dataviz/docs/04-express.md

975 lines
46 KiB
Markdown
Raw Blame History

This file contains invisible Unicode characters

This file contains invisible Unicode characters that are indistinguishable to humans but may be processed differently by a computer. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

This file contains Unicode characters that might be confused with other characters. If you think that this is intentional, you can safely ignore this warning. Use the Escape button to reveal them.

# EXPRESS Choose proper visualization
EXPRESS covers all aspects of choosing the proper visualization in reports and
presentations.
*Proper visualization* means that reports and presentations contain charts
and tables, which convey the desired message along with the underlying facts as quickly
as possible.
This chapter covers utilizing the correct types of charts and tables, replacing
inappropriate visualizations and representations, adding comparisons, and explaining
causes.
1. [Use appropriate object types](#ex-1-use-appropriate-object-types)
2. [Replace inappropriate chart types](#ex-2-replace-inappropriate-chart-types)
3. [Replace inappropriate representations](#ex-3-replace-inappropriate-representations)
4. [Add comparisons](#ex-4-add-comparisons)
5. [Explain causes](#ex-5-explain-causes)
## EX 1 Use appropriate object types
Choosing the appropriate _object type_ is of prime importance for the
comprehension of reports and presentations.
We use tables when looking up data. Tables have a high information density. They
are clear, they are honest, they do not want to highlight, and they typically do
not want to visually convey a certain message. So they do not compete with
charts.
Charts on the opposite are always biased. It is the selection of data, the
selection of the chart type, and the usage of highlighting which makes the
difference. We evaluate charts by asking whether they transfer the intended
message effectively and in a proper way. So charts cannot be replaced by tables.
The following section is about choosing the appropriate types of charts and
tables. It presents in detail different types, layouts, and examples as well as
their proper application.
## EX 1.1 Use appropriate chart types
![Figure EX 1.1: Use appropriate chart types](img/ex-1.1.png)
A _chart_ is a graphical object, in which visualization elements
such as columns, bars, and lines represent data.
This section discusses the types, layout, and examples of _single charts_. _Overlay charts_ _and multiple charts_ are discussed in the CO 4 “[Add elements](06-condense.md#co-4-add-elements)” and CO 5 “[Add objects](06-condense.md#co-5-add-objects)”.
The most important groups of business charts are those showing development over time (charts with horizontal category axes), those showing structural relationships (charts with vertical category axes), and those showing xy charts, scatter plots, and bubble charts (charts with two value axes), see Figure EX 1.1.
Other chart types are of lesser interest in business communication and
will be treated in a later version of the standards.
![Figure EX 1.1-1: Chart Types](img/ex-1.1-1.png)
Looking at charts with horizontal and vertical category axes, the chart
selection matrix displayed in the figure aids in selecting
the right chart type for time series and structure analyses.
In the following sections, the correct usage of _charts with horizontal category axes_, _charts with vertical category axes_, and *charts with two value axes* is discussed in greater detail.
**Charts with horizontal category axes**
Charts with horizontal category axes (short: _horizontal charts_) typically display time series. Use the horizontal category axis as a time axis. Vertically, the visualization elements represent the data per time period or point of time (there is no need to show a vertical value axis as the visualization elements carry their own values). Time category axes run from left to right and show characteristics of period types (e.g. months or years) or points of time (dates).
In general, the data series of a _horizontal chart_ is represented by columns (single, stacked, grouped), vertical pins, horizontal waterfalls, or lines. _Vertical pins_ can be considered very thin columns. Because of their importance, they are dealt with in a separate section.
Here follows the grouping of _horizontal chart types_:
**Single column charts**
![Figure EX 1.1-2: Single column charts](img/ex-1.1-2.png)
In general, _single column charts_ (short: single columns) are used to display the temporal evolvement of one data series.
Single columns consist of:
- **Horizontal category axis:** The _horizontal category axis_ represents with its labels the respective time periods or points of time. The part on “Semantic rules” suggests to use the category width (see width A in the figure) for identifying the period type (see UN 3.3 “[Unify time periods](09-unify.md#un-33-unify-time-periods-and-points-of-time)”).
- **Columns**: One _column_ per time period or point of time extends from the category axis in accordance with the respective value. Columns are displayed in the foreground of the category axis, so that the length of the column is not hidden. The part on “Semantic rules” suggests that the ratio of column width to category width (see ratio B/A in the figure) represents information about the measure type (see UN 3.1 “[Unify measures](09-unify.md#un-31-unify-measures)”).
- **Legends**: As there is only one data series, the legend (name of the data series) is part of the chart title.
- **Data labels**: _Data labels_ name the values of the data series corresponding to the length of the respective columns. Position the labels of positive values above their respective columns, the labels of negative values below.
**Stacked column charts**
![Figure EX 1.1-3: Stacked column charts](img/ex-1.1-3.png)
_Stacked column charts_ (short: stacked columns)
represent more than one data series (e.g. multiple
products, countries, or divisions), see the figure on
the left.
Stacked columns consist of:
- **Horizontal category axis:** See single column charts.
