Empirical Methods in Applied Microeconomics
|
|
- Judith Matthews
- 5 years ago
- Views:
Transcription
1 Empirical Methods in Applied Microeconomics Jörn-Ste en Pischke LSE November Nonlinearity and Heterogeneity We have so far concentrated on the estimation of treatment e ects when the treatment e ect is a constant, i.e. E[y 1i jd i = 1] E[y 0i jd i = 1] = E[y 1i y 0i ] = : However, in reality treatment e ects may be heterogeneous, so that each individual has their own i. In this case, the di erent averages like E[y 1i y 0i ], the population average treatment e ect, E[y 1i jd i = 1] E[y 0i jd i = 1], the treatment e ect on the treated, and E[y 1i jd i = 0] E[y 0i jd i = 0], the treatment e ect on the untreated, may di er from each other. Moreover, with continuous treatments, like in the returns to schooling example, the relationship between y i (earnings) and s i (the number of years of schooling) may be nonlinear, i.e. the return to the 10th year of schooling may be di erent from the 16th year of schooling. Nevertheless, simple linear regression and 2SLS still provide important tools, and possibly the most important tools, to analyze the data, and summarize the results. First, it is important to note that OLS provides the best linear approximation to the population conditional expectation function E[y 1i jx i ]. Hence, linear regression always has a legitimacy, independent of the true functional of E[y 1i jx i ]. The same is not true of other estimators, like GLS, which rely speci cally on the linearity of the regression function. Of course, whether regression estimates something causal depends on how we feel about the population conditional expectation function: if E[y 1i jx i ] represents a causal relationship, then regression helps us estimate that relationship. 1
2 If the population conditional expectation function is non-linear, regression will provide a weighted average derivative (see more on this below). Much recent econometrics has focused on using non-linear models and calculating average derivatives along the non-linear function. This often requires a substantial technical apparatus: OLS gives us such an average derivative right o the bat, is easy to implement, and doesn t involve any researcher choices as to estimtaor speci cs, kernels, bandwidths, etc. It is hence also easy to replicate. One case where the population conditional expectation function is clearly non-linear are linear dependent variable models, like the binary choice model, or regression models for censored variables. Many empirical researchers feel that these cases call for non-linear models like logit or probit for the binary choice model, or tobit for the censored regression model. However, the underlying regression function and their parameters in the probit or logit models are not of particular interest, compared to derivates (or marginal e ects) or the average e ects of switching a dummy regressor o and on, which have a direct interpretation in terms of treatment e ects. When the dependent variable only takes on positive values, as in the case of hours of work, many researchers argue we should be interested in the conditional on positive e ects (COP), which are, for example, estimated by the tobit model. However, these e ects have no direct causal interpretation. COP e ects are estimates of the form E[y i jy i > 0; d i = 1] E[y i jy i > 0; d i = 0]: Note that this conditions on an outcome: y i > 0, e.g. working positive hours. Hence, this identi es a valid causal e ect (in terms of y 1i and y 0i ) plus a selection bias. The reason (and algebra) is the same as we saw before when we conditioned on outcome variables on the right hand side of a regression. In order to get back to causal e ects, we need to combine the COP e ect with the participation e ect (going from y i = 0 to y i > 0) again. Regression automatically gives us a weighted average of these two e ects. Table in the back shows some estimates from a paper by Angrist and Evans (1998) investigating the e ect of the number of children on the employment and hours of mothers. The rst line in columns (2) to (4) show results from regressions of a dummy variable for mothers employment on whether the family has three or more kids. Column (2) is the OLS estimate, columns (3) and (4) are di erent average e ects from a probit model. They are all identical. E ects on work hours, which can be zero or positive, are in columns (2), (5) and (6). The OLS results are again very 2
3 similar to average e ects from a Tobit model. Colums (7) to (10) repeat the same exercise with the number of chilren, rather than just a dummy variable. It is sometimes argued that OLS coe cients will recover the average e ects from a probit as long as the mean of the dependent variable is close to 0.5 but not if it is further in the tail, where the probit function has more curvature. Panel B in the table investigates this case by focusing on college educated women over the age of 30 with older children, who have a much higher employment rate. However, the OLS and probit average e ects are still fairly close in this case, although not as close as in Panel A. 1.1 Controlling for Observables Matching When we talked about regression, we relied on the selection on observables or conditional independence assumption E[y 0i jx i ;d i ] = E[y 0i jx i ]. Instead of exploiting this condition using regression, the most obvious thing to do would be to exploit it directly, and compute, say the e ect of treatment on the treated E[y 1i y 0i jd i = 1] non-parametrically. By the law of iterated expectations and using the CIA E[y 1i y 0i jd i = 1] = E fe[y 1i y 0i jx i ; d i = 1]jd i = 1g E[y 1i y 0i jd i = 1] = E fe[y 1i jx i ; d i = 1] E[y 0i jx i ; d i = 0]jd i = 1g where = E fe[y i jx i ; d i = 1] E[y i jx i ; d i = 0]jd i = 1g = E f X jd i = 1g X = E[y i jx i ; d i = 1] E[y i jx i ; d i = 0]: X is simply the di erence in means between treated and untreated observations with covariate value X i = X. With discrete X s the e ect of treatment on the treated is M = E[y 1i y 0i jd i = 1] = X X X P (X i = Xjd i = 1) where P (X i = Xjd i = 1) is the histogram of X i among the treated. The sample analogue to this is the non-parametric matching estimator. Obviously X is only de ned when there are both treated and untreated observations for a particular covariate value. This means what we are estimating 3
