Biometry regression

73
3
Multiple Regression Analysis: Estimation
In Chapter 2, we learned how to use simple regression analysis to explain a dependent
variable, y, as a function of a single independent variable, x. The primary drawback in
using simple regression analysis for empirical work is that it is very difficult to draw
ceteris paribus conclusions about how x affects y: the key assumption, SLR.4—that all
other factors affecting y are uncorrelated with x—is often unrealistic.
Multiple regression analysis is more amenable to ceteris paribus analysis because it
allows us to explicitly control for many other factors that simultaneously affect the depen-
dent variable. This is important both for testing economic theories and for evaluating policy
effects when we must rely on nonexperimental data. Because multiple regression models
can accommodate many explanatory variables that may be correlated, we can hope to infer
causality in cases where simple regression analysis would be misleading.
Naturally, if we add more factors to our model that are useful for explaining y, then
more of the variation in y can be explained. Thus, multiple regression analysis can be used
to build better models for predicting the dependent variable.
An additional advantage of multiple regression analysis is that it can incorporate fairly
general functional form relationships. In the simple regression model, only one function
of a single explanatory variable can appear in the equation. As we will see, the multiple
regression model allows for much more flexibility.
Section 3.1 formally introduces the multiple regression model and further discusses
the advantages of multiple regression over simple regression. In Section 3.2, we demon-
strate how to estimate the parameters in the multiple regression model using the method
of ordinary least squares. In Sections 3.3, 3.4, and 3.5, we describe various statistical
properties of the OLS estimators, including unbiasedness and efficiency.
The multiple regression model is still the most widely used vehicle for empirical
analysis in economics and other social sciences. Likewise, the method of ordinary least
squares is popularly used for estimating the parameters of the multiple regression model.
3.1 Motivation for Multiple Regression
The Model with Two Independent Variables
We begin with some simple examples to show how multiple regression analysis can be
used to solve problems that cannot be solved by simple regression.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 73

The first example is a simple variation of the wage equation introduced in Chapter 2
for obtaining the effect of education on hourly wage:
wage ϭ b0 ϩ b1educ ϩ b2exper ϩ u, (3.1)
where exper is years of labor market experience. Thus, wage is determined by the two
explanatory or independent variables, education and experience, and by other unobserved
factors, which are contained in u. We are still primarily interested in the effect of educ
on wage, holding fixed all other factors affecting wage; that is, we are interested in the
parameter b1.
Compared with a simple regression analysis relating wage to educ, equation (3.1)
effectively takes exper out of the error term and puts it explicitly in the equation. Because
exper appears in the equation, its coefficient, b2, measures the ceteris paribus effect of
exper on wage, which is also of some interest.
Not surprisingly, just as with simple regression, we will have to make assumptions
about how u in (3.1) is related to the independent variables, educ and exper. However, as
we will see in Section 3.2, there is one thing of which we can be confident: because (3.1)
contains experience explicitly, we will be able to measure the effect of education on wage,
holding experience fixed. In a simple regression analysis—which puts exper in the error
term—we would have to assume that experience is uncorrelated with education, a tenu-
ous assumption.
As a second example, consider the problem of explaining the effect of per student
spending (expend) on the average standardized test score (avgscore) at the high school
level. Suppose that the average test score depends on funding, average family income
(avginc), and other unobservables:
avgscore ϭ b0 ϩ b1expend ϩ b2avginc ϩ u. (3.2)
The coefficient of interest for policy purposes is b1, the ceteris paribus effect of expend
on avgscore. By including avginc explicitly in the model, we are able to control for its
effect on avgscore. This is likely to be important because average family income tends
to be correlated with per student spending: spending levels are often determined by both
property and local income taxes. In simple regression analysis, avginc would be included
in the error term, which would likely be correlated with expend, causing the OLS esti-
mator of b1 in the two-variable model to be biased.
In the two previous similar examples, we have shown how observable factors other
than the variable of primary interest [educ in equation (3.1) and expend in equation (3.2)]
can be included in a regression model. Generally, we can write a model with two inde-
pendent variables as
y ϭ b0 ϩ b1x1 ϩ b2x2 ϩ u, (3.3)
where b0 is the intercept, b1 measures the change in y with respect to x1, holding other
factors fixed, and b2 measures the change in y with respect to x2, holding other factors
fixed.
74 Part 1 Regression Analysis with Cross-Sectional Data
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 74

Multiple regression analysis is also useful for generalizing functional relationships
between variables. As an example, suppose family consumption (cons) is a quadratic func-
tion of family income (inc):
cons ϭ b0 ϩ b1inc ϩ b2inc2 ϩ u, (3.4)
where u contains other factors affecting consumption. In this model, consumption depends
on only one observed factor, income; so it might seem that it can be handled in a simple
regression framework. But the model falls outside simple regression because it contains
two functions of income, inc and inc2 (and therefore three parameters, b0, b1, and b2).
Nevertheless, the consumption function is easily written as a regression model with two
independent variables by letting x1 ϭ inc and x2 ϭ inc2.
Mechanically, there will be no difference in using the method of ordinary least squares
(introduced in Section 3.2) to estimate equations as different as (3.1) and (3.4). Each
equation can be written as (3.3), which is all that matters for computation. There is,
however, an important difference in how one interprets the parameters. In equation (3.1),
␤1 is the ceteris paribus effect of educ on wage. The parameter ␤1 has no such interpreta-
tion in (3.4). In other words, it makes no sense to measure the effect of inc on cons while
holding inc2 fixed, because if inc changes, then so must inc2! Instead, the change in con-
sumption with respect to the change in income—the marginal propensity to consume—
is approximated by
Ϸ ␤1 ϩ 2␤2inc.
See Appendix A for the calculus needed to derive this equation. In other words, the
marginal effect of income on consumption depends on ␤2 as well as on ␤1 and the level
of income. This example shows that, in any particular application, the definitions of
the independent variables are crucial. But for the theoretical development of multiple
regression, we can be vague about such details. We will study examples like this more
completely in Chapter 6.
In the model with two independent variables, the key assumption about how u is
related to x1 and x2 is
E(u͉x1,x2) ϭ 0. (3.5)
The interpretation of condition (3.5) is similar to the interpretation of Assumption SLR.4
for simple regression analysis. It means that, for any values of x1 and x2 in the population,
the average unobservable is equal to zero. As with simple regression, the important part
of the assumption is that the expected value of u is the same for all combinations of x1 and
x2; that this common value is zero is no assumption at all as long as the intercept ␤0 is
included in the model (see Section 2.1).
How can we interpret the zero conditional mean assumption in the previous examples?
In equation (3.1), the assumption is E(u͉educ,exper) ϭ 0. This implies that other factors
affecting wage are not related on average to educ and exper. Therefore, if we think innate
ability is part of u, then we will need average ability levels to be the same across all
⌬cons
⌬inc
Chapter 3 Multiple Regression Analysis: Estimation 75
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 75

combinations of education and experience in the working population. This may or may
not be true, but, as we will see in Section 3.3, this is the question we need to ask in order
to determine whether the method of ordinary least squares produces unbiased estimators.
The example measuring student performance [equation (3.2)] is similar to the wage
equation. The zero conditional mean assumption is E(u͉expend,avginc) ϭ 0, which means
that other factors affecting test scores—
school or student characteristics—are, on
average, unrelated to per student funding
and average family income.
When applied to the quadratic con-
sumption function in (3.4), the zero condi-
tional mean assumption has a slightly
different interpretation. Written literally,
equation (3.5) becomes E(u͉inc,inc2) ϭ 0.
Since inc2 is known when inc is known,
including inc2 in the expectation is redun-
dant: E(u͉inc,inc2) ϭ 0 is the same as E(u͉inc) ϭ 0. Nothing is wrong with putting inc2 along
with inc in the expectation when stating the assumption, but E(u͉inc) ϭ 0 is more concise.
The Model with k Independent Variables
Once we are in the context of multiple regression, there is no need to stop with two inde-
pendent variables. Multiple regression analysis allows many observed factors to affect y.
In the wage example, we might also include amount of job training, years of tenure with
the current employer, measures of ability, and even demographic variables like number of
siblings or mother’s education. In the school funding example, additional variables might
include measures of teacher quality and school size.
The general multiple linear regression model (also called the multiple regression
model) can be written in the population as
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ ␤3x3 ϩ … ϩ ␤kxk ϩ u, (3.6)
where ␤0 is the intercept, ␤1 is the parameter associated with x1, ␤2 is the parameter asso-
ciated with x2, and so on. Since there are k independent variables and an intercept, equa-
tion (3.6) contains k ϩ 1 (unknown) population parameters. For shorthand purposes, we
will sometimes refer to the parameters other than the intercept as slope parameters, even
though this is not always literally what they are. [See equation (3.4), where neither ␤1 nor
␤2 is itself a slope, but together they determine the slope of the relationship between con-
sumption and income.]
The terminology for multiple regression is similar to that for simple regression and is
given in Table 3.1. Just as in simple regression, the variable u is the error term or
disturbance. It contains factors other than x1, x2, …,xk that affect y. No matter how many
explanatory variables we include in our model, there will always be factors we cannot
include, and these are collectively contained in u.
When applying the general multiple regression model, we must know how to interpret the
parameters. We will get plenty of practice now and in subsequent chapters, but it is useful at
A simple model to explain city murder rates (murdrate) in terms of
the probability of conviction (prbconv) and average sentence
length (avgsen) is
murdrate ϭ ␤0 ϩ ␤1prbconv ϩ ␤2avgsen ϩ u.
What are some factors contained in u? Do you think the key
assumption (3.5) is likely to hold?
Q U E S T I O N 3 . 1
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 76

