- Computer users switch tasks extremely frequently, roughly every few minutes by one estimate (Gonzalez & Mark, 2004)
- interruptions occurring at points of higher mental workload are more disruptive and lead to larger resumption lags than those occurring at points of lower mental workload (Bailey & Idbal, 2008)
- alternative explanation = when users are alerted to interruptions at points of higher mental workload, they delay processing of the interruption until they have reached a point of lower mental workload (Salvucci and Bogunovich, 2010)
- A few recent studies (Iqbal & Horvitz, 2007) have suggested that, for deferrable interruptions, users indeed tend to “stabilize task state” before responding.
- However, these studies did not carefully control mental workload, but analyzed workload informally or using hierarchical task models
Salvucci and Bogunovich (2010)
- test their alternative claim using electronic mail customer support task (primary task) and a chat task (interrupting secondary task). Controlled mental workload by having users mentally maintain piece of info. --> windows were sized and positioned to overlap greatly = forced users to switch between windows and so record these task switches
- This type of temporary task-relevant information, called the problem state --> found to act as a constraining bottleneck on multitasking performance: cognition can only maintain problem-state info for one task at a time, and so task switching -> costs from swapping problem states (Corst & Taatgenm 2007)
- The problem-state bottleneck suggests that users will monotask—focus exclusively on the primary task—until task problem state has been eliminated (or at least minimized).
- tested this hypothesis by examining whether users, having received an interrupting chat message, process and respond to the message only at points of minimal mental workload in the primary mail task.
- Supported: when users have the option to defer an interrupting task, they tend to monotask until primary-task mental workload has been minimized. In low work load switched to interrupting task 94% of time vs. only 6% during higher workload
= users can capably handle incoming alerts and defer processing of interruptions until points of lower workload
--> suspect that this ability also generalizes to user self-interruptions and discretionary multitasking (Jin & Dabbish, 2009).
- The maintenance of problem-state information, such as the product name needed while browsing, serves as an important form of workload that can also be associated with a central bottleneck in multitasking behaviour
Great evolution in phones --> from single to multi-purpose devices, but despite this the interaction model for how incoming calls are handled has barely changed
- Leiva et al. (2012) analyzed data from several thousand users over an 18-month period --> smartphone users rarely interrupted by phone calls while they are using other apps (at most 10% of daily app usage). But when they are interrupted, it is massively disruptive and increases the time it takes users to complete the task they were working on.
- attribute to full-screen alerts which interrupt a user’s concurrent app usage.
= large resumption lag after dealing with incoming calls (up to 40s longer to finish the task they were working on prior to dealing with the call.)
- smartphones should move away from using an immediate full-screen notification to signal an in-coming call, because this does not give the user any time to prepare for the interruption.
Instead use a gradual overlay notification, would give user time to prepare for the call
- Iqbal et al. (2011) found that call pre-alerts can reduce the impact that incoming calls have on driving performance (opportunity to think about whether or not it is a good time to take the call)
Bohmer et al. (2014) analysed current smartphones: found that they have two shortcomings that may amplify the disruptiveness of incoming call notifications- 1) full screen notifications, and 2) only provide 2 options, answer or decline
- lab study compared different interfaces that may solve the issue:
1) Postpone UI --> gives 3 options, accept, reject, postpone. Allows user to return to previous app without having to decide whether to react to the incoming call -> increased flexibility/choice. Caller will stay on the line and after short while the app will come to the foreground again for user to accept/reject call.
2) Multiplex UI --> change visual appearance of call notifications. Instead of full screen, divide into two areas, and user again has choice to accept, reject, postpone. Less screen area used for alert = continue on primary task
- compared user experience for each of these UIs when there was an incoming phone call. Primary task = answer a question using three different apps had to memorise and connect pieces of information. Interrupting task: call automatically made after 6s of working on task 5-rounds of the last-letter word generation task
- Found no difference in overall performance for each condition (i.e. resumption lag, speed and errors) but p's found postpone UI more frustrating and needed less mental effort with multiplex UI.
--> p's actively used multiplex to display the call notification in parallel while working on the task = used to defer interruption. Also used postpone option less (because it decreased its value??)
- Replicated finidings in naturalists settings --> p's interestingly requested feature: ability to drag-and-drop the widget to an additional area for declining and sending prewritten text messages to the caller
- Teevan & Heymeyer (2013) positive effect when the caller is aware of the callee’s status --> idea might be to signal the caller as to what is happening (but social implications of declining a call to consider)
- in terms of multiplex UI, change blindness might be an issue (Davies & Beeharee, 2012)
- to further reduce interruptions, should think about changing ringtone to unobtrusive noise
How to make calls less disruptive:
How to know whether to take a call or not:
- Iqbal et al.’s OASIS (2010) system holds non-urgent computer alerts until periods when users are interruptible.
