My Phone and Me: Understanding People's Receptivity to Mobile Notifications
Summary (7 min read)
INTRODUCTION
- Smartphones enable a new form of effortless information awareness.
- Notifications are extremely beneficial to the users: however, at the same time, they are a cause of potential disruption, since they often require users’ attention at inopportune moments.
- The authors believe that interruptibility management systems fail to achieve a very high accuracy in predicting the opportune moment because there is still a lack of understanding concerning the factors influencing the user’s receptivity to mobile notifications in different physical and cognitive situations.
- Finally, the authors observe the substantial role of psychological traits on how a person reacts to a mobile notification, calling for highly personalized interaction between a smartphone and its user.
REASONING ABOUT USERS’ RECEPTIVITY TO MOBILE NOTIFICATIONS
- As suggested by Clark [10], users can respond to an interruption in four possible ways: (i) handle it immediately; (ii) acknowledge it and agree to handle it later; (iii) decline it (explicitly refusing to handle it); (iv) withdraw it (implicitly refusing to handle it).
- 1Certain social norms and expectations from the sender side, however, constrain the flexibility that the receiver has in reacting to a message [30].
- As shown in Figure 2, the authors take three time measurements for each notification: the time of notification arrival (a), the time when the notification is seen (b), and the time when the user accepted (c1) or dismissed (c2) the notification.
- The authors examine the way interruption timing, with respect to the primary task, determines the user’s response to the notification.
DATA COLLECTION
- In order to investigate the nature of disruptive notifications and factors that determine the user’s receptivity to mobile notifications in different physical and cognitive situations, the authors conducted an in-situ field study.
- The My Phone and Me application uses Android’s Notification Listener Service [1] to access notifications and Google’s Activity Recognition API [3] and ESSensorManager [22] to obtain the context information.
- The authors are aware that some notifications are dismissed because they do not require any further action.
- The application did not trigger any questionnaire after 10pm so that the participants do not feel annoyed at responding to the surveys late at night.
- A questionnaire comprises seven multiple-choice and two free-response questions.
Recruitment of the Participants
- The My Phone and Me application was published on the Google Play Store from 12th August 2015.
- It allows users to check statistics on their phone usage and interruptions.
- The authors believe that displaying this information has a minimal interference with users’ actual behavior for interacting with notifications, but provides a valuable functionality in order to make the users keep the application installed on their phones.
- Firstly, the user has to give explicit permission as required by the Android operating system.
- Secondly, the application shows a list of information that is collected and asks for user consent.
DATASET
- The data collection was carried out for around two months, during which the authors collected 19494 notifications, 611 responses for the questionnaire (comprising a set of nine questions listed in Table 2) and 11 personality test results (50 item-based BigFive Factor Markers by Goldberg [16]) from 74 users who installed the My Phone and Me application.
- The authors do not ask them to provide any other demographic information.
- As the authors are primarily using the questionnaire responses, they compared the click rate (i.e., percentage of notifications that are clicked out of total notifications) of the overall notifications with the notifications that were linked to questionnaires.
- Furthermore, in practice, a user might not even receive notifications from each of the sender types during the period of the study.
- Therefore, the authors use the data from 20 users who responded to at least 14 questionnaires, i.e., the minimum number of questionnaires that were answered by users in this set.
UNDERSTANDING RESPONSE TIME
- The key findings of this section are: Users are aware of the notification alerts even when the phone is in silent mode.
- Seen time is fastest when the phone is in vibrate mode and slowest for silent node.
- Notifications are seen fastest when the user is commut- ing and slowest when idle.
- User’s attentiveness increases (reducing the seen time) with the increase in the complexity of an ongoing task.
- The decision time is higher for the notifications from less frequently contacted senders.
The Role of Alert Modality in Perceiving a Notification Alert
- A notification can alert the user by means of vibration, sound and/or flashing LED.
- According to their dataset, when the notifications (with which the questionnaires were linked) were triggered the user’s phone was for 25.54% of the times in the silent mode, 21.50% vibrate mode, 41.94% sound mode and 11.03% sound with vibrate.
