Dr. Stuart WG Derbyshire wrote:
>We can't expect all users of SPM to be experts. Several contributors
>have argued that the researchers, editors, reviewers and readers all
>need to take responsibility for ensuring the methods are
>appropriate. Indeed. But those who provide the stats tools also have
>a responsibility to ensure that the tools are user friendly. If
>diagnostics are necessary (to look at residuals, stds, unthresholded
>maps, whatever else is considered important), but are not easily
>accessible or understandable, then it is likely they will not be
>used or will be used inappropriately.
>
>Similarly, where there is a lot of argument it is difficult for an
>outside user to decide what to do. Somehow we have to provide
>reasonable guidance. I think Tom's internet page describing what he
>would like to see in any imaging paper is good (although it is one
>of my open browsers and I haven't read it in detail yet!). I might
>add that perhaps we should have a list of things all the experts can
>agree a researcher should absolutely do or not do (or risk immediate
>rejection) and then things a reviewer might reasonably look for but
>which are debateable.
>
>
Okay, I snipped out Stuart's disclaimers, so if it sounds
like he's making strong statements, it's probably my
fault.
Now I'm going to make some strong statements:
I agree that the software should be easy to use
and necessary functions should be accessible.
HOWEVER, I think that you are putting the burden of
the science onto the people who are making the
tools. IMO, the researcher should be coming to
the software knowing what he or she wants to do,
including understanding critical issues like
hypothesis rejection, and what assumptions they
are making to do their analysis. Those subjects
are NOT the responsibility of the programmers,
although perhaps programmers should make themselves
more clear that they are only providing tools.
The guy who makes the hammer is not the one who
builds the fine furniture, and the person making
the furniture should not rely on the hammer maker
to tell them when the hammer is the wrong tool,
for example, that a hammer should not be used
to drive a screw.
Does the scientist KNOW that he is assuming that
the residuals are gaussian and iid in time? He
certainly should, and should know what effects,
if any, there are on his analysis since that
assumption is known to be false.
The software can do a lot of the work for you,
but it shouldn't do the planning, understanding,
or interpretation, despite rumors of a new
SPM plugin that writes your methods and results
sections and submits your paper to three journals
automatically.
--
rich hammett SSCC/DIRP/NIMH
National Institutes of Health
Building 10, Room 1D80
10 Center Drive, MSC1148
Bethesda, MD 20892-1148
ph 301-402-8416 fax 301-402-1370
|