Research & DevelopmentWhite Paper

WHP 331

November 9, 2017

Contrasting Delivery Modes for Second Screen TV Content —Push or Pull?

A. Brown, M. Evans and M. Glancywith A. Aizpurua, C. Jay and S. Harper, University of Manchester

Design & Engineering

BRITISH BROADCASTING CORPORATION

BBC Research & DevelopmentWhite Paper WHP 331

Contrasting Delivery Modes for Second Screen TV Content — Push or Pull?

Andy Brown, Michael Evans and Maxine Glancywith Amaia Aizpurua, Caroline Jay and Simon Harper, University of Manchester

Abstract

The use of mobile devices during television viewing is now commonplace, and broadcasters areincreasingly supplying programme-related ‘companion content’. To produce an optimal user expe-rience, it is important to determine how the delivery of companion content affects and is perceivedby the viewer — without this we risk distracting the viewer, leading to frustration and disen-gagement. We present a controlled study investigating how attention, cognitive load, and users’preferences are affected by the provision of two different content delivery modes: pushed and pulled.We find that delivery mode affected the temporal distribution of gaze to the tablet, with a con-sistent viewing pattern for pushed updates, which attracted attention within a few seconds, and amore diverse set of viewing patterns when updates were pulled. Cognitive load was similar in bothconditions, and there was no consensus as to which mode was preferred, but users showed strong,polarised individual preferences. The advantages of each delivery mode are presented as a set ofrecommendations for the delivery of companion content.

Additional key words: Experimental, Evaluation, UX, Television, Distributed Attention,Companion Content, Second Device.

©BBC 2017. All rights reserved.

White Papers are distributed freely on request.

Authorisation of the Chief Scientist is required forpublication.

©BBC 2017. Except as provided below, no part of this document may be reproduced in anymaterial form (including photocoping or storing it in any medium by electronic means) withoutthe prior written permission of BBC Research & Development except in accordance with theprovisions of the (UK) Copyright, Designs and Patents Act 1988.

The BBC grants permission to individuals and organisations to make copies of the entire doc-ument (including this copyright notice) for their own internal use. No copies of this documentmay be published, distributed or made available to third parties whether by paper, electronic orother means without the BBC’s prior written permission. Where necessary, third parties shouldbe directed to the relevant page on BBC’s website at http://www.bbc.co.uk/rd/pubs/whp for acopy of this document.

BBC Research & DevelopmentWhite Paper WHP 331

Contrasting Delivery Modes for Second Screen TV Content — Push or Pull?

Andy Brown, Michael Evans and Maxine Glancywith Amaia Aizpurua, Caroline Jay and Simon Harper, University of Manchester

Executive Summary

Previous work has explored how people divide their attention when they have companion contentpresented on a tablet alongside a TV programme [1] (we define companion content as “curated ad-ditional information supporting a programme that is being broadcast concurrently”). This revealedsome of the factors that influenced where people looked and when they switched their attention.In particular, it revealed that automatically updating the companion content had a strong effecton attention, and this had the potential to disrupt the viewer if presented at an inappropriatetime. Qualitative feedback from participants in those studies also suggested that some people wereunhappy with their lack of control over the companion content.

One solution to this is to give the participants control over what they view on their seconddevice, but this too has potential disadvantages. Does the need to choose what to view and whendistract or confuse viewers? This study, therefore, sought to explore the effects of giving peoplegreater control over the companion content. We compared two methods for presenting companioncontent:

Push: the original method, where the audience members see one segment companion content ata time, and this updates at times designated by the production team.

Pull; where it is possible to see and access all possible segments of companion content at any timeduring the show.

Using the original companion content created for Autumnwatch [46], we explored 3 key researchquestions, with the following results:

RQ1: Does the mode of companion content delivery affect how attention isdistributed between the television and the tablet?

While there was no difference in the overall time spent viewing the second screen between theconditions, there was a difference in the distribution of attention. When companion content ispushed to the user, the automatic updates act as a strong attractor of gaze, drawing viewers awayfrom the television towards the tablet and holding their attention several seconds.

RQ2: Does cognitive load vary according to mode of delivery?

In the pull condition, participants are required to assess the range of companion content availableand select when to view it. The results show that participants chose to match the topic of thecompanion content to that on the television in nearly all cases. The process of assessing, selectingand requesting companion content did not have a measurable impact on the perceived workload,although the interviews revealed that the push condition entailed less decision making than thepull mode. In fact, the median values were slightly lower for the pull mode than for the push forthe ratings for all aspects of task demand except frustration and mental demand, which indicates alower perceived workload for the pull condition. There were no significant differences between the

1

conditions with regard to difficulty managing attention, but the pull condition resulted in feweranswers on the negative end of the scale.

RQ3: Do viewers prefer content to be pushed, or to pull it themselves as re-quired?

The preference data showed that there were no significant differences between the two differentdelivery modes, neither was there a significant difference between participants’ ratings of overallexperience. There was, however, a polarisation in preference between push and pull. Participantswho preferred the pull mode generally did so because they liked to have control over when theylooked at the content. The preference for the push condition was based on the experience beingmore fluid and engaging, and requiring less decision making.

Recommendations

Based on the results of this study, we identify the following recommendations for those creatingcompanion content:

Push or Pull? The results of this study do not support a strong recommendation for either pushor pull. What they do show, however, is that people tend to have a strong personal preferencefor one or the other: pull was liked for the level of control it allows, while push was likedbecause it did not require any user input. Pull would probably make a more suitable default,since participants were less likely to react strongly against it than they were to push. Wemight speculate further that the presentation mode might vary according to the objectivesof the show — more immersive content, such as a drama, might be better served by pushingcompanion content, while pull might work better for factual content with natural breaks.

Timing: Implementing the push behaviour requires careful consideration of update timing, aseach update is likely to attract considerable attention from viewers. Attention builds rapidlyover the 3–5 seconds following the update, then diminishes slowly over the following 20–30seconds.

Layout: The length of these clips meant that there were relatively few companion segments tochoose from in the index page of the pull condition; for a longer show, the index page itselfwill need organisation. Participants’ self-selection of companion content to match televisiontopic suggests that the content could be organised by topic, with index pages per topic pushedto the user, then allowing viewers to pull individual segments.

Notifications: Notifications should be used to mitigate the higher temporal workloads that somepeople felt in the push condition. One approach would be a user-interface component thatindicates when the next update is going to occur, as investigated by [42].

2

1 Introduction

Interacting with a tablet or mobile device while watching television has emerged as a natural way ofconsuming entertainment. The way in which this second device is used varies widely, and includesactivities both related and unrelated to the television programme [2, 3]. Increasingly, a companionapplication is used on the second device to provide extra information that is tightly integrated withthe television programme: this application is used to push programme related content to the user’sdevice. Some examples of these show-related companion applications are ABC’s Grey’s Anatomy,AMC’s The Walking Dead Story Sync, The Conan O’Brien Show’s Team Coco, the BBC Sportsapp or ITV’s Britain’s Got Talent app.

In other cases the interaction is more decoupled: no special application is provided and norelated content is delivered, instead viewers pull information which interests them at times of theirchoice. For instance, going to the Web to find information viewers are interested in, or using socialmedia apps to comment on a particular program.