TABLE 3.1
Terminology for Multiple Regression
y x1, x2, …, xk
Dependent Variable Independent Variables
Explained Variable Explanatory Variables
Response Variable Control Variables
Predicted Variable Predictor Variables
Regressand Regressors
this point to be reminded of some things we already know. Suppose that CEO salary (salary)
is related to firm sales (sales) and CEO tenure (ceoten) with the firm by
log(salary) ϭ ␤0 ϩ ␤1log(sales) ϩ ␤2ceoten ϩ ␤3ceoten2 ϩ u. (3.7)
This fits into the multiple regression model (with k ϭ 3) by defining y ϭ log(salary),
x1 ϭ log(sales), x2 ϭ ceoten, and x3 ϭ ceoten2. As we know from Chapter 2, the parame-
ter ␤1 is the (ceteris paribus) elasticity of salary with respect to sales. If ␤3 ϭ 0, then 100␤2
is approximately the ceteris paribus percentage increase in salary when ceoten increases
by one year. When ␤3 0, the effect of ceoten on salary is more complicated. We will
postpone a detailed treatment of general models with quadratics until Chapter 6.
Equation (3.7) provides an important reminder about multiple regression analysis. The
term “linear” in multiple linear regression model means that equation (3.6) is linear in the
parameters, ␤j. Equation (3.7) is an example of a multiple regression model that, while
linear in the ␤j, is a nonlinear relationship between salary and the variables sales and
ceoten. Many applications of multiple linear regression involve nonlinear relationships
among the underlying variables.
The key assumption for the general multiple regression model is easy to state in terms
of a conditional expectation:
E(u͉x1,x2, …, xk) ϭ 0. (3.8)
At a minimum, equation (3.8) requires that all factors in the unobserved error term be
uncorrelated with the explanatory variables. It also means that we have correctly
accounted for the functional relationships between the explained and explanatory vari-
ables. Any problem that causes u to be correlated with any of the independent variables
causes (3.8) to fail. In Section 3.3, we will show that assumption (3.8) implies that OLS
is unbiased and will derive the bias that arises when a key variable has been omitted from
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 77

the equation. In Chapters 15 and 16, we will study other reasons that might cause (3.8)
to fail and show what can be done in cases where it does fail.
3.2 Mechanics and Interpretation
of Ordinary Least Squares
We now summarize some computational and algebraic features of the method of ordinary
least squares as it applies to a particular set of data. We also discuss how to interpret the
estimated equation.
Obtaining the OLS Estimates
We first consider estimating the model with two independent variables. The estimated OLS
equation is written in a form similar to the simple regression case:
yˆ ϭ ␤ˆ
0 ϩ bˆ
1x1 ϩ ␤ˆ
2x2, (3.9)
where ␤ˆ
0 is the estimate of ␤0, ␤ˆ
1 is the estimate of ␤1, and ␤ˆ
2 is the estimate of ␤2. But
how do we obtain ␤ˆ
0, ␤ˆ
1, and ␤ˆ
2? The method of ordinary least squares chooses the esti-
mates to minimize the sum of squared residuals. That is, given n observations on y, x1,
and x2, {(xi1, xi2, yi): i ϭ 1, 2, …, n}, the estimates ␤ˆ
0, ␤ˆ
1, and ␤ˆ
2 are chosen simultane-
ously to make
͚
n
iϭ1
(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1 Ϫ ␤ˆ
2xi2)2 (3.10)
as small as possible.
In order to understand what OLS is doing, it is important to master the meaning of
the indexing of the independent variables in (3.10). The independent variables have two
subscripts here, i followed by either 1 or 2. The i subscript refers to the observation
number. Thus, the sum in (3.10) is over all i ϭ 1 to n observations. The second index
is simply a method of distinguishing between different independent variables. In the
example relating wage to educ and exper, xi1 ϭ educi is education for person i in the
sample, and xi2 ϭ experi is experience for person i. The sum of squared residuals in
equation (3.10) is ͚
n
iϭ1 (wagei Ϫ ␤ˆ
0 Ϫ ␤ˆ
1educi Ϫ ␤ˆ
2experi)2. In what follows, the i sub-
script is reserved for indexing the observation number. If we write xij, then this means
the ith observation on the jth independent variable. (Some authors prefer to switch the order
of the observation number and the variable number, so that x1i is observation i on variable
one. But this is just a matter of notational taste.)
In the general case with k independent variables, we seek estimates ␤ˆ
0, ␤ˆ
1, …, ␤ˆ
k in
the equation
yˆ ϭ ␤ˆ
0 ϩ ␤ˆ
1x1 ϩ ␤ˆ
2x2 ϩ … ϩ ␤ˆ
kxk. (3.11)
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 78

The OLS estimates, k ϩ 1 of them, are chosen to minimize the sum of squared residuals:
͚
n
iϭ1
(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1 Ϫ … Ϫ ␤ˆ
kxik)2. (3.12)
This minimization problem can be solved using multivariable calculus (see Appendix 3A).
This leads to k ϩ 1 linear equations in k ϩ 1 unknowns ␤ˆ
0, ␤ˆ
1, …, ␤ˆ
k:
͚
n
iϭ1
(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1 Ϫ … Ϫ ␤ˆ
kxik) ϭ 0
͚
n
iϭ1
xi1(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1 Ϫ … Ϫ ␤ˆ
kxik) ϭ 0
͚
n
iϭ1
xi2(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1 Ϫ … Ϫ ␤ˆ
kxik) ϭ 0 (3.13)
и
и
и
͚
n
iϭ1
xik(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1 Ϫ … Ϫ ␤ˆ
kxik) ϭ 0.
These are often called the OLS first order conditions. As with the simple regression
model in Section 2.2, the OLS first order conditions can be obtained by the method of
moments: under assumption (3.8), E(u) ϭ 0 and E(xju) ϭ 0, where j ϭ 1, 2, …, k. The
equations in (3.13) are the sample counterparts of these population moments, although we
have omitted the division by the sample size n.
For even moderately sized n and k, solving the equations in (3.13) by hand calcula-
tions is tedious. Nevertheless, modern computers running standard statistics and econo-
metrics software can solve these equations with large n and k very quickly.
There is only one slight caveat: we must assume that the equations in (3.13) can be
solved uniquely for the ␤ˆ
j. For now, we just assume this, as it is usually the case in
well-specified models. In Section 3.3, we state the assumption needed for unique OLS
estimates to exist (see Assumption MLR.3).
As in simple regression analysis, equation (3.11) is called the OLS regression line
or the sample regression function (SRF). We will call ␤ˆ
0 the OLS intercept estimate
and ␤ˆ
1, …, ␤ˆ
k the OLS slope estimates (corresponding to the independent variables
x1, x2, …, xk).
In order to indicate that an OLS regression has been run, we will either write out equa-
tion (3.11) with y and x1, …, xk replaced by their variable names (such as wage, educ, and
exper), or we will say that “we ran an OLS regression of y on x1, x2, …, xk” or that “we
regressed y on x1, x2, …, xk.” These are shorthand for saying that the method of ordinary
least squares was used to obtain the OLS equation (3.11). Unless explicitly stated other-
wise, we always estimate an intercept along with the slopes.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 79

Interpreting the OLS Regression Equation
More important than the details underlying the computation of the ␤ˆ
j is the interpretation of
the estimated equation. We begin with the case of two independent variables:
yˆ ϭ ␤ˆ
0 ϩ ␤ˆ
1x1 ϩ ␤ˆ
2x2. (3.14)
The intercept ␤ˆ
0 in equation (3.14) is the predicted value of y when x1 ϭ 0 and x2 ϭ 0.
Sometimes, setting x1 and x2 both equal to zero is an interesting scenario; in other cases,
it will not make sense. Nevertheless, the intercept is always needed to obtain a prediction
of y from the OLS regression line, as (3.14) makes clear.
The estimates ␤ˆ
1 and ␤ˆ
2 have partial effect, or ceteris paribus, interpretations. From
equation (3.14), we have
⌬yˆ ϭ ␤ˆ
1⌬x1 ϩ ␤ˆ
2⌬x2,
so we can obtain the predicted change in y given the changes in x1 and x2. (Note how the
intercept has nothing to do with the changes in y.) In particular, when x2 is held fixed, so
that ⌬x2 ϭ 0, then
⌬yˆ ϭ ␤ˆ
1⌬x1,
holding x2 fixed. The key point is that, by including x2 in our model, we obtain a coeffi-
cient on x1 with a ceteris paribus interpretation. This is why multiple regression analysis
is so useful. Similarly,
⌬yˆ ϭ ␤ˆ
2⌬x2,
holding x1 fixed.
E X A M P L E 3 . 1
(Determinants of College GPA)
The variables in GPA1.RAW include college grade point average (colGPA), high school GPA
(hsGPA), and achievement test score (ACT) for a sample of 141 students from a large univer-
sity; both college and high school GPAs are on a four-point scale. We obtain the following
OLS regression line to predict college GPA from high school GPA and achievement test score:
colGPA ϭ 1.29 ϩ .453 hsGPA ϩ .0094 ACT. (3.15)
How do we interpret this equation? First, the intercept 1.29 is the predicted college GPA if
hsGPA and ACT are both set as zero. Since no one who attends college has either a zero high
school GPA or a zero on the achievement test, the intercept in this equation is not, by itself,
meaningful.
More interesting estimates are the slope coefficients on hsGPA and ACT. As expected, there
is a positive partial relationship between colGPA and hsGPA: holding ACT fixed, another point
on hsGPA is associated with .453 of a point on the college GPA, or almost half a point. In
other words, if we choose two students, A and B, and these students have the same ACT
score, but the high school GPA of Student A is one point higher than the high school GPA of
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 80

Student B, then we predict Student A to have a college GPA .453 higher than that of Student
B. (This says nothing about any two actual people, but it is our best prediction.)
The sign on ACT implies that, while holding hsGPA fixed, a change in the ACT score of 10
points—a very large change, since the average score in the sample is about 24 with a stan-
dard deviation less than three—affects colGPA by less than one-tenth of a point. This is a small
effect, and it suggests that, once high school GPA is accounted for, the ACT score is not a
strong predictor of college GPA. (Naturally, there are many other factors that contribute to
GPA, but here we focus on statistics available for high school students.) Later, after we dis-
cuss statistical inference, we will show that not only is the coefficient on ACT practically small,
it is also statistically insignificant.
If we focus on a simple regression analysis relating colGPA to ACT only, we obtain
colGPA ϭ 2.40 ϩ .0271 ACT;
thus, the coefficient on ACT is almost three times as large as the estimate in (3.15). But this
equation does not allow us to compare two people with the same high school GPA; it corre-
sponds to a different experiment. We say more about the differences between multiple and
simple regression later.
The case with more than two independent variables is similar. The OLS regression
line is
yˆ ϭ ␤ˆ
0 ϩ ␤ˆ
1x1 ϩ ␤ˆ
2x2 ϩ … ϩ ␤ˆ
kxk. (3.16)
Written in terms of changes,
⌬yˆ ϭ ␤ˆ
1⌬x1 ϩ ␤ˆ
2⌬x2 ϩ … ϩ ␤ˆ
k⌬xk. (3.17)
The coefficient on x1 measures the change in yˆ due to a one-unit increase in x1, holding
all other independent variables fixed. That is,
⌬yˆ ϭ ␤ˆ
1⌬x1, (3.18)
holding x2, x3, …, xk fixed. Thus, we have controlled for the variables x2, x3, …, xk when
estimating the effect of x1 on y. The other coefficients have a similar interpretation.
The following is an example with three independent variables.
E X A M P L E 3 . 2
(Hourly Wage Equation)
Using the 526 observations on workers in WAGE1.RAW, we include educ (years of education),
exper (years of labor market experience), and tenure (years with the current employer) in an
equation explaining log(wage). The estimated equation is
log(wage) ϭ .284 ϩ .092 educ ϩ .0041 exper ϩ .022 tenure. (3.19)
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 81

