Conference PaperPDF Available

Short-Form Videos Degrade Our Capacity to Retain Intentions: Effect of Context Switching On Prospective Memory

Authors:

Abstract

Social media platforms use short, highly engaging videos to catch users' attention. While the short-form video feeds popularized by TikTok are rapidly spreading to other platforms, we do not yet understand their impact on cognitive functions. We conducted a between-subjects experiment (N= 60) investigating the impact of engaging with TikTok, Twitter, and YouTube while performing a Prospective Memory task (i.e., executing a previously planned action). The study required participants to remember intentions over interruptions. We found that the TikTok condition significantly degraded the users' performance in this task. As none of the other conditions (Twitter, YouTube, no activity) had a similar effect, our results indicate that the combination of short videos and rapid context-switching impairs intention recall and execution. We contribute a quantified understanding of the effect of social media feed format on Prospective Memory and outline consequences for media technology designers to not harm the users' memory and wellbeing.
Short-Form Videos Degrade Our Capacity to Retain Intentions:
Eect of Context Switching On Prospective Memory
Francesco Chiossi
LMU Munich
Munich, Germany
francesco.chiossi@i.lmu.de
Luke Haliburton
LMU Munich
Munich, Germany
luke.haliburton@i.lmu.de
Changkun Ou
LMU Munich
Munich, Germany
research@changkun.de
Andreas Butz
LMU Munich
Munich, Germany
butz@i.lmu.de
Albrecht Schmidt
LMU Munich
Munich, Germany
albrecht.schmidt@i.lmu.de
ABSTRACT
Social media platforms use short, highly engaging videos to catch
users’ attention. While the short-form video feeds popularized by
TikTok are rapidly spreading to other platforms, we do not yet
understand their impact on cognitive functions. We conducted a
between-subjects experiment (
𝑁=
60) investigating the impact
of engaging with TikTok, Twitter, and YouTube while performing
a Prospective Memory task (i.e., executing a previously planned
action). The study required participants to remember intentions
over interruptions. We found that the TikTok condition signi-
cantly degraded the users’ performance in this task. As none of
the other conditions (Twitter, YouTube, no activity) had a similar
eect, our results indicate that the combination of short videos and
rapid context-switching impairs intention recall and execution. We
contribute a quantied understanding of the eect of social media
feed format on Prospective Memory and outline consequences for
media technology designers to not harm the users’ memory and
wellbeing.
CCS CONCEPTS
Human-centered computing
Empirical studies in ubiq-
uitous and mobile computing; Smartphones.
KEYWORDS
Prospective Memory, Social Media, Digital Wellbeing, TikTok
ACM Reference Format:
Francesco Chiossi, Luke Haliburton, Changkun Ou, Andreas Butz, and Al-
brecht Schmidt. 2023. Short-Form Videos Degrade Our Capacity to Retain
Intentions: Eect of Context Switching On Prospective Memory. In Proceed-
ings of the 2023 CHI Conference on Human Factors in Computing Systems
(CHI ’23), April 23–28, 2023, Hamburg, Germany. ACM, New York, NY, USA,
14 pages. https://doi.org/10.1145/3544548.3580778
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for prot or commercial advantage and that copies bear this notice and the full citation
on the rst page. Copyrights for components of this work owned by others than the
author(s) must be honored. Abstracting with credit is permitted. To copy otherwise, or
republish, to post on servers or to redistribute to lists, requires prior specic permission
and/or a fee. Request permissions from permissions@acm.org.
CHI ’23, April 23–28, 2023, Hamburg, Germany
©2023 Copyright held by the owner/author(s). Publication rights licensed to ACM.
ACM ISBN 978-1-4503-9421-5/23/04. . . $15.00
https://doi.org/10.1145/3544548.3580778
1 INTRODUCTION
Social media has exploded into ubiquity in recent years and is cur-
rently used by more than half the world’s population [
51
]. Through
social media feeds, people are frequently exposed to new informa-
tion at an unprecedented rate. The information rate is also continu-
ing to rise with the growing popularity of TikTok and other short
video platforms, which displays a feed of brief, highly engaging
videos. Although this high information density may be desired as
it is benecial for certain applications or appeals to users due to an
instant gratication eect, it can also have serious negative conse-
quences. Digital information overload causes chronic stress [
43
]
and social media use has been shown to have a detrimental impact
on memory performance [
96
]. We are still only beginning to un-
cover the breadth of the impact that social media can have on our
psychological and cognitive functions.
TikTok’s model of short, engaging videos that cause users to
switch contexts rapidly is spreading across social media platforms
(e.g., Instagram Reels [
48
] and YouTube Shorts [
49
]). The wide-
spread use of such video formats has certain known, negative
consequences, including frequent disruptions, especially at the
workplace [
2
]. Online interruptions are the most common in the
workplace [
14
] and are associated with increased workload, chronic
stress, and mental fatigue [
81
]. Specically, this stream of video
information continuously lls our mental buer, which causes us
to eliminate potentially useful information in favor of more su-
percial or irrelevant information provided by the social media
feed. Past research in workplace contexts has demonstrated that
context switching has a detrimental impact on cognitive functions
and task performance [
65
,
109
]. However, the impact of context
switching in social media is not well understood. Zheng
[125]
found
that watching short videos negatively impacts visual short-term
memory, which suggests that we should better understand and
characterize the impact of dierent social media feed formats on
dierent cognitive functions. Prospective Memory (PM) is a funda-
mental cognitive function that describes our ability to remember
to execute a planned action while doing an unrelated task [
39
]. PM
is highly relevant because it enables productive activities, makes
people eective in knowledge work, and accomplishing daily tasks
(e.g., running errands or remembering to attend a meeting). Conse-
quently, we aim to investigate the research question:
RQ:
How do dierent social media feeds impact prospective mem-
ory?
CHI ’23, April 23–28, 2023, Hamburg, Germany Francesco Chiossi, Luke Haliburton, Changkun Ou, Andreas Butz, and Albrecht Schmidt
In this paper, we investigate how dierent social media feed
interruptions impacts PM. We conducted a between-subjects study
where
𝑁=
60 participants simultaneously executed a lexical deci-
sion (LD) task and a PM task in two sessions with a 10-minute break
in the middle. During the break, participants engaged in one of four
activities, depending on their experimental condition: Rest,Twitter,
YouTube, or TikTok. Following the break, participants returned to
the simultaneous LD and PM tasks to evaluate the impact on their
PM. We hypothesized that TikTok would have the most detrimen-
tal eect on PM because it is multi-modal, highly engaging, and
exposes users to rapid changes in context. Our results show that
TikTok signicantly reduces PM performance relative to Rest. On
the other hand, we found that neither Twitter nor YouTube had any
signicant eect on performance. This paper contributes a quanti-
ed measure of the impact of three social media feed formats on
PM, as well as an initial characterization of which social media feed
format have the most signicant impact on performance. Media
technology designers need to understand the detrimental eect of
exposing users to highly engaging rapid context switches, as poor
PM performance can have drastic consequences for users in their
daily lives.
2 RELATED WORK
In this section we provide an overview of PM functioning and the
impact of interruptions. Finally, we report the eects of social media
distraction on cognitive functions related to PM that motivated our
research.
2.1 Prospective Memory
PM is considered as “the ability to remember to perform a previ-
ously planned action at a precise moment in time or following a
specic event while one is engaged in performing another activ-
ity” [
39
]. It is a cognitive function that involves dierent cognitive
processes, i.e., planning, attention, monitoring, and working mem-
ory. Individuals frequently become involved in multiple ongoing
tasks after formulating an intention and, in most everyday contexts,
cannot retain the intention in focal attention [23].
In the PM paradigm developed by Einstein-McDaniel [
31
], par-
ticipants are informed that if a particular target cue appears while
doing an ongoing activity, e.g., a lexical decision task, they should
execute a distinct action, such as pressing a specic key on the
keyboard. Retrieving the delayed intentions requires a monitor-
ing process mediated by bottom-up and top-down processes [
100
].
However, when we have to retain multiple intentions, we need to
be in a preparatory state and actively monitor for the occurrence
of target cues [
32
,
107
]. Specically, monitoring requires allocating
attentional resources for detecting the cue and memory resources
to maintain and rehearse the intentions to execute [
41
]. Proper PM
functioning is closely linked to productivity and safety [
27
] as it
allows the execution of sequential steps such as programming [
94
],
oce work [
53
], or taking or providing medication at the right time
[93].
Before we introduce the role of interruptions in PM, however,
it is important to observe that, unlike any other memory task, in
PM tasks, the recall of information does not occur as a result of
an explicit request by someone [
25
] (as for example in episodic
memory tasks) but must be produced autonomously by the subject,
in a self-initiated manner [
71
]. Therefore, PM tasks do not only
impose attentional demands but also memory demands for keeping
the intention representation active and retrieving it [37, 108].
However, the cognitive demands of PM activities may not be
limited to remembering intentions and monitoring cues. Therefore,
this study focuses on another set of processes likely involved in
PM performance, which has received little attention from PM re-
searchers so far: the eect of temporary interruption of the ongoing
and PM tasks.
2.2 Interruptions in Prospective Memory
As previously stated, PM paradigms encompass a dual-task nature
in which participants are engaged in an ongoing activity while
being asked to act upon perceiving a specic target cue. However,
outside laboratory settings, delays and interruptions frequently
prohibit a person from carrying out an intended action after it is
retrieved. Interruptions are pervasive in everyday life and at work,
but there is a signicant gap in our understanding of how such
disruptions impair PM [
69
]. Specically, in this work, we will fo-
cus on external interruptions. External sources of interruptions
include face-to-face meetings [
80
], instant messenger chats [
40
],
workplace design features [
82
] and more. These examples of exter-
nal interruptions also dier in the channel of interaction (i.e., direct
or via technology [
72
]), the sensory channels involved [
60
] or their
information richness [18].
Multiple studies have demonstrated that after task interruption,
task goals fade from memory, resulting in a long time to resume
and complete the interrupted task, negatively impacting perfor-
mance [
1
,
6
,
76
]. The detrimental behavioral impact of interrup-
tions has been explained in terms of memory for goals theory [
5
],
focusing on memory-based deactivation of the interrupted task, or
theory of attention residue [
56
], where the interruption retains at-
tentional resources to some degree away from the user. Ultimately,
the outcome is similar: interruptions have a negative impact on
performance in the task at hand [
57
,
58
]. Those two theories are
intrinsically and functionally connected to how PM works. Inter-
ruptions implicitly require the involvement of PM processes as,
after interruptions, we have to retrieve what we were engaged in
and execute it [28].
When PM processes are interrupted, we are requested to resume
the interrupted task [
28
]. Then, we allocate residual attentional
resources to monitoring prospective intentions [
103
], and to recall-
ing and executing intentions again upon the appearance of the PM
cue [
24
]. This repeated process with either expected memory fading
or reduced attentional resources might induce a failure in these
processes, resulting in forgetting about the halted work [
69
,
92
]. In
summary, interruptions are a cognitive burden that typically aect
interrupted users’ cognitive capacity to complete the interrupted
task eciently.
2.3 Social Media as a Form of Distraction
Distractions are caused by task-irrelevant stimuli that interrupt
goal-directed behavior [
19
]. Social media distraction is the process
through which social media cues attract people’s attention away
from the task at hand (e.g., working). Social media interruptions
Short-Form Videos Degrade Our Capacity to Retain Intentions CHI ’23, April 23–28, 2023, Hamburg, Germany
dier signicantly from workplace interruptions as they are more
frequent (easily over 100 per day), more complex, and less pre-
dictable [
29
]. This situation might regularly arise due to external
(e.g., persistent notications) or internal cues (e.g., unanswered
messages) [
119
]. Just as the above-mentioned interruptions, social
media distractions also seem to aect memory performance, as
shown in free recall tasks [
36
,
99
]. Those results are consistent with
the idea that social media-induced distractions may harm memory
functioning [
35
]. This detrimental eect might be explained with
theories of memory for goals and attentional residue, as attentional
disengagement interferes with memory encoding [45].
