Keywords

1 Introduction

Since the release of the first iPad in 2010, billions of tablet computers have been sold all over the worldFootnote 1. From playing games at home to presenting slideshows at work, tablet computers are now used for a variety of purposes and in a wide range of contexts [1]. Although many had predicted that tablet PCs sales would overtake the traditional PCs market by 2018Footnote 2, reality has not met expectations with a steady decrease in tablet PC sales since 2015Footnote 3. According to a recent Gartner’s reportFootnote 4, mobile device adoption in the workplace has been lower than expected with workers still massively relying on desktop computer or laptop for their computing needs. For workers, tablet PCs are still too limited when it comes to content creation such as writing and editing document, designing presentation or using spreadsheetFootnote 5. To fight tablet PCs decline, manufacturers have started marketing new devices such as “detachables” and “convertibles” that combine the portability and the touchscreen of a tablet PC with the power of a laptop. A detachable tablet is a device that is composed of a touchscreen and a detachable keyboard (such as Microsoft Surface Pro) while a convertible is a notebook with a touchscreen that can be flipped or folded depending on the model (Lenovo IdeaPad…). With those new devices, users are given the choice between several interaction modes as they can use the regular mouse/keyboard input, the touchscreen only, or a combination of both. However, despite the ever-increasing penetration of these new devices in firmsFootnote 6, only few studies have examined whether the interaction mode can impact user’s performance.

Ostrowski [2] have shown that touch-based interactions could help to reduce the overall load on mental resources compared to mouse and keyboard, thus improving performance on the touch-screen devices. Wang [3] claimed that using gestures mimics common real world gestures and avoids the split attention effect due to the use of the mouse and keyboard. Nevertheless, other researchers did not find any learning gains when tablet PCs are used instead of traditional computers [2]. Since the emergence of touch-screen devices, many studies have compared physical keyboards and on-screen keyboards [4,5,6]. Varcholik et al. [6] found that using a virtual on-screen keyboard to encode text on a device decreases the speed and increases ty** errors when compared to physical keyboards. Furthermore, participants reported being less satisfied with the on-screen keyboard. According to Findlater and Wobbrock [5], users devoted more attention to ty** due to looking at the on-screen keyboard to reduce their error rate. Along with a touch-based interaction mode and an onscreen-keyboard, tablet PCs usually feature smaller displays than laptops with screen sizes ranging from 7 to 11 in. for a “slate” tablet and from 11 to 13 in. for most of the detachables and convertibles. This raises concerns for potential detrimental effects on user’s performance due to increase scrolling, which has been reported to lower text comprehension [7] and task efficiency [8].

In this paper, we investigate the impact of the interaction mode on information search performance, since it has been reported to be one of the most frequently performed activities on IT devices at work [9, 10]. Browsing the web is also a commonly performed activities on tablet PC with around 50% of users who engage in such activity on their touch-screen devices [11]. This paper aims to tackle some of the ergonomic issues raised by the increasing use of touch-screen computers at work by exploring the impact of the interaction modes on users’ behavior while they are seeking information on the internet. In the next section, we will briefly introduce our main theoretical framework and our hypotheses will be outlined. Then, we will describe the experiment conducted and the related method and material. Results will be presented and discussed thereafter.

2 Theoretical Background

2.1 Information Seeking Models

Many models have been developed to describe how people search information [12,13,14,15]. Marchionini’s model [16] describes information searching activity as an eight-step problem-solving process: recognize and accept the problem, define and understand the problem, select the source, formulate the query, execute the query, examine the results, extract information and reflect, iterate or stop the research. While Marchionini was the first to emphasize the iterative nature of the search process in digital environments, his model lacked a clear description of how people interact with the systems and overlooked the role of cognitive and psychological factors [17]. Building on Marchionini’s model, Sharit and colleagues [18] developed a model specifically related to search engines that emphasizes the role of working memory and web-related knowledge on information-seeking (IS) performance. Their model describes IS as a multi-stage process “whereby the problem-solver’s knowledge and other mental representations are manipulated to achieve a goal” [18, p. 3]. First, the problem is identified and broken into several goals and subgoals. Then, search terms are generated and translated into queries in a search engine. Webpages matching the initial goals are visited and the information retrieved is then compared to these goals. The information-seeker stops the research when he/she judges that a sufficient amount of information has been found or when he/she gives up. This process is highly iterative as the mental representations can be modified at each step of the process, thereby leading to a reformulation of the problem and the corresponding goals.

