Role of Product Knowledge
From instructional design literature, it is found that in different conditions, learning can be effective for novices and ineffective for knowledgeable consumers. This is called the expertise reversal effect (Kalyuga & Sweller, 2004). The effect states that interactions exist between how effective certain instructional designs are for different levels of knowledge of a learner. On the one hand, cognitive load can occur when novice learners have to process new information in working memory. On the other hand, more knowledgeable learners who already have knowledge about something can be forced to process unnecessary information, since they already have the knowledge. Learners then are distracted when they have to learn from instructions. Hence, these processes could result in an increase in cognitive load that leads to a reduction of resources in working memory (Kalyuga, 2013; Lee, Kalyuga, & Wales., 2014). According to Kalyuga and Sweller (2004), evidence for the expertise reversal effect is provided when extra textual information is shown in understanding diagrams to high knowledgeable people. The reason for this finding was that when people are knowledgeable, additional text was not required. But, when these explanations were shown, processing of this unnecessary information increased cognitive load.
Thus, it is expected that when product knowledge increases for people in the picture and text condition (visual- visual), cognitive load increases. The explanation for this expectation is that when too much visual information is provided that the person already has knowledge about, cognitive load increases. On the other hand, it is expected that when product knowledge increases in the dual channel condition (visual- auditory), people experience lower cognitive load. The reason is that, in this condition, less visual information is provided that the person already has knowledge about, which leads to a lower cognitive load. Hence, the following hypothesis is presented:
Hypothesis 5: An increase in product knowledge in both the text and pictures condition and the dual
19
2.7
Conceptual Model
Based on the existing literature and proposed hypotheses, a visualization of expected relationships is shown in figure 2.
H5 H4
H1,2,3
Figure 2: Conceptual framework Information
presentation formats
Product knowledge
20
3. RESEARCH DESIGN
This section discusses the design of this research. The chapter will give an overview of the method and procedure of the study. Furthermore, different measures and constructs are elaborated. Finally, the plan of analysis is explained.
3.1
Method
In order to examine the relationship between information presentation in different formats and cognitive load, a between subject design in a virtual online shopping environment is conducted. 200 respondents were randomly assigned to four conditions of information about laptop specifications. The conditions were programmed using the online survey tool Qualtrics. The survey was distributed via e-mail and Facebook. Furthermore, respondents were recruited in the University Library during their lunch break and the Zernike Campus.
3.2
Procedure
Each condition comprised the same product specifications about a virtual laptop. A laptop computer was chosen since it is a high involvement product and it is likely that the search for information is apparent to take place (Mourali, Laroche & Pons, 2005). The laptop specifications were fictional to control for potential relationships to one particular type of laptop. Before each condition was shown, participants were told that they were considering buying a laptop. To make an informed choice, they visited an e-commerce website about a laptop. The participants were told to take the information with them so that, at a later moment, they could discuss the specific information with a friend. After respondents were informed, one of the four conditions was presented. In figure 3 an overview of the design and different conditions is shown and in appendix A an overview of the conditions.
Condition 1- Text only
In this condition, product specifications were presented solely via written text. Thus, information in this condition was presented via one channel only, namely the visual channel.
21 Condition 2- Text & pictures
In the second condition, product specifications were presented via written text and pictures simultaneously. The pictures were presented next to the written text. Therefore, written text and pictures had to be processed at the same time. The same information as in the previous condition was presented here via two ways, namely pictures and written text. The pictures existed of exactly the same information as in the written text. So, the visual channel in figure 1 has to process two forms of information presentation (visual- visual).
Condition 3 & 4- dual channel
The third and fourth condition represent both a dual channel approach. In the third condition, product specifications had to be processed via written text and sound simultaneously. In the fourth condition, the same information had to be processed via pictures and sound. Together, these two conditions form a dual channel approach, since information is simultaneously presented in the visual and audio channel (figure 1). First, participants in the sound conditions were asked to test their sound system, earplugs or headphones. This was done by a female voice that asked the respondents to test their sound system. After participants tested their sound and clicked through, the same voice read the product specifications out loud in both conditions. Whilst listening to the spoken text, the respondent read the text or watched pictures. The length of the spoken text determined the presentation time of each condition. This resulted in a presentation time of 90 seconds, also for the non-sound conditions.