Based on these scenarios, we define companion content as curated additional information sup-porting a programme that is being broadcast concurrently. In such situations, the viewer’s attentionwill be divided between the two devices in a manner that varies according to the television content,the content on the second device, the social environment, and any task the viewer is performing [4].From a broadcaster’s perspective, the aim is to make the viewer feel engaged with the content.The nature of this engagement will vary according to the content itself, but the goal is to have theaudience enjoy the experience and feel that it was a valuable way to spend time. Traditionally, asimple measure of engagement would be the level of attention given to the television screen, but inthe multi-screen world, this no longer holds — the aim is now to allow the viewer to engage acrossdevices, and this implies that attention is split across different information sources. Managingattention in this situation is not necessarily easy, and a poorly designed experience will leave theviewer unable to get the information she wants at the most appropriate time, resulting in frustra-tion and disengagement. The challenge is to design the experience in such a way that the viewer isable to divide attention across the two information streams in an effortless and seamless manner.

To achieve this, it is necessary to have an understanding of how people behave in a dual screenenvironment. In this paper, we describe a study designed to explore the differences between twomethods of providing programme-coupled companion content on a tablet computer. In the first– push – the companion content is updated automatically, at times designed by the productionteam (synchronised with the television), so that all viewers see the same companion content at thesame time, and this is always directly related to the television content. In the second – pull – theviewer is provided with the same companion content, but accesses it via an index page, and canview any of the segments of companion content at any time. We monitored 24 participants in a‘living room’ lab designed for ecological validity, interacting with television and companion contentthat had been designed by a professional production team. The following research questions wereposed as directly relevant to uncover the necessary knowledge and understanding for designing thedelivery of future companion content:

1. Does the mode of companion content delivery affect how attention is distributedbetween the television and the tablet? Viewers may attend to more of the companioncontent if it is pushed, or spend longer looking at an item of interest that has been pulled.

2. Does cognitive load vary according to mode of delivery? It is possible that theadditional level of control provided in the pull condition may be more demanding, or theautomatic updates of the push condition are distracting.

3. Do viewers prefer content to be pushed, or to pull it themselves as required?Understanding which delivery mode is preferred is key to ensuring a successful viewing ex-perience.

3

The contribution of this paper lies in answering these questions. By doing so, we are uncoveringknowledge and building an understanding of:

• How companion content is consumed, in terms of detailed quantified analysis of the distribu-tion of visual attention;

• How demanding people find the process of consuming both television and companion content,in terms of cognitive workload;

• Which method of companion content presentation people prefer.

This understanding is key to identifying the factors that authors of companion content need toconsider when creating their content, and to propose a set of recommendations for how to use thepresentation method to match their creative aims and provide better TV experiences.

2 Background

The use of a ‘second screen’ – usually a tablet or mobile phone – is, for many, now a routine partof watching TV. The advent of interactive television and the mobile Web has allowed TV viewingto move from a primarily passive to a more active domain, where people can control, enrich, share,and transfer television content [5]. A study by Yahoo! and Nielsen reported that, in 2010, 79%of people occasionally or frequently accessed the Internet over a smartphone while watching TV[6]. An ethnographic study, conducted by Vanattenhoven and Geerts, capturing the second-screenhabits of 12 households, found that second-screen use was recorded by 11 of them [7]. A key reasonfor accessing the Internet on a mobile phone is that it fits better with concurrent activities suchas TV [8]. Secondary devices are now used for a wide variety of purposes, including informationseeking and social media use, which is often, but not necessarily, related to the television programmeon screen [9, 10, 11].

This behaviour has started to happen spontaneously as these technologies become an integralpart of everyday life, but it is not entirely consumer-led – broadcasters have also been exploring theuse of a mobile phone, tablet or laptop computer as a secondary display, primarily to aid contentdiscovery and engagement [12]. The secondary device has been used to provide programme guideinformation [13] or material to aid accessibility to disabled users [14] in a manner that does notocclude the main television screen. This can be particularly important in situations where the mainscreen is being shared by a family or other group. Increasingly, by creating and distributing softwareapplications for phones/tablets, or placing specifically related material on the Web, televisionbroadcasters are using the secondary screen to present companion content synchronised with theprimary programme [15].

For broadcasters, this represents a huge opportunity. Within a more standard, desktop com-puting environment, the preference for multiple displays is well-evidenced [16]. Now it is clear thatmany viewers will be using secondary devices while watching TV, and producing complementarycontent for both screens, to provide an even more compelling experience, is becoming a priority.Broadcasters’ dual screen companion applications might add auxiliary material to drama produc-tions, coverage of sport, news, documentaries and entertainment [17, 18, 19], with a variety ofmechanisms used to achieve synchronisation between the two display devices [20, 21].

In order to optimise the dual screen viewing experience, and ensure that companion content isprovided in a comfortable, informative and engaging way, it is important to have a good under-standing of attention and cognition in this complex interaction scenario. Within HCI, we have seenbehaviour with multiple screens investigated from the perspective of modelling task performance[22, 23, 24, 25, 26, 27], studied in the wild [28, 29], and are now seeing a plethora of novel multipledevice interaction paradigms [30, 31, 32, 33, 34, 35, 36, 37, 38]. In the following section, we considerin more detail related research within the domain of entertainment and media.

4

2.1 Related work

Eye tracking studies have provided some important formative results in this field. In an observa-tional study, Holmes et al. used a head-mounted eye-tracker to monitor the attention of peopleviewing a drama or documentary programme with a companion app running on a second-screen[15]. They reported the proportion of attention given to each device (63% on TV, 30% on tablet)and the lengths of the views on each device (means of just over 2s on the TV, 1s on the tablet).They also noted that attention appeared to shift to the tablet when it updated, and when the TVshowed advertising content. A second observational study using free-standing eye trackers to mon-itor attention across two screens reported a similar split of attention between the TV and mobiledevice, with early evidence that both updates to the companion content, and audio exclamationson the TV, were key factors in attracting attention to the tablet and TV respectively [39].

Other research examining the characteristics of TV and companion content has shown thatkeeping displays temporally synchronised, and allowing users to revisit content to recap, bothimprove the dual screen viewing experience [40], and that it may be important to ensure a closevisual match of content to minimize the time necessary to orient attention when switching betweendevices [41].

A key issue with regard to second-screen use is its potential to place additional cognitive de-mands on the viewer. Whilst viewing with companion content is often a positive experience, it isnot necessarily relaxing [17]. Web browsing on a secondary device has been shown to have a nega-tive effect on factual recall and comprehension of news programmes [2], and people have reportedthat attention overload is a common problem when using a second screen [42]. To allow peoplemore control of the dual screen viewing experience, Neate et al. investigated the effects of embed-ding notifications into the TV show, to alert people to the appearance of new companion content[43]. Like [39], they noted that peripheral stimuli on the secondary device appeared to attractattention, and found that the notifications on the TV screen gave them more conscious controlover their orientation of attention. Qualitative evidence in support of TV-based notifications forcompanion content has also been provided by [44].

