The FIRE Project: FIRE3

Size: px
Start display at page:

Download "The FIRE Project: FIRE3"

Transcription

1 The FIRE Project: FIRE3 Richard Butler, Brigham Young University 3 Identification Strategies Largely from Lars Lefgren s BYU (econ 488) class notes. The notation and some of the concepts in this section come from Heckman and Smith (1995), and Chay (2001). Another reference coming to the author s attention after this chapter was drafted, was Angrist s Mostly Harmless Econometrics, which has a more sophisticated discussion of these same issues. 3.1 Covariance May Not Be Causal OLS estimation seems eminently reasonable, but is it also statistically justified as a tool for retrieving casual estimates that can be used for testing theories or predicting outcomes? Note in the simple regression case (with just one slope variable), that the normal equations for ˆβ 1 yields an estimator that converges to cov(x, y)/var(x), where cov(x, y) is the covariance between x and y, and var(x) is the variance of x. Does a positive covariance mean its a casual relationship, in the sense that as x increases, then y will increase? No, covariance doesn t necessarily indicate causality. Consider urban, insured fire-loss damage. In cities, for one-alarm fires, relatively few crews show up to battle the blaze. In a two-alarm fire, more crews come, etc. After we collected a lot of these fire-loss incidents, we regress insured losses on the number of alarms sounded. We find a strong, positive, linear relation: the more alarms, the greater the insured losses. Is a policy implication that we can minimize insured losses by only sounding one alarm for all fires? Of course not. The problem is that anticipated insured losses and number of alarms sounded are jointly determined. The relationship is not causal. Covariance does not imply causality. Another example: do low credit scores causally determine insurance claims? If that were the case, shouldn t insurers loss prevention efforts for some lines of insurance focus on credit counseling? What is an appropriate x and y, vary by questions being addressed and the sample information at hand: one person s x variable may be another person s y variable. As suggested above, there be covariance between any two variables for a number of reasons. The first, and for our purposes, the best reason for the covariance to be positive (or negative) is that there may actually be a unidirectional causal relationship, x causes y. Note the emphasis on getting the direction of causality correct: it is possible that we 1

2 have confused the roles of x and y in our analysis, so rather than mistakingly regressing the number of children on accrued dependent dental costs (so the model assumes accrued dependent dental costs increase the number of children): children i = β 0 + β 1 dental expenses i + µ i the causality is actually just reversed from the specification above, and we should be regressing accrued dependent dental costs on the number of children. The regression immediately above is a case of a reverse causal relationship, the y in our model is actually causing the x in our model specified in the above regression; we have confused their respective roles in the specification immediately above. Frequently, there may some third variable driving both our observed x and y. Come back to the example of credit scores and insurance losses. Suppose that individuals vary by their aversion to risk, with those who are most risk averse being careful with their credit cards and also being relatively more careful while they are driving. Hence, the negative covariance between credit score and insured losses is actually the result of a more complex interaction with one or more other variables not in our model (namely, risk aversion in this case). Finally, there may be a persistent covariance because x and y are jointly determined. This is almost always the case in markets where the relationships between price and quantity is being estimated. For example, the amount of term life insurance policies demanded depends on the premium charged. On the other hand, the number of such policies offered also depends on the premiums. Premiums, and quantity of term insurance coverage transacted, are jointly determined. (This is frequently called endogeneity.) 3.2 Identification: Causal Relationships With Discrete Treatment Variables Consider an outcome, home maintenance expenditures, $H, and a binary treatment variable, D, which is a damage rider for tornado-related property losses. The FIRE researcher wants to estimate the impact of a recent mandate that all homeowners policies in the state must offer a tornado-damage rider if they market insurance in that state (or one could imagine the state government changing insurance in another way: dictating that the rider cost no more than a low, specified maximum amount). The causal response to be estimated is How does a home owner maintain her property after she is chosen at random to get the tornado rider?, or Among those purchasing the rider, what happens when they get the rider relative to what they would do if they had not gotten the rider? (or, what is the Average Treatment Effect for everyone? and what is the Average Treatment Effect for those getting the treatment?) For each individual, the home maintenance expenditures with the rider, is $H 1, or the home maintenance expenditures without rider, is $H 0. In the parlance of this literature, the treatment group is those who have purchased the insurance rider, and the control group are those who didn t purchase the rider. The causal issues here is whether the rider induces a change of behavior in those purchasing the rider (moral hazard), or if those 2

3 purchasing do so because they already don t take care of their property and so know that they are at greater risk if there is a storm (adverse selection). The treatment for a particular individual is $H 1 $H 0. The fundamental problem of causal identification is that we generally observe individuals in the treated state or the untreated state (control state), but not both. The estimation problem for the relevant population is to estimate an average treatment effect (ATE): or, ATE (everyone) = E($H 1 ) E($H 0 ) ATE (for the treated) = E($H 1 D = 1) E($H 0 D = 1) = E($H 1 $H 0 D = 1) Where the conditional expectation has the same meaning as it did with assumption 3 in the first chapter: E($H 1 D = 1) are the average expenditures among those chosing to buy the rider, and E($H 0 D = 1) is the counterfactual that we don t observe: namely, what the average expenditures for those choosing to get the rider would have been if they did not get the rider. This notation allows us to consider how the rider might affect behavior. However, these desirable AT E listed above are not what we usually get by estimating the average mean differences in our samples. The expected outcome of the treated group is the average of their home maintenance expenses, given that they chose the rider E($H 1 D = 1). The expected outcome of the untreated group, those not choosing to buy the rider, is E($H 0 D = 0). In other words, we see the treated outcome of those who chose treatment. We see the untreated outcome of those who chose not to be treated. What does this observed difference in means relate to the AT E we want to estimate? E($H 1 D = 1) - E($H 0 D = 0) = E($H 1 D = 1) -E($H 0 D = 1) + E($H 0 D = 1) - E($H 0 D = 0) = E($H 1 $H 0 D = 1) + E($H 0 D = 1) - E($H 0 D = 0) (Polsky and Nicholson (2003) essentially use this framework to decompose HMO costs.) That is, the observed difference in means of those with and without the rider, equals the AT E for those being treated (E($H 1 $H 0 D = 1)) plus a term representing the sample selection (E($H 0 D = 1) - E($H 0 D = 0)). This sample selection term is the difference in home maintenance expenditures between the counterfactual expenditures for those acquiring the rider but asking what their as if behavior would have been without the rider (E($H 0 D = 1)), and those not acquiring the rider (E($H 0 D = 0)). The sample selection term is the bias between what is observed, and the AT E for the treated. So to get the observed difference equal to the AT E for the treated group, we need E($H 0 D = 1) =E($H 0 D = 0) 3

