Here is an interesting excerpt from a book on advanced physics, called 
"Event-Specific Space-Time" by Philip Gibbs (1998).  It is very relevant to 
my earlier posts on the scientific method and critical analysis of 
information and offers a useful introduction to the pitfalls of simplistic 
research.

Mel Siff

--------------------------------------------------

<<WHAT ABOUT CAUSALITY?

I read a news article recently which reported that family conflict can stunt 
the growth of young children. A survey had shown that parents who divorce or 
separate tend to have smaller children. According to the team who conducted 
the study this is scientific evidence of how conditions in childhood can have 
lifelong consequences.

But how right were they? To conduct the survey someone visited schools and 
measured the height of many children with the same age. The results were then 
compared statistically with the circumstances of their parents. Presumably 
they found a statistically significant negative correlation between height 
and indicators of family conflict such as divorce, thus proving the link. 

Fine so far, but can we conclude that the conflict caused children to be 
smaller? Would it not have been equally valid to conclude that having small 
children leads to divorce? The scientist in charge speculated that stress may 
reduce the amount of growth hormone that young children produce.

In fact he applied his prejudices and drew a conclusion which sounds 
reasonable without realising that the converse was also a possible 
explanation of the survey results. It is not difficult to believe his theory 
but there was nothing from the survey which proved it. In fact the real 
reason behind the correlation may have been one or more third factors such as 
wealth. 

Children of poorer families may have worse standards of nutrition resulting 
in slower growth, and lack of money might also lead to higher divorce rates. 
Another cause may have been a genetic trait which shows up in both the growth 
and temperament of family individuals. Such effects are equally likely to 
show up as a correlation in the survey but the news article said nothing 
about such possibilities.

The difference between the possible conclusions from the survey is not just 
one of semantics. People reading the article could blame their frequent 
family rows for having a small child. Such feelings of guilt are unlikely to 
help the situation. They may have been right but I suspect they would have 
been wrong. Surveys such as this are common and are often reported in the 
media by people who do not appreciate the traps that statistics can lead us 
into. 

When responsible scientists wish to establish causal links between different 
effects they are more careful. For example, when a new drug is tested it is 
necessary to know how effective it is and what side effects it may produce. 
To do this a group of volunteers is selected for trials. The group is divided 
in two at random and one half is given the drug. The other half is given a 
placebo pill which is known to have no effect. Nobody taking part knows which 
group they are in. Both groups are then monitored for possible effects. The 
effect is known to be real if it is significantly more noticeable amongst 
those who took the drug than those who took the placebo. It is then certain 
that taking the drug really caused the effect. 

The difference between this example and the survey is that the choice of who 
got the drug and who did not was controlled. In the survey which claimed to 
link height and family strife there was no control over whose parents were 
divorced which were not so it was impossible to distinguish cause from effect 
or rule out other factors with certainty. >>
---------------------------------------------------------------------

Dr Mel C Siff
Denver, USA
[log in to unmask]


%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%