Lecture 19: Instrumental Variables (Chapter 13.2–13.3) Copyright © 2006 Pearson Addison-Wesley. All rights reserved. Agenda • Review • Randomization • Instrumental Variables (Chapter 13.2) • Using Instrumental Variables (Chapter 13.2) Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-2 Review • If we relax the assumption that explanators are fixed across samples, we confront a number of serious complications. • Stochastic explanators render the mathematics of finding unbiased estimators intractable. • Instead, econometricians search for consistent estimators. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-3 Review (cont.) • The most problematic condition for OLS to be consistent is E(xi i ) 0 which, by the Law of Large Numbers, 1 implies p lim xi i 0. n Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-4 Review (cont.) xiYi xi [ 0 1 X 1 i ] ˆ p lim( 1 ) p lim 2 p lim 2 x x i i 1 x n i i xi i 1 p lim 1 p lim 2 1 x i xi2 n 1 p lim xi i 0 n 1 1 1 Q 1 2 p lim xi n Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-5 Review (cont.) • Asking that key variances be non-zero and bounded is often reasonable (except in certain time series cases). Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-6 Review (cont.) • How reasonable an assumption is E(Xii ) 0 (i.e. Xi and i are uncorrelated)? • In practice, this assumption is often terrible. • When an explanator is correlated with the error term, we call the explanator a “troublesome variable.” Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-7 Review (cont.) • We have examined four complications that induce correlation between X and 1. Omitted Variables Bias 2. Measurement Error 3. Simultaneous Causality 4. Using Lagged Values of the Dependent Variable as Explanators, in the presence of serial correlation Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-8 Review (cont.) • Omitting a variable creates a bias only if: 1. X2 is an explanator of Y (so, when omitted, it becomes a component of the error term). 2. X2 is correlated with X1 (so that X2 creates a correlation between X1 and the error term). Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-9 Review (cont.) • Measurement error also induces a correlation between our included explanator and the error term • Instead of observing Xi , we observe M i Xi vi Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-10 Review (cont.) Model: Yi 0 1 X 1 i Instead of observing X i , we observe M i X i vi We regress Yi 0 1M i i Var ( X i ) X2 1 2 p lim(ˆ1 ) 1 X v2 Var ( X i ) Var (vi ) Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-11 Review (cont.) • Mismeasuring X leads to ATTENUATION BIAS. The estimated coefficient is biased towards 0. • The magnitude of the bias depends on the relative variances of X and v. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-12 Review (cont.) • Under simultaneous causality, X and Y are jointly determined. • Because X and Y are determined simultaneously, X can adjust in response to shocks to Y (). • Thus X will be correlated with Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-13 Review (cont.) • Using lagged dependent variables as explanators is another potential source of correlation between an explanator and the error term. • If there is first-order serial correlation in the error terms, then t-1 is correlated with t • Therefore Yt-1 is correlated with t Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-14 Randomization • In practice, correlation between X and is endemic. • Much of econometric work involves studying the process determining the explanators, to see how they might be correlated with . Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-15 Randomization (cont.) • The ideal X variable has been randomly assigned. • If X has been randomly assigned, then it contains no information about . Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-16 Randomization (cont.) • If X has been randomly assigned, then E( | X) 0 • It follows that E( X) 0 • Moreover, if E( | X) 0 , then OLS is unbiased. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-17 Randomization (cont.) • Laboratory scientists typically work with randomly assigned explanators. • In an experiment, the scientist randomly assigns different subjects to different treatments. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-18 Randomization (cont.) • Economists have been making increasing use of experiments. • Some economists construct carefully controlled markets in computer laboratories (usually with undergraduate subjects). • Other economists insert randomization into actual programs in the field, for example randomly assigning participants to different health care plans. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-19 Randomization (cont.) • See Chapter 15 for a discussion of controlled experiments. • Most econometric research, however, studies field data. • The econometrician observes variables as they are naturally determined, and attempts to draw inferences. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-20 Randomization (cont.) • Occasionally, variables are naturally determined by random chance. • When nature assigns X randomly, econometricians can study a “natural experiment.” Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-21 Randomization (cont.) • For example, Alvin Roth studies centralized computer matching programs to assign medical school graduates to hospital training programs. Graduates and hospitals each submit preference lists, and a computer sorts through the lists and creates matches. • Game theory suggests key properties for such a matching program. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-22 Randomization (cont.) • Roth noticed that in the United Kingdom, each region of the country ran its own centralized matching program. • In some of the regions, organizers had happened to adopt a matching program that possessed good game theoretic properties. Other regions did not adopt good designs. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-23 Randomization (cont.) • Roth speculated that the choice of program design was reasonably random, at least in the sense that program design was not systematically linked to any other properties of the region. • Programs with good game theoretic properties were much more likely to be successful. Centralized programs with poor properties fell into disuse. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-24 Randomization (cont.) • In Roth’s study, advanced econometric methods were unnecessary. • The data naturally provided the “experiment” he needed to observe. • Of course, it is Roth’s duty to argue compellingly that the program rules really are uncorrelated with other aspects of the regions. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-25 Randomization (cont.) • Nature often provides a certain degree of randomization. • For example, suppose we are studying the effect of military service on earnings. • During the Vietnam War, the draft lottery was a largely random process. • Thus, a study of Vietnam War veterans would be more random than a study of veterans from the more recent all-volunteer American army. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-26 Randomization (cont.) • However, service in Vietnam was determined by more than simple random chance. • Some men voluntarily enlisted; others received draft deferments (or simply evaded the draft). • The decision to volunteer, or to avoid being drafted, would undoubtedly be correlated with many other variables. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-27 Randomization (cont.) • Military service in Vietnam, then, consists of two elements: 1. A “clean” random element determined by the draft lottery, that would be uncorrelated with in an earnings regression; and 2. An “unclean” element, determined by active decision-making on the agent’s part, that could easily be correlated with . Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-28 Randomization (cont.) • Vietnam War military service is partially determined by an obviously random process. • Nonetheless, a simple OLS estimate of the effect of military service on earnings would be inconsistent. Military service is also partially determined non-randomly. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-29 Randomization (cont.) • Fortunately, econometricians have discovered a method for separating out the random elements of explanators from the elements that may be correlated with . • Unfortunately, this method requires the data to include an instrumental variable with certain key properties. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-30 Instrumental Variables (Chapter 13.2) • An Instrumental Variable is a variable that is correlated with X but uncorrelated with . • If Zi is an instrumental variable: 1. E( Zi Xi ) ≠ 0 2. E( Zi i ) = 0 Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-31 Instrumental Variables (cont.) • For example, a Vietnam War lottery number is: 1. Correlated with X (“winning” the lottery is highly correlated with military service), but 2. Is uncorrelated with (the lottery number is not connected to cultural, psychological, or economic factors that might affect the decision to join the military, and that might also affect earnings). Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-32 Instrumental Variables (cont.) • The econometrician can use an instrumental variable Z to estimate the effect on Y of only that part of X that is correlated with Z. • Because Z is uncorrelated with , any part of X that is correlated with Z must also be uncorrelated with . Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-33 Instrumental Variables (cont.) • An instrumental variable lets the econometrician find a part of X that behaves as though it had been randomly assigned. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-34 Instrumental Variables (cont.) • For example, let’s revisit the question of how much mortality can be reduced by intensive cardiac treatment. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-35 Instrumental Variables (cont.) Mortalityi 0 1 DiIntensivelyTreated 2 DiFemale ... kUrbani i • If our observable control variables, such as DFemale, were the only differences between patients who received intensive treatment and those who did not, then DIntensivelyTreated would not tell us anything about . OLS would be consistent. