Multidimensional interfaces

Human Efficiency

Leonard Verhoef.
Contact

for

interface & web public transport In Dutch.

with

articles and columns

boek: GUI, webontwerp, psychologie en human efficiency. book: Why designers can't understand their users. consultancy. course: GUI, webdesign, psychology and human efficiency. scan: Why users don’t understand your interface.

e.g.

DRAFT Text

Last update
Oct. 19th 2009

Leonard Verhoef
CHI Nederland Avondlezing
21 oktober; 19:00 - 20:30 met afsluitend een borrel.
Hogeschool Utrecht
Oudenoord 700
Zaal E-1.16

Toegang: Uitverkocht

Gratis voor Chi leden en 10 euro voor niet leden.
Aanmelden activiteiten@chi-nederland.nl

abstract. 1 existing models. 2 multidimensional interfaces. 3 hierarchical structures. 4 vertical structures.

Abstract

Today’s computer interface design focuses on visual, realistic, metaphoric, hierarchical and sequential models (graphical user interface, desktop, menu, tabs and wizards). Mendeleyev changed the structure of knowledge about chemistry not using a visual, realistic and metaphoric model like a wizard steering in a pot on a fire but by using a non-visual, non-realistic and non-metaphoric abstract multidimensional model. Can this approach be applied to interface design?

Multidimensional requirements like mutually exclusive, ordinality, vertical (versus hierarchical) and abstract n-dimensional (versus realistic 3-dimensional) visualisation are analysed, validated empirically and validated in practice. To ensure a general level of the conclusions several types of interfaces of several domains are analysed (main menu structure, traffic signs, public transport route strips, language choice software and the desktop).

1 Existing models

In 1995 Rautenberg and Szabó presented 'a design concept for n-dimensional user interfaces.' They summarize perceptual and physiological knowlegde needed to present information on realistic 3-d interfaces (e.g. color blindness, depth cues, kinesthesis). The article is typical: multidimensional interfaces are implicitly interpreted as realistic 3-dimensional interfaces. They do not present interfaces based on these concepts.

To top.

Ruecker and Liepert (2006) are more specific and do not focus on realistic 3-dimensional interfaces. In “Taking Mendeleyev’s correspondence course: Interface design lessons from the periodic table of the elements.” they compare the way Mendeleyev related and visualized chemical data, to Rich-Prospect Browsing Interfaces.

An example is an interface for Victorian women writers. One or more values for dimensions such as the title of the book, the author, date of publication and geographic area are set, and then the interface filters and presents all cases meeting the criteria set.

An other example of a Rich-Prospect Browsing Interface is pill identification using form, colour, size, etc. Pill photos can be organized in a table combining two variables. The user can combine colour and shape. Any pill colour can go with any pill size, form and characters on the pill.

The focus of Ruecker and Liepert is on multidimensional characteristics of the domain. In this article we will focus on multidimensionality of the interface itself.

2 Multidimensional interfaces

Multidimensionality is simple. A table is a two dimensional structure. Reality is a three dimensional structure. Mendeleyev has shown that multidimensional models can unveil hidden complex structures.

To top.

He used piles of cards to describe the properties of chemical elements so he could arrange them in many independent ways. In the play card deck these dimensions are colours, forms and characters. The elements of these dimensions are mutually exclusive, ordinal and vertical (orthogonal).

2.1 Mutually exclusive elements of a dimension

Theory

One requirement for an efficient organisation of knowledge is that the elements are mutually exclusive. All users and designers should identify an element in the same way. In interfaces these elements can be the options of a menu, a directory and a list of destinations in a station.

To top.

Paap (2001) has written an interesting overview on the design for menus. One of the requirements for a good menu is ‘class-inclusion matching’ which is defined as: ‘… whether the target is an instance of the category specified by an option.’ Meeting these requirements prevents category decisions from being faulty if there is conceptual overlap between the categories, and menu targets might belong to two or more categories. Mandel (1997), refers to mutual exclusiveness for menus in plain language, by stating that options very often seem very similar and yet have very different meanings.

