So what I'm saying is that a theory that cannot be falsified experimentally is not very useful.
The real danger here is that these theories -most of which are supported by no or very slim experimental evidence- are used by governments to formulate economic policies, which in turn affect the life of millions or billions of people. For me, this is a bit like developing an "idealized" theory of nuclear physics and then trying to build a nuclear power-plant based on it. I, personally, would not want to live near one of these ;)
Suppose governments did not formulate economic policies with these imperfect models. What would they do instead?
Edit: this isn't a comment about economics but how economics is used in a political context.
This is what the real science is. Observation and only observation. Reasoning is secondary and unimportant.
Mere observations give no interesting and no useful results. Basically, you'll end up with an ever-growing list of facts.
Science, essentially, is nothing but a process of extracting information (as in the algorithmic information theory, not in Shannon sense) from the data. Reasoning is just a tool, one of the ways of representing this information.
In order to test a model, it is useful to collect new data.
Observation is useful to develop a theory (induction, or "educated guessing"). However, having a theory guides us towards newer, often subtler observations as we attempt to falsify it. Karl Popper called this the theory-ladeness of observation [1]
Of course. But it's merely some heuristics helping to cut down the search space, not an inherent, fundamental property of the universe.
Traditionally, economists have put the facts in a subordinate role and theory in the driver’s seat. Plausible-sounding theories are believed to be true unless proven false, while empirical facts are often dismissed
The error is the assumption that facts are, well, facts. In real life, it's not nearly so black-and-white. Remember back in high school physics, when we did experiments with Newtonian physics, but our answers never quite worked with the theory - not just because of Einstein, but mostly because of (a) measurement error and (b) we were (knowingly or not) assuming point-masses on frictionless planes, etc.
As long as we were willing to write off our errors against those two sources, we could never discover that Newton's theories were incomplete, and we needed relativity to explain differences. We only got to that conclusion via models.
The way that you choose to measure fundamentally drives what your measurements are. And you can only decide what and how to measure if you start off with some model of the world.
So, the corrected wording for that paragraph should be the following:
Plausible-sounding theories are believed to be true as long as their numeric predictions conform to all of the experimental facts within this theory supposed range, up to this theory supposed precision.
And a theory falsified by a new set of facts does not become immediately "false". It just get its range cut down to that range of facts where it is still giving an acceptable prediction precision.
Back to physics analogy, the phlogiston theory is still correct and true, as long as it is applied within the range of the phenomenological thermodynamics. But this does not happen in the classic economics - all those crazy theories are coming from the depths of an insane mind, not from any particular set of facts. They never had any specific range where they were known to be applicable. Therefore, phlogiston is science, and economic theory is not.
I mean, it is possible to measure the effect. But you need to first design the experiment in order to make the measurement. It was necessary for Einstein to sit and think for a long time, to come up with the idea, which others were able to design experiments around.
You can't just say that somebody would have noticed the discrepancy sooner or later. Really, doing experimentation that way is a statistical error, of the sort that's really plaguing the sciences these days.
General relativity is an application of a well known extrapolation heuristics - the simplest possible generalisation of a model. The other possible generalisations were much more complex, and the nature for some reason tend to favour simplicity, so it was reasonable to go that way (as well as it is very reasonable now to go out and test all the possible extensions of the Standard Model, starting with the simplest ones).
"Observation and only observation" is how you build an inchoate collection of useless factoids, it's not how you do science.
Observations are a very important part of the scientific method and you have to be either rooted or anchored in it, but it's only one part of science, and it's by no means the most important product of it. Observations are seeds (from which patterns are recognised and hypotheses formed) and filters (data gathered to test the predictive power of hypotheses).
We call it phenomenology. The very base of any proper science. The rest is just a mechanical work of eliminating the noise from this data.
Science should never produce any knowledge which is not found in the original data (either passive observations or controlled experiments - does not really matter). The scientific method is all about techniques of extracting the most compact representation (i.e., the information) hidden inside this massive data sets. It turns out that the axiomatic theories are very common and compact representations for many areas of science, but not necessarily the universal representation.
What you disregard as 'just a mechanical work' is nothing of the sort. It has required all the ingenuity of the greatest genius in history to get modern science out of the data, and any future advances will require even greater insights.
If anything, the data gathering is the 'mechanical work' part of science.
But yet, this process does not produce any new knowledge on its own. All the knowledge is there, in the data. Waiting to be extracted.
Then, some decades later someone finds the practical application, and the data that matches the model.