This model also points out the role of web-related knowledge that can impact the formulation of the queries as well as the relevance judgment of the webpages retrieved. Furthermore, the authors emphasize the importance of working memory in kee** the user oriented throughout the IS process, especially as the difficulty of the problem-solving task increases. While searching on the web, the user has to maintain the goals in working memory, set and apply a search strategy, as well as process and store information. Both the amount of information that has to be maintained in working memory and the remaining resources available to perform the search can impact the performance in IS [18]. In line with the objectives of this study, it is worthwhile questioning whether seeking information using a specific device could impact the user’s mental resources in terms of working memory and thereby, their performance in IS activity.

2.2 Cognitive Load

Cognitive theories constitute a relevant framework to explore how touch-screen computers can impact users’ mental resources. Originally developed in the field of instructional design, Cognitive Load Theory [19] and her sister theory, the Cognitive Theory of Learning with Media [20], are based on the same assumptions. Firstly, that the human cognitive architecture is made up of two dependent structures: (a) a working memory that actively selects and integrates incoming information with prior knowledge and mental models; (b) a long-term memory that stores a potentially unlimited amount of information in the form of schemata. Secondly that the working memory has limited resources, and activities that demand attention compete for these resources. Cognitive load refers to this limited capacity and is described as the mental cost of a specific task, for a particular individual in a given context. According to these theories, cognitive load is multifactorial and can be divided into three types of cognitive load: intrinsic, extraneous and germane loads. Intrinsic load is related to the material to be learnt (the interactivity of the elements) and the user’s prior knowledge. It is usually related to the task demand and task difficulty. Extraneous load deals with the mental resources devoted to elements that are not directly related to the task at hand, and is often linked to the presentation format. Germane load is described as the mental resources required by schemata acquisition and automation in working memory. Since the working memory capacity is limited, an increase of extraneous load is correlated to a decrease of germane load and, consequently, results in lower learning outcomes or performance.

Different methods allow the measurement of cognitive load: subjective ratings, performance-based measures and psychophysiological measures (see [21] for a review of these methods). Subjective ratings assume that an individual is able to report how much mental effort has invested in the task undertaken. Researchers have used multidimensional scales to discriminate between the different load factors and distinguish intrinsic, extraneous and germane load [22]. Psychophysiological measures have several advantages over subjective ratings. Since they are based on bodily responses, they allow measurement at a high rate and with a high degree of sensitivity [23]. Moreover, they do not require an overt response by the subjects and, in this way, are considered as a direct and “objective” way to infer mental activity. In this paper, we focused on both types of measures, which will be detailed in the Method section.

Given the aforementioned elements, we hypothesize that browsing the web on a touch-screen computer is likely to increase webpage scrolling which will result in a higher demand upon cognitive resources and a rise in extraneous load. Moreover, keying errors are more frequent on onscreen keyboards than on their physical equivalents and therefore, require much more attentional resources.

H1. Using a tablet computer will generate a higher level of extraneous load compared to a laptop

As pointed out by Sharit and colleagues [18], information search on the web requires the employment of a substantial amount of mental resources. The seeker has to keep the goal of the search in working memory while making decisions about the search results’ relevance until the task is successfully achieved. Assuming that using touch-screen computers requires more mental resources and that the working memory has a very limited capacity, we hypothesize that the use of a touch-screen computer hinders user performance in information search when compared to searching on a laptop.

H2. Laptop users will outperform touch-screen computers users in online search.