The advantage of being exclusive is that the user can be sure that the entry has the information he needs; he selects the right option.

Practice

In practice several types of interfaces for the same problem can be found as the table and figures show.

Table x, inclusive and exclusive interfaces

Inclusive

Exclusive

Motor way identification like M60 (UK) and A10 (ring road Amsterdam).

Motor way direction identification like clock wise and anti clock wise on ring roads.

Underground line identification like Bakerloo line London Underground (between Elephant & Castle and Wealdstone)

Underground end destination identification like Porte d'Orleans (going south) and Porte de Clignancourt (going north, same line).

Train identification like Eurostar and Thalys

Train end destination identification (Train to Paris, Train to Amsterdam)

Landmark naming in western cultures like the Alps

Landmark direction naming like in travel cultures: "German mountains" (as inuits in Italy would call the Alps) and "Italian mountains" (as inuits in Germany would call the Alps) (Aporta, 2004).

Figure 1, inclusive elements in a main menu

The options “file” and “edit” are inclusive. When the user wants to edit a file he cannot know which option to select.
(Source: Word Main menu, 2003)

Figure 2, exclusive elements in a main menu

The options “Do” (a list if verbs) and “with” (a list of nouns) are exclusive when the user wants to edit a file. In this case the user wants “to print a print”.
(Source: Verhoef)

Figure 3, inclusive elements in a way finding interface

The options “Ring A10” and “Ring A10” are not mutually exclusive when the user has to decide between left and right lane. The user cannot know which lane to select.
(Source: Real live, ring road Amsterdam exit IJburg, 2009)

To top.

Figure 4, exclusive elements in a way finding interface

The icons clockwise and anti-clockwise are exclusive when the user has to decide between left and right lane.
The user understands which lane to select.
(Source: The ring road entrance icons from the Holslag Verhoef ring road navigation icons set ©)

Empirical validation
To be elaborated.

Design validation
To be elaborated.

2.2 Ordinal dimension

Theory

The disadvantage of exclusive elements is that the user has to inspect all elements to find information. This is no problem when there are only a few options in the main menu or a few trains on the trains indicator. When, however, the number of elements in a list increase searching time will increase as well.

When the elements can be arranged in some quantitative way, this will improve navigation on the dimension. The user can predict the position of an element. Searching for Zulu in an alphabetic list is easy. The longer the list the more advantage an ordinal arrangement gives..

Ordinal arrangements in lists might be: the alphabet, time, steps to be taken, frequency or recency of use. Exclusive elements is a requirement for an ordinal arrangement.

Practice

Figure 5 and 7 show non-ordinal arrangements of elements in an interface.

Figure 5, non-ordinal arrangement in a main menu

Understanding the options of this menu, the user cannot estimate the position of his target option in this list.
The user cannot jump the option he needs.
(Source: Word main menu, 2003)

Figure 2 and 4 show interfaces for the same task, having an ordinal arrangement the user can understand.

Figure 6, ordinal arrangement in a two options main menu

The user can see that the order is alphabetical.
For printing the user might jump to a position at 3/4 of the list
(Source: Verhoef)

Figure 7, non-ordinal arrangement in a list of stations

Right list: for the passenger, not familiar with the geographics of this public transport line, this is a non-ordinal list.
The passenger cannot estimate the position of his destination in the list.
(Source: London Underground Stratford station, 1989)

To top.

Figure 8, ordinal arrangement in a list of stations

Left list: For the passenger not familiar with the geographics of this public transport line this is an ordinal list.
Knowing the alphabet the passenger cannot estimate the position of his destination in the list.
(Source: London Underground Stratford station, 1989)

Empirical validation, sign posts

Spijkers et al. (1985l) found that the best way to structure destinations on a sign post is indeed alphabetically. The table below summarises their findings. This table also defines the experimental conditions of Spijkers et al. in the values for structures as: 1) not-mutually inclusive/random, 2) mutually exclusive/nominal and 3) ordinal.

