The Individual in the Data — the Aspect of Personal Relevance in Designing Casual Data Visualisations

3.1 Personal Data

The first examples are of particular interest to our framework since they first of all encompass visualisations that are concerned with personal relevance, while at the same time the underlying data in itself is “personal”. Therefore, they serve as suitable examples for how the personal relevance of an information can be a guiding influence when designing the actual visualisation. Nonetheless, it has to be pointed out that obviously not every visualisation of personal data is focussing on establishing personal relevance. Many examples are simply visualisations of summaries of personal sensor data and activities and are not concerned with the dimensions of relevance as discussed above (e.g. a generic and basic Fitbit User Interface [16]). The following section will instead focus on a subset of data visualisations that explicitly center around personal experiences – not only in terms of the underlying data, but also in terms of visualisation design, intention, and narrative.

A good example for such research are the works by Alice Thudt, Dominikus Baur, Sheelagh Carpendale et al. on autobiographical visualisations [44], [45], [46]. This line of work focuses on visualising spatio-temporal trajectories (see for example figure 1), which are – in the case of “Visual Mementos: Reflecting Memories with Personal Data” – enriched with photos taken by the individual user or “data collector”. Thudt et al. elaborate, amongst other things, on this concept of visual mementos, which the authors describe as an analogy to physical mementos such as photo albums, travel journals, and souvenirs that all “support personal recall of past events and sharing of experiences” [45, p. 371]. One goal within this visualisation approach is to evoke familiarity with the event represented by the underlying data, for example by revealing “patterns that help the data owners recognize their own history in the visual mapping” [45, p. 371]. This line of work strongly resonates with the proposed concept of inducing relevance by visualising personal spatio-temporal events. The authors reflect that their concept can be a “step towards creating meaning with personal data and reintroducing richness and emotional resonance into our experiences with digital data collections” [45, p. 371].

Similar projects on visualising personal spatio-temporal behaviour for reminiscent exploration have also been conducted by for example Move-o-scope [35]. Move-o-scope is a web platform that allows users to upload their movement data and have it visualised as aggregated movement patterns that support, for example, reminiscence after a journey or the identification of patterns in everyday spatio-temporal behaviour. Otten et al. take a similar approach of aggregating movement data but visualise it with a focus on networks or clusters of spatio-temporal behaviour and the relations between places [36]. Their visualisation “Shifted Maps” supports the discovery of recurring patterns in long-term mobility traces and encourages the user to engage with forgotten places and connections [36].

Furthermore, a thorough literature analysis with additional examples on personal visualisation and personal visual analytics has been conducted by Huang [25]. Without being able to go into detail of the existing examples, they all shed light on the specific narrative of relevance that is at work here: personal data and self-tracking as an “interest in the self”. The concept of continuously and regularly gathering data about oneself, being it by hand, with the help of specialised gadgets, or by using a smartphone app, is most commonly referred to as the “quantified self movement”. It also includes the analysis of the data to produce statistics or visualisations that relate to regular habits, behaviours and feelings [31]. Often, such self-tracking devices and apps are used based on a conscious decision to monitor activities and behaviour in an effort of “self optimisation”. At the same time, the quantified self movement also includes the sharing of data with others, sometimes even publicly. This can be described as a “playful mode of self-surveillance” (ibid.) which introduces disciplinary monitoring on a seemingly voluntary level into formerly non-disciplinary (private) spaces [50]. While the availability of personal data is one of the aspects within a conceptual model of relevance in visualisation, the act – when seen in a critical light – of self-surveillance might also be seen as a bit uncanny. In this context, Foucault’s vocabulary and concept of “self-government” and power (as strategic games) allows us to understand such self-tracking practices not only as a “decipherment of the self by oneself” [18, p. 29] that is built on “self-reflection, self-knowledge, self-examination” (ibid.); Foucault also describes how the individual’s capacity for self-control is linked to power and “forms of political rule and economic exploitation” [28]. However, in Foucault’s understanding, power relations do not necessarily result in a removal of liberty, they can also lead to an ‘empowerment’ or ‘responsibilisation’ of subjects and force them to decision-making in fields of action as part of self-government (ibid.). This reading opens the conceptual space towards an understanding of self tracking as part of a system of power that – besides the threat of exploitation – also carries with it the potential of empowerment and decision-making.

Such critical reflections of self tracking and the gathering of personal data hopefully contribute to a conscious, cautious and informed handling of self tracking devices (or a refusal to use them). Additionally, the interest in collecting personal data exceeds the mere purpose of self improvement: personal data can likewise be used in the examination of spatial behaviour on a collective and broader scale, it can be related to artistic practices and interpretations, and can be intentionally translated into discursive artefacts that relate to such critical discourses [33].