Prior work in HCI has investigated methods for assisting users
in using their phones more intentionally (e.g., [
44
,
111
]), thereby
avoiding being drawn in by the engaging strategies of social media
designers. However, absentminded scrolling continues to be an
issue [
63
], motivating us to investigate the impact of social media
feed designs.
More recently, social media apps started to employ dierent de-
sign strategies, e.g., video autoplay, pull-to-refresh, innite scrolling,
and recommendations, to maximize user engagement and attention
capture [
61
]. These designs create an immediate reward loop by
showing content personalized to the user’s subjective preference
and interest [
123
], based on prior browsing histories and tagged
video classication [
10
,
17
]. The fast pace of switching between
topics, which in the case of short form videos ranges from 15 to 60
seconds, and their often emotional content makes the social media
distractor increase attentional disengagement [
112
,
120
], and there-
fore impairs our capacity to timely and accurately resume the task
at hand. However, there have only been a small number of studies
on problematic short form video viewing habits, partly because
short form video apps have only recently begun to proliferate.
The mechanisms underlying these eects are still not fully un-
derstood. However, recent work showed how a fast information
rate might be linked to cognitive performance impairment in dual-
task [
113
] and working memory settings [
29
,
125
]. Dual-task set-
tings and working memory are intrinsically related to PM, as PM
encompasses multiple tasks to execute, and information to be kept
updated in mind. Further research on the relationship between
new forms of social media content and executive functioning is
necessary, given their increasing pervasiveness and popularity in
our daily life [
51
,
67
]. In this study, we investigated the eect of
dierent social media feeds interruptions on executing previously
planned intentions in a dual-task setting.
3 METHOD
We conducted a lab study using a between-subjects design with
four Interruption conditions Rest,Twitter,YouTube, or TikTok.
The four conditions consist of three popular social media platforms
with varying engagement styles and media formats as well as a
control condition (Rest) where participants are requested to not
engage with any social media and do not perform any other actions.
We selected three social media platforms with large variation in
feed format. TikTok and Twitter require users to rapidly switch
contexts, while YouTube generally consists of longer-format videos
and therefore fewer context switches. The media format also varies
between the three platforms. Both TikTok and YouTube are video-
based while Twitter is primarily text-based (with some photos). We
did not include either Instagram or Facebook in this study, because
their feeds are not suciently dierentiable from TikTok and Twitter
(i.e., they have a combination of videos, images, and text).
3.1 Participants
We recruited
𝑁=
60 participants (35 female, 25 male, aged 19-
34,
𝑀=
24
.
80,
𝑆𝐷 =
3
.
40) through a university mailing list and
social media. All participants were uent German speakers (C2 from
CEFR
1
), which was required for the LD task, and reported normal
or corrected-to-normal vision with no history of any neurological
or psychiatric disorders. The participants all had a high school
education or higher, and reported a weekly average screen time
of 2.04 hours (
𝑆𝐷 =
3
.
37) in the Rest condition, in the TikTok
condition 5.57 hours (
𝑆𝐷 =
2
.
25), in the YouTube condition of
6.75 hours (
𝑆𝐷 =
2
.
49), and in the Twitter condition of 5.51 hours
(
𝑆𝐷 =
2
.
45). Moreover, we collected the screen time associated to
the specic condition participants were allocated to. Specically,
participants allocated to the TikTok condition spent an average of
1 hour and 46 minutes per week on the app (
𝑆𝐷 =.
81). YouTube
participants showed an average screen time of 1 hours and 44
minutes (
𝑆𝐷 =
1
.
94) on YouTube, while Twitter participants spent
an average of 55 minutes (
𝑆𝐷 =.
52) on the Twitter social media
app in the week before participation. Participants were randomly
assigned to a condition containing a platform they use frequently in
their daily life. We randomly assigned participants to either the app
with the highest screen time in the previous week or to Rest. We used
this assignment method so that all participants have a personalized
feed on the platform used in their condition. Rather than attempting
to control the content on each feed, which would likely lead to some
participants being uninterested in the content, we opted for this
ecologically situated approach. The only exception to this allocation
procedure was the YouTube condition, where we let participants
choose one video out of ten options. This approach was chosen in
order to control video duration but still allow participants to choose
from a variety of contents e.g., education, music, and entertainment.
3.2 Tasks
We instructed participants to engage in a Lexical Decision (LD)
task and a Prospective Memory (PM) task simultaneously, based on
Cona et al
. [22]
. A large body of studies has demonstrated that this
combination of tasks is suitable for monitoring prospective memory
processes [
22
,
33
,
98
]. Both tasks were conducted on a computer
monitor (Acer Predator XB241YU 23.8 inch, 165 Hz, 2560 x 1440
pixels) with a mouse and keyboard. We covered all of the keys on
the keyboard except for those used in the experiment (Q,W,E,N,M,
and Spacebar). The experiment was created using PsychoPy [83].
The tasks encompassed 160 LD trials and 16 PM trials (10% of
the LD trials) in two dierent blocks (Pre and Post Interruption), for
a total of 320 LD and 32 PM trials. Each experimental block started
with a xation cross (+) with a pseudorandom duration (1250, 1500,
or 1750 ms) at the center of the screen. Then, a string of letters ap-
peared as a stimulus for maximum 3000 ms. Inter-Stimulus-Interval
1
https://www.coe.int/en/web/common-european-framework- reference-
languages/level-descriptions
CHI ’23, April 23–28, 2023, Hamburg, Germany Francesco Chiossi, Luke Haliburton, Changkun Ou, Andreas Butz, and Albrecht Schmidt
was set to 1000 ms. In the LD task, we displayed sequences of letters
on the monitor and participants had to determine whether each
sequence was a valid word or not. They pressed Nfor words and
Mfor non-words with the index nger and the middle nger of
the right hand, respectively. We counterbalanced the key-response
mapping across participants. Word stimuli were extracted from the
SUBTLEX-DE database [
16
], with a word length ranging from six to
eight letters. Psycholinguistic properties, such as the mean length
and frequency, were matched across experimental sessions [
15
].
Non-words were pseudo-word stimuli created from the used words
by changing one or two letters. Participants were required to re-
spond to word stimuli as fast and accurately as possible. The PM
task is depicted in Figure 1.
Figure 1: PM paradigm. Participants performed a lexical de-
cision task, i.e, performing a decision if one of the present
words was either a word or not, and a PM task, where they
were required to press a specic button in occurrence of a
PM cue word ("blau", "lila" and "grün").
In the PM task, participants were told to remember to carry
out three dierent intentions linked to three dierent PM words.
We instructed participants to press special keys if one of three
keywords appeared on the screen instead of pressing the keys for
the LD task. They were required to press Qfor ’Blau’ (blue), W
for ’Lila’ (purple) and Efor ’Grün’ (green). Between a PM word,
at least 10 LD trials appeared before the rst PM cue and at least
8 LD trials between each pair of PM cues. A practice block with
just the LD task (ve word/non-word trials) was given at the start
of the experiment. Participants used their personal mobile devices
and headphones during the intermission tasks. We implemented
all surveys using Google Forms.
3.3 Interruption Conditions
Each participant completed two rounds of the simultaneous LD and
PM tasks with a 10 minute interruption in the middle, which varied
according to their experimental condition. The four conditions for
the interruption are as follows:
Rest: Participants took a break with no input. We instructed them
not to look at their phones or any other screen.
Twitter: Participants were instructed to scroll through their own
Twitter feed for the entire interruption. A Twitter feed consists
of short texts with occasional photos. The feed switches contexts
rapidly, but does not contain highly engaging video content.
YouTube: We prepared a playlist of 10 minute YouTube videos
in advance from a range of topics including entertainment and
education (e.g., TED Talks). The participants were told to choose
the video that was most interesting to them and watch it for the
entire interruption. Although participants did not view their own
personalized YouTube feeds, this method enabled us to control
for the length of the videos, while still giving them some choice
in content. The YouTube experience generally consists of longer-
format videos with higher production value. By limiting the users to
a single video, they do not switch contexts during the interruption.
TikTok: Participants were instructed to watch videos on their own
TikTok feed for the entire interruption. A TikTok feed consists of
brief videos with sound. The feed is highly engaging and switches
contexts rapidly.
3.4 Procedure
After introducing the study and obtaining informed consent, each
participant proceeded through a beginning survey, four experimen-
tal steps, and nally an ending survey, as shown in Figure 2.
In the beginning survey, we asked participants which social
media platforms they use and the associated screen time. This
information was used to assign participants to a study condition
(Rest,Twitter,YouTube, or TikTok).
First, in the training stage, participants performed the combined
LD and PM task for 10 trials. This stage ensured that the partici-
pants understood the requirements before conducting a measured
task. In TaskPre, participants performed the combined LD and PM
tasks for 176 trials. Following Task
Pre
, the participants engaged
in a break condition for 10 minutes. Depending on their assigned
condition, the participants either engaged with a social media feed
or took a break with no phone use. After the break condition, the
experimenter prompted participants to return to the computer and
perform another 160 trials for the LD and 16 trials for the PM
task (Task
Post
). Finally, participants completed an Ending Survey,
reporting their overall screen time.
3.5 Measures
We measured behavioral data and collected responses to both a
Beginning Survey and an Ending Survey. Each of these data sources
require separate analysis methods. The questionnaires are included
in the supplementary material.
In the beginning survey, we recorded which social media plat-
forms each participant uses (“Which social media platform do you
use between Twitter, TikTok and Youtube?”). Specically, we asked
participant to report how much time they spend in the three social
Short-Form Videos Degrade Our Capacity to Retain Intentions CHI ’23, April 23–28, 2023, Hamburg, Germany
Ending
survey
Beginning
survey
Twier
YouTube
TikTok
Rest
Figure 2: Overview of the experiment timeline. Participants complete a Beginning Survey and are then assigned to one of
4 conditions. After a training phase, participants then perform LD and PM tasks in Task
Pre
, followed by an Interruption
depending on their condition assignment (Rest,Twier,YouTube, or TikTok). Finally, the participants perform the LD and PM
tasks again in TaskPost and complete an Ending Survey.
media apps of interest. This information was used to assign partici-
pants to an appropriate condition, i.e., to the social media app they
spent the most time with, and demographic information. Regarding
the behavioral performance, we measured accuracy and reaction
times (RTs) for both the LD and PM tasks. Lastly, we analyzed the
standard scales (BSMAS for social media addiction [
75
], and SUQ-A
for absent-minded phone use [
66
]) within the questionnaires ac-
cording to their original documentation. We additionally collected
Likert-scale responses on engagement and an Ending Survey for
collecting screen time data (“Please enter your daily average screen
time of the last week (search for Screen Time on iPhone or go to
Digital Wellbeing on Android)”).
3.6 Analyzing Performance Trade-O
We analyzed the reaction time distribution for the correct and
error responses as two dierent distributions [
12
]. To quantify
the behavioral dierences between correct and error responses,
we adopted a Drift-Diusion Model (DDM) [
86
], which has been
leveraged in cognitive science [
46
] and computer graphics [
30
] to
model perceptual decision-making. The DDM model also has been
suggested to model choice and non-choice in LD and PM tasks [
11
,
20
,
115
]. We chose this modeling approach as it is informative
on dierent processing components relevant to PM. Traditional
analysis approaches investigated RTs and task accuracy separately,
proting of only a subset of the available data, while the diusion
model is applied to the joint distribution of RT and accuracy data.
DDM allows for the decomposition of RTs and accuracy into a set
of latent parameters that represent underlying cognitive processes
such as task load [
7
], evidence accumulation [
106
], and decreased
cognitive capacity [
110
]. The DDM assumes that decisions are made
by a Wiener process that accumulates cognitive evidence over time
from a starting point towards two response boundaries.