3 Method

3.1 Sample

Thirty-six students from Arizona State University took part in this experiment. There were 17 males and 19 females with a mean age of 21.36 years (SD = 1.86). They had normal or corrected-to-normal vision and did not report any attention disorder. Participants were given a compensation of 25 dollars for participating in this study and signed an IRB-approved consent form at the beginning of the study.

3.2 Control Variables

Numerous studies have indicated that previous knowledge or experience of the internet, experience of the search engines and of the devices used can impact the efficacy of the search process [24]. Familiarity with the devices was controlled for by asking the participants to report the number of hours spent on each device on a weekly basis (for both laptop and tablet PC). Device self-efficacy was measured via a scale adapted from Compeau and Higgins [25] and modified according to the device to be used. The scale was made of 6 items referring to the ability of the user to use the device (skill based) such as “I could complete a new task using a laptop/tablet PC”. Cronbach’s alpha were .765 for the laptop scale and .772 for the tablet scale respectively. Information search perceived self-efficacy was measured thanks to a 13-item scale (Cronbach’s alpha was .847) developed by Bronstein [26]. Items such as “I can usually find the information I need” assessed users’ perceived ability to perform information search. Participants were asked to rate the extent to which they agreed/disagreed with the statements on a 5-point Likert scale ranging from “Strongly disagree” to “Strongly agree”.

Working memory capacity has been indicated as related to performance in multitasking on the web [27] and information search tasks [28]. In this study, we used the working memory capacity test developed by Oswald et al. [29]. Due to time constraints, only operation span was evaluated for each participant. However, operation span has been specifically related to performance in IS tasks [30]. Sets of arithmetic operations were presented to participants who had to judge whether the problem was true or false (e.g. 25 + 5 = 30). After each problem, a letter was presented and participants had to recall all letters in the right order after the set of equations. Set size ranged from 3 to 7 and there were three administrations for each set size. The test took 15 min on average. Independent samples t-tests performed on the control variables did not yield any significant difference across groups (touch-screen versus laptop) nor gender.

3.3 Measurement Tools, Information Search Tasks and Dependent Variables

Cognitive Load.

Extraneous load represents the load related to the mental resources devoted to the elements irrelevant to the learning tasks [19]. In the usability context, extraneous load has been identified as the load generated by poor layout and usability issues [31] or disorientation in a hypermedia [32]. Based on the usability heuristics [33], extraneous load was measured with 4 items referring to: (1) the ease of navigation on the Web (“Navigating between pages was a problem”); (2) the amount of information displayed (“the amount of information displayed on the screen was appropriate”); (3) the ease of interaction (“It was easy to interact with the device”); (4) the perceived disorientation (“I could identify easily on what page I was and where to go next”). As perceived extraneous load was measured after each information search task, the scores were averaged over the 5 tasks. Cronbach’s alpha coefficients were computed to estimate both the internal consistency of the scales and the reliability of the repeated measurements. All coefficients showed good to very good reliability.

Physiological measures were collected to obtain a more sensitive, reliable and unbiased measure of extraneous load. Techniques such as EEG, galvanic skin response, heart rate variability or fMRI have been used to assess physiological proxies of mental activities but eye-related data is still amongst the most widely explored. Eye-tracking is a non-intrusive and cost-effective method that allows one to observe the user’s attention allocation through the gaze position. Multiple indexes can be collected such as gaze point position, number and frequencies of eye blinks, duration of fixation and saccades (see [34, 35] for a comprehensive review of these measures). The pupillary response is one of the most popular measure of cognitive load due to the fact that the pupillary reflex is under the control of the autonomous nervous system and cannot voluntarily be controlled by the subject. Relationship between an increase of cognitive load and pupil diameter has been described in various contexts (varying from such as simple cognitive tasks [36, 37] to naval simulators [38]; driving [39]; e-learning [40]; e-shop** [41] and an AI web-based tool [42].

