Your statistician is correct.
Why report t to such a degree of accuracy? No one cares what t is, or
knows what it's for. It's used to calculate p. But you don't report
p to any accuracy. This is a hangover from the days when getting the
exact p was hard, but it's not hard now.
You're interested in the difference between two groups, so report
that. And report the CIs, 'cos that's the difference that you're
interested in , the difference in the population.
Imagine you went to the doctor and the doctor said that you should
take some pills, 'cos they would probably make you live longer. How
much longer? You might ask. Well, the doctor says, according to this
recent paper, t = 2.78.
What you want to know is, the difference it's likely to make, and the
likely range of that difference, and maybe the p-value too.
So I would write it as Group A were 5 points higher than Group B (95%
CIs = 3, 7; p = 0.02; d = 0.8). (d is Cohen's d, the standardised
effect size - that's less necessary).
<rant on> The problem with psychologists is that they learn their
statistics from other psychologists, not from statisticians. Fisher
pretty much invented hypothesis testing (as we use it), but his book
was only read by one psychologist - Guilford, who wrote a book about
it. Other psychologists then read Guilford, and wrote books, and so
on. But Guilford made mistakes, and it's taken 50 (or so) years for
those mistakes to get fixed.
There are lots of weird historical things like this - we do things
because that's the way we were taught to do them, and because the
people who taught us were taught to do them, but they're wrong. <rant
off>
Jeremy (A psychologist, but I hang out with statisticians quite a lot).
On 22/03/07, David Hambrook <[log in to unmask]> wrote:
>
>
>
> Hello All,
> This may sound like a very silly and basic question but I'm going to ask
> anyway... I'm in the initial stages of writing up my Masters dissertation
> and I have been told by a statistician at my university that when reporting
> the results of t-tests I should include 95% confidence intervals (CIs). I
> have never had to do this when reporting t-tests in the past. The format I
> would usually use fro reporting is t = 2.578, df = 19, p < 0.05 (for
> example). Have I been wrong all these years??? Any advice would be much
> appreciated.
> Regards,
> Dave.
>
>
> David Hambrook
>
>
>
> MSc Student
>
> Mental Health Studies Programme,
>
> Institute of Psychiatry,
>
> Division of Psychological Medicine and Psychiatry,
>
> Kings College London (Guy's Campus),
>
> 9th Floor Capital House,
>
> 42 Weston Street,
>
> London SE1 3QD
>
> Mob: 0779 5025 868
>
> [log in to unmask] ; [log in to unmask]
>
>
> ________________________________
> The all-new Yahoo! Mail goes wherever you go - free your email address from
> your Internet provider.
--
Jeremy Miles
Learning statistics blog: www.jeremymiles.co.uk/learningstats
|