- This provides evidence that when the phone is in silent mode users are still aware of the notification alerts.
What Factors Influence the Seen Time?
- The authors investigate the role of alert modality, sender and the ongoing task type, complexity and completion level, in influencing the seen time of a notification.
- A Tukey posthoc test (by setting the ↵ = 0.05) revealed that the seen time is statistically significantly higher for silent notifications (average 7.332 mins).
- The classification was done manually, by two coders who initially disagreed on five entries.
- The Impact of Ongoing Task Complexity on Seen Time.
- Thus, the increase in the seen time of notifications is correlated with the decrease in rating of ongoing task’s complexity.
What Factors Influence the Decision Time?
- The authors analyze the effect of the type, complexity and completion level of the ongoing task, and the sender type on the time a user takes to decide how to react to a notification.
- The authors find that neither of these factors have a statistically significant effect on the decision time of notifications with the exception of the sender.
- A Tukey post-hoc test (by setting the ↵ = 0.05) revealed that out of the 11 sender types (shown in Table 2), notifications from partner lead to the fastest decision time (mean DT is 3.315s), followed by immediate family members with an average decision time of 4.891 seconds.
- In other cases users take more time in reading the content before deciding how to handle it.
- The authors hypothesize that this behavior stems from the content of notifications from close friends or family members, which might be more predictable, and a part of a daily routine (e.g., “pick kids from school").
The Role of Notification Presentation
- In their dataset, 2953 (out of 7795) notifications were received when the user was engaged with the phone.
- Here, a high-priority notification is a foreground notification that gets in the way of the user’s ongoing activity and the user cannot perform any action to get it out of the way without clicking or dismissing it (e.g. Viber messages).
- The authors investigate the effect of the notification presentation on the response time (i.e., the sum of seen time and decision time) of a notification.
- The result of a two sample t-test shows that there is a statistically significant effect of notification priority on the response time, t(2951) = 17.694, p < 0.001, with high-priority notifications getting quicker response than low-priority notifications.
WHY A NOTIFICATION BECOMES DISRUPTIVE
- In this section the authors investigate the effect of different factors on the perceived disruption.
- Since the perceived disruption was measured with a 5-point Likert scale, the authors encode the responses as: Strongly disagree=1, Somewhat disagree=2, Neutral=3, Somewhat agree=4 and Strongly agree=5.
- The key findings of this section are: Perceived disruption increases with the increase in the complexity of an ongoing task.
- Messages from subordinates and system messages (where the sender is not a person) are considered as most disruptive.
- Whereas, extended family members are considered as the least disruptive.
The Role of Ongoing Task Complexity
- The authors investigate whether the complexity of an ongoing task is associated with the perceived disruption reported by the users.
- Tau correlation coefficient was computed to assess the relationship between the ongoing task complexity and perceived disruption.
- This demonstrates that the users are likely to get more disrupted by a notification that arrives when they are engaged in an intricate task and less disrupted when they are performing a simple task.
- In their preliminary analysis [27] the authors have found that when users are engaged in complex tasks they also express more of a negative sentiment towards interruptions.
The Role of Ongoing Task Completion Level
- A one-way ANOVA of the reported disruption was carried out for each class of task completion level (starting, in the middle, finishing and not doing anything).
- The results show that the completion level of an ongoing task has a significant impact on the disruption perceived by the users from the notifications, F(3, 451) = 19.43, p < 0.001.
- A Tukey post-hoc test (by setting ↵ as 0.05) reveals that the perceived disruption is the highest when the user is currently involved in a task.
- The perceived disruption is the lowest when the user is starting a task or idle and there is no statistically significant difference between these groups.
- These results show that the perceived disruption when the user is highly engaged in a task is very high not only from the desktop notification, as discussed for example in [13, 24], but also from the mobile notifications.
The Role of Ongoing Task Type
- A one-way ANOVA of the reported disruption is carried out for each type of ongoing task (see Table 2).
- A Tukey post-hoc test (by setting ↵ = 0.05) revealed that the perceived disruption is the highest when the user is working and the lowest while the user is idle.
