David F. Hendry 9Mitchell (1946) allowed the assertion of “measurement without theory” to become
capable of dismissing empirical work without further serious consideration. The
vigorous reply by Vining (1949a, 1949b) still merits reading. With a few honorable
exceptions (such as Atkinson, 2005), even the use of the word “measurement” as a
title for economics’ papers seems to have decreased since (other than in “measure-
ment errors”). Tress (1959) offers a near contemporary analysis of the acrimony
between economics and econometrics at that time, and a possible reconciliation.
Keynes (1939) had asserted that a long list of “preconditions” had to be satisfied
to validate empirical inferences, implicitly arguing that empirical econometrics
must fail unless everything was known in advance (see, e.g., Hendry, 1980). But if
it was impossible to empirically uncover things not already known theoretically,
then no science could have progressed; rather, scholasticism would still rule. There
are several flaws in Keynes’ claims, of which three are the most important.
First, “partial knowledge” can be valuable, and can be learnt from evidence, with
or without prior theories, albeit being subject to revision later. Our understanding
of gravity remains incomplete, but has advanced greatly since Aristotle’s early view
of objects’ natural places (smoke rises, stones fall, as natural to go to heaven or
the centre of the earth, respectively), through Ptolemy’s epicycle theory of plane-
tary motions, Descartes’ vortex theory, and Newtonian inverse-square laws, which
Adam Smith (1795) presciently noted was just a model and not the “truth” (as
most of his contemporaries assumed). Einstein’s relativity theory is still not the
“final answer.” Retrospectively, Aristotle’s theory did not go beyond “explaining”
the phenomena themselves, whereas Newtonian theory, while closely based on
Kepler’s laws of motion of planetary bodies, explained many additional aspects, so
was a clear advance in knowledge, even if later it too was found to be incomplete,
and at relativistic speeds, incorrect. Moreover, despite neither Aristotle’s nor New-
ton’s theories being “true,” both were at least consistent with the observed facts
of their time. The relevant empirical regularities persisted through many theories,
which provided better explanations, often with unanticipated predictions of new
phenomena – genuine “mutual penetration.” Thus, progress is the key to science,
not one-off forging of true laws that hold forever.
Second, if there are invariant features of reality – as in physics and chemistry –
then empirical research can discover them without prior knowledge, as happened
historically in many branches of science. Conversely, if nothing is invariant (an
extreme of Heraclitus of Ephesus supposed view that “reality is change”), neither
economic theories nor econometric models would be of any practical value (see
Hendry, 1995b). Following Bachelier (1900), equity prices have long been viewed
as close to random walks, which may be thought to entail the absence of any
invariants, but if correct – as suggested by some modern theories and empirical
tests of efficient markets – is actually a powerful invariant, as contrasted with a
data generating process whose structure alters every period.
Third, empirical econometrics could still “advance” by rejecting economic the-
ories. This would at least allow economists to focus on theories that were not
yet rejected, if any, and improve those that faced discordant evidence. However,