PHL101: Critical Thinking, Theory and Scientific Theory
A Simplistic view of Scientific Method
- Along with encountering claims made on the basis of observation
or as the conclusions of arguments, we also often encounter claims that
are the products of theories. To evaluate these claims, we need some
sense on how to evaluate theories.
- There are many kinds of things that we call theories. Roughly,
a theory is a collection of claims that we use to describe or explain
some kind of phenomena. A theory typically has as a consequence many
other claims; these are the products or predictions of the theory.
- This general notion of theory is sufficient to describe things as
different as philosophical theories, theological theories, scientific
theories, and mathematical theories. However, there is little agreement
outside of mathematics and science what counts for a good theory.
- We shall limit our discussion then to one kind of theory:
scientific theory. These can be evaluated with criteria that are
clear and relatively uncontroversial. More importantly,
scientific theories are very important to our everyday lives. They provide
our technology and are extremely successful in describing our world.
In scientific method, we sometimes use inductive generalization to
identify causal correlations. One way we can describe the process is:
Generating Scientific Theories
- The basic view:
- We observe (in our sample) that events of kind Q
follows events of kind P;
- We generalize that Qs always follow Ps (that is,
Qs follow Ps throughout our population, which is all
events, including future events);
- We say that Ps cause Qs.
- There are two problems with the basic view:
- Science posits that there is something underlying
the correlation of two events which allows us to say
that one causes another: the working of natural laws.
In other words, not all correlations are taken to be
- It is never clear which observations should be
made until we have a hypothesis about what
correlations we expect (that is, until we have a
hypothesis about natural laws).
Where do we get theories from? There is no set method for creating
philosophical theories or mathematical theories. There is one for
creating scientific theories, however. This is called the
Deductive Nomological Theory of Science, with Falsificationism.
The underlying idea of the method is:
- We develop a hypothesis that there are certain natural
laws or certain effects of natural laws. These hypotheses
must be falsifiable: they must entail predictions which are
in principle testable and could be false.
- We predict observations we can make if the hypothesis is true.
- We check to see if these predictions come true; if they
do, we know that hypothesis may be true and may continue to
believe or use that hypothesis; if the predictions prove
false, we reject the hypothesis.
- We continue to try to falsify the view (that is, we continue
to make and test predictions).
A schematic example:
The "deductive" part here is the inference from hypothesis to
consequences. The "nomological" part is the the presumption that we
are aiming to discover laws ("nomological" means reasoning about
- Suppose H is a hypothesis. H entails predictions
P1 ... Pn that are testable.
- We look to see whether P1 occurs.
- If this predicted phenomenon P1 is
observed to occur, we continue to hold H as a
hypothesis. (We should now also continue to test
whether any other of the predictions P2
... Pn are observed.)
- If any of the predicted consequences is observed
false, we reject the hypothesis H.
Some important corollaries of the deductive nomological theory
Why does it matter that a scientific theory must be "falsifiable"?
- We cannot prove a scientific theory, we can only
show it is better than any of the offered alternatives
(by the criteria of the method and those listed below
for comparing theories).
- Science is creative: you must come up with
hypotheses before you can apply scientific method.
- A scientific theory (that is, a theory about the
physical world, not a logical theory) which is not
falsifiable is a bad theory!
- Many theories of the same phenomenon can pass the
test of the deductive nomological method with
falsificationism. We therefore need a method of
choosing one scientific theory over another.
- What kind of claim is not falsifiable? Examples include:
Most are inclined to believe that what distinguishes these examples
is that they are true either in terms of meaning (analytic) or are
consequences of other premises (logical claims).
- All bachelors are unmarried males.
- Unfalsifiable claims are thus usually presumptions, perhaps
vacuous, that are dressed up like substantive claims. Typically
even the person making the unfalsifiable claim is deceived by their
own terminology and misses the ambiguity.
- Example: economists often claim all motivations are selfish.
Confronted with the observation that some people appear to do
altruistic things, they say these people desire to do altruistic
things and so are selfish. But note: "selfish" on this kind
of view means nothing more than something like, does what one
wants to do. That is fine -- economists should be allowed to define
any terminology they want -- but it is a different meaning
than we usually use for "selfish." Furthermore, to say people
are "selfish" on this view is not to make a claim -- it is to state
an assumption that they make in all their reasoning!