As pointed out by Chen and colleagues, “there can be no single measure that can be recommended as the definitive measure of mental load” (p. 35) [43]. In an attempt to provide reliable measures of cognitive load, along with the eye-related data, cortical brain activity was also measured using an electroencephalograph (EEG). The use of EEG has been validated to measure mental workload and task engagement in various environments and tasks (see [44,45,46] for more details). Berka et al. [44] used quadratic discriminant functional analysis on the decontaminated EEG signal to create a task engagement index with four levels: high engagement, low engagement, distraction and sleep onset. A probability of cognitive state is then provided for each second. According to Berka et al. [45], EEG-engagement is related to visual scanning, sustained attention and information gathering. While some researchers have successfully used this index to show a decrease in cognitive load when using a specific software [47] others did not find any difference in engagement in subjects using video games. In order to find an index that is more suited to the context of web-based information search, we followed the procedure described by Sénécal et al. [48]. They calculated a cognitive load odds based on the EEG-engagement metric using the equation below:

$$ {{\text{Cognitive}}\,\,{\text{Load}}\,\,{\text{Odds}} = \frac{{{\text{Probability}}\,{\text{of}}\,{\text{high}}\,{\text{engagement}} + {\text{Probability}}\,{\text{of}}\,{\text{low}}\,{\text{engagement}}}}{{{\text{Probability}}\,{\text{of}}\,{\text{distraction}} + \,{\text{Probability}}\,{\text{of}}\,{\text{sleep}}\,{\text{onset}}}}} $$

They successfully used this index to discriminate across variations of the user’s cognitive load when visiting or revisiting websites and the impact of website familiarity on cognitive load. Pupil size variation from the baseline and EEG – Cognitive Load Odds were averaged over the tasks in order to obtain a measure for the whole experimental run. For the EEG-Cognitive Load Odds (EEG-CLO), a logarithmic function was applied before averaging over the different tasks. In this experiment, pupil size was considered as a proxy measure of overall load and EEG-CLO index was considered as a measure of extraneous load.

Information Search Tasks.

The search tasks were designed following the simulated work situations principles [49] which means that the task describes the source of information need and the environment, in order to make clear the objectives of the search to the seeker. Effort has been made to generate information search (IS) tasks that were interesting for our population of interest and that motivated them to perform realistic searches. Based on previous research on IS tasks [30, 50], we defined two types of tasks: fact finding and information gathering. The fact finding tasks required the retrieval of one or more specific pieces of information while information gathering tasks were less defined and required the collection of several pieces of information on a given topic. There were three structures for FF tasks: Simple, Hierarchical and Parallel. For FF Simple tasks, one piece of information had to be retrieved to achieve the goal. Hierarchical concerned a deeper search as they had to retrieve a number of pieces of information about the same topic but located at different levels of depth. Conversely, Parallel search dealt with multiple concepts that exist at the same level of depth (breadth search). Table 1 summarizes the tasks and their characteristics.

Table 1. Instructions, type and structure of the tasks
Full size table

Performance Metrics.

Performance was defined as a second order concept that encompasses three dimensions: (1) search outcomes; (2) search effort and (3) depth of search. For each information search task, participants were asked to locate one or several pieces of information using the web-based search engine Google. Once the information was located, they had to clipboard the relevant piece(s) of information and bookmark the corresponding page(s). Search outcome was based on two metrics: the number of elements retrieved weighted by their relevancy and (2) the number of pages bookmarked weighted by their relevancy. All bookmarks and elements in the clipboard were judged as task-relevant. Search effort refers to the effort spent to achieve the goal of the task and is compounded of two elements: the time spent for each bookmark and the time per page. Finally, depth of search reflects the motivation of the participant to seek for information. Since they were allowed an indefinite period to perform the search, one could say that a longer, more motivated search was more likely to end up in a more complete answer, more bookmarks and thus, better performance. Accordingly, we defined depth of search as the time spent on the task, the number of queries formulated and the number of webpages visited. Scores were added up across all tasks in order to compute an overall score for each device.