- **Columns**: The columns (see single column charts) are divided into segments (Excel names them “data points”) representing the data series (stacked columns).
- **Legends**: Legends (names of the data series) are positioned either on the far left side with right-aligned text or on the far right side with left-aligned text. The column segments define their vertical position, centered vertically with the data labels of the respective column segment. If a segment at the far left or far right is missing or has a very small size, its legends might need assisting lines.
- **Data labels**: _Data labels_ name the values of the data series corresponding to the length of the respective column segments. If the sum of the column segments of a category is positive (column pointing upward), the label of the sum is positioned above the respective column, if negative (column pointing downward), it is positioned below.
It must be pointed out that stacked columns should only
be used if all chart values are either positive or
negative.
This chart type might also not be a good choice if the
values of each data series vary too much. The maximum
number of data series (segments of a stacked column) to
be shown depends on the range of how much the values of
each data series vary: More than 5 data series will only
work well in the case of little variations.
Position the data series of central importance or
interest directly on the axis in order to best see its
development over time.
**Grouped column charts**
![Figure EX 1.1-4: Grouped column charts](img/ex-1.1-4.png)
_Grouped column charts_ (short: grouped columns) show, in general, time series for a primary scenario (e.g. AC or FC) in comparison with a reference scenario (e.g. PY or PL). Two columns per category (_grouped columns_) represent these two scenarios.
The columns of the primary scenario and the reference scenario overlap, the reference scenario placed behind the primary scenario either to the left or right of the primary scenario (see bottom chart of the figure as well as the paragraph on ”Scenario comparisons” in UN 4.1 “[Unify scenario analyses](09-unify.md#un-41-unify-scenario-analyses)”). A third scenario could be displayed using a _reference scenario triangle_. All other elements of a grouped column chart are identical to single column charts.
Instead of using grouped columns, the primary scenario
can be represented with a single column with the
reference scenario represented by reference scenario
triangles (see top chart of the figure).
**Horizontal pin charts**
![Figure EX 1.1-5: Horizontal pin charts](img/ex-1.1-5.png)
_Horizontal pin charts_ (short: horizontal pins) are used for the visualization of relative variances in a time series analysis.
Horizontal pins consist of:
- **Horizontal category axis:** see _single column chart_.
- **Pins**: One _pin_ per time period or point of time extends from the category axis to the respective length. The pin has the size of a very thin column. Color the pin green or red corresponding with positive or negative relative variances respectively. The fill of the pinhead represents the primary scenario (see the paragraph on “Scenario comparisons” in UN 4.1 “[Unify scenario analyses](09-unify.md#un-41-unify-scenario-analyses)”). Display the pin in the foreground, so that the length of the pin (see length X in the figure) is not hidden.
- **Legends**: As there is only one data series, the legend (name of the data series) is part of the chart title.
- **Data labels**: _Data labels_ name the values of the data series consistent with the length of the respective pins. Position the labels of positive values above the respective pins, labels of negative values below.
**Horizontal waterfall charts**
_Horizontal waterfall charts_ (short: _horizontal waterfalls_ or _column waterfalls_) analyze root causes, over time, for the change or variance between two or more statuses. Accordingly, horizontal waterfalls consist of two or more _base columns and totals columns_. In between a base column and a totals column there are _contribution columns_ demonstrating what led to the difference between these two columns. The _contribution columns_ start at the end value, i.e. the height, of the preceding column, and show the positive or negative contribution as well as the accumulated contribution of all columns up to the respective point of time.
There are two types of horizontal waterfalls:
![Figure EX 1.1-6: Growth waterfalls](img/ex-1.1-6.png)
**Growth waterfalls**: In _growth waterfalls_, base columns and totals columns represent a stock measure (e.g. headcount, accounts receivable) at different points in time (e.g. end of Q4 2012, 2013 and 2014). The contribution columns in between represent the changes (increases and decreases) over time of this stock measure.
(There is no vertical equivalent to the horizontal _growth waterfall_.)
![Figure EX 1.1-7: Horizontal variance waterfalls](img/ex-1.1-7.png)
**Horizontal variance waterfalls**: In _horizontal variance waterfalls_, base columns and totals columns represent a flow measure (e.g. sales) at different periods in time (e.g. 2015 and 2016) and/or regarding different scenarios (e.g. PL and AC). The contribution columns in between represent the periodical variances between the different time periods and/or scenarios.
The elements of a horizontal waterfall chart are the same as the elements of single column charts. In addition, _assisting lines_ connect the end of a column to the beginning of the succeeding column.
**Line charts**
![Figure EX 1.1-8: Line chart](img/ex-1.1-8.png)
In general, _line charts_ are used for the display
of the temporal evolvement of data series with many data
points.
![Figure EX 1.1-9: Line chart with selective data labels](img/ex-1.1-9.png)
Many data points lead to small category widths. The advantage of line charts over column charts is the simplified display of data (better _data-ink-ratio_). In addition, they can better represent positive and negative values of more than one data series than columns. On the other hand, lines tend to imply a continuous timeline practically non-existent in business communication. Therefore lines should not be used for the presentation of data series with only a few values.