4 that is, no sample selection between groups for the $H 0 side: the counterfactual behavior of those getting the treatment E($H 0 D = 1)), would have have been have been the same as those not getting the treatment E($H 0 D = 0)), if the (D = 1) group had in fact not gotten the treatment (as if the treatment had been randomly assigned). To get the observed difference to equal the AT E for everyone, we also need: E($H 1 D = 1) =E($H 1 D = 0) or, going in the other counterfactual direction, there is no sample selection between groups on the $H 1 side. In summary, we get only if E($H 1 D = 1) - E($H 0 D = 0) = E($H 1 ) -E($H 0 ) E($H 1 D = 1) =E($H 1 D = 0) and E($H 0 D = 1) =E($H 0 D = 0) This means that the average difference between treated and untreated individuals corresponds to the treatment effect only if treated and untreated individuals are the same except for treatment status. That is, there is no sample selection. Suppose that our data generating process is $H i = D i α + X i β + µ i Then, holding X i constant between groups, we get E($H 1 i D = 1, X i ) - E($H 0 i D = 0, X i ) = α + X i β + [E(µ i D = 1) E(µ i D = 0)] Hence, a big problem with biased estimation of the treatment effect (ˆα) is with the unobservables in the model ( [E(µ i D = 1) E(µ i D = 0)]), even when we are holding the observables (X i ) constant. When are we likely to be able to get rid of the sample selection effect, and get consistent AT E estimates? An identifying assumption is what you need to assume in order to identify the parameter of interest. So what is/are the identifying assumptions we can use to help us out? Random assignment. By construction, selection into treatment is random. This means that selection into the treatment is not related to the the ith person s observable (X i ) or unobservable (µ i ) characteristics. What do we identify with random assignment, or what can we learn from a random experiment? E($H 1 $H 0 participated in the random assignment experiment) Identifying assumption. There should be no sample selection effect: participation (getting the tornado rider) is uncorrelated to unobservable characteristics (things in the µ term of the regression). This would be plausible if participation is uncorrelated to observable characteristics, which we check by examining the differences in the means of the D = 1 and D = 0 groups. 4

5 OLS implementation. We model expected expenses as a linear function of observable characteristics and treatment status, and allow non-random selection to effect treatment through a linear function observable characteristics: $H i = X i β + D i α + ɛ i where we are employing matrix notation for generalizability, α is the treatment effect and ɛ is orthogonal to D (treatment status) in this population model. and So that, E($H D = 1, X) = Xβ + α E($H D = 0, X) = Xβ Identifying assumption in this regression context. Treatment is uncorrelated to the error term, e.g. uncorrelated with the unobservables in the model. Assessing plausibility: Treatment status (D, getting the rider under the random assignment experiment) is likely uncorrelated to unobservable factors (ɛ) if it is uncorrelated with observable characteristics (X). So check on the differences in means of the various variables in X between treatment and control status, or simply regress treatment on other X variables (that we did not use in outcome, $H, regression) in the data set, expecting to find an R 2 close to zero. Matching to make the control group comparable to the treatment group. In matching, we simulate random assignment by matching someone with the exact same X in both the D=1 and D=0 groups. (In practice, we match them as close as we can get.) This allows for an arbitrary (non-linear) relationship between X and the outcome. Treatment status is assumed to be random given a particular value of X. That is, the comparison is between a male, age 33 years of age, with a college degree, married, and a full time job who has the rider (D = 1) and a male, age 33 years of age, with a college degree, married, and a full time job who does not have the rider (D = 0) (you can see the problem of finding a E($H 1 D = 1, X) exact match if we additionally included number of children and detailed job occupation). Again, it is assumed that conditioning on the same X removes sample selection effects, so that: E($H 1 D = 1, X) = E($H 1 D = 0, X) and E($H 0 D = 1, X) = E($H 0 D = 0, X) A matching example that mimics these assumptions has been the study of identical twins who ended up with different levels of education. Since many other characteristics are held constant within each twin pair (gender, age, genetic endowment, etc), the resulting differences in earnings associated with differences in education represent a matched analysis of the effect of education on earnings. (Of course, what is not matched is which twin came out first; do you think that this could make a difference?) 5

6 Identifying assumption is that the treatment (insurance rider, D, in our example) is uncorrelated with the error terms of matched observations. Assessing plausibility Do matched observations have comparable characteristics? If we matched on gender, age, schooling and work status, is it also the case that they have similar occupations, wage and salary income, and metropolitan status (suppose these later were unmatched characteristics)? If these unmatched characteristics are also similar, then the matching procedure is much more convincing. Propensity score matching. Since there is a dimensionality problem with matching when there are many variables, or many values within each variable, matching by the means of propensity score is a convenient (sometimes essential) alternative. The Propensity Score Theorem (Rosenbaum and Rubin, 1983) is that if D is randomly assigned conditional on X (this is the notion behind matching, that for two observationally identical individuals, we can treat the insurance rider D as if it were randomly assigned), then it is a random assignment conditional on the propensity score. The propensity score is the likelihood of getting treated (D = 1) conditional on observables X. You estimate it by running a probit or logit on the relevant X. From this, you get a predicted likelihood (ˆp i ) of treatment for everyone in the sample, given their characteristics (again, their X i ). Matching someone who got treated, with someone not treated but with the same propensity score (predicated likelihood of treatment given X i ), results in a match. Then you can get the treatment effect simply by comparing the means of the matched sample. Or you might incorporate the propensity score (or a polynomial of propensity scores, ˆp i ), into your basic estimation ($H) equation. Assessing plausibility Same as for matching in general. Propensity scores can also give a sense of the non-random nature of the treatment (D): generate propensity scores for the whole sample, and look at the boxplots of those scores by treatment status (for the D=1 and D=0 groups), and see how much overlap in the distributions there is. The more overlap, the more like random assignment (on the basis of observables) there is, and the better the comparison. //*** STATA program to estimate propensity scores and do boxplots ***// logit D <BASELINE VARIABBLES, X > predict prob treat graph box prob treat over(d) //*** SAS program to estimate propensity scores and do boxplots ***// proc logistic data=patient variables descending; model D = <BASELINE VARIABLES, X> ; output out=propensity scores pred = prob treat; run; proc boxplot data=propensity scores; plot prob treat*d; run; 6

