Most people running a commercial website look at data such as page views, user ratings and popular search terms. They may well analyse this data through the use of a funnel model that looks at the sales, or other single objective, that the site facilitates.
But governments have to use more complicated measures of success, something illustrated by a system that will visualise usage of Estonia’s state portal, which provides information and a gateway to public services in the country.
“If you look at a river delta, you see a lot of pathways that are criss-crossing, you can zoom in and see from where people are coming and going, what are the relations between different services and articles,” says the portal’s head Raimo Reiman of what is required. He has not found a product that provides this out of the box: “Traditional analytics methods, especially using funnels and things like that, don’t really work for us because we don’t have such clear end-goals for people,” he says.
As well as numerical data, the state portal uses “soft analytics”, including written comments and interviews. As a result of considering all of this, last year it added articles based on life events including starting a family, setting up a company and moving house. Another recent change has made links more visible, which has reduced complaints and raised satisfaction levels – particularly important when an organisation exists to serve the public.
The site’s status as a part of government affects the data-gathering tools it can use. “We’re not using screen recording, mouse tracking or things like that, because we have estimated the risk of data breaches as being too high,” says Reiman. The portal does not require users to log in for many functions and in general aims to keep things as anonymous as possible.
Complex measures of success and the need to serve citizens while protecting their data are all ways in which public-sector organisations can differ from companies in their use of data analysis.
“In theory and in practice in some areas, data analysis is incredibly useful for better understanding the population that you serve, the issues that face the country that you govern and how people are currently interacting with government,” says Gavin Freeguard, head of data and transparency at the Institute for Government. But it should not be seen as providing all the answers: “We talk about ‘data-driven’, which is quite nice as it alliterates,” he says. “I think you want to be data-informed rather than data-driven.”
Quantitative and qualitative information
The best results for government come from using data alongside other kinds of information, both quantitative and qualitative.
Public service – including being fair to all citizens – makes data quality particularly important for government. “There’s been a bit of a propensity recently to focus on the flashy data analytics output you can produce, rather than the fundamentals of the data,” says Jeni Tennison, chief executive of the Open Data Institute. But unless data is high in quality, reliable and representative, the results are unlikely to be useful.
“We’re in a bit of danger of creating stuff that looks really pretty, but doesn’t have substance to it,” says Tennison, adding that one risks is public bodies preferring to use data that is easy to obtain, such as web-scraped material, rather than rigorously-checked statistics.
Ensuring data is analysed fairly is particularly important when data analysis is used to decide where to target services. “The concern is about false positives,” says Richard Puleston, director of strategy, insight and engagement at Essex County Council.
If a system predicts an individual needs the involvement of social workers and they don’t, that intervention could cause harm as well as wasting public money. “I don’t think our algorithms are good enough to give us a level of confidence, and ethically we don’t want to go there,” he says of person-focused targeting.
Instead, work carried out by the Essex Centre for Data Analytics, run by the county council, Essex Police and the University of Essex, generally focuses on neighbourhoods. Its first project focused on lack of school readiness, something which affected about half of the five-year-olds in Basildon suburb Vange, and which can have long-term impacts on education, income and health.
Read more about government, the public sector and data analytics
- DCMS seeks outside experts to review government data sharing ahead of spending review.
- UK government and EU Parliament step up big data analytics policy push.
- Cancer Research UK and government-funded Genomics England use DNA-sequenced big data analytics to develop personalised cancer treatment.
The centre combined data from the county council’s social care, drugs and alcohol and youth offending services, crime information from the police and housing and benefits data from Basildon Borough Council, the second-tier local authority that provides this service, as well as commercial location data from Experian. Data on individuals and households was pseudonymised to remove identifiers and combined to cover output areas covering one or two streets of around 150 households.
Experts predicted the analysis would show that one housing estate in Vange was the main zone for lack of school readiness. But the analysis showed the risks were spread across a wider number of areas: “More than half of them weren’t already known to us,” says Puleston. “What we then did was that share that insight with the community.”
A group called New Generations, run by parents, volunteers and a local headteacher, looked at the data. This led to the establishment of Chicklets, a new nursery with support services for parents, to help young children get ready for school. Following this, around 70% of five-year-olds are now school ready in Vange, taking it from the worst-performing area in Basildon on this measure to the third-best.
