For several years null hypothesis testing (NHT) has been the dominant
form of statistical analysis in psychology. It has also been subject to
periodic criticisms from within the field of psychology. In the past decade
these occasional criticisms have turned into a more or less steady stream
of criticisms and defences of NHT resulting in several journal articles,
at least one edited book, and the convening of an American Psychological
Association Task Force.
The controversy stems from the practice in psychology of treating the
special case of a no effect (nil) hypothesis as the entirety of NHT. This
leads to a variety of justifiably criticized problems with the use of
NHT, which have lead some to call for an outright ban on NHT in psychology.
The APA Task Force recommendations have lead to some changes in undergraduate
and graduate textbooks, (e.g., new/expanded discussions of power and effect
size). These recommendations certainly move psychology toward a more appropriate
use of statistical procedures generally, but they do not solve the underlying
problem of inappropriate NHT.
The solution lies neither in banning NHT nor in relying solely on alternative
procedures, but in "reforming" NHT, replacing a-theoretical
null hypotheses with theoretically meaningful hypotheses. Such a reform
requires that training emphasize parameter estimation and the testing
of theoretical models, an approach that exists in some areas of psychology
and appears to be common in other sciences.
Psychology students are typically introduced to NHT in a parameter estimation
context, but in moving to research designs with two or more conditions
they are frequently taught that the null hypothesis is always a test of
no effect. The notion of parameter estimation is lost, often forever.
A few will be reintroduced to parameter estimation as part of a specialized
data-modelling course in graduate school.
I argue that a greater emphasis on parameter estimation and the testing
of models is the correct reform for NHT in psychology. Such an emphasis
will ensure that the statistical hypothesis being tested matches the substantive
hypothesis of interest. I will discuss the changes that are occurring
in psychology and propose further changes that are still needed.