Randomized Trials: The Ideal Weapon
-
0:00 - 0:05- [Narrator] The path from cause
to effect is dark and dangerous, -
0:06 - 0:09but the weapons
of Econometrics are strong. -
0:10 - 0:15Behold the most powerful,
the sword of random assignment, -
0:15 - 0:18which cuts to the heart
of causal questions. -
0:23 - 0:28We begin with our most powerful
and costliest weapon -- -
0:28 - 0:30randomized trials.
-
0:30 - 0:31- [Student] Awesome.
-
0:31 - 0:35- Every metric's mission begins
with a causal question. -
0:35 - 0:38Clear questions lead
to clear answers. -
0:39 - 0:43The clearest answers come
from randomized trials. -
0:44 - 0:48Let's see how and why
randomized trials provide -
0:48 - 0:51especially convincing answers
to causal questions. -
0:52 - 0:56- [Josh] Like a finely honed sword,
randomized trials cut -
0:56 - 0:58to the heart of a causal problem,
-
0:58 - 1:01creating convincing apples-
to-apples comparisons. -
1:01 - 1:04Yet like any finely made weapon,
-
1:04 - 1:07randomized trials are expensive
and cannot be done quickly. -
1:08 - 1:12- Randomized trials originate
in medical research -
1:12 - 1:16where they're called
randomized clinical trials or RCTs. -
1:17 - 1:21The U.S. Food and Drug Administration
requires drug manufacturers -
1:21 - 1:25to establish the safety
and efficacy of new drugs -
1:25 - 1:26or medical treatments.
-
1:27 - 1:30They do this through a series
of RCTs. -
1:30 - 1:35That's why randomized trials are
said to measure treatment effects. -
1:36 - 1:38You may have contributed
to another kind -
1:38 - 1:40of randomized trial --
-
1:40 - 1:43the A/B tests
Silicon Valley companies use -
1:43 - 1:46to compare marketing strategies.
-
1:46 - 1:50For example, Amazon randomizes
search results in a constant stream -
1:50 - 1:52of hidden experiments.
-
1:52 - 1:53- [Woman] Oh.
- [Man] Interesting. -
1:53 - 1:57- Randomized trials are also
important in education research. -
1:57 - 2:00They've been used to answer
a causal question near and dear -
2:00 - 2:02to my beating teacher's heart --
-
2:02 - 2:05should laptops
and other electronic devices -
2:05 - 2:07be allowed in the classroom?
-
2:07 - 2:10Many see classroom electronics
as a learning aid. -
2:10 - 2:14But others, like me, think they're
a damaging distraction. -
2:14 - 2:15Who's right?
-
2:22 - 2:24Metrics masters teaching
at West Point, -
2:24 - 2:27the military college
that trains American Army officers, -
2:27 - 2:31designed a randomized trial
to answer this question. -
2:31 - 2:33These masters randomly assign
West Point cadets -
2:33 - 2:36into economics classes
operating under different rules. -
2:37 - 2:39Unlike most American colleges,
-
2:39 - 2:42the West Point default
is no electronics. -
2:43 - 2:46For purposes of this experiment,
some students were left -
2:46 - 2:49in such traditional
technology-free classes -- -
2:49 - 2:52no laptops, no tablets,
and no phones! -
2:53 - 2:56This is the control group,
or baseline case. -
2:56 - 3:00Another group was allowed
to use electronics. -
3:00 - 3:03This is the treatment group,
subject to a changed environment. -
3:03 - 3:06The treatment in this case
is the unrestricted use -
3:06 - 3:08of laptops or tablets in class.
-
3:09 - 3:12Every causal question has
a clear outcome -- -
3:12 - 3:16the variables we hope to influence,
defined in advance of the study. -
3:16 - 3:19The outcomes
in the West Point electronics study -
3:19 - 3:20are final exam scores.
-
3:20 - 3:24The study seeks to answer
the following question -- -
3:24 - 3:28what is the causal effect
of classroom electronics on learning -
3:28 - 3:30as measured by exam scores?
-
3:30 - 3:33- [Narrator] West Point economics
students were randomly assigned -
3:33 - 3:36to either the treatment
or control groups. -
3:36 - 3:40Random assignment creates
ceteris paribus comparisons, -
3:40 - 3:44allowing us to draw causal
conclusions by comparing groups. -
3:45 - 3:49The causality revealing power
of a randomized trial comes -
3:49 - 3:53from a statistical property
called the law of large numbers. -
3:53 - 3:55When statisticians
and mathematicians uncover -
3:55 - 3:57something important
and reliably true -
3:57 - 4:01about the natural world,
they call it a law. -
4:02 - 4:05The law of large numbers says
that when the groups randomized -
4:05 - 4:08are large enough, the students
in them are sure to be similar -
4:08 - 4:10on average, in every way.