As in the simple regression case, the coefficients have a percentage interpretation. The only
difference here is that they also have a ceteris paribus interpretation. The coefficient .092
means that, holding exper and tenure fixed, another year of education is predicted to
increase log(wage) by .092, which translates into an approximate 9.2 percent [100(.092)]
increase in wage. Alternatively, if we take two people with the same levels of experience and
job tenure, the coefficient on educ is the proportionate difference in predicted wage when
their education levels differ by one year. This measure of the return to education at least
keeps two important productivity factors fixed; whether it is a good estimate of the ceteris
paribus return to another year of education requires us to study the statistical properties of
OLS (see Section 3.3).
On the Meaning of “Holding Other Factors Fixed”
in Multiple Regression
The partial effect interpretation of slope coefficients in multiple regression analysis can
cause some confusion, so we attempt to prevent that problem now.
In Example 3.1, we observed that the coefficient on ACT measures the predicted dif-
ference in colGPA, holding hsGPA fixed. The power of multiple regression analysis is that
it provides this ceteris paribus interpretation even though the data have not been collected
in a ceteris paribus fashion. In giving the coefficient on ACT a partial effect interpretation,
it may seem that we actually went out and sampled people with the same high school GPA
but possibly with different ACT scores. This is not the case. The data are a random sam-
ple from a large university: there were no restrictions placed on the sample values of
hsGPA or ACT in obtaining the data. Rarely do we have the luxury of holding certain vari-
ables fixed in obtaining our sample. If we could collect a sample of individuals with the
same high school GPA, then we could perform a simple regression analysis relating
colGPA to ACT. Multiple regression effectively allows us to mimic this situation without
restricting the values of any independent variables.
The power of multiple regression analysis is that it allows us to do in nonexperimental
environments what natural scientists are able to do in a controlled laboratory setting: keep
other factors fixed.
Changing More than One Independent Variable Simultaneously
Sometimes, we want to change more than one independent variable at the same time to
find the resulting effect on the dependent variable. This is easily done using equation (3.17).
For example, in equation (3.19), we can obtain the estimated effect on wage when an indi-
vidual stays at the same firm for another year: exper (general workforce experience) and
tenure both increase by one year. The total effect (holding educ fixed) is
⌬log(wage) ϭ .0041 ⌬exper ϩ .022 ⌬tenure ϭ .0041 ϩ .022 ϭ .0261,
or about 2.6 percent. Since exper and tenure each increase by one year, we just add the
coefficients on exper and tenure and multiply by 100 to turn the effect into a percent.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 82

OLS Fitted Values and Residuals
After obtaining the OLS regression line (3.11), we can obtain a fitted or predicted value
for each observation. For observation i, the fitted value is simply
yî ϭ ␤ˆ
0 ϩ ␤ˆ
1xi1 ϩ ␤ˆ
2xi2 ϩ … ϩ ␤ˆ
kxik, (3.20)
which is just the predicted value obtained by plugging the values of the independent vari-
ables for observation i into equation (3.11). We should not forget about the intercept in
obtaining the fitted values; otherwise, the answer can be very misleading. As an example,
if in (3.15), hsGPAi ϭ 3.5 and ACTi ϭ 24, colGPAi ϭ 1.29 ϩ .453(3.5) ϩ .0094(24) ϭ
3.101 (rounded to three places after the decimal).
Normally, the actual value yi for any observation i will not equal the predicted value,
yî: OLS minimizes the average squared prediction error, which says nothing about the
prediction error for any particular observation. The residual for observation i is defined
just as in the simple regression case,
uî ϭ yi Ϫ yî. (3.21)
There is a residual for each observation. If uî Ͼ 0, then yî is below yi, which means that,
for this observation, yi is underpredicted. If uî Ͻ 0, then yi Ͻ yî, and yi is overpredicted.
The OLS fitted values and residuals have some important properties that are immedi-
ate extensions from the single variable case:
1. The sample average of the residuals is zero and so y¯ ϭ yˆ¯.
2. The sample covariance between each independent variable and the OLS residuals
is zero. Consequently, the sample covariance between the OLS fitted values and the
OLS residuals is zero.
3. The point (x¯1, x¯2, …, x¯k, y¯) is always on the OLS regression line: y¯ ϭ ␤ˆ
0 ϩ ␤ˆ
1x¯1 ϩ
␤ˆ
2x¯2 ϩ … ϩ ␤ˆ
kx¯k.
The first two properties are immediate
consequences of the set of equations used
to obtain the OLS estimates. The first
equation in (3.13) says that the sum of the
residuals is zero. The remaining equations
are of the form ͚
n
iϭ1 xijuî ϭ 0, which
implies that each independent variable has
zero sample covariance with uî. Property
(3) follows immediately from property (1).
A “Partialling Out” Interpretation of Multiple Regression
When applying OLS, we do not need to know explicit formulas for the ␤ˆ
j that solve the
system of equations in (3.13). Nevertheless, for certain derivations, we do need explicit
formulas for the ␤ˆ
j. These formulas also shed further light on the workings of OLS.
In Example 3.1, the OLS fitted line explaining college GPA in terms
of high school GPA and ACT score is
colGPA ϭ 1.29 ϩ .453 hsGPA ϩ .0094 ACT.
If the average high school GPA is about 3.4 and the average ACT
score is about 24.2, what is the average college GPA in the sample?
Q U E S T I O N 3 . 2
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 83

Consider again the case with k ϭ 2 independent variables, yˆ ϭ ␤ˆ
0 ϩ ␤ˆ
1x1 ϩ ␤ˆ
2x2. For
concreteness, we focus on ␤ˆ
1. One way to express ␤ˆ
1 is
␤ˆ
1 ϭ Θ͚
n
iϭ1
rî1yi Ι ͞ Θ͚
n
iϭ1
rî
2
1 Ι, (3.22)
where the rî1 are the OLS residuals from a simple regression of x1 on x2, using the sample
at hand. We regress our first independent variable, x1, on our second independent variable,
x2, and then obtain the residuals (y plays no role here). Equation (3.22) shows that we can
then do a simple regression of y on rˆ1 to obtain ␤ˆ
1. (Note that the residuals rî1 have a zero
sample average, and so ␤ˆ
1 is the usual slope estimate from simple regression.)
The representation in equation (3.22) gives another demonstration of ␤ˆ
1’s partial effect
interpretation. The residuals rî1 are the part of xi1 that is uncorrelated with xi2. Another way
of saying this is that rî1 is xi1 after the effects of xi2 have been partialled out, or netted out.
Thus, ␤ˆ
1 measures the sample relationship between y and x1 after x2 has been partialled out.
In simple regression analysis, there is no partialling out of other variables because no
other variables are included in the regression. Computer Exercise C3.5 steps you through
the partialling out process using the wage data from Example 3.2. For practical purposes,
the important thing is that ␤ˆ
1 in the equation yˆ ϭ ␤ˆ
0 ϩ ␤ˆ
1x1 ϩ ␤ˆ
2x2 measures the change
in y given a one-unit increase in x1, holding x2 fixed.
In the general model with k explanatory variables, ␤ˆ
1 can still be written as in equa-
tion (3.22), but the residuals rî1 come from the regression of x1 on x2, …, xk. Thus, ␤ˆ
1
measures the effect of x1 on y after x2, …, xk have been partialled or netted out.
Comparison of Simple and Multiple Regression Estimates
Two special cases exist in which the simple regression of y on x1 will produce the same
OLS estimate on x1 as the regression of y on x1 and x2. To be more precise, write the sim-
ple regression of y on x1 as y˜ ϭ ␤˜
0 ϩ ␤˜
1x1, and write the multiple regression as
yˆ ϭ ␤ˆ
0 ϩ ␤ˆ
1x1 ϩ ␤ˆ
2x2. We know that the simple regression coefficient ␤˜
1 does not usu-
ally equal the multiple regression coefficient ␤ˆ
1. It turns out there is a simple relation-
ship between ␤˜
1 and ␤ˆ
1, which allows for interesting comparisons between simple and
multiple regression:
␤˜
1 ϭ ␤ˆ
1 ϩ ␤ˆ
2␦˜
1, (3.23)
where ␦˜
1 is the slope coefficient from the simple regression of xi2 on xi1, i ϭ 1, ..., n. This
equation shows how ␤˜
1 differs from the partial effect of x1 on yˆ. The confounding term is
the partial effect of x2 on yˆ times the slope in the sample regression of x2 on x1. (See Section
3A.4 in the chapter appendix for a more general verification.)
The relationship between ␤˜
1 and ␤ˆ
1 also shows there are two distinct cases where they
are equal:
1. The partial effect of x2 on yˆ is zero in the sample. That is, ␤ˆ
2 ϭ 0.
2. x1 and x2 are uncorrelated in the sample. That is, ␦˜
1 ϭ 0.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 84