A DDM model includes 4 parameters: the drift rate (
𝜇
), the de-
cision bound (
𝐵
), the non-decision time (
𝑡
) and variance (
𝜎
) com-
ponent. As an interpretation, the drift rate informs the speed and
direction of information accumulation. The drift rate can be inter-
preted as a measure of subjective task diculty: higher (absolute)
drift rates indicate less demanding tasks. Second, the decision bound
characterizes the time needed to make a decision, and with a larger
decision bound, more eort is expected to form a decision. Here,
smaller values imply shorter information uptake and increased er-
roneous responses. Third, the non-decision time captures the time
spent for stimulus processing, but unrelated to the decision, such as
perception of the target stimulus or execution of the response and
stimulus encoding time. Lastly, the variance component models
the uncertainty of a decision process, thus, an increased variance
could result in more diverse reaction times needed to perform a
decision when the PM stimulus is presented. Several studies have
validated these parameters as sensitive to dierent experimental
manipulations, lending credence to their validity [3, 116, 117].
4 RESULTS
In this section, we rst present results on behavioral accuracy us-
ing a Linear Mixed Model (LMM) approach. Second, we employ a
Generalized Linear Mixed Model (GLMM) to investigate dierences
in the reaction times distributions. Finally, depending on normality,
evaluated by the Shapiro-Wilk test [
88
], we report two-way mixed
ANOVA results for parameter analysis on the tted DDM parame-
ters, or ART ANOVAs [
121
] for the non-parametric data. To analyze
subjective responses collected from the engagement, SUQ-A, and
BSMARS questionnaires, we performed a mixed ANOVA on the
Interruption eect, with an additional Bayes factor ANOVA if
the results were non-signicant.
CHI ’23, April 23–28, 2023, Hamburg, Germany Francesco Chiossi, Luke Haliburton, Changkun Ou, Andreas Butz, and Albrecht Schmidt
0.0
0.2
0.5
0.8
1.0
Density (Pre-Interruption)
$Rest Twier ÅYouTube TikTok
Correct
Error
0 1000 2000 3000
RTs (ms)
0.0
0.2
0.5
0.8
1.0
Density (Post-Interruption)
0 1000 2000 3000
RTs (ms)
0 1000 2000 3000
RTs (ms)
0 1000 2000 3000
RTs (ms)
(a) LD Tasks
0.0
0.3
0.6
0.9
1.2
Density (Pre-Interruption)
$Rest Twier ÅYouTube TikTok
Correct
Error
0 1000 2000 3000
RTs (ms)
0.0
0.3
0.6
0.9
1.2
Density (Post-Interruption)
0 1000 2000 3000
RTs (ms)
0 1000 2000 3000
RTs (ms)
0 1000 2000 3000
RTs (ms)
(b) PM Tasks
Figure 3: An overview of reaction time distributions, sepa-
rated by correct and error responses in the LD and PM Task.
The upper row (pre-interruption) shows reaction times be-
fore the interruption. The dierence that are shown are due
to the random assignment and are not linked to the condition.
the lower row (post-interruption) shows the reaction time
after the interruptions in which participants experienced
dierent conditions. The blue distributions show reaction
time associated with correct responses and the red with er-
ror responses. A drastic increase in error responses can be
seen in the TikTok Post-Interruption PM trials (bottom right
diagram).
4.1 Behavioral results
To give an overview of the reaction time distributions, we visualized
the distribution for correct and error responses across dierent
interruption conditions both in LD and PM tasks (see Figure 3).
A drastic increase in error responses can already be seen in the
TikTok Post-Interruption PM trials (bottom right diagram). Below,
we report our previously described statistical analysis of these
distributions.
4.1.1 Behavioral Accuracy. As shown in Figure 4, we inspected the
response accuracy (total number of correct key presses divided by
the total number of key presses) of participants2.
Lexical Decision Task. We conducted an LMM (
𝑅2=
0
.
67) to pre-
dict accuracy with interruption (formula:
accuracy interrupt
+ (1|user_id))
guided by REML and an nloptwrap optimizer and
BIC criteria [
8
,
84
]. The model’s intercept corresponding to Tik-
Tok post interruption (
𝐶𝐼95% =[
0
.
92
,
1
.
04
], 𝑡114 =
30
.
42
, 𝑝 <.
001).
We found non-signicance when comparing to other interruption
conditions: Rest (
𝛽=.
001
, 𝐶𝐼 95% =[−
0
.
09
,
0
.
09
], 𝑡114 =
0
.
03
, 𝑝 =
.
975), Twitter (
𝛽=.
005
, 𝐶𝐼 95% =[−
0
.
09
,
0
.
10
], 𝑡114 =
0
.
11
, 𝑝 =.
913),
and YouTube (
𝛽=
0
.
07
, 𝐶𝐼 95% =[−
0
.
16
,
0
.
02
], 𝑡114 =
1
.
47
, 𝑝 =
.
144) interruption conditions. Those results showed how any of the
investigated interruption conditions impact the behavioral accuracy
in the LD task.
Prospective Memory Task. Similarly, for PM task, we tted an
LMM model (
𝑅2=
0
.
67) to predict accuracy with interruption con-
dition and measure (formula:
accuracy interrupt * measure
+ interrupt + measure + (1|user_id)
). In the comparison
with TikTok post (
𝐶𝐼95% =[
0
.
40
,
0
.
58
], 𝑡110 =
10
.
94
, 𝑝 <.
001) in-
terruption accuracy, we found signicant and positive results for
Rest (
𝛽=
0
.
46
, 𝐶𝐼 95% =[
0
.
33
,
0
.
58
], 𝑡110 =
7
.
25
, 𝑝 <.
001), Twit-
ter (
𝛽=
0
.
49
, 𝐶𝐼 95% =[
0
.
36
,
0
.
61
], 𝑡110 =
7
.
77
, 𝑝 <.
001), YouTube
(
𝛽=
0
.
34
, 𝐶𝐼 95% =[
0
.
22
,
0
.
47
], 𝑡110 =
5
.
42
, 𝑝 <.
001), as well as
pre interruption condition (
𝛽=
0
.
31
, 𝐶𝐼 95% =[
0
.
22
,
0
.
41
], 𝑡110 =
6
.
70
, 𝑝 <.
001). Therefore, we report that in the PM task, behavioral
accuracy after interruption signicantly dropped only in the TikTok
condition, while it remained stable across other conditions.
Behavioral accuracy Comparison between LD and PM tasks. Lastly,
we tted an LMM model (
𝑅2=
0
.
38) to predict accuracy with in-
terrupt and task (formula:
accuracy interrupt * task +
interrupt + task + (1|user_id)
). We found only PM task on
TikTok interruption condition is statistically signicant and nega-
tive (
𝛽=
0
.
28
, 𝐶𝐼 95% =[−
0
.
39
,
0
.
17
], 𝑡230 =
5
.
01
, 𝑝 <.
001) com-
paring to LD task (
𝐶𝐼95% =[
0
.
92
,
1
.
04
], 𝑡230 =
31
.
80
, 𝑝 <.
001) and
other PM tasks (Twitter:
𝛽=
0
.
005
, 𝐶𝐼 95% =[−
0
.
11
,
0
.
10
], 𝑡230 =
0
.
10
, 𝑝 =.
923; YouTube:
𝛽=
0
.
03
, 𝐶𝐼 95% =[−
0
.
13
,
0
.
08
], 𝑡230 =
0.46, 𝑝 =.648).
In sum, the abovementioned results show that in the TikTok
post-interruption trials, participants produced signicantly more
errors in the PM task, whereas the accuracy stayed robust in the
LD task.
2
We further investigated trial by trial accuracy responses by means of a binomial
regression accounting for per item and per participants eects. Results were completely
consistent with the LMM model t for PM and LD tasks. Results from this additional
analysis are available in the supplementary material.
Short-Form Videos Degrade Our Capacity to Retain Intentions CHI ’23, April 23–28, 2023, Hamburg, Germany
$Rest Twier ÅYouTube TikTok
0
20
40
60
80
100
Response Accuracy (%)
Lexical Decision Task
$Rest Twier ÅYouTube TikTok
Prospective Memory Task
Taskpre
Taskpost
Figure 4: Comparison regarding response accuracy. The left gure visualizes pre- and post interruption in Lexical Decision task
for dierent interruption conditions; the right gure visualizes for Prospective Memory task. The response accuracy barely
changes (note the y-axis scale), whereas it drops dramatically for the PM task in the TikTok interruption condition.
4.1.2 Reaction Times.
Lexical Decision Task. In the LD task, we tted a GLMM using
REML and an nloptwrap optimizer on raw RTs considering Inter-
ruption as xed eect and participant and stimulus item (word
stimulus) as a random eects. We performed this analysis on RTs
in both pre and post Interruption. We selected formula
rt
interrupt * measure + interrupt + measure + (1|user_id)
+ (1|stimulus)
for the GLMM with Gamma log link function [
59
],
and guided by BIC criteria [
8
,
84
]. However, we did not report any
signicant results (See supplementary materials or Section 7).
Prospective Memory Task. Using a similar approach, we con-
ducted LMM with similar settings to PM task on raw RTs for both
pre and post Interruption. Here, results mimicked the ones for LD
as we did not report any signicant dierence (See supplementary
materials or Section 7). The distribution of RTs for both correct and
erroneous responses is depicted in Figure 3.
4.1.3 DDM. These observed behavioral accuracy results motivated
us to further inspect participants’ decision behavior in the LD
and PM tasks using DDM for further interpretation. We used Py-
DDM [
101
] to t responses in the LD and PM tasks per participant.
In sum, ANOVAs could not nd any signicance in all model pa-
rameters for the LD task. However, for the PM task, ANOVAs show
signicant dierences in the drift rate, variance, and decision bound
in dierent interruptions, and signicant dierences in the variance
and non-decision time in pre- and post-interruptions. For detailed
results, see Table 1.
Since LD tasks were non-signicant, we only ran ART ANOVA
post-hoc comparisons on PM tasks. Figure 5 shows the visualized
results of these comparisons. We found signicant dierences on
𝜇PM
for TikTok pre- and post-interruption contrasts comparison
(
𝑑 𝑓 =
56
, 𝑆𝐸 =
8
.
456
, 𝑡 =
4
.
683
, 𝑝 <.
001), and signicant dif-
ferences on
𝜇PM
across interruption conditions: TikTok vs. Rest
(
𝑑 𝑓 =
91
.
45
, 𝑆𝐸 =
11
.
66
, 𝑡 =
4
.
18
, 𝑝 =.
002), TikTok vs. Twitter
(
𝑑 𝑓 =
11
.
66
, 𝑆𝐸 =
91
.
45
, 𝑡 =
4
.
67
, 𝑝 <.
001), TikTok vs. YouTube
(
𝑑 𝑓 =
91
.
45
, 𝑆𝐸 =
11
.
66
, 𝑡 =
3
.
54
, 𝑝 =.
014). For
𝜎PM
, we also
found signicant dierences for Twitter (
𝑑 𝑓 =
56
, 𝑆𝐸 =
10
.
184
, 𝑡 =
3
.
283
, 𝑝 =.
044), and signicant dierences on
𝑡PM
for YouTube
(
𝑑 𝑓 =
56
, 𝑆𝐸 =
8
.
470
, 𝑡 =
3
.
487
, 𝑝 =.
014). Across interruption
conditions, we found signicant dierences on
𝜎PM
:TikTok vs. Rest
(
𝑑 𝑓 =
106
.
93
, 𝑆𝐸 =
11
.
52
, 𝑡 =
3
.
72
, 𝑝 =.
009), TikTok vs. Twitter
(
𝑑 𝑓 =
106
.
93
, 𝑆𝐸 =
11
.
52
, 𝑡 =
4
.
08
, 𝑝 =.
002), TikTok vs. YouTube
(
𝑑 𝑓 =
106
.
93
, 𝑆𝐸 =
11
.
52
, 𝑡 =
3
.
58
, 𝑝 =.
014), as well as TikTok vs.
Rest (𝑑𝑓 =112, 𝑆𝐸 =12.04, 𝑡 =4.14, 𝑝 =.002) on decision bound 𝐵.
For a closer look into the TikTok condition, we visualized the
DDM model (see Figure 6) in pre- and post-interruptions for all 15
participants. In pre-interruption, participants have a 20.00% error
rate, the tted DDM model (total loss: 457.13) shows a drift rate
𝜇=
1
.