- When the users are not idle, they perceive least disruption while communicating and doing a personal or maintenance task.
- Since the communication can involve notifications themselves, e.g. two mobile users exchanging WhatsApp messages, the above result is not surprising.
- As shown in a recent study [31], users are receptive to information when they are bored.
UNDERSTANDING THE ACCEPTANCE OF NOTIFICATIONS
- In this section the authors investigate the factors that make the users accept or dismiss a notification.
- The key findings of this section are: Likelihood of the acceptance of a notification decreases with the increase in the perceived disruption.
- Disruptive notifications are accepted because they con- tain useful information.
Procedure
- Through the questionnaires, the authors asked the users the reason for clicking/dismissing a notification (see Table 2).
- In Table 3 and Table 4 the authors calculate the percentage of times each factor was reported as a reason for clicking and dismissing the notifications.
- Since, users may select more than one option, the total count percentage in the table adds up to more than 100%.
- According to these responses, the users mostly accept notifications when they are free, but also the importance of the sender and the usefulness of the content make them accept a notification.
- On the other hand, users avoid attending to notifications that do not contain important, urgent or useful content.
Disruptive Notifications are Likely to be Dismissed
- The authors examine the impact of the disruption caused by the notifications on their likelihood of being accepted.
- In order to quantify this, the authors encoded the response for perceived disruption with the following values: Strongly disagree=1, Somewhat disagree=2, Neutral=3, Somewhat agree=4 and Strongly agree=5.
- The results indicate the likelihood of the acceptance of a notification decreases by 0.581 times (95% confidence interval limits for the slope were [0.497, 0.675]) for a unit increase in the perceived disruption (based on the 5-point Likert scale).
Why are disruptive notifications accepted?
- As discussed above, the disruption perceived by the user makes a notification more likely to be dismissed.
- Table 5 shows the percentage of times each factor was reported by the users for accepting the disruptive notifications.
- As users were allowed to select more than one option, the sum of the percentages in the table adds up to more than 100.
- "Content is important" and "Content is useful" are the most dominant reason provided by the users for clicking the disruptive notifications.
- The authors suspect that these notifications may contain valuable information, but they were not relevant at the moment of delivery.
DOES PERSONALITY MATTER
- Seen time and decision time of notifications.the authors.
- The authors computed the score for the five personality traits (i.e., the so-called Big Five: Extroversion, Agreeableness, Conscientiousness, Neuroticism and Openness) for each of the 11 users who fully completed the exit questionnaire that includes the 50 questions related to personality traits.
- For this computation, the authors used the scoring instructions that come with the personality test [16].
Impact on Reported Disruption
- To quantify the relationship between the five personality traits and the disruption perceived by the users from notifications, the authors fit a linear regression model with the average disruption as a dependent variable, and the five personality traits as independent variables.
- Here, the average disruption is computed as a mean of disruption reported by the user through the questionnaires.
- All responses were encoded with the following values Strongly disagree=1, Somewhat disagree=2, Neutral=3, Somewhat agree=4, and Strongly agree=5.
- Table 6 shows the parameters of the fitted linear regression model.
- The results show that the extroversion personality trait significantly affects the average perceived disruption and that extroverts are more inclined to be disrupted by a notification.
Impact on Notification’s Seen Time and Decision Time
- The authors then investigate whether the personality traits influence the seen and decision time.
- The average decision time of notifications for each user: the average time taken by a user to click/dismiss after viewing a notification.
- The parameters of the fitted linear regression model are shown in Table 7.
- The parameters of the fitted linear regression model are shown in Table 8, and demonstrate that the decision time for a notification is significantly influenced by the user’s Extroversion and Neuroticism personality traits.
- The above results show potential for the interruptibility models to be gen- eralized across groups of users who share the same personality traits.
Deferring Notifications
- Previous studies show that users perceive more disruption from notifications when engaged in intricate tasks and for the first time the authors confirm this for mobile settings in a quantitative way.
