[R] Is R good for not-professional-statistician, un-mathematical clinical researchers?

[Jonathan Baron]

A few comments.

1. What students use.

I've gotten a few undergraduate psychology students to use R to
analyze their data for independent study projects.  They do fine.
I prepare the data for them and show them how to do t tests and
correlations, and how to define new variables from old ones.
They write down what I showed them, and they do it.  They would
have to do the same thing with any data analysis program.

Some of these students will go to graduate school.  Then, when
they analyze data, they will do with we advise all our grad
students to do, "Use whatever statistics program your advisor
uses."  In psychology, that is sometimes R/S, but not often.

But it is often SAS or Matlab.  (Perception people seem to use
Matlab for everything.)  SAS, Matlab, and R all require writing
scripts and looking up things in manuals.  I see no difference in
difficulty, and of course I think R is better for most of what
psychologists do.  For grad students, the way to their hearts is
through their advisors.

Most undergrads do the same thing, using what their advisor uses,
and that is usually Systat or SPSS.  (The faculty who use R, SAS,
and Matlab as not as popular as undergrad research advisors as
those who use Systat and SPSS.)  Whatever they use, it will be a
one-time thing, unless they go to grad school, and then it will
be pretty unlikely that they will wind up using the same program.

My bottom line is that teaching R, as opposed to other options,
does very little harm and has very little long-term benefit, but
perhaps more benefit, because ...

2. The bigger issue.

This is all part of what I see as a larger issue concerning how
research is done.  In the 60s, when I got my formal education, it
was normal for bigshot professors to build their own apparatus,
run their own subjects, and analyze their own data.  I watches my
advisor wire relay racks for operant experiments.  When I got to
do it myself, I found it fun, and I still do everything myself
(except pay my subjects, which the business office won't let me
do).

By contrast, today's young researchers have the attitude that
their role in life is to write grant proposals.  When they hit
the jackpot (which often takes many tries), they then hire people
to do the real work.  They hire programmers to develop their
experiments and statisticians to analyze their data.  Granting
agencies seem to encourage this.  I've seen reviews suggesting
that a statistical consultant be added to a proposal because the
researchers proposed doing t tests (which were all that were
needed).  This attitude percolates down to graduate students and
even undergraduates.

The upshot is that fewer and fewer students are learning the nuts
and bolts of research.  (The one essential thing they all learn
is how to use PowerPoint.)  It isn't just statistics.  It is also
computer programming, electrical engineering, and so on.

The students (and faculty) who take to R are the old fashioned
ones, the ones who also write computer programs for their own
experiments, and like it.  The ones who roll up their sleeves.
Perhaps the downturn in the software industry will encourage some
of these students to stay in acadmic fields rather than saying to
themselves, "Why I am in grad school when I could be making 5
times my income in software?"  Indeed, we are now starting to see
grad-school applications from ex-software-engineers (along with
the steady stream from ex-lawyers).

It would be nice if grant reviewers would encourage this sort of
thing, by questioning big staff budgets rather than asking that
they be bigger.  Social science research could benefit from more,
but smaller, grants.

Teaching R is part of this "battle," since it conveys an attitude
as well as specific knowledge.  I guess that is the main reason I
plan to keep trying to do it.

Jon
-- 
Jonathan Baron, Professor of Psychology, University of Pennsylvania
Home page: http://www.sas.upenn.edu/~baron
R search page: http://finzi.psych.upenn.edu/