Social sciences and the large-scale networks of low-cost instruments for measurement of air quality
One of my principal research drives since I joined the University of Auckland, New Zealand, has been in the development of large-scale networks of low-cost instruments for measurement of air quality. My stay at Durham (October-December 2018) was a wonderful time of welcome and warm friendship and scholarship. It was illuminating in many ways. I had never before had the opportunity to interact closely with scholars from such a range of disciplines. It exposed me to different ways of thinking and looking at the world. It gave me the opportunity to discuss details of my programme with experts in statistics, mathematics and engineering as well as with lawyers, philosophers and social scientists. I also had time to read broadly around the problems that I was thinking about, stimulated by these conversations.
Air quality had leapt into prominence as a serious issue in Europe. The ‘dieselgate’ scandal had erupted in 2015 after Volkswagen was found to have installed illegal ‘defeat devices’ to cheat emissions tests. It became one of the costliest corporate scandals in history. As I began my Fellowship, the WHO head warned that “air pollution is the new tobacco…the simple act of breathing is killing 7 million people every year, yet a smog of complacency pervades the planet”. Since then, the Covid pandemic has highlighted both the marked decrease in traffic-associated pollution and linked health issues that happened as a consequence of lockdowns, and the suspected synergy of particulate air pollution and viral infection spread from the possibility of particles in the air carrying virus-containing aqueous films on their surface. Air pollution as a shared problem has not gone away.
My conversations at Durham highlighted for me that there were two linked research questions: one was my primary research interest – reliable (meaning fit-for-purpose) data from low-cost instruments used in networks – and the other was a social science question about consensus, access, availability, consequent decisions or actions and what ‘fit-for-purpose’ actually meant to different individuals or groups in society. I may have resolved, at least to my own satisfaction, the route to reliable data from low-cost instruments used in networks, and indeed demonstrated this with an extensive network around Los Angeles, in a collaboration with the South Coast Air Quality Management District, but the social science questions remain very much open.
I summarised my thinking around reliability in an article “Low Cost Sensor Networks: How Do We Know the Data Are Reliable?” published in ACS Sensors, 2019, vol 4, pp2558−2565. My reading in the science and engineering literature had led me to into deep waters, floundering helplessly in matrix algebra that I had not visited since I was an undergraduate. The way out, no doubt seeded by my conversations in Durham, was to go back to the basics of the theory of measurement – what actually does it mean to take a measurement and derive some result from it – and to read ideas about the philosophical basis of statements of probability. These ideas emphasised the degree of belief in a result, epitomised by ‘what am I prepared to bet on a particular result’, and the degree of plausibility, focussing on what assumptions have been made in deriving the result, and their plausibility, and evaluation of evidence supporting or contra-indicating the result. Framed that way, attention is immediately focussed on the assumptions either explicitly or implicitly made in going from an observation or measurement to a reported result. If the assumptions are hidden, inappropriate, difficult to determine, or unacceptable, then the believability of the result will be low. There was a need to understand what ‘reliability’ meant in relation to the network purpose. I built on this philosophical framework to devise some very simple and transparent methods to provide reliable data from a hierarchical network comprising a large number of low-cost sensors with a small number of high-quality instrument sites. These results were published in a series of papers. The resultant network continues in use in Southern California and similar networks have been installed in other cities in the USA.
Ultimately, whether sensor network data are plausible is connected to the question of whether they are useful. Here lies a connection between the formal science of measurement and the social science of the consequences. Thus, if the network data reveal patterns which seem plausible based on other experience, or which on investigation have an explanation, then the data themselves can be accepted because they have revealed something useful. For example, correlation of sensor network data with land-use information shows up urban design features that are of importance for personal exposure, leading to understanding of urban design effects on air quality – ‘green walls’, trees, barriers, awnings, parking places, bus lanes and so on. It can also identify specific local effects on air quality, such as high concentrations of pollutants downwind from a transport hub at particular times of day.
The elephant in the room, of course, is what actions might be taken as a consequence of the measurement. ‘Big’ actions at city, national or supra-national scale include bans on diesel-powered vehicles, low-emission zones, congestion charging and access restrictions. These can have a long time to take effect, can result in a feeling of disempowerment of citizens, can be perceived as disproportionately affecting particular social groups and can be vigorously contested. On the other hand, many actions are possible, dependent on the quality, timeliness and spatial resolution of the data. What is missing, I think, is the social science of how people interact with data and how trust in data is connected with actions, prejudices and consequences. Given that the technology is now available and that methods exist by which ‘authority’ can be persuaded to accept that low-cost instrument data are reliable, there might seem to be a basis for cooperation to extend networks to local scale. Perhaps communities could be empowered to assess, effect and change local actions, with local evaluation and reassessment of consequences. There are issues of social justice also: low-income communities are disproportionately impacted by poor air quality because they are the people living alongside major roads, close to industrial sites and near major transport hubs; yet they have less access to instruments, and less time to devote to management and maintenance. There is an ongoing debate: the “Internet of Things” (low-cost sensors) inserts citizens into analysis, measurement and risk assessment; or does it? This is actually hard to do, needing a serious commitment, so in the end could be taken over by specialists. It seems to me that this will be an ongoing conversation and an area in which social science would seem to be central.