4 in practice is actually M = E[y 1i y 0i jd i = 1; 0 < P (D = 1jX i = X) < 1]; i.e. the probability of treatment in the covariate cell cannot be 0 or 1. This is known as common support, and it highlights that matching is only feasible in the region of common support Matching using the Propensity Score Often there are continuous covariates or the covariate vector is high dimensional so that matching on every covariate combination is not feasible. An important result by Rosenbaum and Rubin (1983) implies that it is actually su cient to match on certain functions of the covariates, and in particular on the propensity score P (X i ) = P (d i = 1jX i = X) = E(d i jx i = X), which is the probability of treatment given X i = X. The propensity score theorem says if (y 1i ; y 0i )?d i jx i then (y 1i ; y 0i )?d i jp (X i ). To demonstrate this we will show that P [d i = 1jy 1i ;y 0i ; P (X i )] = P (X i ), implying independence of d i and counterfactual outcomes. P [d i = 1jy 1i ; y 0i ; P (X i )] = E[d i jy 1i ; y 0i ; P (X i )] = EfE[d i jy 1i ; y 0i ; P (X i ); X i ]jy 1i ; y 0i ; P (X i )g = EfE[d i jy 1i ; y 0i ; X i ]jy 1i ; y 0i ; P (X i )g = EfE[d i jx i ]jy 1i ; y 0i ; P (X i )g; by the CIA = EfP (X i )jy 1i ; y 0i ; P (X i )g = P (X i ): In practice, we will not know the propensity score. But it can be estimated in a rst step by a logit or probit regression on the relevant covariates, and possibly interactions of the covariates. We then proceed using the estimated propensity score ^P (X i ). There are various ways to construct estimators of treatment e ects based on matching on the propensity score, see, for example, Imbens (2004) for a survey. One upshot from the literature on matching is that the particular matching estimator used seems to play relatively little role for matching estimates. What is important is the choice of covariates (the CIA really needs to hold), and imposing common support. One useful way to think about propensity score matching estimates is the following weighting approach. It exploits the fact that the CIA implies yi d i E = E[y 1i ] P (X i ) yi (1 d i ) E = E[y 0i ]: (1 P (X i )) 4
5 Therefore, given a scheme for estimating P (X i ); we can construct estimates of the average treatment e ect from the sample analog of yi d i y i (1 d i ) E[y 1i y 0i ] = E P (X i ) 1 P (X i ) (di P (X i ))y i = E : (1) (1 P (X i ))P (X i ) We can similarly calculate the e ect of treatment on the treated from the sample analog of: (d i P (X i ))y i E[y 1i y 0i jd i = 1] = E : (2) (1 P (X i ))P (d i = 1) The idea that you can correct for non-random sampling by weighting by the reciprocal of the probability of selection dates back to Horvitz and Thompson (1952). The Horvitz-Thompson version of the propensity-score approach is appealing since the estimator is essentially automated, with no cumbersome matching required. The Horvitz-Thompson approach also highlights the close link between propensity-score matching and regression. Think of the regression regression of y i on d i, controlling for a saturated model for covariates. This estimator can be written R = E[(d i P (X i ))y i ] E[P (X i )(1 P (X i ))] : (3) The two Horvitz-Thompson matching estimators and the regression estimator are all members of the class of weighted average estimators considered by Hirano, Imbens, and Ridder (2003): E yi d i g(x i ) P (X i ) y i (1 d i ) ; (4) (1 P (X i )) where g(x i ) is a known weighting function. The weighting functions are g(x i ) = 1 g(x i ) = P (X i) P (d i =1) average treatment e ect e ect on the treated g(x i ) = P (X i)(1 P (X i )) E[P (X i )(1 P (X i ))] regression. This highlights that regression will not recover the treatment e ect on the treated but a di erently weighted average treatment e ect. The treatment e ect on the treated puts high weight on the part of the support of X i 5
6 where there are many treated observations (and hence P (X i ) is close to 1). OLS is the e cient estimator if the treatment e ect is the same for all covariate values. To achieve this e icency, OLS puts maximum weight on the observations where p(x i )(1 p(x i )) is large, which is p(x i ) = 0:5. This high variance area of the data carries the most information on the treatment e ect. A similar analogy holds between regression and direct covariate matching Regression versus Matching Given this close analogy between matching and regression, is it worthwhile going down the matching route? Or will simple OLS regression give you a satisfactory answer in many cases, even if you suspect that treatment may vary a lot in the population? I believe that regression will often su ce, and it should always be the rst line of attack. Proponents of matching (be it directly on the covariates or via the propensity score) have recently stressed that the key advantage of matching is that it forces you to consider the common support problem. Relying on regression may inadvertently lead you to extrapolate a lot from the cells that contain only treated or only control observations. In order to investigate this it is useful to consider the empirical example that has played a large role in the di usion of matching methods in econometrics: the evaluation of the National Supported Work (NSW) Demonstration, originally analyzed by Lalonde (1986). Lalonde compared the experimental results from the NSW study to those obtained with alternative methods including regression, using non-experimental control groups drawn from standard data sets like the CPS. Lalonde s conclusions were rather negative, nding that the non-experimental methods yielded variable results which were often not particularly close to the experimental estimates. Dehejia and Wahba (1991) reanalyzed the Lalonde data and found that they could replicate the experimental results well with matching methods using the propensity score. The NSW Demonstration was a program which provided work experience to individuals with particular social and economic problems, like previous unemployment. It was evaluated using a random assignment experiment. Lalonde (1986) created three potential control groups from the CPS (matched to Social Security earnings records ) for the NSW treatment group: The rst is the raw CPS sample, which is fairly representative of the population (CPS-1) as well as two subsamples, which were selected to mirror the characteristics of NSW enrollees more closely, based, for example, 6
7 on previous unemployment experience. We will present some results from both the broad (CPS-1) and the narrowest (CPS-3) comparison group. All samples are limited to men and to those observations for whom both 1974 and 1975 (pre-program) earnings are available. Table (from Dehejia and Wahba, 1991) shows the means of some demographic characteristics, and earnings before the program for the treatment group as well as for the various control groups. The table demonstrates that the NSW program group (and the experimental control group) is younger, less educated, more likely minority, and has much lower earnings than the general population (the CPS-1 sample). The CPS-3 sample matches the treatment group more closely but still shows some di erences, particularly in terms of race and pre-program earnings. Table displays results from various regression estimators of the NSW treatment e ect, using annual 1978 earnings as the outcome variable. The estimates using the experimental control group in column 1 are in the order of $1,600 to $1,800. As would be expected, these estimates vary little depending on the speci cation. Column 2 displays results for the CPS-1 sample. The raw di erence in earnings between NSW participants and the CPS sample is $-8,500, indicating that NSW participants earn substantially less than the program participants. This simply re ects the large selection bias present in the naive comparison with this sample. Successively including demographics and per-program earnings narrows the cap and the treatment e ect rises to a positive $800 in the last row. Results are slightly better in column 3, where we use the CPS-3 control group. This characteristics of this group are much closer to the NSW treatment group, and the raw di erence in earnings is only $-600. The estimate in the last row is close to $1,400, not far from the experimental treatment e ect. The last two columns of the table repeat the exercise using the method advocated by Imbens (2007). He suggests to start by estimating the propensity score, and then limiting the sample to one with enough overlap so that 0:1 < P (X i ) < 0:9. Once this is done, Imbens (2007) nds little di erence in treatment e ects from regression or matching estimators using the same NSW data. He concludes that imposing overlap is important while the particular estimator matters little once this done. We therefore implement this idea by rst selecting the sample, and then running simple regressions within the selected sample again. The same covariates are used in the calculation of the propensity score and in the second step regression. Estimates are displayed in the nal two columns of table 2. With just demographics or just 1975 earnings the Imbens style estimates di er little from the regression estimates in the earlier columns. However, once these 7
