Data Science for Social Good – Practical Examples
Dieser Artikel wurde im Original auf Deutsch von Daniel Kirsch von Data Science for Social Good Berlin veröffentlicht, und hier mit freundlicher Genehmigung und kleinen Anpassungen auf Englisch.
This article gives an overview of use cases and examples how civil society organisations can utilize modern analytical techniques (“Data Science”) for their goals. It was originally published in German by Daniel Kirsch of “Data Science for Social Good Berlin” and is translated and slightly adapted here with kind permission.
Big Data and Data Science have only recently caused significant public controversy. The large newspapers mainly discuss data protection and ethical issues of algorithms – certainly important topics. However, the potential for positive impact through a social-good oriented use of data analysis does not attract much attention.
Which is unfortunate: In many other parts of the world organizations like DataKind, BayesImpact or the “Data Science for Social Good”-Fellowships create social innovation. These initiatives enable NGOs to utilize predictive methods that are usually available only in an online advertisement or algorithmic trading. They also create attractive environments for data scientists to tackle problems of civil society. Data Scientists analyze and visualize for example the success of poverty reduction programmes or help to manage scarce fire protection resources more efficiently.
What exactly is Data Science? And with what kind of problems can Data Scientists help? In our talks with prospective partners, we noticed one common theme: Data Science is a very abstract and vague term. Many organizations we’d like to support don’t know where to start or how to make good use of modern data analysis in their daily work. For this reason, I collected and categorized a range of examples, to give interested readers from civil society organizations a better understanding of Data Science.
If you find yourself in one of the examples and want to know more, don’t hesitate to contact us! We’d like to support you.
The use cases follow this structure: Problem, Solution, Example. Afterwards, there is a list with additional examples, many from DataKind or Data Science for Social Good Chicago.
Analyzing or Predicting Demand
Problem: You want to know how you can reach your target group, but you don’t have the required data to identify potential recipients.
Solution: Data Science helps to find alternative data sets that strongly correlate with previously undetected characteristics of the target group. A model then statistically predicts the demand for a certain group/region.
Example: GiveDirectly needed information about poverty in Kenya, but was lacking official statistics. Data Scientists helped to create a system that automatically detects the roof type in satellite images so that poverty levels could be estimated automatically.
Problem: Data Science helps to estimate the results of your service for each individual case, so you can prioritize cases by impact or urgency. The model can also inform which factors contribute most to the outcome.
Solution: Amnesty International combats human rights abuses. Data Scientists were able to identify patterns in incoming requests for help from the Urgent Action Network. These patterns indicated which cases historically escalated to crises. New cases can now be assigned with an indicator of urgency.
Problem: You want to understand whether your program/campaign/service leads to the desired results.
Solution: Data Scientists can conduct statistical analysis on existing data sets to analyze and visualize results. They can also advise which additional data sets may be useful, in cases where data on the results is not available directly.
Example: The Chicago Alliance to End Homelessness wanted to know which of their programs had the best successes. Data Scientists first defined the success metrics and then visualized the results.
Early Warning Systems
Problem: You want to find out, why recipients of a service or volunteers unexpectedly quit a program.
Solution: Data Science helps to detect patterns in this behaviour so that measures can be taken proactively to re-integrated persons into the program.
Examples: DC Central Kitchen also offers vocational training programs for socially deprived persons. An analysis of their data showed that participants quit the training just shortly before the training officially ends. Further analysis showed that participants with certain characteristics have a higher probability of quitting early. With this knowledge participants at risk can now be approached proactively to keep them in the program.
Fraud Detection / Detecting Misuse
Problem: You want to detect and prevent fraud or misuse of your services.
Solution: Data analysis of usage of your service identifies anomalous or suspicious behavior which may indicate fraud, waste, or misuse.
Examples: The local government of Los Angeles checks applications for social welfare benefits with the help of data analysis for anomalies, to pick cases where an extra review should be conducted.
Problem: Your organization collects large amounts of data, however, this data is only available in unstructured form (e.g. PDF-documents or images).
Solution: Machine Learning (a method of Data Science) or Crowdsourcing, or a combination of both, enriches your data by automatically classifying images or documents or extracting data from them. The information gained here can then be used for other analysis, for example impact prediction.
Examples: UN OCHA works in the field of disaster recovery. In a case of a natural catastrophe they wanted to tag images from social media, if they show damages to infrastructure. Due to limited resources they could only tag a small amount of images. When the Typhoon Bopha struck the Philippines in 2012, the Digital Humanitarians re-used this small number of images to train an algorithm which in turn could tag a much larger number of images in a shorter amount of time.
Problem: Fundraising is another area where Data Science can help. Segmentation of donors and A/B-Testing are only two examples.
Data Science has a lot of potential for civil society organizations – which currently often stays unused. The use cases shown here hopefully make that potential more concrete and explain how data science can be used for the social good.
Out initiative is looking for partner organizations to work on concrete problems and demonstrate how to realize the potential.
The hackathon we are aiming for is modelled after the initial DataKind concept of a DataDive. In an intense, 2-day hackathon interdisciplinary teams of data scientists, developers and designers will work together with domain experts from NGOs and social enterprises to create new insight about their work, solve complex challenges and identify new social impact opportunities. This may be in form of statistical analyses and machine learning models, data visualizations or by linking existing data with new data sets.
Hints and further reading
Another source for ideas and examples is datalook – a portal where data-driven projects with a social-good goal can be published and discussed. The aim is to inspire readers to replicate projects.
DataKind categorizes the use cases differently – into descriptive insights, predictive insights and prescriptive insights. You can read their blog here.
The Nominet Trust (UK) published a report on the topic of Big Data and Social Organizations which is a good addition to this article.