7 Before and after comparison (i.e., simple difference) This strategy takes the difference in outcomes for the same individual before and after treatment. Treatment effect= E($H 1 D = 1, after) E($H 0 D = 1, before) An important issue is what would those receiving treatment had done in the absence of receiving treatment, that is, is it the case that they would have done nothing but for the treatment, namely, that E($H 0 D = 1, after) = E($H 0 D = 1, before)? There is good empirical evidence that this last condition does not always hold. Indeed the Ashenfelter dip is the often noted (first by Ashenfelter, 1978) finding that the average earnings of program participants in employment and training programs usually decline during the period just prior to participation. That is, participant s behavior is different from nonparticipant s behavior, even before participation arises. In terms of regression analysis, the difference in outcomes by treatment status is estimated as an interrupted time series: $H t = β 1 + β 2 trend t + β 3 after t + β 4 (trend t after t ) + ɛ t where after is a dummy variable equaling one in the second period (of the before/after comparison sample). The coefficient on the trend t after t variable, β 4, indicates the effect of the rider on home maintenance expenses, if the identifying strategy is working. Identifying assumption. Is the timing of treatment uncorrelated to other unobserved determinants of the outcome (no Ashenfelter dip?) Intuitively, is the level of the outcome immediately before treatment reflective of what the outcome would have been immediately after treatment in the absence of treatment? Assessing plausibility. Was the timing of the treatment mostly random and unanticipated? Was the trend changing in anticipation of the treatment? Are observables changing at the time of treatment? Panel Data Techniques: Fixed Effects and Lagged Dependent Variables. As we saw in the examples above, causal inference is often ruined by unobservable confounders: in FIRE1, omitted variables or classical measurement error biased the estimates; in the examples above, sample selection biased the estimates. Having repeated observations on the same individuals can help get rid of some of the unobserved, confounding influences (that is, it helps to have a panel data set that has home maintenance expenditures each year for 10 years, for each of 1000 homeowners). Letting A i be time-invariant, unobserved factors for individual i, and assume E($H 0 it A i, X it, D it ) = E($H 0 it A i, X it ), where time trends may now be included in the X it matrix of variables. The above equality says that controlling for the time invariant factors (completed schooling before the sample was begun, family background including parents use of insurance when growing up, genetic factors, etc), then the purchase of the rider, D it, is as if randomly assigned conditional on A i and X it. That is, the expected value of home maintenance expenditures before the 7

8 insurance rider is actually purchased ($H 0 it), will not depend upon whether the rider (D it ) was subsequently purchased, when controlling for A i and X it. So X it is observed, so its easy to condition on those variables, so we are left with conditioning on A i as well. How do we do that? As long as the A i have a linear impact on the outcome ($H), then we can demean the data, household by household just like we did in FIRE1 for our multiple regression example using Y, X 1, X 2, but with panel data sets we take deviations within each household rather than deviations within the whole sample (which we did in FIRE1). Or alternatively, we can just difference the data, year by year, within each household. There are advantages to both (see Wooldridge, 2015). If the error term in our equation is homoskedastic and serially uncorrelated, it is more efficient to use fixed effect models rather than the differencing models. (Statistical routines for demeaning within cohorts are given below in the appropriate FIRE lectures.) Alternatively, researchers have also conditioned on lagged values of the dependent variable, as follows: E($H 0 it $H i,t 1, X it, D it ) = E($H 0 it $H i,t 1, X it ), assuming that such conditioning generated an as if random assignment of the treatment. (There is an old literature on why this might be the case: whatever makes home maintenance expenditures unique this period, also made home maintenance expenditures unique last period within each household.) Since these two forms of conditioning are not nested (one is not a special case of the other), researchers have tried to generalize this conditioning with the following assumption: E($H 0 it A i, $H i,t 1, X it, D it ) = E($H 0 it A i, $H i,t 1, X it ). While this is less restrictive than the previous two assumptions, note that it places more restrictions on the error structured required for consistent estimation of the model. (See Angrist and Pischke, 2009, chapter 5.) Assessing plausibility. For all three versions of the model above, there are several ways to test for plausibility of the identifying assumptions. With sufficient data over time, one can estimate individual cohort time trends (time trends for each i group), and examine whether the results are robust to those trends. Where plausible, this should always be done. Also we can test the plausibility of the assumptions on making the D it treatments as if randomly assigned, by doing Granger (1969) type causality tests. This involves testing whether the treatment, the purchase of the insurance rider D it this year, affected past behavior (it should not, if the identification strategy worked), as well as future behavior. The estimation equation for a Granger test would look something like: $H i,t = A i + X i,t β + m φ=0 γ φd i,t φ + q φ=1 γ +φd i,t+φ + µ i,t 8

9 Treatments given in the future (namely, the q φ=1 γ +φd i,t+φ terms), should have no influence on today s behavior, so that all γ +φ coefficients should be zero. Difference-in-differences. This is a special case of panel data sets, where the treatment is aggregated across individuals. We look at the before and after gain for a treated group and compare it to the before and after gain for a control (untreated) group. Treatment effect= [E($H 1 D = 1, after) E($H 0 D = 1, before)] [E($H 0 D = 0, after) E($H 0 D = 0, before)] The strategy here is to control for the trend in outcome (given by the before/after difference for the control group) not attributable to treatment. If the trend change in home maintenance for the treatment group without treatment, is the same as the trend change in home maintenance for the control group, or [E($H 0 D = 1, after) E($H 0 D = 1, before)] = [E($H 0 D = 0, after) E($H 0 D = 0, before)] then after substituting this into equation above, we get Treatment effect = E($H 1 D = 1, after) E($H 0 D = 1, after) the treatment effect on the treated group. The corresponding regression model is and $H i,t = β 1 + β 2 after t + β 3 treated i + β 4 (after treated) i,t + ɛ i,t $H i = β 2 + β 4 treated i + ɛ i The change in home maintenance expenditures after the institution of the damage rider requirement (where those choosing the rider are the treated group, D=1 ), equals the increase common to both the treated and control groups, β 2, plus the differential change for those getting the treatment, β 4. It is the β 4 coefficient that measures the treatment effect for those getting treated. Identifying assumption The trend of the control group maps out what would have happened to the trend of the treatment group in the absence of treatment. Assessing plausibility. Do the pretreatment trends of the treatment and control group look similar? To check, run the following regression: trend check: $H i,t = α 1 + α 2 trend t + α 3 D i + α 4 (D i trend t ) i,t + µ i,t where D is a dummy variable for those who will eventually choose the tornado rider. Then the joint test for differences in trend is the test for α 3 = 0 and α 4 = 0. Regression Discontinuity. Treatment is strictly assigned on the basis of an observed index. If individuals are above a cutoff, they receive the treatment. If just below, they receive no treatment. For example, suppose there is a development bank that makes 9

