A team from the University of Auckland had a unique opportunity to investigate historical and anonymised bank transaction data to see if they could identify spending patterns in a community following a disease outbreak.
Every day we process millions of transactions and that creates a massive trail of data. This data has typically been used for things like customer segmentation or shopper location – but rarely for a Government outcome.
Professor Shaun Hendy, Director of Te Pūnaha Matatini (a Centre of Research Excellence hosted by the University of Auckland), approached The Innovation Fund with an interesting proposition. He wanted his team to investigate historical and anonymised bank transaction data to see if it was possible to find changes in the spending patterns of a population affected by a disease outbreak. The idea being that if they could find patterns, those patterns could be turned into a model that could be used by Government policy advisors – giving them new insights into how communities respond to large scale events.
The Innovation Fund agreed to provide funding for a paid internship programme, and the data was provided through an encrypted tunnel directly to the secured research lab at Te Punaha Matatini.
The multi-disciplined internship team worked for 10 weeks investigating two historic health outbreaks in New Zealand: The 2016 gastroenteritis outbreak in Havelock North, and the 2012 – 2013 influenza season in Queenstown.
This was the first time the team had the opportunity to work with banking data, which is really granular compared to other data sets available in the media or otherwise.
Westpac provided the team with transaction data that had been through a rigorous de-identification process to ensure no customer details could be compromised.
The dataset provided was estimated to cover 35% of the total number of transactions in New Zealand. Using this data gave team access to a large number of everyday transactions – allowing them to effectively study patterns and behaviours. Data masking treatments were used to prevent identification of individuals or organisations, and to simplify the analysis the transactions were grouped into 20 categories. The research focused on three specific categories – Health, Alcohol and Entertainment.
Tracking the outbreak of gastroenteritis in Havelock North
In August 2016, there was an outbreak of Gastroenteritis in the town of Havelock North caused by contaminated drinking water. Over 1,000 cases were reported – however it’s estimated that 5,500 residents became ill with the disease.
Firstly, the team looked at the financial transactions in the Hastings District in New Zealand. They assumed that during the outbreak, transactions and spending would increase at health merchants and decrease at alcohol and entertainment related merchants. However, they made an interesting discovery: Transactions in the Health category decreased when compared to previous weeks, but spending increased. That means that even though fewer visits were made to health-related shops, more money was spent there than at alcohol and entertainment related merchants.
The investigation then focused on Havelock North, where the vast majority of people affected by the outbreak lived. There was a dramatic drop in transactions and spending the day after the contamination and when a ‘weather bomb’ hit the area. The week after, although the number of transactions decreased the spending did not decrease as much, indicating an increase in the mean spend per transaction.
This could be a result of people not willing to go out for shopping as often as usual, and stocking up on more goods per trip. It also coincides with a severe number of cases reported between 13th and 18th August. Many victims of the illness had to take time off work or school – affecting local businesses.
Interestingly during the second week, transactions in surrounding areas were not affected. This could mean that many businesses in Havelock North were closed all that week as a result of people falling ill, forcing customers to shop outside the town centre.
However, locals from Havelock North spent less during the second week. This was probably a result of local businesses being closed, which meant locals had to travel further out to shop elsewhere – but only picked up the essentials.
In addition, the proportion of spending at health-related merchants also increased in the second week for the entire district.
This meant it would have been difficult to provide an early warning of an outbreak.
Unlike infectious diseases, which can take time to spread, contamination in drinking water affects the population almost instantaneously. So on the other hand, it is possible to observe and track the effect of the outbreak using transaction data.
Tracking the effects of the flu in Queenstown
Next the team investigated the transactions that took place in the Queenstown area in 2013 and 2014, from April to November, to observe the effects of the flu season.
Even with only two years of records, they were able to detect the formation of an annual pattern for the spending behaviour in the area. The data showed that transactions and spending on health increased significantly in the beginning of June – coinciding with the start of winter and flu season. A reason for this may be that although early winter is not the peak of flu season, people may have wanted to stock up as early symptoms started to appear.
Spending in the Health category decreased at the end of June and during the whole month of July – replaced by transactions and spending in alcohol and entertainment related merchants. This coincides with the Winter Festival in June and School holidays in July.
Therefore the conclusion is that the Winter Festival and July school holidays are the main factors that contribute to spending patterns during the winter period.
The results of both studies show that it is possible to identify events in a community by using anonymised payment data.
The research team concluded that the data could be valuable to the following Government agencies:
• Regional councils, ESR and the Ministry of Health for Civil Defence modelling and response (including pandemics)
• Auckland Transport and NZTA for transport modelling and planning
• MPI for biosecurity
• ATEED, Auckland Council, MBIE, and Treasury for urban economic geography planning (investigating agglomeration effects)
• Social Investment Agency, other social sector agencies and Treasury for social investment and well-being programmes”