On the Importance of Visualisation in a Data Driven Society

Daniel Archambault

Newcastle University, U.K.

Keywords:

Visualisation.

Abstract:

Machine learning and data science are receiving signiﬁcant attention and rightly so. The results that can be

produced by distilling large amounts data are amazing. Yet, what society expects is human oversight at an

appropriate level and trust of system results. Oversight and trust are not for machines; oversight and trust are

for humans. Effective solutions thus require careful human-machine collaboration and in turn careful visuali-

sation design. This paper motivates why visualisation design forms a necessary part of a data driven society.

It provides motivation for carefully designed visualisations that must take into account human perceptual fac-

tors, target audience, and the automated processes applied to the information before visualisation. It provides

practical examples where all three must be carefully thought about in order to deliver effective data science.

1 INTRODUCTION

Data science and artiﬁcial intelligence have made re-

markable strides over the past decade with results

far exceeding what we would have expected. These

methods process large amounts of data to achieve

these results. In a modern day, data-driven society,

the data, results of the models, and insights into how

the models work need to be made accessible to di-

verse user communities so that they can be informed

and make data-driven decisions.

Data science is often deﬁned as extracting knowl-

edge and insights from data. Machine learning and

data science methods are essential to solve this prob-

lem. Without them, it would not be possible to pro-

cess data at scale. However, the human side of the

data science equation is often forgotten.

Insight is not for computers;

Insight is for humans.

Machine learning and data science methods are

the engine whereas visualisation is an equally impor-

tant windscreen of a data driven society. Yes, it is true

that we would not get anywhere without scalable data

processing methods and models, but the ﬁnal step of

human consumption is often not emphasised enough.

Governments and societies see the value of (and

fear) the capabilities of artiﬁcial intelligence. They

seek oversight of the models and processes to con-

struct these models and apply them to the large data

https://orcid.org/0000-0003-4978-8479

sets. Again, this is a human-centred problem aligned

with visualisation and not one aligned with machines.

Oversight of models and AI is not for computers;

Oversight of models and AI is for humans.

As these concepts require the human, methods to

expose the models and data from the machine are

a requisite part of the solution. Thus, visualisation

is critical for effective artiﬁcial intelligence, machine

learning, and data science. Visualisation is necessary

for insight into data (and therefore an essential part of

data science) and is required for oversight of models

and AI. In this paper, we explore this question from a

variety of perspectives and my personal experience.

2 IS A TABLE OF STATISTICS

THE ANSWER?

Is it the case that a table of statistics is sufﬁcient

for this problem? Perhaps the simplest form of vi-

sualisation is the best? The importance of visual-

isations complementing calculations has been made

clear many decades ago in statistics (Anscombe,

1973):

A computer should make both

calculations and

graphs. Both sorts of output should be studied; each

will contribute to understanding.

More recent work further illustrates this point.

The Datasaurus Dozen (Matejka and Fitzmaurice,

Archambault, D.

On the Importance of Visualisation in a Data Driven Society.

DOI: 10.5220/0013445500003912

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 20th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications (VISIGRAPP 2025) - Volume 3: VISAPP, pages

7-10

ISBN: 978-989-758-728-3; ISSN: 2184-4321

2017) shows that many very different plots can be

generated for the same set of statistics. Many dif-

ferent patterns – even a dinosaur – could be present

in your data for the same statistics. A similar idea

can be seen in the area of graph drawing (Chen et al.,

2021; Chen et al., 2018) making graph drawings es-

sential and complimentary to graph mining and graph

statistics. Although statistics are important, they need

to be complemented by effective and relevant visuali-

sations for insight.

Also, more studies should be run on the percep-

tual effectiveness of visualisations. The visual chan-

nel used can inﬂuence the amount of order seen in the

data (Chung et al., 2016). Small multiple representa-

tions of dynamic graphs are often faster, with equiv-

alent accuracy, for dynamic graph visualisations (Ar-

chambault et al., 2011). The beneﬁts of keeping a sta-

ble drawing, also known as preservation of the men-

tal map, is surprisingly very difﬁcult to quantify (Ar-

chambault and Purchase, 2016), but there is evidence

that it helps identify speciﬁc nodes and speciﬁc paths

through a graph in a visually scalable way (Archam-

bault and Purchase, 2013).

Thus, complimentary and relevant visualisations

are needed with the proper encoding.

3 CAN DATA PROCESSING AND

VISUALISATION BE APPLIED

INDEPENDENTLY?

Another possibility is that any visualisation can be

applied independently from the data processing algo-

rithm. Unfortunately, this is not the case either as the

data processing inﬂuences how the visualisations are

perceived. Therefore, the automatic processing must

be selected with the visualisation tasks in mind. Ether

the viewer will need to know the limitations of the

automatic processing algorithm or the processing al-

gorithm needs to be selected so the chances of misin-

terpretation is minimised.