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-36 Instrumental Variables (cont.) • However, we reasonably believe that a doctor’s choice to perform intensive cardiac procedures are correlated with many other variables. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-37 Instrumental Variables (cont.) • Doctors might select patients to receive treatment based on their underlying health status. • If health status is an unobservable determinant of mortality, then it is a component of . • OLS will give an inconsistent estimate of the benefits of intensive treatment. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-38 Instrumental Variables (cont.) • To eliminate Omitted Variables Bias, we need to find some determinant of a patient’s receiving intensive cardiac care that is unrelated to mortality. • Is there any element of the cardiac care process that is reasonably random? Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-39 Instrumental Variables (cont.) • Not every hospital is equipped to provide intensive cardiac care. • Some patients live near cardiac care centers. When they have a heart attack, they are more likely to be transported to a center, and thus more likely to receive intensive treatment. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-40 Instrumental Variables (cont.) • McClellan, McNeil, and Newhouse argue that the distance from a patient’s home to the nearest hospital equipped for intensive cardiac care is a valid instrumental variable. • They argue that geographic location is likely to be uncorrelated with unobserved traits that influence mortality. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-41 Instrumental Variables (cont.) • A major task for McClellan, McNeil, and Newhouse is to determine whether or not distance to a cardiac care center really is uncorrelated with mortality (except through the increased likelihood of receiving cardiac treatment). • As a reader of such a study, you should ask yourself whether the authors have convinced you that their instrument is uncorrelated with Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-42 Instrumental Variables (cont.) • When McClellan, McNeil, and Newhouse use their instrumental variable, they find only a small benefit from intensive cardiac care for the marginal patient (some patients may still benefit quite a lot!). Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-43 Instrumental Variables (cont.) • When the economist is worried about measurement error, a good choice of instrument is simply a different measure of the same variable. • The new measure may have its own errors, but these errors are unlikely to be correlated with the mistakes in the first measure, or with any other component of . Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-44 Instrumental Variables (cont.) • For example, Ashenfelter and Rouse were studying the effect of education on earnings. • Their data came from a survey of twins. • They were concerned that individuals might mis-report their own years of schooling, leading to measurement error biases. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-45 Instrumental Variables (cont.) • However, Ashenfelter and Rouse had two separate measures for each individual’s years of schooling. • The survey asked each individual to list both his/her own years of schooling, and also the years of schooling for his/her twin. • The twin’s report of an individual’s schooling served as an instrumental variable for the individual’s self-report. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-46 Instrumental Variables (cont.) • Another example: policy makers are greatly interested in the effects of tax rates on labor force participation (and other taxpayer behaviors). • They would like to run regressions with an individual’s tax rate as an explanator. • However, an individual has some choice over his/her tax rate. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-47 Instrumental Variables (cont.) • Taxpayers who are close to the income threshold for a new tax bracket can choose to limit their taxable income. • For example, they might take more of their pay in the form of untaxed benefits or deferred 401(k) compensation rather than pay higher taxes on the extra compensation. • The ability and desire to adjust taxable income may well be correlated with . Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-48 Instrumental Variables (cont.) • When the government changes the tax rates, the individual’s new tax rate is determined by two elements: 1. The change in tax rates (which is uncorrelated with anything else about the individual), and 2. The individual’s decisions about how to respond to the tax change (which could well be correlated with ). Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-49 Instrumental Variables (cont.) • Public finance economists construct an instrumental variable that captures only the change in tax rates, not the change in behavior. • They use the new tax tables to look up the tax rate individuals would face IF they did NOT change their behavior from before the tax change. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-50 Instrumental Variables (cont.) • The constructed tax rate is correlated with the tax rate the individuals face after the tax change. • The constructed tax rate is uncorrelated with the behavioral adjustments individuals make in response to the tax rate. • Such a constructed instrument is called a simulated instrumental variable. