Loading..
Today is

Combining SPC with Geographic Representations

Click here to go home

In the last chapter, we used colour to represent population. In this chapter we use colour to show signals that occur at the most recent data point (which we will refer to as 'today'). In the first visualisation below, we use a simple mapping to start with - red is any signal over the mean and blue is any signal under the mean. The data span from January 2011 to December 2016 and 'today' is currently set to . You can move the date and one month at a time using these grey buttons. This allows you to see how the patterns look at different times, with different signal patterns, and how they develop. We have also added some buttons at the bottom that will persist as you scroll.

A gapped chart showing whether a signal has been triggered in a neighbourhood. Click on a neighbourhood to see its SPC chart.
The date is .

We can further increase detail by showing the type of signal that occurred 'today'.

A gapped chart showing the type of signal that has be triggered in a neighbourhood.
The date is .

When comparing the above two, the fist graphic remains more clear when rendered at a smaller size when compared to the icon version. This is demonstrated in the visualisations below. The colouring on the right is visible at the second smallest size, whereas the icons only remain clear until the third smallest size. However, neither are clear at the smallest size. This demonstrates how different methods of representing signals can be used depending on the size of the visualisation, something we will discuss later on when we wish to show many maps simultaniously.

A comparison of fill and icon styles at different sizes. While the icons convey more information, it's easier to interprit the filled version at smaller sizes.

We can also use a simple mark to represent signals. Here we use a horizontal line to signify no signal, and a tilted down or up line to signify a signal. We reenforce this with colour. One way in which we can take advantage of this approach is to remove the underlying map and show only the marks.

Here we represent each region purely as a line. We use a discreet mapping for angle - a small angle represents eight over the mean, whereas a large angle represents one over three SD.
The date is .

Rather than representing a discreet value with the angle, perhaps we can convey further information by mapping angle to the mean of the signal. This continuous mapping allows us to see some subtle differences between regions that share the same signal. We can use this mark to further convey information for all neighbourhoods. If there is no signal, then we can show the difference between the value of the latest data point and the process mean. This allows the opportunity to see unexpected patterns in the data. Although for any neighbourhood without a signal, this could be considered noise caused by random fluctuations.

Here angle represents the difference between the process mean and the mean of the signal. For neighbourhoods without a signal, the angle represents the difference between the value of the latest data point and the process mean.
The date is .

We can further enrich the chart by representing the number of data points that exceed an expected signal length as the thickness of the line. So the line appears thicker if, for example, we have a run of nine over the mean (8 + 1) or two over three standard deviations (1 + 1). This allows us to see signals that continue to exhibit persistent negative or positive behaviour, or have not yet been addressed.

The same chart as above but with the addition of representing the number of points exceeding expectation as line thickness.
The date is .

So far we have only visualised a signal that occurred 'today', but what if we want to show some historical aspect or trends in the data? We could show tiny simplified SPC charts in each square (known as sparklines), but these may be difficult to see. Instead we propose a 'trend grid' as an aggregated method of showing all of the signals in a chart. We divide the time points into five bins (columns). If a signal is detected in a bin then a small box is drawn in that column. Its row is determined by the type of signal. Signals under the mean are drawn in the bottom half in green, and signals above are drawn in the top half in red. The further away the signal is from the mean, the further away from the centre point it is drawn. If a bin contains more than one type of signal then we see multiple entries in one column.

A trend grid showing an aggregated view of all signals.
The date is .

Aligning signals vertically requires a lot of space and, therefore, leaves a lot of unused space. Another approach, rather than a grid, is to have two channels (positive and negative). We still bin the time points (our columns), but now we use the height of the channel to show the most 'severe' signal (the one furthest away from the mean), and use opacity to show when multiple signal types exist in a single column. Currently, each channel occupies half of a grid square, but we can reduce that - something that will become useful later on when we show multiple visualisations in the same grid square.

A trend grid showing an aggregated view of all signals.
The date is .

This concludes our look at representing signals, in the next chapter we will look at representing individual processes.