However, such a focus on personal data and the self might make it harder to communicate events on a global scale or bring up events that do not fall within the actual spatio-temporal experience of a user or reader. Yet, in the context of consciously conceiving CDV as a means to prompt identification and inducing personal relevance, the visualisation of personal (spatial) data could allow users to identify their own “position” in space, support reflection of their experiences in the lived-in world and their everyday life. Thereby, the integration or juxtaposition of personal data in CDVs might help to establish a connection to a broader dataset that aggregates both scientific data, as well as personal data by other individuals. On this basis, it is possible to also emphasise the fact that every aggregated dataset is comprised of information on or provided by a number of individuals, that the data is based on a variety of examinations, represents a multitude of views or positions, and in itself is a result of interpretation. Lastly, by integrating personal data and perspectives we wish to lay the ground for emphasising the need for interpretative and critical engagement with visualisations, both on the production as well as the reception side.

3.2 Visualizing the Individual

As mentioned in the last section, most (non-personalised) data visualisations are built on data that is collected by or that concerns bigger groups of people. Thus, every individual is only represented in an aggregated form, for example in a statistical overview on unemployment (see figure 2) or in calculations of population densities in a certain spatial area. In these examples, a variety of individual information or data is combined into aggregated values that are then visualised, for example, in a colour scheme, pattern, or as symbols. This practice of aggregation can be contrasted with a set of visualisations that take an opposing route: to show the individual in the data.

Figure 2

Spatially aggregated visualisation of unemployed individuals [39].

The first example of such an approach that centres around the “individual in the data” also remains in the sphere of spatial data visualisations. In 2010, Dustin Cable created a dot map for the United States (U.S.), in which every dot represents one person and is color-coded by race and ethnicity (see figure 3) [7]. A similar map, but with an aggregation of 100 persons per dot, was produced by the New York Times (NYT) in 2015 [4]. In contrast to traditional choropleth maps (representing e.g. census data, see figure 4) which show the race and ethnicity with the biggest count in a spatial area, such “dot maps” show each individual and, thereby, represent a more thorough and inclusive perspective on the data. By that, in theory, each U.S. citizen (included in the data) would be able to find him- or herself in the map as a dot.

Figure 3

“The Racial Dot Map: One Dot Per Person for the entire U.S.” From left to right a) Low zoom level of the whole U.S. b) Mid zoom level on the area around N.Y.C. c) High zoom level of Manhattan N.Y.C. [7].

Figure 4

Largest Ancestry: 2000 Map [6].

The next two examples are visualisations on gun deaths in the U.S. Similar to the first two cases, these visualisations use one discrete visual entity for each and every single death instead of an aggregated visual entity that represents many (like for example the BBC graphic see figure 5). The company FiveThirtyEight created a visualisation in which each block of two times two pixels represents one gun death in the U.S (see figure 6). The data visualisation firm Periscopic visualised every death that occurred in one year as one line. The latter visualisation goes a step further and not only visualises the number of deaths, but also interpolates the possible lifespan of each victim. Thereby, the visualisation projects – as the project calls it – the “stolen years” (see figure 7). The general approach of representing each individual in the data does not only support direct identification, it also helps to give the audience a better sense for large numbers.

Figure 5

Aggregated visualisation of gun deaths in the U.S. [3].

Figure 6

Visualisation of gun deaths in the U.S. by FiveThirtyEight. One pixel square represents one victim [21].

Figure 7

Visualisation of gun deaths in the U.S. by Periscopic. One line represents one victim [37].

A related design technique that aims at making complex measures easier to grasp is centred around the idea that the illustration of numbers such as monetary figures can be supported by the inclusion of familiar objects from the real world, for example by utilising magnitudes and units that are visually and kinesthetically experienceable [12]. Jacob Harris reflects on such visualisation techniques in a blogpost and also describes different approaches that aim at “humanising” data, all of which share the same motivation: to evoke empathy in the readers. One particular example is summarised as the Wee People [24]. This design technique is closely related to the news value of identification, in that it does represent each individual with an abstracted and minimised human figure, instead of “only” a dot, line, or other “non-human” visual metaphor. On the downside, as Harris points out, such a visualisation becomes hard to read once it surpasses a certain threshold of data points, or when the goal is to visually compare two amounts over time [24].

Figure 8

Illustrations on the difference between the U.S. and Great Britain by the Isotype Institute [17].