Thus far, there has been no formal investigation of how content delivery mode affects visualattention or viewing experience. Whilst there has been some systematic research conducted withregard to the visual presentation of dual screen material [42], much of the previous work hashad a descriptive or observational focus. We build on this foundation with a study that provides acontrolled evaluation of content delivery mode from the perspective of visual attention distribution,perceived cognitive load and user preference. We provide a detailed examination of the distributionof attention following updates, confirming the ‘attention grabbing’ effect that has been notedanecdotally, but not previously quantified. Cognitive load and user preference results demonstratethat broadcasters can retain some freedom in deciding how to deliver companion content, but thatto accommodate both viewers who want an active experience, and those who would prefer it tobe more immersive, it may be necessary to offer different options for delivery mode. These resultsrepresent a timely contribution to user experience in the home environment, at a time when bothTV viewing, and secondary device usage, is on the rise [45].

3 Method

To derive recommendations for the design of companion content, we wanted to investigate behaviourwhilst watching television with a mobile device in a controlled manner, but in a naturalistic envi-ronment. In order to enable participants to experience each type of presentation method, and toreflect on which they preferred, and why, the study design involved each participant viewing onepiece of content with pushed companion content and another with pulled, such that they could rateand reflect on each experience, and compare the two. To minimise bias associated with individualdifferences with regard to the mode of delivery, we followed a mixed design, with all participantswatching the same, single, complete 20 minute clip of a programme. The clip was cut into 2

5

self-contained clips, and each one was presented either experiencing pushed, or pulled companioncontent; presentation order of both clip and delivery mode was counterbalanced.

3.1 Participants

A total of 24 users took part in the study (12 females). The median age of participants was 37 years,with a range of 22-57. Participants were recruited by an external agency. As the stimulus wasnature-themed (see Section3.3), we asked for participants with an interest in natural history andwho enjoyed watching nature programmes. A further inclusion criterion was frequent or regular useof a mobile or tablet while watching television, to find information related to the TV programme.

3.2 Apparatus

The experiment was performed in a user lab that simulates a modern living room. The laboratorywas equipped with cameras and microphones to record participant activity. Four cameras wereused: one was situated above the television screen to record the front view of the participant; asecond camera pointed at the television to record what was being shown; the third camera waspointed down at the tablet, to record what was being shown on the tablet; and the fourth camerawas directed towards the left-hand side of the participant. The television content was displayed ona 47 inch LG television, while the companion content was presented on an Apple iPad through theChrome browser. Both television and companion content were served from a MacBook Pro.

3.3 Stimuli

The stimulus used was related to the Autumnwatch TV program, for which companion contenthad already been created by a professional production team, and broadcast during one episodeof the programme [46]. Autumnwatch presents information about nature in the British Isles inthe autumn, with a weekly programme showing how the wildlife changes over the course of theseason. The episode presented three main topics: dolphins in Cardigan Bay; starlings roostingunder Aberystwyth pier, and sea trout in a Welsh river. It was first broadcast on 18th November2010.

3.3.1 TV content

To minimise practice effects, and maintain a naturalistic viewing experience, the TV programmewas split into two clips (A, B), which were used with their corresponding companion applicationcontent. Each clip covered two topics, and included segments where presenters spoke from anoutdoor studio. Clip A (659 seconds long) started with the topic of ‘starlings’ first, moved to‘dolphins’ and then back to ‘starlings’. Clip B (611 seconds long) started with the topic of ‘starlings’and moved on to a section on ‘sea trout’. The clips were selected to cover two topics, and to beself-contained. Because previously authored companion content was being used, the segments werealso chosen so that the companion content in each clip was as similar as possible with regardto content and frequency. Figure 1 gives an overview of the content on both the TV and thecompanion device.

3.3.2 Companion application content

For each clip we selected the corresponding companion content to present relevant information onthe tablet. As previously mentioned, the content had been designed by the original professionalproduction team, and it was designed to be informative and relatively ‘ambient’. The companioncontent for Clip A consisted of four segments (two for each main topic, ‘starlings’ and ‘dolphins’);that for Clip B contained five segments (three related to ‘starlings’ and two about ‘sea trout’). Allcontent consisted of a single image accompanied by a small amount of text (a single paragraph or

6

one-sentence caption). Interactivity was limited: two segments (A3 and B1) presented a question tothe user and revealed the answer when tapped; two segments (A4 and B4) had no interactivity; twosegments (B2 and B3) had multiple slides that the user could move between. Three segments weremore interactive: A2 had a slider that the user could move back-and-forth to see a visualisation ofstarling flocking, while the other two (A1 and B5) presented an image with further information ondifferent points that could be displayed when tapped (locations to view starlings and the lifecycleof the trout respectively).

(a) Clip A

(b) Clip B

Figure 1: A summary of the TV and companion content for the two clips. The upper half of eachshow what was on the television — sections shaded darker were those where the presenters wereon screen in their outdoor studio; the remainder was location footage. The lower half of each showthe companion content, with red vertical lines indicating where the content was updated in thepush condition.

The same content was used in both conditions, but the delivery mode was varied. Pushedcontent was automatically displayed on the tablet at times originally specified by the productionteam. The content remained on the device until the next segment of content was pushed; thechange was visual only — the screen briefly went white before the new content appeared but therewas no other notification. In the pull condition, the content was accessed via an index page whichallowed participants a completely free choice about when to view it. Figure 2 shows an example ofthe index page, which was considered to be the minimum viable presentation of the content menuto allow viewers to manage scheduling manually.

3.4 Procedure

The experiment took place during January and February 2015. The study was introduced toparticipants, and the nature of the clips and the companion content was explained. Participantswere shown some demo content on the tablet, and were able to try interacting with it. Participantswere encouraged to watch the television as if they were at home, and they were told that they werefree to use, look at and interact with the companion content on the tablet as they wished, or to

7

Figure 2: The index page used for the companion application in the pull condition in Clip B.

completely ignore it.Participants watched both clip A and clip B, each with a different companion content pre-

sentation method. The orders of clip and content delivery mode were counterbalanced so thatparticipants were assigned to one of 4 groups: Group 1 saw A-pull then B-push, Group 2 sawA-push then B-pull, Group 3 B-pull then A-push, and Group 4 B-push then A-pull. After eachvideo clip had finished, participants were asked to self-assess their cognitive workload by meansof the NASA-TLX questionnaire [47], which was presented on the tablet. This was followed by ashort semi-structured interview to collect more in-depth insights and impressions from participantswith regard to their overall experience. Finally, participants were asked to rate their preferredcompanion content provision mode (push or pull) and explain the reasons for their preference. Thesession lasted for approximately one hour. Participants were given a £40 incentive for taking partin the study.

3.5 Data collection and analysis

The following data were collected in order to answer the research questions:

• Video recordings of participant behaviour. These were used to determine the participant’sfocus of visual attention at any point during the experiment.

• A log of every interaction that each user had with the companion content (taps and pagechanges). This supplemented the video data and allowed us to understand when the partici-pants chose to view each page.

• Participants’ ratings of their perceived workload for each condition. Participants were alsoquestioned about how easy or difficult they found it to use the companion content alongsidethe television.

• Likert scale scores concerning the overall experience in each condition, and whether thecompanion content acted more as a distraction from or complement to the television.