Comparing their experimental conditions is obscured because some experimental conditions included more ways of categorisation at the same time. The experimental conditions were not exclusive. Nevertheless the table confirms the conclusion that a random structure is less efficient than an ordinal (alphabetical) structure.

Table way-finding performance and values for cognitive structure



Structure	Correct		Values
random		49%	not nominal
random+categoric		79%	not nominal+nominal
direction +alphab.		62%	not nominal +ordinal
direction+categoric+alphab.		76%	not nominal+nominal +ordinal
categoric+alphab.		87%	nominal +ordinal
alphab.		91%	+ordinal

To top.

Empirical validation route strip

To be elaborated.

3 Hierarchical structures for dimensions

So far the requirements for one dimension were discussed. Systems are complex and one dimension might not be sufficient to represent the system.

More dimensions are a solution and it has to be determined how to arrange these dimensions. There are several options.

Theory

In information technology it is common practice to arrange information in a hierarchical way, e.g. a menu and a directory. These models are compatible with ancient linear feed-forward models of human cognition (Spivey, 2007). The inconvenience of a hierarchical structure might not be a problem for users when there are only a few options to select.

To top.

In complex systems navigation might become a problem. It should be noted that these traditional hierarchical structures might limit user’s performance because they are not compatible with human physiology. Human physiological structures are not hierarchical but parallel and network like. In nature escaping an attacking predictor by immediately climbing a tree might be more effective than following a hierarchical tree to find the best escape.

Multidimensional interfaces can be compatible with the continuous dynamics of human cognition. Figure 9 and 10 show examples of hierarchical and parallel interface solutions for selecting language on a ticket vending machine.

Practice

The Belgium train ticket vending machine has a hierarchical structure (see Figure 9). Selection of language is the first obligatory step.

Figure 9, hierarchical solution for language choice on a train ticket vending machine.

The first step is compulsory: select language.
(Source: Belgium touch screen train ticket vending machine)

To top.

To be compatible with the parallel information processing of humans Netherlands Railways train ticket vending machine does not have a hierarchical structure (see figure 10). The passenger can change any option chosen, at any time. There is no need to return to the main menu. Language can be changed any time.

Figure 10, parallel solution for language on a train ticket vending machine

Language can be changed any time.
(Source: Touch screen ticket vending machine Netherlands Railways)

Empirical validation

The theory is that the human information processing is not hierarchical but a parallel. This suggests incompatibility with hierarchical interfaces like menus. The consequences of this incompatibility can be tested with the language selection option on train ticket vending machines.

At Schiphol airport we observed 76 passengers changing language or giving up the operation. Not noticing the language switch button might be the reason. The default language of the machine is Dutch. The data are in the table below.

To top.

Most users first switch language within 3.4 seconds (80%).

8% of the users switch language after 4.3 seconds and after having made several selections in the Dutch mode. They search for a language button only when they get problems they can't solve. For those users the parallel interface is no problem.

12% of the users give up the operation for several reasons. The empirical data suggest that the probability of problems because of not noticing the language switch button is button is 0%.

These data suggests that a parallel solution can be used.

Figure x, the two train ticket vending machines observed for the language switch at Schiphol Airport

Language switch
n			Time to:
61	80%	3,4 sec.	language switch at first button pressed
6	8%	4,3 sec.	language switch after first button press
9	12%		language is not switched, no ticket is bought
76	100%
Language is not switched, no ticket is bought
n			No ticket because:
4	44%		"Just looking at the machine"
3	33%		"Machine does not have my payment method"
1	11%		"Machine does not have my ticket"
1	11%		Reason unknown
0	0%		Dutch nor English speaking
0	0%		Dutch not, English yes, language button not noticed
9	100%

To top.

