Quiz 4
An observational study done at several large hospitals found that women who
had ultrasound gave birth to babies with lower weight, on average, than women
who did not have ultrasound. Because of this, the researchers advised
discontinuing ultrasound because it was unsafe. Do the data support this
conclusion?
About half of you answered this incorrectly. So study these types of
questions closely.
Solution:
The key here are the words "observational study". The researchers are
making a cause-and-effect conclusion: if we give ultrasound, the baby's
birthweight will be reduced. If we don't give ultrasound, it won't.
You can NEVER make such conclusions based on an observational study.
So you get 2 out of 5 points for saying "no".
The reason? Because there could be confounding variables that we didn't
account for. If that's all you said, then give yourself 1 point.
If you gave a plausible confounding factor as an example, give yourself 3
points. If your proposed confounding factor was a little vague or maybe
not entirely convincing, give yourself somewhere between 1 and 3 points.
An example of a good and plausible confounding factor: mothers whose
babies were at risk were more likely to be given ultrasound in order to examine
the health of their fetus. The weight of these babies was lower because,
on the whole, they were less healthy to begin with. Note that this
confounding factor both explains the assignment of values of the treatment
variables (unhealthy babies more likely to get ultrasound) and the response
variable (unhealthy babies more likely to weigh less).
Let's dissect this a bit. The treatment variable here is "ultrasound", and
the factors are whether or not the mother received it. So the "Treatment
group" consists those mothers who got ultrasound, and the "control group"
are those who did not. The response variable is the baby's birthweight. In
this case, we observed that the birthweight was lower, on average, among
the treatment group.
For a controlled experiment, we would have to assign mothers to one group
or another. In this case, mothers (or their doctors) chose which group
they went into based on a variety of factors that we can only guess at. For
this reason, the control and treatment groups could differ in a number of
ways, other than the ultrasound. And therefore we can't conclude that
the ultrasound caused the dip in birthweight.
Some answers that would have received partial credit. (1) "The mothers'
eating habits might have come into play." Well, yes. That explains
why some babies had lower weights than others. But it doesn't explain
why those babies were more likely to get ultrasound. This explanation would
get 0 out of the 3 remaining points. To be a confounding factor, it
must explain both the treatment and the response. (2) "There could
be a personality trait that makes people want ultrasound and leads to a lower
weight baby." Well, yes, that would explain it. But this isn't
a terribly compelling explanation because its hard to imagine what that personality
trait might be. So this explanation would get maybe 2 out of the remaining
3 points.
P.S. Later, a controlled, randomized study was done and found a protective
effect of ultra-sounds.
Practice
If you're unsure of your answers, email them (along with the problem statement)
to me or your TAs.
1. A study of young children found that those with more body fat tended to
have more "controlling" mothers; th eSan Franciso Chronicle (November 9,
1994) concluded that "Parents of Fat Kids Should Lighten Up."
a) Was this an observational study or a controlled experiment?
b) Did the study find an association between mother's behavior and her child's
level of body fat?
c) Do the data support the Chronicle's advice on child-rearing? Explain.
2. (This one's tricky. You probably won't guess the confounding variable.)
Over the period 1964-1970, the suicide rate in England fell by about one-third.
During this period, a volunteer welfar organization called "The Samaritans"
was expanding rapidly. One investigator thought that the Samaritans
were responsible for the declien in suicides. He did an observational
study to prove it. This study was based on 15 pairs of towns. To
control for confounding, the towns in a pair were matched on the variables
regarded as important. One town in each pair had a branch of the Samaritans;
the other did not. On the whole, the towns with the Samaritans had
lower suicide rates. So the Samaritans prevented suicides. Or
did they?
Discussion of 2.