Even though simple and multiple regression estimates are almost never identical, we
can use the above formula to characterize why they might be either very different or quite
similar. For example, if ␤ˆ
2 is small, we might expect the multiple and simple regression
estimates of ␤1 to be similar. In Example 3.1, the sample correlation between hsGPA and
ACT is about 0.346, which is a nontrivial correlation. But the coefficient on ACT is fairly
little. It is not surprising to find that the simple regression of colGPA on hsGPA produces
a slope estimate of .482, which is not much different from the estimate .453 in (3.15).
E X A M P L E 3 . 3
[Participation in 401(k) Pension Plans]
We use the data in 401K.RAW to estimate the effect of a plan’s match rate (mrate) on the
participation rate (prate) in its 401(k) pension plan. The match rate is the amount the firm
contributes to a worker’s fund for each dollar the worker contributes (up to some limit); thus,
mrate ϭ .75 means that the firm contributes 75 cents for each dollar contributed by the
worker. The participation rate is the percentage of eligible workers having a 401(k) account.
The variable age is the age of the 401(k) plan. There are 1,534 plans in the data set, the aver-
age prate is 87.36, the average mrate is .732, and the average age is 13.2.
Regressing prate on mrate, age gives
prate ϭ 80.12 ϩ 5.52 mrate ϩ .243 age.
Thus, both mrate and age have the expected effects. What happens if we do not control for
age? The estimated effect of age is not trivial, and so we might expect a large change in the
estimated effect of mrate if age is dropped from the regression. However, the simple regres-
sion of prate on mrate yields prate ϭ 83.08 ϩ 5.86 mrate. The simple regression estimate of
the effect of mrate on prate is clearly different from the multiple regression estimate, but the
difference is not very big. (The simple regression estimate is only about 6.2 percent larger than
the multiple regression estimate.) This can be explained by the fact that the sample correla-
tion between mrate and age is only .12.
In the case with k independent variables, the simple regression of y on x1 and the multi-
ple regression of y on x1, x2, …, xk produce an identical estimate of x1 only if (1) the OLS
coefficients on x2 through xk are all zero or (2) x1 is uncorrelated with each of x2, …, xk.
Neither of these is very likely in practice. But if the coefficients on x2 through xk are small,
or the sample correlations between x1 and the other independent variables are insubstantial,
then the simple and multiple regression estimates of the effect of x1 on y can be similar.
Goodness-of-Fit
As with simple regression, we can define the total sum of squares (SST), the explained sum
of squares (SSE), and the residual sum of squares or sum of squared residuals (SSR) as
SST ϵ ͚
n
iϭ1
(yi Ϫ y¯)2
(3.24)
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 85

SSE ϵ ͚
n
iϭ1
(yî Ϫ y¯)2
(3.25)
SSR ϵ ͚
n
iϭ1
uî
2. (3.26)
Using the same argument as in the simple regression case, we can show that
SST ϭ SSE ϩ SSR. (3.27)
In other words, the total variation in {yi} is the sum of the total variations in {yî} and in
{uî}.
Assuming that the total variation in y is nonzero, as is the case unless yi is constant in
the sample, we can divide (3.27) by SST to get
SSR/SST ϩ SSE/SST ϭ 1.
Just as in the simple regression case, the R-squared is defined to be
R2 ϵ SSE/SST ϭ 1 Ϫ SSR/SST, (3.28)
and it is interpreted as the proportion of the sample variation in yi that is explained by the
OLS regression line. By definition, R2 is a number between zero and one.
R2 can also be shown to equal the squared correlation coefficient between the actual
yi and the fitted values yî. That is,
R2 ϭ . (3.29)
[We have put the average of the yî in (3.29) to be true to the formula for a correlation coef-
ficient; we know that this average equals y¯ because the sample average of the residuals is
zero and yi ϭ yî ϩ uî.]
An important fact about R2 is that it never decreases, and it usually increases when
another independent variable is added to a regression. This algebraic fact follows because,
by definition, the sum of squared residuals never increases when additional regressors are
added to the model. For example, the last digit of one’s social security number has noth-
ing to do with one’s hourly wage, but adding this digit to a wage equation will increase
the R2 (by a little, at least).
The fact that R2 never decreases when any variable is added to a regression makes it a
poor tool for deciding whether one variable or several variables should be added to a model.
The factor that should determine whether an explanatory variable belongs in a model is
whether the explanatory variable has a nonzero partial effect on y in the population. We
Θ͚
n
iϭ1
(yi Ϫ y¯) (yî Ϫ yˆ¯)Ι
2
Θ͚
n
iϭ1
(yi Ϫ y¯)2
ΙΘ͚
n
iϭ1
(yî Ϫ yˆ¯)2
Ι
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 86

will show how to test this hypothesis in Chapter 4 when we cover statistical inference.
We will also see that, when used properly, R2 allows us to test a group of variables to see
if it is important for explaining y. For now, we use it as a goodness-of-fit measure for a
given model.
E X A M P L E 3 . 4
(Determinants of College GPA)
From the grade point average regression that we did earlier, the equation with R2 is
colGPA ϭ 1.29 ϩ .453 hsGPA ϩ .0094 ACT
n ϭ 141, R2 ϭ .176.
This means that hsGPA and ACT together explain about 17.6 percent of the variation in college
GPA for this sample of students. This may not seem like a high percentage, but we must
remember that there are many other factors—including family background, personality, qual-
ity of high school education, affinity for college—that contribute to a student’s college per-
formance. If hsGPA and ACT explained almost all of the variation in colGPA, then performance
in college would be preordained by high school performance!
E X A M P L E 3 . 5
(Explaining Arrest Records)
CRIME1.RAW contains data on arrests during the year 1986 and other information on 2,725 men
born in either 1960 or 1961 in California. Each man in the sample was arrested at least once prior
to 1986. The variable narr86 is the number of times the man was arrested during 1986: it is zero
for most men in the sample (72.29 percent), and it varies from 0 to 12. (The percentage of men
arrested once during 1986 was 20.51.) The variable pcnv is the proportion (not percentage) of
arrests prior to 1986 that led to conviction, avgsen is average sentence length served for prior
convictions (zero for most people), ptime86 is months spent in prison in 1986, and qemp86 is
the number of quarters during which the man was employed in 1986 (from zero to four).
A linear model explaining arrests is
narr86 ϭ ␤0 ϩ ␤1pcnv ϩ ␤2avgsen ϩ ␤3ptime86 ϩ ␤4qemp86 ϩ u,
where pcnv is a proxy for the likelihood for being convicted of a crime and avgsen is a mea-
sure of expected severity of punishment, if convicted. The variable ptime86 captures the incar-
cerative effects of crime: if an individual is in prison, he cannot be arrested for a crime out-
side of prison. Labor market opportunities are crudely captured by qemp86.
First, we estimate the model without the variable avgsen. We obtain
narr86 ϭ .712 Ϫ .150 pcnv Ϫ .034 ptime86 Ϫ .104 qemp86
n ϭ 2,725, R2 ϭ .0413.
This equation says that, as a group, the three variables pcnv, ptime86, and qemp86 explain
about 4.1 percent of the variation in narr86.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 87

Each of the OLS slope coefficients has the anticipated sign. An increase in the proportion of
convictions lowers the predicted number of arrests. If we increase pcnv by .50 (a large increase
in the probability of conviction), then, holding the other factors fixed, ⌬narr86 ϭ Ϫ.150(.50)
ϭ Ϫ.075. This may seem unusual because an arrest cannot change by a fraction. But we can
use this value to obtain the predicted change in expected arrests for a large group of men. For
example, among 100 men, the predicted fall in arrests when pcnv increases by .50 is Ϫ7.5.
Similarly, a longer prison term leads to a lower predicted number of arrests. In fact, if
ptime86 increases from 0 to 12, predicted arrests for a particular man fall by .034(12) ϭ .408.
Another quarter in which legal employment is reported lowers predicted arrests by .104, which
would be 10.4 arrests among 100 men.
If avgsen is added to the model, we know that R2 will increase. The estimated equation is
narr86 ϭ .707 Ϫ .151 pcnv ϩ .0074 avgsen Ϫ .037 ptime86 Ϫ .103 qemp86
n ϭ 2,725, R2 ϭ .0422.
Thus, adding the average sentence variable increases R2 from .0413 to .0422, a practically
small effect. The sign of the coefficient on avgsen is also unexpected: it says that a longer
average sentence length increases criminal activity.
Example 3.5 deserves a final word of caution. The fact that the four explanatory vari-
ables included in the second regression explain only about 4.2 percent of the variation in
narr86 does not necessarily mean that the equation is useless. Even though these variables
collectively do not explain much of the variation in arrests, it is still possible that the OLS
estimates are reliable estimates of the ceteris paribus effects of each independent variable
on narr86. As we will see, whether this is the case does not directly depend on the size
of R2. Generally, a low R2 indicates that it is hard to predict individual outcomes on y with
much accuracy, something we study in more detail in Chapter 6. In the arrest example,
the small R2 reflects what we already suspect in the social sciences: it is generally very
difficult to predict individual behavior.
Regression through the Origin
Sometimes, an economic theory or common sense suggests that ␤0 should be zero, and so
we should briefly mention OLS estimation when the intercept is zero. Specifically, we
now seek an equation of the form
y˜ ϭ ␤˜
1x1 ϩ ␤˜
2x2 ϩ … ϩ ␤˜
kxk, (3.30)
where the symbol “~” over the estimates is used to distinguish them from the OLS esti-
mates obtained along with the intercept [as in (3.11)]. In (3.30), when x1 ϭ 0, x2 ϭ 0, …,
xk ϭ 0, the predicted value is zero. In this case, ␤˜
1, …, ␤˜
k are said to be the OLS estimates
from the regression of y on x1, x2, …, xk through the origin.
The OLS estimates in (3.30), as always, minimize the sum of squared residuals, but
with the intercept set at zero. You should be warned that the properties of OLS that
we derived earlier no longer hold for regression through the origin. In particular, the
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 88