3.4 Apparatus

In the laptop condition, we used a Dell Latitude E6540 with an Intel Core i5 quad processor and 8 GB RAM. The monitor was a 15.4 in large screen with a resolution of 1920 × 1080. No external mouse or keyboard was provided in order to make the participants use the integrated keyboard and touchpad. For the tablet PC condition, we used a Lenovo Yoga 13 tablet PC with an Intel Core i7 and 8 GB RAM. The monitor was 13.3 in large with a resolution of 1600 × 900. The tablet PC was folded so that participants were only provided with the touch-screen to interact with the device. Both computers were running Windows 8 and used Google Chrome as their default internet browser.

EEG measurement involves detecting the fluctuation of voltage potential generated by large groups of neurons in the brain. The EEG signal was acquired using the B-Alert X10 device from Advanced Brain Monitoring. This device allows us to remotely acquire data of brain activity using a wireless set of nine electrodes (F3, F4, Fz, C3, C4, Cz, P3, P4 and POz) sampled at 256 Hz. B-Alert proprietary software uses an artifact decontamination algorithm to account for electrical interferences, eye blinks or motor movements. It computes two composite metrics: a cognitive workload index and a cognitive state index (see [44] for the technical details).

Eye tracking data was collected via a Tobii X2-60 remote eye tracker sampled at 60 Hz. For the laptop condition, the tracker was attached below the screen in order to track the eyes even when the eye-lids were partially closed. For the touch-screen condition, both the tablet PC and the tracker were fixed on a tailor-made mobile device stand that allowed the users to use the touchscreen without putting their hands/arms in front of the tracker. iMotions software version 5.7 was used to display the questionnaires and tasks and allowed the integration and synchronization of the EEG and eye tracking signals.

3.5 Protocol

The study was conducted in a testing room at Arizona State University. The participants were first escorted into the lab by the examiner, then they were seated and given an informed consent form. Once the form was viewed and signed, the study procedure was explained and the participant had to complete the computer-based working memory capacity test. Then, the EEG headset was placed onto the participant’s head, an impedance check was run before starting the ABM calibration procedure. This calibration takes around 15 min and consisted of 3 computerized tasks in which visual (colorful shapes) and audio stimuli had to be identified.

Then, the eye tracking was calibrated and participants had to gaze at a blank screen for a 10-s baseline. Before moving to the information search tasks, they had to fill out an online questionnaire including questions about their age, gender and the control scales (cf. supra). Instructions regarding the information search tasks were provided and a video played showing a trial task and explaining what was expected to perform the task. The actual tasks were then displayed in a randomized order. For each task, they were asked to retrieve one or more pieces of information, clipboard the relevant items and bookmark the corresponding pages until they considered they had provided enough relevant elements to the problem presented. No time limit was set to complete the tasks. After each search task, they had to report their level of cognitive load for the task just performed. Once all tasks were completed, participants were thanked for their participation, compensated, and given information on obtaining the results of the study. The whole experimental run took around 2 h.

4 Results

All data transformations were performed on the open-source software R [51] and statistical analyses using the statistical suite SPSS. The assumption of normality was met for all the variables included in our analyses. Independent samples t-test were used to analyze the differences in scores across the two conditions. Table 2 provides the mean and standard deviations for the main performance metrics as well as the t-statistics, corresponding p-values and effect sizes.

Table 2. Means (SD) and results from the t-tests for the time on task, depth of search, efficacy and search effort
Full size table