10 disaster aid available to local regions within less developed countries whose risk management index, RMI exceeds a certain value, say 30 (let s suppose they have evaluated the availability of aid for a 1000 such disaster incidences, scattered widely). To assess the value of disaster aid in promoting subsequent economic recovery, we compare the real economic growth of countries just below the cutoff value (and barely not qualifying for the aid) with those whose RMI is just above 30, and hence barely eligible for the aid, at times when disasters strike and such aid is needed. The difference between the just below, and just above, country growth levels identifies the treatment effect. In practice, one controls for the level of the index as well. Is effect identified for all regions? No, just those near the cutoff; hence, it is a local average treatment effect (LATE). Regression for this type of identification strategy is growth i,t = β 1 + β 2 RMI i,t + β 3 D i,t + β 4 (RMI D) i,t + ɛ i,t where D i,t = 1 if the region receives aid (at the time of the disaster, they the must have a RMI of 30 or greater), and D i,t = 0 if the region does not receive aid (RMI < 30) at the time disaster strikes. Assume that while the cutoff value of 30 is well established, and many regional entities RMI populate either side of the cutoff. The regression is estimated for regional governments just around the cutoff value. The causal affect of receiving aid on growth is given by ˆβ 3 and ˆβ 4. Identifying assumption The unobserved characteristics on either side of the discontinuity are comparable. Assessing plausibility. Show that observables are comparable on either side of the cutoff. Print out and examine the data to confirm there are no suspicious jumps in the density of observations near the cutoff. Instrumental variables. Return to our regression of home maintenance on getting the tornado rider for the homeowner s policy, discussed earlier: $H i = β 1 + D i α + ɛ i Further, let s assume that treatment D is correlated with the residual (for any one of a number of reasons, including simultaneous equation structure, measurement error, or omitted variable bias in the structural, data generating process). This violates the standard OLS assumption leading to biased estimates of α. plim ˆα OLS = α + cov(d, ɛ)/var(d) We can overcome this problem if we have an instrument, Z, that is correlated to treatment but uncorrelated to the residual: Cov(D, Z) 0; Cov(ɛ, Z) = 0 The intuition underlying this condition is that Z only affects the outcome, home maintenance expenditures, through its correlation with the insurance rider. Using this instrument, we can perform two stage least squares, as indicated in the first chapter. We regress treatment on the instrument in what is called a first-stage regression. 10

11 D i = γ 1 + Z i γ 2 + ν i In the second stage, we regress the outcome on the predicted treatment. $H i = β 0 + ˆD i α + ɛ i [[where ˆD i = ˆγ 1 + Z iˆγ 2 ]] The IV estimate converges to the following: plim ˆα = α + cov(z, ɛ)/cov(z, D) Note that if the IV assumptions hold, the last term equals zero. When will the IV estimator be more or less biased than the OLS estimator? Identifying assumption. The instrumental variable only affects the outcome, home maintenance expenditures, through the decision to acquire the insurance rider. Suppose you had data on whether the individual insured s parents bought insurance riders. This might work as an instrument, if it increased the likelihood of a buyer acquiring a tornadorider, but had no effect on home maintenance expenditures otherwise. The instrument has to be uncorrelated to the unobserved determinants of outcome. Assessing plausibility. You need to have a compelling reason about why the instrumental variable only affects the rider, without any other influence on home maintenance expenditures. Show that instrument is uncorrelated with other observable determinants of outcome. Heterogeneous Treatment Effects. So far in these first two lectures, we have assumed the the model coefficients are stable against observations (so we have i subscripts on the coefficients, as well as the variables). Suppose otherwise and we also allow treatment effects to differ across individuals. Then our data generating regression looks like: $H i = β i + D i α i + ɛ i What does the IV estimate of α converge to? To simplify the issue, assume where Z has two values (0 and 1) and D has two values (0 and 1). Let s further assume that there are two types of individuals each forming one half of the population. Type A s (T = A) treatment status responds to Z: P r(d = 1 Z = 1, T = A) > P r(d = 1 Z = 0, T = A). prob(you got a rider parents did) > prob(you got a rider parents did not) Type B s (T = B) treatment status does not respond to Z: P r(d = 1 Z = 1, T = B) = P r(d = 1 Z = 0, T = B). probability of rider not affected by parents purchases We ll assume that α is the same within types but differs across types (i.e. α A does not equal α B ). Then what does this estimator converge to? plim(ˆα) = 11

12 {[.5E($H Z = 1, T = A) +.5E($H Z = 1, T = B)] [.5E($H Z = 0, T = A) +.5E($H Z = 0, T = B)]}/ {[.5P r(d Z = 1, T = A) +.5P r(d Z = 1, T = B)] [.5P r(d Z = 0, T = A) +.5P r(d Z = 0, T = B)]} Or, plim(ˆα) = {E($H Z = 1, T = A) E($H Z = 0, T = A)}/ {P r(d Z = 1, T = A) P r(d Z = 0, T = A)} Or, plim(ˆα) = {α A P r(d = 1 Z = 1, T = A) α A P r(d = 1 Z = 0, T = A)}/ {α A P r(d = 1 Z = 1, T = A) α A P r(d = 1 Z = 0, T = A)} Or, plim(ˆα) = α A Note that the instrumental variable here gives us a local average treatment effect: this estimator converges to that part of the population whose responses are sensitive to variations in the instrumental variable (e.g., the As in the population, but not the Bs). This is a general result for instrumental variables. Structural Econometric Models. There is no clear dividing line between structural econometric modeling and non-structural modeling. Basically, structural modeling tries to tie the estimation equation more fully to economic theory (and sometimes, statistical theory), in explaining how variations in x k affects the output y, and sometimes, how the unobservable variables, µ affect y as well. As noted in Reiss and Wolak (2007), there are three things that this attention to economic structure in that statistical modeling of relationships attempts to accomplish: First, structural modeling may be able to make inferences about unobserved behavioral relationships that could not be retrieved from nonexperimental data, without reference to the structure. Such use of modeling is ubiquitous in FIRE projects. Second, structural models provide justification for speculation on counterfactual outcomes or provide policy simulations. This is inevitably done in actuarial science and in studies of regulation, most often implicitly. Finally, structural models are often employed to distinguish the implications of competitive theories. But as Reiss and Wolak note the only sense in which one can test the two theories is to ask whether one of these ways of combining the same economic and stochastic primitives provides a markedly better description of observed or out-of-sample data. But even when physics, and metaphysics, conspire to make our theories unassailably true, the data used to estimate those theories always include unobservables that are not wholly accounted for by our theories. Therefore, all of the identification insights discussed above apply to the best of structural models. Hence, we will not distinguish further between structural and non-structural models below. Though we will discuss Butler and 12