-
4:11 - 4:14This means that groups of students
randomly divided can be expected -
4:14 - 4:18to have similar family background,
motivation, and ability. -
4:19 - 4:22Sayonara selection bias,
at least in theory. -
4:23 - 4:26In practice, the groups randomized
might not be large enough -
4:26 - 4:28for the law
of large numbers to kick in. -
4:29 - 4:32Or the researchers might have
messed up the random assignment. -
4:32 - 4:35As in any technical endeavor,
even experienced masters -
4:35 - 4:37are alert for a possible foul-up.
-
4:38 - 4:40We therefore check for balance,
-
4:40 - 4:43comparing student background
variables across groups -
4:43 - 4:45to make sure
they indeed look similar. -
4:47 - 4:49- [Narrator] Here's
the balance check for West Point. -
4:50 - 4:52This table has two columns --
-
4:52 - 4:55one showing data
from the control group -
4:55 - 4:57and one showing data
from the treatment group. -
4:58 - 5:02The rows list a few of the variables
that we hope are balanced -- -
5:02 - 5:06sex, age, race, and high school GPA,
among others. -
5:07 - 5:09The first row indicates
what percentage -
5:09 - 5:12of each group is female.
-
5:12 - 5:16It's 17% for the control group
and 20% for the treatment group. -
5:18 - 5:21Kamal, how does balance
look for GPA? -
5:22 - 5:25- [Kamal] Controls have
a GPA of 2.87, -
5:25 - 5:29while treated have
a GPA of 2.82, pretty close. -
5:30 - 5:34- [Narrator] Happily the two groups
look similar all around. -
5:34 - 5:36- How big does our sample have to be
-
5:36 - 5:37for the law
of large numbers to kick in? -
5:38 - 5:39- [Narrator] The West Point study,
-
5:39 - 5:43which involved about 250 students
in each group, -
5:43 - 5:45is almost certainly big enough.
-
5:45 - 5:48There are no hard
and fast rules here. -
5:48 - 5:51In another video, you'll learn
how to confirm the hypothesis -
5:51 - 5:55of group balance
with formal statistical tests. -
5:55 - 5:56- [Man] Exciting.
-
6:00 - 6:02- The heart of the matter
in this table -
6:02 - 6:04is the estimated treatment effect.
-
6:05 - 6:07Remember the treatment
in this case is permission -
6:07 - 6:09to use electronics in class.
-
6:10 - 6:14Treatment effects compare averages
between control and treatment groups. -
6:15 - 6:18The group Allowed Classroom
Electronics had -
6:18 - 6:22an average final exam score
0.28 standard deviations -
6:22 - 6:24below the score for students
in the control group. -
6:26 - 6:28How big an effect is this?
-
6:28 - 6:32Social scientists measure test scores
in standard deviation units -
6:32 - 6:35because these units are easily
compared across studies. -
6:35 - 6:39We know from a long history
of research on classroom learning -
6:39 - 6:41that 0.28 is huge.
-
6:41 - 6:46A decline of 0.28 is like taking
the middle student in the class -
6:46 - 6:48and pushing him or her down
to the bottom third. -
6:49 - 6:52How certain can we be
that these large results -
6:52 - 6:53are meaningful?
-
6:53 - 6:56After all, we're looking
at a single randomized division -
6:56 - 6:58between treatment
and control groups. -
6:58 - 7:01Other random splits might have
produced something different. -
7:01 - 7:04- We therefore quantify
the sampling variance -
7:04 - 7:06in estimates of causal effects.
-
7:06 - 7:08- What is sampling variance?
-
7:09 - 7:11- [Narrator] Sampling variance
tells us how likely -
7:11 - 7:16a particular statistical finding
is to be a chance result, -
7:16 - 7:19rather than indicative
of an underlying relationship. -
7:20 - 7:23Sampling variance is summarized
by a number, -
7:23 - 7:27called the standard error
of the estimated treatment effect. -
7:27 - 7:29- [Students muttering] I don't get it.
- What is she even talking... ? -
7:29 - 7:33- Don't worry -- we'll cover
this important idea in depth later. -
7:33 - 7:34- [Student] Great.
-
7:34 - 7:36- Just remember the smaller
the standard error, -
7:36 - 7:39the more conclusive the findings.
-
7:39 - 7:42On the other hand,
if the standard error is large -
7:42 - 7:45relative to the effect
we're trying to estimate, -
7:45 - 7:47there's a pretty good chance
we'd get something different -
7:47 - 7:50if we were to rerun the experiment.