The inclusion of human figures in data visualisations or graphics is not a new approach, whatsoever. Most famously, Marie Neurath, Otto Neurath, Gerd Arntz, and their team established the “Vienna Method of Pictorial Statistics” in the 1920ies and 1930ies, also known as “Isotype” (International System Of Typographic Picture Education). They not only theorised how to visually communicate abstract phenomena to the public, but also applied their Isotype system in public exhibition projects. The language of pictograms which were developed by Neurath and their colleagues were used to explain data on societal structures and developments to the public. In their visualisations, each pictogram represents a certain count of units (see figure 8). It is therefore an intermediate path between completely aggregated visualisations and the previously named examples where the focus lied on revealing the individual. Neurath and their team used visual metaphors that resemble the underlying entity to support a more intuitive understanding of the represented units. This pictorial visualisation technique is still popular today as it allows to create relatable illustrations in contrast to more abstract information visualisation techniques like bar charts. The New York Times used the technique for example to illustrate casualties in the iraq conflict (See figure 9 left) [1], as did Neil Halloran in his video on the fallen of world war II (see figure 9 right) [22].

Figure 9

Left: Visualisation by the New York Times illustrating 31 days of Iraq [1]; right: Still from “the fallen of world war II” video [22].

Boy et al. have recently explored this spectrum of visualizing the individual in the data in casual contexts [5]. In their study they developed a model that differentiates between abstract representations of individuals in the data (e.g. dots or lines), abstracted icons, and realistic shapes of human beings. While we used relevance as our conceptual model to discuss the design and perception of visualisations in casual contexts, Boy et al. explore how the design spectrum (abstract shapes, abstracted icons, realistic shapes) affects the level of empathy in the visualisation’s audience. The concept of empathy has recently attracted the interest of the visualisation community. It was discussed, amongst others, by the “Responsible Data Forum on Visualizations” in early 2016 [51] and the conversation was then continued by Jon Schwabish in his PolicyVis Podcast, where he debated the topic with Kim Rees and Mushon Zer-Aviv [41].

While relevance focuses mostly on the guidance of personal attention and interest, the concept of empathy extends this and can be regarded as a precursor to action. High relevance could, thus, be a first step and an influential factor in later creating the feeling of empathy and it seems useful to integrate empathy into our conceptual model of personal relevance. Following this thought, it is plausible to suggest that it might be easier to associate with an individual that is more similar to us (e.g. on a level of spatial, social, or political distance) both when addressing personal relevance as well as in the creation of empathic encounters.

4.1 Taking Perspective Through Multidimensional Scaling

In a n-dimensional data space, each dimension can represent a certain attribute. When thinking about CDV and the question of relevance, our first aspect within the design model is to allow the individual reader to relate to some of these data attributes. This might include a relation by religion, political affiliation, ethnicity, residence, occupation, language, or more trivial aspects such as taste in music or choice of hobbies. By allowing an individual to relate themselves to at least some of the n-dimensions of attributes, they might be encouraged to mentally position themselves within a data space. This act of “mental self-positioning” is similar to multidimensional scaling in the sense that the individual searches for a location in the n-dimensional space that is characterised by data attributes with the greatest similarity to themselves (see figure 10). As an example for such a space metaphor, we would like to go back to Dustin Cable’s dot map [7]. The dot map’s data space reduces the complexity of citizenship and identity to two dimensions: ethnicity and space (e.g. residence). On the spatial dimension, individuals can mentally position themselves, for example, in an area that is similar to their residence and thus evokes “closeness”. Or they can mentally position themselves in an area that they feel closely connected to on an emotional level. The same is true on the second dimension, the ethnicity. By combining the two dimensions and relating them to their own identity attributes, individuals can establish a relation to the displayed data. In the case of the dot map, this might be the identification of the area where one lives and a comparison of one’s own ethnicity with the other represented ethnicities in the same area. Or one could relate to other areas where there is an agglomeration of people that identify with the same ethnicity as oneself and thus establish a connection and feeling of “closeness”. Through both dimensions, a connection to each individual can be established. When applying this approach to other dimensions of data, the recipient needs to be able to take on an individual perspective on the overall data and mentally position themselves within this data space.

Figure 10

Reduction of dimensions through multidimensional scaling, the user herself represented in blue other data points are represented in grayscale, the brighter the farther away from the user.

Figure 11

Left: applying KNN with K=1 and K=4; right: applying K-Means clustering with K=3.

To extend the statistical metaphor, we could assume that in some cases certain attributes might be unknown or uncertain from the user’s perspective. Through a k-nearest-neighbour (KNN) approach, a visualisation or rather algorithm could then suggest certain attributes (see figure 11 left) that is based on similar items. As an example, a KNN approach could calculate a likely political orientation by having a user position themselves within a selection of opinions on certain topics. If an overwhelming number of items is present in a data set, using k-means clustering could group items. Thereby, a subset of items is created that shares some similarity with the individual perspective and can therefore serve as a selection which the user can identify herself with (see figure 12). Such a personal perspective can then serve as an entry point to the overall data space.