13 Lambson (2015), both because their estimation equation derives from a structural specification (their first order condition for portfolio maximization), and because its structural specification (a non-expected utility theory) implicitly contradicts assumptions implicit in much of the subsequent empirical studies we will be examining. But first some advice and a couple of useful, if somewhat out of the way, tools. 13

EMERGING MARKETS - Lecture 2: Methodology refresher

EMERGING 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 information

Econometrics 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 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 information

Empirical approaches in public economics

Empirical 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 information

Potential Outcomes Model (POM)

Potential 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 information

Controlling for Time Invariant Heterogeneity

Controlling for Time Invariant Heterogeneity Controlling for Time Invariant Heterogeneity Yona Rubinstein July 2016 Yona Rubinstein (LSE) Controlling for Time Invariant Heterogeneity 07/16 1 / 19 Observables and Unobservables Confounding Factors

More information

Selection on Observables: Propensity Score Matching.

Selection 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 information

Quantitative Economics for the Evaluation of the European Policy

Quantitative 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 information

Econometrics. Week 8. Fall Institute of Economic Studies Faculty of Social Sciences Charles University in Prague

Econometrics. Week 8. Fall Institute of Economic Studies Faculty of Social Sciences Charles University in Prague Econometrics Week 8 Institute of Economic Studies Faculty of Social Sciences Charles University in Prague Fall 2012 1 / 25 Recommended Reading For the today Instrumental Variables Estimation and Two Stage

More information

IV Estimation and its Limitations: Weak Instruments and Weakly Endogeneous Regressors

IV Estimation and its Limitations: Weak Instruments and Weakly Endogeneous Regressors IV Estimation and its Limitations: Weak Instruments and Weakly Endogeneous Regressors Laura Mayoral IAE, Barcelona GSE and University of Gothenburg Gothenburg, May 2015 Roadmap Deviations from the standard

More information

Causal Inference Lecture Notes: Causal Inference with Repeated Measures in Observational Studies

Causal 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 information

14.74 Lecture 10: The returns to human capital: education

14.74 Lecture 10: The returns to human capital: education 14.74 Lecture 10: The returns to human capital: education Esther Duflo March 7, 2011 Education is a form of human capital. You invest in it, and you get returns, in the form of higher earnings, etc...

More information

EC402 - Problem Set 3

EC402 - Problem Set 3 EC402 - Problem Set 3 Konrad Burchardi 11th of February 2009 Introduction Today we will - briefly talk about the Conditional Expectation Function and - lengthily talk about Fixed Effects: How do we calculate

More information

WISE International Masters

WISE International Masters WISE International Masters ECONOMETRICS Instructor: Brett Graham INSTRUCTIONS TO STUDENTS 1 The time allowed for this examination paper is 2 hours. 2 This examination paper contains 32 questions. You are

More information

Development. ECON 8830 Anant Nyshadham

Development. ECON 8830 Anant Nyshadham Development ECON 8830 Anant Nyshadham Projections & Regressions Linear Projections If we have many potentially related (jointly distributed) variables Outcome of interest Y Explanatory variable of interest

More information

A Course in Applied Econometrics Lecture 14: Control Functions and Related Methods. Jeff Wooldridge IRP Lectures, UW Madison, August 2008

A Course in Applied Econometrics Lecture 14: Control Functions and Related Methods. Jeff Wooldridge IRP Lectures, UW Madison, August 2008 A Course in Applied Econometrics Lecture 14: Control Functions and Related Methods Jeff Wooldridge IRP Lectures, UW Madison, August 2008 1. Linear-in-Parameters Models: IV versus Control Functions 2. Correlated

More information

Applied Econometrics Lecture 1

Applied Econometrics Lecture 1 Lecture 1 1 1 Università di Urbino Università di Urbino PhD Programme in Global Studies Spring 2018 Outline of this module Beyond OLS (very brief sketch) Regression and causality: sources of endogeneity

More information

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data?

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data? When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data? Kosuke Imai Department of Politics Center for Statistics and Machine Learning Princeton University

More information

AGEC 661 Note Fourteen

AGEC 661 Note Fourteen AGEC 661 Note Fourteen Ximing Wu 1 Selection bias 1.1 Heckman s two-step model Consider the model in Heckman (1979) Y i = X iβ + ε i, D i = I {Z iγ + η i > 0}. For a random sample from the population,

More information

Job Training Partnership Act (JTPA)

Job Training Partnership Act (JTPA) Causal inference Part I.b: randomized experiments, matching and regression (this lecture starts with other slides on randomized experiments) Frank Venmans Example of a randomized experiment: Job Training

More information

Causal Inference with Big Data Sets

Causal 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 information

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data?

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data? When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Longitudinal Data? Kosuke Imai Princeton University Asian Political Methodology Conference University of Sydney Joint

More information

More on Roy Model of Self-Selection

More on Roy Model of Self-Selection V. J. Hotz Rev. May 26, 2007 More on Roy Model of Self-Selection Results drawn on Heckman and Sedlacek JPE, 1985 and Heckman and Honoré, Econometrica, 1986. Two-sector model in which: Agents are income

More information

ECON Introductory Econometrics. Lecture 17: Experiments

ECON 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 information

STOCKHOLM UNIVERSITY Department of Economics Course name: Empirical Methods Course code: EC40 Examiner: Lena Nekby Number of credits: 7,5 credits Date of exam: Friday, June 5, 009 Examination time: 3 hours

More information

1 Impact Evaluation: Randomized Controlled Trial (RCT)

1 Impact Evaluation: Randomized Controlled Trial (RCT) Introductory Applied Econometrics EEP/IAS 118 Fall 2013 Daley Kutzman Section #12 11-20-13 Warm-Up Consider the two panel data regressions below, where i indexes individuals and t indexes time in months:

More information

Finding Instrumental Variables: Identification Strategies. Amine Ouazad Ass. Professor of Economics

Finding Instrumental Variables: Identification Strategies. Amine Ouazad Ass. Professor of Economics Finding Instrumental Variables: Identification Strategies Amine Ouazad Ass. Professor of Economics Outline 1. Before/After 2. Difference-in-difference estimation 3. Regression Discontinuity Design BEFORE/AFTER

