Metrology Part III – The “soft” sciences are harder than the “hard” sciences.

Keywords: measurement, metrology, reductionism, SI units, reality

Recap

I’ve had various discussions about science where people have made statements like, “That can’t be measured” and “Science can’t tell us anything about that.” Since I have yet to be presented with a well-defined concept that science is not capable of analyzing (vs. necessarily already having analyzed) I feel such statements need to be addressed. This is the third in a series about metrology – the science of measurement.

In Metrology is not about the weather. Part I – How to weigh a potato, I introduced the most important thing to know about a measurement: there is always an error in both accuracy and precision. I then outlined the traceability of measurement standards from the International Organization for Standardization (ISO) to a produce scale used to weight a potato. The final section discussed how the formalization of measurement error is also a formalization of my three philosophical foundations discussed in Am I a figment of your imagination?  

In Metrology Part II – How do I measure thee? Let me count the ways, I introduced the International System of Units (SI) including how an infinite number of units can be derived from the seven base units. I then introduced the formal generalized definition of a measure, which allows expansion beyond SI based units such as those of the International System of Quantities. Of relevance here, is the distinction between physical units and conceptual units (my terminology). I also briefly mentioned statistics as a methodology that analyzes sets of measurements.

In this installment I will argue that the so called “soft” sciences are still physical sciences. This includes discussing the difference between the types of measurements the different sciences use. This leads to a further generalization of units, culminating in a discussion of fictional units.

Hard and Soft Sciences

The so called “hard” sciences start with physics, chemistry, and biology and then cascade down through many an application such as geology, astronomy, and medicine. The so called soft sciences are usually related to things like society and psychology – the social sciences that look at the way people act and think. The term “soft” science is sometimes used because of the impression that they are less rigorous than the “hard” sciences. In the extreme case, some people don’t even consider them sciences at all. I will argue that they are, in fact, sciences, and that they can be rigorous. I will also argue that, contrary to popular belief, social sciences are, in essence, physical sciences since they measure functions of physical brains. In fact, the main difference between hard and soft sciences is that the soft sciences are much harder than the “hard” sciences. We’ve discovered the Higgs boson but we haven’t solved world hunger.

Some standard characteristics expected from science are peer review, reproducibility, and self-correction. Science results are expected to be written down, submitted for peer review for validation, and published for historical record. Ideally, such results are reproducible. That is, someone else performing the same study would get the same results. The next level is for related studies to validate a larger hypothesis. If there are inconsistencies in related studies, then the general results (possibly even specific results) are reexamined with the intent of revising the general hypothesis. This is the self-correcting nature of science. The Reproducibility Project by the Open Science Collaboration evidences these characteristics of science in the social sciences. The project identified the 100 most commonly cited papers in psychology and went about trying to reproduce them. The results are a good example of a half-empty-half-full debate. There were enough of the studies that could not be reproduced that some people are claiming there is a reproducibility “crisis”. However, considering that psychology was (arguably) first formalized as a science by Sigmund Freud circa 1900 – that is, it is less than 150 years old – I find it remarkable that there was a reasonable percentage of studies that held up to reproducibility. The Reproducibility Project directly implements the concept of self-correction.

Soft science measurement types

The essence of science is experimentation and measurement. In order to understand the soft sciences, it is necessary to realize that the types of measurements made are significantly different than those made by the hard sciences. (This might be why some soft science experiments are called “studies” rather than “experiments.”) One of the big differences is that psychological and sociological studies often collect data by asking people questions; we don’t usually pull out a yardstick to measure what people think or feel. (However, the ability to measure people’s inner thoughts is increasing. An interesting example is the Harvard Implicit Association Test which analyzes biases you might not know you have.) This can make such data collection seem “soft” since people change their mind, are not always conscious of how they feel about something, and out-and-out lie. But these are factors that are handled with increasing sophistication.

Within sociology, or any study of groups of people, the approach is to do statistical analysis across lots of people. Since people answer questions of the type “From 1 to 10 …” differently, the use of distributions and other statistical results are the appropriate presentation of the data. But this is really just a variant on the types of statistical methods that are used in many of the hard sciences. The much ballyhooed confirmation of the existence of the Higgs Boson is an example. The result is from analyzing many particle interactions to provide a statistical inference with a certain level of confidence. It might be argued that the error associated with the Higgs Boson are smaller than that often associated with sociological surveys, but the foundational methodology is not significantly different.

As pointed out in Metrology Part I, a calibration trail is used to ensure the error of measurements of SI units. Social sciences don’t have a foundational standard equivalent to the intrinsic SI standards. But there are calibration methods just the same (even if practitioners wouldn’t use the term). The most basic error controls are simply to have large numbers of participants and to have appropriate diversity. Unfortunately, neither of these control factors has been implemented as well as they should have been. This is getting better. An example of improvement is the recognition that a lot of psychological studies have been done on college students. This has led to the acronym WIERD science (Western, Educated, Industrialized, Rich, and Democratic). Realistically, the description of this type of science should probably include white and male as well. But the growing awareness of this bias is another example of self-correction.