To program the spoken text into the online survey, the audio fragments were recorded and uploaded online via SoundCloud.com. Via this music stream website, an embedded link was copied into two of the four conditions. To let the audio play automatically when the product page was presented, the embedded URL had to be adjusted with the code ‘automatic=1’. However, at this point the music link (indicated by a large play button) was shown very large above the product page and participants had the option to stop the music. To disable that participants could stop the spoken text before the voice was done, the link width and height were set equally to five. After the adjustment the play button was hidden.
3.3
Measurements
Recall22 load (Hong et al., 2004; Paas & van Merrienboer, 1994). Furthermore, recall is related to infer how successful the e-commerce website is in presenting product information to users. In addition, the ability of recalling product information on webpages can be used as an indicator of how successful a website is in presenting product information to consumers (Hong et al., 2004). In this study, aided recall is used. With aided recall a retrieval cue will be provided to let participants recall information (Lwin & Morrin, 2012). Aided recall is easy to interpret, namely an answer is wrong or right. The aided recall test consisted of 20 multiple-choice questions (e.g. How many inches was the laptop screen?). The multiple-choice questions required the respondents to choose the right answer out of a range of four (Schüler et al., 2012). The choices were analysed by counting the number of wrong answers. The number of wrong answers is chosen since many wrong answers could mean that people experience high cognitive load, and therefore make more mistakes. According to Brunken, Plass and Leutner (2003), measuring respondents’ performance is the best direct measure to infer that people are affected by cognitive load.
Pre-test
The four conditions were extensively pretested. First, the amount of product specifications had to be set for the study. Several lists of different amounts of product specifications were developed and tested on reading speed. Next, several tests of different amounts and difficulties were developed and pretested by 20 students. Finally, a pre-test was executed incorporating the four conditions. The scores on the test in different conditions yielded expected score differences. At the end, the information resulted in 25 product specifications and 20 multiple-choice questions. These multiple- choice questions were asked directly after the 90 seconds of information presentation in each condition.
Interaction effects
For the interaction effect of product knowledge, the 7-item scale of Laroche, Bergeron and Goutaland (2003) was adopted (alpha=0.85). The participants had to answer 7 statements about their specific knowledge of laptops (e.g. I use this product often in my life). The questions were asked by using a 7- point Likert scale and ranged from 1, strongly disagree to 7, strongly agree.
Other measures
23 knowledge about the participants in the survey, other demographic questions were asked (e.g. gender, age, province, profession, and education).
3.4
Method of Analysis
Construct validityThe multitasking and product knowledge measurement scales used in the analysis need to be validated. This validation is needed to infer that the measurements truly represent the constructs used in this study. To investigate the reliability of the constructs, Cronbach’s Alpha will be used. According to Peterson (1994), a value of .7 is seen as the minimum.
Mean scores
In this study, the number of wrong answers is used as a dependent variable to infer cognitive load. First, to detect differences per condition, mean recall scores per condition will be measured and model free evidence is provided.
Kruskal- Wallis H test
Second, to infer that the mean scores and deviations are significantly different per category, an ANOVA test can be performed. To perform an ANOVA test, different assumptions have to be adhered to. According to Malhotra (2010), (1) the different conditions of the independent variable should be fixed, (2) the error term of the dependent variable should have a normal distribution with a zero mean and a constant variance, (3) the error terms should be uncorrelated. However, it is expected that the distribution of the error terms do not follow a normal distribution since the dependent variable is count data. If the data is indeed not normally distributed, the Kruskal- Wallis H test will be performed. This test is a rank based nonparametric test that can be used to determine if there are statistically significant differences between the sum ranks of three or more groups. In the test, the smallest number of the dependent variable gets a rank of 1. The largest number gets a rank of N. where N is the total number of values in all groups. Furthermore, one assumption of performing the Kruskal- Wallis H test is that the distributions in the conditions have the same distribution shape. The following equation estimates the Kruskal- Wallis H test statistic:
(1)𝐻 = (N − 1) = ∑𝑖=1𝑔𝑁𝑖 (𝑟𝑖− 𝑟)2 ∑ ∑ (𝑟𝑖𝑗− 𝑟)2
𝑗=1𝑛𝑖 𝑖=1
𝑔
24 Where 𝑁𝑖 is the number of observations in group i, 𝑟𝑖𝑗 is the rank (among all observations) of observation j from group i and N is the total number of observations across all conditions.
