=Paper=
{{Paper
|id=Vol-2413/paper05
|storemode=property
|title=
Visualization of Multi Key Performance Indicators by Dynamic Chernoff Faces
|pdfUrl=https://ceur-ws.org/Vol-2413/paper05.pdf
|volume=Vol-2413
|authors=Eduard Gerhardt
}}
==
Visualization of Multi Key Performance Indicators by Dynamic Chernoff Faces
==
https://ceur-ws.org/Vol-2413/paper05.pdf
Visualization of Multi Key Performance Indicators by
Dynamic Chernoff Faces
Eduard Gerhardt 1[0000-0002-2755-640X]
1
University of Applied Sciences Coburg, Friedrich-Streib-Str. 2, 96450 Coburg, German
eduard.gerhardt@hs-coburg.de
Abstract. Digitalization and Big Data have arrived in almost all areas of daily
life. Data are today the new oil (1). But not only the flood of data has increased,
the data have become more complex and heterogeneous. M aking a fact -based
decision is getting much more difficult. A meaningful aggregation of data to in-
formation for a better overview and for a better understanding by human beings
has become more important. One possible aggregation method of information is
a better visualization. The visualization method which is analyzed is Chernof
Faces. It is a method of glyph-based visualizations of multi-dimensional space
was developed in 70´s by Hermann Chernoff (7). They consist of different facial
features to which KPIs are assigned. A KPI is e.g. assigned to the mouth. The
larger the value of this KPI is the bigger changes the mouth its shape and vice
versa. Each facial feature has a different effect on humans. The result is a human
emotion from happy to sad. Chernoff Faces thus combine several key figures into
a facial expression, which people can quickly perceive and interpret. One prob-
lem of Chernoff faces theie static nature. As the KPIs are statically assigned to
the same face parts. What to do, if the importance of the KPIs changes for the
company? Instead of maximizing profit, maximizing sales takes center stage. The
relative importance of face features remains the same! One possible solution
would be dynamic Chernoff Faces, in which software decides company -specific
or situation-specific, which KPI are assigned to which facial features and thus
provides an overall evaluation.
Keywords: Key Performance Indicators, Data aggregation, Information visuali-
zation, Chernoff Faces, dynamic Chernoff Faces
1 Data, Information, Knowledge
The complexity of the modern world can be seen in the dashboards of modern cars.
Ford's Model T had only one speedometer. The information provided by the speedom-
eter was enough to operate the car. In current vehicles significantly more sensors are
installed, which generate much more raw data while driving. This operational data is
summarized into several pieces of information that the driver understands. In addition
to the speedometer, the engine temperature, time of arrival, average speed, distance to
the destination, etc. are constantly displayed to the driver. These information give him
a very good basis to make fact-based decisions for operating the car. The driver can
Proceedings of the XXII International Conference “Enterprise Engineering and Knowledge
M anagement” April 25-26, 2019, M oscow, Russia
speed up if the dashboard informs him that the estimated arrival time is later then
planed. The possible questions which arise here:
Is the driver able to evaluate the large number of information pieces s imultaneously?
Are the drivers not distracted from driving the car by the huge amount of infor-
mation?
Will the quality of the car driving be improved by this additional information?
1.1 Information Pyramid
The terms information and data are often falsely equated, though they have fundamen-
tally different meanings. Data (lat. Dare: give) are individual values which are repre-
sented by means of characters and appear in numerical, alphabetic, alphanumeric form
or just in form of other signs . They are produced and processed by machines. For ex-
ample, the numbers 500 or D are data coded in the first case with Arabic and in the
second case with Roman characters.
Information (lat. Informatio: explanation) are data which people only understand
when they are explained semantically. The information is the basis for knowledge be-
cause it contains messages or meanings (Fig. 1, Industrial Age).
Fig. 1. Changes in information pyramid
For example, the number 500 becomes an information when it is assigned a semantic
meaning. This can be 500 contract terminations, 500 EURO basic salary or 500 liters
of water consumption. The knowledge is a consequent action based on information .