46, variance
𝜎=
1
.
89, decision bound
𝐵=
1
.
94, and non-
decision time
𝑡=
461ms. In the post-interruption, participants have
a total of 50.98% error rate, the tted DDM (total loss: 461.46) has a
drift rate
𝜇=
0
.
000, variance
𝜎=
2
.
35, decision bound
𝐵=
1
.
71, and
non-decision time
𝑡=
681ms. These results show that our partici-
pants tended to give more correct responses (
𝜇=
1
.
46) before the
TikTok interruption. However, after the interruption, participants
had an equally probable decision tendency towards correct and
error responses given a decision (
𝜇=
0
.
000). Furthermore, the non-
decision time increased from 461ms (pre) to 681ms (post), variance
increased from 1.89 (pre) to 2.35 (post), and the decision bound was
reduced from 1.94 (pre) to 1.71 (post).
4.2 Subjective results
We present subjective results separately for the reported engage-
ment, SUQ-A, and BSMARS. Classical statistical inference was sup-
plemented with Bayes Factors (BFs). This was done to establish
the equivalence of interruptions on some of the dependent vari-
ables, which can be seen as a way of conrmatory testing of the
Null-hypothesis [114].
CHI ’23, April 23–28, 2023, Hamburg, Germany Francesco Chiossi, Luke Haliburton, Changkun Ou, Andreas Butz, and Albrecht Schmidt
Table 1: An overview of analyzed results using two-way ANOVAs on tted DDM parameters. Signicant results are highlighted
in bold font. In terms of interruption, the results indicate signicant dierences in the drift rate
𝜇
, variance
𝜎
, and decision
bound
𝐵
in the PM task but no signicance in the LD task. For pre- and post-interruption, the ANOVA showed a signicant
dierence in variance 𝜎and non-decision time 𝑡in the PM task but no signicance in the LD task.
Interruption Pre-Post Interruption ×Pre-Post
𝑓 𝑑 𝑓 𝐹 𝑝 𝜔2/𝜂2𝑓 𝑑 𝑓 𝐹 𝑝 𝜔 2/𝜂2𝑓 𝑑𝑓 𝐹 𝑝 𝜔 2/𝜂2
𝜇PM 3 56 4.078 .011 0.133 1 56 2.420 .125 0.024 3 56 6.466 .001 0.215
𝜎PM 3 56 4.233 .009 0.139 1 56 12.593 .001 0.167 3 56 4.712 .005 0.157
𝐵PM 3 56 3.020 .037 0.092 1 56 0.025 .874 0.017 3 56 2.385 .079 0.065
𝑡PM 3 56 1.695 .179 0.034 1 56 15.851 <.001 0.204 3 56 2.692 .055 0.078
𝜇LD 3 56 2.520 .067 0.119 1 56 1.033 .314 0.018 3 56 0.785 .508 0.040
𝜎LD 3 56 1.615 .196 0.03 1 56 1.591 .212 0.010 3 56 0.697 .558 0.015
𝐵LD 3 56 1.812 .155 0.039 1 56 0.069 .794 0.016 3 56 0.398 .755 0.031
𝑡LD 3 56 1.517 .220 0.025 1 56 0.607 .439 0.007 3 56 0.636 .595 0.019
$ Æ
0
1
2
3
4
5
6
7
8
***
*
***
**
Dri (µ)
$ Æ
0.5
1.0
1.5
2.0
2.5
3.0
3.5
4.0
4.5
*
*
**
**
Variance (σ)
$ Æ
0.6
0.8
1.0
1.2
1.4
1.6
1.8
2.0
2.2**
Bound (B)
$ Æ
0.2
0.4
0.6
0.8
1.0
*
Non-decision Time (t)
pre
post
Figure 5: Post-hoc comparisons regarding tted DDM parameters in the PM task. A larger drift rate
𝜇
means a stronger decision
tendency toward correct responses, a larger variance
𝜎
means higher uncertainty in a decision, and a larger bound
𝐵
requires
more eort to form a decision. A larger non-decision time means less eciency to start a decision.
4.2.1 Engagement. We performed a mixed one-way ANOVA on the
subjectively reported engagement in each condition. We could not
nd any signicant dierences between Interruption conditions
(
𝐹3,56 =
2
.
59
, 𝑝 =.
062). Hence, we further executed a Bayes factor
ANOVA against the Null-hypothesis. The result specically shows
𝐵𝐹01 =
0
.
893, which means that H1 is only 1.12 times more likely
to occur than H0. This result implies insucient evidence that
engagement is dierent between conditions [114].
4.2.2 SUQ-A and BSMARS. We performed a mixed one-way
ANOVA on the reported SUQ-A values but could not nd a sig-
nicant dierence (
𝐹3,56 =
2
.
267,
𝑝=.
091) across the dierent In-
terruption conditions. A subsequent Bayes factor analysis gives
𝐵𝐹01 =
1
.
213, meaning that H0 is only 1.213 times more likely than
H1. Similarly, a mixed one-way ANOVA on the reported BSMARS
could not nd signicant dierences (
𝐹3,56 =
1
.
065,
𝑝=.
371) across
the Interruption conditions. A subsequent Bayes factor gives
𝐵𝐹01 =
3
.
903, meaning that H0 is 3.903 times more likely than
H1 [114].
5 DISCUSSION
We evaluated the impact of dierent social media feed modalities on
PM performance and RTs. In our study, participants were instructed
to simultaneously perform a LD and PM task with an interruption
that varied according to the experimental condition (Rest,Twitter,
YouTube, or TikTok). We did this to understand whether and how
dierent social media feed designs, with varying media modali-
ties and context-switch frequencies, impact cognitive performance
in terms of PM. We will rst summarize our results, then relate
our ndings to the literature on how PM functions and on its rela-
tion with interruptions. We will conclude by summarizing possible
consequences for media technology designers and by highlighting
future work in the eld.
Short-Form Videos Degrade Our Capacity to Retain Intentions CHI ’23, April 23–28, 2023, Hamburg, Germany
0
10
Counts
accuracy: 80.00%
Fied DDM for TikTok Pre-interruption RTs
-2
-1
0
1
2
Decision Process
decision upper bound B= 1.94
decision lower bound B=1.94
non-decision
time t= 461ms
dri µ= 1.46
variance σ= 1.89
0 500 1000 1500 2000 2500 3000
RTs (ms)
0
10
Counts
error: 20.00%
0
10
Counts
accuracy: 49.02%
Fied DDM for TikTok Post-interruption RTs
-2
-1
0
1
2
Decision Process
decision upper bound B= 1.71
decision lower bound B=1.71
non-decision
time t= 681ms
dri µ= 0.00
variance σ= 2.35
0 500 1000 1500 2000 2500 3000
RTs (ms)
0
10
Counts
error: 50.98%
Figure 6: DDM visualizations for speed-accuracy tradeo
before (left) and after TikTok interruption. The blue his-
tograms on top show the distribution of correct responses,
while the red histograms at the bottom show the error re-
sponse distribution. Bold histograms represent correct and
erroneous responses, and transparent histograms are DDM
simulations. The middle gures are DDM simulated decision
process. In TikTok’s pre-interruption, the measured response
accuracy was 80.00%, with a drift rate
𝜇=
1
.
46 towards correct
responses. In the post- Tiktok interruption, the measured
response accuracy drops to 49.02%, and the tted DDM has a
zero drift rate. TikTok’s post-interruption has an equal ten-
dency toward correct and error responses.
5.1 Interpreting the Results
TikTok signicantly degraded PM performance in terms of correct
vs. erroneous responses. In fact, participants in the TikTok condi-
tion were only slightly better than randomly guessing after the
interruption. We can therefore conclude that the TikTok condition
had a signicant negative impact on PM, while neither Twitter nor
YouTube had any observable eect. We originally hypothesized that
TikTok would have a larger detrimental eect on PM than the other
social media formats because it is highly engaging. However, we
found no signicant dierence in subjective engagement scores
between any conditions, so it appears that the eect is not caused
by participants being more engaged in the TikTok feed. Further-
more, we investigated if PM functioning might be impacted by
video format or by rapid changes in context. However, as neither
YouTube nor Twitter degraded PM performance, this result seems to
point towards the detrimental impact of the combination of those
two features, as shown in the TikTok interruption. However, this
result could be caused by other factors. In sum, we found that Tik-
Tok signicantly degrades PM performance, but further research is
required to understand the exact mechanism underlying this eect.
Additionally, we quantied participants’ social media addiction
(BSMARS) and absentminded smartphone use (SUQ-A) to investi-
gate whether the change in PM was inuenced by variations in our
sample populations. However, neither of these measures showed
a signicant dierence across interruption conditions. Therefore,
based on these three results, performance in the TikTok condition
was not aected by measured individual dierences of the partici-
pants but rather by characteristics of the feed itself. It is theoretically
possible that users of specic apps, such as TikTok, might share
specic characteristics or cognitive behaviors, such as a dierential
use of social media that might impact psychological and cognitive
features at dierent levels [
78
,
102
]. Although we found no eect
from dierences in SUQ-A or BSMARS, we cannot exclude that
there might have been hidden mediatory variables, such as daily
negative aect [
99
]. Therefore, future work is necessary to compare
the interruption eect on users unfamiliar with a specic app. This
would further investigate the relationship between PM and spe-
cic social media formats and help uncover the mechanism driving
social media-based PM degradation.
5.2 Impact of Social Media on PM
Our study exposed participants to dierent social media feeds for-
mat. Our main objective was to determine whether social media
interruptions impact PM retrieval and monitoring to identify crit-
icalities for technology design and manage the adverse eects of
social media interruptions. Short-form videos represent multimodal
and emotional stimuli, whose content is often tailored to their con-
sumer [
26
]. They are salient interruptions that quickly divert atten-
tion. Thus, when a PM task is interrupted, people might not have
enough time or be too distracted to resume the intention explicitly.
This is because such dynamic visual and auditory features require
attention focus for eective information processing during video
watching [
74
]. However, such attentional resources demanded by
the video format have been shown to impact the participant’s ca-
pacity to either identify PM cues for triggering intention execution
or keep PM intentions active in mind.
In the PM task, we reported three interesting results that cor-
roborate our interpretation of the detrimental eect of short-term
videos. First, the PM cue detection accuracy decreased by almost
40% after users were interrupted by short-form videos. After the
TikTok interruption, participants tried to retrieve the intention,
but this goal might be eeting because of the attentional demands
of the interruption. Even if participants were aware of the need
to retrieve the intention, they were not able to associate the PM
cue word with the associated button to press. This interpretation
CHI ’23, April 23–28, 2023, Hamburg, Germany Francesco Chiossi, Luke Haliburton, Changkun Ou, Andreas Butz, and Albrecht Schmidt
is in line with previous research showing that dividing attention
impaired PM performance [
34
,
73
] and visual working memory per-
formance [
125
]. Moreover, as the interruption recruited attentional
resources, the interruption duration did not leave enough time to
retrieve and successfully resume the intention. Participants were
vulnerable to the interruption and not ready to resume the task, as
they were cognitively engaged with the interruption. This result
is in line with earlier research showing that participants who re-
trieve an intention sometimes forget to execute it even after shorter
interruptions [28, 70].
Second, the DDM parameter analysis allowed us to examine the
cost and interference eects of the interruption. DDM integrates
two behavioral features, which are usually in a compensatory rela-
tionship (i.e., accuracy and response speed), into psychologically
meaningful process parameters [
87
]. The rst parameter extracted
with this approach, the drift rate, mimicked the results we obtained
with behavioral accuracy post-interruption. Drift rate represents
processing eciency and therefore, it is selectively inuenced by
the task demands on participants [
91
]. In our study, short-form
videos increased the memory demand on participants, making it
harder to process each alternative, leading to reduced evidence
accumulation rates for making a decision. The TikTok interruption
absorbed resources that would otherwise be allocated to the PM
task, thereby slowing processing eciency [
104
,
105
]. This inter-
pretation is specically supported by the fact that we did not nd
any signicant dierence when analyzing the parameters from the
DDM in the LD task.