Negative Binomial Poisson Regression
Third, since the dependent variable is count data, it can be expected that it follows a Poisson distribution. To infer that the data indeed follows a Poisson distribution, several assumptions have to be met: (1) the data is not normally distributed, (2) the number of wrong answers cannot produce non-negative counts. If the expected count could entail negative values, ordinary least squares would be more suited for analysis (Gardner, Mulvey & Shaw, 1995). (3) The estimated lambda (λ) is lower than 10. Lambda is measured by dividing all observations by the total number of respondents, and is referred to as the mean. (4) The mean equals the variance. However, if the fourth assumption does not hold, the estimated Poisson model could be overdispersed. In the case of overdispersion, the heterogeneity across respondents is larger than expected by the Poisson model. Therefore, the heterogeneity is caused by variance in the Poisson model and other unexplained variance. To account for this unexplained variance, the Negative Binomial Poisson Regression could be more suited to fit the data. Under the Poisson, the Lambda (λ) is assumed to be constant or homogeneous over different conditions. However, by defining a specific distribution for λ, heterogeneity within the conditions is now allowed.
The negative binomial distribution is a member of the so-called Poisson mixture models (Karlis & Xekalaki, 2005). The equation of the Negative Binomial Poisson model is adopted from Lawless (1987) and is written as follows:
(2)𝑃𝑟 = (𝑌 = 𝑦|𝑥) =(𝑦 + 𝑎−1) 𝑦! (𝑎−1) ( 𝑎𝜇(𝑥) 1 + 𝑎𝜇(𝑥)) 𝑦 ( 1 1 + 𝑎𝜇(𝑥)) 𝑎−1
Equation 2: Econometric form of the Negative Binomial Poisson Regression model
Where Y is the dependent variable the number of wrong answers, which is a count. y= 0,1,2,3… number of wrong answers, 𝑥 represents the explanatory variables, 𝜇 is the mean given by Lambda (λ) and 𝑎 is the dispersion parameter that cannot become zero (𝑎 ≥0).
25 Model fit
To determine the model fit, the Likelihood ratio test will be performed and information criteria will be estimated. The different information criteria are estimated as follows:
(3) AIC= -2 ln(L) + 2K (4) BIC=-2 ln(L) + ln(N) K (5) CAIC=-2 ln(L) + (ln(N)+1) K
Equation 3,4,5: Econometric forms of the information criteria
Where L is the likelihood measure, K the number of parameters and N the number of observations in the dataset.
In general, the model that yields scores closer to zero is the model with the best fit. However, for large sample sizes, the Bayesian Information Criterion (BIC) and the Consistent Akaike Information Criterion (CIAC) are the best suited. According to Andews and Currim (2003), the reason for this is that these criteria add a larger penalty parameter, compared to the AIC score, when more variables are used in the model.
Hypothesis testing
26
4. RESULTS
In this part, the results of the analyses are presented. First, descriptive statistics are shown. Then, the scales used in this research are validated. Subsequently, mean scores per condition will be explained. After that, a Kruskal- Wallis H test will be performed. Finally, a Negative Binomial Poisson Regression is estimated including the moderation effect of product knowledge.
4.1
Descriptive Analysis
Data was randomly collected over 200 participants. Of these participants, 162 completed the online survey. From these 162 participants, missing values were treated by making use of casewise deletion. Casewise deletion is applied when any respondent with a missing value is deleted from the analysis (Malhotra, 2010). In total, three respondents did not fill in all questions; hence these cases were excluded from the analyses. The original sample of 200 respondents resulted in a dataset of 159 respondents suited for analyses. Furthermore, out of range data was not found in the dataset.
Table 1 below provides a summary of the respondents and demographics. 67.9% of the participants was male (N=108) and 32.1% was female (N=51). Furthermore, 28.8% of all participants followed a HBO education (N=45) and 57.2% a WO education (N=91). Of the total respondents, 55.3% was currently participating in a study (N=55.3) and 36,5% were employed (N=58). The average age of respondents was 25 years old and most respondents were located in the category of 18-24 years (N=89, 55.6%).