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-51 Checking Understanding • Suppose you are studying the effect of price on the demand for cigarettes, using a cross-section of different states’ cigarette consumption and average price. • You would like to regress CigarettesSoldi 0 1 Pricei i where i indexes each state Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-52 Checking Understanding (cont.) CigarettesSoldi 0 1 Pricei i where i indexes each state • Because Pricei is endogenous, you need to instrument. Which of these variables would be suitable? 1. Each state’s cigarette excise tax 2. A measure of each state’s anti-smoking laws 3. Each state’s sales tax Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-53 Checking Understanding (cont.) 1. Each state’s cigarette excise tax – Cigarette excise taxes are surely correlated with cigarette prices. However, they also reflect the level of anti-smoking sentiment in the state (MA has a tax of $1.51 per pack, NC has a tax of $0.05 per pack). Anti-smoking sentiment is an omitted determinant of consumption, so excise taxes are correlated with . Excise taxes are not a valid instrument. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-54 Checking Understanding (cont.) 2. A measure of state anti-smoking laws – State anti-smoking laws might be correlated with price, but only through their effect on cigarette demand in the state. Such measures are an explanator of cigarette consumption; moreover, they are also a proxy for state anti-smoking sentiment. Anti-smoking laws are a component of , and would make a terrible instrument. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-55 Checking Understanding (cont.) 3. Each state’s sales tax – State sales taxes are correlated with cigarette prices. Higher sales taxes raise the prices of all goods. – There is no reason to expect sales taxes to have any other effect on cigarette consumption, or to be correlated with any other determinant of consumption. – State sales taxes are a reasonable instrument. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-56 Using Instrumental Variables (Chapter 13.2) • Instrumental variables are NOT the explanator of interest. • We want to know the effect of military service on earnings, not just the effect of “winning” the draft lottery. • We want to know the effect of intensive cardiac care on mortality, not just the effect of living near a cardiac care center. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-57 Using Instrumental Variables (cont.) • We do NOT simply use instrumental variables as proxies for the explanator of interest. • Instead, we use IV’s as a tool to tease out the “random” (or at least uncorrelated) component of X. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-58 Using Instrumental Variables (cont.) • Let’s construct a consistent IV estimator for the case of measurement error. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-59 DGP with E(Xii ) ≠ 0 Yi 0 1 X i i E( i ) 0 Var( i ) 2 Cov( i , j ) 0 for i j E( X i i ) 0, M i X i vi 1 (xi 2 ) 2X < n E(vi ) 0 Var(vi ) v2 Cov(vi ,v j ) 0 for i j Cov(vi , X i ) 0 Cov(Z i , X i ) 0 Cov(Z i , i ) 0 Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-60 Using Instrumental Variables • If Xi were uncorrelated with i , we would want to weight more heavily observations with a high xi value. • We know that Zi is correlated with the “clean” part of Xi , so now we want to weight more heavily observations with a high zi value. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-61 Using Instrumental Variables (cont.) ˆ IV Copyright © 2006 Pearson Addison-Wesley. All rights reserved. ziYi zi mi 19-62 Using Instrumental Variables (cont.) Generalizing away from measurement error: Yi 0 1 X i i E( X i i ) 0 E(Z i i ) 0 E(zi xi ) 0 Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-63 Using Instrumental Variables (cont.) ˆ IV Copyright © 2006 Pearson Addison-Wesley. All rights reserved. ziYi zi xi 19-64 Checking Understanding ˆ IV ziYi zi xi IV OLS ˆ ˆ How does differ from the that would result from regressing Yi 0 1Z i i Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-65 Checking Understanding (cont.) ziYi IV ˆ zi xi ziYi OLS ˆ 2 zi The estimators differ in the denominators. Only z appears in the denominator of ˆ OLS , i in the form of zi2 . Both zi and xi appear in the denominator of ˆ IV , in the form of z x . i i Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-66 Using Instrumental Variables • What is the expectation of IV? ziYi zi ( 0 1 X i i ) IV ˆ E ( 1 ) E E zi xi zi xi zi X i zi i 1 E E zi xi zi xi zi i 1 E zi xi Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-67 Using Instrumental Variables (cont.) • What is the expectation of IV? zi i IV ˆ E ( ) 1 E zi xi Because Cov( X i , i ) 0, the bias term cannot be eliminated. IV is biased in the same direction as the bias in OLS. