Evoviz

The DIKW hierarchy

The data, information, knowledge, and wisdom (DIKW) pyramid conceptualizes data-driven decision making as an incremental process of connecting and contextualizing data. The pyramid shape implies ‘reduction’ or ‘distillation’ up the hierarchy, but of what and how? Value or utility perhaps? It is a widely employed conceptualization, but representations vary in number and taxonomy of levels, or whether a continuum, and whether feedback is included or not. There is also little consensus for what components flow up the hierarchy and how.

DIKW is also conceptualized using other visual metaphors, including against axes of ‘context’ and ‘understanding’ and divided by time, as network, with feedback loops, and linkage to methods and behaviors.

DIKW diagrams provide different perspectives on the general process of creating knowledge and acting upon it from data, which is a multi-faceted undertaking that involves the connection and contextualization of data, generating knowledge. This process is underlain by research, ‘doing’, and reflection.

Frické (2009) criticizes the DIKW hierarchy as:

• Assuming data is ‘correct’ but all data are liable to error, uncertainty, and bias.
• ‘Information’ may be defined in many different ways.
• Theory is required for knowledge but is not part of the DIKW hierarchy.
• ‘Tacit’ information is derived from experience not data.

These criticisms are pertinent reminders of the danger of ‘trusting’ data, the limitations of data-driven solutions that ignore unmeasured or qualitative aspects of a system, and that all models are simplifications from particular perspectives. However, ‘theory’ may be included by simply enclosing the diagrams in a box as it applies to all aspects of the process.

Map/graphic communication model

The map communication model developed during and after the Second World War as a information-based conceptualization of map design in an age of paper maps. While originating in geography and before computers, it may also be applied to graphics in general.

The ‘graphic communication model’ is conceptualized as a linear process in which the aim is to efficiently communicate a purposeful ‘message’ to a user via a graphic, which acts as an ‘interface’. Success depends on:

1. A clearly defined purpose that is understood by the designer.
2. The availability and selection of appropriate data.
3. Appropriately encoding of data into a graphical form that communicates the message.
4. The user must be able to ‘read’ the graphic, i.e. differentiate and associate graphic elements.
5. The user infers meaning from the graphic.
6. If user receives the message efficiently, then the graphic has met its purpose, otherwise it has not.

There are many challenges to overcome and opportunities for things to go wrong. For example,

1. The purpose may be miss-interpreted by the designer.
2. Data may be limited or simply not exist in the required form, and varies in quality and cost requiring re-consideration of commissioner expectations.
3. The designer must have the knowledge and skills to graphically encode the message in an appropriate manner for its use.
4. The user must be able to ‘read’ the graphic, but users and use conditions vary. Accessibility need should have been considered in 3, but are not always. Alternative means of communication may be required for different users, devices, and contexts.
5. Users may infer different meanings from the same graphic because of circumstance, custom, or preconceived belief.

The map communication model emphasizes message communication through a static interface. It ignores the the questions of power asked by critical cartographers including,

1. What is its ‘real’ purpose – the stated purpose may not reflect actual purpose.
2. How was it created – what data was used and why? were stakeholders consulted?
3. What it shows – does it present a particular perspective that marginalizes groups or facts?

Today, many maps and graphics are published as interactive digital interfaces that allow the user to explore data, find patterns, and construct their own meaning. Constructed meaning and subsequent action are conditional on the information presented that may not always be the same when dynamically drawn, i.e. ‘personalized’ in some way.

Concepts in the more general DIKW model may be mapped to elements in the more specific graphic communication model, the most obvious being both are underpinned by ‘data’. The designer selects data that is encoded as graphic information that the user decodes and contextualizes to make a decision. The user may have no direct knowledge of the data that underlies the graphic; however, from the user’s perspective, the graphics are ‘data’ which they mentally connect using prior knowledge, intuition, and meta-information such as titles and legends.

Whereas, to the user, the graphics are ‘data’, from which they construct meaning conditioned by their own knowledge, which hopefully allows wise decisions to be made. My analogies between the DIKW hierarchy and the map communication and geovisualization cube models may seem a bit tenuous, and may indeed be, but the DIKW hierarchy does make you think about how understanding is ‘distilled’ from data.

Exploratory data analysis

Exploratory data analysis (EDA) is the application of graphical approaches to discover and understand pattern in data.

“The greatest value of a picture is when it forces us to notice what we never expected to see” (Tukey 1977).

EDA is both a ‘way of thinking’ and a practical approach to engaging with data that emphasizes iterative exploration, discovery, reflection, and understanding. In EDA, iteration refines understanding of the data, whereas in the ‘waterfall’ design model, which is similar to the map/graphic communication model, iteration refines graphic design. In reality, both iterations are required for effective visualization.