Table 1: Overview demographics
4.2
Scale Validation
To measure the internal consistency of the used constructs, reliability tests were performed. Product knowledge and multitasking were measured using standard items from validated scales (Laroche et al., 2003; David et al., 2013). Since in both scales one question was reversed, this question had to be turned the other way around. To test for reliability of the constructs, a Cronbach’s Alfa was performed. In table 2, an overview is shown including the different Cronbach’s Alpha measures. For product knowledge, the Cronbach’s Alfa (α=.86) was almost similar as in the Laroche et al. (2003) study (α=.85). Hence, no questions were deleted. Next to the reliability test of product knowledge,
Gender Freq. Perc. School Freq Perc Age Freq Perc Profession Freq Perc
Male 108 67.9 Secondary
school 15 9.4 18-24 89 55.6
In-paid
employment 58 36.5 Female 51 32.1
VMBO/ MAVO 8 5.0 24-34 64 40.0 Entrepreneur 9 5.7
HBO 45 28.3 45-54 1 0.6 Student 88 55.3
WO 91 57.2 >55 5 3.1 Unemployed 4 2.5
27 the 6-item construct of multitasking was tested for reliability (David et al., 2013) (α=0.65). The initial Cronbach’s Alpha resulted in a score of (α=0.645). After deleting question 1, the reliability of the construct improved to (α=0.695). This score improved the initial Alpha in the David et al. (2013) research (α=0.65). Since the new 5-item scale showed to be more reliable than the 6-item construct and almost resulted in an Alpha of 0.7, one item was deleted from the analysis.
Questions Alpha Source
Product knowledge
1. I frequently use/consume this product in my life 2. I have much information about this product 3. I have much experience purchasing this product 4. In general, my knowledge of this product is strong 5. I am informed about this product 6. Compared to friends, my knowledge of this product is strong
7. Compared to experts my knowledge of this product is strong
α=.857 Laroche et al., (2003)
Multitasking
1. I Focus on only one thing at a time
2. I multitask with multiple media at the same time
3. I have multiple browser windows/tabs open at the same time
4. I have multiple applications/programs open at the same time 5. I have multiple emails open at the same time
6. I have multiple chat windows open at the same time
α=
.645 Alpha if item 1 deleted:.695
David et al., (2013)
Table 2: Scale validation product knowledge and multitasking
To get a better understanding of the data, the product knowledge and multitasking experience scale scores were averaged per respondent. As the table shows, almost 40% of the respondents consider themselves experienced regarding laptops (answers 5-7). Almost 60% of the respondents have moderate or low knowledge about laptops (answer 1-4). Next to that, most respondents score a 4 on multitasking. Hence, it can be concluded that the majority of respondents have experience with multitasking. The variables are graphically shown in figure 4.
Figure 4: Product knowledge and multitasking experience score 0 20 40 60 2 3 4 5 6 7 N u m b e r o f r e sp o n d e n ts
Mean product knowledge score
Product knowledge
0 50 100 150 2 3 4 5 N u m b e r o f r e sp o n d e n ts28
4.3
Mean Scores
First, decriptives of the respondents per condition and dependent variable are presented in table 3.
Table 3: Descriptives per condition
As table 3 shows, the number of respondents per condition were respectively 44 in the text only condition, 42 in the text and pictures condition, 37 in the text and sound condition and 36 in the Pictures and sound condition. One would expect the number of respondents to be equally divided over the different conditions. However, differences between groups are due to the fact that not every respondent finished the survey, and data had to be
cleaned prior to the analysis. Next to that, the mean number of wrong answers per condition are shown. The scores range from 5.6 wrong answers in the text only condition and 8.5 wrong answers in the text and pictures condition. The mean scores are in line with what is expected, since in the text only condition, the least amount of wrong answers are given. Second, the text and pictures condition resulted in the highest mean wrong answers. Third, both sound conditions
result in lower mean recall scores compared to the text and picture condition. Hence, this also confirms expectations, namely that when information is presented including sound, respondents remember information better, and thus make fewer mistakes. An overview of the scores per condition is shown in figure 5.
