The purely statistical approach, i. e. statistical analysis in conjunction with Big Data, to the study of the human mind and the resulting behaviour is – once again – predominant in psychology, cognitive sciences, economics and related fields. The basic idea is that models that make use of sophisticated statistical analysis in conjunction with large databases allow for the increasingly precise predictions of future events.
The problem is that these models can function without any insight into the underlying principles that may explain the occurrence of these events. Given that a scientific explanation is given if and only if the complex visible phenomena are reduced to the simple invisible principles (call them “Laws” if you like), it seems that the statistical approach violates the idea of a scientific explanation in a fundamental way, because the do not give any insight into the simple invisibles – the underlying principles of nature.
On the other hand, the result of a scientific explanation is supposed to be a scientific theory. The sole criterion for the “value and justification” of a scientific theory, according to Albert Einstein, rests solely on the correspondence between the consequences from the theory and the facts as they are. It seems that the statistical approach – trivially – strives to achieves this goal. The models would, if successful, give the precise way in which events will occur. Alas, without any insight about why the events occur.
There is a simple (and uncharitable) explanation for the predominance of the statistical approach: Human behaviour is undetermined, i. e. there is no set of laws that allow one to deduce what a human will do next. This is a result that follows from basic economic analysis and can be confirmed by everyone. If you love bananas and hate apples and if you are given the choice between a banana and an apple, one has to conclude that you will chose the banana in all cases. Yet, we all know that we could simply change our minds and take the apple instead.
Economists have long recognized that the only way to predict future behaviour – which is the holy grail of economics and politics – can only be achieved by either extrapolate the probability of future behaviour from past behaviour or by straight up manipulation (or by any conceivable combination of these two ways). While the later is taken care of by advertising, public relation, education and so on, the former seems to be what sophisticated statistical analysis in conjunction with Big Data seems to be perfect for.
The question then is: Is there any saving grace for the statistical approach (without the incorporation of actual principles) or are these models simply an overly hyped way for scientists to serve economic and political interests?
I would like to stress that statistical models that include underlying principles are standard in the natural sciences. What makes the above mentioned approach stand out is the rejection of the principles except for the most basic assumption that somehow we extrapolate only from experience and nothing else.
It appears you might have poor experience in using statistics, or rather misuse of statistics. Statistics is never meant to explain the "human mind". In social science, it has been use as an analytical tool to test the underlying proposition or model that an investigator assert could explain the human behavior (an outward manifest of the human mind in response to outside stimulus, i.e. given a certain stimulus in the envirinment, how does human response? If that response follows a certain pattern, we called that pattern a behavior and attempt to find a model to predict that behavior, not the mind). Statistical test is the test of the model, never testing whether statistics can prove the function of the mind, but the behavior as the output of the mind.
The error in your logic comes from the failure to differentiate "form" and "abstract". The data or observation is form, the model used to predict future data is abstract. With existing data or future data, the proposed model can be transform from abstract equation to constructed reality, i.e. predicted value. How accurate this value is could be tested by statistics and comparing the result to the actual data. If the test shows that the proposed model fail, it is the failure of the proposed model (abstract) not a failure in statistics. We cannot blame a field of science or statistics for the inadequacy of the investigator who is not well trained in the subject or a novice in using statistics.
The difference between natural science and social science is that in nature, the law of nature is more predictable than in human behavior- - -a subject of most social science research. You may find unsatisfactory in the failure of predictive models in social science. For instance, despite all research done in economics, we still have economic crisis, financial crisis, etc. How can we not predicted with greater accuracy with the statistical tests and tools that we have at our disposal? It is not the fault of statistics; it is the failure of the model. In natural science, in contrast, statistics works well in testing the prediction of natural phenomenon because all occurences in nature is "form", not abstract. Nature is not an idea, it is fact. Predicting human behavior in social science is an idea, not fact. The actual behavior is fact. When we compare the facts (actual behavior) to the idea (model), sometimes the model fails. This is a failure of the idea, not the tool to test the idea.
Is human behavior predictable? Testable by statistics? To a large extent "yes." When we are hungry, we eat. When we are thirsty, we drink. In these simple cases, it is easy to predict and test. However, in more complex situation, predictive tools becomes less accurate. However, this is not a failure in statistics, this is human failure in not finding the correct model.
Dear Paul Louangrath ,
I should have stated more clearly that the issue is not with statistics per se, but rather with a particular use of statistical models in an explanatory context. To give an example: Consider a branch of contemporary economics, Econometrics. Econometric models are supposed to give functional relations between economic variables (usually in some complicated way that we can ignore here). As any standard textbook will tell you, these models can be used in conjunction with actual explanatory theories, but they can also be used without any theoretical understanding. In this case, they simply approximate relations between variables, while they give no explanation why the relation holds.
Models of this kind are used, for example in language acquisition or cognitive computational neurosciences, where the explicit goal is to actually understand the brain/mind.
You wrote: “The error in your logic comes from the failure to differentiate "form" and "abstract". The data or observation is form, the model used to predict future data is abstract. With existing data or future data, the proposed model can be transform from abstract equation to constructed reality, i.e. predicted value. How accurate this value is could be tested by statistics and comparing the result to the actual data. If the test shows that the proposed model fail, it is the failure of the proposed model (abstract) not a failure in statistics.”
I think this states the problem: What you call “form” is simply observational data. Consider again the model for scientific explanations mentioned in the question. The observational data are the complex visible phenomena. The task of a scientist is to find the actual mechanism that underlies the observational complexity. Kinda like Kepler’s model of the solar system that describes the movement of the planets in terms of a precise and simple mathematical model (Kepler’s Laws). Now, the problem with the purely statistical approach (as in the econometric models without theoretical underpinning) is that they do not yield the mechanism, they only approximate data. Thus, they fulfil the criterion for a successful scientific theory in the sense above in a trivial way, but they fail completely as a scientific explanation (again, it is not about statistics, but rather about scientific understanding).
You wrote: “How can we not predicted with greater accuracy with the statistical tests and tools that we have at our disposal? It is not the fault of statistics; it is the failure of the model. In natural science, in contrast, statistics works well in testing the prediction of natural phenomenon because all occurences in nature is "form", not abstract. Nature is not an idea, it is fact. Predicting human behavior in social science is an idea, not fact.”
Here we have to distinguish between the notion of success in predicting behaviour and the notion of success in understanding behaviour. I strongly disagree with you (as far as I understand your distinction between “form” and “abstract”). The social sciences have to deal with nature as well, because we are organisms and as such part of the natural world. Trying to make a distinction here is futile (although quiet common). I think the main difference is that while there is some understanding of the underlying principles in the natural sciences, there is no understanding in terms of underlying principles with regard to something as complex as human behaviour.
Let me end by pointing out that my main motivation for this question is not social sciences, but rather approaches to understanding of human cognition, for example in neuroscience, linguistics and philosophy. It is my impression that in these areas, ideas resurface that are just a restatement in modern terms of what is known as behaviorism in psychology, structuralism in linguistics or empiricism in philosophy. The point of my question is to see if one can make clear why exactly these ideas fail.
Best,
Sven Beecken
For an indept analysis of some of the issues, see Berwick et al. Poverty of the stimulus revisited.
https://www.ncbi.nlm.nih.gov/pubmed/21824178
For an overview of the goals of cognitive computational neurosciences, see, for example Nikolaus Kriegeskorte & Pamela K. Douglas Cognitive computational neuroscience. https://www.nature.com/articles/s41593-018-0210-5
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Statistical Analysis
- Data Analysis
- Dataset
- Big Data