Understanding the Use of Narrative Patterns by Novice Data Storytellers

Tom Blount

, Laura Koesten

, Yuchen Zhao

and Elena Simperl

University of Southampton, U.K.

King’s College London, U.K.

Imperial College London, U.K.

Keywords:

Data Story, Human-Data Interaction, Narrative Patterns, Data Visualisation.

Abstract:

Data stories are about communicating data, tailored to a speciﬁc audience, with a compelling narrative. Cre-

ating them requires a mix of data science and design skills, which can be difﬁcult for beginners. Patterns

can help, as they provide tried-and-tested solutions to commonly occurring challenges. ‘Narrative patterns’

are a particular class of patterns that support data-storytellers in structuring the presentation of data within

their story, aiding them in effectively communicating with their audience. Our aim is to understand how such

patterns are applied in practice and identify ways they could be of greater use, especially for people new to the

ﬁeld. To this end, we conduct a review of 67 data stories, created by both professional data storytellers and by

postgraduate university students studying data-science, to analyse their use of narrative patterns. Starting from

a collection of narrative patterns from the literature, we explore which patterns are used more often, either on

their own or in combination, and which ones beginners struggle with. From the ﬁndings we derive recommen-

dations on how to reﬁne some of the less accessible patterns and for training and tool support, which would

allow wider audiences to articulate their data insights effectively.

1 INTRODUCTION

Data storytelling has become a top priority in many

professional roles. Major media outlets, including the

New York Times, the Wall Street Journal or Reuters

have set up data journalism teams to provide infor-

mation and analysis about the important issues of the

day, and publish their data stories via dedicated ac-

counts on social media. Brands have discovered info-

graphics and other visual renderings of data as a way

to boost trafﬁc — for instance, to promote Narcos,

a show that tells the story of Pablo Escobar, Netﬂix

launched a data story that talks about the economy

of Columbian cocaine trade in a socially engaging

way.Scientists are increasingly mindful of the impact

of their work in a broader context and use data sto-

ries to inform and raise awareness of science-related

topics (Eccles et al., 2008; Kwan-Liu Ma et al., 2012).

In this paper, we follow Lee et al. (2015)’s def-

inition of ‘data stories’ as facts backed by data, vi-

sualisations, and/or, annotation, with meaningful nar-

ration. We distinguish stories from other forms of

data-related communication such as charts or dash-

boards, which tend to focus on the data and the means

to represent it visually. By contrast, data stories re-

quire their authors to take a broader, holistic view of

what they are trying to say, and outline the basic struc-

ture of their message before deciding on the visuals to

be rendered. Studies found them more intuitive and

engaging than less thought-through combinations of

charts and text (Gershon and Page, 2001). Interactiv-

ity creates a vast array of possibilities to engage with

the underlying data and tailor the storyline to the in-

terests of the reader (Kosara and Mackinlay, 2013).

Creating data stories requires a mix of data science

and design skills, which can be difﬁcult for begin-

ners (Lee et al., 2015). Patterns can help, as they pro-

vide tried-and- tested solutions to commonly occur-

ring challenges. In this paper, we focus on a particu-

lar class of patterns called ‘narrative patterns’, which,

in a data context, describe the order and manner in

which the storyteller communicates the data to their

audience (Branston and Stafford, 2010; Bach et al.,

2018a).

Narrative patterns have been explored in depth in

literature and media studies (Branston and Stafford,

2010). Narrative guidance has shown to have a

positive effect on writing tasks (Kim and Monroy-

Hernandez, 2016). In a data context, such patterns

have emerged only recently Bach et al. (2018a) and

are not extensively supported by methodologies and

tools. Convinced of their utility, our aim with the

present study is to get a better understanding of how

patterns are applied in practice and identify ways they

128

Blount, T., Koesten, L., Zhao, Y. and Simperl, E.

Understanding the Use of Narrative Patterns by Novice Data Storytellers.

DOI: 10.5220/0010121601280138

In Proceedings of the 4th International Conference on Computer-Human Interaction Research and Applications (CHIRA 2020), pages 128-138

ISBN: 978-989-758-480-0

could be of greater use, especially for people new to

the ﬁeld. Our paper is organised according to the fol-

lowing research questions:

RQ1: Which patterns do storytellers prefer?

RQ2: Are certain patterns used in combination?

RQ3: Which patterns are not used correctly by

beginners?

RQ4: How do preferences differ between begin-

ners and experienced professionals?

Starting from the collection of 18 narrative pat-

terns proposed by Bach et al. (2018a), we analyse a

sample of 67 data stories at both ends of the expe-

rience spectrum: 43 stories created by data-science

students who have taken a data visualisation course,

and 24 award-winning stories. For the student stories,

we consider the reported usage of narrative patterns,

as well as the actual usage to identify patterns that

are popular with beginners, both on their own and in

combination, and those that beginners struggle with.

We compare them with the patterns found in the work

of rather experienced storytellers to get a sense of pat-

tern uptake and areas where more training and support

is needed for ﬂedgling practitioners.

We propose recommendations on how to reﬁne

some of the less accessible patterns from Bach et al.,

including new categories, support for additional data

manipulation features, and built-in interactive ele-

ments, which would increase their ease of use and al-

low wider audiences to articulate their data insights

more effectively.

2 RELATED WORK

We start with a brief overview of patterns and infor-

mation design, followed by an account of narrative

patterns that apply speciﬁcally to data stories, and

a review of tool support, following the classiﬁcation

of genres and structures proposed by Segel and Heer

(2010).

