We are pleased to announce that the winners of the competition are Nina Di Cara from Population Health Sciences and Tiff Massey, Analyst from Ernst and Young with their project ‘Is loneliness associated with movement for education?’. The specific research question assumes that in most cases, movement for primary and secondary education is associated with upward social mobility. That is, moving to try to get into a better school than is available in their current local area.  

The team’s research question was ‘Is community-level loneliness associated with the quality of local schools, and how far can this be attributed to the movement of families pursuing upward social mobility through education?’  

The winning team explored several models and created novel metrics to explore the relationship between loneliness and movement education. They found the population change caused by moving of children aged 4-15 has a small impact on loneliness in communities. They hypothesised that the reason children of this age move, is mostly to pursue better educational opportunities and so movement for the purpose of education in primary and secondary students is associated with loneliness. We will hear more about the details of the analysis in Nina Di Cara’s upcoming blog, to be published on the ONS Data Science Campus website. 

Nina Di Cara said “We were so excited at the opportunity to take part in the data challenge, especially since it gave us the chance to try out using open data to answer a question that has real-life significance. It was a lot of fun to work together and challenge ourselves – we both learned a lot by taking part so winning was a bonus!”  

Jasmine Grimsley, Senior Data Scientist at the ONS Data Science Campus, said “Congratulations to the winners of this year’s Jean Golding Institute Loneliness data challenge which provided an opportunity for students to use their cutting-edge analysis skills to answer current questions relevant to government. Students brought together alternative data sources, admin data, and combined it with existing open government data in novel ways.

“At the Data Science Campus we want to work with people from across the country to try new ways of analysing data to provide new information which can inform decisions. The methods our winners used are exciting and will help in future explorations of how the country can make better use of its data.”

The winners received £1,000 in prize money and have also been invited to the Office for National Statistics (ONS) Data Science Campus to share new ideas for data analysis. They will also have the opportunity to present their findings and spend a “Day in the life” of a Government Data Scientist. Furthermore, their work will be showcased on the Data Science Campus website in blog form.   

The two runners-up each receiving £250 are Angharad Stell and Robert Eyre.  

More about the competition

The Office for National Statistics have developed a loneliness index using open prescription data which is available at the MSOA (Middle layer Super Output Area in the ONS coding system) level across England. These data also provide information to identify MSOA’s that are within geographical clusters where the loneliness index is high or low. We would like to understand if the mobility of people for education is associated with the risk for being in a high or low cluster. The movement of people for education can be locally or across a great distance.

In this competition, we challenged participants to put forward a research question related to loneliness and movement for education, and answer it using the loneliness dataset provided (see below) alongside other suggested data sources.  Read more

The Jean Golding Institute data competitions

We run a number of competitions throughout the year – to find out more take a look at Data competitions.

 

Blog written by Jonathan Rougier, Professor of Statistical Science, University of Bristol

In a computer experiments we run our experiment in silico, in situations where it would be expensive or illegal to run them for real.

Computer code which is used as an analogue for the underlying system of interest is termed a simulator; often we have more than one simulator for a specified system. I have promoted the use of ‘simulator’ over the also-common ‘model’, because the word ‘model’ is very overloaded, especially in Statistics (see Second-order exchangeability analysis for multimodel ensembles).

Parameters and Calibration

The basic question in a computer experiment is how to relate the simulator(s) and the underlying system. We need to do this in order to calibrate the simulator’s parameters to system observations, and to make predictions about system behaviour based on runs of the simulator.

Parameters are values in the simulator which are adjustable. In principle every numerical value in the code of the simulator is adjustable, but we would usually leave physically-based values like the gravitational constant alone. It is common to find parameters in chunks of code which are standing-in for processes which are not understood, or which are being approximated at a lower resolution. In ocean simulators, for example, we distinguish between ‘molecular viscosity’, which is a measurable value, and ‘eddy viscosity‘, which is the parameter used in the code.

The process of adjusting parameters to system observations is a statistical one, requiring specification of the ‘gap’ between the simulator and the system, termed the discrepancy, and the measurement errors in the observations. In a Bayesian analysis this process tends to be called calibration. When people refer to calibration as an inverse problem it is usually because they have (maybe implicitly) assumed that the simulator is perfect and the measurement error is Normal with a simple variance. These assumptions imply that the Maximum Likelihood value for the parameters is the value which minimizes the sum of squared deviations. But we do not have to make these assumptions in a statistical analysis, and often we can use additional insights to do much better, including quantifying uncertainty.