Puleston says this demonstrates the value of data analysis at neighbourhood level. “Sometimes, when you have conversations with people, they say if you’re not identifying people or households it’s not of any value. That’s rubbish,” he says, with neighbourhoods being the ideal level for a community service such as the Chicklets nursery.
The centre has carried out data analysis on individual businesses in Essex, predicting which ones are likely to become of interest to the police, trading standards or licensing services, but targeting businesses does not have the privacy issues involved in targeting people.
As well as considering the ethics of data analysis, with the centre in the process of recruiting an independent ethics board, Puleston says it looks for projects where there is good data already available and where data analysis could lead to action.
On the former, he says there is a lack of data collected by the public sector on loneliness, making it hard to tackle in this way. On the latter, it is known that boys whose fathers are in prison are more likely to enter the criminal justice system, but there is no easy way to intervene: “Sometimes we can create insight without having the wherewithal to address the problem,” says Puleston.
He adds that it helps to have those delivering services work alongside data scientists: “There needs to be some creative tension between those two groups.”
Healthcare is perhaps the most obvious area of public service for the adoption of data analysis, given that medical science is largely built on this. The UK government has been led by data and science in reacting to the coronavirus epidemic over recent weeks, making a celebrity out of the UK’s chief medical officer Chris Whitty.
But politics can trump data analysis. David Nutt, professor of neuropsychopharmacology at Imperial College London, was sacked as the government’s chief advisor on drugs in 2009 after saying policy in this area was not based on evidence. Nutt’s research found that legal alcohol was more harmful to society than illegal drugs, although heroin was rated as having the greatest damage on individuals.
“The logical conclusion is, if government drugs policy is about harms, alcohol should be the primary focus,” Nutt writes in his new book Drink? The new science of alcohol and your health. “But for political reasons, this evidence has been ignored.”
The data shows that British deaths from liver disease, most of which are caused by alcohol, are five times greater than they were 40 years ago, while those from all other diseases have halved. Alcohol costs a third of what it did in 1970 in real terms and consumption has risen by about 50%, while duties are inconsistent as they are based on type of drink, adding 20p per unit of alcohol to the cost of wine but just 6p a unit to extra-strength cider.
“I think [it] is probably the most severe failure ever of UK health policy,” writes Nutt. There is no sign of change: in his first budget speech on 11 March, chancellor of the exchequer Rishi Sunak froze the UK’s alcohol duties, to cheers from MPs.
Data analysis may only be used intermittently in setting health policy, but it is of great value in managing healthcare services. However, this presents different problems.
“We have a universal health system, which means you could argue that the NHS should maybe have the best health data in the world,” says Josh Keith, senior fellow in data analytics at the Health Foundation, a healthcare charity. “But we find in practice that datasets are fragmented and incomplete, so they offer only a partial view of the healthcare experience of individuals and at a system level.”
Again, things can be improved by organisations sharing their data. The Greater Manchester Health and Social Care Partnership was set up as part of the area’s devolved control of health and social care: “We do much more now as a system, rather than as a group of people who talk occasionally,” says Graham Beales, the partnership’s head of business intelligence.
Beales has introduced Salesforce-owned Tableau’s data visualisation software with implementation work by Oklahoma-based Interworks to move analytics away from spreadsheets. This has included setting up a single dashboard for accident and emergency services across Greater Manchester, based at an emergency care hub hosted by North West Ambulance Service NHS Trust.
Information from each hospital on its A&E service is means users can spot problems before they become serious and even redirect ambulances carrying non-critical patients to quieter services, meaning those patients will wait for less time despite longer journeys. The partnership is also working to improve children’s mental health services through better population analysis.
While he sees many opportunities to do more with data, Beales is cognisant of its drawbacks, saying that the NHS needs to stop looking for non-existent average patients and start considering the wide range of health interactions people experience rather than just individual courses of treatment.
“People are, by their very nature, not very linear,” he says – the main thing that makes public-sector data analysis more complicated, more sensitive but also potentially more meaningful than much of what is carried out in the private sector.