2.1 Patterns and Information Design

Design patterns provide repeatable, reusable solutions

to recurring design problems in virtually any area of

design (Borchers, 2000). They help beginners by pro-

viding a common, accepted language that captures the

intent behind a design (Gamma et al., 2002). They

facilitate interdisciplinary work (Borchers, 2000) and

computer support (Budinsky et al., 1996). Closer to

our ﬁeld, there is a huge body of literature in areas

such as software engineering (Gamma, 1995), user

interface design (Granlund et al., 2001), or ontology

design (Gangemi and Presutti, 2009) that deﬁnes, ap-

plies and assesses the use of patterns.

Information design has established approaches

and guidelines to choose the best medium and struc-

ture to communicate insights to a given audience.

This process involves different skill sets and levels

of complexity (Segel and Heer, 2010). Data visual-

isation recommends speciﬁc classes of charts and vi-

sual encoding to support speciﬁc types of cognitive

tasks (Ware, 2012) and warns against the effects of

misleading or ineffective charts when presenting data

to broad audiences (Kong et al., 2019).

When building a data story, the designer needs to

consider the main message and its intended audience,

structure the storyline and decide which media types

are best suited to support a particular part of the plot.

Breaking down this complex process into sub-tasks

is useful especially for beginners, and ultimately en-

ables the development of tools to support storytellers

in their work.

2.2 Narrative Patterns for Data Stories

There is extensive literature discussing the formalisa-

tion of storytelling for different genres (Branston and

Stafford, 2010; Reagan et al., 2016). Data stories, as a

relatively new form of media, follow their own regular

patterns to construct and communicate meaning, for

instance by organising charts into speciﬁc sequences

that are known to aid understanding and decision sup-

port. These patterns can be genre-speciﬁc (e.g. in-

stance data games) or apply more widely to speciﬁc

types of data (e.g. time series) or presentation modes

(e.g. slides) (Bach et al., 2018a). Bach et al. have in-

troduced design patterns for data comics, a data sto-

rytelling genre that is said to be more effective, en-

gaging, and easier to understand and recall than in-

fographics (Bach et al., 2018b; Wang et al., 2019).

Brehmer et al. (2017) have investigated timeline pat-

terns used in data stories. (Tang et al., 2019) have

explored how people transform a narrative into hand-

drawn storyline visualisations. Hullman et al. (2017)

have evaluated how people order charts in a sequence.

These studies have advanced the ﬁeld and contributed

to our knowledge about how patterns are applied to

practice to construct the ﬂow of data stories. Unlike

them, we focus on beginners, and explore their use

and understanding of a range of narrative patterns in

stories they designed on their own, on topics of their

choice, using relevant data.

Our paper refers to the collection of 18 narrative

patterns introduced by Bach et al. (2018a), which are

based on existing data stories in the literature and the

Understanding the Use of Narrative Patterns by Novice Data Storytellers

129

web. As shown in Table 1, they can be broadly cat-

egorised into the ﬁve (potentially overlapping) cate-

gories:

• Argumentation (Arg): reasoning systematically

to support messages and arguments.

• Flow (Flw): helping structure the sequence of

messages and arguments.

• Framing (Frm): the way facts and events in a

story are perceived and understood through narra-

tion.

• Emotion (Etn): enhancing readers’ ability to un-

derstand and share the feelings and experiences

important to the story.

• Engagement (Egm): the feeling of being part of

the story, of being connected to it and being in

control over the interaction with the story’s con-

tent.

2.3 Narrative Patterns in Storytelling

Tools

While multiple frameworks and tools for data story-

telling have been proposed (Cruz and Machado, 2011;

Kim et al., 2018; Gratzl et al., 2016; Bongshin Lee

et al., 2013; Satyanarayan and Heer, 2014; Amini

et al., 2017; Kim et al., 2019; Metoyer et al., 2018;

Hullman et al., 2013; Gao et al., 2014), only a few re-

cent ones implicitly support narrative patterns. Data-

Toon (Kim et al., 2019) and DataClips (Amini et al.,

2017) help build the ﬂow of a data story. The auto-

matic transition function in DataToon analyses two

related charts and generates a visual transition be-

tween them, which helps the author to gradually re-

veal their differences. DataClips allows to adjust the

animation speed of stories. Other works support fram-

ing and emotion patterns — for instance, Zhao et al.

(2015) propose to use comic character to present data

stories, which breaks the ‘virtual wall’ between story

and reader. Metoyer et al. (2018) apply text process-

ing techniques to automatically extract key facts that

can help link data points with sports players, show-

ing the human element behind the data. Our study

systematically analysed the use of a range of narra-

tive patterns to provide recommendations about how

they could be made more accessible by beginners and

supported by tools (Kim et al., 2019).

3 METHODOLOGY

3.1 Participants

The participants in our study were students from a

12 week postgraduate data visualisation course with

both taught and lab components, which covered top-

ics such as human-data interaction, basic types of

charts, visual perception, misleading with charts, in-

teractivity, and storytelling. The lectures include dif-

ferent classes of patterns and best practices, includ-

ing the narrative patterns from Bach et al. (2018a).

The students were also introduced todata visualisation

tools and libraries (e.g. Tableau and D3.js), which

equipped them with the skills required to implement

a story. To aid the understanding of patterns, we pro-

vided ‘design cards’, which described each pattern

and how it could be used alongside a screenshot with

a real-world example and the link to access the exam-

ple online.