4.4
Kruskal- Wallis H test
ANOVA tests could not be performed since the standardized residuals of the number of wrong answers were not normally distributed. To test for normality, the Kolmogorov- Smirnov (K-S) and Shapiro- Wilk test (S-W) were performed and presented in appendix B. The tests showed a non-significant result for the K-S test (P=.059) and a non-significant result for the S-W test (P=.006). According to Razali and Wah (2011), the S-W test compared to the K-S test has more power. Therefore, the S-W test is interpreted and it can be concluded that residuals of the number of wrong answers are not normally distributed. Since a violation of the normality assumption was detected, nonparametric analysis of variance was performed. One assumption of performing the Kruskal- Wallis H test is that
Text only Text and pictures
Text and sound
Pictures and sound
Number of respondents per condition 44 42 37 36
Recall (mean number of wrong answers) 5.6 8.5 7.5 6.9
Figure 5: Overview mean recall scores
0 2 4 6 8 10
29 the distributions in the conditions have the same shape. Plots for the different conditions were developed and showed to be distributed the same per condition.
In this research, two conditions were aimed at processing information including sound. First, differences between these two groups were analysed. In table 4 the comparisons of the mean ranks are presented. Since the number of wrong answers do not provide significant differences (χ2 =.972 P=.324), the data does not show evidence to support the third hypothesis. Hence, Processing two pieces of information via pictures and sound did not result in less wrong answers compared to processing two pieces of information via written text and sound.
Dual channels separated Chi- square P. value Mean rank
Text and sound versus pictures and sound .972 .324 39.41 vs 34.53 Table 4: Overview Kruskal- Wallis H test dual channel
Because these conditions both can be seen as a dual channel approach, they were grouped for subsequent analysis. The grouped conditions are now referred to as ‘dual channel’. The scores in table 3 strengthen this choice given that the recall scores in both conditions are more or less the same (7.5 vs. 6.9) and these scores are located between the only text condition and text and pictures condition (5.6 and 8.5). Also, the difference in mean values shows not to be extensively large (4.12).
Results Kruskal- Wallis H test
The results of the Kruskal- Wallis H test on the number of wrong answers are presented in table 5. The Kruskal- Wallis H test showed a statistically significant difference in the rank scores of wrong answers between the different presentation conditions (χ2= 12.707 P=.002). The mean rank number of wrong answer score of 60.75 in the text
only condition, 95.33 in the text and pictures condition and 82.78 in the dual channel condition. The significant value of the test means that the different conditions have different ranks of sums. The highest mean rank of the text and picture condition explains that in that condition, the highest numbers of wrong answers are ranked.
However, the results of the K-W test did not show any significant differences between the three groups separately. To show differences between specific groups, post hoc tests were performed. To perform any of the post hoc tests, each condition was compared separately in the K-W test by selecting cases. The results of the post hoc test are shown in table 6.
Table 5: Kruskal- Wallis H test for number of wrong answers
Chi-square Df. P. value Number of
wrong answers 12.707 2 .002
Conditions N Mean rank
Text only 44 60.75
Text and
pictures 42 95.33
30 Chi square P. value Mean rank1
1: Text only versus text and pictures 12.121 .000*** 34.39 vs. 50.05
2: Text only versus dual channel 6.344 .012** 48.86 vs. 65.11
3: Dual channel versus text and pictures 2.007 .157 54.67 vs. 63.79
Table 6: Overview post hoc tests Kruskal- Wallis H test (*** significant at the 1% level ** significant at the 5% level)
The tests showed significant differences in the mean rank scores between the text only vs. text and pictures condition (χ2 =12.121 P=.000). Furthermore, comparing the mean rank of the text and picture condition (50.05) with the mean rank in the text only condition (34.39), yields that in the text only condition lower values are ranked. This means that in the text only condition, less wrong answers were given compared to the text and pictures condition. Next to that, comparing the text only condition versus the dual channel condition results in significant differences (χ2 =6.344 P=.012). Furthermore, comparing the mean rank scores of the text only condition (48.86) with the dual channel condition (65.11), concludes that, when information is processed via dual channels, scores are ranked higher. This means that in the dual channel condition, more wrong answers were given compared to the text only condition. At last, comparing the dual channel condition with the text and pictures condition did not result in significant differences in rank scores (χ2 =2.007 P=.157). However, the mean rank of the dual channel condition (54.67) seems to be lower compared to the text and pictures condition (63.79). Thus, since the two conditions are not significantly different, the differences cannot be interpreted.