• A subjective rating of preference for push or pull presentation.

• Audio recordings of the interviews. These were transcribed so that thematic analysis couldbe applied, in order to identify the reasons behind participants’ preferences.

The data were used to address the research questions as follows:

8

1: ‘Does the mode of companion content delivery affect how attention is distributed between thetelevision and the tablet?’

In order to answer this research question, it is necessary to monitor visual attention. Whilethis can be done using eye-tracking, analysis of the video recordings is less obtrusive as it allowsparticipants to behave more naturally. In this case, the focus of attention throughout the experienceis determined by slow-motion analysis of the the composite video recording of the four camera feeds.For each participant, the time (to 0.1s resolution) of each shift of the viewer’s attention (betweentelevision, tablet, and off-screen) was recorded. The precision of this method is estimated at 0.5s.These data were used to calculate the total amount of attention given to each device. In addition,the proportion of participants looking at either device at any time during either clip could becalculated, to understand how attention varied over time.

2: ‘Does cognitive load vary according to mode of delivery?’Perceived cognitive load data were collected using the NASA TLX workload questionnaire,

which was completed twice by each participant, immediately after each clip finished. The ques-tionnaire was presented on the tablet: the participant was presented with a bar for each question,with ends labelled ‘high’ and ‘low’; tapping on the bar displayed a slider which could be moved towhatever position the participant felt appropriate. No scale was shown to the user, but the sliderhad 21 discrete positions.

The TLX questionnaire was followed by the semi-structured interview, which explored in moredetail how difficult the participants found the experience, and how they felt about it generally.Some questions were answered using Likert scales, while others were purely qualitative. The Likertscale questions were as follows (answered on a scale of 1 to 7, with lower numbers representing aworse experience):

• How difficult or easy did you find it to follow the Autumwatch clip while interacting with theadditional content on the tablet?

• How difficult or easy was it to get the information you wanted from the tablet whilst followingthe content on the TV?

• How difficult or easy did you find it to manage your attention between the two screens?

Further questions in the interview probed participants about their overall views, to understandwhether they felt the companion content added value to the experience. In the push condition par-ticipants were asked whether they felt the timings of the updates were appropriate, and in the pullcondition they were asked how they decided when to access the companion content. Participantswere also asked if they remembered any times at which they felt their attention was drawn to onedevice or the other.

3: ‘Do viewers prefer content to be pushed, or to pull it themselves as required?’Participants were directly asked for quantifiable answers to three questions relating to their

preference. The first were Likert scale questions (as above) asked for each condition:

• How would you rate the overall experience?

• Did you feel that the content on the tablet worked as a distraction from the main clip or didyou feel that it complemented the main content?

The third question, near the end of the session, asked participants to compare the two methods,and to express their preference for one or the other on a simple (quantifiable) scale, followed-upwith questions designed to elicit richer answers. Participants were first presented with a slider(which worked in the same way to those used for the TLX), with ‘push’ at one end and ‘pull’ atthe other, and asked to place the slider on the scale to indicate which mechanism they preferredusing the distance from the centre to indicate how strongly that preference was felt. This was

9

followed up by asking for the reasons behind their answer. In addition to these direct questions,the semi-structured interview was designed to elicit opinions about each mode of presentation, andit was anticipated that the discussion would reveal preferences and the reasons behind them.

4 Results

The results are reported in line with the previously outlined research questions (see Section 1).Section 4.1 presents data showing how the mode of delivery affects the distribution of attention,Section 4.2 describes how mode of delivery affects perceived cognitive load, and Section 4.3 considersviewer preference.

4.1 Does the mode of companion content delivery affect how attention is dis-tributed between the television and the tablet?

A quantitative analysis of the data relating to the overall division of attention between the televisionand the tablet, and how the allocation of attention varied over time, is reported. This section alsoreports on whether, in the pull condition, participants matched their choice of companion contentto the topic on the television.

4.1.1 Overall division of attention

The overall split of attention was assessed by calculating the total time that people spent looking atthe television, the tablet, or off-screen. As shown in Table 1, the television received more attentionthan the tablet under both conditions, but whether the content was pushed or pulled to the tabletdid not affect the overall length of time people spent viewing it.

Television Tablet Off-screenCondition median SD median SD median SD

Push 69.5 9.6 30.4 7.9 0.3 4.6Pull 68.8 10.7 28.8 10.4 0.1 4.2

Table 1: Median divisions of attention (% of total time) between television, tablet, and off-screen,for each condition.

4.1.2 Distribution of attention

Figure 3 demostrates how the proportion of people looking at the tablet varied during each clipunder each condition. The upper plots show the push condition, and are annotated with lines thatindicate when the companion content was updated. Casual observation of these plots suggests thatautomatic updates in the companion content resulted in an almost immediate increase in the levelof attention given to the tablet.

Additional annotations on Figure 3 highlight the periods of time when the tablet received highlevels of attention. These are defined as those time periods (longer than 5 seconds) when morethan 36% of participants were viewing the tablet (this threshold was calculated by dividing thedata into 0.5s slices then ranking them by the level of attention in each slice; the value of thethird quartile was 36%). Again, casual inspection of these graphs indicates that not only was eachautomatic update followed by glances at the tablet, but that people’s attention was maintained formany seconds.

The inferences drawn from these graphs can be tested statistically by comparing the level ofattention received by the tablet before and after an update. To do this, we compare the attentiongiven to the tablet in the 10 seconds leading up to each update with that in a 10 second window after

10

(a) Clip A

(b) Clip B

Figure 3: A comparison of the level of attention received by the tablet over each clip under the push(upper half, blue) and pull (lower half, green) conditions. The black vertical lines show when thecompanion content changed automatically; the yellow bands represent high attention time periods.

the update. The ‘after’ window was chosen to be the period between 5 and 15s after the update,to allow time for participants to notice the change (the exact way in which attention changesfollowing the update is shown in Figure 4). Update times were selected by the production teamso that the companion content was synchronised with the topic under discussion on the television.As the time for update 1 was designed to occur 3s after the start of clip B, it is not possible tocalculate the level of attention in the 10s before. Therefore, update 1 in clip B is not included inthis analysis. For each participant we calculate the time they are viewing the tablet in each of these10s windows, then we calculate the median of these values across participants. This was repeatedfor the same time windows in the pull condition – i.e., the same points in the television programme– despite there being no automatic update. The before and after windows are compared using the

11

Wilcoxon signed-rank test. Table 2 presents these medians for clips A and B under each condition,highlighting those updates where there was a significant difference in attention between the beforeand after 10s windows.

push pullupdate before after p before after p

A1 2.3 4.2 0.21 4.2 2.6 0.52A2 0.0 8.6 0.004 0.0 0.0 0.14A3 0.7 8.7 0.004 2.0 8.3 0.21A4 1.0 7.2 0.007 0.0 0.0 0.29

B2 0.0 8.3 0.006 0.0 4.8 0.94B3 0.8 6.6 0.005 0.0 0.0 1.00B4 0.7 4.5 0.009 4.6 1.6 0.12B5 0.0 9.9 0.004 0.0 0.0 0.42

Table 2: Attention given to the tablet before and after the time each update occurred. Values givethe median total time (s) the device was attended by participants in the 10s before and 5–15s aftereach time. The two periods were compared using the Wilcoxon signed-rank test, bold indicatesthat p < 0.05; N=11.