OLS residuals no longer have a zero sample average. Further, if R2 is defined as 1 Ϫ
SSR/SST, where SST is given in (3.24) and SSR is now ͚
n
iϭ1 (yi Ϫ ␤˜
1xi1 Ϫ … Ϫ␤˜
kxik)2,
then R2 can actually be negative. This means that the sample average, y¯, “explains” more
of the variation in the yi than the explanatory variables. Either we should include an inter-
cept in the regression or conclude that the explanatory variables poorly explain y. In order
to always have a nonnegative R-squared, some economists prefer to calculate R2 as the
squared correlation coefficient between the actual and fitted values of y, as in (3.29). (In
this case, the average fitted value must be computed directly since it no longer equals y¯.)
However, there is no set rule on computing R-squared for regression through the origin.
One serious drawback with regression through the origin is that, if the intercept ␤0 in the
population model is different from zero, then the OLS estimators of the slope parameters
will be biased. The bias can be severe in some cases. The cost of estimating an intercept
when ␤0 is truly zero is that the variances of the OLS slope estimators are larger.
3.3 The Expected Value of the OLS Estimators
We now turn to the statistical properties of OLS for estimating the parameters in an under-
lying population model. In this section, we derive the expected value of the OLS estima-
tors. In particular, we state and discuss four assumptions, which are direct extensions of
the simple regression model assumptions, under which the OLS estimators are unbiased
for the population parameters. We also explicitly obtain the bias in OLS when an impor-
tant variable has been omitted from the regression.
You should remember that statistical properties have nothing to do with a particu-
lar sample, but rather with the property of estimators when random sampling is done
repeatedly. Thus, Sections 3.3, 3.4, and 3.5 are somewhat abstract. Although we give
examples of deriving bias for particular models, it is not meaningful to talk about
the statistical properties of a set of estimates obtained from a single sample.
The first assumption we make simply defines the multiple linear regression (MLR) model.
Assumption MLR.1 (Linear in Parameters)
The model in the population can be written as
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ … ϩ ␤kxk ϩ u, (3.31)
where ␤0, ␤1, …, ␤k are the unknown parameters (constants) of interest and u is an unob-
servable random error or disturbance term.
Equation (3.31) formally states the population model, sometimes called the true model,
to allow for the possibility that we might estimate a model that differs from (3.31). The
key feature is that the model is linear in the parameters ␤0, ␤1, …, ␤k. As we know, (3.31)
is quite flexible because y and the independent variables can be arbitrary functions of the
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 89

underlying variables of interest, such as natural logarithms and squares [see, for example,
equation (3.7)].
Assumption MLR.2 (Random Sampling)
We have a random sample of n observations, {(xi1,xi2,…,xik,yi): i ϭ 1,2,…,n}, following the
population model in Assumption MLR.1.
Sometimes, we need to write the equation for a particular observation i: for a randomly
drawn observation from the population, we have
yi ϭ ␤0 ϩ ␤1xi1 ϩ ␤2xi2 ϩ … ϩ ␤kxik ϩ ui. (3.32)
Remember that i refers to the observation, and the second subscript on x is the variable
number. For example, we can write a CEO salary equation for a particular CEO i as
log(salaryi) ϭ ␤0 ϩ ␤1log(salesi) ϩ ␤2ceoteni ϩ ␤3ceoteni
2 ϩ ui. (3.33)
The term ui contains the unobserved factors for CEO i that affect his or her salary. For
applications, it is usually easiest to write the model in population form, as in (3.31). It
contains less clutter and emphasizes the fact that we are interested in estimating a popu-
lation relationship.
In light of model (3.31), the OLS estimators ␤ˆ
0, ␤ˆ
1, ␤ˆ
2, …, ␤ˆ
k from the regression of y
on x1, …, xk are now considered to be estimators of ␤0, ␤1, …, ␤k. We saw, in Section 3.2,
that OLS chooses the estimates for a particular sample so that the residuals average out to
zero and the sample correlation between each independent variable and the residuals is zero.
Still, we need an assumption that ensures the OLS estimators are well defined.
Assumption MLR.3 (No Perfect Collinearity)
In the sample (and therefore in the population), none of the independent variables is constant,
and there are no exact linear relationships among the independent variables.
Assumption MLR.3 is more complicated than its counterpart for simple regression because
we must now look at relationships between all independent variables. If an independent vari-
able in (3.31) is an exact linear combination of the other independent variables, then we say
the model suffers from perfect collinearity, and it cannot be estimated by OLS.
It is important to note that Assumption MLR.3 does allow the independent variables
to be correlated; they just cannot be perfectly correlated. If we did not allow for any cor-
relation among the independent variables, then multiple regression would be of very lim-
ited use for econometric analysis. For example, in the model relating test scores to edu-
cational expenditures and average family income,
avgscore ϭ ␤0 ϩ ␤1expend ϩ ␤2avginc ϩ u,
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 90

we fully expect expend and avginc to be correlated: school districts with high average
family incomes tend to spend more per student on education. In fact, the primary moti-
vation for including avginc in the equation is that we suspect it is correlated with expend,
and so we would like to hold it fixed in the analysis. Assumption MLR.3 only rules out
perfect correlation between expend and avginc in our sample. We would be very unlucky
to obtain a sample where per student expenditures are perfectly correlated with average
family income. But some correlation, perhaps a substantial amount, is expected and cer-
tainly allowed.
The simplest way that two independent variables can be perfectly correlated is when
one variable is a constant multiple of another. This can happen when a researcher inad-
vertently puts the same variable measured in different units into a regression equation.
For example, in estimating a relationship between consumption and income, it makes
no sense to include as independent variables income measured in dollars as well as
income measured in thousands of dollars. One of these is redundant. What sense would
it make to hold income measured in dollars fixed while changing income measured in
thousands of dollars?
We already know that different nonlinear functions of the same variable can appear
among the regressors. For example, the model cons ϭ ␤0 ϩ ␤1inc ϩ ␤2inc2 ϩ u does
not violate Assumption MLR.3: even though x2 ϭ inc2 is an exact function of x1 ϭ inc,
inc2 is not an exact linear function of inc. Including inc2 in the model is a useful way to
generalize functional form, unlike including income measured in dollars and in thou-
sands of dollars.
Common sense tells us not to include the same explanatory variable measured in dif-
ferent units in the same regression equation. There are also more subtle ways that one
independent variable can be a multiple of another. Suppose we would like to estimate an
extension of a constant elasticity consumption function. It might seem natural to specify
a model such as
log(cons) ϭ ␤0 ϩ ␤1log(inc) ϩ ␤2log(inc2) ϩ u, (3.34)
where x1 ϭ log(inc) and x2 ϭ log(inc2). Using the basic properties of the natural log (see
Appendix A), log(inc2) ϭ 2иlog(inc). That is, x2 ϭ 2x1, and naturally this holds for all
observations in the sample. This violates Assumption MLR.3. What we should do instead
is include [log(inc)]2, not log(inc2), along with log(inc). This is a sensible extension of the
constant elasticity model, and we will see how to interpret such models in Chapter 6.
Another way that independent variables can be perfectly collinear is when one inde-
pendent variable can be expressed as an exact linear function of two or more of the other
independent variables. For example, suppose we want to estimate the effect of campaign
spending on campaign outcomes. For simplicity, assume that each election has two can-
didates. Let voteA be the percentage of the vote for Candidate A, let expendA be campaign
expenditures by Candidate A, let expendB be campaign expenditures by Candidate B, and
let totexpend be total campaign expenditures; the latter three variables are all measured in
dollars. It may seem natural to specify the model as
voteA ϭ ␤0 ϩ ␤1expendA ϩ ␤2expendB ϩ ␤3totexpend ϩ u, (3.35)
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 91

in order to isolate the effects of spending by each candidate and the total amount of spend-
ing. But this model violates Assumption MLR.3 because x3 ϭ x1 ϩ x2 by definition. Trying
to interpret this equation in a ceteris paribus fashion reveals the problem. The parameter
of ␤1 in equation (3.35) is supposed to measure the effect of increasing expenditures by
Candidate A by one dollar on Candidate A’s vote, holding Candidate B’s spending and
total spending fixed. This is nonsense, because if expendB and totexpend are held fixed,
then we cannot increase expendA.
The solution to the perfect collinearity in (3.35) is simple: drop any one of the three
variables from the model. We would probably drop totexpend, and then the coefficient on
expendA would measure the effect of increasing expenditures by A on the percentage of
the vote received by A, holding the spending by B fixed.
The prior examples show that Assumption MLR.3 can fail if we are not careful in spec-
ifying our model. Assumption MLR.3 also fails if the sample size, n, is too small in rela-
tion to the number of parameters being
estimated. In the general regression model
in equation (3.31), there are k ϩ 1 param-
eters, and MLR.3 fails if n Ͻ k ϩ 1. Intu-
itively, this makes sense: to estimate k ϩ 1
parameters, we need at least k ϩ 1 obser-
vations. Not surprisingly, it is better to
have as many observations as possible,
something we will see with our variance calculations in Section 3.4.
If the model is carefully specified and n Ն k ϩ 1, Assumption MLR.3 can fail in rare
cases due to bad luck in collecting the sample. For example, in a wage equation with edu-
cation and experience as variables, it is possible that we could obtain a random sample
where each individual has exactly twice as much education as years of experience. This
scenario would cause Assumption MLR.3 to fail, but it can be considered very unlikely
unless we have an extremely small sample size.
The final, and most important, assumption needed for unbiasedness is a direct exten-
sion of Assumption SLR.4.
Assumption MLR.4 (Zero Conditional Mean)
The error u has an expected value of zero given any values of the independent variables. In
other words,
E(u͉x1,x2, …,xk) ϭ 0. (3.36)
One way that Assumption MLR.4 can fail is if the functional relationship between the
explained and explanatory variables is misspecified in equation (3.31): for example, if we
forget to include the quadratic term inc2 in the consumption function cons ϭ ␤0 ϩ ␤1inc
ϩ ␤2inc2 ϩ u when we estimate the model. Another functional form misspecification
occurs when we use the level of a variable when the log of the variable is what actually
shows up in the population model, or vice versa. For example, if the true model has
In the previous example, if we use as explanatory variables expendA,
expendB, and shareA, where shareA ϭ 100и(expendA/totexpend)
is the percentage share of total campaign expenditures made by
Candidate A, does this violate Assumption MLR.3?
Q U E S T I O N 3 . 3
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 92

log(wage) as the dependent variable but we use wage as the dependent variable in our
regression analysis, then the estimators will be biased. Intuitively, this should be pretty
clear. We will discuss ways of detecting functional form misspecification in Chapter 9.
Omitting an important factor that is correlated with any of x1, x2, …, xk causes
Assumption MLR.4 to fail also. With multiple regression analysis, we are able to include
many factors among the explanatory variables, and omitted variables are less likely to be
a problem in multiple regression analysis than in simple regression analysis. Neverthe-
less, in any application, there are always factors that, due to data limitations or ignorance,
we will not be able to include. If we think these factors should be controlled for and they
are correlated with one or more of the independent variables, then Assumption MLR.4
will be violated. We will derive this bias later.
There are other ways that u can be correlated with an explanatory variable. In
Chapter 15, we will discuss the problem of measurement error in an explanatory vari-
able. In Chapter 16, we cover the conceptually more difficult problem in which one or
more of the explanatory variables is determined jointly with y. We must postpone our
study of these problems until we have a firm grasp of multiple regression analysis under
an ideal set of assumptions.
When Assumption MLR.4 holds, we often say that we have exogenous explanatory
variables. If xj is correlated with u for any reason, then xj is said to be an endogenous
explanatory variable. The terms “exogenous” and “endogenous” originated in simulta-
neous equations analysis (see Chapter 16), but the term “endogenous explanatory vari-
able” has evolved to cover any case in which an explanatory variable may be cor-
related with the error term.
Before we show the unbiasedness of the OLS estimators under MLR.1 to MLR.4, a word
of caution. Beginning students of econometrics sometimes confuse Assumptions MLR.3 and
MLR.4, but they are quite different. Assumption MLR.3 rules out certain relationships
among the independent or explanatory variables and has nothing to do with the error, u. You
will know immediately when carrying out OLS estimation whether or not Assumption
MLR.3 holds. On the other hand, Assumption MLR.4—the much more important of the
two—restricts the relationship between the unobservables in u and the explanatory variables.
Unfortunately, we will never know for sure whether the average value of the unobservables
is unrelated to the explanatory variables. But this is the critical assumption.
We are now ready to show unbiasedness of OLS under the first four multiple regres-
sion assumptions. As in the simple regression case, the expectations are conditional on
the values of the explanatory variables in the sample, something we show explicitly in
Appendix 3A but not in the text.
Theorem 3.1 (Unbiasedness of OLS)
Under Assumptions MLR.1 through MLR.4,
E(␤ˆ
j) ϭ ␤j, j ϭ 0, 1, …, k, (3.37)
for any values of the population parameter ␤j. In other words, the OLS estimators are unbi-
ased estimators of the population parameters.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 93