Line charts cannot be “stacked” in order to show structure like in stacked column charts. In the place of line charts for “stacked data”, *area charts*offer a good solution (there is no layout concept for area charts in this version of the guide yet).
Line charts with more than three intersecting lines tend to be confusing. Instead, several smaller charts with one line each could be placed next to one another (small multiples), particularly when the general trends of the lines are to be analyzed not the direct comparison of two data series (e.g. in comparing seasonal developments of several years), see also EX 2.4 “[Replace spaghetti charts](04-express.md#ex-24-replace-spaghetti-charts)”.
Line charts consist of:
- **Horizontal category axis:** See _single column chart_. The semantic rules in part 3 suggest to use the category width (see width A in the first figure) for identifying the period type (see UN 3.3 “[Unify time periods](09-unify.md#un-33-unify-time-periods-and-points-of-time)”).
- **Lines**: One or more _lines_ with _line markers_ represent the values of the respective data series. Use line thickness, line color, and line markers for meaning, see part “Semantic rules”.
- **Legends**: _Legends_ label the data series. If the line chart shows only one data series, include the legend in the chart title. If the line chart shows two or more data series, the legend should be positioned to the right of the far right data point (left-aligned text, see first figure) or the left of the far left data point (right-aligned text, see second figure). Alternatively position legends close to the lines at any other place in the chart.
- **Data labels**: _Data labels_ name the values of the respective data points. If possible, label maximum values (peaks) above the line markers and minimum values (valleys) below the line markers. In many practical applications it is not necessary to clutter the line chart by labeling every data point, see second figure on the left and SI 5.3 “[Avoid unnecessary labels](05-simplify.md#si-53-avoid-unnecessary-labels)”.
**Other horizontal charts**
Other chart types with horizontal category axes are *boxplot charts* (range charts) and _area charts_. There is no specific notation concept for these chart types yet however it can be easily derived from the notation concept of column and line charts.
**Charts with vertical category axes**
Charts with vertical category axes (_vertical charts_)
typically show structural data. In general, present structural
data of one period or one point of time in the form of
_bars_.
Use the vertical category as a structure axis. Horizontally, the visualization elements represent the data per structure element (there is no need for a horizontal value axis as the visualization elements carry their own values). Structure axes run from top to bottom and show characteristics of structures (e.g. products or countries). The sequence of these elements depends on the intended analysis; see the UNIFY section about “Structure analyses”.
In general, the data series of a vertical chart is represented by
(horizontal) _bars_ (single, stacked, grouped), by
_horizontal pins_, or by _waterfall bars_. Do not
use lines in vertical charts as they could be interpreted as
trends or developments, which do not exist in structure
analyses.
_Horizontal pins_ can be considered very thin bars, but
because of their importance are dealt with in a
separate section. A chart with horizontal pins is called a
_vertical pin chart_.
Here follows the grouping of _vertical chart types_:
**Single bar charts**
![Figure EX 1.1-10: Single bar charts](img/ex-1.1-10.png)
In general, _single bar charts_ (short: single
bars) are used for the structural analysis of one data
series (e.g., products, countries, or divisions) for one
period or one point in time.
Single bar charts consist of:
- **Vertical category axis:** The
_vertical category axis_ with its labels
represents the respective structure elements such as
countries, products, etc. The category width (see
width A in figure) should be the same
for corresponding analyses.
- **Bars**: One _bar_ per structure element extends from the category axis to the length representing the respective value. Display the bars in the foreground of the category axis, so that the length of the bar is not hidden. The part on “Semantic rules” suggests that the ratio of bar width to category width (see ratio B/A in figure) represents information about the measure type (see UN 3.1 “[Unify measures](09-unify.md#un-31-unify-measures)”).
- **Legends**: As there is only one data
series, the legend (name of the data series) is part
of the chart title.
- **Data labels**: _Data labels_
name the values of the data series consistent with
the length of the respective bars. Position the
labels of positive values at the right hand side of
the respective bars, the labels of negative values
at the left hand side.
**Stacked bar charts**
![Figure EX 1.1-11: Stacked bar charts](img/ex-1.1-11.png)
Stacked bar charts (short: stacked bars) represent more
than one data series (e.g., products, countries, or
divisions) for one period or one point in time.
Stacked bar charts consist of:
- Vertical category axis: See single bar charts.
- **Bars**: The bars (see single bar
charts) are divided into segments (Excel names them
“data points”) representing the data series (stacked
bars).
- **Legends**: Legends (names of the data
series) are positioned either above the top stacked
bar or below the bottom stacked bar, with the bar
segments defining their horizontal position: they
are horizontally centered with the data labels of
the respective bar segment. If a segment at the top
or bottom is missing or has a very small size, its
legend might need assisting lines.
- **Data labels**: _Data labels_
name the values of the data series corresponding to
the length of the respective bar segment. If the sum
of the bar segments of a category is positive (bar
pointing to the right), the label of the sum is
positioned to the right hand side of the respective
bar. If the sum is negative (bar pointing to the
left), the label of the sum is positioned to the
left hand side of the respective bar.
