So having said that, the last kind of study
design we're going to talk about this lecture
is the ecological study. It isn't really an
experiment. It's probably a kind of observational
study as well, but very often it's spoken
about in the same breath as an experiment.
The defining characteristic of an ecological
study is that the unit of analysis is a population
or a group and not an individual. Here is an
example, populations with high proportions
of pets are also populations with low frequencies
of asthma. Think about that for a second.
That will lead you to probably surmise that
owning a pet would reduce your chances of
having asthma. The problem is you can't really
make that conclusion. You can't say that an
individual living in one of these towns with
high amounts of pets is likely to not have
asthma. We can't make predictions about individuals
based upon these population averages. This
is what we call the ecological fallacy. An
ecological fallacy is when we infer something
about an individual based upon observations
about a group, so what we learn from an ecological
study cannot be used to make conclusions about
individual behavior. I would say the ecological
fallacy is the foundation of racism. You observe
behavior of a group and you assume the individuals
from that group must also have the same characteristics
or behaviors. Another example, it has
been observed that the greater the proportion
of immigrants in a given US state, the lower
the average literacy level of that state.
You might conclude then that immigrants have
a lower level of literacy. It's not true.
Individual immigrant residents are more literate
than the native residents, what's going on?
A number of scenarios, I encourage you to
investigate further if this interests you,
but the important observation here is that
observations made about a population cannot
be extrapolated to individual characteristics
from that population.
So let's go over what we've learned. What
distinguishes an experiment from an observational
study, do you remember? Well in an experiment
the researcher, you perhaps, manipulates a
variable in the study, such as deciding who
gets the intervention and who gets the placebo.
Maybe that decision is made by the flip of
a coin, but you were involved in doing the
flipping at the very least. What's considered
the gold standard of evidence? The RCT is,
because of its controlled nature. We can control
extraneous variables and control confounding
quite well, so it has a high degree of internal
validity, hence a high degree of causal proof.
Why is random allocation important? Well because
it allows for the equal distribution of biasing
factors, not just the ones you know about,
especially the ones you don't know about. And again
that improves internal validity. Why is blinding
important? Well blinding is important to control
for certain biases, especially the Rosenthal
and Hawthorne effects. And how is a quasi-experiment
different from an RCT? It lacks randomization,
that's the 'R' part. And how is a natural
experiment different from an RCT? Do you remember?
Allocation in a natural experiment is determined
by an external force, like a natural disaster
or a law of some kind. And what's the distinguishing
trait of an ecological study?