00:01
How do I account for these two biases, Hawthorne
effect and Rosenthal effect? We account for
them with blinding. Blinding is when some
parties in the experiment don't know who got
the drug versus who got the control group
therapy. There are different kinds of blinding.
00:23
Single-blindedness is when the subjects themselves
don't know whether or not they're getting
the drug or the other therapy, how could they
not know? Well sometimes the therapy is disguised
in such a way to make it indistinguishable
from the control group therapy. Let's say
it's a placebo pill, so you're getting a pill,
that's the new drug and the control group
is getting a placebo, which looks exactly
the same but contains nothing that's medically
effective, you can't tell which one it is.
Obviously if you stop to think about it, there
are some cases in which you cannot single
blind. Can you imagine a case when you're
running an RCT and there's no possible way
to blind the subject to whether or not they're
getting the therapy? I can imagine it right
now. It's possible to do an RCT on the effects
of massage, maybe on pain management, so one
group is going to get a massage and the other
group is going to get what, there's nothing
I can imagine that resembles a massage sufficiently
that you don't know you're not getting a massage,
so some things you cannot blind for. So again
single-blindedness is when the subjects themselves
don't know what therapy they're getting. Double
blindness is when neither the subjects, nor
the investigators know which subjects are
getting what. How did this happen? Well they
seal it off in envelopes. We decided beforehand,
in the randomization process, which group
will be the test group and which group will
be the control group and that's sealed off.
And the investigator does not know until the
seal is broken at the end of the study. Is
there such a thing as triple blinding, what
you think? Can you think of a third party
in this relationship that may be requires
blinding of their own? Well there is, the
data analyst. We have subjects, we have investigators
and we have the people who work with the data.
So modern RCTs sometimes require this third
type of blinding to maintain that everything
is as objective as possible.
02:19
Now let's talk about placebo for a second.
I mentioned that sometimes the control group
will get this thing called a placebo. The
word literally means "I please", because a
placebo is meant to please a person. You take
this, a drug that does not work and maybe
it pleases you. So a placebo by definition
is a simulated or medically ineffective treatment
designed to deceive a patient into thinking
they're getting an actual treatment. Traditionally
it was what was given to all RCT control groups,
but we no longer do that typically, we use
the next standard treatment, as I mentioned.
Now you've probably heard of something called
the placebo effect. Sometimes we call this
the physiological effect produced by placebo.
03:05
Now that particular definition is useful because
it speaks to the idea that there is a physiological
effect involved. Your body is responding to
a placebo thinking that it is actually a therapy.
03:17
There's some conflict amongst the specialists
over whether or not it is a physiological
response. Some people argue it is entirely
a perceptive response that maybe the placebo
effect only works on things like perceived
pain. You can do your own research on this
topic, it is a fascinating field to look into.
Now imagine we have a study that looks like
an RCT, but it lacks the random allocation,
what do you think we call that? Well we call
that quasi-experiment. A quasi-experiment is
an experiment that lacks the randomized nature
of the allocation and very often it lacks
blinding as well. But quasi-experiments are
enormously useful and really quite popular
in international health studies and in program
evaluation studies. So here's how it works,
we have a study population, much like the
randomized controlled trial example. We don't
randomize, but we decide who gets in the treatment
group versus who gets in the control group. Again,
this is not a cohort study, remember a cohort
study did not involve randomization either,
but cohort study the subjects chose where
to go, they chose whether or not they are
going to be exposed to the thing in question
or not. In a quasi-experiment, much like every
experiment, the investigator chooses which
group they're in, not randomly in this
case. Here is an example. Let's say we're
looking at Rwandan communities and Rwandan
communities are beset with a problem with
waterborne diseases and sometimes the tap
water is turbid resulting in a need to boil
the water. People often don't know how to
intensify turbid water or how long
to boil it. So let's say we introduce an educational
radio broadcast in Rwanda to test whether
or not teaching people how to identify and
boil this water can reduce in fewer cases
of waterborne disease. By the way this is
a classic case for a quasi-experimental design,
any sort of media intervention is ready-made
for a quasi-experimental design. This is how
it works. We have our Rwandan communities,
we choose two of them, not randomly, I choose
one that's going to receive my radio broadcast,
I choose another that's not going to receive
the radio broadcast. I wait some time and
I see whether or not there is lower or higher
cases or incidence of waterborne diseases.
Keep in mind there are some things I cannot
control in this scenario. I cannot prevent
people from traveling from one community to
the other. I cannot prevent people from communicating
to each other, say "Hey did you hear about
that radio broadcast?" And I cannot control
for other factors that may be influencing
the rate of waterborne disease, like an NGO
showing up and also giving a private lecture
on how to identify turbid water. Having said
all that, the quasi-experimental design is
an excellent design for program evaluation
or evaluating large-scale community interventions
like this.
06:18
Let's take a moment now to talk about internal
versus external validity, they're both measurements
of what we call validity, which is an important
concept in scientific proof, but they're distinctly
different. Internal validity is the extent
to which the associations or relationships
within your study are real. Did you measure
something that's real. So in other words,
it is the extent to which the causal relationships
are meaningful. Is the approximate truth of
the relationship observed meaningful? The
way that we establish internal validity is
to say, did I control for confounding? Did
I control for any of the other biases that
we've talked about in another lecture on bias?
External validity on the other hand is when
I'm able to generalize from my findings to
the rest of the world, to my reference population.
07:06
An externally valid study has results that
are applicable to everybody else. Okay