Practical validation

A practical advantage of a hierarchical first language selection is that there is no possibility that the language switch is not noticed. The price to be paid is that all users have to make this step. In the Schiphol observations this step took 3. 4 seconds for experienced users. The mean time needed to select a ticket is 11.0 seconds (n=73, Verhoef, 2008). Purchasing time and number of ticket vending machines to install is 31% higher for the hierarchical language step only for these figures.

The observed probability of not noticing the language button is 0.0%. On the machine investigated there are several options to reduce the probability of not noticing the language switch more. In the interface observed the first step shows 50% information in conspicuous formats (large size, colourfully, dynamics) and information that is not needed for the first step (e.g. the current language button, the 'cancel' button, advertisements, conspicuous payment method information). In the design there are several options to increase conspicuity of the language option (e.g. larger format).

These practical considerations suggest that the parallel solution can be applied and is more efficient than the hierarchical solution.

To top.

4. Vertical structure for dimensions

Vertical or orthogonal means: an element can have any value of any combination of dimensions. When the data are put in a table, all cells of the table are filled. A playing card can have any colour, any character and any form; these dimensions are vertical.

To top.

The interface should allow for all theoretical combinations. In practice there might be no need to fill all cells. On a ticket vending machine there might be a fixed price to buy a ticket for a dog. No destination or class type might be required. In a computer interface it should be possible to perform all commands with all elements. In practice there might be no need to spell check the pictures.

4.1 Realistic 3-d visualisation

Theory

Realistic visualisations like desktop, menu, bars, and icons and concrete motor movements like pointing are common practice in interface design. The 3-d interfaces have the same characteristics. It is supposed that realism makes user friendly because the user is familiar with real life. This supposition is effective only when the task and structure of the real life metaphor is similar to the task and structure to be performed with the interface.

Physical laws such as gravity, size, form and a 3-d real life, impose restrictions to the realistic interface. The same applies for perceptual laws such as visual distance, depth perception and kinesthesis as Rautenberg and Szabó (1995) pointed out.

The motor and perceptual operations needed to manipulate the concrete objects also impose restrictions to the cognitive operations the user has to perform. There is a large difference between muscular pointing operations with concrete objects on a desktop and brain operations with abstract objects of a mental system. A simple example is that the motor activity of pointing to a concrete icon is less efficient in long lists than a more cognitive operation like typing a name that is in an alphabetic list.

To top.

Most theories on human cognitive development describe a sequence of motor, visual, language and mental operations (Piaget, Bruner, Gal'perin). When there is an understanding problem at a higher level a common solution is to descend to a lower cognitive level. When the mental operation of 2+1 is not understood by a child, one should switch to a lower level: "You have 2 cookies and you get another one" (language level), or lower: "Here you see 2 and 1 cookies)" (perceptual level) or even lower: "Here you have 2 and 1 cookies" (a concrete level that allows motor operations) (Van Eerde en Verhoef 1978). Giving children cookies only, to learn arithmetic, will not bring them to mental operations with numbers. This suggests that giving users realistic concrete interfaces like desktops, will not bring them to mental operations and understanding working with a complex and abstract system like today's software.

A final consideration is that from a design point of view new technology requires a new interface. The car that replaced the horse does not have a lead and a saddle.

The mind set of a realistic 3-dimensional interface hides the options of more dimensional interface solutions. In some cases multi-dimensional is equivalent to realistic 3-dimensional (e.g. Apple's new multi-dimensional desktop). What if the interface should be n-dimensional? Beyond moving cookies there is abstract mental arithmetic. What is beyond moving icons on a on a not-dimensional desktop?

Practice

Figure x, 4-d realistic desktop

The desktop is called a 4-d-desktop. It is at least a 2x4d desktop as each single desktop has two dimensions (x and y).
(Source: http://www.tinydl.com/software/50226-four-dimensional-4d-desktop-yodm-3d-v132.html)

To top.

Picture to be elaborated.

Figure x, an experimental non realistic dimensional desktop
(Source: ...)

Empirical validation

There are several options for designers of a desktop.