8 covariates are combined the treatment e ect estimates in the CPS-1 sample are somewhat closer to the experimental estimates than the pure regression estimates ( nal two rows). This is not true for the CPS-3 sample. This indicates that prior sample selection is not really necessary once we use the propensity score method to select the sample, and, in fact, it may be detrimental. The results also highlight that the 0:1 < P (X i ) < 0:9 rule may result in an empty sample. This happens in case where we use only the 1975 earnings for the CPS-1 sample. Since this indicates that the set of covariates is not very powerful to create enough variation in the propensity score it is a useful warning sign about the covariates. Table 3.32 also displays the means in the CPS-1 sample after pre-selecting on the propensity score using the full set of covariates. The comparison with the experimental group demonstrates that imposing common support does a good job in balancing most covariates. What to take away from this exercise? First and foremost, it is impressive what a good job regression does in the CPS-1 sample with a fairly limited number of covariates (and without any non-linearities or interaction terms other than age squared). Clearly this sample is extremely di erent from the treatment group, so the potential for extrapolating from outside the common support seems great in this example. Nevertheless, the estimate in the last row closes 90% of the gap in the raw data. Preselecting the sample by emulating the program admission criteria (as in the case of the CPS-3) sample yields even better estimates, as good as the propensity score method. This preselection seems like a sensible route since there is little a priori reason to start with the CPS-1 sample. The second nding is that the choice of covariates is more important than the choice of estimation method, and the choice of the original control sample once enough covariates are being used (compare di erences across rows versus di erences across columns). Finally, the propensity score method is able to improve on regression with the CPS-1 sample. The estimate of $1,400 may be signi cantly di erent from $800 for policy purposes (although the di erence is not huge compared to the standard errors). For example, these two estimates might yield di erent conclusions from a cost-bene t calculation for the program. This means that matching methods may have a role in a carefully designed study of treatment e ects. Nevertheless, this should not distract from the fact that regression should play an important role in evaluating programs like the NSW: it is simple to implement and hence to replicate, transparent, and it squarely puts the focus on the key issue of covariate selection, rather than other estimation issues. Hence, regression should be the starting point of any analysis invoking the CIA, and 8
9 often may be the nal word. 1.2 IV In analogy to the previous discussion we can ask what IV estimates when the treatment e ect is heterogeneous. In order to gain insight on this, consider the simplest case of a binary instrument and a binary endogenous regressor. Return to the IV assumptions we discussed earlier: Assumption 1 (Random assignment) z is as good as randomly assigned Assumption 2 (Exclusion) Y (z; D) = Y (z 0 ; D) 8z; z 0 ; D Assumption 3 (First stage) E(D(1) D(0)) 6= 0. To these assumptions add: Assumption 4 (Monotonicity) D i (1) D i (0) 0 8i The fourth assumption says that we need the instrument to act similarly on all observations: if it raises the probability of treatment for some individuals, it can t lower it for others (alternatively we could have D i (1) D i (0) 0). With these assumptions we get the LATE theorem (Imbens and Angrist, 1994). E(y i jz i = 1) E(y i jz i = 0) E(d i jz i = 1) E(d i jz i = 0) = E[y 1i y 0i jd i (1) > d i (0)]: The LATE theorem says that IV estimates the average e ect of the treatment among the subpopulation of individuals for whom d i (1) >d i (0). This subpopulation consists of those whose treatment status is changed by the instrument because the inequality is strict. Hence, it is a local average treatment e ect, and it depends on the particular instrument being used. With heterogeneous treatment e ects di erent instruments will in general identify di erent LATEs. One implication of this is that there is no overidenti cation test: the over-id test in essence tells us whether the LATEs for di erent instruments are the same, so it becomes a test of homogeneity. A quick proof of the LATE theorem is as follows: The exclusion restriction let s us write E(y i jz i = 1) = E[y 0i + (y 1i y 0i )d i jz i = 1], which equals 9
10 E[y 0i + (y 1i y 0i )d 1i ] by random assignment. Similarly E(y i jz i = 1) = E[y 0i + (y 1i y 0i )d 0i ]. Hence E(y i jz i = 1) E(y i jz i = 0) = E[y 0i + (y 1i y 0i ) d 1i ] E[y 0i + (y 1i y 0i ) d 0i ] = E[(y 1i y 0i ) (d 1i d 0i )] By monotonicity d 1i d 0i equals either zero or one so that E[(y 1i y 0i ) (d 1i d 0i )] = E[(y 1i y 0i ) jd 1i > d 0i ]P (d 1i > d 0i ): A similar argument shows that E(d i jz i = 1) E(d i jz i = 0) = P (d 1i >d 0i ), which completes the proof. In order to understand this result, it is useful to consider four subpopulations: d 1i = 0 1 d 0i = 0 never-taker complier 1 de er always-taker Never-takers are never treated, regardless of the value of the instrument. Always takers always take the treatment, also regardless of the instrument. Since the treatment status of never-takers and always-takers doesn t change, they do not contribute to the estimate of the treatment e ect (because of the exclusion restriction). Hence E(y i jz i = 1) E(y i jz i = 0) is only a ected by the compliers and de ers, the groups which have their treatment status changed by the instrument. Notice that compliers have their treatment status switched on, and de ers have their treatment status switched o as the instrument changes from 0 to 1. The estimate E(y i jz i = 1) E(y i jz i = 0) will be an average of the treatment e ect of the two groups. Since we don t know the relative size of these (unobserved) subpopulations, it would be di cult to interpret this average. The monotonicity assumption rules out the de ers. Hence, the e ect E(y i jz i = 1) E(y i jz i = 0) is just due to the compliers, i.e. the group with d 1i >d 0i. Hence, IV estimates the treatment e ect for compliers with the instrument. 1.3 Internal versus external validity The key of evaluation research is to estimate the causal e ect of a treatment, and to enhance our understanding of the treatment in question. The rst goal in analyzing an evaluation question is always to avoid selection bias. To put it di erently, the rst goal is the internal validity of the estimates. 10
