Understanding if current local boundaries are most effective

Freegle is a UK-wide platform where people can give away and ask for things that would otherwise be thrown away. Instead of donating an unwanted item to a charity shop, an user (the offerer) posts the item on the plaform, for another user (the replier) to express interest for the item. Once an agreement has been reached, the exchange of the item happens, usually by the replier collecting the item in person. Any user can give away items, collect other items and ask for items.

The request

The aim of the work was to analyse the existing local geographical areas of Freegle communities and data about the item exchanges that have happened, to see whether we can identify natural “communities of interest” and suggest boundary changes. This would help make more reuse happen, which is what Freegle is all about.

The approach

Initial discussion was made to understand and explore the user data. It was agreed that two main parts of the data would be useful for analysis. The first is the spatial information of the offerers and repliers, as this allows their interactions to be visualised on a map. The second is the number of messages of offerers and repliers who have interacted, even if the interaction did not result in a successful exchange.

After obtaining the relevant data, exploratory results were presented in subsequent meetings, and the hypothesis and analyses were refined, with the help of Edward’s understanding of the Freegle data. Carrying out the analyses in such an iterative fashion led to satisfactory statistical models and visualisations being agreed on.

The result

On the spatial analysis front, the network of interacting users is visualised, with some suggestions of network modelling provided, to enable comparison with the existing communities of Freegle.
On the non-spatial analysis front, the effect on the probability of a successful exchange by both the number of messages frequency and the physical distance between users is quantified by a modified logistic regression model. Specifically, such probability has an upper limit (which is smaller than 1) even if the users are very close to each other and interact frequently. Not only is such discovery equally interesting and unusual, it also suggests room for raising the successful exchange rate.

Impact and benefits

I loved collaborating on this project with Clement.  Far too often, charities can drown in a sea of data, draw unjustified conclusions, or simply ignore the valuable data which they possess for lack of expertise to analyse it. 

Applying professional statistical expertise from Clement, enabled by the Royal Statistical Society’s programme, enabled us to identify some key insights which we can take action on immediately (e.g. the number of message exchanges which indicates a higher chance of success), as well as some more avenues to explore.  We wouldn’t have been able to make progress on this on our own.

I’d highly recommend this programme to anyone with data they’d like to get a grip on.