Conducting the study in the context of a data visu-

alisation course gave us access to a relatively homo-

geneous population, with known and comparable data

science and design skills, reducing the effects of vary-

ing skill levels in the understanding and application of

patterns.

3.2 Data

The participants created data stories with at least three

different charts, on a topic of their choice, using li-

braries and tools they felt most comfortable with, as

part of their ﬁnal coursework. They were asked to

document their work, motivate the choice of charts

and structure of the story, and comment on which nar-

rative patterns they found useful. After discarding in-

complete reports, we were left with 43 data stories.

In addition, we collected 24 data stories created

by professional data storytellers from the Data Jour-

nalism Awards

between 2014 and 2019. To create

the sample, we examined the stories that were recip-

ients of an award during this time and included all

those that were accessible online at the time of writ-

ing. While we do not have an account of the patterns

the authors intended to use, we make the assump-

tion that they were capable of using patterns correctly

based on their professional qualiﬁcations.

3.3 Methods

Each of the 43 data stories authored by beginners

were accompanied by a self-reported account of the

https://datajournalismawards.org

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications

130

Table 1: Narrative design patterns and the categories they belong to Bach et al. (2018a).

Pattern name (acronym) Arg Flw Frm Etn Egm Description

Addressing the audience (addr) X X X Allowing the audience to become part of a narrative.

Breaking the 4th wall (wall) X X Subjects in a narrative unexpectedly addressing the audi-

ence.

Call for action (call) X Asking the audience to take actions to solve the issues pre-

sented in the narrative.

Compare (compare) X Showing multiple visualisations juxtaposed and highlight-

ing the difference between them.

Concretise (concretise) X X Transforming abstract concepts or numbers into solid and

known references.

Convention breaking (conv) X Breaking established graphical convention to convey sur-

prising information.

Defamiliarisation (defam) X Presenting known and familiar objects in an unexpected

way to make the audience to rethink in a different way.

Exploration (explore) X X Giving audience the freedom to actively interact with data.

Familiarisation (fam) X X Making the narrative more personal based on the knowl-

edge about the audience.

Gradual reveal (reveal) X X X Unfolding a narrative in a hierarchical way (e.g., different

granularity or subsets).

Humans-behind-the-dots (hu-

mans)

X Connecting the used data with the subjects (e.g., persons,

characters) behind the data.

Make-a-guess (guess) X X Asking the audience to take part in the narrative to ﬁnd out

the conclusions.

Physical metaphor (metaphor) X Using direction and space in visualisations to convey dif-

ferent kinds of information.

Repetition (repeat) X X Using the same type of visualisations to present an effect

repeatedly through different data dimensions.

Rhetorical question (question) X X X X Presenting the argument of a narrative as a question.

Silent data (silent) X Emphasising the argument of a narrative by de-

emphasising or hiding some data.

Speed-up/slow-down (speed) X X X Using the speed of animations to show the change in the

intensity and urgency.

Users-ﬁnd-themselves (users) X X Asking the audience to ﬁnd out the conclusion of a narra-

tive by themselves.

narrative patterns applied. We reviewed these reports

manually to identify those that were more or less pop-

ular, either individually or in combination. We then

looked at each story to spot discrepancies between

self-reported and actual usage, which gave us an idea

of the patterns novices might struggle with. We used

the work of Bach et al. (2018a) as well as the con-

tent of the design cards and other course material the

students received as a frame of reference to ascer-

tain expected usage and deviations from it in prac-

tice. Finally, we compared the use of patterns between

novices and experts by analysing the set of 24 award-

winning stories regarding the narrative patterns they

applied.

4 RESULTS

We report the usage of individual patterns as well as

of pattern combinations within data stories and how

much the self-reported usage overlapped with actual

usage. Finally we discuss pattern usage by profes-

sional data storytellers and identify commonalities

and differences

As discussed in Section 2, the 18 patterns belong

to ﬁve categories: argumentation, ﬂow, framing,

emotion, and engagement. We found that the stu-

dent participants (N = 43) chose patterns mostly in

the ﬁrst two groups, speciﬁcally compare and ex-

plore, as well as pattern combinations from these

two groups. For most patterns, both popular and less

popular, we could identify discrepancies between re-

ported and actual usage, which suggest that additional

support may be needed or at issues with the patterns

themselves. Award-winning storytellers share pref-

erences with the less experienced participants, though

they also apply more advanced patterns which the stu-

dents found challenging.

4.1 Overall Usage

As shown in Figure 1, the median number of uniquely

used patterns per story was three, out of a set of 18.

Understanding the Use of Narrative Patterns by Novice Data Storytellers

131

Figure 1: Distributions of the number of used patterns (left,

mean = 3.65, median = 3.00, standard deviation = 1.75)

reported by the participants and the percentage of cor-

rectly used patterns among all reported used patterns (right,

mean = 65%, median = 67%, standard deviation = 30%).

Notches indicate 95% conﬁdence interval around median.

In addition, some patterns were not applied well: the

median share of correctly used patterns was 67%.

Figure 2 shows the detailed reported usage. Each

column in the heat map is a narrative design pattern

and each row is the reported usage from a participant.

A green coloured cell means that the actual usage of

the pattern matches the reported usage. An orange

coloured cell means that there is a discrepancy be-

tween them. The wall and speed pattern were not

reportedly used by anyone.