Figure 4 visualises the effect of updates by overlaying the attention across the times of the fourupdates in each of clips A and B, for the push and pull conditions. These show that in the caseof the push mode, the attention right after the updates rose and then progressively decreased overthe following minute. In contrast, in the case of the pull mode, the attention remained relativelyconstant across these times.

4.1.3 Coverage of companion content

In the push condition, all segments of companion content were displayed to the user; while inthe pull condition, participants viewed an index page and had to choose to view each piece ofcompanion content. The interaction logs reveal that all participants visited all segments in clip A;in clip B, segments B1 and B2 were visited by everyone, while B2 was not visited by one (P15),B3 was not visited by three (P1, P15, P20) and B4 was not visited by 3 (P3, P13, P24). Someparticipants viewed a segment two or three times. P01 visited all segments of clip B at the start,then did not interact again with the companion content.

Table 3 details how many participants interacted with each interactive segment under eachcondition, and gives the median number of touches on the tablet for each. Overall, the total numberof segments interacted with was the same, but there were differences in individual segments. Thebiggest difference was found for segment A2, an interactive slider showing a simulation of starlingflocking — the tablet was touched more than twice as many times in the push condition as inthe pull condition. On the other hand, segment B5 was interacted with by 7 people in the pullcondition, compared with one in the push condition. The total amount of interaction (as measuredby touches) was 422 touches in the pull condition and 597 touches in the push condition; thisdifference arose almost entirely from clip A (193 in pull vs. 367 in pull). As a whole, it is not clearthat the way in which participants accessed the information affected how much people were willingto interact with it.

4.1.4 Coordination of companion and TV content

In the push condition, participants were presented with new companion content at times that theproduction team deemed suitable; these times were selected so that the companion content com-plemented the topic under discussion on the television. In the pull condition, however, participantswere free to view any of the segments of companion content at any time during the clip (although

12

(a) Push

(b) Pull

Figure 4: Overlays of the overall attention data from the times around the push updates in clipsA and B. The data are aligned so that t = 0 corresponds to the time in the clip that the updateis scheduled in the push condition. The mean proportion of people attending at any given relativetime is plotted as a line.

13

Participants TouchesSegment Push Pull Push Pull p

A1 11 7 4 1 0.0066A2 12 10 9.5 3.5 0.0022A3 9 12 4 5

B1 8 11 1 3.5B2 12 10 7 3B3 12 8 6 7B5 1 7 0 1.5 0.0071

Total 65 65 597 422

Table 3: This table presents the number of participants (from 12) who interacted with each in-teractive segment under each condition, and the median number of touches made on the tablet ineach. The final column gives the p value for the Wilcoxon signed-rank test comparing the numberof touches per person, where a significant difference (p < 0.05) was found between conditions. Thetotal values for touches give the total number of touches across all participants.

the order of links on the index page matched that used in the push condition). Analysis of theinteraction data indicates that people almost exclusively matched the theme of the companioncontent with the topic currently on the television. Although the segments were generally looked atin a different order to that provided, only two participants viewed a segment of companion contentat a time when the relevant topic was not under discussion (e.g., viewing companion content aboutstarlings while the TV presenters were discussing trout).

4.2 Does cognitive load vary according to mode of delivery?

Overall workload was measured by means of the NASA-TLX scale, which consists of 6 subscales:mental demand, physical demand, temporal demand, performance, effort and frustration. Partic-ipants were asked to self-assess their perceived workload after interacting with each presentationmode. Figure 5 summarises the ratings given by participants; these did not differ significantly(Wilcoxon Signed-Rank Test, p > 0.05) between push and pull presentation modes for any of thesix subscales.

Some of the questions that participants answered in the semi-structured interview were capturedon Likert scales. Three of these related to workload, asking participants how easy or difficultthey found it to follow the TV while watching the iPad, or vice versa, and how easy they foundmanaging their attention. As with the workload scores, there were no significant differences betweenconditions, but the pull condition resulted in fewer answers on the negative end of the scale. Figure6 summarises these results.

4.3 Do viewers prefer content to be pushed, or to pull it themselves as required?

After having interacted with both companion content presentation modes, participants were askedto show their preferred option on a 15-point scale, from 1 (push) to 15 (pull). Figure 7 shows thatnine participants preferred the push mode, eleven preferred the pull mode and two participants didnot show any preference. The median value was 9.5 — slightly on the pull side of the scale.

Two of the Likert scale questions that participants answered were related to preference. Thesewere about the quality of the overall experience, and whether the companion content was seenas distracting or complementary to the TV. Figure 8 summarises these results. There were nosignificant differences between conditions, but the pull condition resulted in fewer answers on thenegative end of the scale. For example, 5 of the 24 participants considered the companion content

14

Figure 5: Box plot comparing NASA TLX workload results.

Figure 6: Box plot showing distributions of the answers to 3 questions during the semi-structuredinterview. Answers were given on a 7-point Likert scale.

in the push condition to be distracting (i.e., 1–3 on the Likert scale), compared to none for the pullcondition.

4.3.1 Qualitative analysis

The semi-structured interviews allowed us to capture some interesting insights into why partic-ipants preferred one presentation mode or the other. In order to understand these, ThematicAnalysis [48] was applied to the transcriptions of the final post-interaction interviews. Four themeswere identified: control, effort, fluidity and engagement. The first related to the pull condition;

15

Figure 7: Distribution of the ratings given by participants reflecting their preference over push (1)or pull (15).

the other three were benefits of the push condition. This reveals that while overall the preferenceswere balanced, the reasons behind them were not: while most participants cited the same reasonfor preferring pull, there was a wider range of benefits identified with having content pushed.

To ensure the reliability of the results, a second coder reviewed the transcripts and applied thecoding scheme to the data. Cohen’s Kappa statistic was computed, to measure the average level ofagreement between both coders: a mean coefficient of 0.88 was obtained, indicating a substantiallevel of agreement. The themes are described below.

Participants who preferred the pull mode valued having control of the companion content,as they could choose what to look at on the tablet and when (15 mentions, 11 participants):

“more free will [. . . ] you can go at your own pace, you can go after triggers, after TVthat’s relevant to you ” (P04)

Figure 8: Box plot showing distributions of the answers to two questions during the semi-structuredinterview. Answers were given on a 7-point Likert scale.