MacEachren and Kraak’s geovisualization cube

The map communication model was developed before computers enabled the publication of dynamic interactive maps and graphics, and does not directly address variation in users or use purpose, which are subsumed into the design process. MacEachren and Kraak’s (1997) geovisualization cube considers visualization along three axes, ‘user’, ‘interaction’, and ‘information’.

Geovisualisation purposes are located on the diagonal from expert high-interaction exploration of the unknown to the low-interaction/static presentation of the known to the public, with analysis and synthesis as intermediate steps. The cube provides a useful generalization of how purposes correlate with the axes dimensions, but the sequence does not recognize that the ‘public’ are increasingly engaging in ‘citizen science’, so may be acting as ‘experts’, and that static printing remains a useful tool in exploration, analysis, and synthesis for large datasets where screen size is limited. Try finding a route across wild terrain without being able to view a large area at the same time. The ‘user’ and ‘information’ axes may be better conceptualized as distributions, with exploration ‘more likely’ to be done by experts than the public, and unknown information ‘more likely’ to be interactively explored than statically presented.

The geovisualisation cube locates four ‘processes’ in three dimensions whereas DIKW locates four ‘products’ in two dimensions. Both share the dimension of ‘understanding’ and DIKW ‘connection’ may be mapped to the diagonal linking the general processes in the cube levels in the DIKW hierarchy and the methodological tools and behaviors associated with them. The processes may also be mapped to the steps in the graphic communication model as the designer explores available data, analyzes or otherwise processes it, and then synthesizes different data streams into a graphic presentation.

Data visualisation literacy

Data visualization literacy (Börner et al. 2019) proposes that in the information age, the ability to ‘read’ and ‘write’ data graphics is as important as being able to read and write text. Creating ‘good’ graphics requires an understanding of design concepts (equivalent to grammar and spelling) and practical experience (equivalent to reading and writing). Design is an iterative process for both the designer and stakeholders.

Dave’s conceptual model

DIKW, the map communication and waterfall models, the geovisualization cube, and data literacy, along with the criticisms of them, combined with personal experience, have structured how I think about data literacy, data, and its visualization from my perspective. My approach is similar to that advocated by data visualization literacy, but is from the perspective of the analyst rather than the system as a whole. In data visualization literacy, the analyst travels around the loop, whereas I take the analyst/designer’s perspective as they travel around the loop, having ‘conversations’ with the data and stakeholders.

My analyst-centric perspective is based on three principles:

1. To understand data you need some understanding of the system and how it was collected.
2. Graphic exploration is key to finding, understanding, analyzing, and presenting pattern within data.
3. Pattern in the data tells you about the system and the data.

Tasks are often conditioned by the need to meet a particular remit specified or negotiated with stakeholders who may well be paying you. My job is to meet the remit, or in my case, I have been lucky enough to have been able to create my own remit, allowing me to experiment with a wide range of data and purposes. In my conceptual model, understanding comes from direct experience of the world, ‘views’ onto data, and information from documentary sources. We take existing conceptual models of how the world works and adapt them to specific systems and use needs. Our conceptual models drive the selection of data and subsequent analysis, including visualization, but also structure how data is collected. Primary data may be tailored to need, whereas secondary data collection is underpinned by the conceptual model and needs of others. Data exploration, processing, and analysis is an iterative process that involves viewing data from multiple perspectives, identifying and critically evaluating observed patterns in relation to patterns predicted from conceptual models and knowledge of how the data was collected.

I believe that to understand data that describes a system, you need a reasonable understanding of how that system operates as a means of ‘ground-truthing’ to the real world. This is especially important when working alone, but what is ‘reasonable’? If you work on multiple systems, then you cannot be an expert in all of them, and if you are not an expert, how do you know if you know enough to make confident inferences? If you make strong claims based on flawed reasoning, then you are a victim of the Dunning-Kruger effect that predicts greater confidence from those with lower competence. The converse may also occur, where a lack of confidence in your own knowledge and judgment delays or prevents publication – a form of imposter syndrome. You are trapped in a dilemma, but it is better to be cautiously brave than a rabbit frozen by a headlight.

Direct experience of a system is beneficial, but not essential. Years of looking at digital data of the same place or thing can only take you so far. Data visualization is cerebral, whereas experience is both visceral and cerebral. It allows correspondence between data and experienced reality to be observed and assessed, facilitating improved understanding of the system and how representative the data is of it.

Now repeat the mantra:

References

Tukey, J. W. (1977). Exploratory data analysis. Addison-Wesley Publishing Company. https://consoleflare.com/blog/wp-content/uploads/2022/09/Exploratory-Data-Analysis-1977-John-Tukey.pdf