4.2 Preference of Narrative Patterns

Based on the overall usage, we further analysed the

participants’ preferences of each individual pattern, in

order to answer RQ1, i.e., which patterns are most

favoured. As shown in Figure 3, compare has the

highest usage (33), followed by explore (26). The

ﬁrst helps making a point by showing differences in

the data, while the second is a means to structure the

ﬂow of a story.

Reported usage of eight of the 18 patterns, mostly

from the framing and emotion categories, was

sparse, with 5 or less participants commenting on

them. These include: humans, metaphor, repeat,

silent, guess, and defam, while wall and speed were

not mentioned at all. The reasons for this are varied.

A pattern such as humans is relevant only when the

data is about people, which was not always the case

in our sample. Using repeat requires the authors to

ﬁnd multiple dimensions in the data and to plot these

dimensions through the same type of visualisation to

build up an argument. Finally, guess asks for feed-

back from the audience, which makes the story more

complex to develop.

As noted earlier, wall and speed were not men-

tioned by any participants. The former requires the

presence of a narrator, who directly (and sometimes

unexpectedly) addresses the audience. This imposes a

particular authorial style on the narrative, which may

not be suitable for all media types, and less familiar

to students reading a data science degree. The speed

pattern can only be realised with the use of animations

Figure 2: Reported usage from 43 novice participants. Each

column represents a narrative pattern, each row represents a

participant.

or videos, which again are more complex to develop.

4.3 Preferred Pattern Combinations

To address RQ2, we examined which patterns were

most often mentioned in combination in the students’

reports.In line with the most popular patterns dis-

cussed earlier, the pair explore-compare achieved the

highest reported usage. As noted before, explore is

a ﬂow pattern that helps a story unfold by allowing

the audience to freely engage with data. Compare

is about argumentation and juxtaposes two charts to

hint at differences and trends. This corresponds to

a design where the author does not dictate the order

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications

132

Figure 3: Beginners’ pattern preferences. Compare and ex-

plore have the highest number of usage. Wall and speed

were not used by any participants.

in which the reader would need to inspect the story,

but rather allows them to look around freely, while

placing directly related charts in proximity to each

other (Kanizsa, 1979). Another commonly used ﬂow-

argumentation sequence was reveal-compare. The

students also reported applying a mix of different ﬂow

patterns, such as reveal-explore, to build up their sto-

ries. Both reveal-compare and reveal-explore match

the Martini glass story model proposed by Segel and

Heer (2010), which uses a tight narrative ﬂow (reveal)

at the beginning to set the stage and bring the main

points across and then allows the audience to freely

explore.

4.4 Usage Correctness

To answer RQ3, we investigated whether the novice

participants applied their reported narrative patterns

in their stories correctly. Figure 4 shows the percent-

age of correct usage of each pattern. Guess, com-

pare, and question achieved the highest scores. Some

of them, like compare are relatively easy to use and

were used extensively. Others, like guess and ques-

tion, were mentioned less often (as shown in Fig-

ure 3), but most of those who applied them, were able

to do so correctly.

In four cases (call, reveal, addr, and fam) less

than half of the participants applied them well, which

suggests challenges by beginners. To understand

these challenges, we reviewed the stories and iden-

tiﬁed common themes in the errors made.

The most persistent mistake when using call was

that participants did not communicate the kind of ac-

tion needed from the audience explicitly. In partic-

ipants’ reports, they expressed that the ﬁnal conclu-

sions of their stories would make their audience think

about the situation and may change the audience’s be-

haviour. However, they did not explicitly built in a

‘call’ in their narratives, or the speciﬁc subsequent

Figure 4: Percentage of correct usage of patterns. Guess,

compare, and question have the highest percentage of cor-

rect usage. Wall and speed are not shown here, as they were

not reported by the participants.

action they desired their audience to take. Similarly,

some participants chose to use addr, but failed to talk

to their audience explicitly. Both cases could be tack-

led by speciﬁc guidance, or, as we will discuss later

on, through bespoke tool support with templates that

would explicitly require the designer to name the ‘call

for action’ or reach out to the audience or include best

practice examples.

Participants struggled with reveal, one of the most

popular patterns (as shown in Figure 3). We noticed

a mismatch between the linear structure several par-

ticipants used to connect different parts of their story

and the hierarchical model assumed by reveal. This

pattern requires the designer to ﬁrst process the data

according to this hierarchy, which is in turn reﬂected

in the build-up of the story.

Although some participants tried to use fam to

bring their stories closer to their audience, many of

them failed to explicitly ask the audience to provide

additional information for personalisation. Instead,

the authors’ made their own assumptions about their

audience.

A second group of patterns, consisting of concre-

tise, defam, silent, and users, were only used cor-

rectly 50% of the time. A common issue for concre-

tise was that the participants failed to transform the

more general concepts they were trying to convey via

familiar references. Instead, they simply presented

the information in a chart (e.g., showing countries on

a map), but did not explicitly connect the abstract in-

formation with the frame of reference. In the case of

defam, which was mentioned only in a few reports,

the participants presented the data creatively, but the

choices they made were not a good ﬁt for the data.

This suggests that introducing patterns may interfere

with best practice on the use of standard charts.

Understanding the Use of Narrative Patterns by Novice Data Storytellers

133

Some participants confused silent and compari-

son. For the former, they actually used comparison,

especially when the difference between the relevant

variables was obvious. No data was hidden or de-

emphasised. For users, the incorrect cases did not

explicitly encourage the audience to engage with the

story further. Neither did these make use of interac-

tive elements to facilitate this.

