I sent a reply directly to the questioner, but as others have replied to 
the list here is what I said (slightly updated to take account of some 
replies).

I am surprised that so far everyone has seemingly accepted the principle of 
presenting CIs as a+/- b. I would argue very strongly against the use of 
+/- here. Apart from the often documented ambiguity when authors give a+/-b 
without indicating what the quantity after +/- is, i feel that for a CI one 
wishes to see explicitly the limits, rather than have to work them out in 
one's head.

The BMJ, Lancet and some other UK journals will not allow +/- usage in 
any/most contexts, although the usage seem ubiquitous in US journals. It is 
very unusual though to see confidence intervals given using +/- in any 
medical journal. I presume that the style is common in journals in other 
fields.

That said, if you quote the limits, the issue of units does not go away, 
although it does remove the ambiguity associated with percentages that 
others have pointed out. We have to decide whether to say that the estimate 
was 20% (95% CI 19.5% to 20.5%) or whether we can remove one or more of 
those % signs (or units).  (I added the '95%' here to show that there is a 
further % to present.)  Although the units for the CI will generally be the 
same as those for the point estimate (percentages are the possible 
exception), I think it makes sense to repeat the units once within the 
brackets.

Here I prefer the separator 'to' to a hyphen because when both limits are 
negative one gets awful expressions such as -6.2 - -2.2.

The British Medical Journal has put its excellent style manual on line at 
http://bmj.com/advice/35.html

The entry for Confidence Interval reads:

>confidence intervals - because of the possible risk of confusion with 
>minus signs use "to," not the hyphen, in confidence intervals (and other 
>ranges that may contain negative values); "...(confidence interval -7.0 to 
>0.9 kPa)..."
>Be vigilant when it could be only one or perhaps more than one interval; 
>don't say "upper confidence interval" (this would be the upper confidence 
>limit)
>Where do the units go?
>60% (45% to 70%)
>45 mg/ml (20 to 50 mg/ml)
>45 (20 to 50) mg/ml
>All these are ok. Percentage and the degree sign are close up to the 
>numbers so go with all numbers. Whether to repeat the unit depends on the 
>length--of the unit itself, and of the material in parentheses:
>45 mg/ml (95% confidence interval 20 to 50 mg/ml)
>95% CI is ok in column headings in table, or in figures; no need to spell 
>out in footnote. But spell out in legends and text

Parts of this are based on a paper I co-wrote many years ago*.

The middle section addresses the question posed about units. the third 
option is in addition to those mentioned earlier. Overall, the answer seems 
to be that it doesn't matter as long as the presentation is not ambiguous, 
but aesthetics may play a part.

Doug Altman

* Gardner MJ, Altman DG. Confidence intervals rather than P values: 
estimation rather than hypothesis testing. British Medical Journal 
1986;292:746–750.


At 13:21 01/05/2002 -0400, Steve Bousquin wrote:
Settle (or join) a debate:  should a confidence interval with units be
written as, for example:

20.0 +/- 0.5%   *OR*   20.0% +/- 0.5%


20.0 +/- 0.5 m  *OR*  20.0 m +/- 0.5 m


Thanks!


Steve

At 13:21 01/05/2002 -0400, Steve Bousquin wrote:
>Settle (or join) a debate:  should a confidence interval with units be
>written as, for example:
>
>20.0 +/- 0.5%   *OR*   20.0% +/- 0.5%
>
>
>20.0 +/- 0.5 m  *OR*  20.0 m +/- 0.5 m
>
>
>Thanks!
>
>
>Steve

_____________________________________________________

Doug Altman
Professor of Statistics in Medicine
Cancer Research UK Medical Statistics Group
Centre for Statistics in Medicine
Institute of Health Sciences
Old Road, Headington
Oxford OX3 7LF, UK

email: [log in to unmask]
Tel: 01865 226799
Fax: 01865 226962
www: http://www.ihs.ox.ac.uk/csm/