More information

Flexible Estimation of Treatment Effect Parameters

Flexible 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 information

Econ 2148, fall 2017 Instrumental variables I, origins and binary treatment case

Econ 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 information

Final Exam. Economics 835: Econometrics. Fall 2010

Final Exam. Economics 835: Econometrics. Fall 2010 Final Exam Economics 835: Econometrics Fall 2010 Please answer the question I ask - no more and no less - and remember that the correct answer is often short and simple. 1 Some short questions a) For each

More information

On the Use of Linear Fixed Effects Regression Models for Causal Inference

On the Use of Linear Fixed Effects Regression Models for Causal Inference On the Use of Linear Fixed Effects Regression Models for ausal Inference Kosuke Imai Department of Politics Princeton University Joint work with In Song Kim Atlantic ausal Inference onference Johns Hopkins

More information

Gov 2002: 4. Observational Studies and Confounding

Gov 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 information

Dealing With Endogeneity

Dealing With Endogeneity Dealing With Endogeneity Junhui Qian December 22, 2014 Outline Introduction Instrumental Variable Instrumental Variable Estimation Two-Stage Least Square Estimation Panel Data Endogeneity in Econometrics

More information

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Panel Data?

When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Panel Data? When Should We Use Linear Fixed Effects Regression Models for Causal Inference with Panel Data? Kosuke Imai Department of Politics Center for Statistics and Machine Learning Princeton University Joint

More information

Longitudinal Data Analysis. RatSWD Nachwuchsworkshop Vorlesung von Josef Brüderl 25. August, 2009

Longitudinal Data Analysis. RatSWD Nachwuchsworkshop Vorlesung von Josef Brüderl 25. August, 2009 Longitudinal Data Analysis RatSWD Nachwuchsworkshop Vorlesung von Josef Brüderl 25. August, 2009 Longitudinal Data Analysis Traditional definition Statistical methods for analyzing data with a time dimension

More information

STOCKHOLM UNIVERSITY Department of Economics Course name: Empirical Methods Course code: EC40 Examiner: Lena Nekby Number of credits: 7,5 credits Date of exam: Saturday, May 9, 008 Examination time: 3

More information

Regression Discontinuity Designs.

Regression 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 information

II. MATCHMAKER, MATCHMAKER

II. 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 information

Simple Regression Model. January 24, 2011

Simple Regression Model. January 24, 2011 Simple Regression Model January 24, 2011 Outline Descriptive Analysis Causal Estimation Forecasting Regression Model We are actually going to derive the linear regression model in 3 very different ways

More information

Treatment Effects. Christopher Taber. September 6, Department of Economics University of Wisconsin-Madison

Treatment 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 information

Lecture 4: Linear panel models

Lecture 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 information

Econ 673: Microeconometrics Chapter 12: Estimating Treatment Effects. The Problem

Econ 673: Microeconometrics Chapter 12: Estimating Treatment Effects. The Problem Econ 673: Microeconometrics Chapter 12: Estimating Treatment Effects The Problem Analysts are frequently interested in measuring the impact of a treatment on individual behavior; e.g., the impact of job

More information

Rockefeller College University at Albany

Rockefeller College University at Albany Rockefeller College University at Albany PAD 705 Handout: Simultaneous quations and Two-Stage Least Squares So far, we have studied examples where the causal relationship is quite clear: the value of the

More information

Wooldridge, Introductory Econometrics, 4th ed. Chapter 15: Instrumental variables and two stage least squares

Wooldridge, Introductory Econometrics, 4th ed. Chapter 15: Instrumental variables and two stage least squares Wooldridge, Introductory Econometrics, 4th ed. Chapter 15: Instrumental variables and two stage least squares Many economic models involve endogeneity: that is, a theoretical relationship does not fit

More information

Econometric Causality

Econometric Causality Econometric (2008) International Statistical Review, 76(1):1-27 James J. Heckman Spencer/INET Conference University of Chicago Econometric The econometric approach to causality develops explicit models

More information

Write your identification number on each paper and cover sheet (the number stated in the upper right hand corner on your exam cover).

Write 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 information

Introduction 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 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 information

1 Motivation for Instrumental Variable (IV) Regression

1 Motivation for Instrumental Variable (IV) Regression ECON 370: IV & 2SLS 1 Instrumental Variables Estimation and Two Stage Least Squares Econometric Methods, ECON 370 Let s get back to the thiking in terms of cross sectional (or pooled cross sectional) data

More information

PSC 504: Differences-in-differeces estimators

PSC 504: Differences-in-differeces estimators PSC 504: Differences-in-differeces estimators Matthew Blackwell 3/22/2013 Basic differences-in-differences model Setup e basic idea behind a differences-in-differences model (shorthand: diff-in-diff, DID,

More information

Linear Models in Econometrics

Linear Models in Econometrics Linear Models in Econometrics Nicky Grant At the most fundamental level econometrics is the development of statistical techniques suited primarily to answering economic questions and testing economic theories.

More information

New Developments in Econometrics Lecture 11: Difference-in-Differences Estimation

New 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 information

Principles Underlying Evaluation Estimators

Principles 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 information

Online Appendix to Yes, But What s the Mechanism? (Don t Expect an Easy Answer) John G. Bullock, Donald P. Green, and Shang E. Ha

Online Appendix to Yes, But What s the Mechanism? (Don t Expect an Easy Answer) John G. Bullock, Donald P. Green, and Shang E. Ha Online Appendix to Yes, But What s the Mechanism? (Don t Expect an Easy Answer) John G. Bullock, Donald P. Green, and Shang E. Ha January 18, 2010 A2 This appendix has six parts: 1. Proof that ab = c d

More information

WISE MA/PhD Programs Econometrics Instructor: Brett Graham Spring Semester, Academic Year Exam Version: A

WISE MA/PhD Programs Econometrics Instructor: Brett Graham Spring Semester, Academic Year Exam Version: A WISE MA/PhD Programs Econometrics Instructor: Brett Graham Spring Semester, 2015-16 Academic Year Exam Version: A INSTRUCTIONS TO STUDENTS 1 The time allowed for this examination paper is 2 hours. 2 This

More information

FNCE 926 Empirical Methods in CF

FNCE 926 Empirical Methods in CF FNCE 926 Empirical Methods in CF Lecture 11 Standard Errors & Misc. Professor Todd Gormley Announcements Exercise #4 is due Final exam will be in-class on April 26 q After today, only two more classes