- Thus, in order to benefit users, the OS should offer more flexibility to them for setting the busy moments so that only time-critical notifications could be triggered.
- This would potentially allow interruptibility management (IM) systems to learn patterns to predict the user’s engagement with complex tasks and prioritize interruptions accordingly.
Improving Notification Presentation
- The authors found that users become more attentive at busy moments, but are likely to perceive most notifications as disruptive and dismiss them.
- Disruptive notifications tend to be accepted if they contain useful content.
- The presentation of notification summaries could be adapted to help users quickly decide whether to click or dismiss notifications, e.g. by highlighting notifications from priority contacts or with priority content that is learnt over time by an IM system.
Building a Personality-Dependent Interruptibility Model
- The authors observed that the perceived disruption, seen and decision time are influenced by the user’s personality traits.
- This demonstrates the potential to take the personality trait into account in interruption models and, for example, generalise interruptibility models across groups of users who share the same personality traits.
- These findings can be exploited to design more effective machine learning algorithms for intelligent notifications.
LIMITATIONS
- Most limitations of the work presented in this paper stem from their decision to collect data in the wild, with the minimum amount of intervention from their users.
- Moreover, in case a notification arrives when the user is already engaged with the phone, the authors assume that the user has seen the notification.
- Further, since their users are not confined to a laboratory, the authors are limited to self-reported level of disruption from a notification.
- The density of notifications negatively impacts the sentiment towards individual notifications [28].
- Finally, while the work is the first to their knowledge to uncover the role of individual psychological traits on mobile interruptibility, it is important to note that the authors related the traits with the reported interruptibility.
CONCLUSIONS
- The contributions of this study are threefold.
- First, the authors have confirmed the validity of some past desktop interruptibility studies in a mobile setting.
- The authors have analysed the data to show that the response time of a notification in the mobile environment is not only influenced by an ongoing task’s type, completion level and task complexity, but also by the notification’s alert modality, presentation and sender-recipient relationship.
- Moreover, the relationship with the sender influences the user’s decision on accepting a notification or not.
- Finally, different people exhibit different reactions and the authors observe a substantial role of the individual psychological traits on how a person reacts to a mobile notification.
Did you find this useful? Give us your feedback
Citations
2,071 citations
232 citations
177 citations
Cites background from "My Phone and Me: Understanding Peop..."
...In addition to reducing participant burden, recent developments in human-computer interaction reveal the possibilities of using data captured by mobile phone sensors to identify ideal timing, or the “opportune moments” [54-57], to send prompts in order to minimize interruption or participant’s engagement in competing activities....
[...]
135 citations
125 citations
References
4,777 citations
"My Phone and Me: Understanding Peop..." refers methods in this paper
...For this computation, we used the scoring instructions that come with the personality test [16]....
[...]
...The data collection was carried out for around two months, during which we collected 19494 notifications, 611 responses for the questionnaire (comprising a set of nine questions listed in Table 2) and 11 personality test results (50 item-based BigFive Factor Markers by Goldberg [16]) from 74 users who installed the My Phone and Me application....
[...]
...Additionally, the My Phone and Me application asks the users to take a personality test based on the 50 item Big-Five Factor Markers from the International Personality Item Pool, de- veloped by Goldberg [16]....
[...]
2,071 citations
959 citations
"My Phone and Me: Understanding Peop..." refers background in this paper
...At the same time, studies have shown that users cannot ignore their smartphones for a long time, because they start feeling stressed and anxious about missing important information until they finally pick up the phone to check for any new notifications [26]....
[...]
662 citations
"My Phone and Me: Understanding Peop..." refers background in this paper
...Multitasking is fundamental in workplaces, where task switching happens every few minutes [17], but also in the private sphere, where an increasing number of personal computing devices mediate the flow of data, be it of entertainment, social or informative nature....
[...]
652 citations
Additional excerpts
...To explain why the interruptions are disruptive, Altmann and Trafton propose the Memory for Goals model that explains how users’ intention move the necessary mental state of the problem between the foreground and the background of their attention and how such state deteriorates when kept in the background [6]....
[...]