The dominant statistical model for relating the simulator and the system is the best input model, which asserts that there is a best value for the parameters, although we do not what it is. Crucially, the best value does not make the simulator a perfect analogue of the system: there is still a gap. I helped to formalize this model, working with Michael Goldstein and the group at Durham University (e.g. Probabilistic formulations for transferring inferences from mathematical models to physical systems and Probabilistic inference for future climate using an ensemble of climate model evaluations). Michael Goldstein and I then proposed a more satisfactory reified model which was better-suited to situations where there was (or could be) more than one simulator (Reified Bayesian modelling and inference for physical systems). The paper has been well-cited but the idea has not (yet) caught on.

In a Bayesian analysis, calibration and prediction tend to be quite closely related, particularly because the same model of the gap between the simulator and the system has to be used for both calibration (using historical system behaviour) and prediction (future system behaviour). There are some applications where quite simplistic models have been widely used, such as ‘anomaly correction’ in paleoclimate reconstruction and climate prediction (See Climate simulators and climate projections).

Emulators

Calibration and prediction are fairly standard statistical operations when the simulator is cheap enough to run that it can be embedded ‘in the loop’ of a statistical calculation. But many simulators are expensive to run; for example, climate simulators on super-computers run at about 100 simulated years per month. In this case, each run has to be carefully chosen to be as informative as possible. The crucial tool here is an emulator, which is a statistical model of the simulator.

In a nutshell, carefully-chosen (expensive) runs of the simulator are used to build the emulator, and (cheap) runs of the emulator are used ‘in the loop’ of the statistical calculation. Of course, there is also a gap between the emulator and the simulator.

Choosing where to run the simulator is a topic of experimental design.

Early in the process, a space-filling design like a Latin Hypercube is popular. As the calculation progresses, it is tempting to include system observations in the experimental design. This is possible and can be very advantageous, but the book-keeping in a fully-statistical approach can get quite baroque, because of keeping track of double-counting – see Bayes linear calibrated prediction for complex systems. It is quite common in a statistical calculation to split learning about the simulator on the one hand, and using the emulator to learn about the system on the other, for pragmatic reasons (Comment on article by Sanso et al (PDF)).

Sometimes the emulator will be referred to as the surrogate simulator, particularly in Engineering. Often the surrogate is a flexible fitter with a restricted statistical provenance (e.g.’polynomial chaos (PDF)‘). This makes it difficult to use surrogates for statistical calculations, because a well-specified uncertainty about the simulator is a crucial output from an emulator. Statistics and Machine Learning have widely adopted the Gaussian process as a statistical model for an emulator.

Gaussian processes can be expensive to compute with, especially when the simulator output is high-dimensional, like a field of values (Efficient emulators for multivariate deterministic functions). The recent approach of inducing points looks promising  (On sparse variational methods and the Kullback-Leibler divergence between stochastic processes (PDF)).

Emulators have also been used in optimization problems. Here the challenge is to approximately maximize an expensive function of the parameters; I will continue to refer to this function as the ‘simulator’. Choosing the parameter values at which to run the simulator is another experimental design problem. In the early stages of the maximization the simulator runs are performed mainly to learn about the gross features of the simulator’s shape, which means they tend to be widely-scattered in the input space. But as the shape becomes better known (i.e., the emulator’s uncertainty reduces), the emphasis shifts to homing-in on the location of the maximum, and the simulator runs tend to concentrate in one region. There are some very effective statistical criteria for managing this transition from explore to exploit. This topic tends to be known as ‘Bayesian optimization’ in Machine Learning, see Michael Osborne’s page for some more details.

 

 

Blog written by Chris Adams, Teaching Fellow, School of Chemistry, University of Bristol

This project was funded by the annual Jean Golding Institute seed corn funding scheme.