Graph sampling methods take a large graph and

generate a representative sample for visualisation or

other processing purposes. If the end goal is visu-

alisation, it is important to understand that the sam-

pling method chosen inﬂuences how the sample is

perceived. For example, relative degree is important

for perceiving nodes as high degree in a sample (Wu

et al., 2017). In terms of edge bundling, often edges

are bundled together based on similar edge properties

but this can create the appearance of false connections

and patters in the data where there are none (Lhuillier

et al., 2017). If the idea is to group edges together

with similar origins and destinations together then

this visualisation is suitable. However, if underly-

ing paths and topology is required for understanding,

edge bundling techniques that group edges with un-

derlying paths (Wallinger et al., 2022; Wallinger et al.,

2023) would be more suitable which can be com-

puted while simultaneously drawing the graph (Ar-

chambault et al., 2024).

In summary, the data processing algorithm also

can inﬂuence the visualisation. This should be se-

lected with the task and user background in mind so

that the chances of a misleading visualisation are min-

imised.

4 CASE STUDY: VISUALISING

MODELS FOR DIVERSE

AUDIENCES DURING

COVID-19

Visualisation was embedded into the UK response

to COVID-19 through RAMPVis (Chen et al., 2022;

Dykes et al., 2022): a volunteer project to support vi-

sualisation of pandemic modelling for a variety of au-

diences. The pandemic response was a concrete ex-

ample of what data scientists do. Data science meth-

ods, models, and visualisation were visible on a daily

basis in media across the world. During this time, new

strategies for developing visualisations were created

for a wide range of audiences.

In our case, we were tasked on visualisation of

simulations of policies on contact tracing data. In the

early stages of the pandemic, there was no time for

careful selection of problems or design of bespoke so-

lutions. Also, the communities that we were working

with did not know that visualisation was a ﬁeld and

how it could help. Fortunately, we did have some

experience with visualising contact tracing that we

could draw upon (Baumgartl et al., 2021), but not for

the speciﬁc problem of simulating policies on contact

tracing simulation and understanding how these mod-

els performed. In our initial sessions, existing visual-

isation tools were applied out of the box (Simonetto

et al., 2020; Arleo et al., 2022) to the simulations and

presented to pandemic modelling teams to discover

what was necessary (Sondag et al., 2022). These ses-

sions provided buy in for visualisations as initial re-

sults could be seen and errors in the simulations could

be discovered. For example, visualisations produced

from MulitDynNoS (Arleo et al., 2022) were able to

illustrate that too many random cases (imported from

outside) were set in the simulator when travel was al-

ready restricted. As the situation evolved, more be-

VISIGRAPP 2025 - 20th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications

spoke tools could be developed using standard visu-

alisation methodologies. This resulted in a ﬁnal visu-

alisation (Sondag et al., 2022) that could be used to

make sense of simulated contact tracing policies.

In a second project, we were tasked with the idea

of communicating pandemic modelling ideas to chil-

dren through interactive visualisations (Mcneill et al.,

2023). Working with Technocamps, we developed

an interactive visualisation method to help children

explore disease modelling concepts to gain a better

understanding what data scientists would do during a

pandemic. The visualisation was used by 100 young

people every month at its height, providing a real

data science problem that was relevant to their expe-

riences.

5 CONCLUSION

In summary, both insight and oversight are not for

computers; they are for humans. A way to achieve

communication of data a models inside a computer

to humans is through visualisation. Thus, visuali-

sation forms an equally important part of data sci-

ence and AI. This has been known for a long time

in statistics, but sometimes requires repeating. These

visualisations cannot be applied blindly without con-

sidering the underlying algorithms used to process

the data as it inﬂuences how the visualisation is per-

ceived. A case study on how this work was applied

during the COVID-19 response was discussed along

with human-centred development strategies and con-

siderations for communicating data science concepts

to diverse audiences.

ACKNOWLEDGEMENTS

For the purpose of open access, the authors have ap-

plied a Creative Commons Attribution (CC-BY) li-

cence to any Author Accepted Manuscript version

arising from this submission.

REFERENCES

Anscombe, F. (1973). Graphs in statistical analysis. The

American Statistician, 27(1):17–21.

Archambault, D., Liotta, G., N

ollenburg, M., Piselli, T.,

Tappini, A., and Wallinger, M. (2024). Bundling-

Aware Graph Drawing. In 32nd International Sym-

posium on Graph Drawing and Network Visualization

(GD 2024), volume 320, pages 15:1–15:19.

Archambault, D., Purchase, H., and Pinaud, B. (2011). An-

imation, small multiples, and the effect of mental map

preservation in dynamic graphs. IEEE Transactions

on Visualization and Computer Graphics, 17(4):539–

552.