11 Internal validity means that we actually get an estimate of the causal e ect of the treatment under study. Random assignment experiments tend to score well on internal validity. The Tennessee STAR experiment is likely to deliver internally valid estimates of sending children to classrooms of size 15 compared to 22 for very young children in Tennessee in the mid 1980s. Of course, this question is not of as much interest as the broader question: do smaller classes raise student achievement? For example, are the STAR estimates valid for the neighboring state of Kentucky, for other countries than the US? Do they still hold in 2005? Are the e ects the same in 8th grade? Are the e ects the same for reducing class size from 37 to 30 students? These are questions of external validity. We like to generalize and extrapolate from what we have learned in one setting. Whether we can do that is a question of the external validity of the estimates. Random assignment experiments are designed to overcome threats to internal validity. The external validity of an experiment may or may not be good, depending on how special the setting is. In medical drug trials, external validity may not be a big issue. If I try a cancer drug on a large enough group of cancer patients, the results are likely to hold for similar cancer patients elsewhere as well. In social experiments this may be less likely. If the e ect of class size is non-linear, the STAR results may not hold for larger class sizes. If the class size e ect is heterogenous, the STAR results may not hold in states with a very di erent make-up of student population. If the class size matters because students in big classes are more likely to be disruptive, and whether a student is disruptive depends on their age, then the STAR experiment may not be informative for older students, etc. 11
ECONOMETRICS II (ECO 2401) Victor Aguirregabiria. Spring 2018 TOPIC 4: INTRODUCTION TO THE EVALUATION OF TREATMENT EFFECTS
ECONOMETRICS II (ECO 2401) Victor Aguirregabiria Spring 2018 TOPIC 4: INTRODUCTION TO THE EVALUATION OF TREATMENT EFFECTS 1. Introduction and Notation 2. Randomized treatment 3. Conditional independence
More informationFlexible Estimation of Treatment Effect Parameters
Flexible Estimation of Treatment Effect Parameters Thomas MaCurdy a and Xiaohong Chen b and Han Hong c Introduction Many empirical studies of program evaluations are complicated by the presence of both
More informationSelection on Observables: Propensity Score Matching.
Selection on Observables: Propensity Score Matching. Department of Economics and Management Irene Brunetti ireneb@ec.unipi.it 24/10/2017 I. Brunetti Labour Economics in an European Perspective 24/10/2017
More informationPSC 504: Instrumental Variables
PSC 504: Instrumental Variables Matthew Blackwell 3/28/2013 Instrumental Variables and Structural Equation Modeling Setup e basic idea behind instrumental variables is that we have a treatment with unmeasured
More informationLecture 8. Roy Model, IV with essential heterogeneity, MTE
Lecture 8. Roy Model, IV with essential heterogeneity, MTE Economics 2123 George Washington University Instructor: Prof. Ben Williams Heterogeneity When we talk about heterogeneity, usually we mean heterogeneity
More informationEconomics 241B Estimation with Instruments
Economics 241B Estimation with Instruments Measurement Error Measurement error is de ned as the error resulting from the measurement of a variable. At some level, every variable is measured with error.
More informationIdenti cation of Positive Treatment E ects in. Randomized Experiments with Non-Compliance
Identi cation of Positive Treatment E ects in Randomized Experiments with Non-Compliance Aleksey Tetenov y February 18, 2012 Abstract I derive sharp nonparametric lower bounds on some parameters of the
More informationEmpirical approaches in public economics
Empirical approaches in public economics ECON4624 Empirical Public Economics Fall 2016 Gaute Torsvik Outline for today The canonical problem Basic concepts of causal inference Randomized experiments Non-experimental
More informationRecitation Notes 6. Konrad Menzel. October 22, 2006
Recitation Notes 6 Konrad Menzel October, 006 Random Coefficient Models. Motivation In the empirical literature on education and earnings, the main object of interest is the human capital earnings function
More informationSensitivity checks for the local average treatment effect
Sensitivity checks for the local average treatment effect Martin Huber March 13, 2014 University of St. Gallen, Dept. of Economics Abstract: The nonparametric identification of the local average treatment
More informationMichael Lechner Causal Analysis RDD 2014 page 1. Lecture 7. The Regression Discontinuity Design. RDD fuzzy and sharp
page 1 Lecture 7 The Regression Discontinuity Design fuzzy and sharp page 2 Regression Discontinuity Design () Introduction (1) The design is a quasi-experimental design with the defining characteristic
More informationInstrumental Variables. Ethan Kaplan
Instrumental Variables Ethan Kaplan 1 Instrumental Variables: Intro. Bias in OLS: Consider a linear model: Y = X + Suppose that then OLS yields: cov (X; ) = ^ OLS = X 0 X 1 X 0 Y = X 0 X 1 X 0 (X + ) =)
More informationII. MATCHMAKER, MATCHMAKER
II. MATCHMAKER, MATCHMAKER Josh Angrist MIT 14.387 Fall 2014 Agenda Matching. What could be simpler? We look for causal effects by comparing treatment and control within subgroups where everything... or
More informationSection 10: Inverse propensity score weighting (IPSW)
Section 10: Inverse propensity score weighting (IPSW) Fall 2014 1/23 Inverse Propensity Score Weighting (IPSW) Until now we discussed matching on the P-score, a different approach is to re-weight the observations
More informationWhat s New in Econometrics. Lecture 1
What s New in Econometrics Lecture 1 Estimation of Average Treatment Effects Under Unconfoundedness Guido Imbens NBER Summer Institute, 2007 Outline 1. Introduction 2. Potential Outcomes 3. Estimands and
More informationQuantitative Economics for the Evaluation of the European Policy
Quantitative Economics for the Evaluation of the European Policy Dipartimento di Economia e Management Irene Brunetti Davide Fiaschi Angela Parenti 1 25th of September, 2017 1 ireneb@ec.unipi.it, davide.fiaschi@unipi.it,
More informationEmpirical Methods in Applied Economics Lecture Notes
Empirical Methods in Applied Economics Lecture Notes Jörn-Ste en Pischke LSE October 2005 1 Regression Discontinuity Design 1.1 Basics and the Sharp Design The basic idea of the regression discontinuity
More informationEvaluating Nonexperimental Estimators for Multiple Treatments: Evidence from a Randomized Experiment
Evaluating Nonexperimental Estimators for Multiple Treatments: Evidence from a Randomized Experiment Carlos A. Flores y Oscar A. Mitnik z December 22, 2008 Preliminary and Incomplete - Comments Welcome
More informationThe problem of causality in microeconometrics.
