Open Data can help us work out how much London Fire Fighters spend on defenseless kittens

Wednesday, 26 April 2017

Open Data can help us work out how much London Fire Fighters spend on defenseless kittens

This second post leads on from Smart Cities: Pushing Open Data with Power BI. We will analyse the problem and look at the steps taken by the London Fire Fighters after seeing the results to try to tackle the situation.


We are now looking towards what information we want to extract from the data to give us the answers that we are seeking. There are some questions which are clear from the start of the analysis, however, there are others that arise as the data gives us more information.

The problem: The increase in the number of interventions by the fire fighting team has been increasing (along with the associated cost) which is what raised the alarm and led us to this analysis.

Figure 1: Campaign image from 2016.
We will start by proposing a hypotheses that we can try to contrast with the data. With the conclusions that we reach, we will search for strategies and define initiatives that will allow us to come up with solutions for the problem or to minimize its effects.

  • Hypothesis 1: The number of services continues to increase each year. If there is no corrective measure put in place, the cost of the service will continue to increase.
  • Hypothesis 2: The type of animal that is involved in the incident is firstly of importance when determining whether it is necessary for the fire fighters to be deployed. The location of the incident (rural or urban) can also be related with the type of animal.
The first step for us is to look at the data. This data can then be loaded as a table to allow us to see the names of the fields (some of which will be descriptive) and apply some filters. Our previous in-depth study has helped us decide which fields will give each report more relevant information. 

*With a more complex analysis, we would be at the selection phase for what attributes would give us a greater information gain, allowing us to segment the data in a more effective way. For example, determining what attributes allow us to group and predict the values of the "IncidentNominalCost".

To work on the first hypothesis we opted for the "Line Chart". We selected the sum of the CallYear field and dragged the mouse down to "Axis" so that we could display this value as part of the vertical axis and then "PumpCount" was dragged under the "Values" section to be part of the horizontal axis. If we simply select the fields from the list, they will be added in an order that we have not picked, however if we drag them directly to their final position it makes our results much more straight forward.

Captura campos
Figure 2: The CallYear field should appear below the "Axis" tag, whilst the "PumpCount" field should appear under the "Values" tag.

This lets us access the first dashboard from the report, which allows us to assess the evolution in the number of rescue services deployed per year.

Figure 3: The evolution of the number of emergency cases per year.

To give a more efficient analysis for the graphic, various filters give us extra help. Given that we only have data from the first quarter of 2017, we have applied a filter to shorten the year.

Figure 4: Example of an advanced filter. This one only shows years prior to 2016.

By applying this filter, the new graphic would be as follows:

Figure 5: The evolution of the number of emergency cases per year (filtered view).

We can also present the data in different ways, The "Funnel" feature makes it easier to assess the value of the total number of services carried out yearly. The ability to change between both representations is easy due to the dashboard function. Power BI does all the work for us, making it a smooth process.

Figure 6: Graph example of the "Funnel" function.

We can clearly see an increase in action between 2009 and 2011. This increase was then followed by a decrease which we will soon explain, and finally followed by an increase again in 2016. The sample is not that extensive however, but we can notice a trend for the gradual increase in the number of cases.

The yearly cost for the service is not exactly proportional, as the cost stems from the number of dedicated hours and can vary with each case. It is clear from the graph that 2011 saw the same trend in the evolution of cost as 2015.

Figure 7: The evolution of the yearly cost

To work on the second hypothesis we went for the "Pie Chart" visualisation. Start off by selecting the "CallYear" field and drag this under the "Axis" label so that this value is on the vertical axis. "PumpCount" is then dragged below the "Values" label so that it appears on the horizontal axis.

Figure 8: Number of services according to the type of animal

From the Pie Chart, we can clearly see that most incidents occurred with small animals. By taking a closer look at each segment we can see the concrete evidence alongside the total percentage (49.61% cats, 17.91% birds and 17.79 dogs).

If we translate this information to the cost and return to the "Clustered Column Chart" visualisation, we can see that the public spending dedicated on saving cats from 2009-2016 was 866,834 GBP! From that astonishing amount, 115,404 specifically comes from 2016.

Figure 9: The cost per animal in 2016

The period of 2009 to 2016 also seen 307,418 GBP dedicated to rescuing birds and 11,084 GBP to saving squirrels. The time period of 2009-2010 only saw the firefighters face 34 cases involving this small group of rodents.

The "Animal Group Parent" field has given us a great deal of information relating to the distribution of warnings. We will complete and broaden this analysis with a geographical layer.

To analyse the geographical distribution, we chose the "Map" visualisation. First, selecting the "Borough" field and dragging it under the "Location" label, the "AnimalGroupParent" field under the "Legend" label and the PumpCount field under the "Size" field.