More information

Chapter 11. Regression with a Binary Dependent Variable

Chapter 11. Regression with a Binary Dependent Variable Chapter 11 Regression with a Binary Dependent Variable 2 Regression with a Binary Dependent Variable (SW Chapter 11) So far the dependent variable (Y) has been continuous: district-wide average test score

More information

Truncation and Censoring

Truncation and Censoring Truncation and Censoring Laura Magazzini laura.magazzini@univr.it Laura Magazzini (@univr.it) Truncation and Censoring 1 / 35 Truncation and censoring Truncation: sample data are drawn from a subset of

More information

Econometrics with Observational Data. Introduction and Identification Todd Wagner February 1, 2017

Econometrics with Observational Data. Introduction and Identification Todd Wagner February 1, 2017 Econometrics with Observational Data Introduction and Identification Todd Wagner February 1, 2017 Goals for Course To enable researchers to conduct careful quantitative analyses with existing VA (and non-va)

More information

The returns to schooling, ability bias, and regression

The 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 information

Lecture 8. Roy Model, IV with essential heterogeneity, MTE

Lecture 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 information

Statistical Inference with Regression Analysis

Statistical Inference with Regression Analysis Introductory Applied Econometrics EEP/IAS 118 Spring 2015 Steven Buck Lecture #13 Statistical Inference with Regression Analysis Next we turn to calculating confidence intervals and hypothesis testing

More information

Predicting the Treatment Status

Predicting the Treatment Status Predicting the Treatment Status Nikolay Doudchenko 1 Introduction Many studies in social sciences deal with treatment effect models. 1 Usually there is a treatment variable which determines whether a particular

More information

Lecture 9: Panel Data Model (Chapter 14, Wooldridge Textbook)

Lecture 9: Panel Data Model (Chapter 14, Wooldridge Textbook) Lecture 9: Panel Data Model (Chapter 14, Wooldridge Textbook) 1 2 Panel Data Panel data is obtained by observing the same person, firm, county, etc over several periods. Unlike the pooled cross sections,

More information

2) For a normal distribution, the skewness and kurtosis measures are as follows: A) 1.96 and 4 B) 1 and 2 C) 0 and 3 D) 0 and 0

2) For a normal distribution, the skewness and kurtosis measures are as follows: A) 1.96 and 4 B) 1 and 2 C) 0 and 3 D) 0 and 0 Introduction to Econometrics Midterm April 26, 2011 Name Student ID MULTIPLE CHOICE. Choose the one alternative that best completes the statement or answers the question. (5,000 credit for each correct

More information

Applied Microeconometrics (L5): Panel Data-Basics

Applied Microeconometrics (L5): Panel Data-Basics Applied Microeconometrics (L5): Panel Data-Basics Nicholas Giannakopoulos University of Patras Department of Economics ngias@upatras.gr November 10, 2015 Nicholas Giannakopoulos (UPatras) MSc Applied Economics

More information

LECTURE 2: SIMPLE REGRESSION I

LECTURE 2: SIMPLE REGRESSION I LECTURE 2: SIMPLE REGRESSION I 2 Introducing Simple Regression Introducing Simple Regression 3 simple regression = regression with 2 variables y dependent variable explained variable response variable

More information

Instrumental Variables and the Problem of Endogeneity

Instrumental Variables and the Problem of Endogeneity Instrumental Variables and the Problem of Endogeneity September 15, 2015 1 / 38 Exogeneity: Important Assumption of OLS In a standard OLS framework, y = xβ + ɛ (1) and for unbiasedness we need E[x ɛ] =

More information

Recitation Notes 5. Konrad Menzel. October 13, 2006

Recitation 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 information

Lecture 11/12. Roy Model, MTE, Structural Estimation

Lecture 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 information

Comments on: Panel Data Analysis Advantages and Challenges. Manuel Arellano CEMFI, Madrid November 2006

Comments on: Panel Data Analysis Advantages and Challenges. Manuel Arellano CEMFI, Madrid November 2006 Comments on: Panel Data Analysis Advantages and Challenges Manuel Arellano CEMFI, Madrid November 2006 This paper provides an impressive, yet compact and easily accessible review of the econometric literature

More information

Dynamics in Social Networks and Causality

Dynamics 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 information

Applied Statistics and Econometrics

Applied Statistics and Econometrics Applied Statistics and Econometrics Lecture 7 Saul Lach September 2017 Saul Lach () Applied Statistics and Econometrics September 2017 1 / 68 Outline of Lecture 7 1 Empirical example: Italian labor force

More information

Basic econometrics. Tutorial 3. Dipl.Kfm. Johannes Metzler

Basic econometrics. Tutorial 3. Dipl.Kfm. Johannes Metzler Basic econometrics Tutorial 3 Dipl.Kfm. Introduction Some of you were asking about material to revise/prepare econometrics fundamentals. First of all, be aware that I will not be too technical, only as

More information

ECO375 Tutorial 8 Instrumental Variables

ECO375 Tutorial 8 Instrumental Variables ECO375 Tutorial 8 Instrumental Variables Matt Tudball University of Toronto Mississauga November 16, 2017 Matt Tudball (University of Toronto) ECO375H5 November 16, 2017 1 / 22 Review: Endogeneity Instrumental

More information

Econometric Analysis of Cross Section and Panel Data

Econometric Analysis of Cross Section and Panel Data Econometric Analysis of Cross Section and Panel Data Jeffrey M. Wooldridge / The MIT Press Cambridge, Massachusetts London, England Contents Preface Acknowledgments xvii xxiii I INTRODUCTION AND BACKGROUND

More information

Transparent Structural Estimation. Matthew Gentzkow Fisher-Schultz Lecture (from work w/ Isaiah Andrews & Jesse M. Shapiro)

Transparent Structural Estimation. Matthew Gentzkow Fisher-Schultz Lecture (from work w/ Isaiah Andrews & Jesse M. Shapiro) Transparent Structural Estimation Matthew Gentzkow Fisher-Schultz Lecture (from work w/ Isaiah Andrews & Jesse M. Shapiro) 1 A hallmark of contemporary applied microeconomics is a conceptual framework

More information

Difference-in-Differences Estimation

Difference-in-Differences Estimation Difference-in-Differences Estimation Jeff Wooldridge Michigan State University Programme Evaluation for Policy Analysis Institute for Fiscal Studies June 2012 1. The Basic Methodology 2. How Should We

