A new interdisciplinary approach to the study of social sciences | Introduction #
Hard, Social, Sciences #
In the study of phenomena it’s common to look for reductionist perspectives. In the context of social sciences, such vectors of analysis were especially loved by positivist and neo-positivist thinkers, who always looked to “harden” sociology, economics, political sciences and their many branches into the mold of more rigorous disciplines, such as physics and chemistry.
As is often the case in the history of social sciences, this approach fell partly flat, as it tried to squeeze the proverbial square peg in a round hole: as complicated physics or chemistry may be, they enjoy a much more clearly visible deterministic flavour than the social sciences. An assumption can be made -and indeed is made- that an experiment can be repeated and that similar results can be expected from each repetition, assuming minimal variation in the starting conditions, which can be partially -dare I say mostly- controlled.
On this last point -control- social sciences have their biggest weakness. Because in observing a person we see a unique individual, and not “a specimen of human”, we can’t rightly make perfectly accurate inferences from single instances of experiments, and because people are different both from each other and from themselves (across time), we can’t properly replicate the same experiment more than once without incurring in some degree of inaccuracy
Positivist Approach #
Obviously correct sampling techniques can minimise systemic biases, and so we can assume that for large enough sample sizes we will obtain results we can rightly make inferences from. However, while doing so, and thus diving deep into the positivist approach to the social sciences, we lose part of the connection with the casual relationship we are trying to study in the first place. We can observe statistically significant correlations and try to guess the relationship between the variables the change of which we observed, but little more.
Interpretative Approach #
On the other side of the spectrum, we see the interpretative paradigm of sociology. A more personal branch of study based on the concepts of Verstehen and typology. If positivism, especially if it devolves into variable-sociology, is akin to fitting a square peg into the proverbial round hole, Interpretative sociology often faces of the risk of trying to make a needle fit into it; of course, it passes without a hitch, but there’s very little potential for inference to be made out of it.
Current-day Practice #
While the interpretative approach is far more accurate in describing the exact phenomenon it’s studying (it’s internal validity is very high), the positivist approach is great at making broad inference, but rarely neatly fits any real situation (it’s external validity is very high).
This brought modern-day researchers to adopt a hybrid approach, using a more interpretative paradigm to conduct exploratory research, and then to use its findings to try to construct valid operationalizations of social variables, to try to conduct broader research to make inferences about how society at large works.
It’s worthy of note that this debate isn’t only present in the social sciences. More individualised approaches to research are present in other fields, like medicine, where recently some Eastern traditional techniques (such as meditation) are being incorporated in western systems as randomised control trials (RCTs) started to show their effectiveness. In this case, the influx may (and has) lead some to reconsider the validity of heuristics, even in a “harder” science, such as medicine (which, while it isn’t physics, is certainly considered more rigorous than the social sciences).
The Value of Heuristics #
While “heuristics” is often negatively charged, especially in scientific fields, I personally find it undeniable that it’s ubiquitous in any field that isn’t strictly reductionist (meaning most fields). While in some fields (such as physics) it may be limited, in other fields, such as the Social Sciences, I would argue that it’s abundant. This is the case because of the dominance of the positivist paradigm within social sciences. It’s often easy in reading economics, sociology, or political science papers to lose track of the fact that humans are being talked about, and not puppets in a perfect model. It’s true that some fields (such as behavioural economics) are trying to avoid falling into variable-sociology, but they’re still relatively new and not the norm.
Lets quickly establish what I’ll be meaning by “heuristics” specifically:
A heuristic is the pragmatic abstraction of a process that attempts to explain its effects, but does not provide a clear explanation for the underpinning causal relationships that explain it
Seen in this lens, we can imagine a “broad”, or “high-level” heuristic, or a “focused”, or “low-level” heuristic. Fields that are more reductionist minimise the scope (or level) or their heuristics to almost make them disappear, while newer, “softer” sciences still rely on broad, high-level heuristics.
Heuristics and also be simply “bad” or “good”. An example of good heuristic is the ancient Greek understanding that a balanced diet, in balancing the humors of the body, would make one healthier. An example of a bad one would be that blood lettings, in balancing the humors of the body would heal one from sickness.
In the present day, we very much rely on heuristics when predicting economic effects using macroeconomic models, even if grounded in microeconomic principles. We rely on heuristics when making prediction about the social effect of policy. We rely on heuristics when making any prediction using a model of which we don’t fully understand the casual grounding of the effects.
To pose an example, we broadly know that inflation can be tackled with a rise in the interest rates, but this knowledge is more akin to the ancient Indian notion that certain diets would tackle the symptoms of diabetes, than to the knowledge that insulin can restore the natural absorption of sugars. We’re not acting at random, and our actions often have positive consequences, but when our heuristic fails, we may cause more damage than good.
For example, lets take austerity policy. According to the models we had in 2008, 1929, and every time it was applied, it was supposed to work, and help us out of the economic crisis we were precipitating into. However, as it later became apparent, that was not the case. We can see similar mistakes in other heuristic systems, such as Ayurveda (there are sub-practices within Ayurveda that prescribe toxic doses of lead).
However, we must be careful not to throw out the baby with the bath water: just as some aspects of Ayurvedic medicine can and are incorporated into western medicine through careful analysis, the same could potentially be done with our heuristic systems in the Social Sciences.
Solutions #
This is where we run into the issue I presented at the beginning: it’s much, much harder to conduct rigorous experimentation in the Social Sciences due to the fact that we’re studying an ever-changing, extremely complex system made of the interactions of ever-changing, extremely complex individuals. In this series, I try to explore different options
There’s already branches of social sciences tackling this issue, such as Behavioural economics, which tries to incorporate more micro-level elements into the study of economics to account for the quirks of human beings as individuals, rather than considering them rational actors. This is trying to move away from the heuristic, and push it to a lower level, where the detail we lose from adopting it is lower.
Another very interesting approach is the prospect of using specialised Large Machine-Learning Models. This has generally been unfeasible, but it’s very much a possibility, seeing the current tooling. I personally believe that an approach utilising Machine Learning, almost by definition a “heuristics machine”, has enormous potential to expand the field into a more well-rounded science, better incorporating its micro and macro-level analysis by either:
- leaning into the heuristics and utilising a large amount of data to train very accurate models, or
- utilising specialised models to analyse broad populations at a depth generally only possible for case studies.
It’s important to note that Machine Learning Models are pretty much black-boxes, so if the aim it to reduce the level of heuristics we’re utilising, the second option is probably preferable, although less straight-forward than the first. So let’s expand on it.
The general idea of it is to utilise the adaptability of a LLM to conduct a more flexible poll, with questions changing and adapting depending on the input from the user. This allows to simulate the depth of Case Studies. It also potentially retains the increased reproducibility that we see when we apply technological solutions to problems, thus making it easily scalable to maintain large sample sizes.
We will be analysing possible solutions in future posts.
Reply by Email