The information 500 contract terminations may e.g. lead to the act ion of improving the
service processes for customers to increase the competitiveness.
3
1.2 Operational Data
Due to the increasing digitalization of business processes, significantly more process
data are generated in companies. For the running the business processes Online Trans-
action Processing Systems (OLTP) is used. With every business process e.g. every de-
livered customer order, every paid invoice or every hired employee operational data are
automatically generated by OLTP systems. These mass data can also be called raw
data. They are often difficult for humans to understand. They rarely provide an im-
portant insight for the management and therefore they are summarized , aggregated into
key performance indicators (KPIs). This process of aggregation can be also described
as data refinement (2 S. 591). Data must become information, otherwise people will
not understand them and they will be not able to initiate knowledge-based actions.
1.3 Key Performance Indicators
Key performance indicators, as aggregated data, have the task to capture quantitatively
recorded facts in concentrated form and to serve as a control and steering instrument
for managers (3 S. 19-20). Through the key performance indicators operational raw
data are transformed into information. People are the primary users of these infor-
mation.
In the information age two developments take place. On o ne side many companies
diversificated their business e.g. transition at VW from one Car-Company “Beetle” to
many services company. On the other side companies digitalized many different busi-
ness processes, which previously either did not exist or were hand led conventionally.
These two developments increased the variance of operational raw data. The result is
the inflation of the information which people must use for the operational and strategic
decisions.
At this point one phenomenon of information age is revealed. The transition fro m
the Ford T to modern cars , which was accompanied by the enrichment the car dash-
board with information happens now to the business. Managers have to analyze a lot
more information to make decisions. The first phenomena of this development can be
spotted in the late 90`s with the introduction of Kaplan and Norton's Balanced Score-
card as a result of the transition from the industrial to the information age (4). Whereas
in the industrial age, financial indicators such as profit, sales, costs, and ROI were
enough to control the business activity. In the information age much more indicators
are needed to run the company in a balanced way. Kaplan and Norton combined many
different and isolated indicators into a multi KPI system, such as the balanced scorecard
with perspectives as finances, processes, employees, sustainability, etc. Each perspec-
tive can hold several key performance indicators. This leads to inflation of KPIs. At
this point the same questions , regarding the driver of a modern car, can be raised. Are
managers able to process large number of information pieces simultaneously? Are they
distracted by the information amount?
The inflation of the KPIs requires a further aggregation level or meaningful visuali-
zation of multi KPIs (Fig. 1, Information Age) so that people can understand and pro-
cess this information. Will the quality of decisions increase? If that does not happen,
there is a risk that multi KPI systems will not be understood by humans and will degen-
erate into data.
Due to the flood of KPIs, methods of data visualization as a further aggregation level
have to be analyzed in detail. Data visualization is the systematic, rule-based, external,
permanent and graphic representation of information in order to gain insights, to de-
velop understanding and to communicate to people (5 S. 1). Humans are at the forefront
of visualization. As a possible visual aggregation of multi KPIs, Chernoff faces are
considered below. The central question is: "Is it possible to use Chernoff-faces to ag-
gregate multiple KPIs to make faster and better decisions?"
2 Visualization as an Aggregation of the Key Figures
2.1 Basics of Informati on Visualization
Seeing is done in different steps. When light falls on the retina, it is transmitted to the
visual cortex. Already during the transport, the information is partly processed. This
includes edge recognition, orientation recognition, segmentation, motion detection and
color processing. Significantly, these processes take place without direct attention, that
means that they are pre-conscious. These processes are very fast and parallel (6 S. 13,
21).
Only through attention, information is actively filtered out. The subsequent pro-
cesses do not use the full information that has come to the retina. So -called feature maps
are created for each feature that is detected in the first phase. There are e.g. a map
highlighting the red color, a map for movement, a map for horizontally oriented objects,
etc. These maps serve as a basis for attention (6 S. 150-153). Looking at the KPI total
turnover of € 1,000,000, the retina only picks up the points of light for the first time.
Late they are split into the different cards. A map is created which filters out only the
edges of each letter and number. Additional cards are created for the color of the paper,
for the color of the font and so on.