Similarly to the drift rate, the variance parameter also showed
the same results as the drift rate. Participants had increased un-
certainty about which intentions to execute, i.e., which associated
button to press. This can be seen in the light of reduced attentional
capacities. Theories of attention associate resource allocation to
preparatory responses and increased processing speed [
55
]. The
time needed for accumulating evidence and making a decision is
inuenced by the attention directed towards it. Thus, short-form
video consumption competed for such resources and resulted in
decreased evidence accumulation in the PM task [
9
]. The drift rate
and variance results align with the overlapping attentional and
memory demands that aect accuracy and RT variability via longer
RTs during PM attentional lapses [47].
Finally, we found decreased decision bound in the short-form
video compared to the control condition (Rest). This implies that
participants chose a more liberal response criterion, and therefore,
they accepted less evidence to make a response after the TikTok
interruption. As a result, participants failed to retrieve the intention
and, therefore, made a less informed decision, impairing their accu-
racy. They were not ready, given their excessive memory load, to
monitor the PM cues and make a conscious decision [
11
]. Taken to-
gether, those results point towards how PM is vulnerable to context-
switching, specically in the form of engaging short-form videos.
5.3 Consequences for Media Technology
Designers
Our results have consequences in two major areas: mitigating the
adverse eects of short-form videos on PM and intentionally ex-
ploiting the associated PM degradation.
5.3.1 Mitigate the adverse eects of short-form videos on PM. We
show that interruptions with short-form video (such as but not lim-
ited to TikTok) signicantly degrade PM post-interruption, which
means that a user is likely to have degraded performance when
they return to their primary task. Prospective memory is a crucial
aspect of daily life and is susceptible to interruption. Therefore, we
argue that a social media feed that negatively impacts cognitive
performance in the real world is generally not desired and could
be classied as a Dark Pattern. Gray et al. [38] outlined ve strate-
gies for dark patterns: nagging, obstruction, sneaking, interface
interference, and forced action. Degrading cognitive performance
in the real world does not fall into any of these categories but per-
haps deserves its own classication. Recently, multiple platforms
are introducing features inspired by the TikTok feed (e.g., Insta-
gram Reels and YouTube shorts). By increasing the use of this dark
pattern across the social media landscape, social media designers
might impact PM performance across a broader audience. Addi-
tional research is required to understand how to combat this eect,
but we can suggest some logical rst approaches. Past work in
HCI has used digital reminders [
118
] and other memory aids [
97
]
to improve PM performance. Moreover, digital reminder systems
have been shown to support rehearsal for highly-specied inten-
tions [13], such as remembering to purchase a specic item in the
supermarket. Similar reminder-based approaches could help users
remember tasks after a social media interruption in a productivity
context. Alternately, proactive reminder systems [
122
] or Digital
Self-Control Tools [
64
] can anticipate the social media feed engage-
ment and suggest positive and healthier interruptions [
95
]. This
would, rst, support the overall eciency of digital reminders for
task management and, second, overcome challenges in support of
PM as intentions are generally triggered by either specic events or
times [
50
]. However, in line with past work on mindful smartphone
use [
62
,
111
], we highlight the fundamental necessity for users to
maintain agency and be permitted to use their smartphones and
social media feeds in a way that aligns with their own needs and
values.
Finally, since we only observed PM degradation in the TikTok
condition, it follows that interspersing short-form video feeds with
other media formats may mitigate this eect. Investigating interven-
tions to prevent the disruption and generally enhance prospective
memory is important from a practical standpoint [
68
] and may
potentially improve cognitive performance and well-being for a
generation of internet users.
5.3.2 Leverage the impact of short-form videos to intentionally make
users forget information. Aside from avoiding adverse eects, tech-
nology designers could exploit PM degradation by intentionally
engaging short-form videos as interruptions in strategic situations.
Such interruptions could be employed to help a user forget informa-
tion. For example, this eect could be used in a video game where
the designer wants to challenge the player by disrupting their abil-
ity to remember a task they need to accomplish. Distraction is a
recommended game design technique to increase immersion or the
impact of surprising scenes [
90
], but it could also increase diculty.
To operationalize this, a game designer could strategically insert
short cutscenes to distract a player from a task they are supposed
Short-Form Videos Degrade Our Capacity to Retain Intentions CHI ’23, April 23–28, 2023, Hamburg, Germany
to remember to perform. This intentional distraction could increase
diculty while entertaining the user with engaging scenes.
This approach could also be employed therapeutically to help a
user forget about unhappy memories. Distraction through media,
such as video games, has been eectively used for mental health
recovery applications [
21
]. In practice, this could be combined
with an aective computing paradigm that monitors when a user is
unhappy and intervenes with short videos to disrupt their memories.
However, additional research is required to determine whether
short-form videos could be used for such applications.
Finally, designers could employ short-form videos to help users
transfer their focus to a new task. Research on task-switching shows
that the initial task interferes with the second task because users
maintain residual attention on features that are no longer rele-
vant [
89
]. This could be implemented by triggering a series of
short-form videos any time a user switches their focus to a new
task, which may reduce the residual attention on the previous task
and thereby reduce the time required to fully focus on the new task.
There is a need for additional future work to determine whether
short-form videos could be used to diminish the adverse eects of
residual attention.
5.4 Limitations & Future Work
Our results must be viewed in light of certain limitations, which
we reect on below.
We did not evaluate every social media platform, however, we
chose three of the currently most widespread platforms, which
largely vary in their feed format. Other major social media plat-
forms are either very similar to platforms we have already chosen
(e.g., Mastodon shares nearly all feed features with Twitter), or
consist of a combination of features from the platforms we selected
(e.g., Instagram and Facebook both have ‘Reels’, which are user-
generated short-format videos similar to TikTok, Facebook feeds
also contain longer videos, similar to YouTube, and both have text
and image posts of variable lengths, similar to Twitter).
Second, users did not browse their own feeds in the YouTube
condition. Rather, they freely chose a video from a pre-dened list
of options. This approach was chosen to control for video duration
and equalize the interruption length, as dierent interruption length
has shown to have dierential eects on information encoding and
retrieval [
4
,
77
,
79
], thus compromising between time demands and
content preference. Future work will address how the eect of long
format videos can impact PM with content chosen by the users and
for dierent duration.
Additionally, our situated experimental design presented con-
founding variables. We chose three popular social media feeds that
varied in terms of media content. Twitter mostly relies on text con-
tent with sporadic images, Youtube on video content of medium
duration (
11.7 minutes)
3
, while TikTok also employs video con-
tent but with a short duration (15 seconds 3 minutes). The social
media feeds also diered in the pace of the presented content, i.e.,
context switching and media format. In our study, we did not at-
tempt to quantify how each feed characteristic (number of context
3
https://www.statista.com/statistics/1026923/youtube-video-category-average-
length/
switches and media format) individually contributes to PM degra-
dation. Therefore, we did not employ a condition that combines
text content as Twitter and no context switching as YouTube, e.g., a
10-minutes reading article interruption. However, media modality
has shown to dierentially impact memory processing depending
on textual or visual stimulation [
54
], or when such content is com-
bined [
124
], interfering with automatic and control processing [
85
].
Thus, now that our research has identied that dierent social me-
dia feeds have dierent impacts on PM, this should motivate future
work to investigate this eect on a more ne-grained level. We
propose a future experiment that systematically varies the pace of
context switching as well as the type of media modality.
Finally, the lexical decision task is an established ongoing task
when investigating PM [
31
], but it may not be the most ecologically
valid choice. Therefore, future work should aim to extend our results
to real-world tasks, which might bring to light additional eects that
remained hidden in our laboratory setting. One potential direction
in such a more realistic setting is the design of reminders that
could counteract the detrimental eect of interruptions, which has
successfully been shown in other contexts [
52
]. Thus, it could be
desirable to compare the eect of encoding reminders with a no-
reminder condition and a pause condition that allows participants
ample time to encode. Some reminder cues, but not others, have
been shown to improve PM performance in PM paradigms [42].
6 CONCLUSION
In this paper, we investigated the impact of social media on prospec-
tive memory. We conducted a between-subjects study with 60 par-
ticipants comparing three social media feeds (TikTok,Twitter, and
YouTube) and a Rest condition as a control. Our results show that
short video streams such as TikTok have a signicant detrimental
impact on prospective memory performance. Specically, users
showed a worsened speed/accuracy trade-os as compared to all
other experimental conditions. The other platforms do not signif-
icantly aect performance. Interestingly, social media addiction,
absent-minded phone, and perceived engagement did not have
any relationship or inuenced accuracy across conditions. This
allowed us to disentangle the eect of individual dierences from
the detrimental one of short-form videos on PM. We contribute
an empirical understanding of the impact of dierent social media
feeds on PM and discuss consequences for technology designers to
create engaging experiences without negatively impacting users.
7 OPEN SCIENCE
We encourage readers to reproduce and extend our results and anal-
ysis methods. Therefore our experimental setup, collected datasets,
and analysis scripts are available on Github 4.
ACKNOWLEDGMENTS
We would like to thank Martina Gluderer for her support during
data collection and the submission of this work. Francesco Chiossi
was supported by the Deutsche Forschungsgemeinschaft (DFG, Ger-
man Research Foundation), Project-ID 251654672-TRR 161. Luke
Haliburton was supported by the Bavarian Research Alliance asso-
ciation ForDigitHealth.
4https://github.com/mimuc/media-prospective- memory
CHI ’23, April 23–28, 2023, Hamburg, Germany Francesco Chiossi, Luke Haliburton, Changkun Ou, Andreas Butz, and Albrecht Schmidt
REFERENCES
[1]
Piotr D Adamczyk and Brian P Bailey. 2004. If not now, when? The eects
of interruption at dierent moments within task execution. In Proceedings of
the SIGCHI conference on Human factors in computing systems. 271–278. https:
//doi.org/10.1145/985692.985727
[2]
Shamel Addas and Alain Pinsonneault. 2018. Theorizing the Multilevel Eects
of Interruptions and the Role of Communication Technology. Journal of the
Association for Information Systems 19, 11 (Nov. 2018). https://aisel.aisnet.org/
jais/vol19/iss11/2
[3]
Philip A Allen, Mei-Ching Lien, Eric Ruthru, and Andreas Voss. 2014. Multi-
tasking and aging: do older adults benet from performing a highly practiced
task? Experimental aging research 40, 3 (2014), 280–307. https://doi.org/10.1080/
0361073X.2014.896663
[4]
Erik M Altmann and Christian D Schunn. 2019. Integrating decay and in-
terference: A new look at an old interaction. In Proceedings of the Twenty-
Fourth Annual Conference of the Cognitive Science Society. Routledge, 65–70.
https://doi.org/10.4324/9781315782379-49
[5]
Erik M Altmann and J Gregory Trafton. 2002. Memory for goals: An activation-
based model. Cognitive science 26, 1 (2002), 39–83. https://doi.org/10.1207/
s15516709cog2601_2
[6]
Erik M Altmann, J Gregory Trafton, and David Z Hambrick. 2014. Momentary
interruptions can derail the train of thought. Journal of Experimental Psychology:
General 143, 1 (2014), 215. https://doi.org/10.1037/a0030986
[7]
B Hunter Ball and Andrew J Aschenbrenner.2018. The importance of age-related
dierences in prospective memory: Evidence from diusion model analyses.
Psychonomic Bulletin & Review 25, 3 (2018), 1114–1122. https://doi.org/10.3758/
s13423-017- 1318-4
[8]
Dale J. Barr, Roger Levy, Christoph Scheepers, and Harry J. Tily. 2013. Random
eects structure for conrmatory hypothesis testing: Keep it maximal. Journal
of Memory and Language 68, 3 (2013), 255–278. https://doi.org/10.1016/j.jml.
2012.11.001
[9]
Russell J Boag, Luke Strickland, Shayne Loft, and Andrew Heathcote. 2019.
Strategic attention and decision control support prospective memory in a com-
plex dual-task environment. Cognition 191 (2019), 103974. https://doi.org/10.