Issues with explore, the second most mentioned

pattern (Figure 3) were also caused by the lack of

interaction, as the authors did not embed interaction

in their stories to allow their audience to tailor their

story (by setting values, selecting and de-selecting

variables, or blending out some of the charts)

4.5 Pattern Usage by Professionals

We assume that the patterns used in the 24 award-

winning stories in our sample are the result of the au-

thors’ professional training and experience. As in the

stories created by data storytellers in training, com-

pare is the most used pattern (see Figure 5). This sup-

ports the general understanding that a data story, more

so than a chart or a dashboard, requires the author to

take a broader, holistic view of the message they try

to bring across, and structure the story to make con-

sistent, coherent arguments that support their claims.

Two other patterns, explore and reveal, were also

common, conﬁrming their utility in building up the

ﬂow of some of the most compelling stories in the

ﬁeld.

However, there were also some differences. The

more experienced authors applied repeat and concre-

tise often (ranked 2nd and 4th respectively). As dis-

cussed earlier, concretise resorts to well known refer-

ences to bring complex concepts across, and the prob-

lematic cases failed to make this connection. The pat-

tern repeat involves different dimensions in the data

to substantiate the same argument. While using this

type of structure is a matter of choice, the pattern

proved less popular with the beginners. Only ﬁve of

them attempted to use it, though four of them did so

correctly. Conversely, following the taught compo-

nent of the course which included a session on narra-

tive patterns, the students successfully used some pat-

terns, such as conv, defam, guess, and silent, which

were not used in the award-winning stories. Therefore

we believe that providing training on narrative pat-

terns in an accessible way can make data storytellers

aware of the full spectrum of designs and lead to more

diverse stories.

In terms of pattern combinations, profession-

als also resort to ﬂow-argumentation pairs such

as repeat-compare, explore-compare, and reveal-

Figure 5: Narrative pattern usage in award-winning stories.

Compare is the most used pattern by professional story-

tellers. Repeat and concretise are more used by profes-

sional storytellers than novice storytellers.

compare. They chose repeat more than explore to

serve the ﬂow function in ﬂow-argumentation com-

binations. Meanwhile, ﬂow-ﬂow combinations such

as repeat-explore and reveal-explore were among

the most used combinations by professionals.

5 DISCUSSION

We found clear preferences for applying some narra-

tive patterns, both individually and in combination, in

both groups of data storytellers. We also established

that beginners struggled with some of the patterns,

even when these were used extensively.

Patterns of ﬂow and argumentation are the core

of a data stories, though experienced authors can han-

dle some of the more advanced patterns (for instance

those that require additional data or references, or in-

teractive elements) better than novices. We now dis-

cuss our recommendations for using narrative patterns

to help data storytellers with varying levels of ex-

perience and point to some core issues that need to

be addressed regarding patterns’ deﬁnition, tools, and

training.

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications

134

5.1 Supporting Popular Patterns

The primary use of narrative patterns is to assist the

creation of data stories (Bach et al., 2018a). In our

study, people were presented with the patterns in the

form of design cards. In real-world data storytelling

tools, such assistance could be embedded as function-

alities, including interactive elements that describe or

recommend useful patterns and assist with their ap-

plication. More advanced tools could use these pat-

terns as templates to help build up a story, and point to

their likely impact on particular audience. Guidance

could include data from empirical studies on improve-

ments in attention, understanding, memorability, en-

gagement etc.

Our results show that both beginners and very ad-

vanced authors preferred to use patterns for argu-

mentation and ﬂow. Some tools have already started

to consider providing narrative support (Kim et al.,

2019). We welcome this and suggest recommending

users narrative patterns and pattern combinations at

different stages during the creative process. Recom-

mendations could depend on the ‘section’ of the story

being edited. For example, at the beginning of a story,

patterns such as question and fam are a good way to

make the audience engage with the content in a per-

sonalised manner, while call works well after stating

a conclusion. Reﬁning this recommendation process

warrants further investigation into the factors that af-

fect author’s preferences and the impact of particular

patterns on the reader.

5.2 Errors in Usage and How to Fix

Them

One of our ﬁndings is that there are common discrep-

ancies between the reported usage and the actual us-

age of patterns by less experienced authors. For some

patterns such as compare, most of the participants

could use and identify them correctly. Most cases

where errors were made fall into one of three cate-

gories: pattern confusion, missing data, and miss-

ing interaction.

5.2.1 Pattern Confusion

Some patterns proved more difﬁcult to apply than oth-

ers. For example, some participants failed to dis-

tinguish between addr and question. The distinc-

tion between the two is fairly subtle — according

to the framework by Bach et al. (2018a), both can

serve framing, emotion, and engagement. The latter

uses a speciﬁc way of inviting the audience to par-

ticipate in the narrative. We believe further studies

would be useful to understand whether this distinction

is needed in practice across professional roles with

different backgrounds and levels of expertise in infor-

mation design who need to communicate data in their

daily work.

5.2.2 Missing Data

Some patterns require additional data manipulation.

For example, reveal would beneﬁt from functional-

ities that process the data at varying levels of gran-

ularity. Concretise needs authors to relate abstract

notions to known references to enhance comprehen-

sion. The known references, however, are not always

contained in the data used to create a story. One pos-

sible solution to address this is to recommend known

references based on the contextual information avail-

able. For example, Riederer et al. have proposed the

use of analogies to help people relate to numerical

data (Riederer et al., 2018).

