? Why "Burden of proof" ?

[Larry Trask]

On Sat, 7 Aug 1999 ECOLING at aol.com wrote:

> I respectfully disagree with the person who
> recently argued between the following two
> alternatives that either could be an appropriate "null hypothesis"
> (though I am greatly appreciative of that same person's
> many other contributions!)

It was me.

>>    No languages are related.
>> <snip>
>>    All languages are related.

> Rather, the real null hypothesis is something like

> "We do not know whether all languages are related
> (or whether there was polygenesis)" (etc.)

No, I'm afraid not.  This last statement is not a hypothesis, but merely
an observation of the present state of our knowledge.  As such, it is
not subject to test.  A null hypothesis must be something we can test.

A null hypothesis is not a statement of anyone's belief.  It is an
analytical tool set up to permit investigation.  Having set it up, we
then attempt to find sufficient evidence to disconfirm it.  If we can do
that, we have made progress: we know that the null hypothesis is false,
at least in certain cases, and that its contradictory is therefore true
in those cases.

Here's an example from a book I've been reading.  The author, a
statistician and a baseball fan, is interested in finding out whether
the performance of baseball players is affected by the nature of the
ballparks in which they play -- that is, he's interested in what
baseball historians call the `park effect'.  Accordingly, he sets up his
null hypothesis as follows:

	A baseball player's performance is unaffected by the parks
	he plays in.

Now, this is *not* the author's belief.  Quite the contrary: he makes it
clear that he personally believes this hypothesis to be false.  But he
can't achieve anything merely by declaring it to be false.  So,
naturally, what he does is to state it, and then to go on to assemble
evidence against it.  As it happens, he is able to assemble so much
evidence against it that he feels safe in concluding that it is false,
and that its contradictory must be true.  That's how things work.

Same in comparative linguistics.  If we want or hope to establish that
certain languages are related, we must take as our null hypothesis the
statement I set out earlier:

	No languages are related.

By assembling persuasive evidence against this hypothesis in particular
cases, we succeed in disconfirming it for those cases, and therefore in
establishing that certain languages *are* related.  For example, in the
case of the languages we call the Germanic languages, so much evidence
can be adduced against the null hypothesis that it becomes untenable for
those languages, and its contradictory must be accepted: the Germanic
languages really are related.

In general, the appropriate null hypothesis is the hypothesis you hope
to disconfirm, not the one you hope to confirm.  Suppose we adopted
instead the opposite null hypothesis:

	All languages are related.

To investigate this, we would have to seek to disconfirm it by
assembling persuasive evidence against it, at least in some cases.
And that, as I pointed out earlier, is an impossibility: we cannot prove
that *any* languages are absolutely unrelated.  Hence no progress is
possible.

To reiterate: a null hypothesis is not a statement of belief.  In fact,
the null hypothesis is usually the very opposite of what we hope or
suspect is the truth, the conclusion we would like to establish.

As both my baseball example and my linguistic example suggest, a
suitable null hypothesis is normally stated in the negative: it declares
that there is *no* relation between the objects of inquiry.  This is so
because we are normally interested in showing that these things *are*
related.

Even when we really do want to show that things are unrelated, the null
hypothesis is still stated in the negative.  Suppose I want to show that
there is no correlation between star-signs and personality.  The
appropriate null hypothesis is the negative:

	Star-signs are unrelated to personality.

I will then go on to show -- I hope -- that no evidence can be assembled
that disconfirms the null hypothesis, and hence that there is no reason
to reject the null hypothesis this time.

> Any attempt to force anyone else to accept something
> that embodies a CLAIM (as both of the first two alternatives
> above do), is MANIPULATING the discourse,
> instead of dealing with facts.

No, not at all.  Stating a null hypothesis is *not* an attempt to force
anyone to believe it.

> If we don't know, then we don't know, it's as simple as that.

Well, yes, but, if we want to escape from our ignorance, we must go
about our investigations in an orderly way.

> I personally don't give a hoot what anyone wants
> to "assume", or tell me to assume, in the absence of
> data justifying such an assumption.

Again: a null hypothesis is neither a statement of belief nor an attempt
at intellectual intimidation.  It is merely an analytical tool.

> Using a "burden of proof" argument is merely
> a way of trying to get someone to accept a conclusion
> in the absence of evidence.

Hardly.  The burden of proof is always on the person who wants to defend
the contradictory of the appropriate null hypothesis.  If I hope to
persuade my colleagues that Basque and Burushaski are related, then the
null hypothesis is that they are *not* related, and it is up to me to
assemble enough evidence to disconfirm the null hypothesis in the eyes
of my colleagues.  That's how comparative linguistics works.

Larry Trask
COGS
University of Sussex
Brighton BN1 9QH
UK

larryt at cogs.susx.ac.uk