There are several interfaces for the input activities using a desktop. The most frequently used interface is pointing. Within pointing there are several interface solutions, e.g. activating an item clicking once or twice. Another type of input interface is entering letters. Within entering letters there are several solutions as well, e.g. how to solve the letter input problem when two options start with the same letter.

There are several interfaces for the visual appearance of the desktop. The most frequently used is icons with text. Other options are no words or no icons.


There are several interface concepts for communicating with computers. The most frequently used is a desktop, other metaphoric options are a dashboard and a control room.

There are several tasks users perform with the desktop. The most frequently performed task is probably finding an item. Finding an item in a set of user specified loactions is the focus of the semantic the semantic desktop (Nadeem, D., 2007). Other tasks are maintenance (clean up, installing organizing directories, backup, updating), monitoring (new mails, warning, error messages).

This combination of interface design options, tasks and interactions between them impairs the discussion on the efficiency of the desktop. Designing a desktop for empirical comparison includes taking many decisions. The result is that there are many ’What ifs’ in the discussion of the results. What ever a desktop is and how ever it has been designed, it is a two dimensional area and on this characteristic we will focus now.

4.2 N-d tabular visualisation Theory

To be elaborated..

Practice

Picture to be elaborated.

Figure x, non-vertical arrangement of icons on a desktop

The desktop has two dimensions: an x-axis and a y-axis. However, these dimensions are not used by the interface to arrange the icons.
(Source: ...)

Picture to be elaborated.

Figure x, vertical arrangement of icons on a desktop

The y-axis of this desktop is used for an alphabetic arrangement of the elements. Frequency of use might be another option.
The x-axis of this desktop is used for file type; programs left and data right. (Source: ...)

Figure x, non-vertical arrangement on a ticket vending machine

The visual and conceptual dimension on this ticket vending machine is not prominent.
The passenger might not know where to start.
For previous options he has to press the red arrow button (Terug).
(Source: Belgium touch screen train ticket vending machine, approx. 2004)

To top.

Figure x, a vertical arrangement on a ticket vending machine

The x-axis is used for the task-steps-dimensions.
The y-axis is used for the parameters of the task step dimension.
(Source: NS touch screen train ticket vending machine)

http://webdemo.ns.nl/webdemo.e2000/).

Empirical validation

To be elaborated.

4.3 N-d form visualisation
Theory

To be elaborated.

Practice

Picture to be elaborated.

Figure x, planning overview, no dimensions in form
(Source: Excel-like planning)

To top.


Picture to be elaborated.

Figure x, planning overview, form used for dimensions
(Source: Nootenboom Verhoef)

Empirical validation

To be elaborated.

References To be elaborated.

To top.

Contact

Leonard Verhoef.
+31 (30) - 231 44 97
Parkstraat 19
3581 PB Utrecht
Nederland

humanefficiency.nl
verhoef@humanefficiency.nl

Chamber of commerce, trade register, subscription number: 39057871.

	Practice Picture to be elaborated. Figure x, non-vertical arrangement of icons on a desktop The desktop has two dimensions: an x-axis and a y-axis. However, these dimensions are not used by the interface to arrange the icons. (Source: ...)		Picture to be elaborated. Figure x, vertical arrangement of icons on a desktop The y-axis of this desktop is used for an alphabetic arrangement of the elements. Frequency of use might be another option. The x-axis of this desktop is used for file type; programs left and data right. (Source: ...)
	Figure x, non-vertical arrangement on a ticket vending machine The visual and conceptual dimension on this ticket vending machine is not prominent. The passenger might not know where to start. For previous options he has to press the red arrow button (Terug). (Source: Belgium touch screen train ticket vending machine, approx. 2004) To top.		Figure x, a vertical arrangement on a ticket vending machine The x-axis is used for the task-steps-dimensions. The y-axis is used for the parameters of the task step dimension. (Source: NS touch screen train ticket vending machine) http://webdemo.ns.nl/webdemo.e2000/).