5.2.3 Missing Interaction

Apart from additional data, many patterns would ben-

eﬁt from interaction. For example, exploration as-

sumes the reader would tailor the story to their needs

or interests, for instance by choosing variables or

specifying values. Familiarisation is based on the

information that authors actively elicit from their au-

dience. Many of the problematic cases failed to build

in such necessary active interaction. Therefore, we

suggest that tools that want to use narrative patterns

as assistance need to consider providing the embed-

ded interaction together with the patterns, thereby en-

abling data storytellers to use the patterns more easily.

5.3 Support for Pattern Combinations

Our analysis revealed that some combinations of nar-

rative patterns were repeatedly used for purposes of

ﬂow and argumentation. These include explore-

compare and reveal-compare for the less advanced

storytellers, and repeat-compare, explore-compare,

and reveal-compare for those more experienced.

We suggest that data visualisation tools should

consider support for structuring the ﬂow of a story,

and link different parts of the ‘plot’ to argu-

ments. Existing models such as Kosara’s Claim-

Fact-Conclusion model (Kosara, 2017), could serve

as a starting point to template pattern combinations.

By these means, authors could be assisted in reusing

multiple patterns in a coherent way to create more

complex stories.

Understanding the Use of Narrative Patterns by Novice Data Storytellers

135

5.4 Training Novices to Use More

Complex Patterns

In our comparison between the stories created by

more or less advanced authors, we realised that pro-

fessionals preferred to use more complex patterns

(e.g., concretise and repeat) that required additional

data and interaction. These patterns were not gen-

erally used or handled well by novices, which sug-

gests the use of complex narrative patterns may re-

quire more formalised training.

However, we also found that some of the students

could use some patterns (e.g., silent and guess) that

were not used by the professionals. We believe that

providing narrative patterns as support in data story-

telling can encourage authors to use complex patterns.

More complex patterns could be supported with ad-

ditional examples and training material, to allow the

creation of richer, more compelling stories.

Apart from providing narrative patterns, demon-

strating examples of how they are applied (both cor-

rect and incorrect) would help novices adopt more

complex patterns in their stories and avoid possi-

ble mistakes. For example, many of the 24 award-

winning stories are good examples to show how to

use concretise and repeat. Thus, we suggest to ex-

tract narrative pattern usage as case studies to train

novice storytellers in addition to the design cards we

have used on the course.

5.5 Unexpected Effects

Although the provided narrative patterns helped the

participants with their story creation work, we noticed

some unexpected effects caused by the usage of cer-

tain patterns. For example, when some participants

used compare or repeat (meaning that they juxta-

posed multiple visualisations), the axes of different

visualisations were not normalised according to the

new overall range, which adds cognitive effort in un-

derstanding the visualisation.

Dimara et al. (2017) and Diakopoulos (2018)

both discuss the unintended effects which may re-

sult from anchoring charts to a narrative. For exam-

ple, charts connected by narrative patterns may sug-

gest an implied relationships between them, irrespec-

tive of whether one exists in reality. Thus whether,

and to what degree narrative patterns might cause bias

in peoples’ perception, should be further investigated

and, more importantly, can be made explicit to both

authors and audiences during pattern instruction or by

storytelling tools.

6 LIMITATIONS

In this paper, we analysed the usage of narrative pat-

terns in data stories by both novices and experienced

authors. Here we note the limitations of our study.

Our participants were all students from a

postgraduate-level data visualisation course. They

were taught the use of narrative patterns and their ad-

vantages for design. As such, they do not fully rep-

resent the general population of professionals who in-

creasingly have to use visual means to communicate

data to different audiences. However, they can be seen

as an approximation as we learned anecdotally that

the majority of them had not used or heard of narra-

tive patterns for data stories before their course.

Another limitation is the way we reviewed the 24

stories created by data journalists. For the students,

we had access to documentation which explicitly re-

ferred to the patterns they considered. RQ4 is by con-

trast based only on the stories themselves and we had

no information about how aware the authors were of

emerging patterns and their thinking process. Instead,

we assume that the authors, based on their skill sets,

are aware of best practice and, consciously or not, ap-

ply patterns such as those suggested by Bach et al.

(2018a), which draws upon experiences in data jour-

nalism. Further qualitative studies would be needed

to add context to the observed differences between the

two groups of storytellers and to understand if, why,

and how advanced authors choose to apply and mix

patterns. This would help reﬁne the pattern collec-

tion, and the way the individual patterns are deﬁned

and taught.

Furthermore, our analysis did not allow us to dif-

ferentiate between participants’ understanding of the

patterns and pattern uptake. Further work could in-

vestigate how patterns as well as pattern instructions

differ in their complexity and might so inﬂuence us-

age preferences. Similarly, future work could include

qualitative methods to understand potential barriers to

pattern usage by novices in more depth.

7 CONCLUSIONS

In this paper, we analysed the application and prefer-

ences of narrative patterns in the creation of data sto-

ries. Our results showed that there are preferred pat-

terns, both when using patterns individually as well as

in combination. We found discrepancies between re-

ported and actual usage for all narrative patterns. Pat-

terns that help with argumentation or ﬂow are ap-

plied extensively, also in combination.