It must be pointed out that stacked bars should only be
used if all chart values are either positive or
negative.
This chart type might also not be a good choice if the
values of each data series vary too much. The maximum
number of data series (segments of a stacked bars) to be
shown depends on the range of how much the values of
each data series vary: More than 5 data series will only
work well in the case of little variations.
Position the data series of central interest directly at
the axis in order to best see its structure.
**Grouped bar charts**
![Figure EX 1.1-12: Grouped bar charts](img/ex-1.1-12.png)
In general, _grouped bar charts_ (short: grouped
bars) show structure analyses for a primary scenario
(e.g. AC or FC) in comparison to a reference scenario
(e.g. PY or PL). Two bars per category (_grouped
bars)_ represent _these_ two scenarios.
The bars of the primary scenario and the reference scenario overlap, the reference scenario placed behind the primary scenario either above or below (see the bottom chart of the figure as well as the paragraph on “Scenario comparisons” in UN 4.1 “[Unify scenario analyses](09-unify.md#un-41-unify-scenario-analyses)”).
A third scenario could be displayed using a _reference scenario triangle_. All other elements of a grouped bar chart are identical to a single bar chart.
Alternatively, instead of grouped bars, the primary
scenario can be represented with a single bar and the
reference scenario by reference scenario triangles (see
top chart of figure).
**Vertical pin charts**
![Figure EX 1.1-13: Vertical pin charts](img/ex-1.1-13.png)
_Vertical pin charts_ (short: vertical pins) are used for the visualization of relative variances in a structure analysis.
Vertical pins consist of:
- Vertical category axis: see _single bar chart_.
- **Pins**: One _pin_ per structure element extends from the category axis to the respective length. The pin has the size of a very thin bar. It is colored green or red when representing positive or negative relative variances respectively. The fill of the pinhead represents the primary scenario (see the paragraph on “Scenario comparisons” in UN 4.1 “[Unify scenario analyses](09-unify.md#un-41-unify-scenario-analyses)”). Display pins in the foreground, so that the length of the pin (see length X in the figure) is not hidden.
- **Legends**: As there is only one data series, the legend (name of the data series) is part of the chart title.
**Data labels**: _Data labels_ name
the values of the data series corresponding to the
length of the respective pins. Position the labels of
positive values at the right hand side of the respective
pins, the labels of negative values at the left hand
side.
**Vertical waterfall charts**
_Vertical waterfalls charts_ (in short: _vertical waterfalls_ or _bar waterfalls_) analyze structural root causes for the difference between two or more statuses. Accordingly, vertical waterfalls consist of two or more _base bars_ and _totals bars_. In between a base bar and a totals bar there are _contribution bars_ representing the contribution to the difference between these two bars. Starting from the top base bar, _contribution bars_ always start at the end of the preceding bar, showing positive or negative individual contributions of the respective structure element as well as the accumulated contribution resulting in the next totals bar.
There are two types of vertical waterfalls:
![Figure EX 1.1-14: Calculation waterfalls](img/ex-1.1-14.png)
**Calculation waterfalls**: In _calculation waterfalls_, base bars and totals bars represent base and result measures (e.g. sales and EBIT) whereas the contribution bars in between represent the additions and subtractions of other measures (e.g. financial income and direct cost) in a calculation scheme. More complex calculation schemes (e.g. profit and loss calculation) can have intermediate subtotals bars.
(There is no horizontal correspondence to the vertical
_calculation waterfall_.)
![Figure EX 1.1-15: Vertical variance waterfalls](img/ex-1.1-15.png)
**Vertical variance waterfalls**: In _vertical variance waterfalls_, base bars and totals bars represent values at different periods or points in time (e.g. January 1, 2013 and January 1, 2014) and/or different scenarios (e.g. PY and AC). The contribution bars in between represent the variances in structure between the different times and/or scenarios.
The elements of vertical waterfalls are the same as the elements of single bar charts. In addition, _assisting lines_ connect the end of a bar with the beginning of the succeeding bar.
**Remainder bar**
![Figure EX 1.1-16: Remainder bar](img/ex-1.1-16.png)
If a large number of elements need to be presented, then
only the most important elements can be displayed in one
chart or on one page. In order to make the analyses
exhaustive, sort the elements by descending size,
accumulating the smallest elements, which cannot be
depicted, in a _remainder bar_ (“rest of…”). Separate the remainder bar from the
other bars by a wider gap (see gap C in the figure on
the left).
Note: This remainder bar has to be excluded from some Structure analyses such as averaging, ranking, and selecting.
**Other vertical charts**
Other chart types with vertical category axes are
_vertical boxplot charts_ (range charts). There
is no specific notation concept for this chart type yet
however it can be derived from the notation of the
standard bar charts.
In general, do not use lines and areas in vertical charts
as they might underline a continuum of data non-existent
in business communication.
**Charts with two values axes**
![Figure EX 1.1-17: Charts with two values axes](img/ex-1.1-17.png)
_Charts with two value axes_ show two-dimensional
positioning of visualization elements, which can provide new and
interesting insights. *Scattergrams* arrange points
in a two-dimensional coordinate system.