- There is in the notion of unfalsifiability a deep
implicit fact about the natural world. Logical facts are
necessary facts, it appears, but natural facts about the
physical world appear to be different kinds of facts. They
are contingent: there is some consistent description of the
world in which a natural fact is other than it is. For
example, it is true that Earth has one moon, but one could
consistently describe a world in which Earth has two moons.
Compare this to 2+2=4. You cannot consistently describe a way
the world could be in which 2+2=5; you would contradict
yourself. The mathematical fact is necessary; the physical
fact is not.
- To reiterate and stress this point: The most
pernicious aspect of unfalsifiable claims is that they are
usually either mathematical/logical claims, or stipulated (as
in the case of definitions) claims. If someone makes an
unfalsifiable claim with the presumption that it is a
scientific claim about the natural world, they are making a
mistake of confusing one thing for another: they are confusing
definitions or logic for the natural world. They typically do
not realize that they are making this mistake.
- Unfalsifiable claims may sometimes also arise because
a claim is so vague, or because someone allows continual revision
of the standards of testing the claim.
Choosing between theories
The Deductive Nomological Method with Falsificationism does not
guarantee that we will have only one theory. (We use "theory" to mean
at least a collection of one or more hypotheses.) There may be very
many different and contradictory hypotheses that pass the method as
explanations of some one phenomenon. To choose between theories, we
use three ranked criteria:
These are ranked in terms of importance, as noted. We will give up
consistency for predictive power, and give up on simplicity for
- Predictive power. Theory A has more predictive power
than theory B if A predicts everything that B predicts and
- Consistency with existing scientific theory. If theory
A and B have similar predictive power, but A is consistent
with other theories we believe and B is not consistent with
those theories, we will prefer A.
- Simplicity. If theory A and theory B have similar
predictive power, and they are both consistent with theories
that we already believe, but theory A is simpler than theory
B, we will prefer theory A.
What about "Nomological"? A note about Induction
- Scientific theory can be interpreted as positing that we
learn through induction facts about the world. Inductive
reasoning when we conclude from many observations that a
certain correlation has occurred to the claim that in future
observations that correlation will occur.
- In the case of the deductive nomological method with
falsificationism, this means that as we observe and confirm
many of the predictions of our hypothesis, we conclude
that we should continue to believe H, which in turn means
that we believe that other (e.g., future) predictions of
H are also going to be observed.
- Induction has been very controversial in philosophy.
"The Problem of Induction" asks, why assume the future
will be like the past?
- The talk about past and future here is in fact
misleading. Induction is really just an instance of the kind
of generalizations from observations we discussed in our
first section. That is, it is reasoning from a sampe of
observations to a population of observations. Of course,
we checked our sample (otherwise it would not be our sample)
so that occurred in the past, and we generalize to other
observations that we have not made, so those are in the
future, but time is not really relevant.
- Thus, induction assumes not that the future is like
the past but that statistical generalizations can be made
in a reliable way.
- What is special is that our population of potential
observations in science is often not well defined or bounded
in a clear way. We are thus often forced to assume that
the correlations we are studying converge on one
- Note that if we are wrong in our inductive reasoning, and
our past observations were not representative, this is
presumably because there was some kind of bias in our earlier
sample. So, it is not a refutation of induction to argue that
we may mistakenly sometimes see patterns that later prove not
to be there; we can make sense of this possibility by
observing that samples are not always representative. Lacking
a pattern is itself a kind of thing that can be represented or
misrepresented by a sample.
The Duhem Thesis
Pierre Duhem recognized that sometimes we discover things that are
inconsistent with our hypotheses but we are inclined to keep our
hypotheses. This is because, he realized, we never really test a
single hypothesis, but rather a number of them as a group. We assume,
for example, that our measurement instruments are working well, that
we have not made a mistake in our mathematics, and so on. But it then
follows that when a prediction is made, and then found false, and one
of these other assumptions could be false.
A more sophisticated version of the deductive nomological method with
falsificationism thus looks something like this:
- Suppose H1 is our new hypothesis.
H1 and a range of additional hypotheses
H2, H3, H4...
Hn entail predictions P1
... Pn that are testable.
- We look to see whether P1 occurs.
- If these predicted phenomenon is observed to be
true, we continue to hold H1 as a
hypothesis and move on to test other predictions.
- If any of the predicted consequences is observed
false, we reject the one or more of H1...
Claims Derived from Theory