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications

136

Our ﬁndings suggest several methods of support-

ing the application of narrative patterns in data sto-

rytelling tools, and ways to further improve the ex-

isting patterns to increase their accessibility and ease

of use. We also highlight possible issues that need

to be addressed for a more comprehensive pattern up-

take by authors of data stories. We believe that our

ﬁndings are useful for developers of data storytelling

tools, who want to use narrative patterns as assistance

in their systems.

There are two key areas of future work to be un-

dertaken. Firstly, a deeper investigation into pattern-

usage by professionals may reveal more information

about the differences we noted, their overall process

in constructing stories, and they way in which they

conceptualise the patterns that they use. Secondly,

exploring how storytellers (and novices in particular)

can be supported when using these patterns, to mit-

igate the issues we have highlighted. This may be

achieved by directly integrating our recommendations

into visualisation and storytelling tools (such as Excel

or Tableau) to more clearly tutorialise the storycraft-

ing process, and to recommend more than just charts

based on a dataset: for example, relevant sequences

of charts, charts showing different levels of detail in

the data, or charts that prompt the user to add a value

or spot an outlier.

We believe that to advance the information de-

sign ﬁeld we need a much better understanding of the

building blocks of data stories, and how they are used

in practical settings by both novices and profession-

als. Patterns are the basis on which data is commu-

nicated to the audience. Patterns go beyond single

chart selection and choice of visual encoding; patterns

guide authors in how to compose an engaging story.

ACKNOWLEDGEMENTS

This work was supported by the Data Stories project

(EPSRC EP/P025676/1) and the They Buy For You

project (EU Horizons 2020 #780247).

REFERENCES

Amini, F., Henry Riche, N., Lee, B., Monroy-Hernandez,

A., and Irani, P. (2017). Authoring Data-Driven

Videos with DataClips. IEEE Transactions on Visu-

alization and Computer Graphics, 23(1):501–510.

Bach, B., Stefaner, M., Boy, J., Drucker, S., Bartram, L.,

Wood, J., Ciuccarelli, P., Engelhardt, Y., Koeppen, U.,

and Tversky, B. (2018a). Narrative design patterns for

data-driven storytelling. In Data-Driven Storytelling,

pages 125–152. AK Peters/CRC Press.

Bach, B., Wang, Z., Farinella, M., Murray-Rust, D., and

Henry Riche, N. (2018b). Design patterns for data

comics. In Proceedings of the 2018 CHI Confer-

ence on Human Factors in Computing Systems, pages

38:1–38:12, New York, NY, USA. ACM.

Bongshin Lee, Kazi, R. H., and Smith, G. (2013). Sketch-

Story: Telling More Engaging Stories with Data

through Freeform Sketching. IEEE Transactions on

Visualization and Computer Graphics, 19(12):2416–

2425.

Borchers, J. O. (2000). A pattern approach to interaction de-

sign. In Proceedings of the 3rd Conference on Design-

ing Interactive Systems: Processes, Practices, Meth-

ods, and Techniques, DIS ’00, pages 369–378, New

York, NY, USA. ACM.

Branston, G. and Stafford, R. (2010). The Media Student’s

Book. Routledge.

Brehmer, M., Lee, B., Bach, B., Henry Riche, N., and Mun-

zner, T. (2017). Timelines Revisited: A Design Space

and Considerations for Expressive Storytelling. IEEE

Transactions on Visualization and Computer Graph-

ics, 23(9):2151–2164.

Budinsky, F. J., Finnie, M. A., Vlissides, J. M., and Yu,

P. S. (1996). Automatic code generation from design

patterns. IBM systems Journal, 35(2):151–171.

Cruz, P. and Machado, P. (2011). Generative Story-

telling for Information Visualization. IEEE Computer

Graphics and Applications, 31(2):80–85.

Diakopoulos, N. (2018). Ethics in data-driven visual story-

telling. In Data-Driven Storytelling, pages 233–248.

AK Peters/CRC Press.

Dimara, E., Bezerianos, A., and Dragicevic, P. (2017). Nar-

ratives in Crowdsourced Evaluation of Visualizations:

A Double-Edged Sword? In Proceedings of the 2017

CHI Conference on Human Factors in Computing Sys-

tems, pages 5475–5484, New York, NY, USA. ACM.

Eccles, R., Kapler, T., Harper, R., and Wright, W.

(2008). Stories in GeoTime. Information Visualiza-

tion, 7(1):3–17.

Gamma, E. (1995). Design Patterns: Elements of Reusable

Object-oriented Software. Pearson Education India.

Gamma, E., Helm, R., Johnson, R., and Vlissides, J. (2002).

Design patterns: Abstraction and reuse of object-

oriented design. In Software Pioneers: Contributions

to Software Engineering, pages 701–717. Springer

Berlin Heidelberg, Berlin, Heidelberg.

Gangemi, A. and Presutti, V. (2009). Ontology design pat-

terns. In Handbook on Ontologies, pages 221–243.

Springer.

Gao, T., Hullman, J., Adar, E., Hecht, B., and Diakopoulos,

N. (2014). Newsviews: An automated pipeline for

creating custom geovisualizations for news. In Pro-

ceedings of the SIGCHI Conference on Human Fac-

tors in Computing Systems, CHI ’14, pages 3005–

3014, New York, NY, USA. ACM.

Gershon, N. and Page, W. (2001). What Storytelling Can

Do for Information Visualization. Communications of

the ACM, 44(8):31–37.