The problem of causality in microeconometrics. Andrea Ichino University of Bologna and Cepr June 11, 2007 Contents 1 The Problem of Causality 1 1.1 A formal framework to think about causality....................................
More informationMatching using Semiparametric Propensity Scores
Matching using Semiparametric Propensity Scores Gregory Kordas Department of Economics University of Pennsylvania kordas@ssc.upenn.edu Steven F. Lehrer SPS and Department of Economics Queen s University
More informationEmpirical Methods in Applied Economics
Empirical Methods in Applied Economics Jörn-Ste en Pischke LSE October 2007 1 Instrumental Variables 1.1 Basics A good baseline for thinking about the estimation of causal e ects is often the randomized
More informationPotential Outcomes Model (POM)
Potential Outcomes Model (POM) Relationship Between Counterfactual States Causality Empirical Strategies in Labor Economics, Angrist Krueger (1999): The most challenging empirical questions in economics
More informationMatching Techniques. Technical Session VI. Manila, December Jed Friedman. Spanish Impact Evaluation. Fund. Region
Impact Evaluation Technical Session VI Matching Techniques Jed Friedman Manila, December 2008 Human Development Network East Asia and the Pacific Region Spanish Impact Evaluation Fund The case of random
More informationIntroduction to causal identification. Nidhiya Menon IGC Summer School, New Delhi, July 2015
Introduction to causal identification Nidhiya Menon IGC Summer School, New Delhi, July 2015 Outline 1. Micro-empirical methods 2. Rubin causal model 3. More on Instrumental Variables (IV) Estimating causal
More informationLecture 11 Roy model, MTE, PRTE
Lecture 11 Roy model, MTE, PRTE Economics 2123 George Washington University Instructor: Prof. Ben Williams Roy Model Motivation The standard textbook example of simultaneity is a supply and demand system
More informationRecitation Notes 5. Konrad Menzel. October 13, 2006
ecitation otes 5 Konrad Menzel October 13, 2006 1 Instrumental Variables (continued) 11 Omitted Variables and the Wald Estimator Consider a Wald estimator for the Angrist (1991) approach to estimating
More informationESTIMATING AVERAGE TREATMENT EFFECTS: REGRESSION DISCONTINUITY DESIGNS Jeff Wooldridge Michigan State University BGSE/IZA Course in Microeconometrics
ESTIMATING AVERAGE TREATMENT EFFECTS: REGRESSION DISCONTINUITY DESIGNS Jeff Wooldridge Michigan State University BGSE/IZA Course in Microeconometrics July 2009 1. Introduction 2. The Sharp RD Design 3.
More informationMC3: Econometric Theory and Methods. Course Notes 4
University College London Department of Economics M.Sc. in Economics MC3: Econometric Theory and Methods Course Notes 4 Notes on maximum likelihood methods Andrew Chesher 25/0/2005 Course Notes 4, Andrew
More informationA Course in Applied Econometrics. Lecture 2 Outline. Estimation of Average Treatment Effects. Under Unconfoundedness, Part II
A Course in Applied Econometrics Lecture Outline Estimation of Average Treatment Effects Under Unconfoundedness, Part II. Assessing Unconfoundedness (not testable). Overlap. Illustration based on Lalonde
More informationBlinder-Oaxaca as a Reweighting Estimator
Blinder-Oaxaca as a Reweighting Estimator By Patrick Kline A large literature focuses on the use of propensity score methods as a semi-parametric alternative to regression for estimation of average treatments
More informationEvaluating Nonexperimental Estimators for Multiple Treatments: Evidence from a Randomized Experiment
Evaluating Nonexperimental Estimators for Multiple Treatments: Evidence from a Randomized Experiment Carlos A. Flores y Oscar A. Mitnik z May 5, 2009 Preliminary and Incomplete Abstract This paper assesses
More informationCausal Inference Lecture Notes: Causal Inference with Repeated Measures in Observational Studies
Causal Inference Lecture Notes: Causal Inference with Repeated Measures in Observational Studies Kosuke Imai Department of Politics Princeton University November 13, 2013 So far, we have essentially assumed
More informationA Measure of Robustness to Misspecification
A Measure of Robustness to Misspecification Susan Athey Guido W. Imbens December 2014 Graduate School of Business, Stanford University, and NBER. Electronic correspondence: athey@stanford.edu. Graduate
More informationECON 594: Lecture #6
ECON 594: Lecture #6 Thomas Lemieux Vancouver School of Economics, UBC May 2018 1 Limited dependent variables: introduction Up to now, we have been implicitly assuming that the dependent variable, y, was
More informationExercise Sheet 4 Instrumental Variables and Two Stage Least Squares Estimation
Exercise Sheet 4 Instrumental Variables and Two Stage Least Squares Estimation ECONOMETRICS I. UC3M 1. [W 15.1] Consider a simple model to estimate the e ect of personal computer (P C) ownership on the
More informationCausal Inference with Big Data Sets
Causal Inference with Big Data Sets Marcelo Coca Perraillon University of Colorado AMC November 2016 1 / 1 Outlone Outline Big data Causal inference in economics and statistics Regression discontinuity
More informationThe returns to schooling, ability bias, and regression
The returns to schooling, ability bias, and regression Jörn-Steffen Pischke LSE October 4, 2016 Pischke (LSE) Griliches 1977 October 4, 2016 1 / 44 Counterfactual outcomes Scholing for individual i is
More informationEconometrics of causal inference. Throughout, we consider the simplest case of a linear outcome equation, and homogeneous
Econometrics of causal inference Throughout, we consider the simplest case of a linear outcome equation, and homogeneous effects: y = βx + ɛ (1) where y is some outcome, x is an explanatory variable, and
More informationRegression Discontinuity
Regression Discontinuity Christopher Taber Department of Economics University of Wisconsin-Madison October 16, 2018 I will describe the basic ideas of RD, but ignore many of the details Good references
More informationGov 2002: 4. Observational Studies and Confounding
Gov 2002: 4. Observational Studies and Confounding Matthew Blackwell September 10, 2015 Where are we? Where are we going? Last two weeks: randomized experiments. From here on: observational studies. What
More informationA Course in Applied Econometrics. Lecture 5. Instrumental Variables with Treatment Effect. Heterogeneity: Local Average Treatment Effects.