In our previous empirical examples, Assumption MLR.3 has been satisfied (because
we have been able to compute the OLS estimates). Furthermore, for the most part, the
samples are randomly chosen from a well-defined population. If we believe that the spec-
ified models are correct under the key Assumption MLR.4, then we can conclude that OLS
is unbiased in these examples.
Since we are approaching the point where we can use multiple regression in serious
empirical work, it is useful to remember the meaning of unbiasedness. It is tempting, in
examples such as the wage equation in (3.19), to say something like “9.2 percent is an
unbiased estimate of the return to education.” As we know, an estimate cannot be unbi-
ased: an estimate is a fixed number, obtained from a particular sample, which usually is
not equal to the population parameter. When we say that OLS is unbiased under Assump-
tions MLR.1 through MLR.4, we mean that the procedure by which the OLS estimates
are obtained is unbiased when we view the procedure as being applied across all possible
random samples. We hope that we have obtained a sample that gives us an estimate close
to the population value, but, unfortunately, this cannot be assured. What is assured is that
we have no reason to believe our estimate is more likely to be too big or more likely to
be too small.
Including Irrelevant Variables in a Regression Model
One issue that we can dispense with fairly quickly is that of inclusion of an irrelevant
variable or overspecifying the model in multiple regression analysis. This means that
one (or more) of the independent variables is included in the model even though it has no
partial effect on y in the population. (That is, its population coefficient is zero.)
To illustrate the issue, suppose we specify the model as
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ ␤3x3 ϩ u, (3.38)
and this model satisfies Assumptions MLR.1 through MLR.4. However, x3 has no effect
on y after x1 and x2 have been controlled for, which means that ␤3 ϭ 0. The variable x3
may or may not be correlated with x1 or x2; all that matters is that, once x1 and x2 are con-
trolled for, x3 has no effect on y. In terms of conditional expectations, E(y͉x1,x2,x3) ϭ
E(y͉x1,x2) ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2.
Because we do not know that ␤3 ϭ 0, we are inclined to estimate the equation
including x3:
yˆ ϭ ␤ˆ
0 ϩ ␤ˆ
1x1 ϩ ␤ˆ
2x2 ϩ ␤ˆ
3x3. (3.39)
We have included the irrelevant variable, x3, in our regression. What is the effect of includ-
ing x3 in (3.39) when its coefficient in the population model (3.38) is zero? In terms of the
unbiasedness of ␤ˆ
1 and ␤ˆ
2, there is no effect. This conclusion requires no special derivation,
as it follows immediately from Theorem 3.1. Remember, unbiasedness means E(␤ˆ
j) ϭ ␤j
for any value of ␤j, including ␤j ϭ 0. Thus, we can conclude that E(␤ˆ
0) ϭ ␤0, E(␤ˆ
1) ϭ ␤1,
E(␤ˆ
2) ϭ ␤2, and E(␤ˆ
3) ϭ 0 (for any values of ␤0, ␤1, and ␤2). Even though ␤ˆ
3 itself will
never be exactly zero, its average value across all random samples will be zero.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 94

The conclusion of the preceding example is much more general: including one or more
irrelevant variables in a multiple regression model, or overspecifying the model, does not
affect the unbiasedness of the OLS estimators. Does this mean it is harmless to include
irrelevant variables? No. As we will see in Section 3.4, including irrelevant variables can
have undesirable effects on the variances of the OLS estimators.
Omitted Variable Bias: The Simple Case
Now suppose that, rather than including an irrelevant variable, we omit a variable that
actually belongs in the true (or population) model. This is often called the problem of
excluding a relevant variable or underspecifying the model. We claimed in Chapter 2
and earlier in this chapter that this problem generally causes the OLS estimators to be
biased. It is time to show this explicitly and, just as importantly, to derive the direction
and size of the bias.
Deriving the bias caused by omitting an important variable is an example of mis-
specification analysis. We begin with the case where the true population model has two
explanatory variables and an error term:
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ u, (3.40)
and we assume that this model satisfies Assumptions MLR.1 through MLR.4.
Suppose that our primary interest is in ␤1, the partial effect of x1 on y. For example, y
is hourly wage (or log of hourly wage), x1 is education, and x2 is a measure of innate abil-
ity. In order to get an unbiased estimator of ␤1, we should run a regression of y on x1 and
x2 (which gives unbiased estimators of ␤0, ␤1, and ␤2). However, due to our ignorance or
data unavailability, we estimate the model by excluding x2. In other words, we perform a
simple regression of y on x1 only, obtaining the equation
y˜ ϭ ␤˜
0 ϩ ␤˜
1x1. (3.41)
We use the symbol “~” rather than “^” to emphasize that ␤˜
1 comes from an underspeci-
fied model.
When first learning about the omitted variable problem, it can be difficult to distin-
guish between the underlying true model, (3.40) in this case, and the model that we actu-
ally estimate, which is captured by the regression in (3.41). It may seem silly to omit the
variable x2 if it belongs in the model, but often we have no choice. For example, suppose
that wage is determined by
wage ϭ ␤0 ϩ ␤1educ ϩ ␤2abil ϩ u. (3.42)
Since ability is not observed, we instead estimate the model
wage ϭ ␤0 ϩ ␤1educ ϩ v,
where v ϭ ␤2abil ϩ u. The estimator of ␤1 from the simple regression of wage on educ
is what we are calling ␤˜
1.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 95

We derive the expected value of ␤˜
1 conditional on the sample values of x1 and x2. Deriv-
ing this expectation is not difficult because ␤˜
1 is just the OLS slope estimator from a sim-
ple regression, and we have already studied this estimator extensively in Chapter 2. The
difference here is that we must analyze its properties when the simple regression model
is misspecified due to an omitted variable.
As it turns out, we have done almost all of the work to derive the bias in the simple
regression estimator of ␤˜
1. From equation (3.23) we have the algebraic relationship ␤˜
1 ϭ
␤ˆ
1 ϩ ␤ˆ
2d˜
1, where ␤ˆ
1 and ␤ˆ
2 are the slope estimators (if we could have them) from the mul-
tiple regression
yi on xi1, xi2, i ϭ 1, . . . ,n (3.43)
and d˜
1 is the slope from the simple regression
xi2 on xi1, i ϭ 1, . . . ,n. (3.44)
Because d˜
1 depends only on the independent variables in the sample, we treat it as fixed
(nonrandom) when computing E(␤˜
1). Further, since the model in (3.40) satisfies
Assumptions MLR.1 to MLR.4, we know that ␤ˆ
1 and ␤ˆ
2 would be unbiased for ␤1 and
␤2, respectively. Therefore,
E(␤˜
1) ϭ E(␤ˆ
1 ϩ ␤ˆ
2d˜
1) ϭ E(␤ˆ
1) ϩ E(␤ˆ
2)d˜
1
ϭ ␤1 ϩ ␤2d˜
1,
(3.45)
which implies the bias in ␤˜
1 is
Bias(␤˜
1) ϭ E(␤˜
1) Ϫ ␤1 ϭ ␤2d˜
1. (3.46)
Because the bias in this case arises from omitting the explanatory variable x2, the term on
the right-hand side of equation (3.46) is often called the omitted variable bias.
From equation (3.46), we see that there are two cases where ␤˜
1 is unbiased. The first
is pretty obvious: if ␤2 ϭ 0—so that x2 does not appear in the true model (3.40)—then
␤˜
1 is unbiased. We already know this from the simple regression analysis in Chapter 2.
The second case is more interesting. If ␦˜
1 ϭ 0, then ␤˜
1 is unbiased for ␤1, even if ␤2 0.
Because ␦˜
1 is the sample covariance between x1 and x2 over the sample variance of x1,
␦˜
1 ϭ 0 if, and only if, x1 and x2 are uncorrelated in the sample. Thus, we have the impor-
tant conclusion that, if x1 and x2 are uncorrelated in the sample, then ␤˜
1 is unbiased. This
is not surprising: in Section 3.2, we showed that the simple regression estimator ␤˜
1 and
the multiple regression estimator ␤ˆ
1 are the same when x1 and x2 are uncorrelated in the
sample. [We can also show that ␤˜
1 is unbiased without conditioning on the xi2 if E(x2͉x1)
ϭ E(x2); then, for estimating ␤1, leaving x2 in the error term does not violate the zero
conditional mean assumption for the error, once we adjust the intercept.]
When x1 and x2 are correlated, ␦˜
1 has the same sign as the correlation between x1 and
x2: ␦˜
1 Ͼ 0 if x1 and x2 are positively correlated and ␦˜
1 Ͻ 0 if x1 and x2 are negatively corre-
lated. The sign of the bias in ␤˜
1 depends on the signs of both ␤2 and ␦˜
1 and is summarized
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 96