This visual information is stored in iconic memory for a short notice and is filtered
out by the attention, then processed accordingly in the visual and verbal working
memory. Here it comes to the usual restrictions of 3 to 5 objects of memory (6 S. 22,
180, 311, 377, 383). This means that people can simultaneously perceive, interpret and
evaluate a maximu m of 3 to 5 key performance indicators. As the KPIs become more
complex, it becomes more difficult or even impossible to completely utilize the upper
limit (5 object of memory). The modern multi KPI systems, such as e.g. the Balanced
Scorecard far exceed the biologically limited number of memory items. At this point,
the next problem of the information age is revealed. On the one hand, the amount of
key performance indicators which are relevant to run the company in a balanced way
is increasing permanently. On the other hand, humans are encountering biological lim-
its of information processing that nature has given. The car driver of modern cars and
the manager of a up to date companies are reaching biological limits to process multi
KPIs simultaneously because there are too many KPIs.
5
2.2 Information Visualization by Chernoff Faces
The idea of glyph-based visualizations is that single KPIs of a multi KPI system are
shown graphically, e.g. represented by dashes, symbols (6 S. 163). Chernoff Faces is a
method of glyph-based visualizations of multi-dimensional space was developed in
70´s by Hermann Chernoff (7). They consist of different facial features to which KPIs
are assigned. A KPI is e.g. assigned to the mouth. The larger the value of this KPI is
the bigger changes the mouth its shape and vice versa (Fig. 2).
Fig. 2. Chernoff Face
Each facial feature has a different effect on humans. The mouth shape, for example,
change the perceived emotion from happy to sad. Chernoff Faces thus combine several
key figures into a facial expression, which people can quickly perceive and interpret.
Chernoff claimed that faces are part of everyday’s life for people and therefore even
small changes are easily recognized. He suggested that cartoon or cartoon faces are
sufficient for the recognition of emotions. It was also mentioned that Chernoff Faces
are perceived as a whole, as an emotion. (7 S. 363). Thus, individual key figures are
perceived as a unit respectively as a system. This goal is also tracked by aggregation of
data or by multi KPI systems.
An important task of visualizations is to highlight the important details to draw at-
tention to them. This feature is called salience or relative importance. People focus on
different facial features. The importance of the individual parts of the face to transmit
an emotion is as follows:
1. Curvature of mouth,
2. Size of the eyes
3. Form of the chin
4. Height of the face,
5. Angle of the eyes,
6. Length of the nose and
7. Length of the eyebrows. (8 S. 210)
The relative importance of facial features means that most important KPIs must be as-
signed to the most salient face feature (Table 1)
Table 1. Assignment of KPI to facial features based on their importance
Rank Part of the face KPI Max
1 Curvature of mouth Annual Revenue 100.000 €
2 Size of the eyes Annual cost 25.000 €
3 Form of the chin Profit 75.000 €
4 Height of the face ROI 2,50 €
6 Length of the nose Customer Satisfactory 10,0
7 Length of the eyebrows employee satisfaction 10,0
Based on the assignments in the Table 1 following Chernoff Faces are generated
(Fig. 3).
Fig. 3. Best and worst Chernoff faces, if all KPIs are perfect or miserable
Looking at these Chernoff Faces a human can much faster evaluate the overall com-
parison. The face on the left looks happier than the face on the right. That means that
the situation of the company described by the KPIs in the column “Max” is better the
situation of the company in the column “Min”. The evaluation speed of multi KPIs is
the most important advantage of the Chernoff Faces.
Chernoff Faces have a side effect of non-linearity. This can also lead to the fact that
desired effects are not observed clearly enough, even distracted from it (7 S. 363). For
example, an important metric may be on the nose, but the observer is distracted by the
variety of other facial features and instead focuses on the mouth and eye areas. This
reduces the information of the nose and instead looks for salient facial features.
The next problem is that Chernoff faces are static. What is to do, if the importance
of the KPIs changes for the company? Instead of maximizing profit, maximizing sales
takes center stage. The relative importance of face features remains the same!