16

“I’d rather be in control of the research rather than being dictated to” (P11)

“it’s just simple, it’s unrushed, you don’t feel like – you sort of feel like you have toread it when it’s being rushed along [. . . ] I like the fact that you can just choose to readwhat you want to read as the programme goes on” (P12)

Participants who preferred the push mode found the experience entailed less effort to get theinformation, as the information was pushed to them automatically. They perceived it as a moreconvenient way of getting the content, requiring less decision making (14 mentions, 9 participants):

“I chose the automatic one because it took the decision making away from you at thetime” (P08)

“It was very unique really for me because you’re actually given the information ratherthan having to look for it [. . . ] Just more relaxing really [. . . ] it wasn’t the same effortinvolved” (P09)

“I just liked the convenience of the automatic [. . . ] it’s nicer having the informationthat’s relevant just come up so you don’t have to look for it [. . . ] I found it reallyconvenient and enjoyable. Yeah less effort and information comes up that you wouldn’thave to look for [. . . ] you don’t have to sort of distract yourself to go and look for it.It’s just there popping up so you can enjoy watching the programme and then also seeany other information that pops up automatically” (P13)

Some of the participants who preferred the push mode perceived it as a smoother experiencewith a better flow than the pull mode (5 mentions, 4 participants):

“I found it more fluid I have to say [. . . ] a little bit more smoother [. . . ] I found theactual experience more fluid and more relaxing” (P09)

“It was just easy flowing” (P14)

“I think the pace, the kind of flow was a lot better” (P18)

Some participants with a preference for the push mode talked about having a more engagingexperience than when the companion content was pulled (6 mentions, 5 participants):

“I liked it because I was kept engaged, it was giving me the information – it complimentedwhat was going on on the screen” (P07)

“I preferred the automatic sort of feature where it would go through [. . . ] you get moreinvolved then [. . . ] I quite like the idea of getting involved as well while you’re watching”(P14)

“makes you feel that you’re a part of the program [. . . ] So you’re sort of engaged in ita bit more, whereas the first one” [the pull mode] “you are opting in when you wantto” (P22)

5 Discussion

This experiment aimed to explore the effect of giving viewers control over which segments ofcompanion content to view at any given time, comparing this to the condition where companioncontent is automatically updated at times pre-determined by the production team and designed toprovide a high-level of synchronisation between devices. Here we discuss the results with respect towhether the method of presenting the companion content had an impact on how people allocatedtheir attention; which condition was perceived as demanding higher cognitive load, and which onewas preferred by viewers (see Section 1).

17

5.1 Does the mode of companion content delivery affect how attention is dis-tributed between the television and the tablet?

While there was no difference in the overall time spent viewing the second screen between theconditions, there was a difference in the distribution of attention. When companion content ispushed to the user, the automatic updates act as a strong attractor of gaze, drawing viewers awayfrom the television towards the tablet and holding their attention several seconds. This resultextends previous work, which reported anecdotally that updates may have this effect [39], byproviding quantitative evidence that can be used by production teams.

5.2 Does cognitive load vary according to mode of delivery?

In the pull condition, participants are required to assess the range of companion content availableand select when to view it. The results show that participants chose to match the topic of thecompanion content to that on the television in nearly all cases. The process of assessing, selectingand requesting companion content did not have a measurable impact on the perceived workloadaccording to the ratings from the NASA-TLX, although the interviews revealed that the pushcondition entailed less decision making than the pull mode. In fact, the median values were slightlylower for the pull mode than for the push for the ratings in all subscales of the NASA-TLX exceptfor the frustration and mental demand, which indicates a lower perceived workload for the pullcondition. There were no significant differences between the conditions with regard to difficultymanaging attention, but the pull condition resulted in fewer answers on the negative end of thescale.

5.3 Do viewers prefer content to be pushed, or to pull it themselves as required?

The preference data showed that there were no significant differences between the two differentdelivery modes, neither was there a significant difference between participants’ ratings of overallexperience. There was, however, a polarisation in preference between push and pull. Participantswho preferred the pull mode generally did so because they liked to have control over when theylooked at the content. The preference for the push condition was based on the experience beingmore fluid and engaging, and requiring less decision making.

While it was not the primary aim of this research to explore which factors in the contentinfluenced how people allocate their attention, some of these have emerged during the course of theanalysis. In the pull condition, people could decide when to view additional content, and it wasobserved that this was typically done during periods of low visual interest on the TV, such as whenthe presenters were talking to each other in the studio, or when they wanted additional informationabout something that had interested them in the television programme. It was also noted that, inboth conditions, verbal audio cues are important triggers for switching attention from the secondscreen back to the television. It will ultimately be important not only to understand which factorstrigger of attention shifts, but to understand the interplay between them, to ensure that contentis presented in a way that minimises distraction, and increases the likelihood of attention shifts torelevant information.

6 Recommendations and future work

The use of a second device while watching television has emerged as a natural behaviour due to theubiquity of tablet and mobile devices, and carefully designed companion content has the potentialto greatly enhance the television viewing experience. Based on the results of this study, we identifythe following recommendations for those creating companion content:

Push or Pull? The results of this study do not support a strong recommendation for either pushor pull. What they do show, however, is that people tend to have a strong personal preference

18

for one or the other: pull was liked for the level of control it allows, while push was likedbecause it did not require any user input. Pull would probably make a more suitable default,since participants were less likely to react strongly against it than they were to push. Wemight speculate further that the presentation mode might vary according to the objectivesof the show — more immersive content, such as a drama, might be better served by pushingcompanion content, while pull might work better for factual content with natural breaks.

Timing: Implementing the push behaviour requires careful consideration of update timing, aseach update is likely to attract considerable attention from viewers. Attention builds rapidlyover the 3–5 seconds following the update, then diminishes slowly over the following 20–30seconds.

Layout: The length of these clips meant that there were relatively few companion segments tochoose from in the index page of the pull condition; for a longer show, the index page itselfwill need organisation. Participants’ self-selection of companion content to match televisiontopic suggests that the content could be organised by topic, with index pages per topic pushedto the user, then allowing viewers to pull individual segments.

Notifications: Notifications should be used to mitigate the higher temporal workloads that somepeople felt in the push condition. One approach would be a user-interface component thatindicates when the next update is going to occur, as investigated by [42].

7 Limitations

The stimulus used in this study was related to the Autumnwatch program and was created by amajor international broadcaster as a genuine and varied dual-screen experience. While the partic-ipants were people with an interest in natural history, who enjoyed watching nature programmesand were used to use the mobile or tablet while watching TV, we hypothesise that the outcomes aregeneralisable to broader audiences interested in other factual programs with similar characteristicsto the Autumnwatch program, and to the companion content format we used. Nevertheless, furtherresearch is required to confirm these recommendations for other genres of TV programs.

An important limitation of this study is that we only monitored visual attention. Auditoryattention is an important part of the experience, but due to the fact the companion contentemployed was purely visual in nature, and the methodological challenges in objectively determiningthe location of auditory attention in this situation, we only tracked visual attention.

8 Conclusions

This study explored two methods for presenting companion content to television viewers: in one(push) the viewer receives updates automatically, at a time determined by a production team; inthe other (pull), viewers had access to all companion content throughout the show, and made theirown decisions about when to view each segment.

While there was no clear overall preference for push or pull, there was a polarisation: most usersstrongly preferred one mechanism or the other. People preferred the pull mode due to the controlit provided, and the push mode because it was perceived as a more fluid and engaging experiencethat required less effort in terms of decision making. In the push condition, it was demonstratedthat automatic updates had a strong effect on viewers’ attention, pulling it to the tablet almostimmediately and keeping it there for several seconds.

These outcomes are an important step in building an empirical understanding of how to produceand present companion content. They indicate that there is an appetite for the high-qualitycompanion content, curated by the production team, and that it has a place alongside self-directedbrowsing.