Granlund,

A., Lafreni

ere, D., and Carr, D. A. (2001). A

pattern-supported approach to the user interface de-

Understanding the Use of Narrative Patterns by Novice Data Storytellers

137

sign process. In International Conference on Human-

Computer Interaction: 05/08/2001-10/08/2001.

Gratzl, S., Lex, A., Gehlenborg, N., Cosgrove, N., and

Streit, M. (2016). From Visual Exploration to Story-

telling and Back Again. Computer Graphics Forum,

35(3):491–500.

Hullman, J., Diakopoulos, N., and Adar, E. (2013). Con-

textiﬁer: Automatic Generation of Annotated Stock

Visualizations. In Proceedings of the SIGCHI Confer-

ence on Human Factors in Computing Systems, pages

2707–2716, New York, New York, USA. ACM.

Hullman, J., Kosara, R., and Lam, H. (2017). Finding a

Clear Path: Structuring Strategies for Visualization

Sequences. Computer Graphics Forum, 36(3):365–

375.

Kanizsa, G. (1979). Organization in vision: Essays on

Gestalt perception. Praeger Publishers.

Kim, J. and Monroy-Hernandez, A. (2016). Storia: Sum-

marizing social media content based on narrative the-

ory using crowdsourcing. In Proceedings of the 19th

ACM Conference on Computer-Supported Coopera-

tive Work & Social Computing, CSCW ’16, pages

1018–1027, New York, NY, USA. ACM.

Kim, N. W., Bach, B., Im, H., Schriber, S., Gross, M., and

Pﬁster, H. (2018). Visualizing Nonlinear Narratives

with Story Curves. IEEE Transactions on Visualiza-

tion and Computer Graphics, 24(1):595–604.

Kim, N. W., Henry Riche, N., Bach, B., Xu, G., Brehmer,

M., Hinckley, K., Pahud, M., Xia, H., McGufﬁn, M. J.,

and Pﬁster, H. (2019). DataToon: Drawing Dynamic

Network Comics With Pen + Touch Interaction. In

Proceedings of the 2019 CHI Conference on Human

Factors in Computing Systems, pages 105:1–105:12,

New York, New York, USA. ACM.

Kong, H.-K., Liu, Z., and Karahalios, K. (2019). Trust

and recall of information across varying degrees of

title-visualization misalignment. In Proceedings of

the 2019 CHI Conference on Human Factors in Com-

puting Systems, CHI ’19, pages 346:1–346:13, New

York, NY, USA. ACM.

Kosara, R. (2017). An Argument Structure for Data Sto-

ries. In Proceedings of the Eurographics/IEEE VGTC

Conference on Visualization: Short Papers, pages 31–

35, Goslar Germany, Germany. Eurographics Associ-

ation.

Kosara, R. and Mackinlay, J. (2013). Storytelling: The Next

Step for Visualization. Computer, 46(5):44–50.

Kwan-Liu Ma, Liao, I., Frazier, J., Hauser, H., and Kostis,

H.-N. (2012). Scientiﬁc Storytelling Using Visual-

ization. IEEE Computer Graphics and Applications,

32(1):12–19.

Lee, B., Henry Riche, N., Isenberg, P., and Carpendale, S.

(2015). More Than Telling a Story: Transforming

Data into Visually Shared Stories. IEEE Computer

Graphics and Applications, 35(5):84–90.

Metoyer, R., Zhi, Q., Janczuk, B., and Scheirer, W. (2018).

Coupling Story to Visualization: Using Textual Anal-

ysis As a Bridge Between Data and Interpretation. In

23rd International Conference on Intelligent User In-

terfaces, pages 503–507, New York, New York, USA.

ACM.

Reagan, A. J., Mitchell, L., Kiley, D., Danforth, C. M., and

Dodds, P. S. (2016). The emotional arcs of stories are

dominated by six basic shapes. EPJ Data Science,

5(1–12):31.

Riederer, C., Hofman, J. M., and Goldstein, D. G. (2018).

To Put That in Perspective: Generating Analogies

That Make Numbers Easier to Understand. In Pro-

ceedings of the 2018 CHI Conference on Human Fac-

tors in Computing Systems, pages 548:1–548:10, New

York, New York, USA. ACM.

Satyanarayan, A. and Heer, J. (2014). Authoring Narrative

Visualizations with Ellipsis. Computer Graphics Fo-

rum, 33(3):361–370.

Segel, E. and Heer, J. (2010). Narrative Visualization:

Telling Stories with Data. IEEE Transactions on Visu-

alization and Computer Graphics, 16(6):1139–1148.

Tang, T., Rubab, S., Lai, J., Cui, W., Yu, L., and Wu, Y.

(2019). iStoryline: Effective Convergence to Hand-

drawn Storylines. IEEE Transactions on Visualization

and Computer Graphics, 25(1):769–778.

Wang, Z., Wang, S., Farinella, M., Murray-Rust, D., Henry

Riche, N., and Bach, B. (2019). Comparing Effective-

ness and Engagement of Data Comics and Infograph-

ics. In Proceedings of the 2019 CHI Conference on

Human Factors in Computing Systems, pages 253:1–

253:12, New York, New York, USA. ACM.

Ware, C. (2012). Information Visualization: Perception for

Design. Elsevier.

Zhao, Z., Marr, R., and Elmqvist, N. (2015). Data Comics:

Sequential Art for Data-Driven Storytelling. Techni-

cal report.

CHIRA 2020 - 4th International Conference on Computer-Human Interaction Research and Applications

138