Non-linearity, the statics of the Chernoff Faces and the lack of standard software to
implement the method probably have prevented the spread of this method in practice,
despite very good research results. One possible solution would be dynamic Chernoff
Faces, in which software decides company-specific or situation-specific, which KPI are
assigned to which facial features and thus provides an overall evaluation.
7
2.3 Information Visualization by Dynamic Chernoff Faces
Dynamic Chernoff Faces use the idea that the most important KPIs should be assigned
to the most salient facial features (8 S. 210). The salience order of facial features is
determined bilocally and was defined in the previous chapter. But how can the rating
of the KPIs be determined?
A very simple key figure system consisting of three KPIs with actual and target val-
ues is displayed in Table 2.
Table 2. M ulti Key Performance Indicator System
Target Relative
Weighting
KPI Actual Target achieve- Im- Rank
factor
ment portance
Revenue 40% 160.000 220.000 -27% 0,109 2
Cost 20% 110.000 150.000 27% 0,053 3
Profit 40% 50.000 70.000 -29% 0,114 1
Weighting factors are defined individually by the managers. They can be derived
from the corporate strategy. For a company that wants to penetrate the market quickly,
sales are more important than profits and costs. A company that would like to consoli-
date itself would define profit as the most important measure. In this case the company
pursues revenue/profit maximizing strategy. Cost reduction is inferior.
According to the actual data the company did not hit the target for revenue and profit,
but exceeded the target for the cost by 27%. How to assess the particular performance
based on this KPI system? Which indicator is more important for the total performance
of the company? A possible algorithm for evaluating individual measures in the multi -
measure system can be defined as follows:
𝑇𝑎𝑟𝑔𝑒𝑡−𝐴𝑐𝑡𝑢𝑎𝑙
Revaltive Importance of KPI = | ∗ 𝑊𝑒𝑖𝑔ℎ𝑡 |
𝑇𝑎𝑟𝑔𝑒𝑡
The most important KPI is the one with the highest absolute value. As a result, key
figures that deviated significantly from the target value and have higher weight are per-
ceived as more important key figures. So here is the order of KPIs: profit, revenue and
costs. The next step is to assign the KPIs to the face features according their salience :
Profit to the mouth, revenue to the eyes and cost to the chin. The face emotion based
on single face features will deliver the assessment of the total performance of the com-
pany.
2.4 Empirical Test
The question "Do dynamic Chernoff faces lead to faster and more correct evaluation of
multi KPIs systems?" can be answered with an empirical test. The test was conducted
with 168 subjects by Tim Stahringer at the University of Applied Sciences in Coburg
(9). Subjects had to compare companies based on KPI systems. The test subjects were
presented several fictitious companies. There was always an objective ranking between
companies, such as Table 3: Company 1 is better than company 2.
Table 3. KPI systems of two different companies presented as grid
Company 1
Weighting
KPI Actual Target
factor
Revenue 40% 160.000 220.000
Cost 20% 110.000 150.000
Profit 40% 50.000 70.000
Company 2
Weighting
KPI Actual Target
factor
Revenue 20% 190.000 200.000
Cost 40% 180.000 185.000
Profit 40% 10.000 15.000
A group of subjects is presented the company's KPIs in form of grid as in Table 3,
the other group in the form of bullet graphs (Fig. 4), while another group was presented
the same key figures in the form of dynamic Chernoff Faces (Fig. 4Fig. 5). The subjects
had to evaluate the individual performance of many different companies and to rank
the companies.
9
Fig. 4. KPI systems of two different companies presented as Bullet-Graph
Fig. 5. KPI systems of two different companies presented as dynamic Chernoff Faces
The results of the survey were evaluated by means of Kendall Tau. The Tau = 1
means that the subject has ranked in a pairwise comparison correctly and the Tau = -1
means the rating was wrong. Per subject an average Tau was calculated. The Table 4
demonstrates that the subject S6 has made the evaluation of all three company pairs
correctly and that the subject S5 evaluated everything wrong.