1016/j.cognition.2019.05.011
[10]
Jesús Bobadilla, Fernando Ortega, Antonio Hernando, and Abraham Gutiérrez.
2013. Recommender systems survey. Knowledge-based systems 46 (2013), 109–
132. https://doi.org/ttps://doi.org/10.1016/j.knosys.2013.03.012
[11]
C Dennis Boywitt and Jan Rummel. 2012. A diusion model analysis of task
interference eects in prospective memory. Memory & Cognition 40, 1 (2012),
70–82. https://doi.org/10.3758/s13421-011- 0128-6
[12]
Gene A Brewer. 2011. Analyzing response time distributions: Methodological
and theoretical suggestions for prospective memory researchers. Zeitschrift für
Psychologie/Journal of Psychology 219, 2 (2011), 117. https://doi.org/10.1027/
2151-2604/a000056
[13]
R. N. Brewer, M. R. Morris, and S. E. Lindley. 2017. How to Remember What to
Remember: Exploring Possibilities for Digital Reminder Systems. 1, 3, Article
38 (sep 2017), 20 pages. https://doi.org/10.1145/3130903
[14]
Duncan P Brumby, Christian P Janssen, and Gloria Mark. 2019. How do inter-
ruptions aect productivity? In Rethinking productivity in software engineering.
Springer, 85–107. https://doi.org/10.1016/j.ijhcs.2014.11.003
[15]
Marc Brysbaert, Matthias Buchmeier, Markus Conrad, Arthur M Jacobs, Jens
Bölte, and Andrea Böhl. 2011. The word frequency eect: a review of recent
developments and implications for the choice of frequency estimates in German.
Experimental psychology 58, 5 (2011), 412. https://doi.org/10.1027/1618-3169/
a000123
[16]
Johannes L Busch, Femke S Haeussler, Frank Domahs, Lars Timmermann, Immo
Weber, and Carina R Oehrn. 2022. German normative data with naming latencies
for 283 action pictures and 600 action verbs. Behavior research methods 54, 2
(2022), 649–662. https://doi.org/10.3758/s13428-021- 01647-w
[17]
Zhuang Chen, Qian He, Zhifei Mao, Hwei-Ming Chung, and Sabita Maharjan.
2019. A study on the characteristics of douyin short videos and implications for
edge caching. In Proceedings of the ACM Turing Celebration Conference-China.
1–6. https://doi.org/10.1145/3321408.3323082
[18]
Jan Chong and Rosanne Siino. 2006. Interruptions on software teams: a com-
parison of paired and solo programmers. In Proceedings of the 2006 20th an-
niversary conference on Computer supported cooperative work. 29–38. https:
//doi.org/10.1145/1180875.1180882
[19]
Wesley C Clapp and Adam Gazzaley. 2012. Distinct mechanisms for the impact
of distraction and interruption on working memory in aging. Neurobiology of
aging 33, 1 (2012), 134–148. https://doi.org/10.1016/j.neurobiolaging.2010.01.012
[20]
Anna-Lisa Cohen and Jason L. Hicks. 2017. Characterization of Prospective
Memory and Associated Processes. Springer International Publishing, Cham,
41–60. https://doi.org/10.1007/978-3- 319-68990- 6_3
[21]
Michelle Colder Carras, Anna Kalbarczyk, Kurrie Wells, Jaime Banks, Rachel
Kowert, Colleen Gillespie, and Carl Latkin. 2018. Connection, meaning, and
distraction: A qualitative study of video game play and mental health recovery
in veterans treated for mental and/or behavioral health problems. Social Science
& Medicine 216 (Nov. 2018), 124–132. https://doi.org/10.1016/j.socscimed.2018.
08.044
[22]
Giorgia Cona, Francesco Chiossi, Silvia Di Tomasso, Giovanni Pellegrino,
Francesco Piccione, Patrizia Bisiacchi, and Giorgio Arcara. 2020. Theta and
alpha oscillations as signatures of internal and external attention to delayed
intentions: A magnetoencephalography (MEG) study. NeuroImage 205 (Jan.
2020), 116295. https://doi.org/10.1016/j.neuroimage.2019.116295
[23]
Giorgia Cona, Cristina Scarpazza, Giuseppe Sartori, Morris Moscovitch, and
Patrizia Silvia Bisiacchi. 2015. Neural bases of prospective memory: a meta-
analysis and the “Attention to Delayed Intention”(AtoDI) model. Neuroscience
& Biobehavioral Reviews 52 (2015), 21–37. https://doi.org/10.1016/j.neubiorev.
2015.02.007
[24]
Gabriel I Cook, J Thadeus Meeks, Arlo Clark-Foos, Paul S Merritt, and Richard L
Marsh. 2014. The role of interruptions and contextual associations in delayed-
execute prospective memory. Applied Cognitive Psychology 28, 1 (2014), 91–103.
https://doi.org/10.1002/acp.2960
[25]
Fergus I. M. Craik, Richard Govoni, Moshe Naveh-Benjamin, and Nicole D.
Anderson. 1996. The eects of divided attention on encoding and retrieval
processes in human memory. Journal of Experimental Psychology: General 125,
2 (1996), 159–180. https://doi.org/10.1037/0096-3445.125.2.159
[26]
James Davidson, Benjamin Liebald, Junning Liu, Palash Nandy, Taylor Van Vleet,
Ullas Gargi, Sujoy Gupta, Yu He, Mike Lambert, Blake Livingston, et al
.
2010.
The YouTube video recommendation system. In Proceedings of the fourth ACM
conference on Recommender systems. 293–296. https://doi.org/10.1145/1864708.
1864770
[27]
R Key Dismukes. 2012. Prospective memory in workplace and everyday
situations. Current Directions in Psychological Science 21, 4 (2012), 215–220.
https://doi.org/10.1177/0963721412447621
[28]
Rahul M Dodhia and Robert K Dismukes. 2009. Interruptions create prospective
memory tasks. Applied Cognitive Psychology: The Ocial Journal of the Society
for Applied Research in Memory and Cognition 23, 1 (2009), 73–89. https://doi.
org/10.1002/acp.1441
[29]
Elmira Zahmat Doost and Wei Zhang. 2022. Mental workload variations during
dierent cognitive oce tasks with social media interruptions. Ergonomics 0, 0
(2022), 1–17. https://doi.org/10.1080/00140139.2022.2104381
[30]
Budmonde Duinkharjav, Praneeth Chakravarthula, Rachel Brown, Anjul Patney,
and Qi Sun. 2022. Image Features Inuence Reaction Time: A Learned Proba-
bilistic Perceptual Model for Saccade Latency. ACM Trans. Graph. 41, 4, Article
144 (jul 2022), 15 pages. https://doi.org/10.1145/3528223.3530055
[31]
Gilles O Einstein and Mark A McDaniel. 2005. Prospective memory: Multiple
retrieval processes. Current Directions in Psychological Science 14, 6 (2005),
286–290. https://doi.org/10.1111/j.0963-7214.2005.00382.x
[32]
Gilles O Einstein, Mark A McDaniel, Ruthann Thomas, Sara Mayeld, Hilary
Shank, Nova Morrisette, and Jennifer Breneiser. 2005. Multiple processes in
prospective memory retrieval: factors determining monitoring versus sponta-
neous retrieval. Journal of Experimental Psychology: General 134, 3 (2005), 327.
https://doi.org/10.1037/0096-3445.134.3.327
[33]
Gilles O. Einstein, Mark A. McDaniel, Ruthann Thomas, Sara Mayeld, Hilary
Shank, Nova Morrisette, and Jennifer Breneiser. 2005. Multiple Processes in
Prospective Memory Retrieval: Factors Determining Monitoring Versus Spon-
taneous Retrieval. Journal of Experimental Psychology: General 134, 3 (2005),
327–342. https://doi.org/10.1037/0096-3445.134.3.327
[34]
Gilles O Einstein, Rebekah E Smith, Mark A McDaniel, and Pat Shaw. 1997.
Aging and prospective memory: the inuence of increased task demands at
encoding and retrieval. Psychology and aging 12, 3 (1997), 479. https://doi.org/
10.1037/0882-7974.12.3.479
[35]
Annie Beth Fox, Jonathan Rosen, and Mary Crawford. 2009. Distractions,
distractions: does instant messaging aect college students’ performance on
a concurrent reading comprehension task? CyberPsychology & Behavior 12, 1
(2009), 51–53. https://doi.org/10.1089/cpb.2008.0107
[36]
Scott T Frein, Samantha L Jones, and Jennifer E Gerow. 2013. When it comes to
Facebook there may be more to bad memory than just multitasking. Computers
in Human Behavior 29, 6 (2013), 2179–2182. https://doi.org/10.1016/j.chb.2013.
04.031
[37]
Brian A Gordon, Jill T Shelton, Julie M Bugg, Mark A McDaniel, and Denise
Head. 2011. Structural correlates of prospective memory. Neuropsychologia 49,
14 (2011), 3795–3800. https://doi.org/10.1016/j.neuropsychologia.2011.09.035
[38]
Colin M. Gray, Yubo Kou, Bryan Battles, Joseph Hoggatt, and Austin L. Toombs.
2018. The Dark (Patterns) Side of UX Design. In Proceedings of the 2018 CHI
Conference on Human Factors in Computing Systems (CHI ’18). Association for
Computing Machinery, New York, NY, USA, 1–14. https://doi.org/10.1145/
3173574.3174108
[39]
Yvonne C. T. Groot, Barbara A. Wilson, Jonathan Evans, and Peter Watson.
2002. Prospective memory functioning in people with and without brain injury.
Journal of the International Neuropsychological Society 8, 5 (July 2002), 645–654.
https://doi.org/10.1017/S1355617702801321
[40]
Ashish Gupta, Han Li, and Ramesh Sharda. 2013. Should I send this message?
Understanding the impact of interruptions, social hierarchy and perceived task
Short-Form Videos Degrade Our Capacity to Retain Intentions CHI ’23, April 23–28, 2023, Hamburg, Germany
complexity on user performance and perceived workload. Decision Support
Systems 55, 1 (2013), 135–145. https://doi.org/10.1016/j.dss.2012.12.035
[41]
Melissa J Guynn. 2003. A two-process model of strategic monitoring in event-
based prospective memory: Activation/retrieval mode and checking. Inter-
national Journal of Psychology 38, 4 (2003), 245–256. https://doi.org/10.1080/
00207590344000178
[42]
Melissa J. Guynn, Mark A. Mcdaniel, and Gilles O. Einstein. 1998. Prospective
memory: When reminders fail. Memory & Cognition 26, 2 (March 1998), 287–298.
https://doi.org/10.3758/BF03201140
[43]
Dorothee Hefner and Peter Vorderer. 2016. Digital Stress: Permanent Con-
nectedness and Multitasking. In The Routledge Handbook of Media Use and
Well-Being: International Perspectives on Theory and Research on Positive Media
Eects. Taylor & Francis, 484. https://doi.org/10.4324/9781315714752
[44]
Alexis Hiniker, Sungsoo (Ray) Hong, Tadayoshi Kohno, and Julie A. Kientz.