Figure 12

Left: applying KNN with K=1 and K=4; right: applying K-Means clustering with K=3.

4.2 Overcoming Limitations by Overview and Zoom / Details

The concept described above presents us with an approach that aims at highlighting the personal relevance of a dataset to an individual, or, to be more precise, allow the individual to establish a personal relation to the data at hand. At the same time, these approaches distort the perspective on the overall data space. By encouraging the individual who interacts with such a visualisation to take on a distinct and personal perspective onto the data space, their focus might be limited to reflecting their own relation to the data which could then lead to a strongly biased perspective and render the global perspective or non-personalised significance invisible. When thinking of ways to overcome this, we can keep in mind the climate change communication literature discussed above. Climate change communication tries to increase relevance by introducing local perspectives, which are then reconnected to the global perspective in order to incorporate the local into the big picture. Turning Ben Shneiderman’s “visualisation mantra” (overview first, zoom and filter, details-on-demand) [42] on it’s head, a detailed and personal perspective could serve as an entry point that triggers personal association and supports personal relevance. In a second step, zoom and filter interactions could help to understand the “depth” of the data space and the associated topics by also introducing perspectives that go beyond the individual’s mental position. This intermediate stage might, for example, include a multitude of individual positions within the data space and thus lead over to more aggregated views. The third step can then introduce an overview on the global perspective and highlight the non-personalised significance of the topic at hand. Preferably, when applied in a visualisation, these three stages would support continuous shifts of perspective.

4.3 Visualising the Individual and Using Less Abstract Visualisations to Increase Identification

While the previous design techniques are primarily focussed on the data, the last technique focuses on the visual representation of the data itself. The related work has presented us with two connected approaches to visualise data in a more relatable way: visualising individual data points with abstract shapes or using visual metaphors that resemble the actual entity (e.g. humans or familiar objects from the physical world). The main idea suggests that, instead of showing aggregated visualisations, the individual data point within a display of data has a “higher resolution” and provides more level of detail and singularity, which then allows the recipient to relate to the individual cases behind the data points. Building upon this general idea, the design spectrum laid out by Boy et al. (see figure 13) suggests that less abstract visualisations might in turn lead to increased identification and relevance [5]. Such an approach that aims at humanising abstract data was also illustrated by the NYT’s icon visualisations that introduces a personal perspective on the casualties in the Iraq war [1].

Figure 13

Figure adapted from [5] illustrating the design spectrum from generic to unique.

4.4 Challenges of Abstract Data

With the illustration of those three approaches within a design space we wish to transfer the theoretical concept of personal relevance in CDVs into an applicable concept. The implications in the design space are based on the conceptual model that we introduced in the first part of the paper and influenced by our reading of the related work. Most of the related work that we used to discuss some general ideas and concepts are either based on personal data (collected by individuals) or they are centred around human-related data. In both cases, the user’s connection to the underlying data is very obvious, as the data either portrays their own data or another individual’s data. The next section now illustrates how our concept of personal relevance as a design space for visualisations could also be applied to an abstract dataset that has no direct connection to either personal data or human-related data. Or, to be more precise, how we can establish a personal connection to an abstract dataset by, for example, introducing factors or data of personal relevance.

4.5 The Individual in the Data: Relevance and Personal Perspectives

With the following use case we illustrate how the design techniques described in the last section can be combined and applied to an actual dataset. We will particularly exemplify how this design space also allows us to address personal relevance with a dataset that is not directly connected to human identities, attributes, or traits. The basis for this visual exploration is air pollution data from the city of London [30]. Traditionally, visualisations of air pollution either represent aggregated overviews or static maps of air pollution. We chose this example since it is quite abstract and, similar to the example of climate change, only affects a subset of the population directly, at least to such a degree that it is perceivable. A majority of the population is only indirectly affected, even though everyone obviously contributes to the problem at hand.

The first visualisation bears some resemblance to the traditional pollution maps (see figure 14). By adding multiple interaction elements, we then enable the user to locate themselves in the data or, rather, on the map. A locator plugin allows the user to identify and zoom onto their current position. In a similar way, a georeferencing plugin allows the user to search for an address (e.g. home, a friend’s house, work, a neighbourhood considered for relocation). Furthermore, prominent and well known locations are added as navigational cues. As previously discussed, this example highlights a typical problem in communicating pollution and climate change: on the local level, some people might not seem directly affected at first sight. A person living in a minor road near to a public park might be under the impression that air pollution is not a problem for them, thereby ignoring the overall impact on the whole city.

