Click here to go home
So far, we have looked at signals, but information regarding the processes can affect how we should interpret the signal. For example, areas of low crime can be more susceptible to false signals, whereas signals in high-crime areas may require more attention. By displaying process information, we can show the number of processes in the data, as well as the variability. Below, each square shows a mini-SPC chart showing only the processes as rectangles, centred on the mean, and with the height of the rectangle sized by the control limits. This allows us to see the number of processes and amount of variability within a neighbourhood, but does not allow us to make any comparisons between neighbourhoods.
We can further enhance this figure by showing the mean of the processes, as well as adding a vertical line between processes to visually enforce the change.
To provide a relationship between processes, we can use a global measure of variance. So now we can show how much variation exists between the control limits and the mean. The disadvantage of this approach is that it places greater emphasis on neighbourhoods with low crime.
Another way in which we can make comparisons between neighbourhoods is to set each mini-SPC to have the same minimum and maximum values on the y-axis. Not only can we now compare the amount of variability between neighbourhoods, we can also see the amount of crime a neighbourhood has compared to its neighbours. In contrast to the visualisation above, this method now places greater emphasis on neighbourhoods with high crime.
Finally, rather than showing crime frequency, we can show crime rate (based on neighbourhood population). However, we believe this places too much emphasis on city centres, where census population counts are low but crime is high. Therefore, without daytime and nighttime population data, we believe crime frequency to be a more informative measure to visualise.
Now that we have established that we're going to look at crime frequency with a global y-axis, we can think about reintroducing NPU boundary data. One approach is to colour each process by its NPU, this way we can even remove the background and leave a faint outline.
The issue with the above approach is that it reenforces 'artificial' NPU boundaries, such that if some correlation exists between neighbourhoods, the colour may hinder the observation of such correlation if it exists across multiple NPUs. To mitigate this issue, we faintly colour the background by NPU. Using this method, we hypothesise that the processes still dominate the attention, but the NPU colouring acts as a passive indicator.
This concludes our investigation into processes. Our next step is to combine the signal and process information to create a rich overview for each neighbourhood.