Econometrics – II
BA (H) Econ Core– Spring, 2024 (February to May)
Chapter 15: Instrumental Variables Estimation and two stage
least squares
Wooldridge, J.M. (2015) Introductory Econometrics, 6e
Tirtha Chatterjee
Omitted variables in a simple regression model
• Based on methods we have learnt, when faced with the prospect of omitted variables
bias (or unobserved heterogeneity)
• we can ignore the problem and suffer the consequences of biased and inconsistent estimators;
• we can try to find and use a suitable proxy variable for the unobserved variable
• But both methods have shortcomings
• Ignoring the problem might not be the correct way- biased and inconsistent estimates
• Second it is not always possible to find a good proxy.
• There is a better approach- Instrumental variable approach
Tirtha Chatterjee
Instrumental variable approach
• This leaves the unobserved variable in the error term, but rather than estimating the
model by OLS, it uses an estimation method that recognizes the presence of the
omitted variable.
• This is what the method of instrumental variables does.
• Consider the problem of unobserved ability in a wage equation for working adults.
• A simple model is log 𝑤𝑎𝑔𝑒 = 𝛽! + 𝛽"𝑒𝑑𝑢𝑐 + 𝛽#𝑎𝑏𝑖𝑙 + 𝑒
• where e is the error term.
Tirtha Chatterjee
Instrumental variable approach
• Suppose that a proxy variable is not available (or does not have the properties needed
to produce a consistent estimator of 𝛽").
• Then, we put abil into the error term, and we are left with the simple regression model
log 𝑤𝑎𝑔𝑒 = 𝛽! + 𝛽"𝑒𝑑𝑢𝑐 + 𝑢
• where u contains abil.
• OLS estimation will give us a biased and inconsistent estimator of 𝛽" results if educ
and abil are correlated.
• But we can estimate the model to get unbiased and consistent estimates if we can find
an instrumental variable for educ.
Tirtha Chatterjee
Instrumental variable approach
• A simple regression model is written as 𝑦 = 𝛽! + 𝛽"𝑥 + 𝑢
• where we think that x and u are correlated (have nonzero covariance): 𝐶𝑜𝑣(𝑥, 𝑢) ≠ 0
• In order to obtain consistent estimators of 𝛽! and 𝛽" when x and u are correlated, we
need a new variable (z) that satisfies certain properties.
• z is an instrumental variable for x, or an instrument for x if it satisfies the following
• instrument exogeneity - z is exogenous in equation : z is uncorrelated with u, that is, 𝐶𝑜𝑣 𝑧, 𝑢 =
0
• z should have no partial effect on y (after x and omitted variables have been controlled for), and z should be
uncorrelated with the omitted variables
• as instrument relevance - z is relevant for explaining variation in x - z is correlated with x, that is,
𝐶𝑜𝑣(𝑧, 𝑥) ≠ 0
Tirtha Chatterjee
Can we empirically test the two assumptions
• instrument exogeneity is difficult to test empirically.
• We must maintain 𝐶𝑜𝑣 𝑧, 𝑢 = 0 by appealing to economic behavior or introspection.
• instrument relevance can be tested, given a random sample from the population.
• The easiest way to do this is to estimate a simple regression between x and z.
• In the population, we have
𝑥 = 𝜋! + 𝜋"𝑧 + 𝑣
• We should be able to reject the null hypothesis- 𝐻!: 𝜋"=0 against the two-sided
alternative 𝐻! : 𝜋" ≠ 0, at a sufficiently small significance level (say, 5% or 1%).
• If this is the case, then we can be fairly confident that instrument relevance holds.
Tirtha Chatterjee
Instrument for education
log 𝑤𝑎𝑔𝑒 = 𝛽! + 𝛽"𝑒𝑑𝑢𝑐 + 𝑢
• An instrumental variable z for educ must be (1) uncorrelated with ability (and any
other unobserved factors affecting wage) and (2) correlated with education.
• The last digit of an individual’s Social Security Number almost certainly satisfies
the first requirement: it is uncorrelated with ability because it is determined randomly.
• However, it is not correlated with education- because of the way it is determined
• Therefore it makes a poor instrumental variable for educ because it violates instrument
relevance assumption
Tirtha Chatterjee
Instrument for education
log 𝑤𝑎𝑔𝑒 = 𝛽! + 𝛽"𝑒𝑑𝑢𝑐 + 𝑢
• IQ cannot be used as an IV although it was a good proxy for ability
• A proxy variable for abil say, IQ is a good proxy because it is highly corrrelated with abil.
• An instrumental variable must be uncorrelated with abil.
• Therefore, while IQ is a good candidate as a proxy variable for abil, it is not a good instrumental
variable for educ because it violates the instrument exogeneity requirement.
• Similarly another variable- family background- say mother’s education can be highly
correlated with child’s education but child’s ability could also be driven by mother’s
education and hence is not a good IV
• Another IV choice for educ is number of siblings while growing up (sibs).
• Typically, having more siblings is associated with lower average levels of education.
• Thus, if number of siblings is uncorrelated with ability, it can act as an instrumental variable for
educ.
Tirtha Chatterjee
IV in policy impact studies- Binary endogenous & IV
• Suppose we want to estimate the effect of being a veteran of the Vietnam War had on
lifetime earnings.
log 𝑒𝑎𝑟𝑛𝑠 = 𝛽! + 𝛽"𝑣𝑒𝑡𝑒𝑟𝑎𝑛 + 𝑢
• where veteran is a binary variable.
• The problem with estimating by OLS is that there may be a self-selection problem
• Perhaps people who get the most out of the military choose to join, or the decision to join is
correlated with other characteristics that affect earnings.
• These will cause veteran and u to be correlated.
• An instrument could be Vietnam draft lottery provided a natural experiment
• Young men were given lottery numbers that determined whether they would be called to serve in
Vietnam.
• draft lottery number is uncorrelated with the error term u as lottery was randomly assigned
• But those with a low enough number had to serve in Vietnam, so that the probability of being a
veteran is correlated with lottery number.
Tirtha Chatterjee
Estimating the effect of smoking on birth weight
• Without other explanatory variables, the model is
log 𝑏𝑤𝑔ℎ𝑡 = 𝛽! + 𝛽"𝑝𝑎𝑐𝑘𝑠 + 𝑢
• where packs is the number of packs smoked by the mother per day.
• Packs could be correlated with other health factors or the availability of good prenatal
care
• so that packs and u might be correlated.
• Instrument for packs could be average price of cigarettes in the state of residence
• We will assume that cigprice and u are uncorrelated
• even though state support for health care could be correlated with cigarette taxes
• Packs & cigprice are negatively correlated - cigprice can be used as an IV if it was a normal good
• To check this, we regress packs on cigprice, using the data in BWGHT:
Tirtha Chatterjee
Estimating the effect of smoking on birth weight
• We find no significant relationship b/w smoking during pregnancy and cigarette prices
• But cigarette is not a normal good. Hence packs and cigprice are not correlated
• We should not use cigprice as an IV for packs.
• But what happens if we do? The IV results would be
• The coefficient on packs is huge and of an unexpected sign.
• The standard error is also very large, so packs is not significant.
• Estimates are meaningless because cigprice fails one requirement of an IV- relevance
• Price was a weak instrument- the problem was of “low” (but not zero) correlation
between z and x.
Tirtha Chatterjee