A Course in Applied Econometrics Lecture 5 Outline. Introduction 2. Basics Instrumental Variables with Treatment Effect Heterogeneity: Local Average Treatment Effects 3. Local Average Treatment Effects
More informationFour Parameters of Interest in the Evaluation. of Social Programs. James J. Heckman Justin L. Tobias Edward Vytlacil
Four Parameters of Interest in the Evaluation of Social Programs James J. Heckman Justin L. Tobias Edward Vytlacil Nueld College, Oxford, August, 2005 1 1 Introduction This paper uses a latent variable
More informationNew Developments in Econometrics Lecture 11: Difference-in-Differences Estimation
New Developments in Econometrics Lecture 11: Difference-in-Differences Estimation Jeff Wooldridge Cemmap Lectures, UCL, June 2009 1. The Basic Methodology 2. How Should We View Uncertainty in DD Settings?
More informationGroupe de lecture. Instrumental Variables Estimates of the Effect of Subsidized Training on the Quantiles of Trainee Earnings. Abadie, Angrist, Imbens
Groupe de lecture Instrumental Variables Estimates of the Effect of Subsidized Training on the Quantiles of Trainee Earnings Abadie, Angrist, Imbens Econometrica (2002) 02 décembre 2010 Objectives Using
More informationExperiments and Quasi-Experiments
Experiments and Quasi-Experiments (SW Chapter 13) Outline 1. Potential Outcomes, Causal Effects, and Idealized Experiments 2. Threats to Validity of Experiments 3. Application: The Tennessee STAR Experiment
More informationIV Estimation WS 2014/15 SS Alexander Spermann. IV Estimation
SS 2010 WS 2014/15 Alexander Spermann Evaluation With Non-Experimental Approaches Selection on Unobservables Natural Experiment (exogenous variation in a variable) DiD Example: Card/Krueger (1994) Minimum
More informationDynamics in Social Networks and Causality
Web Science & Technologies University of Koblenz Landau, Germany Dynamics in Social Networks and Causality JProf. Dr. University Koblenz Landau GESIS Leibniz Institute for the Social Sciences Last Time:
More informationIntroduction to Econometrics
Introduction to Econometrics STAT-S-301 Experiments and Quasi-Experiments (2016/2017) Lecturer: Yves Dominicy Teaching Assistant: Elise Petit 1 Why study experiments? Ideal randomized controlled experiments
More informationRegression Discontinuity
Regression Discontinuity Christopher Taber Department of Economics University of Wisconsin-Madison October 24, 2017 I will describe the basic ideas of RD, but ignore many of the details Good references
More informationLecture 11/12. Roy Model, MTE, Structural Estimation
Lecture 11/12. Roy Model, MTE, Structural Estimation Economics 2123 George Washington University Instructor: Prof. Ben Williams Roy model The Roy model is a model of comparative advantage: Potential earnings
More informationEMERGING MARKETS - Lecture 2: Methodology refresher
EMERGING MARKETS - Lecture 2: Methodology refresher Maria Perrotta April 4, 2013 SITE http://www.hhs.se/site/pages/default.aspx My contact: maria.perrotta@hhs.se Aim of this class There are many different
More informationInstrumental Variables in Action: Sometimes You get What You Need
Instrumental Variables in Action: Sometimes You get What You Need Joshua D. Angrist MIT and NBER May 2011 Introduction Our Causal Framework A dummy causal variable of interest, i, is called a treatment,
More informationWhat s New in Econometrics? Lecture 14 Quantile Methods
What s New in Econometrics? Lecture 14 Quantile Methods Jeff Wooldridge NBER Summer Institute, 2007 1. Reminders About Means, Medians, and Quantiles 2. Some Useful Asymptotic Results 3. Quantile Regression
More informationThe Problem of Causality in the Analysis of Educational Choices and Labor Market Outcomes Slides for Lectures
The Problem of Causality in the Analysis of Educational Choices and Labor Market Outcomes Slides for Lectures Andrea Ichino (European University Institute and CEPR) February 28, 2006 Abstract This course
More informationWhat do instrumental variable models deliver with discrete dependent variables?