Results showed that, when averaging time across all tasks, there was no statistically significant difference in time between the two devices. Regarding performance in information search tasks, our results indicated that the depth of search and efficacy were lower for those who used a touch-screen PC compared to those who used a laptop. Fewer webpages (M = 2.74) were visited and fewer queries were formulated (M = 3.14) on the touch-screen condition compared to the laptop condition. The t-test analysis showing significant differences for both the number of webpages visited (p = .005, d = 0.96) and the number of queries generated (p = .004, d = 1.04). Similarly, the participants who used the touch-screen PC obtained a worse overall efficacy with almost half the number of elements copied into the clipboard (M = 1.43) than in the laptop condition (M = 3.46). Those on the touch-screen PC bookmarked 25% less webpages (M 6.88) than those on the laptop (M = 9.29). Regarding the search effort, the results indicated that touch-screen group spent significantly more time for each web page bookmarked than their counterparts on the laptop (p = .021, d = 1.10). A significant difference was also found when looking at the average time spent per task. As showed in Table 2, 122.35 s were spent on each webpage visited on a touch-screen PC against only 73.08 s when visited on a laptop (p < .001, d = 1.32).

We hypothesized that these differences in performance might be related to variations of cognitive load. Table 2 shows that using the touch-screen PC led to a higher level of perceived extraneous load (M = 3.32) compared to a laptop (M = 2.55), this difference being statistically significant (p = .008, d = 0.96).

Along with this self-reported measure, cognitive load was measured by two physiological metrics: pupil size variation from the baseline and EEG cognitive load (EEG-CLO). Linear Mixed Model analyses were performed using the MIXED procedure in SPSS (see [52] for an introduction to those models). The device used was defined as fixed factors. To account for within-subject variability related to physiological data and the non-independence of repeated measures, a random intercept was defined along with the fixed effect. Relationships amongst the residuals were directly estimated in the model using a Variance Component covariance matrix. Intra Class Correlation coefficients were computed by dividing the unexplained variance of the residuals by the variance of the random factor, which gives the percentage of variance explained by the random effect.

Table 3 shows a decrease in pupil diameter for those who used the tablet PC compared to those who searched on the laptop. Regarding EEG data, results showed a significant increase of EEG-CLO for the touch-screen condition (M = 1.87) than for the laptop condition (M = 1.04); F = 5.27, p = .028. Finally. ICC coefficients indicated that for the two measures, the random intercept factor explained the majority of the residual variance with coefficients of 74.9% and 87.7%. Pearson’s product-moment correlation analyses were run to assess the relationships amongst loads but no significant correlations were found.

Table 3. Means (SD) and linear mixed model analyses of the pupil size variation and the EEG-based index of cognitive load
Full size table

5 Discussion

In an effort to explore the impact of the use of touch-screen PCs, we investigated whether seeking information on the web differs when performed on a touch-screen computer versus on a laptop. A lab-based study was carried out to shed light on this issue as well as to explore the underlying mechanisms that could explain the impact of a specific device on information search (IS) performance. We hypothesized that using a tablet PC would decrease performance in IS compared to a laptop, since a touch-screen PC requires more mental resources to be used, thereby reducing those available to perform the search.

The results of this study provide evidence that using a touch-screen computer to seek information on the web results in a drop in performance in IS (H2 is supported). Fewer queries were formulated, fewer webpages were visited and therefore, participants retrieved fewer relevant elements to achieve the goals of the tasks. They either did not find any relevant element or they provided a less complete answer than those who had sought information on a laptop. It must be noted that all elements saved into the clipboard and all pages bookmarked were task-relevant, regardless of the device used.

While fewer webpages were visited, the touch-screen PC group spent much more time on each webpage which could indicate that reading required more effort. Consequently, those users might have engaged in a more in-depth exploration of the webpage, trying not to jump from one page to another too quickly so as to minimize the mental effort in navigating the web. Also, of interest is that one would expect that the search should have taken longer as a result of the strain generated by the device, yet our results showed no difference on the average time spent per task. A possible explanation for this might be that participants interrupted the search process earlier because they were frustrated or disoriented in the task. As described by Sharit’s model [18], the ability to stay oriented is a critical determinant of the success of the search process.