![Figure EX 1.1-18: Bubble charts](img/ex-1.1-18.png)
*Bubble charts* (portfolio charts) have bubbles
instead of points and use the bubble area to show a third
dimension. A fourth dimension could be
presented via pie segments within the bubbles (bubble pie
charts).
Besides _scattergrams_ and _bubble charts_ there
are other chart types with two value axes, e.g. charts with
horizontal axes representing a continuous timeline (instead of
fixed time categories) and charts with columns or bars of
variable width.
There are no specific notation rules for charts with two value axes yet. An appropriate notation concept for these chart types can be derived from the notation of column charts, bar charts and line charts with their data visualization elements, legends, data labels, and axes.
## EX 1.2 Use appropriate table types
A *table* is a communication object in which data is arranged
in two dimensions, i.e. (vertical) _columns and_ (horizontal)
_rows_. The _row header_ (row name) describes the content
of a row, the _column header_ (column name) the content of a
column. The data are positioned at the intersections of rows and columns
called _table cells_.
“One-dimensional tables” (tables with one or more columns but without row
headers) are called _lists_ and are not covered here.
_Table types_ are defined by a set of _columns_ and a set
of _rows_ in order to fulfill specific analytic and/or reporting
purposes.
**Column types**
Column types are columns with similar content falling under
similar column headers. Typical column types are _time
columns_ (with monthly or yearly data), _scenario
columns_ (with actual or plan data) and _variance
columns_ (ΔPL or ΔPY).
The following layout principles apply to all column types:
- **Width**: Columns belonging to a certain
column type should have an identical width. This column
width should not depend on the text length of the respective
column header.
- **Orientation**: Right-align columns with
numerical data. Left-align columns with non-numerical data
(e.g. texts or product names). _Column headers_ have
the same orientation as the rest of the column. Headers for
combined columns can be centered or even left-aligned to
increase legibility.
- **Vertical lines and gaps**: Vertical lines
separating different columns should be very light or even
omitted. Use white vertical lines or white vertical gaps to
mark the columns. In the following figures, different widths
of these white lines resp. gaps are being used to separate
and group columns. Separate columns of the same type by a
narrow gap (see gap B1 in the figure in section “Scenario columns” et seq.). Use a
wider gap to separate a group of similar columns from the
next group (see gap B2 in the figure in section “Row header columns” et seq.).
Additional layout principals depend on the _column types_
described below.
**Row header columns**
![Figure EX 1.2-1: Row header columns](img/ex-1.2-1.png)
Row header columns contain the header texts of the rows.
Often, these columns are positioned at the very left of
a table. In most cases, row header columns are much
wider than other column types.
Keep the texts of the row headers short by using
abbreviations or footnotes in order to omit too wide
tables.
Use a wider gap (see width B2 in the figure)
to separate the _row header column_ from columns
with numbers.
**Scenario columns**
![Figure EX 1.2-2: Scenario columns](img/ex-1.2-2.png)
_Scenario columns_ show data for scenarios
(e.g. previous year, plan, actual). Use the same width
for all scenario columns (depending on the number of
digits).
For the sequence of scenario columns see
UN 4.1 “[Unify scenario ana­lyses](09-unify.md#un-41-unify-scenario-analyses)”.
**Variance columns**
![Figure EX 1.2-3: Variance columns](img/ex-1.2-3.png)
_Variance columns_ show data of absolute variances (e.g. ΔPL, ΔPY) or relative variances (e.g. ΔPL%, ΔPY%).
**Time columns**
![Figure EX 1.2-4: Time columns](img/ex-1.2-4.png)
_Time columns_ show _time periods_ (for
flow measures) or _points of time_ (for stock
measures).
Use a temporal order from left to right for the
sequence of the time columns (e.g. Jan, Feb, Mar, or
2013, 2014, 2015).
**Measure columns**
![Figure EX 1.2-5: Measure columns](img/ex-1.2-5.png)
_Measure columns_ show measures
such as sales, headcount, or equity.
Displaying long measure names in column headers can be
challenging. As the column width should not depend on
the length of the measure name, use the abbreviations
defined in the glossary instead.
Use a wider gap after intermediate results to expose the
calculation scheme (see width B2 in the figure on the
left).
**Structure columns**
![Figure EX 1.2-6: Structure columns](img/ex-1.2-6.png)
_Structure columns_ show the elements of a structure dimension (e.g. countries or products).
**“Thereof” columns**
![Figure EX 1.2-7: “Thereof” columns](img/ex-1.2-7.png)
If details of an aggregated column are shown in one or
more column not totaling to the aggregated column, these
columns are called “thereof” columns.
The design of the _thereof columns_ should differ
from other columns. E.g. use a smaller font (see X in
the figure) to expose a _thereof
column_ and do not separate it from the mother
column (see columns _AL3_ and _AL3.1_ in
the figure) in order to show that it is part
of it. A _thereof column_ is positioned at the
right hand side of the mother column.