What do instrumental variable models deliver with discrete dependent variables? Andrew Chesher Adam Rosen The Institute for Fiscal Studies Department of Economics, UCL cemmap working paper CWP10/13 What
More informationInstrumental Variables
Instrumental Variables Yona Rubinstein July 2016 Yona Rubinstein (LSE) Instrumental Variables 07/16 1 / 31 The Limitation of Panel Data So far we learned how to account for selection on time invariant
More informationSimple Estimators for Semiparametric Multinomial Choice Models
Simple Estimators for Semiparametric Multinomial Choice Models James L. Powell and Paul A. Ruud University of California, Berkeley March 2008 Preliminary and Incomplete Comments Welcome Abstract This paper
More informationIntroduction: structural econometrics. Jean-Marc Robin
Introduction: structural econometrics Jean-Marc Robin Abstract 1. Descriptive vs structural models 2. Correlation is not causality a. Simultaneity b. Heterogeneity c. Selectivity Descriptive models Consider
More informationLecture Notes on Measurement Error
Steve Pischke Spring 2000 Lecture Notes on Measurement Error These notes summarize a variety of simple results on measurement error which I nd useful. They also provide some references where more complete
More informationMid-term exam Practice problems
Mid-term exam Practice problems Most problems are short answer problems. You receive points for the answer and the explanation. Full points require both, unless otherwise specified. Explaining your answer
More informationLecture 4: Linear panel models
Lecture 4: Linear panel models Luc Behaghel PSE February 2009 Luc Behaghel (PSE) Lecture 4 February 2009 1 / 47 Introduction Panel = repeated observations of the same individuals (e.g., rms, workers, countries)
More informationECO Class 6 Nonparametric Econometrics
ECO 523 - Class 6 Nonparametric Econometrics Carolina Caetano Contents 1 Nonparametric instrumental variable regression 1 2 Nonparametric Estimation of Average Treatment Effects 3 2.1 Asymptotic results................................
More informationEconomics 620, Lecture 18: Nonlinear Models
Economics 620, Lecture 18: Nonlinear Models Nicholas M. Kiefer Cornell University Professor N. M. Kiefer (Cornell University) Lecture 18: Nonlinear Models 1 / 18 The basic point is that smooth nonlinear
More informationWageningen Summer School in Econometrics. The Bayesian Approach in Theory and Practice
Wageningen Summer School in Econometrics The Bayesian Approach in Theory and Practice September 2008 Slides for Lecture on Qualitative and Limited Dependent Variable Models Gary Koop, University of Strathclyde
More informationEconomics 620, Lecture 19: Introduction to Nonparametric and Semiparametric Estimation
Economics 620, Lecture 19: Introduction to Nonparametric and Semiparametric Estimation Nicholas M. Kiefer Cornell University Professor N. M. Kiefer (Cornell University) Lecture 19: Nonparametric Analysis
More informationEnvironmental Econometrics
Environmental Econometrics Syngjoo Choi Fall 2008 Environmental Econometrics (GR03) Fall 2008 1 / 37 Syllabus I This is an introductory econometrics course which assumes no prior knowledge on econometrics;
More informationMicroeconometrics. Bernd Süssmuth. IEW Institute for Empirical Research in Economics. University of Leipzig. April 4, 2011
Microeconometrics Bernd Süssmuth IEW Institute for Empirical Research in Economics University of Leipzig April 4, 2011 Bernd Süssmuth (University of Leipzig) Microeconometrics April 4, 2011 1 / 22 Organizational
More informationEvaluating Nonexperimental Estimators for Multiple Treatments: Evidence from Experimental Data
Evaluating Nonexperimental Estimators for Multiple Treatments: Evidence from Experimental Data Carlos A. Flores y Oscar A. Mitnik z September 2009 Abstract This paper assesses the e ectiveness of unconfoundedness-based
More information1 Correlation between an independent variable and the error
Chapter 7 outline, Econometrics Instrumental variables and model estimation 1 Correlation between an independent variable and the error Recall that one of the assumptions that we make when proving the
More informationEducation Production Functions. April 7, 2009
Education Production Functions April 7, 2009 Outline I Production Functions for Education Hanushek Paper Card and Krueger Tennesee Star Experiment Maimonides Rule What do I mean by Production Function?
More informationECON Introductory Econometrics. Lecture 17: Experiments
ECON4150 - Introductory Econometrics Lecture 17: Experiments Monique de Haan (moniqued@econ.uio.no) Stock and Watson Chapter 13 Lecture outline 2 Why study experiments? The potential outcome framework.
More informationRank preserving Structural Nested Distribution Model (RPSNDM) for Continuous
Rank preserving Structural Nested Distribution Model (RPSNDM) for Continuous Y : X M Y a=0 = Y a a m = Y a cum (a) : Y a = Y a=0 + cum (a) an unknown parameter. = 0, Y a = Y a=0 = Y for all subjects Rank
More informationThe Generalized Roy Model and Treatment Effects
The Generalized Roy Model and Treatment Effects Christopher Taber University of Wisconsin November 10, 2016 Introduction From Imbens and Angrist we showed that if one runs IV, we get estimates of the Local
More informationEconometrics II. Nonstandard Standard Error Issues: A Guide for the. Practitioner
Econometrics II Nonstandard Standard Error Issues: A Guide for the Practitioner Måns Söderbom 10 May 2011 Department of Economics, University of Gothenburg. Email: mans.soderbom@economics.gu.se. Web: www.economics.gu.se/soderbom,
More informationMethods to Estimate Causal Effects Theory and Applications. Prof. Dr. Sascha O. Becker U Stirling, Ifo, CESifo and IZA
Methods to Estimate Causal Effects Theory and Applications Prof. Dr. Sascha O. Becker U Stirling, Ifo, CESifo and IZA last update: 21 August 2009 Preliminaries Address Prof. Dr. Sascha O. Becker Stirling
More informationPrinciples Underlying Evaluation Estimators
The Principles Underlying Evaluation Estimators James J. University of Chicago Econ 350, Winter 2019 The Basic Principles Underlying the Identification of the Main Econometric Evaluation Estimators Two
More informationEcon 2148, fall 2017 Instrumental variables I, origins and binary treatment case
Econ 2148, fall 2017 Instrumental variables I, origins and binary treatment case Maximilian Kasy Department of Economics, Harvard University 1 / 40 Agenda instrumental variables part I Origins of instrumental
More informationBounds on Average and Quantile Treatment E ects of Job Corps Training on Participants Wages
Bounds on Average and Quantile Treatment E ects of Job Corps Training on Participants Wages German Blanco Food and Resource Economics Department, University of Florida gblancol@u.edu Carlos A. Flores Department
More informationECONOMETRICS FIELD EXAM Michigan State University May 9, 2008
ECONOMETRICS FIELD EXAM Michigan State University May 9, 2008 Instructions: Answer all four (4) questions. Point totals for each question are given in parenthesis; there are 00 points possible. Within
More informationBounds on Population Average Treatment E ects with an Instrumental Variable
Bounds on Population Average Treatment E ects with an Instrumental Variable Preliminary (Rough) Draft Xuan Chen y Carlos A. Flores z Alfonso Flores-Lagunes x October 2013 Abstract We derive nonparametric
More informationRewrap ECON November 18, () Rewrap ECON 4135 November 18, / 35
Rewrap ECON 4135 November 18, 2011 () Rewrap ECON 4135 November 18, 2011 1 / 35 What should you now know? 1 What is econometrics? 2 Fundamental regression analysis 1 Bivariate regression 2 Multivariate
More informationSingle-Equation GMM: Endogeneity Bias
Single-Equation GMM: Lecture for Economics 241B Douglas G. Steigerwald UC Santa Barbara January 2012 Initial Question Initial Question How valuable is investment in college education? economics - measure
More informationSIMULATION-BASED SENSITIVITY ANALYSIS FOR MATCHING ESTIMATORS
SIMULATION-BASED SENSITIVITY ANALYSIS FOR MATCHING ESTIMATORS TOMMASO NANNICINI universidad carlos iii de madrid UK Stata Users Group Meeting London, September 10, 2007 CONTENT Presentation of a Stata
More informationWrite your identification number on each paper and cover sheet (the number stated in the upper right hand corner on your exam cover).
