Return to Video

Introduction to Instrumental Variables (IV)

  • Not Synced
    - [Instructor] The path from cause
    to effect is dark and dangerous.
  • Not Synced
    But the weapons
    of Econometrics are strong.
  • Not Synced
    [Attack] with fierce
    and flexible instrumental variables
  • Not Synced
    when nature blesses you
    with fortuitous random assignment.
  • Not Synced
    [gong rings]
  • Not Synced
    Randomized trials are the surest
    path to ceteris parabus comparisons.
  • Not Synced
    Alas, this powerful tool
    is often unavailable.
  • Not Synced
    But sometimes, randomization
    happens by accident.
  • Not Synced
    That's when we turn
    to instrumental variables,
  • Not Synced
    IV for short.
  • Not Synced
    - [Voice whispers] Instrumental
    variables.
  • Not Synced
    - [Instructor] Today's lesson
    is the first of two on IV.
  • Not Synced
    Our first IV lesson begins
    with a story of schools.
  • Not Synced
    - [Josh] Charter schools
    are public schools
  • Not Synced
    freed from daily district oversight
    and teacher union contracts.
  • Not Synced
    The question of whether charters
    boost achievement
  • Not Synced
    is one of the most important
  • Not Synced
    in the history
    of American education reform.
  • Not Synced
    - The most popular charter schools
    have more applicants
  • Not Synced
    so the luck of the lottery draw
    decides who's offered a seat.
  • Not Synced
    A lot is at stake for the students
    vying for their chance.
  • Not Synced
    Waiting for the lottery results
    brings up lots of emotions
  • Not Synced
    as was captured
    in the award-winning documentary
  • Not Synced
    "Waiting For Superman."
  • Not Synced
    - [Mother] Don't cry.
    You're gonna make Mommy cry.
  • Not Synced
    Okay?
  • Not Synced
    - Do charters really provide
    a better education?
  • Not Synced
    Critics most definitely say no,
  • Not Synced
    arguing that charters enroll
    better students to begin with,
  • Not Synced
    smarter or more motivated
    so differences in later outcomes
  • Not Synced
    reflects selection bias.
  • Not Synced
    - [Kamal] Wait, this one seems easy.
  • Not Synced
    In a lottery, winners
    are chosen randomly,
  • Not Synced
    so just compare winners and losers.
  • Not Synced
    - [Student] Obviously.
  • Not Synced
    - On the right track, Kamal,
  • Not Synced
    but charter lotteries
    don't force kids into
  • Not Synced
    or out of a particular school.
  • Not Synced
    They randomize offers
    of a charter seat.
  • Not Synced
    Some kids get lucky.
  • Not Synced
    Some kids don't.
  • Not Synced
    If we just wanted to know
    the effect of charter school offers,
  • Not Synced
    we could treat this
    as a randomized trial.
  • Not Synced
    But we we're interested
    in the effects
  • Not Synced
    of charter school attendance,
    not offers.
  • Not Synced
    And not everyone
    who is offered, accepts.
  • Not Synced
    IV turns the effect of being offered
    a charter seat into the effect
  • Not Synced
    of actually attending
    a charter school.
  • Not Synced
    - [Student] Cool.
  • Not Synced
    - Oh nice.
  • Not Synced
    - Let's look at an example,
    a charter school from
  • Not Synced
    the Knowledge Is Power
    Program, or KIPP for short.
  • Not Synced
    This KIPP school is in Lynn,
  • Not Synced
    a faded industrial town
    on the coast of Massachusetts.
  • Not Synced
    The school has
    more applicants than seats
  • Not Synced
    and therefore picks its students
    using a lottery.
  • Not Synced
    From 2005 to 2008,
    371 fourth and fifth graders
  • Not Synced
    put their names
    in the KIPP/Lynn lottery,
  • Not Synced
    253 students won a seat at KIPP,
  • Not Synced
    118 students lost.