Our project was a collaboration between the Schools of Computer Science and Chemistry. The computer scientist side stems from the Epic Kitchens project, which used head-mounted GoPro cameras to capture video footage from the user’s perspective of people performing kitchen tasks. They then used the resulting dataset to set challenges to the computer vision community: can a computer learn to recognise tasks that are being done slightly differently by different people? And if it can, can the computer learn to recognise whether the procedure is being done well? Conceptually this is not so far from what we as educators do in an undergraduate student laboratory; we watch the students doing their practicals, and then make judgements about their competency. Since chemistry is essentially just cooking, we joined forces to record some video footage of undergraduates doing chemistry experiments. Ultimately, one can imagine the end point being a computer trained to recognize if an experiment was being done well and providing live feedback to the student; like a surgeon doing an operation wearing a camera that can guide them. This is way in the future though….

There were some technical aspects that we were interested in exploring – for example, chemistry often involves colourless liquids in transparent vessels. The human eye generally copes with this situation without any problem, but it’s much more of a challenge for computers. There were also some educational aspects to think about – we use videos a lot in the guidance that we give students, but these are not first person, and are professionally filmed. How would footage of real students doing real experiments compare? It was also interesting to have recordings of what the students actually do (as opposed to what they’re told to do) so we can see at which points they deviate from the instructions.

We used the funding to purchase a couple of GoPros to augment those we already had, and to fund two students to help with the project. Over the course of a month, we collected film of about 30 different students undertaking the same first year chemistry experiment, each film being about 16 GB of data (thanks to the rdsf for looking after this for us). It was interesting to see how the mere fact of wearing the camera affected the student’s behaviour; several of them commented that they made mistakes which they wouldn’t normally have done, simply because they were being watched. As somebody who has sat music exams in the recent past I can testify that this is true….

One of the research students then sat down and watched the films, creating a list of which activities were being carried out at what times, and we’re now in the process of feeding that information to the computer and training it to recognize what’s going on. This analysis is still ongoing, so watch this space….

The Jean Golding Institute seed corn funding scheme

The JGI offer funding to a handful of small pilot projects every year in our seed corn funding scheme – our next round of funding will be launched in the Autumn of 2019. Find out more about the funding and the projects we have supported in our projects page.

We are pleased to announce the winners of the Curiosity Challenge are Oliver Davis and his team here at the University of Bristol: Zoe Reed, Nina Di Cara, Chris Moreno-Stokoe, Helena Davies, Valerio Maggio, Alastair Tanner and Benjamin Woolf.

The team will be collaborating with We The Curious on a prototype, which is due to be ready in October and will be going live to audiences when the new exhibition at We The Curious opens next year. Oliver Davis is an Associate Professor and Turing Fellow at Bristol Medical School and the Medical Research Council Integrative Epidemiology Unit (MRC IEU), where he leads a research team in social and genetic data science. Together his team and We The Curious will develop a ‘Curiosity Toolkit’ for a public audience called ‘Can machines understand emotion?’ The team’s toolkit will invite audiences to:

• Share the idea that humans can produce data that help to classify emotions
• Recognise that humans produce lots of data every day that expresses how they feel, and that researchers can use these data to teach machines to interpret those feelings
• Experience a live example of how the type of data they produce can contribute to an Artificial Intelligence (AI) solution to a problem
• Understand why researchers need computers to help them to analyse huge volumes of data
• Contribute to and influence current research being undertaken by the team
• Appreciate how these data can be used on a large scale to understand population health.

Oliver Davis says “Our toolkit will guide participants through the process of teaching machines how to recognise human emotion, using a series of five activities. This is directly relevant to our current research using social media data in population cohorts to better understand both mental health difficulties and positive emotions such as happiness and gratitude.”

Helen Della Nave at We The Curious said “The enthusiastic response from researchers to work with our public audiences was fantastic. Working with Oliver’s team will give audiences the opportunity to influence development of a new database of emotions and support future research. We are very excited about our audiences having the opportunity to get actively involved in this project.”

Through this competition, We The Curious have also offered Rox Middleton, a Post-doctoral Research Fellow at the University of Bristol a research residency for her project ‘How to use light’.

For further details of the competition requirements and background, see Curiosity Challenge.

The Jean Golding Institute data competitions

We run a number of competitions throughout the year – to find out more take a look at Data competitions.

Blog written by Hen Wilkinson, School for Policy Studies at the University of Bristol.

The project was funded by the annual Jean Golding Institute seed corn funding scheme. It emerged from Hen’s ESRC funded PhD research, supported by the SWDTC and School for Policy Studies.