4.5
Negative Binomial Regression
Given that the number of wrong answers is count data, it could mean that the number of wrong answers follows a Poisson distribution. A Poisson regression models the probability of a given number of events occurring in a fixed time interval. One assumption used to infer that the data is Poisson distributed is that the standardized residuals
should not follow a normal distribution. In subsection 4.4, the normality test of the unstandardized residuals were elaborated. Since the S-W test resulted in significance, the null hypothesis that the data is normally distributed can be rejected. Therefore it is assumed that the data is non-normally distributed. Additional proof is shown in figure 6 where the observed residuals do not closely fit the expected normal values.
1
31 However, this does not directly imply that the data follows a Poisson distribution. Another assumption is that the expected count cannot produce non-negative number of wrong answers. Since respondents could not have minus one answer wrong, the number of wrong answers cannot be lower than zero. Hence, this assumption also holds.
Furthermore, the estimated mean should be lower than ten (λ<10). The mean is estimated by the total number wrong answers (k= 1126) in the data divided by the number of respondents (n=159) in the data. If the estimated mean would result in λ>10, one can assume that the data follows a normal distribution. In this research, the estimated mean is (λ=7,08) which is lower than 10. Hence, it is not likely that the distribution follows a normal distribution.
Another underlying assumption of the Poisson model is that the mean and the variance are equal. To test if this assumption holds, the Lagrange Multiplier Test was performed. The outcomes of this test determine if the Negative Binomial Poisson distribution ancillary parameter is 0. Table 7 below shows that the mean is smaller than the variance and so results in overdispersion.
Parameter <1 Parameter >0 Non-directional
Ancillary Parameter 1.000 .000 .000
Table 7: Lagrange multiplier test
Estimation Negative Binomial Poisson Regression
The likelihood ratio chi-square provides a test of the overall model comparing this model to a model without any predictors (the "null" model). In the omnibus test, the P- value indicates if the model is a significant improvement compared to the null model. The test shows a significant change in the chi-square value, thus a significant improvement compared to the null model (P=.000). In table 8 the results of the Omnibus test are presented.
In table 9, the different estimated models are shown. The first estimated model provides estimates of the different conditions only. Next, in the second model, the conditions and the main effect of product knowledge are shown. Third, the theoretical developed model including the interaction effects is presented. Furthermore, model fit scores are shown.
Omnibus test Likelihood ratio Df. P.value
59.734 5 .000***
32
Variables Model 1 (conditions only) Model 2 (conditions + product knowledge)
Model 3 (conditions + interaction effect)
Exp (B) P. value Exp (B) P. value Exp (B) P. value
Intercept 7.151 .000*** 7.149 .000*** 7.115 .000***
Text only .760 .000*** .760 .000*** .764 .001***
Text and pictures 1.215 .004*** 1.216 .004** 1.216 .004**
Dual channel 1 1 . 1 .
Product knowledge 1.005 .838 1.060 .150
Text only*Prkn .961 .554
Text and pictures*Prkn .880 .036**
Dual channel* Prkn 1 .
Model fit Model 1 Model 2 Model 3
Df. 2 3 5
Log likelihood -453,252 -453,231 -450,990
AIC 912.504 914.462 913.980
BIC 921.711 926.638 932.393
CAIC 924.711 930.738 938.393
Model 1: Conditions only
As table 9 shows, each condition is significantly different compared to the dual channel approach. The model shows for the different conditions significant values; text only (P=.000, Exp (B)=.760) and for text and pictures (P=.004 Exp (B)=1.215). Hence, if information is presented via text only, the expected count (λ) of number of wrong answers is .760 times smaller compared to presenting information via dual channels. Therefore, 24% less wrong answers are given. Second, if information is presented via text and pictures compared to the dual channel approach, the expected count (λ) of number of wrong answers is 1.215 times larger. Hence, an increase of 21,5% wrong answers is expected when information is presented via the dual channel approach compared to information via text and pictures.