Figure 14

Interactive map for exploring air pollution in London, through navigational cues (upper right), current location plugin (left), georeferencing plugin (right) or slippy map interaction.

In order to overcome this spatial bias, we explore the temporal dimension in two additional visualisations which both focus on the bigger picture. The goal of both visualisations was to allow users to relate their own activities on a temporal dimension to the air pollution in the whole city of London. Instead of only displaying spatio-temporal difference between weekdays and weekends, the first visualisation (see figure 15) goes a step further and emphasises the individual’s connection to air pollution. It therefore connects a time use survey with the pollution data. Time use surveys are conducted via diaries in which the participants record their activities in certain time intervals. The UK time survey uses a 10-minutes interval, resulting in 144 activity descriptions per day [11]. Weekday and weekends show very distinct patterns. By separating these two categories and then correlating the named activities with the pollution data, we are able to highlight the impact of traveling (e.g. commuting) on air pollution. We therefore selected an exemplary roadside sensor from the city of London and combined it with the time use survey (see figure 15). Thereby, we wish to encourage people to reflect their own activities throughout a day to either “find themselves” in the data, or assume a different perspective “through the eyes of others”. We acknowledge that this comparison is difficult on many levels. First, the time use survey data is not restricted to London but represents data from the whole of the UK. Second, air pollution caused by motorised travel is a strong contributor to urban air pollution, but not the only one. After analysing the overall temporal patterns of all sensors in London we can say that the general pattern of two peaks on weekdays during rush hour in the morning and afternoon, as well as a slowly rising and descending curve on the weekend applies to most locations, beside the inner city. The inner city usually shows high traffic throughout the day, which is only higher on weekdays.

Figure 15

Comparing time use survey data (bottom) with a focus on travel activities (dark purple) and air pollution (NO2) data (top) for a sensor in Putney High Street.

Figure 16

Visualisation of commutes in London. On the left we see a video with three perspectives, on the right the map shows the current location, the overall route and the pollution distribution indicated as a heat map. On the bottom we see a timeline visualisation of the pollution exposure. The location shown in this example is the area of Stockwell in south London. The colour scale for the level of pollution is shown in figure 14.

In an attempt to increase the factor of identification even more, we developed a novel visualisation technique that combines video with a geospatial and a timeline visualisation (see figure 16). We recorded videos of bike commutes from several parts of London into central London. The GPS trajectories of those commutes were combined with the (almost) real-time pollution maps of London. Thereby we were able to create a timeline of pollution exposure for the cyclist. In this case we decided to only use a generic visual indicator for the individual moving through the city: a dot on the map that shows the current position. In this case, the humanising factor is integrated into the visualisation by synchronising the dot that indicates the position with the video images which, for each position on the route, allows the user to view the surrounding from the cyclist’s point of view. As a result, this visualisation gives a very vivid example of the exposure of air pollution on a person moving through the city. It furthermore illustrates the traffic impact on air pollution. A good example is the intersection at Elephant and Castle, which usually exhibits high congestions during rush hour. The videos clearly show the dense traffic, while the map and timeline correspondingly show the high pollution levels (see figure 17).

Figure 17

Same partition as in figure 16, this time for the location of Elephant and Castle in south London. The color scale is shown in figure 14.

Through the examples above we illustrated how some aspects laid out in the design space for personal relevance could be applied to a very abstract and non-personal dataset. We did so by exploring ways that might allow the users to find themselves in the data or identify with aspects of it, especially through the factors of space, time, and daily routines. In the process of experimenting with ideas, we realised that such techniques can only be applied up to a certain level of complexity (see for example figure 18, which presents the recipient with a complex cluster of overlapping levels of information).

Through the discussion of these examples we identified additional perspectives and potential approaches that are based on the design space. One of these approaches could, for example, include using k-means to cluster the time use survey data based on activity patterns which would allow users to pick a group they can identify with. Another solution would be to use less abstract values. As an example, instead of only using abstract pollution scales, information on the actual impact on individuals could be included in the visualisation. This could include health related information, such as the impact of air pollution on persons with heart diseases or asthma, or information related to air quality and its influence on sporting performance or general well-being.

Figure 18

A more complex visualisation of all air pollution measured by sensors in London, organized in an abstracted spatial grid (the blue line indicates the river Thames). Each sensor is visualised through a 24-hour spider graph. When selecting a sensor, the respective location and the surroundings are visualised in a miniature map on the lower right.