Projects
Projects: Custom Search |
||
| Reference Number | ES/Y010787/1 | |
| Title | DEBIAS - DEveloping a framework to measure and correct BIASes in human mobility data extracted from digital footprints | |
| Status | Started | |
| Energy Categories | Other Cross-Cutting Technologies or Research (Demographics) 10%; Energy Efficiency (Transport) 20%; Not Energy Related 70%; |
|
| Research Types | Basic and strategic applied research 100% | |
| Science and Technology Fields | SOCIAL SCIENCES (Sociology) 50%; PHYSICAL SCIENCES AND MATHEMATICS (Computer Science and Informatics) 50%; |
|
| UKERC Cross Cutting Characterisation | Not Cross-cutting 70%; Sociological economical and environmental impact of energy 30%; |
|
| Principal Investigator |
Professor F Rowe Geography and Planning University of Liverpool |
|
| Award Type | Standard | |
| Funding Source | ESRC | |
| Start Date | 15 August 2024 | |
| End Date | 14 August 2025 | |
| Duration | 12 months | |
| Total Grant Value | £150,520 | |
| Industrial Sectors | ||
| Region | North West | |
| Programme | Data & Infrastructure - Services | |
| Investigators | Principal Investigator | Professor F Rowe , Geography and Planning, University of Liverpool (99.999%) |
| Other Investigator | Dr C Cabrera-Arnau , Geography and Planning, University of Liverpool (0.001%) |
|
| Industrial Collaborator | Project Contact , Northeastern University, USA (0.000%) |
|
| Web Site | ||
| Objectives | This project aims to develop a generalisable framework to measure, assess and correct existing biases in human mobility data extracted from Digital Footprints (DFs). We will develop this framework using location data from Meta-Facebook, Twitter and Huq for the UK. The framework will be reproducible and transferable, and designed to use aggregate, privacy-preserving origin-destination mobility data from any geographical setting. This data structure ensures anonymisation and is a common way for data holders to share sensitive data, supporting thus the long-term vision of the ESRC in providing a framework sustain the use of robust and ethical DF Data (DFD) for research. The project delivers this vision by tackling three objectives:1. To quantify biases in spatial population counts derived from DFD arising from differences in the access and use of the digital technology, and identify key population attributes associated with these biases. First, we will design a statistical indicator to measure the extent of bias in DF-based population counts for subnational areas. Focusing on a digital platform (e.g. Meta-Facebook), we will compute the share of active user population over the total population by area, to determine its local coverage levels. Second, we will use machine learning to model the relationship between our measure of population bias and area-level demographic, socioeconomic and geographic attributes. The analysis will identify the most important population features characterising biases in DFD from a given digital platform, and inform our modelling analysis in WP-II to generate bias-corrected DF human mobility data.2. To create correction factors to mitigate biases in spatial population count data, and generate bias-adjusted, DF-derived human mobility count data in the form of origin-destination matrices. Building on work by the PI, we will develop a two-stage Bayesian modelling framework. In the first stage, we will produce a two-level Bayesian model to estimate sub-national bias-adjusted DF-derived mobility counts. The first level will use our measure from WP-I to estimate and isolate existing biases in our DF-derived mobility counts arising from differences in the access and use of digital technology. The second will produce bias-adjusted mobility counts in a spatial interaction modelling framework. Mobility counts will be modelled as a function of socioeconomic, demographic and geographic attributes of origin and destination areas. In a second stage, a separate spatial interaction model will be estimated to impute missing mobility counts which have not been captured by our DFs, because of GPS jamming, geographic restrictions or privacy constraints.3. To assess the validity of our bias-adjusted DF-derived mobility counts against comparable official statistical sources of human mobility data. We will correlate our bias-adjusted DF-derived mobility counts with corresponding data from the 2021 census, mid-year population estimates and labour force survey at different spatial scales. We will focus on two forms of human mobility using data from official sources (i.e. internal migration and travel-to-work). We will assess six theoretically relevant measures of mobility. We expect high correlation between our bias-adjusted DF-derived mobility counts and reported official counts capturing broad similar trends in mobility. We do not expect an exact match in numbers as a range of measurement issues prevents this.The delivery of these objectives will provide a framework to enable the production of more reliable mobility data from DFs to monitor and predict human mobility patterns. Reliable DF-based mobility data are essential to support a range of societal decisions relating to retail catchments, housing, transport demand, carbon emissions and transmissible diseases. We will also produce an open-source software package and training materials to facilitate the implementation and ensure the legacy of our framework. | |
| Abstract | Our project, DEBIAS, aims to develop a generalisable framework to quantify and adjust existing biases in Digital Footprint Data (DFD) on human mobility. To this end, we will use DFD on human mobility obtained from users of the social media platforms Facebook and X (previously Twitter), and from smartphone applications collected by the company Huq. We will use data from the UK, but our framework will be reproducible and transferable to any DF source and geographical setting. Our framework will rely on aggregate human mobility data capturing flows of people between origins and destinations. Key benefits of using this data structure are that these data are more easily accessible. They help overcome ethical concerns ensuring anonymisation and represent a common format used by data providers to share DFD on mobility.Why mobility? Understanding how humans move is key to supporting appropriate policy responses to address population issues, carbon emission, urban planning, service delivery, public health and disaster management. DFD, such as location data collected from smartphone apps offer a unique opportunity to analyse population movements at high geographic and temporal granularity, with extensive coverage in near real-time. Research leveraging DFD has have a transformative impact expanding existing theories and developing new analytical tools and infrastructure of social and spatial human behaviour across the social sciences.Challenge: Biases in DFD have represented a major methodological barrier to reaping their benefits, contributing to scepticism and deterring wider usage of DFD. Biases mainly exist due to differences in: 1) the access and usage of the digital technologies used to collect the data (e.g. only 70% of the British population uses Facebook); and, 2) the demographic and socioeconomic profiles of users of the technology (e.g. Twitter has a young adult and male-dominated user profile mainly from urban areas). As such, human mobility data derived from DFD offer a partial representation of the overall population, limiting our capacity to draw conclusions about the overall local population.Promise: DEBIAS will deliver a framework to adjust biases in DF-derived mobility data and an open-source software package and training materials to implement it. These outcomes will contribute to delivering the Smart Data Research UK (SDRUK) programme aim of unlocking the power of new forms of data for research and innovation to tackle social challenges by 1) enabling the monitorisation and prediction of patterns of human mobility by facilitating robust real-time analysis based on DFs; 2) augmenting the technical social science research capacity in the use of DFD; 3) expanding existing theories and developing new explanations on spatial human behaviour by supporting research on highly granular space-time mobility patterns; and, 4) supporting the long-term access to robust and ethical DF-derived mobility data. DEBIAS will thus contribute to SDRUK-specific objectives by providing secure data access, safeguarding public trust, and building capability for cutting-edge research. To maximise impact, we will engage with direct beneficiaries of our work: a) researchers and analysts; b) public sector agencies; and, c) commercial stakeholders or third sector organisations engaged in data for public good initiatives and working in mobility, transport and migration. We will establish an advisory board representing expertise from the commercial (Meta), academic (Northeastern U.) and transgovernmental (UN) sectors, to inform the development of our software tool and training materials. Working with Meta and UN-IOM will enable the exploration of opportunities for long-term strategic partnerships for data access and application of our approach to new problems. We will disseminate and increase the awareness of our work via research articles, presentations and workshops targeting different audiences and adopting open science principles | |
| Data | No related datasets |
|
| Projects | No related projects |
|
| Publications | No related publications |
|
| Added to Database | 02/10/24 | |