- Ter Hofte (2007) has applied this idea to managing telephone calls by building a predictive model that blocks calls to users when they are actively engaged in an activity.
- TellingCalls by Granndhi et al. (2011) conveys information about the call between caller and callee (e.g. the call’s topic).
- Mark et al. (2014) in situ tracking study using online activity logging and experience sampling (probing user throughout the day). Measured online interaction with custom-built software, mouse/keyboard activity, computer sleep mode, and scales for engagement, challenge, affect and arousal
- P's more focused (peaks mid-afternoon) than bored (peaks early afternoon)
- Happiest doing rote work (focused work more stress) and so being in a state of flow/focus doesn't = most happiness
- attentional states shift as online activities change
- Mondays = most bored day (blue monday)
- = people may gradually move into a focused state --> raises question of whether the interruptions break focus or if the lack of focus comes first, leading to interruption susceptibility
- results can be used to inform the design of workplace tools so they can promote more focus during use.
- highly educated sample (all with degree) = ability to generalise?But, 32 p's is more than double other work observation studies and data collected for 5 full days = analyse variability of attentional states across range of contexts
- Construct validity? attentional states labels were based on p's literally just rating how engaged/challenged they felt at that moment
Jin and Dabbish (2009) looked at self-interruption on computer
- found typical information worker is interrupted every 12mins and half the time they're interrupting themselves --> most focus has been on external influences
- identified 7 categories of self-interruption based on observations and retrospective interviews of people using a computer to complete their normal work tasks
- breaks, recollections, and routines - primarily initiated by the user’s internal cognitive processes,
- adjustments, triggers, waits, and inquiries motivated by the situation
e.g. several users initiated a break because of internal feelings of boredom, whereas waits were initiated because of the primary task status.
- +ve/-ve of interruptions: inquiries, breaks, and adjustments facilitated the primary task by providing valuable info/creating an environment that encouraged increased productivity. But, self-interruptions also harmed the primary task because of context- switching costs and delay.
- since can be both +ve and -ve, how do we design systems to reduce interruptions?? must target specific characteristics of different types of interruptions
- Limitation: observer presence may have influenced p behavior e.g. more focused because being watched. Also small no. p's, in future should increase this and vary age and occupation = more generalisable
Important to consider real world context e.g. in the workplace:
- Nature of workplace is dynamic, can cause attentional states of workers to change depending on many factors (e..g (task-at-hand, interactions, affective state, interruptions, online activities --> changes in these = interruptions?)
- Schaufelli et al. (2002) engagement = a state of mind where one feels absorbed and dedicated in work --> users’ self-ratings of engagement found to have situational validity. Focus, boredom, and rote (repetitive) work affect engagement and challenge
Why do people want to multitask and self-interrupt?
- Czerwinski, et al. (2004) diary-based study on task- switching --> 40% of the task switches reported were self-initiated and did not involve proceeding to the next logical task.
- Emotional states:
- switch activities so as to attain an emotional equilibrium
- self interrupt to move to a particular activity that enables them to continue their current attentional state and reduce tension from the current activity
- Law of diminishing returns = in all productive processes, adding more of one factor of production, while holding all others constant, will at some point yield lower per-unit returns
- multitasking is often the rational thing to do --> by sharing our time between different tasks we maximise productivity/reward
- Jarvstad et al. (2012) knowing when to move on from a task:
- compared timing decisions for a perceptual task with timing decisions for more cognitive tasks --> performance highly similar across tasks = knowledge can be acquired, and used to make timing decisions, in an equally efficient way regardless of whether that knowledge is derived through perceptual or cognitive experience
Gonzales and Mark (2004): observational study (in-situ) at an investment management company
- used participant observations and long interviews of 14 people over 7m. Day-to-day operations of one team in charge of developing, testing and supporting major financial software modules to be used by the client
- Found info work is very fragmented --> people spend an average of 3mins working on any single event before switching to another event. Informal interactions average four and a half minutes each.
- People spend on average somewhat more than two minutes on any use of electronic tool, application, or paper document before they switch to use another tool.
- People interrupt their work themselves (internal interruptions) about as much as they are interrupted by external influences.
- Most interruptions are due to face-to-face interactions
- argue that it makes sense to understand how time is distributed among working spheres, activities that are thematically connected for the individual --> found that working spheres are also highly fragmented: people spend on the average 11.5mins in continuous work on a project or theme before they switch to another.