Another calibration method is the use of priming. This is when what immediately precedes a question can affect how someone answers a question.  I was introduced to this concept in the ‘70s as a freshman in college. We were asked to take what everyone called the “Carrot Test”. This is because you were asked multiple times whether or not you like your carrots raw or cooked. My understanding is that, although the basic idea of priming has held up, some of the original results and levels of influence have not. There is quite a science on how to write surveys to eliminate priming, leading or biased questions, and other potential flaws.

Finally, I’ll point out the A/B tests that have been done by social media (mentioned in the docudrama Social Dilemma). This is where algorithms on social media platforms have implemented two-option studies on people to see which option maximizes attention. This provides a calibration of psychological methods that allow changing behavior at a level never seen before.

Reduction to SI units

I’ve stated that all physical units are derivatives of the SI base units. So, if I want to claim that social sciences are physical sciences, it is necessary to reduce survey questions to SI units. The main argument is that brains are physical entities. Responses to surveys are a product of brain activity. I’ll use lie detectors as an example of this reduction since there is a popular image of someone taking a lie detector test.

One of the main tools used in lie detectors is an Electroencephalography (EEG). The cover photo is an example. (Note that current lie detection technology is not infallible which is why it cannot be used as legal evidence.) An EEG measures electrical activity (electron movement) of the brain. As such, the unit of measurement of an EEG is ultimately reducible to the ampere. In order to get the wavy lines there is also time (seconds) involved. Arguably volume (meters cubed) is involved since an EEG is of a particular brain. Thus, if necessary, you can reduce an EEG to some derived SI unit. A key point here is that a lie detector measures physical (SI) units and tells us something about a person’s thoughts and actions.

Although survey questions can be much more complicated than a yes or no answer, the basic principle still applies. With enough work and, realistically, more advanced science then we currently have, it would be possible to reduce the answer to a survey question to SI units. And the technology is getting closer. Some awesome examples are mind controlled prosthetics. But there has also been an experiment of technologically aided 3-way direct brain to brain communication. This requires reading what a person is thinking and writing that thought into someone else’s brain.

A key point here is that it is not necessary to formally reduce measurements to SI units in order for them to be useful. Even in the hard sciences, units of measure are usually referenced by special names rather than their specific SI equivalence.

Another key point is that it isn’t always necessary to make measurements at the molecular level to calculate a system’s properties. It is theoretically possible to calculate the miles per gallon of a car by analyzing its mass and the engine’s energy output based on molecular level analysis. But it isn’t practical and the result can be obtained by measuring the distance covered and the amount of fuel used. Similarly, there is no reason to look at a person at a molecular level in order to analyze their actions. If you want to know what someone thinks, a good place to start is by asking. (If only more people would do this outside of science, a lot of relationship problems might be more easily resolved.)

Physical, conceptual, and fictional units

I want to expand on, or generalize, the idea that it is not necessary to analyze everything at the molecular level in order for an analysis to be useful. This involves the idea of conceptual vs. physical units (again, these are my terms).

In a previous post I said that “apple” is a unit. But, as a unit, it isn’t precise in the way an SI unit is. After all, no two apples have exactly the same volume, weight, and dimensions. Yet we have little problem recognizing an apple when we pick one up. The concept of apple is generic. A specific apple is an instantiation of the generic concept. A specific apple is the physical unit associated with the conceptual unit.  More abstract conceptual units are bits and bytes. You can’t really hold a bit in your hand. You can hold instantiations of bits in your hand in the sense that you can pick up a disk drive that has electromagnetic instances of bits. But these are only bits because of their interpretation as data. Unlike an instance of an apple, there is nothing inherent about a configuration of electrons that makes them a bit. Yet conceptually manipulating bits and instantiating them in computers has led to extremely useful results. Further generalizations and abstractions of units are numbers and math in general. You can’t hold a one or an equation in your hand. We don’t normally call numbers “units” but, in the context of this discussion, there isn’t any real distinction between numbers and bits. Numbers are instantiated in the same way as bits or apples. It just doesn’t make any difference what it’s two of – it can be two bits or two apples or two of anything.

My point is that conceptual units are very useful. The distinction I’m making between physical and conceptual units reflects the modelism of my three philosophical foundations I keep returning to. Bits and numbers are used to model objective reality. By conceptually manipulating the model, there is potential to learn more about objective reality, although the results of the manipulation have to be tested against objective reality. If the results hold up, then we might be able to manipulate objective reality through instantiations of those conceptual objects.

One last thing is to make a distinction between conceptual units and fictional units. We might think of “unicorn” as a conceptual unit in that there are many images of unicorns – unicorns are conceivable. But they are purely fictional since there are no instantiations of them. I want to emphasize that there is nothing wrong with contemplating fictional things. They can be very entertaining. That is, they can affect the real world by producing pleasure (and also pain). But too many people don’t seem to understand that just imagining something doesn’t make it real. And thinking that fictional things are real causes real harm in the real world. This is where science comes into play. Science helps us distinguish fiction from reality.

A skeptic, someone who doubts, tends to assume something is fictional until reason and evidence support its reality.

(The cover photo was taken from Wikipedia.)