This continues from the previous post on the Downfall of Rhetoric in 20th Century
Even though our goal is to explain how apparently objective looking statistics conceal arbitrary and subjective judgments, the path we take requires a detour through “epistemology”, or the theory of knowledge. Instead of the deep discussion provided by Putnam (2002, Collapse of the Fact/Value Dichotomy), we will take a shortcut, and look at how these philosophical debates and controversies about have shaped the way that social sciences in general, and statistics in particular, have conceived of the relationship between the numbers we analyzed and the real world that generates these numbers. The wide gap between the philosophers and other intellectuals can be seen clearly in their respective views regarding logical positivism. One of the lifetime advocates of logical positivism, A.J. Ayer, eventually came to the realization that “it was all wrong”. Another sympathizer and proponent, Bas Von Fraasen, opens his book “The Scientific Image” by saying that this philosophy had a “spectacular crash”:
“Today, however. no one can adhere to any of these philosophical positions to any large extent. Logical positivism, especially, even if one is quite charitable about what counts as a development rather than a change of position, had a rather spectacular crash. So let us forget these labels which never do more than impose a momentary order on the shifting sands of philosophical fortune. and let us see what problems are faced by an aspirant empiricist today.”
Despite clear acknowledgements of its failure, typical non-philosophers neither understand the evolution of positivism under pressures created by scientific discoveries, nor understand the reasons for its eventual abandonment. Despite general awareness that this philosophy has collapsed, a recent survey by Hands (2007) finds that economists continue to believe in the central tenets of logical positivism. As I have argued in Zaman (2012), the foundations of econometrics are solidly built and logical positivist principles. After the collapse of positivism, it became essential to re-examine these foundations, and re-build the discipline on a different set of fundamental principles. This revolution still remains to be carried out. Some aspects of the change required are discussed in Simpson’s Paradox. We plan to deal with a more elementary aspect, namely the relation between the data we use and the realities of the external world that these numbers are supposed to measure. But first, we must look at how the fact/value dichotomy is viewed by the general, non-philosophical public. It useful to note that we are expositing and providing a critique of the popular understanding of positivist philosophy, which must be differentiated from the more sophistical philosophical versions.
It is easily understood that there are “facts” : (F) the number of students who took the SAT in the USA in 2019 was 2,220,087. It is also true that there are values, like the “golden rule”: (V) do unto others as you would have them do unto you. There is indeed a sharp separation between these two statements; F is objective and can be verified by any independent observer – all would come to the same conclusion. V is subjective and different people can have different opinions about whether or not it is true or false. Furthermore, there is no way to establish whether or not V is true; there is no method for checking values against objective empirical realities in the world around us, to see if it is true or false. The key argument that Hilary Putnam makes is that this distinction exists and is valid, but it is not a dichotomy. To be more explicit, it is not true that all statements can be classified into one of these two categories. Facts and Values both exist, but the vast majority of propositions we deal with in our lives and in our knowledge disciplines cannot be classified as being either a fact or a value. Furthermore, given a statement in which both facts and values are entangled, we cannot pry the two apart to create two statements, one of which is purely factual while the other is purely value.
Putnam argues that wrong conclusions have been drawn because a distinction has been inflated into a dichotomy. Treating the distinction between facts and values as a dichotomy leads to disastrous results. Once we show that something is not a “fact” – that is, it does not have any direct translation to an observable aspect of external reality – then we are forced to conclude that it must be a “value”, and hence not part of reliable human knowledge. As Putnam has shown, in most of the knowledge that we use to conduct our daily lives, facts and values are “inextricably entangled”. All our lives we are faced with major decisions like “which college should I go to?”, “which person should I marry?”, “which job should I apply for?”. For making such decisions, it would be useful to have an objective ranking over the choices that we have. However, as we will show later in this paper, objective rankings are not possible when there are multiple dimensions involved. For example, if one college has a strong math department, while the other has a strong english department, then choice among the two will have to be based on my personal preferences regarding the balance between the two skills I would like to acquire. However, this subjective decision is not PURELY a value judgment. There are many facts we take into consideration in arriving at such decisions. Contrary to the conception that economics is purely positive – based purely on facts – while values are used by policy makers, the facts presented to the policy makers are created via a mixture of facts, and subjective decisions regarding how to weight the different facts, in order to combine them into a single number.
Just as individual decisions are based on mixtures of facts and value, so collective choices by communities are based on mixtures of facts and values. Every nation has a large amount of wealth in terms of land, water, infrastructure, as well as skilled human beings capable of learning and producing objects. Each nation faces choices in terms of where to spend energies to achieve best results in the future. In making these choices about how much to invest in factories, how much in education, and so on, we must make subjective judgements. There is no way to avoid making value judgments when decisions require choosing over multidimensional characteristics. The positivist point of view, almost universally advocated by economists and econometricians, is that we can separate the objective and the subjective. The econometricians should present purely objective facts to the policy makers, while the policy makers use their subjective values to make decisions. Our goal in this paper is to show that this separation cannot be done. The “facts” we present to policy makers require us to make arbitrary choices. It is impossible to do otherwise, because reducing multidimensional characteristics to a single number always involves making subjective decisions regarding the relative weights of the different dimensions. At the same time it is impossible to directly present the complete and unadulterated purely objective data, because this would be incomprehensible in raw format. Any procedure for “reducing” masses of data to a small and manageable set of numbers to guide policy requires subjective decisions. Thus, nearly all of the numbers currently in use by statisticians and econometricians are mixtures of facts and values, and it is impossible to avoid doing this mixing.
In the remainder of the paper, we move from abstract philosophical consideration to practical illustrations, to show how numbers we routinely use and regard as objective, conceal value judgments. Those who are aware of how these values are built into the manufacture of statistics can use this knowledge to deceive people. They can bake in their own value judgments into the statistics which they manufacture, while maintaining an appearance of objectivity that is automatically created by the use of quantitative data. This may be reason why by far the most popular book on statistics, with more than 1.5 million copies sold, is the time-revered classic by Darrel Huff called “How to Lie with Statistics”.