Table 4. Calculation of average tau by subject
Subject Tau for pair- Tau for pair- Tau for pair- Average tau
wise com- wise com- wise com-
parison 1 parison 2 parison 3
S1 +1 -1 +1 +1/3
S2 +1 +1 -1 +1/3
S3 +1 -1 -1 -1/3
S4 -1 +1 -1 -1/3
S5 -1 -1 -1 -1
S6 +1 +1 +1 +1
S7 +1 -1 -1 -1/3
Total average tau -0,05
Kendall's Tau has a value range of [-1,1]. The value 1 corresponds to the perfect
truth, the value -1means the exact opposite of the truth and 0 equal to coincidence (10
S. 81-85). In the example of Table 4 is the total average tau equal to -0.05, which is
approximately coincidental.
The results of the survey at the University of Applied Sciences in Coburg can be
summarized as followed. Grids and Bullet Graphs were compared with dynamic
Chernoff Faces. All versions were based on the same KPI system with the same values.
For Bullet Graphs, the average tau was 0,28 and the average response time was 34
seconds. For the dynamic Chernoff Faces, the average tau was 0,71 and the average
response time was 9 seconds. For the most common method to present KPIs the grid
was the average tau 0,33 and the average response time was 27 seconds. All results
were significant. The study has confirmed that the visualization of multichannel sys-
tems with Chernoff Faces has resulted in faster and more accurate scores than bullet
graphs and grids. That means the transformation of single values of KPIs into a face
expression with a specific emotion helps humans to proceed mu lti KPIs simultaneously.
This could be an answer to information inflation.
2.5 Conclusion
The Chernoff Faces reflect a multi KPI system as a unit! The survey also showed if the
differences between compared company are very small, the Chernoff Faces could look
almost identical, which can be a problem. Restricting the Chernoff Faces to seven met-
rics because of the most salient face features can be remedied by incorporating less
salient facial features. The biological restriction of humans up to 3- to 5 KPIs in the
iconic memory can be bypassed by presenting the KPIs in form of Chernoff Faces .
Thus, dynamic Chernoff-Faces provide a big added value, as more information can be
perceived by humans at the same time.
References
1. Krzanich, Brian. Data is the New Oil in the Future of Automated Driving. Data is the New
Oil in the Future of Automated Driving. [Online] 11 07, 2017. https://newsroom.in-
tel.com/editorials/krzanich-the-future-of-automated-driving/.
2. Gerhardt, Eduard. M ultidimensionale Datenanalyse. WISU-Wirtschaftsstudium. 5 2016, S.
590-596.
3. Reichmann, Thomas. Controlling mit Kennzahlen und M anagement-Tools. Vah-len : s.n.,
2006.
4. Kaplan, Robert und Norton, David. Balanced Scorecard: Strategien erfolgreich umsetzen.
s.l. : Schäffer-Poeschel Verlag, 1997.
5. Lengler, Ralph und Epller, M artin J. Towards a periodic table of visualization methods for
management. . IASTED Proceedings of the Conference on Graphics and Visualization in
Engineering (GVE 2007). Clearwater, Florida, USA. 2007.
6. Ware, Colin. Information visualization : Perception for design. Amsterdam, Bos-ton : 3rd
ed.: Elsevier/M K, 2013.
7. The Use of Faces to Represent Points in k-Dimensional Space Graphically. Chernoff, Her-
man. 1973, Journal of the American Statistical Association 68 (1973), S. 361–368.
8. A M ULTIVARIATE GRAPHIC DISPLAY FOR RE-GIONAL ANALYSIS. Huff, David
L. und Black, William. Elsevier : s.n., 1978, Bde. S. 199–218.
9. Stahringer, TIm. Konzeption und prototypische Umsetzung der Visualisierung von M ul-
tikennzahlensystemen mit dynamsichen Chernoff Faces. [Online] 2018. http://tstah-
ringer.de/Stahringer-Chernoff-Faces.pdf.
10.Kendall, M aurice G. A new measure of rank correlation. Biometrika. 1938, Bde. 1/2, S.
81–93, 30.