2016. MyTime: Designing and Evaluating an Intervention for Smartphone Non-
Use. In Proceedings of the 2016 CHI Conference on Human Factors in Computing
Systems (CHI ’16). Association for Computing Machinery, New York, NY, USA,
4746–4757. https://doi.org/10.1145/2858036.2858403
[45] Yoolim Hong and Andrew Leber. 2016. Attentional disengagement suppresses
visual long-term memory. Journal of Vision 16, 12 (2016), 1017–1017. https:
//doi.org/10.1167/16.12.1017
[46]
Sebastian S Horn, Ute J Bayen, and Rebekah E Smith. 2011. What can the
diusion model tell us about prospective memory? Canadian Journal of Experi-
mental Psychology/Revue canadienne de psychologie expérimentale 65, 1 (2011),
69. https://doi.org/10.1037/a0022808
[47]
Andreas Ihle, Paolo Ghisletta, and Matthias Kliegel. 2017. Prospective memory
and intraindividual variability in ongoing task response times in an adult lifespan
sample: The role of cue focality. Memory 25, 3 (2017), 370–376. https://doi.org/
10.1080/09658211.2016.1173705
[48]
Instagram. 2020. Introducing Instagram Reels. https://about.instagram.com/
blog/announcements/introducing-instagram- reels-announcement
[49]
Chris Jae. 2020. Building YouTube Shorts, a new way to watch & create on
YouTube. https://blog.youtube/news-and- events/building-youtube-shorts/
[50]
Theodor Jager and Matthias Kliegel. 2008. Time-based and event-based prospec-
tive memory across adulthood: Underlying mechanisms and dierential costs
on the ongoing task. The Journal of general psychology 135, 1 (2008), 4–22.
https://doi.org/10.3200/GENP.135.1.4-22
[51]
Simon Kemp. 2022. Digital 2022: Global Overview Report. https://datareportal.
com/reports/digital-2022- global-overview- report
[52]
Dagmar Kern, Paul Marshall, and Albrecht Schmidt. 2010. Gazemarks: gaze-
based visual placeholders to ease attention switching. Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems (2010). https://doi.org/10.
1145/1753326.1753646
[53]
Lia Kvavilashvili and Laura Fisher. 2007. Is time-based prospective remembering
mediated by self-initiated rehearsals? Role of incidental cues, ongoing activity,
age, and motivation. Journal of Experimental Psychology: General 136, 1 (2007),
112. https://doi.org/10.1037/0096-3445.136.1.112
[54]
Annie Lang, Robert F Potter, and Paul D Bolls. 1999. Something for nothing:
Is visual encoding automatic? Media Psychology 1, 2 (1999), 145–163. https:
//doi.org/10.1207/s1532785xmep0102_4
[55]
Robert Langner, Michael B Steinborn, Anjan Chatterjee, WalterSturm, and Klaus
Willmes. 2010. Mental fatigue and temporal preparation in simple reaction-time
performance. Acta psychologica 133, 1 (2010), 64–72. https://doi.org/10.1016/j.
actpsy.2009.10.001
[56]
Sophie Leroy. 2009. Why is it so hard to do my work? The challenge of attention
residue when switching between work tasks. Organizational Behavior and
Human Decision Processes 109, 2 (2009), 168–181. https://doi.org/10.1016/j.
obhdp.2009.04.002
[57]
Sophie Leroy and Theresa M Glomb. 2018. Tasks interrupted: How anticipating
time pressure on resumption of an interrupted task causes attention residue
and low performance on interrupting tasks and how a “ready-to-resume” plan
mitigates the eects. Organization Science 29, 3 (2018), 380–397. https://doi.
org/10.1287/orsc.2017.1184
[58]
Sophie Leroy and Aaron M Schmidt. 2016. The eect of regulatory focus
on attention residue and performance during interruptions. Organizational
Behavior and Human Decision Processes 137 (2016), 218–235. https://doi.org/10.
1016/j.obhdp.2016.07.006
[59]
Steson Lo and Sally Andrews. 2015. To transform or not to transform: Using
generalized linear mixed models to analyse reaction time data. Frontiers in
psychology 6 (2015), 1171. https://doi.org/10.3389/fpsyg.2015.01171
[60]
Sara A Lu, Christopher D Wickens, Julie C Prinet, Shaun D Hutchins, Nadine
Sarter, and Angelia Sebok. 2013. Supporting interruption management and
multimodal interface design: Three meta-analyses of task performance as a
function of interrupting task modality. Human factors 55, 4 (2013), 697–724.
https://doi.org/10.1177/001872081347629
[61]
Kai Luko, Ulrik Lyngs, Himanshu Zade, J Vera Liao, James Choi, Kaiyue Fan,
Sean A Munson, and Alexis Hiniker. 2021. How the design of youtube inuences
user sense of agency. In Proceedings of the 2021 CHI Conference on Human Factors
in Computing Systems. 1–17. https://doi.org/10.1145/3411764.3445467
[62]
Kai Luko, Cissy Yu, Julie Kientz, and Alexis Hiniker. 2018. What Makes
Smartphone Use Meaningful or Meaningless? 2, 1 (2018), 26. https://doi.org/10.
1145/3191754
[63]
Ludmila Lupinacci. 2020. ‘Absentmindedly scrolling through nothing’: liveness
and compulsory continuous connectedness in social media. Media, Culture & So-
ciety (July 2020), 0163443720939454. https://doi.org/10.1177/0163443720939454
[64]
Ulrik Lyngs, Kai Luko, Petr Slovak, Reuben Binns, Adam Slack, Michael Inzlicht,
Max Van Kleek, and Nigel Shadbolt. 2019. Self-control in cyberspace: Applying
dual systems theory to a review of digital self-control tools. In proceedings of
the 2019 CHI conference on human factors in computing systems. 1–18. https:
//doi.org/10.1145/3290605.3300361
[65]
Gloria Mark, Daniela Gudith, and Ulrich Klocke. 2008. The cost of interrupted
work: More speed and stress. 107–110. https://doi.org/10.1145/1357054.1357072
[66]
Jeremy Marty-Dugas, Brandon Ralph, Jonathan Oakman, and Daniel Smilek.
2018. The Relation Between Smartphone Use and Everyday Inattention.
Psychology of Consciousness: Theory, Research, and Practice 5 (March 2018).
https://doi.org/10.1037/cns0000131
[67]
Bernard R McCoy. 2016. Digital distractions in the classroom phase II: Student
classroom use of digital devices for non-class related purposes. (2016).
[68]
Mark A. McDaniel and Gilles O. Einstein. 2007. Prospective Memory: An Overview
and Synthesis of an Emerging Field. SAGE Publications. https://doi.org/10.4135/
9781452225913
[69]
Mark A McDaniel, Gilles O Einstein, Thomas Graham, and Erica Rall. 2004.
Delaying execution of intentions: Overcoming the costs of interruptions. Applied
Cognitive Psychology: The Ocial Journal of the Society for Applied Research in
Memory and Cognition 18, 5 (2004), 533–547. https://doi.org/10.1002/acp.1002
[70]
Mark A McDaniel, Gilles O Einstein, Amy C Stout, and Zack Morgan. 2003.
Aging and maintaining intentions over delays: do it or lose it. Psychology and
Aging 18, 4 (2003), 823. https://doi.org/10.1037/0882- 7974.18.4.823
[71]
Mark A. McDaniel, Melissa J. Guynn, Gilles O. Einstein, and Jennifer Breneiser.
2004. Cue-Focused and Reexive-Associative Processes in Prospective Memory
Retrieval. Journal of Experimental Psychology: Learning, Memory, and Cognition
30, 3 (2004), 605–614. https://doi.org/10.1037/0278-7393.30.3.605
[72]
Daniel C McFarlane and Kara A Latorella. 2002. The scope and importance of
human interruption in human-computer interaction design. Human-Computer
Interaction 17, 1 (2002), 1–61. https://doi.org/10.1207/S15327051HCI1701_1
[73]
Deborah Mcgann, Judi A Ellis, and Alan Milne. 2002. Conceptual and perceptual
processes in prospective remembering: Dierential inuence of attentional
resources. Memory & cognition 30, 7 (2002), 1021–1032. https://doi.org/10.3758/
BF03194320
[74]
Xiongkuo Min, Guangtao Zhai, Jiantao Zhou, Xiao-Ping Zhang, Xiaokang Yang,
and Xinping Guan. 2020. A multimodal saliency model for videos with high
audio-visual correspondence. IEEE Transactions on Image Processing 29 (2020),
3805–3819. https://doi.org/10.1109/TIP.2020.2966082
[75]
Lucia Monacis, Valeria de Palo, Mark D.Griths, and Maria Sinatra. 2017. Social
networking addiction, attachment style, and validation of the Italian version of
the Bergen Social Media Addiction Scale. Journal of Behavioral Addictions 6, 2
(June 2017), 178–186. https://doi.org/10.1556/2006.6.2017.023
[76]
Christopher A Monk and David G Kidd. 2008. The eects of brief interruptions
on task resumption. In Proceedings of the Human Factors and Ergonomics Society
Annual Meeting, Vol. 52. SAGE Publications Sage CA: Los Angeles, CA, 403–407.
https://doi.org/10.1177/154193120805200443
[77]
Christopher A Monk, J Gregory Trafton, and Deborah A Boehm-Davis. 2008.
The eect of interruption duration and demand on resuming suspended goals.
Journal of experimental psychology: Applied 14, 4 (2008), 299. https://doi.org/10.
1037/a0014402
[78]
Christian Montag, Haibo Yang, and Jon D Elhai. 2021. On the psychology of
TikTok use: A rst glimpse from empirical ndings. Frontiers in public health 9
(2021), 641673. https://doi.org/10.3389/fpubh.2021.641673
[79]
Phillip L Morgan, John Patrick, Samuel M Waldron, Sophia L King, and Tanya
Patrick. 2009. Improving memory after interruption: Exploiting soft constraints
and manipulating information access cost. Journal of experimental psychology:
Applied 15, 4 (2009), 291. https://doi.org/10.1037/a0018008
[80]
Michael A Nees and Anjali Fortna. 2015. A comparison of human versus
virtual interruptions. Ergonomics 58, 5 (2015), 852–856. https://doi.org/10.1080/
00140139.2014.990934
[81]
Brid O’Conaill and David Frohlich. 1995. Timespace in the workplace: Dealing
with interruptions. In Conference companion on Human factors in computing
systems. 262–263. https://doi.org/10.1145/223355.223665
[82]
Greg R Oldham, Carol T Kulik, and Lee P Stepina. 1991. Physical environments
and employee reactions: eects of stimulus-screening skills and job complexity.
Academy of Management Journal 34, 4 (1991), 929–938. https://doi.org/10.5465/
256397
[83]
Jonathan W Peirce. 2007. PsychoPy—psychophysics software in Python. Journal
of neuroscience methods 162, 1-2 (2007), 8–13. https://doi.org/10.1016/j.jneumeth.
2006.11.017
CHI ’23, April 23–28, 2023, Hamburg, Germany Francesco Chiossi, Luke Haliburton, Changkun Ou, Andreas Butz, and Albrecht Schmidt
[84]
Heng Peng and Ying Lu. 2012. Model selection in linear mixed eect models.
Journal of Multivariate Analysis 109 (2012), 109–129. https://doi.org/10.1016/j.
jmva.2012.02.005
[85]
Thomas E Powell, Hajo G Boomgaarden, Knut De Swert, and Claes H de Vreese.
2019. Framing fast and slow: A dual processing account of multimodal framing
eects. Media Psychology 22, 4 (2019), 572–600. https://doi.org/10.1080/15213269.
2018.1476891
[86]
Roger Ratcli. 1978. A theory of memory retrieval. Psychological Review 85
(1978), 59–108. https://doi.org/10.1037/0033-295X.85.2.59
[87]
Roger Ratcli. 2013. Parameter variability and distributional assumptions in
the diusion model. Psychological review 120, 1 (2013), 281. https://doi.org/10.
1037/a0030775
[88]
Nornadiah Mohd Razali, Yap Bee Wah, et al
.
2011. Power comparisons of shapiro-
wilk, kolmogorov-smirnov, lilliefors and anderson-darling tests. Journal of
statistical modeling and analytics 2, 1 (2011), 21–33.
[89]
Franziska R. Richter and Nick Yeung. 2012. Memory and Cognitive Control
in Task Switching. Psychological Science 23, 10 (Oct. 2012), 1256–1263. https:
//doi.org/10.1177/0956797612444613 Publisher: SAGE Publications Inc.
[90]
Scott Rogers. 2014. Level Up! The Guide to Great Video Game Design. John Wiley
& Sons. Google-Books-ID: UT5jAwAAQBAJ.
[91]
Jan Rummel, Beatrice G Kuhlmann, and Dayna R Touron. 2013. Performance
predictions aect attentional processes of event-based prospective memory.