Formatmall skapad: 2011-12-01 Uppdaterad: 2015-03-06 / LP Department of Economics Course name: Empirical Methods in Economics 2 Course code: EC2404 Semester: Spring 2015 Type of exam: MAIN Examiner: Peter
More informationThe problem of causality in microeconometrics.
The problem of causality in microeconometrics. Andrea Ichino European University Institute April 15, 2014 Contents 1 The Problem of Causality 1 1.1 A formal framework to think about causality....................................
More informationImplementing Matching Estimators for. Average Treatment Effects in STATA
Implementing Matching Estimators for Average Treatment Effects in STATA Guido W. Imbens - Harvard University West Coast Stata Users Group meeting, Los Angeles October 26th, 2007 General Motivation Estimation
More informationInstrumental Variables in Action
Instrumental Variables in Action Remarks in honor of P.G. Wright s 150th birthday Joshua D. Angrist MIT and NBER October 2011 What is Econometrics Anyway? What s the difference between statistics and econometrics?
More informationEconometrics Lecture 5: Limited Dependent Variable Models: Logit and Probit
Econometrics Lecture 5: Limited Dependent Variable Models: Logit and Probit R. G. Pierse 1 Introduction In lecture 5 of last semester s course, we looked at the reasons for including dichotomous variables
More informationh=1 exp (X : J h=1 Even the direction of the e ect is not determined by jk. A simpler interpretation of j is given by the odds-ratio
Multivariate Response Models The response variable is unordered and takes more than two values. The term unordered refers to the fact that response 3 is not more favored than response 2. One choice from
More informationfinite-sample optimal estimation and inference on average treatment effects under unconfoundedness
finite-sample optimal estimation and inference on average treatment effects under unconfoundedness Timothy Armstrong (Yale University) Michal Kolesár (Princeton University) September 2017 Introduction
More informationTreatment Effects. Christopher Taber. September 6, Department of Economics University of Wisconsin-Madison
Treatment Effects Christopher Taber Department of Economics University of Wisconsin-Madison September 6, 2017 Notation First a word on notation I like to use i subscripts on random variables to be clear
More informationIntroduction to Econometrics
Introduction to Econometrics Lecture 3 : Regression: CEF and Simple OLS Zhaopeng Qu Business School,Nanjing University Oct 9th, 2017 Zhaopeng Qu (Nanjing University) Introduction to Econometrics Oct 9th,
More information1 A Non-technical Introduction to Regression
1 A Non-technical Introduction to Regression Chapters 1 and Chapter 2 of the textbook are reviews of material you should know from your previous study (e.g. in your second year course). They cover, in
More informationA Course on Advanced Econometrics
A Course on Advanced Econometrics Yongmiao Hong The Ernest S. Liu Professor of Economics & International Studies Cornell University Course Introduction: Modern economies are full of uncertainties and risk.
More informationA Course in Applied Econometrics Lecture 18: Missing Data. Jeff Wooldridge IRP Lectures, UW Madison, August Linear model with IVs: y i x i u i,
A Course in Applied Econometrics Lecture 18: Missing Data Jeff Wooldridge IRP Lectures, UW Madison, August 2008 1. When Can Missing Data be Ignored? 2. Inverse Probability Weighting 3. Imputation 4. Heckman-Type
More informationRegression Discontinuity Designs.
Regression Discontinuity Designs. Department of Economics and Management Irene Brunetti ireneb@ec.unipi.it 31/10/2017 I. Brunetti Labour Economics in an European Perspective 31/10/2017 1 / 36 Introduction
More informationTime Series Models and Inference. James L. Powell Department of Economics University of California, Berkeley
Time Series Models and Inference James L. Powell Department of Economics University of California, Berkeley Overview In contrast to the classical linear regression model, in which the components of the
More informationPrediction and causal inference, in a nutshell
Prediction and causal inference, in a nutshell 1 Prediction (Source: Amemiya, ch. 4) Best Linear Predictor: a motivation for linear univariate regression Consider two random variables X and Y. What is
More informationCausal Inference with General Treatment Regimes: Generalizing the Propensity Score
Causal Inference with General Treatment Regimes: Generalizing the Propensity Score David van Dyk Department of Statistics, University of California, Irvine vandyk@stat.harvard.edu Joint work with Kosuke
More informationResearch Design: Causal inference and counterfactuals
Research Design: Causal inference and counterfactuals University College Dublin 8 March 2013 1 2 3 4 Outline 1 2 3 4 Inference In regression analysis we look at the relationship between (a set of) independent
More information