By Charlotte Service

Although data has been around in its various forms for centuries, its role within society has become increasingly significant in recent years. The incessant advancements in the realms of technology has propelled our ability to store and share information into a constant state of progression. A profound collection of data and numbers allow us to draw conclusions for all kinds of scenarios. Accordingly, we have seen statistics and models amalgamated into the core infrastructure of government policy making. Yet, given recent political choices and the ongoing pandemic, it is intuitive to consider the potential consequences that can arise from data-based decisions lacking in human agency. 

The relationship between science and politics is as close as it ever has been. Boris Johnson’s chief advisor Dominic Cummings’ view of British bureaucracy is no secret. Cummings has laid bare his vision for the future of the civil service, advocating plans to cut back the cabinet and probing ways to bypass Whitehall. Fundamentally, he believes it’s time to put more faith in the data, with fewer arts graduates and more scientists. At a glance, this seems both reasonable and in keeping with the modern world – perhaps even a more efficient and impartial way to run the government. However, an influx of artificial intelligence threatens to replace practical knowledge that only people can provide; common sense, that which provides stable foundations for policymaking and hence, is essential for good governance.

There is no denying the usefulness of data analysis, both in its ability to make sense of a vast collection of information and predict potential outcomes. Nevertheless, the interpretation of the data and context of the matter in question are of particular importance as without proper real-life considerations, forecasts can become – at worst – mindless and – at best –inaccurate predictions. Scientific models tend to rely heavily on their assumptions and thus, the methodology behind them must be thoroughly scrutinized. A deficient understanding of the relevant affair and a fixation on headline numbers is the reason why many algorithms designed to solve problems are more often than not seen to create them. This dilemma, combined with Britain’s ever-increasing reliance on science to determine its next move, can perhaps explain why the government – from travel restrictions to exam results – has acquired a tendency to take the so-called U-turn. 

The students who did not sit exams this year, leaving the fate of their results in the hands of the government, are just one example of the collateral damage that results from data-driven decisions. An algorithm designed to alleviate grade inflation left students with significantly poorer grades than they had been predicted – meaning many failed to attain the standard for their university offers. 40 per cent of A-level grades were downgraded from teachers’ predictions, with the grading system most heavily burdening those from lower socio-economic backgrounds. Naturally, this left the government with little choice but to adjust grades, this time basing them on the past performance of students instead of the previous academic record of their school. Common sense would have prompted this approach in the first place, but the government chose to trust the algorithm. Additionally, prior to the modification of exam results, universities accepted many students who did not make their offers based on the assumption that results were an unfair representation of academic ability. However, upon the alterations, universities were forced to accept more students than they were prepared for, with even less time to figure out how to deal with such a large influx of students. Predominantly, the point is not that the algorithm was unsuccessful in its attempts to evade a particular problem, but rather that a lack of consideration for other relevant factors rules out the bigger picture, and in turn, has the potential to induce a string of complications carrying much greater weight.

The influence of science in the political sphere has been evident throughout the pandemic. The news has been littered with charts, graphs and statistics daily, all on display to justify the UK government’s next move. Despite the prospects of an initial “herd immunity” strategy for Britain back in March, it soon materialised that such an approach wouldn’t be sufficient to overcome COVID-19 in the eyes of data science. Professor Neil Ferguson and his team at Imperial College-London (ICL) presented a model which necessitated the enforcement of mandatory restrictions on normal social interaction in order to avoid what he predicted would reach over  500,000 deaths.  As a consequence, Britain entered into a strict lockdown and we witnessed the government completely backtrack.

Despite the heavy economic and social cost of lockdown, you might say that Ferguson’s data handling had a positive effect on the whole of the UK. Nonetheless, it is worth noting that the scientist has a record for proposing some rather extreme predictions, most famously done so during the foot and mouth epidemic, where his work induced the mass culling of over 6.5 million animals. Many say that the professor’s modelling was “not fit for purpose” and led to the unnecessary killing of livestock. As a result, questions have been raised as to whether or not Ferguson’s Covid figures were similarly extraneous and hyperbolic.

In order to ascertain the reliability of Ferguson’s modelling, it seems astute to evaluate the country which, in contrast to the rest of Europe, avoided the imposition of the formal lockdown copiously recommended by the World Health Organisation. With the reputation of being a rather non-conformist nation, it was no surprise that the Swedes took a more relaxed approach to the pandemic. Though those vulnerable to the virus were told to remain at home and gatherings of more than 50 people were banned, schools, pubs, restaurants and gyms have remained open in Sweden throughout the pandemic. Hence, the Swedes have maintained a form of normality that the vast majority of the world has become alienated from today. 

Anders Tegnell, the doctor behind Sweden’s lack of lockdown, was well aware of the longevity of COVID-19. Having previously worked with Ebola in Africa, Tegnell’s practical experience in the field of disease management meant that he anticipated the epidemic’s potential to emanate extensive social damage. However, his recommended approach to the handling of coronavirus was undeniably controversial, particularly so when the number of Swedish cases began to rise dramatically during the summer months.  But as cases have started to rise again throughout the rest of Europe, Sweden’s current number of cases per capita are 90 percent below their peak in late June and the Swedes approach has earned them a lower mortality rate than the UK, Spain and France. The Swedish doctor claims that because of his country’s long-term focus, their forthcoming spread of the virus will be much less volatile and perilous than those who enforced strict lockdown measures. 

Only time will tell whether or not Sweden was right in its decision to minimize social restrictions over the past months. Nonetheless, Sweden’s approach can be utilised to examine the predictive performance of Ferguson’s epidemiology model. Though the model was used exclusively to make predictions for the UK and United States, it can be theoretically adapted to construct forecasts for any other nation; this is done simply through altering the inputs, such country population and policy imposition dates.  

At the beginning of April, researchers at Uppsala University elected to do just that, releasing an epidemiological model for Sweden that adapted the ICL COVID-19 model from Ferguson and his colleagues. Once again, the model produced results which insisted upon the immediate adoption of strict lockdown policies similar to those currently implemented by the rest of Europe. Otherwise, Sweden was to foresee a median mortality of 96,000 by the end of June. The proposed mitigation measures of the ICL model were said to be capable of reducing this mortality by approximately three-fold, as well as preventing the collapse of the Swede’s healthcare system. However, Sweden has not adhered to the recommended restrictions and their death toll currently stands at 5,918. Though Sweden’s stats are far from desirable, with a mortality rate significantly higher than that their neighbouring countries with similarly low-population densities, generally the figures do not speak well for the performance of the ICL model. In assuming the Uppsala researchers’ adaptation of the model was correct, the discrepancy of the predictions relative to the prevailing figures truly emphasises the need for clarity and scrutinization in model methodology, as well as the significance of underlying assumptions in data modelling.   

Though the ICL model produced forecasts that were undeniably exaggerated, the rationale behind it was sound. The above findings simply highlight the limitations of data, not necessarily that Ferguson’s recommended restrictions were inappropriate; they stress the importance of gaining a balance between theory and experience when it comes to policy decision-making. The statistical facts cannot be denied, but context cannot be neglected. Real life isn’t rational, and it isn’t always something that an algorithm can predict accurately. As long as politicians remain fixated on setting targets and meeting deadlines, they risk losing sight of the bigger picture and consequently, will sacrifice the potential benefits of combining both calculations and common sense. 

The views expressed in this article are the author’s own and may not reflect the opinions of The St Andrews Economist.