To begin: performing research on ‘work-from-home’ (as I work from home)
I’ve spent the last few weeks undertaking research on where people work (focussing on people’s impressions after switching from office-work to home-work), preparing and modifying courses for on-line delivery (in particular some introductory stats courses, hence this post!), delivering some lectures, and sorting out administrative and orientation activities for the new term.
All remote and/or on my laptop.
Fortunately (or unfortunately) I am therefore a subject of my own research. So I must be careful not to let my personal impressions of remote work shade my analysis: yet, I cannot help but use my own experience as a filter through which observations are ‘read’. This is a normal part of research, and one that all researchers, whether dealing with qualitative information or numbers, need to be aware of.
Indeed, since my research involves both numbers (a large survey) and in-depth interviews, I am well aware of how personal opinions, wishes and feelings can shade what is read into numbers just as much as it can shade what is read into interviews. This does NOT mean that it is possible to make numbers and interviews say whatever I want them to: there is real (but fuzzy) information in data and interviews. It does mean, however, that at every stage choices are made about what to emphasize and what to look for. If one is not careful these choices can be guided, then overwhelmed, by personal circumstances and beliefs.
Furthermore, as all researchers know, it is never possible to state anything categorically: we work on averages and probabilities. We can make statements of quasi-certainty (e.g. data show that there is a 99.9% chance that, in early July in Montreal, workers in professional services had, on average, reduced the average time they spent working in offices by 50% – yes, 50% is an average of an average), but :
i) this quasi-certainty is not absolute. There is a possibility that the result appears by chance, by fluke.
ii) this quasi-certainty is qualified. The data are for a place: whilst there is little reason to doubt that broadly similar results would obtain elsewhere, that is not certain; and, the data are for a particular point in time (end of June, early July): there are good reasons to think that in September things are a little different.
iii) the reliability of the 50% figure is not absolute: any such number has a margin of error. There is a 95% chance that the real number (i.e. the number for the entire population of workers, not just for my sample) is somewhere between, say, 35% and 65%.
iv) there is uncertainty in the measurement: the accuracy of the measure of ‘time spent in the office” is far from absolute – it is each interviewee’s estimation.
Trump’s issues with studies of climate change, inequality, public health,….
What is true for my research is true for climate science (yes, there have been many replications of study results, but climate projections are nevertheless hedged by probabilities), and also for studies on public health, inequality and discrimination. These types of study get under Trump’s skin, and he wishes them either to disappear or to come to conclusions diametrically opposite to those reported.
Such studies can provide two types of evidence:
i) in-depth analysis and discussion of specific examples of climate change impact, inequality or discrimination (I will focus on these, but the same applies to public health). From these examples, processes can be uncovered and illustrated, but general statements can’t be made (unless these general statements are to the effect that an idea thought to be universal is not in fact universal, because a counter-example has been found);
ii) statistical studies showing that (for example), on average, one group (such as women) is very likely to be paid less than another group (such as men).
(I will not discuss projections and models here: their results are a similar to results of statistical analyses in that they provide orders of magnitude and probabilities).
To take the example of earnings, a result showing that there is a 99.99% probability that women are paid less than men for the same job does NOT show that every women is paid less than every man. Of course there are many very high-paid women who are paid more than low-paid men: but overall there are fewer high-paid women than high-paid men. Likewise, even in the same job, there are many women paid more than men : but, on average, women are paid less. There are fewer women in the higher pay grades, even though there are some.
So are scientists against Trump?
So, are scientists against Trump? Well, as a researcher, I don’t think so: results showing that climate change has a high probability of being caused by humans, and that it is most probably accelerating forest fires, are accurate. Likewise, reports showing that, on average, women are paid less than men are also accurate.
But their accuracy lies in the fact they assign a probability to climate change and to its causes. Likewise, the accuracy of data showing women are paid less, on average, than men lies in the fact that they show women are more likely to be paid less, not that every woman is paid less than every man.
One of the easy ways that Trump and his supporters have to invalidate science is to pick on absolute statements such as ‘Climate change IS caused by human activity‘; ‘Fires ARE accelerated by climate change‘; ‘Women ARE paid less than men‘ and show – quite rightly – that they are untrue. Indeed, Trump and his ilk use the type of evidence outlined in point i above: they use counter-examples to disprove absolute statements. They can do this to good effect because statements about inequality, climate change and other politically charged issues are often presented as absolutes rather than as high probabilities.
For example, in the diagram above, Trump (if he could decipher it) and his ilk can point to all the low-paid men in the red part of the graph (who are genuinely low-paid, and who liberal-minded people should care about), and compare them to the fewer women in the blue part of the graph.
Trump & Co. would thereby accurately disprove the statement that ‘women are paid less than men‘: but they don’t disprove the statement ‘on average women are paid less than men‘. Since most headline claims are that ‘Women are paid less than men‘ (this packs more punch and also corresponds to campaigners’ beliefs), Trump and his ilk are correct, however much some may find this politically and socially unpalatable.
The big problem for equality campaigners (and for climate scientists, pandemic specialists and so on) is that Trump can marshall many specific examples that ‘disprove’ the idea that climate change is upon us, that women are discriminated against, that wearing masks slows the propagation of Covid-19, and so on.
Given that many people are more or less innumerate (it is ‘OK’ to find maths and stats tedious, far less ‘OK’ to refuse to learn how to read), then it is difficult to argue against Trump’s claims: his claims are often (not always) technically correct (but usually totally improbable, as well as being politically unpalatable to many liberals ). Similarly, absolute statements about climate change and inequality are technically incorrect (but often highly probable, as well as being politically palatable to many liberals).
The personal wishes and desires of scientists (and campaigners)
One item that current movements about climate, inequalities, racism etc… need to beware of is that the personal desires and politics of engaged researchers can interfere with their careful presentation of results: because many climate change advocates badly want climate action, because equality advocates badly want inequality to be recognised and dealt with, they resort to absolute statements that are easily disproved. These absolute statements are good for headlines, they pack a punch, but they are fodder for Fox news and other groups who – for whatever reason – wish to deny the strong (but not absolute) evidence of systemic biases, rising inequalities and human-caused climate change. Absolute statements can be disproved with a single counter-example (Karl Popper’s epistemology of falsification is all about this): but subtle statements are not listened to….
Of course, researchers are well aware of these problems, but subtle arguments get drowned out by all sides of the ‘debate’. The temptation is strong to overstate one’s case, to deny factually correct counter-examples, to not deal with unpalatable realities (such as low-paid men and high-paid women – which of course tell us nothing about averages or systemic biases, but which should not be ignored either) and to omit the subtleties and fuzziness of actual research results. Punchy absolutes can grab headlines and generate support: but in the longer term they set research up for easy dismissal ‘à la Trump’.
In my own research – dealing with less pressing issues than climate change and inequality – I also need to be careful. I am working from home, and have many frustrations. But whatever my own experience and beliefs, I must be attentive to what my data and interviewees reveal, and to not overstate or misrepresent the evidence.
Naturally, with careful statements and equivocal findings, I am unlikely to become a research superstar! Maybe I should use these unusual pandemic circumstances to announce the (inevitable) end of cities as we know them, and reveal the dawn of a new suburban age… provided this is disproved after my best-seller has sold a few million copies, who cares!