February 2021

Sun Mon Tue Wed Thu Fri Sat
  1 2 3 4 5 6
7 8 9 10 11 12 13
14 15 16 17 18 19 20
21 22 23 24 25 26 27
28            
Blog powered by Typepad

« SLU Economics Department Homepage | Main | Relevant and Insightful: Lopez on the Pigou Tax »

Comments

Feed You can follow this conversation by subscribing to the comment feed for this post.

Fred,

Of course the problem with the internet is that we have tons of information, but knowledge is what is needed. To turn information into knowledge we need judgement, and judgement is guided by theory. In the end, it is, as Mises taught us, theory that will enable us to understand. Correlation without causal theories to make sense of them will just remain blurry data patterns. The choice is never between no theory and theory, but always between articulated and defended theory versus inarticulate and undefended theory. Theory is always there -- against a point Mises stressed in Epistemological Problems against the historists --- to think in lanaguage, he argued, is to already think in theory.

Quiz: what great thinker did he rely on to make that point?

Pete

Kant?

Explanatory theories do not fall out of the data, that is an inductive fallacy. The analogy with google is false, google uses search strategies (theories) to find objects that exist. The same applies to the species "discovered" by Venter. That is stamp collecting, not science.

Finding a new or strange object or phenomenon, given the gift of wonder, may provoke thought, the awareness of a problem which is the starting point for serious thought and systematic investigation. But then it is the thoughts that count and the search for significant data - data that makes a difference. At that point the massive data sifting algorithms may be helpful, but only if the preliminary thinking and theorising have been well done.

Science does not advance by discovering objects or theories, it advances by inventing abstract generalisations that explain more and stand up to tests better than the alternatives. Correlation will never be enough for logical reasons, regression models demonstrate that correlations come and go depending on the number of variables in the equation and the form of the model.

Observations are theory-dependent and theories are under-determined by the data. There will always be alternative theories to account for any pattern of data. When you get to the point of testing alternative theories the bulk of data is useless: all the evidence for Newton's theory was also evidence for Einstein's theory. Etc etc.

But we couldn't live without google and the power of calculation means that we can do things before morning tea that used to take rooms full of calculators all week. We just have to be asking the right questions.

On what data did you rely on to formulate that theory, Rafe?

*What data...

An interesting question Erick, and one that appears to be based on the idea that theories can "rely" on data in some inductivist or historicist sense.

I suppose it relies on every instance of advances in scientific knowledge that I know about. Plus an appraisal of various theories of knowledge and methodology.

More to the point, can you offer data that refute the theory. Or even a single datum:)

Correlational analysis cannot show causal relationships - and somehow this is a good thing?

Knowing that A is related to B isn't enough. How do you separate out what is cause and what is effect? Theories allow us to test and determine which causes contribute to which effects.

A stilted example is the old standby about correlations: Number of Churches is proportionally related to the number of drunks in a given city. Why? Because a confounding variable called population is underlying the relationship. Deep data analysis could reveal this - and that would be helpful to someone developing theories as it would help pare down the variables that need to be considered.

For me the massive data analysis seems like something that will support existing science methods and jumpstart investigations in complex areas - not something that replaces or removes the need for theory and experimentation.

The comments to this entry are closed.

Our Books