Archambault, D. and Purchase, H. C. (2013). Mental map

preservation helps user orientation in dynamic graphs.

In Graph Drawing (GD ’12), pages 475–486.

Archambault, D. and Purchase, H. C. (2016). Can anima-

tion support the visualisation of dynamic graphs? In-

formation Sciences, 330:495–509.

Arleo, A., Miksch, S., and Archambault, D. (2022). Event-

based dynamic graph drawing without the agonizing

pain. Computer Graphics Forum, 41(6):226–244.

Baumgartl, T., Petzold, M., Wunderlich, M., Hohn, M., Ar-

chambault, D., Lieser, M., Dalpke, A., Scheithauer, S.,

Marschollek, M., Eichel, V. M., Mutters, N. T., Con-

sortium, H., and Landesberger, T. V. (2021). In search

of patient zero: Visual analytics of pathogen transmis-

sion pathways in hospitals. IEEE Transactions on Vi-

sualization and Computer Graphics, 27(2):711–721.

Chen, H., Soni, U., Lu, Y., Huroyan, V., Maciejewski, R.,

and Kobourov, S. (2021). Same stats, different graphs:

Exploring the space of graphs in terms of graph prop-

erties. IEEE Transactions on Visualization and Com-

puter Graphics, 27(3):2056–2072.

Chen, H., Soni, U., Lu, Y., Maciejewski, R., and Kobourov,

S. (2018). Same stats, different graphs. In Graph

Drawing and Network Visualization (GD ’18), pages

463–477.

Chen, M., Abdul-Rahman, A., Archambault, D., Dykes, J.,

Ritsos, P., Slingsby, A., Torsney-Weir, T., Turkay, C.,

Bach, B., Borgo, R., et al. (2022). RAMPVIS: An-

swering the challenges of building visualisation ca-

pabilities for large-scale emergency responses. Epi-

demics, 39:100569.

Chung, D. H. S., Archambault, D., Borgo, R., Edwards,

D. J., Laramee, R. S., and Chen, M. (2016). How

ordered is it? on the perceptual orderability of visual

channels. Computer Graphics Forum, 35(3):131–140.

Dykes, J., Abdul-Rahman, A., Archambault, D., Bach, B.,

Borgo, R., Chen, M., Enright, J., Fang, H., Firat, E. E.,

Freeman, E., et al. (2022). Visualization for epidemi-

ological modelling: challenges, solutions, reﬂections

and recommendations. Philosophical Transactions of

the Royal Society A, 380(2233):20210299.

Lhuillier, A., Hurter, C., and Telea, A. (2017). State of the

art in edge and trail bundling techniques. Computer

Graphics Forum, 36(3):619–645.

Matejka, J. and Fitzmaurice, G. (2017). Same stats, differ-

ent graphs: Generating datasets with varied appear-

ance and identical statistics through simulated anneal-

ing. In Proceedings of the 2017 CHI Conference on

Human Factors in Computing Systems, CHI ’17, page

1290–1294.

Mcneill, G., Sondag, M., Powell, S., Asplin, P., Turkay, C.,

Moller, F., and Archambault, D. (2023). From asymp-

tomatics to zombies: Visualization-based education of

disease modeling for children. In Proceedings of the

2023 CHI Conference on Human Factors in Comput-

ing Systems, CHI ’23.

On the Importance of Visualisation in a Data Driven Society

Simonetto, P., Archambault, D., and Kobourov, S. (2020).

Event-based dynamic graph visualisation. IEEE

Transactions on Visualization and Computer Graph-

ics, 26(7):2373–2386.

Sondag, M., Turkay, C., Xu, K., Matthews, L., Mohr, S.,

and Archambault, D. (2022). Visual analytics of con-

tact tracing policy simulations during an emergency

response. Computer Graphics Forum, 41(3):29–41.

Wallinger, M., Archambault, D., Auber, D., Nollenburg,

M., and Peltonen, J. (2022). Edge-Path Bundling:

A Less Ambiguous Edge Bundling Approach . IEEE

Transactions on Visualization & Computer Graphics,

28(1):313–323.

Wallinger, M., Archambault, D., Auber, D., N

ollenburg, M.,

and Peltonen, J. (2023). Faster edge-path bundling

through graph spanners. Computer Graphics Forum,

42(6):e14789.

Wu, Y., Cao, N., Archambault, D., Shen, Q., Qu, H., and

Cui, W. (2017). Evaluation of graph sampling: A vi-

sualization perspective. IEEE Transactions on Visual-

ization and Computer Graphics, 23(1):401–410.

VISIGRAPP 2025 - 20th International Joint Conference on Computer Vision, Imaging and Computer Graphics Theory and Applications