19

9 Acknowledgments

We would like to thank the EPSRC, the BBC, and Manchester Informatics for their fundingcontribution to this work via the Synchronised Attention Grant (EP/K503782/1), the MultipleScreen Orchestration Grant (EP/H500154/1) and the IDInteraction Grant (EP/M017133/1)

References

References

[1] A. Brown, R. Jones, M. Glancy, M. Evans, Armstrong M., C. Jay, and S. Harper. R&D3729 — Understanding how people split attention over two devices. Technical report, BBCResearch & Development, 2015.

[2] Anna Van Cauwenberge, Gabi Schaap, and Rob van Roy. TV no longer commands our fullattention: Effects of second-screen viewing and task relevance on cognitive load and learningfrom news. Comput. Human Behav., 38:100–109, September 2014.

[3] Duncan P. Brumby, Helena Du Toit, Harry J. Griffin, Ana Tajadura-Jimenez, and Anna L.Cox. Working with the television on: An investigation into media multitasking. In Proceed-ings of the Extended Abstracts of the 32Nd Annual ACM Conference on Human Factors inComputing Systems, CHI EA ’14, pages 1807–1812, New York, NY, USA, 2014. ACM.

[4] Marianna Obrist, Regina Bernhaupt, and Manfred Tscheligi. Interactive tv for the home: Anethnographic study on users’ requirements and experiences. Int. J. Hum. Comput. Interaction,24(2):174–196, 2008.

[5] Pablo Cesar, Dick C. Bulterman, and A. J. Jansen. Usages of the secondary screen in aninteractive television environment: Control, enrich, share, and transfer television content. InProceedings of the 6th European Conference on Changing Television Environments, EUROITV’08, pages 168–177, Berlin, Heidelberg, 2008. Springer-Verlag.

[6] Ashmeed Ali, Edward Wong, Radha Subramanyam, Paul Cushman, Brian Morel, Alice KimHawari, Mandar Shinde, Gateley Meeker, David Gill, Aaron Paquette, and Jake Moskowitz.Mobile shopping framework. the role of mobile devices in the shopping process. Technicalreport, Yahoo! and The Nielsen Foundation, 2010.

[7] Jeroen Vanattenhoven and David Geerts. Second-screen use in the home: An ethnographicstudy. In EuroITV 2012 10th European Interactive TV Conference, pages 162–173. FraunhoferInstitute for Open Communication System, 2012.

[8] Stina Nylander, Teres Lundquist, and Andreas Brannstrom. At home and with computeraccess: Why and where people use cell phones to access the internet. In Proceedings of theSIGCHI Conference on Human Factors in Computing Systems, CHI ’09, pages 1639–1642,New York, NY, USA, 2009. ACM.

[9] Emmanuel Tsekleves, Roger Whitham, Koko Kondo, and Annette Hill. Bringing the televisionexperience to other media in the home: An ethnographic study. In Proceedings of the SeventhEuropean Conference on European Interactive Television Conference, EuroITV ’09, pages 201–210, New York, NY, USA, 2009. ACM.

[10] John Rooksby, Mattias Rost, Alistair Morrison, Marek Bell, Matthew Chalmers, and TimothySmith. Practices of parallel media: using mobile devices when watching television. In Designingwith Users for Domestic environments: Methods, Challenges: Lessons Learned, 2014.

20

[11] Evelien D’heer, Cedric Courtois, and Steve Paulussen. Everyday life in (front of) the screen:The consumption of multiple screen technologies in the living room context. In Proceedingsof the 10th European Conference on Interactive Tv and Video, EuroiTV ’12, pages 195–198,New York, NY, USA, 2012. ACM.

[12] Jhilmil Jain, Michael Evans, and Vinoba Vinayagamoorthy. Exploring and enhancing the userexperience for tv. In CHI ’13 Extended Abstracts on Human Factors in Computing Systems,CHI EA ’13, pages 3187–3190, New York, NY, USA, 2013. ACM.

[13] Leon Cruickshank, Emmanuel Tsekleves, Roger Whitham, Annette Hill, and Kaoruko Kondo.Making interactive tv easier to use: Interface design for a second screen approach. The DesignJournal, 10(3):41–53, 2007.

[14] Vinoba Vinayagamoorthy, Penelope Allen, Matt Hammond, and Michael Evans. Researchingthe user experience for connected tv: A case study. In CHI ’12 Extended Abstracts on HumanFactors in Computing Systems, CHI EA ’12, pages 589–604, New York, NY, USA, 2012. ACM.

[15] Michael E. Holmes, Sheree Josephson, and Ryan E. Carney. Visual attention to televisionprograms with a second-screen application. In Proceedings of the Symposium on Eye TrackingResearch and Applications, ETRA ’12, pages 397–400, New York, NY, USA, 2012. ACM.

[16] Xiaojun Bi and Ravin Balakrishnan. Comparing usage of a large high-resolution display tosingle or dual desktop displays for daily work. In Proceedings of the SIGCHI Conference onHuman Factors in Computing Systems, CHI ’09, pages 1005–1014, New York, NY, USA, 2009.ACM.

[17] Santosh Basapur, Gunnar Harboe, Hiren Mandalia, Ashley Novak, Van Vuong, and CrystaMetcalf. Field trial of a dual device user experience for itv. In Proceddings of the 9th Inter-national Interactive Conference on Interactive Television, EuroITV ’11, pages 127–136, NewYork, NY, USA, 2011. ACM.

[18] Janet Murray, Sergio Goldenberg, Kartik Agarwal, Tarun Chakravorty, Jonathan Cutrell,Abraham Doris-Down, and Harish Kothandaraman. Story-map: Ipad companion for longform tv narratives. In Proceedings of the 10th European Conference on Interactive Tv andVideo, EuroiTV ’12, pages 223–226, New York, NY, USA, 2012. ACM.

[19] Pedro Centieiro. Bringing the sport stadium atmosphere to remote fans. In Proceedings ofthe 21st ACM International Conference on Multimedia, MM ’13, pages 1063–1066, New York,NY, USA, 2013. ACM.

[20] C. Howson, E. Gautier, P. Gilberton, A. Laurent, and Y. Legallais. Second screen tv syn-chronization. In Consumer Electronics - Berlin (ICCE-Berlin), 2011 IEEE InternationalConference on, pages 361–365, Sept 2011.

[21] N.Q.K. Duong, C. Howson, and Y. Legallais. Fast second screen tv synchronization combiningaudio fingerprint technique and generalized cross correlation. In Consumer Electronics - Berlin(ICCE-Berlin), 2012 IEEE International Conference on, pages 241–244, Sept 2012.

[22] Roman Radle, Hans-Christian Jetter, Mario Schreiner, Zhihao Lu, Harald Reiterer, andYvonne Rogers. Spatially-aware or spatially-agnostic?: Elicitation and evaluation of user-defined cross-device interactions. In Proceedings of the 33rd Annual ACM Conference on Hu-man Factors in Computing Systems, CHI ’15, pages 3913–3922, New York, NY, USA, 2015.ACM.