Model 2: Conditions and product knowledge
As in the first model, the model shows for the different conditions almost the same significant values; text only (P=.000, Exp (B)=.760) and for text and pictures (P=.004 Exp (B)=1.216). Thus, when information is presented via text only, the expected count (λ) of number of wrong answers is .760 times smaller compared to presenting information via dual channels. Hence, when information is presented via text only compared to the dual channel approach, 24% less wrong answers are given. Second, if information is presented via text and pictures compared to the dual channel approach, the expected count (λ) of number of wrong answers is 1.216 times larger. Hence, an increase of 21,6% wrong answers is expected when information is presented via the dual channel approach compared
33 to information via text and pictures. It should be noted that these probabilities are only the case when product knowledge is incorporated in the model. Furthermore, product knowledge was mean centered. The main effect of product knowledge has no significant influence on the number of wrong answers (P=.838). So, an increase in product knowledge compared to the mean value does not result in significant differences on the number of wrong answers given.
Comparing the parameter estimates of the first and second model, the exponential betas do not show large differences. Hence, this means that including product knowledge as a main effect does not lead to large estimated differences in the estimates of the separate conditions.
Model 3: Conditions and interaction effect of product knowledge
In line with the first and second model, both individual conditions show significant values; text only (P=.001, Exp (B)=.764) and for text and pictures (P=.004 Exp (B)=1.216). Thus, when information is presented via solely text, the expected count (λ) of number of wrong answers is .764 times smaller compared to presenting information via dual channels. Hence, when information is presented using text only compared to the dual channel approach, 23,6% less wrong answers are given. Second, if information is presented via text and pictures compared to the dual channel approach, the expected count (λ) of number of wrong answers is 1.216 times larger. Hence, an increase of 21,6% wrong answers is expected when information is presented via the dual channel approach compared to information via text and pictures. It should be noted that these expected counts are estimated when product knowledge is mean centered and interaction effects are incorporated in the model. Next, in line with the second model, product knowledge has no significant influence on the number of wrong answers (P=.150).
Interaction
34 In figure 7 the combined effects of the conditions including interactions are shown. The effects are calculated by using the estimated betas for each variable. In appendix C an overview of the betas are shown. The counts are estimated as follows:
𝑒𝑥𝑝𝑒𝑐𝑡𝑒𝑑 𝑐𝑜𝑢𝑛𝑡 = 𝑒𝑐+(𝛽𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛𝑖)+(𝛽𝑝∗𝑃𝑠𝑐𝑜𝑟𝑒)+(𝛽𝑖𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛∗𝑃𝑠𝑐𝑜𝑟𝑒)
Where 𝑒 is the exponent, 𝑐 is the constant, 𝛽𝑐𝑜𝑛𝑑𝑖𝑡𝑖𝑜𝑛𝑖 is the estimated parameter (𝛽) per condition,
𝛽𝑝 is the estimated parameter (𝛽) of the main effect product knowledge, 𝑃𝑠𝑐𝑜𝑟𝑒 is the product knowledge score relative to the mean (range from -3 to 3) and 𝛽𝑖𝑛𝑡𝑒𝑟𝑎𝑐𝑡𝑖𝑜𝑛 is the estimated interaction parameter (𝛽). In figure 7 the effects of product knowledge for the different conditions are shown.
The figure shows a decrease in number of wrong answers in the text and pictures condition when product knowledge increases. On the other hand, when product knowledge increases in the dual channel condition, the expected number of wrong answers also increases. In addition, comparison of the text and pictures with the dual channel condition shows that if product knowledge is <1.5 relative to the mean (0), presenting information via the dual channel approach leads to a lower amount of wrong answers. However, if product knowledge becomes >1.5, more wrong answers are expected in the dual channel condition.
Furthermore, the text only condition shows to be constant over time for different values of product knowledge. However, since no significant interaction effect was found between the text only and dual channel condition, this result should carefully be interpreted.