Consciousness and cognition 22, 3 (2013), 729–741. https://doi.org/10.1016/j.
concog.2013.04.012
[92]
Jan Rummel, Ann-Katrin Wesslein, and Thorsten Meiser. 2017. The role of
action coordination for prospective memory: Task-interruption demands aect
intention realization. Journal of Experimental Psychology: Learning, Memory,
and Cognition 43, 5 (2017), 717. https://doi.org/10.1037/xlm0000334
[93]
Penelope M Sanderson and Tobias Grundgeiger. 2015. How do interruptions
aect clinician performance in healthcare? Negotiating delity, control, and
potential generalizability in the search for answers. International Journal of
Human-Computer Studies 79 (2015), 85–96. https://doi.org/10.1016/j.ijhcs.2014.
11.003
[94]
Premjit K Sanjram and Khan Azizuddin. 2010. Attention and programmer char-
acteristics in prospective memory: an investigation of habit intrusion error in
programmer multitasking. In Proceedings of the 28th Annual European Conference
on Cognitive Ergonomics. 307–310. https://doi.org/10.1145/1962300.1962365
[95]
Farzan Sasangohar, Birsen Donmez, Patricia Trbovich, and Anthony C Easty.
2012. Not all interruptions are created equal: positive interruptions in healthcare.
In Proceedings of the Human Factors and Ergonomics Society Annual Meeting,
Vol. 56. SAGE Publications Sage CA: Los Angeles, CA, 824–828. https://doi.org/
10.1177/10711813125611
[96]
Daniel L. Schacter. 2012. Constructive memory: past and future. Dialogues in
Clinical Neuroscience 14, 1 (March 2012), 7–18. https://doi.org/10.31887/DCNS.
2012.14.1/dschacter
[97]
Hendrik Schulze. 2003. MEMOS: an interactive assistive system for prospective
memory decit compensation-architecture and functionality. In Proceedings of
the 6th international ACM SIGACCESS conference on Computers and accessibility
(Assets ’04). Association for Computing Machinery, New York, NY, USA, 79–85.
https://doi.org/10.1145/1028630.1028645
[98]
Michael K. Scullin, Mark A. McDaniel, Jill T. Shelton, and Ji Hae Lee. 2010.
Focal/nonfocal cue eects in prospective memory: Monitoring diculty or
dierent retrieval processes? Journal of Experimental Psychology: Learning,
Memory, and Cognition 36, 3 (2010), 736–749. https://doi.org/10.1037/a0018971
[99]
Neika Sharian and Laura B Zahodne. 2021. Daily associations between social
media use and memory failures: The mediating role of negative aect. The
Journal of general psychology 148, 1 (2021), 67–83. https://doi.org/10.1080/
00221309.2020.1743228
[100]
Jill Talley Shelton, Michael K Scullin, and Jessica Y Hacker. 2019. The multipro-
cess framework: Historical context and the “dynamic” extension. In Prospective
memory. Routledge, 10–26. https://doi.org/10.4324/9781351000154
[101]
Maxwell Shinn, Norman H Lam, and John D Murray. 2020. A exible framework
for simulating and tting generalized drift-diusion models. ELife 9 (2020),
e56938. https://doi.org/10.7554/eLife.56938
[102]
Cornelia Sindermann, Jon D Elhai, and Christian Montag. 2020. Predicting
tendencies towards the disordered use of Facebook’s social media platforms:
On the role of personality, impulsivity, and social anxiety. Psychiatry Research
285 (2020), 112793. https://doi.org/10.1016/j.psychres.2020.112793
[103] Rebekah E Smith. 2003. The cost of remembering to remember in event-based
prospective memory: investigating the capacity demands of delayed intention
performance. Journal of Experimental Psychology: Learning, Memory, and Cogni-
tion 29, 3 (2003), 347. https://doi.org/10.1037/0278-7393.29.3.347
[104]
Rebekah E Smith. 2008. Connecting the past and the future: Attention, memory,
and delayed intentions. (2008).
[105]
Rebekah E Smith. 2010. What costs do reveal and moving beyond the cost debate:
Reply to Einstein and McDaniel (2010). (2010). https://doi.org/10.1037/a0019183
[106]
Rebekah E Smith, Sebastian S Horn, and Ute J Bayen. 2012. Prospective memory
in young and older adults: The eects of ongoing-task load. Aging, Neuropsy-
chology, and Cognition 19, 4 (2012), 495–514. https://doi.org/10.1080/13825585.
2013.827150
[107]
Rebekah E Smith, R Reed Hunt, Jennifer C McVay, and Melissa D McConnell.
2007. The cost of event-based prospective memory: salient target events. Journal
of Experimental Psychology: Learning, Memory, and Cognition 33, 4 (2007), 734.
https://doi.org/10.1037/0278-7393.33.4.734
[108]
Rebekah E Smith and Shayne Loft. 2014. Investigating the cost to ongoing tasks
not associated with prospective memory task requirements. Consciousness and
cognition 27 (2014), 1–13.
[109]
J. B. Spira and J. B. Feintuch. 2005. The cost of not paying attention: How in-
terruptions impact knowledge worker productivity. New York, NY: Basex. Tech-
nical Report. Basex Inc, New York, NY. 21 pages. http://iorgforum.org/wp-
content/uploads/2011/06/CostOfNotPayingAttention.BasexReport1.pdf
[110]
Luke Strickland, Shayne Loft, and Andrew Heathcote. 2019. Evidence accumu-
lation modeling of event-based prospective memory. In Prospective Memory.
Routledge, 78–94.
[111]
Nađa Terzimehić, Luke Haliburton, Philipp Greiner, Albrecht Schmidt, Heinrich
Hussmann, and Ville Mäkelä. 2022. MindPhone: Mindful Reection at Unlock
Can Reduce Absentminded Smartphone Use. In Designing Interactive Systems
Conference (DIS ’22). Association for Computing Machinery, New York, NY, USA,
1818–1830. https://doi.org/10.1145/3532106.3533575
[112]
Esther Thorson and Annie Lang. 1988. The eects of videographic complexity
on memory for televised information. International Communication Association,
New Orleans, LA (1988).
[113]
Melina R Uncapher, Lin Lin, Larry D Rosen, Heather L Kirkorian, Naomi S
Baron, Kira Bailey, Joanne Cantor, David L Strayer, Thomas D Parsons, and
Anthony D Wagner. 2017. Media multitasking and cognitive, psychological,
neural, and learning dierences. Pediatrics 140, Supplement_2 (2017), S62–S66.
https://doi.org/10.1542/peds.2016-1758D
[114]
Don van Ravenzwaaij, Rei Monden, Jorge N Tendeiro, and John Ioannidis. 2019.
Bayes factors for superiority, non-inferiority, and equivalence designs. BMC
medical research methodology 19, 1 (2019), 1–12. https://doi.org/10.1186/s12874-
019-0699- 7
[115]
Mikkel C. Vinding, Jonas Kristoer Lindeløv, Yahui Xiao, Raymond C. K. Chan,
and Thomas Alrik Sørensen. 2021. Volition in prospective Memory: Evidence
against dierences between free and xed target events. Consciousness and
Cognition 94 (Sep 2021), 103175. https://doi.org/10.1016/j.concog.2021.103175
[116]
Andreas Voss and Jochen Voss. 2008. A fast numerical algorithm for the estima-
tion of diusion model parameters. Journal of Mathematical Psychology 52, 1
(2008), 1–9. https://doi.org/10.1016/j.jmp.2007.09.005
[117]
Eric-Jan Wagenmakers, Roger Ratcli, Pablo Gomez, and Gail McKoon. 2008. A
diusion model account of criterion shifts in the lexical decision task. Journal
of memory and language 58, 1 (2008), 140–159. https://doi.org/10.1016/j.jml.
2007.04.006
[118]
Yao Wang and Manuel A. Pérez-Quiñones. 2014. Exploring the role of prospec-
tive memory in location-based reminders. In Proceedings of the 2014 ACM In-
ternational Joint Conference on Pervasive and Ubiquitous Computing: Adjunct
Publication (UbiComp ’14 Adjunct). Association for Computing Machinery, New
York, NY, USA, 1373–1380. https://doi.org/10.1145/2638728.2641718
[119]
Henry H Wilmer, Lauren E Sherman, and Jason M Chein. 2017. Smartphones and
cognition: A review of research exploring the links between mobile technology
habits and cognitive functioning. Frontiers in psychology 8 (2017), 605. https:
//doi.org/10.3389/fpsyg.2017.00605
[120]
Lisa Wirz and Lars Schwabe. 2020. Prioritized attentional processing: Acute
stress, memory and stimulus emotionality facilitate attentional disengagement.
Neuropsychologia 138 (2020), 107334. https://doi.org/10.1016/j.neuropsychologia.
2020.107334
[121]
Jacob O. Wobbrock, Leah Findlater, Darren Gergle, and James J. Higgins.
2011. The Aligned Rank Transform for Nonparametric Factorial Analyses
Using Only Anova Procedures. In Proceedings of the SIGCHI Conference on
Human Factors in Computing Systems (Vancouver, BC, Canada) (CHI ’11). As-
sociation for Computing Machinery, New York, NY, USA, 143–146. https:
//doi.org/10.1145/1978942.1978963
[122]
Neil Yorke-Smith, Shahin Saadati, Karen L Myers, and David N Morley. 2009.
Like an intuitive and courteous butler: A proactive personal agent for task
management. In Proceedings of The 8th International Conference on Autonomous
Agents and Multiagent Systems-Volume 1. 337–344. https://doi.org/10.5555/
1558013.1558059
[123]
Jing Zeng, Chrystal Abidin, and Mike S Schäfer. 2021. Research perspectives on
TikTok and its legacy apps: introduction. International Journal of Communication
15 (2021), 3161–3172. https://doi.org/10.5167/uzh-205427
[124]
Jie Zhang, Tongtong Xue, Shaobo Liu, and Zhijie Zhang. 2022. Heavy and light
media multitaskers employ dierent neurocognitive strategies in a prospective
memory task: An ERP study. Computers in Human Behavior 135 (2022), 107379.
https://doi.org/10.1016/j.chb.2022.107379
[125]
Mengyu Zheng. 2021. Inuence of Short Video Watching Behaviors on Visual
Short-Term Memory. Atlantis Press, 1855–1859. https://doi.org/10.2991/assehr.
k.211220.314
... We approach these questions using short-form videos because those are currently the most engaging type of shared social media content 11 . To be more precise, we consider three TikTok videos because the platform has over a billion users and uses short-form videos (about 3-60 seconds) as a communication medium [33] and has been used in previous HCI (Human-Computer Interaction) studies [15]. 11 https://www.forbes.com/advisor/business/social-media-statistics/#source, ...
... Figure 2 shows the realization of the space and dimensionality conditions. For increasing validity, we tested the conditions in the three most liked TikTok videos according to Wikipedia 13 , showing a dancing man 14 , lip-syncing 15 , and a drawing video 16 , which resulted in a total of 12 × 3 = 36 trials per participant. In this work, we define a private space as a space accessible only for and managed by certain people [5] that can be individualized and personalized [20], semi-public as privately owned spaces accessible by a certain structural group formed by, e.g., common interests or shared activities [78], and public space as a physically and socially accessible space by everyone that is centrally managed by the city or communal authority [70,82]. ...
... last accessed September 12, 2023, © Jamie Big Sorrel Horse. 15 social media content. We also asked participants to indicate where they would feel comfortable sharing this AR content for each video. ...
Preprint
Full-text available
Augmented Reality (AR) is evolving to become the next frontier in social media, merging physical and virtual reality into a living metaverse, a Social MediARverse. With this transition, we must understand how different contexts (public, semi-public, and private) affect user engagement with AR content. We address this gap in current research by conducting an online survey with 110 participants, showcasing 36 AR videos, and polling them about the content's fit and appropriateness. Specifically, we manipulated these three spaces, two forms of dynamism (dynamic vs. static), and two dimensionalities (2D vs. 3D). Our findings reveal that dynamic AR content is generally more favorably received than static content. Additionally, users find sharing and engaging with AR co