**Remainder columns**
![Figure EX 1.2-8: Remainder columns](img/ex-1.2-8.png)
If the set to be presented in the columns has too many
elements, accumulate the less important or smaller
elements in a _remainder_ column (e.g. 10 columns
for the top 10 countries and a remainder column titled
“Rest of world” or “RoW”).
In the figure, the _remainder column_ “Other cost” has the same vertical gaps B1 as the
other measure columns.
**“Percent of” columns**
![Figure EX 1.2-9: “Percent of” columns](img/ex-1.2-9.png)
Use “_Percent of”_ columns to present important
data of another column as shares of a given total. A
typical example for a “_percent of_” column is
data of a profit and loss statement as a percentage of
sales.
“Percent of” columns have a smaller font size (see X)
than the other columns.
**Totals columns**
![Figure EX 1.2-10: Totals columns](img/ex-1.2-10.png)
Position columns displaying _totals of a group of
columns_ (e.g. quarters totaling in years or
products totaling in product groups) at the right hand
side of the columns belonging to this group. The design
of the _totals columns_ should differ from other
columns, e.g. highlighted by bold fonts or by light gray
background.
The column types described before refer to
_single_ columns. The following paragraphs
present _combined_ columns i.e.
_hierarchical_ and _nested_ columns.
**Hierarchical columns**
![Figure EX 1.2-11: Hierarchical columns](img/ex-1.2-11.png)
Hierarchies in dimensions may call for columns showing
multiple levels. If possible, the sibling elements
belonging to the same parent element of a dimension
should be homogenous, mutually exclusive, and
collectively exhaustive.
Separate parents by appropriate means, e.g. wider gaps. Display the parent columns at the right hand side of their child _columns (like totals columns)._
In the figure, a wider gap B2
separates the two years (with four quarters each)
from each other.
**Nested columns**
![Figure EX 1.2-12: Nested columns](img/ex-1.2-12.png)
In _nested columns_, two column types are combined
in such a way that the columns of one type repeat
iteratively within every column of the other type.
Separate the resulting groups of columns by appropriate
means, e.g. wider gaps.
In the figure, wider gaps B2
separate the four years (with AC and PL data each)
from each other.
**Row types**
_Row types_ are rows with similar content falling under
similar row headers. Typical row types are _measure rows_
(e.g. sales, cost, profit) or _structure rows_ (e.g.
Italy, France, UK).
The following layout principles apply to all row types:
- **Height**: Rows belonging to a row type should
have an identical height (see height A in the figure in
section “measure rows” et seq.).
- **Horizontal lines**: Separating rows by light
horizontal lines will increase the legibility.
Additional layout principals depend on the row types described
below.
_Time periods and points of time_, _scenarios_,
and _variances_ should be displayed in columns rather
than in rows.
**Column header rows**
![Figure EX 1.2-13: Column header rows](img/ex-1.2-13.png)
Column header rows contain the header texts of the
columns. In most cases, these rows are positioned at the
very top of a table. In order to group columns two and
more column header rows might be necessary. If
necessary, abbreviate column header texts in order to
fit in the preferred column width. Alternatively keep
column headers short by using footnotes.
In the figure the _column header row_
uses two lines in order to fit the column header texts
in the narrow columns.
**Measure rows**
![Figure EX 1.2-14: Measure rows](img/ex-1.2-14.png)
_Measure rows_ show measures
such as sales, headcount, or equity.
Separate rows showing final or intermediate results of a
calculation scheme (_results rows_ _or totals
rows_) by solid lines. Display results rows in
bold font or highlight them with light gray background.
An additional gap B below a results row will increase
legibility.
**Structure rows**
![Figure EX 1.2-15: Structure rows](img/ex-1.2-15.png)
Structure rows show elements of a structure dimension (e.g. countries or products).
**“Thereof” rows**
![Figure EX 1.2-16: “Thereof” rows](img/ex-1.2-16.png)
If details of an aggregated row are shown in one or more
rows not totaling to the aggregated row, these rows are
called “thereof” rows. Place the aggregated
_above_ the “thereof” rows (in contrast to the
_totals_ _row_ positioned _below_
the rows of its group).
The design of the _thereof rows_ should differ
from other rows. E.g. in the figure, the
_thereof row_ is of smaller height, written in a
smaller font (see X), not separated by a horizontal
line, and has a right-aligned row header.
**Remainder rows**
![Figure EX 1.2-17: Remainder rows](img/ex-1.2-17.png)
If the structure dimension to be presented in the rows
outline has too many elements, accumulate the less
important or smaller elements in a _remainder row_ (e.g. 7 rows for the top 7 countries and a
remainder titled “Rest of world”).
Exclude remainder rows from some of the Structure analyses such as averaging, ranking, and
selecting.
In the figure, the _remainder row_ has
the same row height A as the other structure rows of
this table example.
**“Percent of” rows**
![Figure EX 1.2-18: “Percent of” rows](img/ex-1.2-18.png)
Use “_Percent of”_ rows to present important data
of another row as shares of a given total. A typical
example for a “_percent of_” row is gross profit
as a percentage of sales.