0 2 4 6 8 10 12 -3 -2 -1 0 1 2 3 Co u n t Product knowledge
Expected wrong number of answers
Text & pictures Textonly Dual- channel
35 Model fit
First, to compare the goodness of fit between the estimated models, the likelihood ratio test was performed. The test is based on the likelihood, which indicates how well the model fits the data. The difference between the log likelihood values between the three models did not result in larger 𝑥2 values compared to the critical values. Hence, the three models showed not to be significantly different in terms of goodness of fit.
Second, for each model, information criteria were estimated to determine the model fit. The model fit scores are shown in table 9. In line with the likelihood ratio tests, the interaction model does not show a better model fit. The fact that the information criteria do not show a better fit is likely to be caused by the fact that not all interaction effects were significant in model one. However, since an interaction effect was hypothesized and main effects of the conditions were significant in each model, the interaction model was used to test hypotheses.
4.6
Hypothesis Testing
Main effectsBased on the three estimated Negative Binomial Poisson models, it is consistently found that the text only condition resulted in lower number of wrong answers compared to the dual channel condition and text and picture condition. Hence, based on this outcome, it can be inferred that the text only condition results in the least cognitive load. Therefore, hypothesis 1 is supported, since a person’s cognitive load in the single-subtask condition was lower than a person’s cognitive load in the multitasking condition. Next to that, data was found to infer that the probability of occurrence of wrong number of answers is lower in the visual-visual condition compared to the visual- audio condition. Therefore, processing two pieces of information via dual channels (visually- auditory) result in less cognitive load compared to processing two pieces of information in only one channel (visually- visually). Hence, hypothesis 2 is supported. Furthermore, data was not found to support hypothesis 4 as product knowledge as a main effect did not result in differences in the number of wrong answers.
Interaction effects
36 and pictures condition and the dual channel condition, led to a lower cognitive load in the text and pictures condition compared to the dual channel condition.
In table 10, an overview of the hypotheses is presented.
Hypothesis Supported
1 A person’s cognitive load in the single-subtask condition will be lower than a person’s
cognitive load in the multitasking condition. Yes
2 Processing two pieces of information via separate channels (visually- auditory) results in less cognitive load compared to processing two pieces of information in only one channel (visually- visually).
Yes
3 Processing two pieces of information via pictures and sound results in less cognitive
load compared to processing two pieces of information via written text and sound. No
2
4 When a person has more product knowledge, cognitive load is expected to decrease. No
5 An increase in product knowledge in both the text and pictures condition and the dual
channel condition, leads to a higher cognitive load in the text and pictures condition
compared to the dual channel condition.
No
Table 10: overview hypotheses
2
37
5. DISCUSSION
The purpose of this study was to examine the effect of different information formats on cognitive load among different levels of product knowledge in an e-commerce setting. In terms of theoretical contributions, this study is the first to look at literature on multitasking, cognitive theories and bring this into a marketing setting. Cognitive load was inferred by measuring information recall. Whilst this study builds on previous literature on information processing by studying different ways of presenting the same information (Sweller, 2003; Mayer, 2014), it also explores the influences of product knowledge (Kalyuga, 2013; Lee et al., 2014). In this section, first the theoretical implications are given. Second, the practical implications are discussed.
5.1
Theoretical Implications
Main effects
First, this study examined the differences between conducting multiple subtasks compared to one subtask. Presenting solely text versus multitasking situations resulted in better recall scores in the text only condition. The Reasoning for this is that resources that were otherwise allocated to one task, now had to be divided over different tasks. The fewer resources available to process more than one piece of information resulted in a higher cognitive load. This is in line with expectations and prior research of Wickens (2002), Borst et al. (2010) and Wang et al., 2012). This study has shown that less incorrect answers were given when information was presented only via text.
Second, this research provides evidence that processing two pieces of information on an e-commerce website via separate channels (visually- auditory) results in less cognitive load compared to processing two pieces of information in only one channel (visually- visually). Evidence was found that the number of wrong answers was lower when information was presented simultaneously with sound compared to the condition where pictures and text were shown concurrently. This finding suggests that using sound during the search for product information results in a more efficient information search and effective recall of information. This result is in line with research of Baddeley and Hitch (1974), Sweller (2003) and Mayer (2014).