-
7:50 - 7:54You can think of the standard error
as a way to gauge how much -
7:54 - 7:57we can trust the result we're seeing.
-
7:57 - 7:58- [Students] Okay.
-
7:58 - 8:01- [Narrator] In this study,
the relevant standard error is 0.1. -
8:02 - 8:05- For now it's enough to learn
a simple rule of thumb. -
8:05 - 8:07When an estimated
treatment effect is more -
8:07 - 8:09than double its standard error,
-
8:09 - 8:12the odds this non-zero result is due
to chance are very low, -
8:12 - 8:14roughly 1 in 20.
-
8:15 - 8:18Because this is so unlikely,
we say that estimates that are two -
8:18 - 8:21or more times larger
than the associated standard errors -
8:21 - 8:23are statistically significant.
-
8:24 - 8:28- Camilla, is the treatment effect
in the West Point study -
8:28 - 8:30statistically significant?
-
8:31 - 8:36- The standard error is 0.10
and the treatment effect is 0.28, -
8:36 - 8:40so 0.28 is more than two times larger
than 0.10, so yeah. -
8:40 - 8:42- Correct.
-
8:42 - 8:46The lost learning caused
by electronics use in Econ 101 -
8:46 - 8:50is therefore both large
and statistically significant. -
8:50 - 8:51- [Man] Nice.
-
8:57 - 9:00- Randomized trials usually provide
the most convincing answers -
9:00 - 9:01to causal questions.
-
9:01 - 9:04When this weapon
is in our tool kit, we use it. -
9:05 - 9:07Random assignment allows us
to claim ceteris -
9:07 - 9:10has indeed been made paribus.
-
9:10 - 9:13But randomized trials
could be tough to organize. -
9:13 - 9:15They may be expensive,
time-consuming, -
9:15 - 9:18and, in some cases,
considered unethical. -
9:18 - 9:21So masters look
for convincing alternatives. -
9:21 - 9:25The alternatives try to mimic
the causality revealing power -
9:25 - 9:28of a randomized trial
but without the time, trouble, -
9:28 - 9:31and expense
of a purpose-built experiment. -
9:31 - 9:35These alternative tools are applied
in real world scenarios -
9:35 - 9:37that mimic random assignment.
-
9:39 - 9:43- [Narrator] You're on your way
to mastering Econometrics. -
9:43 - 9:44Make sure this video sticks
-
9:44 - 9:47by taking a few
quick practice questions. -
9:47 - 9:51Or, if you're ready,
click for the next video. -
9:51 - 9:54You can also check out MRU's website
-
9:54 - 9:57for more courses,
teacher resources, and more.
- Title:
- Randomized Trials: The Ideal Weapon
- Description:
-
MIT’s Josh Angrist—aka Master Joshway—introduces us to our most powerful weapon: randomized trials!
Randomized trials originate in medical research, where they’re called “randomized clinical trials” or “RCTs.” That’s why randomized trials are said to measure “treatment effects”.
Josh covers a fascinating study from West Point that seeks to answer a common question using randomized trials: Are our devices more distracting than useful when it comes to learning?
In this video, we cover the following:
-What's the difference between control and treatment groups?
-How to "check for balance"
-The Law of Large Numbers
-Treatment effects
-Standard errors and statistical significance***INSTRUCTOR RESOURCES***
High school teacher resources: https://bit.ly/2RcCsta
Professor resources: https://bit.ly/2XgdHQG
EconInbox: https://bit.ly/2yzTjzz***MORE LEARNING***
Try out our practice questions: https://bit.ly/2xVuvSg
See the full course: https://bit.ly/2Xc88CQ
Receive updates when we release new videos: https://bit.ly/2VaLTdH
More from Marginal Revolution University: https://bit.ly/39QeWZo - Video Language:
- English
- Team:
- Marginal Revolution University
- Project:
- Mastering Econometrics
- Duration:
- 10:01
Retired user edited English subtitles for Randomized Trials: The Ideal Weapon | ||
Kirstin Cosper edited English subtitles for Randomized Trials: The Ideal Weapon | ||
Kirstin Cosper edited English subtitles for Randomized Trials: The Ideal Weapon | ||
Kirstin Cosper edited English subtitles for Randomized Trials: The Ideal Weapon | ||
Kirstin Cosper edited English subtitles for Randomized Trials: The Ideal Weapon | ||
Kirstin Cosper edited English subtitles for Randomized Trials: The Ideal Weapon | ||
Kirstin Cosper edited English subtitles for Randomized Trials: The Ideal Weapon | ||
Kirstin Cosper edited English subtitles for Randomized Trials: The Ideal Weapon |