TABLE 3.2
Summary of Bias in ␤˜1 When x2 Is Omitted in Estimating Equation (3.40)
Corr(x1,x2) > 0 Corr(x1,x2) < 0
␤2 Ͼ 0 positive bias negative bias
␤2 Ͻ 0 negative bias positive bias
in Table 3.2 for the four possible cases when there is bias. Table 3.2 warrants careful study.
For example, the bias in ␤˜
1 is positive if ␤2 Ͼ 0 (x2 has a positive effect on y) and x1 and
x2 are positively correlated, the bias is negative if ␤2 Ͼ 0 and x1 and x2 are negatively cor-
related, and so on.
Table 3.2 summarizes the direction of the bias, but the size of the bias is also very
important. A small bias of either sign need not be a cause for concern. For example, if
the return to education in the population is 8.6 percent and the bias in the OLS esti-
mator is 0.1 percent (a tenth of one percentage point), then we would not be very con-
cerned. On the other hand, a bias on the order of three percentage points would be
much more serious. The size of the bias is determined by the sizes of ␤2 and ␦˜
1.
In practice, since ␤2 is an unknown population parameter, we cannot be certain
whether ␤2 is positive or negative. Nevertheless, we usually have a pretty good idea about
the direction of the partial effect of x2 on y. Further, even though the sign of the correla-
tion between x1 and x2 cannot be known if x2 is not observed, in many cases, we can make
an educated guess about whether x1 and x2 are positively or negatively correlated.
In the wage equation (3.42), by definition, more ability leads to higher productivity
and therefore higher wages: ␤2 Ͼ 0. Also, there are reasons to believe that educ and
abil are positively correlated: on average, individuals with more innate ability choose
higher levels of education. Thus, the OLS estimates from the simple regression equation
wage ϭ ␤0 ϩ ␤1educ ϩ v are on average too large. This does not mean that the estimate
obtained from our sample is too big. We can only say that if we collect many random sam-
ples and obtain the simple regression estimates each time, then the average of these esti-
mates will be greater than ␤1.
E X A M P L E 3 . 6
(Hourly Wage Equation)
Suppose the model log(wage) ϭ ␤0 ϩ ␤1educ ϩ ␤2abil ϩ u satisfies Assumptions MLR.1
through MLR.4. The data set in WAGE1.RAW does not contain data on ability, so we estimate
␤1 from the simple regression
log(wage) ϭ .584 ϩ .083 educ
n ϭ 526, R2 ϭ .186.
(3.47)
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 97

This is the result from only a single sample, so we cannot say that .083 is greater than ␤1; the
true return to education could be lower or higher than 8.3 percent (and we will never know
for sure). Nevertheless, we know that the average of the estimates across all random samples
would be too large.
As a second example, suppose that, at the elementary school level, the average score
for students on a standardized exam is determined by
avgscore ϭ ␤0 ϩ ␤1expend ϩ ␤2povrate ϩ u, (3.48)
where expend is expenditure per student and povrate is the poverty rate of the children in
the school. Using school district data, we only have observations on the percentage of stu-
dents with a passing grade and per student expenditures; we do not have information on
poverty rates. Thus, we estimate ␤1 from the simple regression of avgscore on expend.
We can again obtain the likely bias in ␤˜
1. First, ␤2 is probably negative: there is ample
evidence that children living in poverty score lower, on average, on standardized tests.
Second, the average expenditure per student is probably negatively correlated with the
poverty rate: the higher the poverty rate, the lower the average per student spending, so
that Corr(x1, x2) Ͻ 0. From Table 3.2, ␤˜
1 will have a positive bias. This observation has
important implications. It could be that the true effect of spending is zero; that is, ␤1 ϭ 0.
However, the simple regression estimate of ␤1 will usually be greater than zero, and this
could lead us to conclude that expenditures are important when they are not.
When reading and performing empirical work in economics, it is important to master
the terminology associated with biased estimators. In the context of omitting a variable
from model (3.40), if E(␤˜
1) Ͼ ␤1, then we say that ␤˜
1 has an upward bias. When E(␤˜
1)
Ͻ ␤1, ␤˜
1 has a downward bias. These definitions are the same whether ␤1 is positive
or negative. The phrase biased towards zero refers to cases where E(␤˜
1) is closer to
zero than ␤1. Therefore, if ␤1 is positive, then ␤˜
1 is biased towards zero if it has a
downward bias. On the other hand, if ␤1 Ͻ 0, then ␤˜
1 is biased towards zero if it has an
upward bias.
Omitted Variable Bias: More General Cases
Deriving the sign of omitted variable bias when there are multiple regressors in the esti-
mated model is more difficult. We must remember that correlation between a single
explanatory variable and the error generally results in all OLS estimators being biased.
For example, suppose the population model
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ ␤3x3 ϩ u (3.49)
satisfies Assumptions MLR.1 through MLR.4. But we omit x3 and estimate the model as
y˜ ϭ ␤˜
0 ϩ ␤˜
1x1 ϩ ␤˜
2x2. (3.50)
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 98

Now, suppose that x2 and x3 are uncorrelated, but that x1 is correlated with x3. In other
words, x1 is correlated with the omitted variable, but x2 is not. It is tempting to think that,
while ␤˜
1 is probably biased based on the derivation in the previous subsection, ␤˜
2 is unbi-
ased because x2 is uncorrelated with x3. Unfortunately, this is not generally the case: both
␤˜
1 and ␤˜
2 will normally be biased. The only exception to this is when x1 and x2 are also
uncorrelated.
Even in the fairly simple model above, it can be difficult to obtain the direction of bias
in ␤˜
1 and ␤˜
2. This is because x1, x2, and x3 can all be pairwise correlated. Nevertheless, an
approximation is often practically useful. If we assume that x1 and x2 are uncorrelated, then
we can study the bias in ␤˜
1 as if x2 were absent from both the population and the estimated
models. In fact, when x1 and x2 are uncorrelated, it can be shown that
E(␤˜
1) ϭ ␤1 ϩ ␤3 .
This is just like equation (3.45), but ␤3 replaces ␤2, and x3 replaces x2 in regression (3.44).
Therefore, the bias in ␤˜
1 is obtained by replacing ␤2 with ␤3 and x2 with x3 in Table 3.2. If
␤3 Ͼ 0 and Corr(x1, x3) Ͼ 0, the bias in ␤˜
1 is positive, and so on.
As an example, suppose we add exper to the wage model:
wage ϭ ␤0 ϩ ␤1educ ϩ ␤2exper ϩ ␤3abil ϩ u.
If abil is omitted from the model, the estimators of both ␤1 and ␤2 are biased, even if we
assume exper is uncorrelated with abil. We are mostly interested in the return to educa-
tion, so it would be nice if we could conclude that ␤˜
1 has an upward or a downward bias
due to omitted ability. This conclusion is not possible without further assumptions. As an
approximation, let us suppose that, in addition to exper and abil being uncorrelated, educ
and exper are also uncorrelated. (In reality, they are somewhat negatively correlated.)
Since ␤3 Ͼ 0 and educ and abil are positively correlated, ␤˜
1 would have an upward bias,
just as if exper were not in the model.
The reasoning used in the previous example is often followed as a rough guide for
obtaining the likely bias in estimators in more complicated models. Usually, the focus
is on the relationship between a particular explanatory variable, say, x1, and the key
omitted factor. Strictly speaking, ignoring all other explanatory variables is a valid prac-
tice only when each one is uncorrelated with x1, but it is still a useful guide. Appendix
3A contains a more careful analysis of omitted variable bias with multiple explanatory
variables.
3.4 The Variance of the OLS Estimators
We now obtain the variance of the OLS estimators so that, in addition to knowing the cen-
tral tendencies of the ␤ˆ
j, we also have a measure of the spread in its sampling distribution.
Before finding the variances, we add a homoskedasticity assumption, as in Chapter 2. We
do this for two reasons. First, the formulas are simplified by imposing the constant error
͚
n
iϭ1
(xi1 Ϫ x¯1)xi3
͚
n
iϭ1
(xi1 Ϫ x¯1)2
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 99

variance assumption. Second, in Section 3.5, we will see that OLS has an important effi-
ciency property if we add the homoskedasticity assumption.
In the multiple regression framework, homoskedasticity is stated as follows:
Assumption MLR.5 (Homoskedasticity)
The error u has the same variance given any values of the explanatory variables. In other words,
Var(u͉x1,…,xk) ϭ ␴2.
Assumption MLR.5 means that the variance in the error term, u, conditional on the
explanatory variables, is the same for all combinations of outcomes of the explanatory
variables. If this assumption fails, then the model exhibits heteroskedasticity, just as in the
two-variable case.
In the equation
wage ϭ ␤0 ϩ ␤1educ ϩ ␤2exper ϩ ␤3tenure ϩ u,
homoskedasticity requires that the variance of the unobserved error u does not depend on
the levels of education, experience, or tenure. That is,
Var(u͉educ, exper, tenure) ϭ ␴2.
If this variance changes with any of the three explanatory variables, then heteroskedastic-
ity is present.
Assumptions MLR.1 through MLR.5 are collectively known as the Gauss-Markov
assumptions (for cross-sectional regression). So far, our statements of the assumptions
are suitable only when applied to cross-sectional analysis with random sampling. As we
will see, the Gauss-Markov assumptions for time series analysis, and for other situa-
tions such as panel data analysis, are more difficult to state, although there are many
similarities.
In the discussion that follows, we will use the symbol x to denote the set of all inde-
pendent variables, (x1, …, xk). Thus, in the wage regression with educ, exper, and tenure
as independent variables, x ϭ (educ, exper, tenure). Then we can write Assumptions
MLR.1 and MLR.4 as
E(y͉x) ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ … ϩ ␤k xk,
and Assumption MLR.5 is the same as Var(y͉x) ϭ ␴2. Stating the assumptions in this way
clearly illustrates how Assumption MLR.5 differs greatly from Assumption MLR.4.
Assumption MLR.4 says that the expected value of y, given x, is linear in the parameters,
but it certainly depends on x1, x2, …, xk. Assumption MLR.5 says that the variance of y,
given x, does not depend on the values of the independent variables.
We can now obtain the variances of the ␤ˆ
j, where we again condition on the sample
values of the independent variables. The proof is in the appendix to this chapter.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 100