More information

What s New in Econometrics. Lecture 1

What 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 information

Instrumental Variables

Instrumental Variables Instrumental Variables Department of Economics University of Wisconsin-Madison September 27, 2016 Treatment Effects Throughout the course we will focus on the Treatment Effect Model For now take that to

More information

Ec1123 Section 7 Instrumental Variables

Ec1123 Section 7 Instrumental Variables Ec1123 Section 7 Instrumental Variables Andrea Passalacqua Harvard University andreapassalacqua@g.harvard.edu November 16th, 2017 Andrea Passalacqua (Harvard) Ec1123 Section 7 Instrumental Variables November

More information

Instrumental Variables

Instrumental 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 information

Logistic regression: Why we often can do what we think we can do. Maarten Buis 19 th UK Stata Users Group meeting, 10 Sept. 2015

Logistic regression: Why we often can do what we think we can do. Maarten Buis 19 th UK Stata Users Group meeting, 10 Sept. 2015 Logistic regression: Why we often can do what we think we can do Maarten Buis 19 th UK Stata Users Group meeting, 10 Sept. 2015 1 Introduction Introduction - In 2010 Carina Mood published an overview article

More information

Applied Quantitative Methods II

Applied Quantitative Methods II Applied Quantitative Methods II Lecture 10: Panel Data Klára Kaĺıšková Klára Kaĺıšková AQM II - Lecture 10 VŠE, SS 2016/17 1 / 38 Outline 1 Introduction 2 Pooled OLS 3 First differences 4 Fixed effects

More information

ECO220Y Simple Regression: Testing the Slope

ECO220Y Simple Regression: Testing the Slope ECO220Y Simple Regression: Testing the Slope Readings: Chapter 18 (Sections 18.3-18.5) Winter 2012 Lecture 19 (Winter 2012) Simple Regression Lecture 19 1 / 32 Simple Regression Model y i = β 0 + β 1 x

More information

The Simple Linear Regression Model

The Simple Linear Regression Model The Simple Linear Regression Model Lesson 3 Ryan Safner 1 1 Department of Economics Hood College ECON 480 - Econometrics Fall 2017 Ryan Safner (Hood College) ECON 480 - Lesson 3 Fall 2017 1 / 77 Bivariate

More information

PSC 504: Instrumental Variables

PSC 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 information

IV Estimation and its Limitations: Weak Instruments and Weakly Endogeneous Regressors

IV Estimation and its Limitations: Weak Instruments and Weakly Endogeneous Regressors IV Estimation and its Limitations: Weak Instruments and Weakly Endogeneous Regressors Laura Mayoral IAE, Barcelona GSE and University of Gothenburg Gothenburg, May 2015 Roadmap of the course Introduction.

More information

Applied Econometrics (MSc.) Lecture 3 Instrumental Variables

Applied Econometrics (MSc.) Lecture 3 Instrumental Variables Applied Econometrics (MSc.) Lecture 3 Instrumental Variables Estimation - Theory Department of Economics University of Gothenburg December 4, 2014 1/28 Why IV estimation? So far, in OLS, we assumed independence.

More information

The FIRE Project: FIRE2

The FIRE Project: FIRE2 The FIRE Project: FIRE2 Richard Butler, Brigham Young University 2 Population Model and Insights into Informative Relationships Again, the data generating process (DGP) is a description of how an outcome,

More information

Causal Inference with General Treatment Regimes: Generalizing the Propensity Score

Causal 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 information

Analysis of Panel Data: Introduction and Causal Inference with Panel Data

Analysis of Panel Data: Introduction and Causal Inference with Panel Data Analysis of Panel Data: Introduction and Causal Inference with Panel Data Session 1: 15 June 2015 Steven Finkel, PhD Daniel Wallace Professor of Political Science University of Pittsburgh USA Course presents

More information

LECTURE 10. Introduction to Econometrics. Multicollinearity & Heteroskedasticity

LECTURE 10. Introduction to Econometrics. Multicollinearity & Heteroskedasticity LECTURE 10 Introduction to Econometrics Multicollinearity & Heteroskedasticity November 22, 2016 1 / 23 ON PREVIOUS LECTURES We discussed the specification of a regression equation Specification consists

More information

Lecture Notes 12 Advanced Topics Econ 20150, Principles of Statistics Kevin R Foster, CCNY Spring 2012

Lecture Notes 12 Advanced Topics Econ 20150, Principles of Statistics Kevin R Foster, CCNY Spring 2012 Lecture Notes 2 Advanced Topics Econ 2050, Principles of Statistics Kevin R Foster, CCNY Spring 202 Endogenous Independent Variables are Invalid Need to have X causing Y not vice-versa or both! NEVER regress

More information

Introduction to Econometrics

Introduction 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 information

Introduction to Regression Analysis. Dr. Devlina Chatterjee 11 th August, 2017

Introduction to Regression Analysis. Dr. Devlina Chatterjee 11 th August, 2017 Introduction to Regression Analysis Dr. Devlina Chatterjee 11 th August, 2017 What is regression analysis? Regression analysis is a statistical technique for studying linear relationships. One dependent

More information

Econometrics Problem Set 3

Econometrics Problem Set 3 Econometrics Problem Set 3 Conceptual Questions 1. This question refers to the estimated regressions in table 1 computed using data for 1988 from the U.S. Current Population Survey. The data set consists

More information

Education Production Functions. April 7, 2009

Education 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 information

An overview of applied econometrics

An overview of applied econometrics An overview of applied econometrics Jo Thori Lind September 4, 2011 1 Introduction This note is intended as a brief overview of what is necessary to read and understand journal articles with empirical

More information

ECO 310: Empirical Industrial Organization Lecture 2 - Estimation of Demand and Supply

ECO 310: Empirical Industrial Organization Lecture 2 - Estimation of Demand and Supply ECO 310: Empirical Industrial Organization Lecture 2 - Estimation of Demand and Supply Dimitri Dimitropoulos Fall 2014 UToronto 1 / 55 References RW Section 3. Wooldridge, J. (2008). Introductory Econometrics:

More information

Ch 7: Dummy (binary, indicator) variables

Ch 7: Dummy (binary, indicator) variables Ch 7: Dummy (binary, indicator) variables :Examples Dummy variable are used to indicate the presence or absence of a characteristic. For example, define female i 1 if obs i is female 0 otherwise or male

More information

Teaching Causal Inference in Undergraduate Econometrics

Teaching Causal Inference in Undergraduate Econometrics Teaching Causal Inference in Undergraduate Econometrics October 24, 2012 Abstract This paper argues that the current way in which the undergraduate introductory econometrics course is taught is neither

More information