[23] James R. Wallace, Stacey D. Scott, and Carolyn G. MacGregor. Collaborative sensemakingon a digital tabletop and personal tablets: Prioritization, comparisons, and tableaux. In

21

Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI ’13,pages 3345–3354, New York, NY, USA, 2013. ACM.

[24] Fabian Hennecke, Wolfgang Matzke, and Andreas Butz. How screen transitions influence touchand pointer interaction across angled display arrangements. In Proceedings of the SIGCHIConference on Human Factors in Computing Systems, CHI ’12, pages 209–212, New York,NY, USA, 2012. ACM.

[25] Dugald Hutchings. An investigation of fitts’ law in a multiple-display environment. In Pro-ceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI ’12, pages3181–3184, New York, NY, USA, 2012. ACM.

[26] Xiaojun Bi, Seok-Hyung Bae, and Ravin Balakrishnan. Effects of interior bezels of tiled-monitor large displays on visual search, tunnel steering, and target selection. In Proceedingsof the SIGCHI Conference on Human Factors in Computing Systems, CHI ’10, pages 65–74,New York, NY, USA, 2010. ACM.

[27] Dagmar Kern, Paul Marshall, and Albrecht Schmidt. Gazemarks: Gaze-based visual place-holders to ease attention switching. In Proceedings of the SIGCHI Conference on HumanFactors in Computing Systems, CHI ’10, pages 2093–2102, New York, NY, USA, 2010. ACM.

[28] Tero Jokela, Jarno Ojala, and Thomas Olsson. A diary study on combining multiple infor-mation devices in everyday activities and tasks. In Proceedings of the 33rd Annual ACMConference on Human Factors in Computing Systems, CHI ’15, pages 3903–3912, New York,NY, USA, 2015. ACM.

[29] Nicholas Chen, Francois Guimbretiere, and Abigail Sellen. Graduate student use of a multi-slate reading system. In Proceedings of the SIGCHI Conference on Human Factors in Com-puting Systems, CHI ’13, pages 1799–1808, New York, NY, USA, 2013. ACM.

[30] Xiang ’Anthony’ Chen, Tovi Grossman, Daniel J. Wigdor, and George Fitzmaurice. Duet:Exploring joint interactions on a smart phone and a smart watch. In Proceedings of theSIGCHI Conference on Human Factors in Computing Systems, CHI ’14, pages 159–168, NewYork, NY, USA, 2014. ACM.

[31] Peter Hamilton and Daniel J. Wigdor. Conductor: Enabling and understanding cross-deviceinteraction. In Proceedings of the SIGCHI Conference on Human Factors in Computing Sys-tems, CHI ’14, pages 2773–2782, New York, NY, USA, 2014. ACM.

[32] Olivier Chapuis, Anastasia Bezerianos, and Stelios Frantzeskakis. Smarties: An input systemfor wall display development. In Proceedings of the 32Nd Annual ACM Conference on HumanFactors in Computing Systems, CHI ’14, pages 2763–2772, New York, NY, USA, 2014. ACM.

[33] Michael Nebeling, Theano Mintsi, Maria Husmann, and Moira Norrie. Interactive developmentof cross-device user interfaces. In Proceedings of the 32Nd Annual ACM Conference on HumanFactors in Computing Systems, CHI ’14, pages 2793–2802, New York, NY, USA, 2014. ACM.

[34] Jishuo Yang and Daniel Wigdor. Panelrama: Enabling easy specification of cross-device webapplications. In Proceedings of the SIGCHI Conference on Human Factors in ComputingSystems, CHI ’14, pages 2783–2792, New York, NY, USA, 2014. ACM.

[35] Pei-Yu (Peggy) Chi and Yang Li. Weave: Scripting cross-device wearable interaction. InProceedings of the 33rd Annual ACM Conference on Human Factors in Computing Systems,CHI ’15, pages 3923–3932, New York, NY, USA, 2015. ACM.

22

[36] Steven Houben and Nicolai Marquardt. Watchconnect: A toolkit for prototyping smartwatch-centric cross-device applications. In Proceedings of the 33rd Annual ACM Conference onHuman Factors in Computing Systems, CHI ’15, pages 1247–1256, New York, NY, USA,2015. ACM.

[37] Ken Hinckley, Morgan Dixon, Raman Sarin, Francois Guimbretiere, and Ravin Balakrishnan.Codex: A dual screen tablet computer. In Proceedings of the SIGCHI Conference on HumanFactors in Computing Systems, CHI ’09, pages 1933–1942, New York, NY, USA, 2009. ACM.

[38] Jakob Bardram, Sofiane Gueddana, Steven Houben, and Søren Nielsen. Reticularspaces:Activity-based computing support for physically distributed and collaborative smart spaces.In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems, CHI ’12,pages 2845–2854, New York, NY, USA, 2012. ACM.

[39] Andy Brown, Michael Evans, Caroline Jay, Maxine Glancy, Rhianne Jones, and Simon Harper.Hci over multiple screens. In CHI ’14 Extended Abstracts on Human Factors in ComputingSystems, CHI EA ’14, pages 665–674, New York, NY, USA, 2014. ACM.

[40] David Geerts, Rinze Leenheer, Dirk De Grooff, Joost Negenman, and Susanne Heijstraten. Infront of and behind the second screen: Viewer and producer perspectives on a companion app.In Proceedings of the 2014 ACM International Conference on Interactive Experiences for TVand Online Video, TVX ’14, pages 95–102, New York, NY, USA, 2014. ACM.

[41] Christian Valuch, Ulrich Ansorge, Shelley Buchinger, Aniello Raffaele Patrone, and OtmarScherzer. The effect of cinematic cuts on human attention. In Proceedings of the 2014 ACMInternational Conference on Interactive Experiences for TV and Online Video, TVX ’14, pages119–122, New York, NY, USA, 2014. ACM.

[42] Timothy Neate, Matt Jones, and Michael Evans. Designing attention for multi-screen TVexperiences. Proc. 2015 Br. HCI Conf. - Br. HCI ’15, (August):285–286, 2015.

[43] Timothy Neate, Matt Jones, and Michael Evans. Mediating attention for second screen com-panion content. In Proceedings of the 33rd Annual ACM Conference on Human Factors inComputing Systems, CHI ’15, pages 3103–3106, New York, NY, USA, 2015. ACM.

[44] Pedro Almeida, Jorge Abreu, Telmo Silva, Lıgia Duro, Monica Aresta, and Rita Oliveira.Notification Mechanisms In Second-Screen Scenarios - Towards a Balanced User Experience.IEEE, 2015.

[45] D. Docherty et al. Final Report and Recommendations. Futur. Innov. Telev. Technol. TaskForce, 2014.

[46] Theo Jones. Designing for second screens: The autumnwatch companion. http://www.bbc.

co.uk/blogs/researchanddevelopment/2011/04/the-autumnwatch-companion---de.

shtml, last accessed 2013-12-12, 2011.

[47] Sandra G. Hart. Nasa-Task Load Index (Nasa-TLX); 20 Years Later. In Human Factors andErgonomics Society Annual Meeting, volume 50, 2006.

[48] Virginia Braun and Victoria Clarke. Using thematic analysis in psychology. Qualitative Re-search in Psychology, 3(2):77–101, 2006.

23