- Mechanisms should be flexible to enable people to group particular documents and applications but at the same time there should be recognition that many applications are shared among working spheres (e.g. an e-mail client or scheduling tools).
- when working spheres use very different information resources and applications, it will be convenient to have mechanisms that save the state of the information device particular to that working sphere, making it easier to resume work.
2 lags associated with an interruption:
- 1) interruption lag between alert for secondary task and beginning secondary task = time to start interrupted task
- 2) Resumption lag between end of secondary task and resume of primary task = time to resume the primary task
- Memory for goals (Altmann & Tafton, 2002) --> goal-directed behaviour requires maintenance of problem state information and problem state information is represented as declarative memory elements.
-Declarative memory is defined by activation and activation declines over time, so may fall below the retrieval threshold = memory elements decay over time unless they are actively rehearsed = resumption errors will occur
= interruption duration effect
- Hodgett & Jones (2006): people are slower to resume Tower of London task when it has been suspended for longer
- Monk et al. (2008): longer interruptions led to slower task resumptions of VCR task --> because the activation level of the retrieved item is closer to the interference level
- Trafton et al. (2011) have focused on sequence errors made in the execution of routine data-entry tasks. interruptions -> increase in frequency of errors compared to noninterruption trials. most of the incorrect actions were in close proximity to the correct step (i.e., participants were most likely to either skip/repeat one step ahead/behind).
- = some of the negative effects of interruptions can be mitigated by active goal rehearsal --> can be before an interrupting task is attended to (Trafton et al., 2003) or concurrently with the interrupting task (Salvucci, Monk, & Trafton, 2009).
Brumby et al. (2013) 2 expts. where p's were interrupted by a cognitively demanding secondary mental arithmetic task while working on a routine sequential data-entry task
- Expt 1: time cost of making a resumption error following an interruption was varied (investigate whether tasks resumed more/less quick depending on relative error cost)
- Expt 2: forced slower task resumptions by way of a post-interruption lockout that stops the participant from resuming the primary task immediately after dealing with the interruption --> find out whether more/fewer errors would be made
- following interruption period had to recall what it was they were supposed to do next --> would be difficult because the sequential data-entry task interface does not provide explicit cues to aid successful resumption and the high cognitive load of the interrupting task would limit active goal rehearsal during the interruption period itself (Cades et al., 2007)
- 1) when resumption errors carried a sig. time-cost penalty, participants made fewer errors and resumed the primary task more slowly than when errors incurred a low-time cost penalty = speed-accuracy trade-off --> striking result because a number of previous studies have shown that people cannot be easily motivated to make fewer errors in the execution of routine procedural tasks (e.g., Back et al., 2006)
- shows resumption errors different to post-completion errors --> resumption errors reflect an error in memory retrieval
- 2) lockout period = sig. reduction in resumption errors, thought because it prohibited fast, inaccurate task resumptions and allowed them time to recall where they were in the task sequence before the interruption (once lockout over, p's resumed task quickly--> they were preparing to execute planned action)
- consistent with studies that have also shown benefits to using lockout periods to encourage users to use more memory-intensive strategies (Gray et al., 2006) and plan extended sequences of actions (O’Hara & Payne, 1998, 1999; Svendsen, 1991)
= longer resumption lages following an interruption reduce likelihood that resumption errors will be made
- means Altmann and Trafton's (2002) memory for goals theory doesn't easily account for why long resumption lags = fewer resumption errors, because it assumes that memory representations decay over time
- should alter the memory retrieval mechanism so that increasing the time allowed for retrieval would result in fewer resumption errors being made: with increased time, multiple episodic memory representations might be slowly recovered in order to resume the task correctly
- Limitations: within subjects = learning effect, and not naturalistic as lack of control on how and when p's choose to deal with interruptions
- we seem eager to receive notifications; we frequently check phones (Oulasvirta et al., 2012)
- we often switch to other tasks to improve productivity (Jin & Dabbish, 2009)
- evidence that having workers take breaks between tasks improves their performance (Dai et al. 2015) and that
- breaking tasks into smaller chunks improves attentiveness (Cheng et al. 2015)
- we are annoyed when we receive too many notifications, especially during focused work
- interruptions often disrupt activity, hindering performance and provoking errors
- we find it difficult to return to a previous task after having been interrupted
- Gillie and Broadbent (1989) found that the nature and complexity of an interruption affects how much performance will be disrupted.
- O’Connaill and Frohlich (1995) found that 41% of the time people do not resume their original task after an interruption.
- Interruptions have been found to be beneficial as well as disruptive