Winnie, 
the paper doesn't provide any info on whether the predicted cumulative 
mortality after 6 months in the two groups is significantly different or not -- 
although following a discussion of the shapes of the curves it does 
say "Further analysis using a Cox model showed that these differences due to 
bed sharing are statistically significant". I understand this to mean that it is 
the shapes of the estimated cumulative curves in the graph (from 0-6 months) 
that are significantly different -- not the 1 vs 2 per 10,000 end points (and I 
am *assuming* that if the end points were significantly different this would 
have been reported). You'd have to look at the paper to see waht I mean 
about the curves on the graph (Figure 3). (And thanks to Karleen for posting 
the link to it).

But we also need to think about what 'statistical significance' tells us in this 
situation, and consider whether it is an important issue here. If we accept 
p<0.05 as our acceptable level of statistical significance (as many studies do) 
then 'a statistically significant difference' means that there is a 95% 
probablility that the difference between the two things we are comparing is 
a 'real' difference and not a chance one (and by setting our p at <0.05 we are 
saying we are prepared to accept that we might be making a mistake because 
there is a 5% probability of our observed difference not being a 'real' 
difference.)

Unfortunately this tells us nothing about whether the difference between 1 
and 2 deaths per 10,000 is sufficiently clinically, ethically, or practically 
significant to justify advice to breastfeeding mothers regarding bed-sharing 
and SIDS. This is a real issue in the public understanding of science/research. 
If a difference is statistically significant does that mean we are compelled to 
act on it, and if a difference isn't statistically significant does that mean we 
are safe to ignore it? Absolutely not -- it just tells us how confident we can be 
about the existence of a difference -- not whether that difference is 
important. Unfortunately due to the multiple meanings of the term 'significant' 
we all too often interpret 'statistical significance' as denoting 'real-world 
significance'. I have always argued that in this case 'real-world significance' 
has more to do with how a 1 in 10,000 increased risk of SIDS stacks up 
against any benefits lost from not bedsharing and against all those other risks 
we may (blithely or otherwise) take with our infants' lives. 

Regards, Helen

--------------
Helen Ball
Anthropology & Parent-Infant Sleep Lab
Durham University, UK
www.dur.ac.uk/sleep.lab

             ***********************************************

Archives: http://community.lsoft.com/archives/LACTNET.html
Mail all commands to [log in to unmask]
To temporarily stop your subscription: set lactnet nomail
To start it again: set lactnet mail (or [log in to unmask])
To unsubscribe: unsubscribe lactnet or ([log in to unmask])
To reach list owners: [log in to unmask]