“Percent of” rows have a smaller font size (see X) than
the other rows.
**Totals rows**
![Figure EX 1.2-19: Totals rows](img/ex-1.2-19.png)
Place rows displaying _totals of a group of rows_ (e.g. countries totaling in regions or products
totaling in product groups) below the rows of this group
and separated them by solid lines.
The design of the _totals rows_ should differ from
other rows, e.g. highlighted by bold fonts or by light
gray background.
The row types described before refer to _single_
rows. The following paragraphs present _combined_
rows i.e. _hierarchical_ and _nested_
rows.
**Hierarchical rows**
![Figure EX 1.2-20: Hierarchical rows](img/ex-1.2-20.png)
Hierarchies in dimensions may call for rows showing
multiple levels. If possible, the sibling elements
belonging to the same parent element of a dimension
should be homogenous, mutually exclusive, and
collectively exhaustive.
Separate parents by appropriate means, e.g. wider gaps
(see additional gap B in the figure).
Display the parent rows _below_ their child
rows (like _totals rows_).
**Nested rows**
![Figure EX 1.2-21: Nested rows](img/ex-1.2-21.png)
In _nested rows_, two types of rows are combined
in such a way that the rows of one type repeat
iteratively within every row of the other row type.
Separate the resulting groups of rows by appropriate
means, e.g. wider gaps (see additional gap B in the
figure).
**Table types**
![Figure EX 1.2: Table types](img/ex-1.2.png)
Table types are distinguished by their analytic purpose in time series tables, variance tables and cross tables. Tables serving more than one analytic purpose are called combined tables.
**Time series tables**
![Figure EX 1.2-22: Time series tables](img/ex-1.2-22.png)
_Time series tables_ are used for time series analyses, combining time columns with measure rows or structure rows.
A typical example for a _time series table_ is a
sales analysis by countries (rows) and years (columns).
**Variance tables**
![Figure EX 1.2-23: Variance tables](img/ex-1.2-23.png)
_Variance tables_ are used for scenario analyses, combining scenario columns and variance columns with measure rows or structure rows.
A typical example for a _variance table_ is a
sales analysis by countries (rows) showing different
scenarios and different variances (columns).
**Cross tables**
![Figure EX 1.2-24: Cross tables](img/ex-1.2-24.png)
_Cross tables_ are used for Structure analyses, combining structure columns with structure rows.
A typical example of a _cross table_ is a sales
table with countries in rows and products in columns.
**Combined tables**
![Figure EX 1.2-25: Combined table 1](img/ex-1.2-25.png)
_Combined tables_ are used for multiple analyses. A combined table uses more than one _column type_ and/or more than one _row type_ presented either side by side or nested.
The first figure shows a hierarchical
structure of countries on three levels in the rows. The
columns are nested: scenarios and variances are the same
for both time periods _November_ and
_January_November_.
![Figure EX 1.2-26: Combined table 2](img/ex-1.2-26.png)
The second figure shows the measures of a
calculation scheme in the rows. The columns are nested:
The four quarters and the annual total are the same for
both years.
![Figure EX 1.2-27: Combined table 3](img/ex-1.2-27.png)
The third figure shows the same rows as the
second one (measures of a calculation scheme). The
nested columns now show PY and AC data as well as
absolute and relative variances for two markets.
## EX 2 Replace inappropriate chart types
Inappropriate charts make it hard to perceive the message. Knowing the correct
usage of chart types helps in replacing inappropriate visualizations, such as
pie charts, speedometer visualizations, radar charts, and spaghetti charts, with
those chart types better suited.
## EX 2.1 Replace pie and ring charts
![Figure EX 2.1: Replace pie and ring charts](img/ex-2.1.png)
_Pie_ and _ring charts_ are [circular charts](http://en.wikipedia.org/wiki/Circle) dividing some total into [sectors](http://en.wikipedia.org/wiki/Circular_sector) of relative proportion, but there are better ways to illustrate the numerical proportions of segments, e.g. bar charts or charts with stacked columns, see Figure EX 2.1.
_Pie charts_ allow for one-dimensional analyses only, and therefore seldom convey revealing insights. However, some useful applications for pie charts exist, for example when market sizes and/or market shares for one period need to be allocated to certain regions on a map (see CH 3.3 “[Avoid misleading colored areas in maps](07-check.md#ch-33-avoid-misleading-colored-areas-in-maps)”). As opposed to column or bar charts, pie charts can be positioned on a specific point on a map.
## EX 2.2 Replace gauges, speedometers
![Figure EX 2.2: Replace gauges, speedometers](img/ex-2.2.png)
Often found as part of a so-called dashboard, _speedometers_ are
probably one of the most useless visualizations out there. They take up
way too much space and have often confusing color coding and scaling. In
general, bar charts showing the respective structures or columns charts
showing the respective development over time are better choices, see
Figure EX 2.2.