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-68 Using Instrumental Variables (cont.) • What is the probability limit of IV? 1 p lim ziYi ziYi IV n ˆ p lim( ) p lim z x i i p lim 1 zi xi n 1 p lim zi ( 0 1 X i i ) n 1 p lim zi xi n Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-69 Using Instrumental Variables (cont.) • What is the probability limit of IV? 1 1 1 p lim zi xi p lim zi i IV n n ˆ p lim( 1 ) 1 p lim zi xi n By the Law of Large Numbers.... 1Cov( Z i , X i ) Cov( Z i , i ) Copyright © 2006 Pearson Addison-Wesley. All rights reserved. Cov( Z i , X i ) 1 19-70 Using Instrumental Variables (cont.) • What is the probability limit of IV? Cov( Z i , X i ) Cov( Z i , i ) IV ˆ p lim( 1 ) 1 Cov( Z i , X i ) Cov( Z i , X i ) If Cov( Z i , X i ) 0, the denominator equals 0, and the p lim does not exist. If Cov( Z , ) 0, then ˆ IV is inconsistent. i i Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 1 19-71 Using Instrumental Variables (cont.) • The asymptotic variance of IV is 1 2 p lim zi 1 2 1 2 Var(Zi ) n 2 n n Cov(Zi , X i )2 1 p lim zi xi n • The greater the covariance between X and Z, the lower the asymptotic variance. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-72 Using Instrumental Variables (cont.) • In the multiple regression case, we may have more than one explanator that is correlated with the error term (i.e. more than one troublesome variable). • However, an instrumental variable only needs to instrument for one of the troublesome variables. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-73 Using Instrumental Variables • A variable Zi can instrument for a particular troublesome explanator, XRi, if: 1. Cov( Zi,XRi ) ≠ 0 2. Cov( Zi,i ) = 0 • Zi must be correlated with the troublesome variable for which it instruments, but need not be correlated with all of the troublesome variables. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-74 Using Instrumental Variables (cont.) • To estimate a multiple regression consistently, we need at least one instrumental variable for each troublesome explanator. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-75 Using Instrumental Variables (cont.) • When we have just enough instruments for consistent estimation, we say the regression equation is exactly identified. • When we have more than enough instruments, the regression equation is over identified. • When we do not have enough instruments, the equation is under identified (and inconsistent). Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-76 Review (cont.) • In practice, correlation between X and is endemic. • Much of econometric work involves studying the process determining the explanators, to see how they might be correlated with Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-77 Review (cont.) • The ideal X variable has been randomly assigned. • If X has been randomly assigned, then it contains no information about • However, true randomization is relatively uncommon. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-78 Review (cont.) • Often, an explanator is partially determined in a way that is random, or at least uncorrelated with • However, the explanator is also influenced by omitted variables, or determined endogenously, or is in some other way correlated with Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-79 Review (cont.) • Fortunately, econometricians have discovered a method for separating out the random elements of explanators from the elements that may be correlated with . • Unfortunately, this method requires the data to include an instrumental variable with certain key properties. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-80 Review (cont.) • An Instrumental Variable is a variable that is correlated with X but uncorrelated with . • If Zi is an instrumental variable: 1. E(ZiXi ) ≠ 0 2. E(Zii ) = 0 Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-81 Review (cont.) • If Xi were uncorrelated with i , we would want to weight more heavily observations with a high xi value. • We know that Zi is correlated with the “clean” part of Xi , so now we want to weight more heavily observations with a high zi value. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-82 Review (cont.) ˆ IV Copyright © 2006 Pearson Addison-Wesley. All rights reserved. ziYi zi xi 19-83 Review (cont.) • What is the probability limit of IV? Cov( Z i , X i ) Cov( Z i , i ) IV ˆ p lim( 1 ) 1 1 Cov( Z i , X i ) Cov( Z i , X i ) If Cov( Z i , X i ) 0, the denominator equals 0, and the p lim does not exist. If Cov( Z , ) 0, then ˆ IV is inconsistent. i i Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 1 19-84 Review (cont.) • The asymptotic variance of IV is 1 2 p lim zi 1 2 1 2 Var(Zi ) n 2 n n Cov(Zi , X i )2 1 p lim zi xi n • The greater the covariance between X and Z, the lower the asymptotic variance. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-85 Review (cont.) • To estimate a multiple regression consistently, we need at least one instrumental variable for each troublesome explanator. Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-86 Review (cont.) • When we have just enough instruments for consistent estimation, we say the regression equation is exactly identified. • When we have more than enough instruments, the regression equation is over identified. • When we do not have enough instruments, the equation is under identified (and inconsistent). Copyright © 2006 Pearson Addison-Wesley. All rights reserved. 19-87