Through this, we can see the distribution of warnings related to cats and dogs is fairly evenly distributed in more urban zones:

Figure 10: Geographical distribution of the warnings for cat (yellow) and dog (orange) rescue.

Regarding other types of animals of bigger size such as cows, bulls and deer, the cases are more spread out and related to rural areas, as was expected. The cases involving larger animals see a more vital role for the firemen to help resolve the situation.

Figure 11: Geographic distribution of warnings related to the rescue of larger animals like bulls (red), cows (grey), and deer (blue).

For the geographic distribution as we have previously mentioned the parameters of the "AnimalGroupParent" has given us the best information when deciphering the data and nearly always prioritises the services.

If any information appears outside of the given fields, it could be due to duplicates or mistakes in their name or postcode. City names can also coincide with others on the other side of the Atlantic. In any of the above cases, we can easily exclude any mistakes from the graph.

Figure 12: Examples of data that have clearly been duplicated due to name or postal code.

PowerBI also allows us to visualise segmentation. We can segment the data from the report with a concrete value, for example by year or geographic location. As an example, we will segment the number of services carried out in 2016 related to small animals (dogs, cats, birds, hedgehog, hamsters, squirrel, and ducks).

To carry out this segmentation we used the "Slicer" visualisation. We then selected the "Borough" field and then the rest of the canvas panels automatically offer the corresponding information which is linked to the district.

Figure 13: A view of the canvas with two different visualisations (table and funnel) under the Borough "City of London"

We can have yet to examine the data further with PowerBI and develop a text analysis included in the "FinalDescription" field. Thanks to the native integration with R, PowerBI can help us to group and analyse with better detail the information that has been previously mentioned by the RSPCA and rectify occurrences like the "Trapped" or "Stuck" fields.

Conclusions: Taking Action

The previous analysis has allowed us to confirm and reformulate the first hypothesis as: 

"If measures are not taken, the inadequate consumption of public resources in this type of service will continue to increase year by year"

Following this statement we can also add that:

  • The people of London are definitely animal lovers (the data that we have analysed echoes this statement)
  • The good citizen who calls the firemen to help a pet doesn't pay the cost from their pocket...or do they?
It seems obvious to state that the citizen who raises the alarm is NOT aware of the incurred cost in terms of actually carrying out the action. They do not take into account the waste of public money and even worse the bad use of an emergency service. Calling the firemen to rescue a hamster, a trapped parrot or helping a cat get down safely from a rooftop is not an effective use of public money.

A public awareness campaign was proposed to combat this problem and result in a more efficient use of public resources and better service to citizens.

2012: The Campaign

July 2012 seen the London Fire Brigade launch the campaign:

Figure 14: 2012 London Fire Brigade Campaign: "I am an animal, get me out of here".

The main aim of the campaign was to educate citizens that they could continue to be "good citizens" and help animals in dangerous situations without wasting public resources.

The campaign had two axes:
  • On one hand making the people aware of the cost of these services to firemen.
  • The ad then aimed to show people the viable alternative to using the fire brigade; RSPCA (Royal Society for the Prevention of Cruelty to Animals).
The positive effects of this campaign were seen immediately as there was a reflection in the number of calls registered in 2012.

Figure 15: A news piece regarding the decrease in call outs to rescue animals.

However, 2015 saw the rise of the trend again which was tackled through social media campaigns by the firemen.

Figure 16: 2012 Fire Brigade Campaign in London

Figure 17: BBC Campaign.

February 2017 saw an interactive map become published:

Figure 18: Interactive Map

It became clear that these campaigns need to be periodically circulated to ensure they do not lose their effectiveness.

Final Conclusion

The greater availability of open data in terms of public services combined with the various tools that allow us to analyse patterns and visualize said data can quickly translate into cost saving and satisfied citizens who are involved with their surroundings.

This has been a simple yet effective example with tangible results. Using the previously collected data for 2016 we could have estimated an approximate 215,160 GBP spent on episodes related to domestic or small animals.

If we could harness the potential with all data that businesses and public institutions store we could apply Data Science to allow us to improve our lives and public spaces. We should start to benefit from the use of data! 

1 comment:

  1. Current Web technology allows governments to share with the public a variety of information in unlimited quantities on demand. Technology is also available to allow citizens to bring issues of concern to the attention of local, regional and national governments. However, exploiting these capabilities within government systems is a challenge that encompasses environmental, policy, legal, and cultural issues. Establishing effective eGovernment requires openness, transparency, collaboration and skill in taking advantage of the capabilities of the World Wide Web. The rich potential for two-way dialogue between citizens and government creates a need for global leadership. The W3C has an opportunity to provide guidance in support of eGovernment objectives by promoting existing open Web standards and noting the challenges external to the Web and technology. There is also role for the W3C to facilitate the development and vetting of new open Web standards needed by governments in context.
    custom essay writing service