  • Not Synced
    A year later, lottery winners had
    much higher match scores
  • Not Synced
    than lottery losers.
  • Not Synced
    But remember,
    we're not trying to figure out
  • Not Synced
    whether winning a lottery
    makes you better at math.
  • Not Synced
    We want to know if attending KIPP
    makes you better at math.
  • Not Synced
    Of the 253 lottery winners,
    only 199 actually went to KIPP.
  • Not Synced
    The others chose
    a traditional public school.
  • Not Synced
    Similarly of the 118 lottery losers,
    a few actually ended up at KIPP.
  • Not Synced
    They got an offer later.
  • Not Synced
    So what was the effect of test scores
    of actually attending KIPP?
  • Not Synced
    - [Student] Why can't we just
    measure their math scores?
  • Not Synced
    - [Instructor] Great question.
  • Not Synced
    Who would you compare them to?
  • Not Synced
    - [Student] Those who didn't attend.
  • Not Synced
    - [Instructor] Is attendance random?
  • Not Synced
    - [Camilla] No.
  • Not Synced
    - Selection bias.
  • Not Synced
    - [Instructor] Correct.
  • Not Synced
    - [Otto] What?
  • Not Synced
    - [Instructor] The KIPP offers
    are random so we can be confident
  • Not Synced
    of ceteris parabus,
    but attendance is not random.
  • Not Synced
    The choice to accept the offer
    might be due to characteristics
  • Not Synced
    that are related
    to math performance.
  • Not Synced
    Say, for example,
    that dedicated parents
  • Not Synced
    are more likely
    to accept the offer.
  • Not Synced
    Their kids are also more likely
    to do better in math,
  • Not Synced
    regardless of school.
  • Not Synced
    - [Student] Right.
  • Not Synced
    - [Instructor] IV converts
    the offer effect
  • Not Synced
    into the effect of KIPP attendance,
  • Not Synced
    adjusting for the fact
    that some winners go elsewhere
  • Not Synced
    and some losers manage
    to attend KIPP anyway.
  • Not Synced
    Essentially, IV takes
    an incomplete randomization
  • Not Synced
    and makes the appropriate
    adjustments.
  • Not Synced
    How? IV describes a chain reaction.
  • Not Synced
    Why do offers affect achievement?
  • Not Synced
    Probably because they affect
    charter attendance
  • Not Synced
    and charter attendance
    improves math scores,
  • Not Synced
    the first link in the chain
    called the first stage
  • Not Synced
    is the effect of the lottery
    on charter attendance.
  • Not Synced
    The second stage is the length
    between attending a charter
  • Not Synced
    and an outcome variable,
  • Not Synced
    in this case, math scores.
  • Not Synced
    The instrumental variable
    or instrument for short
  • Not Synced
    is the variable
    that initiates the chain reaction.
  • Not Synced
    The effect of the instrument
    on the outcome is called
  • Not Synced
    the reduced form.
  • Not Synced
    This chain reaction can be
    represented mathematically.
  • Not Synced
    We multiply the first stage,
    the effect of winning
  • Not Synced
    on attendance, by the second stage,
  • Not Synced
    the affect of attendance on scores.
  • Not Synced
    And we get the reduced form,
  • Not Synced
    the effect of winning
    the lottery on scores.
  • Not Synced
    Reduced form and first stage
    are observable and easy to compute.
  • Not Synced
    However, the effect of attendance
    on achievement
  • Not Synced
    is not directly observed.
  • Not Synced
    This is the causal effect
    we're trying to determine.
  • Not Synced
    Given some important assumptions
    we'll discuss shortly,
  • Not Synced
    we can find the effect
    of KIPP attendance
  • Not Synced
    by dividing the reduced form
    by the first stage.
  • Not Synced
    This will become more clear
    as we work through an example.
  • Not Synced
    - [Student] Let's do this.
  • Not Synced
    - A quick note on measurement.
  • Not Synced
    We measure achievement
    using standard deviations,
  • Not Synced
    often denoted
    by the Greek letter sigma (σ).