## EX 2.3 Replace radar and funnel charts
![Figure EX 2.3: Replace radar and funnel charts](img/ex-2.3.png)
So-called _radar charts_ (also called _net charts_ or
_spider charts_) are frequently used for evaluating
purposes. Having no advantage over bar charts and having, actually, many
weaknesses, use them only for two-dimensional analyses (e.g. comparing
young-old with rich-poor). Willard C. Brinton wrote almost 100 years
ago: “This type of chart should
be banished to the scrap heap. Charts on rectangular ruling are easier
to draw and easier to understand.”
Of course, if the circular arrangement has meaning (such as the compass
direction), this kind of chart can be very valuable, but these types of
analysis are not typical in business reporting.
_Funnel charts_ are misleading when the size of the area displayed
does not correspond to the respective numerical values an issue
applying also to other artificial chart forms (e.g. spheres) in which
the length, area, or volume do not correspond to the numerical values.
## EX 2.4 Replace spaghetti charts
![Figure EX 2.4: Replace spaghetti charts](img/ex-2.4.png)
A chart with more than three or four intersecting lines (“spaghetti chart”) can be more confusing than several smaller charts with one line each placed next to one another (small multiples), particularly when evaluating the shape or the trend of the lines, see Figure EX 2.4.
However, when needing to compare exactly the height of data points of
several lines, spaghetti charts cannot be avoided.
## EX 2.5 Replace traffic lights
![Figure EX 2.5: Replace traffic lights](img/ex-2.5.png)
“Traffic lights” with green, red, and yellow areas are a popular form of
visualization but contain little information per area used. However,
they can be used for analyses showing “yes or no” decisions or
situations similar to real traffic lights. In all other cases replace
them with more suitable means of (analog) representation such as bar
charts, see Figure EX 2.5.
## EX 3 Replace inappropriate representations
From a perceptual perspective, avoid all visual representations requiring time
consuming analyses or additional explanations, particularly the popular use of
merely conceptual representations and all forms of texts, including bullet
lists.
## EX 3.1 Prefer quantitative representations
![Figure EX 3.1: Prefer quantitative representations](img/ex-3.1.png)
Due to the time constraints usually involved with presentations,
conceptual graphs prove less suitable than charts, photos, maps, etc.
For a one-hour presentation, do not use more than three or four
conceptual representations. Do this only if they are absolutely
essential for comprehension. The audience will understand charts and
pictures (photos, drawings, etc.) better and faster, see Figure EX
3.1.
## EX 3.2 Avoid text slides in presentations
![Figure EX 3.2: Avoid text slides in presentations](img/ex-3.2.png)
Avoid all forms of text slides in presentations. Texts should either be
recited or written in a handout. A few exceptions to this rule are
specific texts being discussed such as definitions, quotes, etc. In
general, all forms of lists (bullet points) should appear only in the
written handout, not projected on the screen. True, if someone sees and
hears something simultaneously, he remembers it better than when he just
hears it, but bear in mind texts are not considered something merely to
be seen they must be read and understood, see Figure EX 3.2.
## EX 4 Add comparisons
Visual perception is strongly based on setting one perceived object in relation
to another. Adding meaningful comparisons helps the eye evaluate faster, the
main purpose of charts.
## EX 4.1 Add scenarios
![Figure EX 4.1: Add scenarios](img/ex-4.1.png)
Scenarios such as “plan” and “previous year” are the most common references for comparison purposes. Add them whenever available. Use a standardized scenario notation for faster comprehension, see Figure EX 4.1.
## EX 4.2 Add variances
![Figure EX 4.2: Add variances](img/ex-4.2.png)
Having added scenarios for comparison purposes, the visualization of variances makes it easier to evaluate the situation. Use a standardized notation of variances for faster comprehension, see Figure EX 4.2.
## EX 5 Explain causes
Present data more understandable by showing interrelations, i.e. causes and dependencies. Seeing the entire context, especially extreme values and deviant values, helps to explain causes. Details increase not only the level of credibility but also comprehension. Use charts to prove, explain, and render something plausible, not to serve merely as decoration. This section focuses on the explanation of causes by using tree structures, clusters, and correlations. A more structured approach to increasing information density is discussed in the chapter “CONDENSE Increase information density”.
## EX 5.1 Show tree structures
![Figure EX 5.1: Show tree structures](img/ex-5.1.png)
The presentation of the assumptions or basic data upon which a business analysis is based, results not only in better understanding, but also makes it more convincing. A good choice is the display of calculated measures in a tree structure, see Figure EX 5.1 (see also CO 5.2 “[Show related charts on one page](06-condense.md#co-52-show-related-charts-on-one-page)”).
## EX 5.2 Show clusters
![Figure EX 5.2: Show clusters](img/ex-5.2.png)
With the help of clusters in two-dimensional and three-dimensional forms,
large amounts of data very often can provide interesting and new
insights, see Figure EX 5.2.
## EX 5.3 Show correlations
![Figure EX 5.3: Show correlations](img/ex-5.3.png)
When comparing several data series, correlations are often sought in
order to better understand the interrelations. Suitable rankings and
comparisons can facilitate the understanding of patterns,
see Figure EX 5.3.
[← Organize content](02-structure.md) | [Avoid Clutter →](05-simplify.md)