With regard the Cognitive Load Theory framework [19], we suggested that this drop in performance might be caused by an increase of extraneous load. Our findings are in line with this claim since participants who used the touch-screen PC reported a higher level of perceived extraneous load (H1 is supported) than those on the laptop. It means that they experienced difficulties with navigating between pages and interacting with the device and they judged that the amount of information displayed on the screen was not optimal. It can therefore be assumed that the smaller screen, the on-screen keyboard and the touch-based interaction mode specific to those type of devices may impose a burden on the user’s mental resources. Interestingly, we obtained these results despite using a medium-sized touch-screen PC (13.3”) although some devices feature much smaller screens (below 10”). Using such devices would have certainly resulted in more marked differences.

Along with these self-reported measures, we gathered physiological data in order to get more sensitive, reliable and unbiased measures of cognitive load. While there is still much debate with respect to the multimodal measures of cognitive load [43], our findings provide some interesting food for thought. First, it is noteworthy that our proxy measures were not correlated together which indicates that there are not likely to be related to the same underlying factors. As expected, a negative pupil size variation was found in the touch-screen condition. Given that pupil size variation has been related to information processing [34] in working memory, this could account for the decrease in both depth of search and efficacy. Our EEG-based measure of cognitive load (EEG-CLO) was higher in the touch-screen condition, which supports our assumption that using to search information on the internet leads to a higher level of extraneous load. However, there is still a lack of understanding about the specific elements that generate this burden. It is possible that averaging the EEG-CLO index over the whole experimental run has hidden subtle relevant variations of the mental load during the tasks. For instance, previously an EEG study has shown differences in mental workload when people select links or read text online [53], events that occur and must be analyzed at a higher temporal resolution. Further investigation should be undertaken to precisely identify what are the main elements users are struggling with when searching the web on a touch-screen device.

5.1 Limits

The present research has several limitations that should be noted. First, as with most lab-based studies, our sample size was relatively small (due to the duration of the experimental procedure). Our results showed, however, that despite this limited sample size, our statistical analyses mostly yielded small p-values and meaningful effect sizes. Second, our findings should be cautiously interpreted when generalizing to other populations since our participants were only young adults. Older adults and elderly may be less familiar with touch-screen devices and may be more disoriented while using such devices [54]. Third, a folded convertible laptop was used instead of a slate tablet PC. Although participants had to use gestures to interact with this device, they were not allowed to handle it due to the use of the eye tracking system. One might argue this may have hampered the interaction and worsened performance. On the other hand, the similarity of the devices (same operating system, same browser) limited the number of potential confounding variables. Furthermore, since the size of the convertible laptop screen was greater than a standard slate tablet PC, we believe that it could have “softened” our results by reducing the need for scrolling. Replicating this study on an off-the-shelf slate tablet PC should thus be considered. Finally, the study was conducted in a controlled environment and participants were asked to perform only one information search at a time. In the workplace, it is most likely than workers are using their devices on the go while doing several tasks in parallel. As multitasking has a strong negative impact on memory encoding [55], it raises the possibility that the impact of using a tablet could be more pronounced in such conditions.

5.2 Future Research

To develop a full picture of the impact of touch-screen devices in the workplace, additional studies will be needed that extend the nature of the tasks involved, as well as taking into account the effect of multitasking on users’ mental resources. It would be interesting to extend the test scenario by including tasks such as checking and writing emails or word processing, in order to best encompass the variety of the tasks performed on IT devices at work. There is also much room for further progress in refining our understanding of the objective measurement of cognitive load. Future research should seek to extend our understanding of the relationships between physiological measures and the underlying cognitive processes in the context of web browsing. Another important avenue for further research concerns the effect of motivational factors on users’ behavior. Techniques such as galvanic skin response or emotion recognition software could be considered to obtain objective measures of users’ emotions or cognitive engagement while using different devices.

5.3 Conclusion

The present research has demonstrated that the device used to search online can impact users’ performance. Regardless of this, it is undeniable that touch-screen computers such as slate tablet PC, detachables and convertibles have the potential to be of great benefit in the workplace, thanks to their portability, usability and interactivity. However, they might not be suited to every activity required in the workplace and special care needs to be taken when introducing new technology into the firms.