  • Not Synced
    One σ is a huge move
    from around the bottom 15%
  • Not Synced
    to the middle of most
    achievement distributions.
  • Not Synced
    Even a 1/4 or 1/2 σ
    difference is big.
  • Not Synced
    - [Instructor] Now we're ready
    to plug some numbers
  • Not Synced
    into the equation
    we introduced earlier.
  • Not Synced
    First up, what's the effect
  • Not Synced
    of winning the lottery
    on math scores?
  • Not Synced
    KIPP applicant math scores
    are a third of a standard deviation
  • Not Synced
    below the state average in
    the year before they apply to KIPP.
  • Not Synced
    But a year later, lottery winners
    score right at the state average
  • Not Synced
    while the lottery losers
    are still well behind.
  • Not Synced
    With an average score
    around - 0.36 σ.
  • Not Synced
    The effect of winning the lottery
    on scores is the difference
  • Not Synced
    between the winners' scores
    and the losers' scores.
  • Not Synced
    Take the winners'
    average math scores,
  • Not Synced
    subtract the losers'
    average math scores,
  • Not Synced
    and you will have 0.36 σ .
  • Not Synced
    Next stop: what's the effect
    of winning the lottery on attendance?
  • Not Synced
    In other words,
    if you win the lottery,
  • Not Synced
    how much more likely are you
    to attend KIPP than if you lose?
  • Not Synced
    First, what percentage
    of lottery winners attend KIPP?
  • Not Synced
    Divide the number of winners
    who attended KIPP
  • Not Synced
    by the total number
    of lottery winners -- that's 78%.
  • Not Synced
    To find the percentage
    of lottery losers who attended KIPP,
  • Not Synced
    we divide the number of losers
    who attended KIPP
  • Not Synced
    by the total number
    of lottery losers -- that's 4%.
  • Not Synced
    Subtract 4 from 78, and we find
    that winning the lottery
  • Not Synced
    makes you 74%
    more likely to attend KIPP.
  • Not Synced
    Now we can find
    what we're really after,
  • Not Synced
    the effect of attendance on scores,
    by dividing 0.36 by 0.74.
  • Not Synced
    Attending KIPP raises math scores
  • Not Synced
    by 0.48 standard deviations
    on average.
  • Not Synced
    That's an awesome achievement game,
  • Not Synced
    equal to moving
    from about the bottom
  • Not Synced
    to the middle
    of the achievement distribution.
  • Not Synced
    - [Student] Whoa, half a sig.
  • Not Synced
    - [Instructor] These estimates
    are for kids opting in
  • Not Synced
    to the KIPP lottery,
    whose enrollment status
  • Not Synced
    is changed by winning.
  • Not Synced
    That's not necessarily
    a random sample
  • Not Synced
    of all children in Lynn.
  • Not Synced
    So we can't assume
    we'd see the same effect
  • Not Synced
    for other types of students.
  • Not Synced
    - [Students] Huh.
  • Not Synced
    - But this effect
    on keen for KIPP kids
  • Not Synced
    is likely to be a good indicator
    of the consequences
  • Not Synced
    of adding additional charter seats.
  • Not Synced
    - [Student] Cool.
    - [Student] Got it.
  • Not Synced
    - IV eliminates selection bias
    but like all of our tools,
  • Not Synced
    the solution builds on a set
    of assumptions
  • Not Synced
    not to be taken for granted.
  • Not Synced
    First, there must be
    a substantial first stage --
  • Not Synced
    that is the instrumental variable,
    winning or losing the lottery,
  • Not Synced
    must really change the variable
    whose effect we're interested in --
  • Not Synced
    here, KIPP attendance.
  • Not Synced
    In this case, the first stage
    is not really in doubt.
  • Not Synced
    Winning the lottery makes
    KIPP attendance much more likely.
  • Not Synced
    Not all IV stories are like that.