Collaborative working is central to tackling the world’s complex problems but is not easy to sustain

Power dynamics and inequalities play out in all directions, in the relationships between individuals just as much as between organisations. By making ‘hot spots’ visible in group interactions it becomes easier to acknowledge and work with points of conflict that will inevitably arise and to deal with them in a creative and sustainable manner.

While researching ‘the space between’ individuals and organisations, qualitative researcher Hen Wilkinson and data scientist Bobby Stuijfzand developed a new methodology using computer software to visualize energy shifts in group interactions. Listening to audio recordings of groups working together on a task, the impact of nonverbal elements in the group interaction was striking, with dynamics between participants influenced just as much by the nonverbal content of laughs, silences, sighs, asides and interruptions, as by the words spoken.

Visualizing shifts of energy – a new approach in qualitative research

Following this observation, the ambition to visualize these tangible shifts of ‘energy’ in the groups took hold. To date, little attention has been paid to generating computer visuals in qualitative research, so creating a rigorous, systematic visualization of energy shifts was lengthy, challenging and exciting. For more detail on the rationale and methodology we developed over the course of two years and to view the final interactive versions of the design, see Visualizing energy shifts in group interactions. Among the many challenges we faced were finding and adapting an instrument to use with small and interactive qualitative datasets; establishing interrater reliability; identifying what was meant by ‘energy’; deciding which nonverbal elements to visualize; and how to present the resulting data.

On the website we present four 5-minute visualized extracts of group interaction, each drawn from a different group discussion, two of which were held in the UK and two in the Netherlands. Each extract of data is five minutes long, made up of 2.5 minutes of interaction either side of a central mid-point clash or strong challenge in the group. The five minutes of data were then scored by a team of raters listening independently to audio clips of the extract divided into meaning units, which are shown as ‘topic shifts’ on the visualizations. In this way, the qualitative data was converted into numerical values for three main variables – levels of mood and engagement as they shifted over a set period of time.

The support of seed corn funding from the Jean Golding Institute allowed us to work on the presentation of the visualizations, from realising an interactive website which showed how the numerical data we used was reached to refining the aesthetics of the design to encourage maximum engagement with the graphs and clarity of understanding in the viewer. Initial images were generated using ggplot2, a data visualization package for the statistical programming language ‘R’ – see Initial visualizations.

Following the generation of these first images, we explored the significance of data presentation through extensive design research, working with designer Derek Edwards. This drew on multiple sources in a visual exploration of accessibility, of the impact of colour, of multi-layered research and into the use of pattern, texture, animation and shape in displaying qualitative data. Slides from the design research show some of the various considerations we were reflecting on:

The initial images generated with ‘R’ were then refined using D3.js, a powerful and well-regarded software library used extensively to create interactive data visualizations on the web. Refining the aesthetics of the design was important to the project, both in terms of encouraging maximum engagement with the graphs and in terms of data clarity. Each graph contains multiple layers of information, from group participant engagement levels to the overall mood of the group, points of topic shift in the group discussions and dropdown text boxes of the verbal interactions between participants at any topic shift point.

The example below – visualizing a strikingly bad-tempered interaction – uses the final design we settled on (see Visualizing energy shifts in group interactions) once all considerations had been taken into account. The ‘energy line’ running through the centre of the graph is a composite of engagement and mood results and is cut across by a second nonverbal indicator of group dynamic – incidents of laughter illustrating both use and function. As outlined in the methodology sketch, we developed a categorisation for types of laughter heard in this study ranging from cohesive (green) through self-focused (yellow) to divisive (red). In this group, laughter can be seen to anticipating the shifts in mood from positive (green) to negative (red) and back again.

This project has sparked considerable interest, both in terms of its early-day implications for qualitative and mixed methods research and in terms of its potential as an applied tool for teams, organisations and collaborations to use. Further funding in 2019 through an Impact Award has enabled the interdisciplinary team working on the project to embark on further developments and connections.

We are fully aware of the work-in-progress nature of this approach and are very interested to receive feedback, comments, ideas for future applications from anyone out there! If you would like more information on this visualization project or have a comment to share, please contact the lead researcher, Hen Wilkinson, via hen.wilkinson@bristol.ac.uk.

The Jean Golding Institute seed corn funding scheme

The JGI offer funding to a handful of small pilot projects every year in our seed corn funding scheme – our next round of funding will be launched in the Autumn of 2019. Find out more about the funding and the projects we have supported.