  • Not Synced
    Second, the instrument
    must be as good
  • Not Synced
    as randomly assigned
    meaning lottery winners and losers
  • Not Synced
    have similar characteristics.
  • Not Synced
    This is the independence assumption.
  • Not Synced
    Of course, KIPP lottery wins
    really are randomly assigned.
  • Not Synced
    Still, we should check for balance
    and confirm that winners and losers
  • Not Synced
    have similar family backgrounds,
  • Not Synced
    similar aptitudes and so on.
  • Not Synced
    In essence, we're checking
    to ensure KIPP lotteries are fair
  • Not Synced
    with no group of applicants
    suspiciously likely to win.
  • Not Synced
    Finally, we require
    the instrument change outcomes
  • Not Synced
    solely through
    the variable of interest,
  • Not Synced
    in this case, attending KIPP.
  • Not Synced
    This assumption is called
    the exclusion restriction.
  • Not Synced
    - IV only works if you can satisfy
    these three assumptions.
  • Not Synced
    - I don't understand
    the exclusion restriction.
  • Not Synced
    How could winning the lottery
    affect math scores
  • Not Synced
    other than by attending KIPP?
  • Not Synced
    - [Student] Yeah.
  • Not Synced
    - [Instructor] Great question.
  • Not Synced
    Suppose lottery winners
    are just thrilled to win,
  • Not Synced
    and this happiness motivates them
    to study more and learn more math
  • Not Synced
    regardless of where
    they go to school.
  • Not Synced
    This would violate
    the exclusion restriction
  • Not Synced
    because the motivational effect
    of winning is a second channel
  • Not Synced
    whereby lotteries
    might affect test scores.
  • Not Synced
    While it's hard
    to rule this out entirely,
  • Not Synced
    there's no evidence
    of any alternative channels
  • Not Synced
    in the KIPP study.
  • Not Synced
    - IV solves the problem
    of selection bias
  • Not Synced
    in scenarios like the KIPP lottery
    where treatment offers are random
  • Not Synced
    but some of those offered opt out.
  • Not Synced
    This sort of intentional
    yet incomplete random assignment
  • Not Synced
    is surprisingly common.
  • Not Synced
    Even randomized clinical trials
    have this feature.
  • Not Synced
    IV solves the problem
    of non-random take up
  • Not Synced
    in lotteries or clinical research.
  • Not Synced
    But lotteries are not the only
    source of compelling instruments.
  • Not Synced
    Many causal questions can be
    addressed by naturally occurring
  • Not Synced
    as good as randomly
    assigned variation.
  • Not Synced
    Here's a causal question for you --
  • Not Synced
    do women who have children early
    in their careers suffer
  • Not Synced
    a substantial earnings penalty
    as a result?
  • Not Synced
    After all, women earn less than men.
  • Not Synced
    We could, of course, simply
    compare the earnings of women
  • Not Synced
    with more and fewer children.
  • Not Synced
    But such comparisons are fraught
    with selection bias.
  • Not Synced
    If only we could randomly assign
    babies to different households.
  • Not Synced
    Yeah, right,
    sounds pretty fanciful.
  • Not Synced
    Our next IV story -- fantastic
    and not fanciful --
  • Not Synced
    illustrates an amazing,
    naturally-occurring instrument
  • Not Synced
    for family size.
  • Not Synced
    - [Instructor] You're on your way
    to mastering Econometrics.
  • Not Synced
    Make sure this video sticks
  • Not Synced
    by taking a few
    quick practice questions.
  • Not Synced
    Or, if you're ready,
    click for the next video.
  • Not Synced
    You can also check out
    MRU's website for more courses,
  • Not Synced
    teacher resources, and more.
  • Not Synced
    ♪ [music] ♪
Title:
Introduction to Instrumental Variables (IV)
Description:

more » « less
Video Language:
English
Team:
Marginal Revolution University
Project:
Mastering Econometrics
Duration:
12:57

English subtitles

Revisions Compare revisions