Theorem 3.2 (Sampling Variances of the OLS Slope Estimators)
Under Assumptions MLR.1 through MLR.5, conditional on the sample values of the indepen-
dent variables,
Var(␤ˆ
j) ϭ , (3.51)
for j ϭ 1, 2, …, k, where SSTj ϭ ͚
n
iϭ1 (xij Ϫ x¯j)2 is the total sample variation in xj, and R2
j is the
R-squared from regressing xj on all other independent variables (and including an intercept).
Before we study equation (3.51) in more detail, it is important to know that all of the
Gauss-Markov assumptions are used in obtaining this formula. Whereas we did not need
the homoskedasticity assumption to conclude that OLS is unbiased, we do need it to val-
idate equation (3.51).
The size of Var(␤ˆ
j) is practically important. A larger variance means a less precise
estimator, and this translates into larger confidence intervals and less accurate hypothe-
ses tests (as we will see in Chapter 4). In the next subsection, we discuss the elements
comprising (3.51).
The Components of the OLS Variances: Multicollinearity
Equation (3.51) shows that the variance of ␤ˆ
j depends on three factors: ␴2, SSTj, and Rj
2.
Remember that the index j simply denotes any one of the independent variables (such as
education or poverty rate). We now consider each of the factors affecting Var(␤ˆ
j) in turn.
THE ERROR VARIANCE, S2. From equation (3.51), a larger ␴2 means larger variances
for the OLS estimators. This is not at all surprising: more “noise” in the equation (a larger
␴2) makes it more difficult to estimate the partial effect of any of the independent variables
on y, and this is reflected in higher variances for the OLS slope estimators. Because ␴2 is a
feature of the population, it has nothing to do with the sample size. It is the one component
of (3.51) that is unknown. We will see later how to obtain an unbiased estimator of ␴2.
For a given dependent variable y, there is really only one way to reduce the error vari-
ance, and that is to add more explanatory variables to the equation (take some factors out
of the error term). Unfortunately, it is not always possible to find additional legitimate fac-
tors that affect y.
THE TOTAL SAMPLE VARIATION IN xj, SSTj. From equation (3.51), we see that
the larger the total variation in xj is, the smaller is Var(␤ˆ
j). Thus, everything else being
equal, for estimating ␤j, we prefer to have as much sample variation in xj as possible. We
already discovered this in the simple regression case in Chapter 2. Although it is rarely
␴2
SSTj(1 Ϫ Rj
2)
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 101

possible for us to choose the sample values of the independent variables, there is a way
to increase the sample variation in each of the independent variables: increase the sample
size. In fact, when sampling randomly from a population, SSTj increases without bound
as the sample size gets larger and larger. This is the component of the variance that sys-
tematically depends on the sample size.
When SSTj is small, Var(␤ˆ
j) can get very large, but a small SSTj is not a violation of
Assumption MLR.3. Technically, as SSTj goes to zero, Var(␤ˆ
j) approaches infinity. The
extreme case of no sample variation in xj, SSTj ϭ 0, is not allowed by Assumption MLR.3.
THE LINEAR RELATIONSHIPS AMONG THE INDEPENDENT VARIABLES, Rj
2. The term
Rj
2 in equation (3.51) is the most difficult of the three components to understand. This term
does not appear in simple regression analysis because there is only one independent vari-
able in such cases. It is important to see that this R-squared is distinct from the R-squared
in the regression of y on x1, x2, …, xk: Rj
2 is obtained from a regression involving only
the independent variables in the original model, where xj plays the role of a dependent
variable.
Consider first the k ϭ 2 case: y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ u. Then, Var(␤ˆ
1) ϭ ␴2/
[SST1(1 Ϫ R1
2)], where R1
2 is the R-squared from the simple regression of x1 on x2 (and an
intercept, as always). Because the R-squared measures goodness-of-fit, a value of R1
2 close
to one indicates that x2 explains much of the variation in x1 in the sample. This means that
x1 and x2 are highly correlated.
As R1
2 increases to one, Var(␤ˆ
1) gets larger and larger. Thus, a high degree of linear
relationship between x1 and x2 can lead to large variances for the OLS slope estimators.
(A similar argument applies to ␤ˆ
2.) See Figure 3.1 for the relationship between Var(␤ˆ
1)
and the R-squared from the regression of x1 on x2.
In the general case, Rj
2 is the proportion of the total variation in xj that can be explained
by the other independent variables appearing in the equation. For a given ␴2 and SSTj, the
smallest Var(␤ˆ
j) is obtained when Rj
2 ϭ 0, which happens if, and only if, xj has zero sam-
ple correlation with every other independent variable. This is the best case for estimating
␤j, but it is rarely encountered.
The other extreme case, Rj
2 ϭ 1, is ruled out by Assumption MLR.3, because
Rj
2 ϭ 1 means that, in the sample, xj is a perfect linear combination of some of the other
independent variables in the regression. A more relevant case is when Rj
2 is “close” to
one. From equation (3.51) and Figure 3.1, we see that this can cause Var(␤ˆ
j) to be large:
Var(␤ˆ
j) → ϱ as Rj
2 → 1. High (but not perfect) correlation between two or more inde-
pendent variables is called multicollinearity.
Before we discuss the multicollinearity issue further, it is important to be very clear
on one thing: a case where Rj
2 is close to one is not a violation of Assumption MLR.3.
Since multicollinearity violates none of our assumptions, the “problem” of multi-
collinearity is not really well defined. When we say that multicollinearity arises for esti-
mating ␤j when Rj
2 is “close” to one, we put “close” in quotation marks because there is
no absolute number that we can cite to conclude that multicollinearity is a problem. For
example, Rj
2 ϭ .9 means that 90 percent of the sample variation in xj can be explained by
the other independent variables in the regression model. Unquestionably, this means that
xj has a strong linear relationship to the other independent variables. But whether this trans-
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 102

lates into a Var(␤ˆ
j) that is too large to be useful depends on the sizes of ␴2 and SSTj. As
we will see in Chapter 4, for statistical inference, what ultimately matters is how big ␤ˆ
j is
in relation to its standard deviation.
Just as a large value of Rj
2 can cause a large Var(␤ˆ
j), so can a small value of SSTj. There-
fore, a small sample size can lead to large sampling variances, too. Worrying about high
degrees of correlation among the independent variables in the sample is really no different
from worrying about a small sample size: both work to increase Var(␤ˆ
j). The famous
University of Wisconsin econometrician Arthur Goldberger, reacting to econometricians’
obsession with multicollinearity, has (tongue in cheek) coined the term micronumerosity,
which he defines as the “problem of small sample size.” [For an engaging discussion of mul-
ticollinearity and micronumerosity, see Goldberger (1991).]
Although the problem of multicollinearity cannot be clearly defined, one thing is clear:
everything else being equal, for estimating ␤j, it is better to have less correlation between
xj and the other independent variables. This observation often leads to a discussion of how
Var(b1)ˆ
0
1R1
2
FIGURE 3.1
Var(␤ˆ1) as a function of R1
2.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 103

to “solve” the multicollinearity problem. In the social sciences, where we are usually
passive collectors of data, there is no good way to reduce variances of unbiased estima-
tors other than to collect more data. For a given data set, we can try dropping other inde-
pendent variables from the model in an effort to reduce multicollinearity. Unfortunately,
dropping a variable that belongs in the population model can lead to bias, as we saw in
Section 3.3.
Perhaps an example at this point will help clarify some of the issues raised concerning
multicollinearity. Suppose we are interested in estimating the effect of various school
expenditure categories on student performance. It is likely that expenditures on teacher
salaries, instructional materials, athletics, and so on, are highly correlated: wealthier
schools tend to spend more on everything, and poorer schools spend less on everything.
Not surprisingly, it can be difficult to estimate the effect of any particular expenditure cat-
egory on student performance when there is little variation in one category that cannot
largely be explained by variations in the other expenditure categories (this leads to high
Rj
2 for each of the expenditure variables). Such multicollinearity problems can be miti-
gated by collecting more data, but in a sense we have imposed the problem on ourselves:
we are asking questions that may be too subtle for the available data to answer with any
precision. We can probably do much better by changing the scope of the analysis and
lumping all expenditure categories together, since we would no longer be trying to esti-
mate the partial effect of each separate category.
Another important point is that a high degree of correlation between certain indepen-
dent variables can be irrelevant as to how well we can estimate other parameters in the
model. For example, consider a model with three independent variables:
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ ␤3x3 ϩ u,
where x2 and x3 are highly correlated. Then Var(␤ˆ
2) and Var(␤ˆ
3) may be large. But the
amount of correlation between x2 and x3 has no direct effect on Var(␤ˆ
1). In fact, if x1 is
uncorrelated with x2 and x3, then R1
2 ϭ 0
and Var(␤ˆ
1) ϭ ␴2/SST1, regardless of how
much correlation there is between x2 and
x3. If ␤1 is the parameter of interest, we do
not really care about the amount of corre-
lation between x2 and x3.
The previous observation is important
because economists often include many
control variables in order to isolate the
causal effect of a particular variable. For
example, in looking at the relationship
between loan approval rates and percent
of minorities in a neighborhood, we might
include variables like average income, average housing value, measures of creditwor-
thiness, and so on, because these factors need to be accounted for in order to draw
causal conclusions about discrimination. Income, housing prices, and creditworthiness
are generally highly correlated with each other. But high correlations among these con-
trols do not make it more difficult to determine the effects of discrimination.
Suppose you postulate a model explaining final exam score in
terms of class attendance. Thus, the dependent variable is final
exam score, and the key explanatory variable is number of classes
attended. To control for student abilities and efforts outside the
classroom, you include among the explanatory variables cumula-
tive GPA, SAT score, and measures of high school performance.
Someone says, “You cannot hope to learn anything from this
exercise because cumulative GPA, SAT score, and high school per-
formance are likely to be highly collinear.” What should be your
response?
Q U E S T I O N 3 . 4
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 104

Variances in Misspecified Models
The choice of whether or not to include a particular variable in a regression model can
be made by analyzing the tradeoff between bias and variance. In Section 3.3, we derived
the bias induced by leaving out a relevant variable when the true model contains two
explanatory variables. We continue the analysis of this model by comparing the variances
of the OLS estimators.
Write the true population model, which satisfies the Gauss-Markov assumptions, as
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ u.
We consider two estimators of ␤1. The estimator ␤ˆ1 comes from the multiple regression
yˆ ϭ ␤ˆ
0 ϩ ␤ˆ
1x1 ϩ ␤ˆ
2x2. (3.52)
In other words, we include x2, along with x1, in the regression model. The estimator ␤˜
1 is
obtained by omitting x2 from the model and running a simple regression of y on x1:
y˜ ϭ ␤˜
0 ϩ ␤˜
1x1. (3.53)
When ␤2 0, equation (3.53) excludes a relevant variable from the model and, as we saw
in Section 3.3, this induces a bias in ␤˜
1 unless x1 and x2 are uncorrelated. On the other hand,
␤ˆ
1 is unbiased for ␤1 for any value of ␤2, including ␤2 ϭ 0. It follows that, if bias is used
as the only criterion, ␤ˆ
1 is preferred to ␤˜1.
The conclusion that ␤ˆ1 is always preferred to ␤˜1 does not carry over when we bring
variance into the picture. Conditioning on the values of x1 and x2 in the sample, we have,
from (3.51),
Var(␤ˆ
1) ϭ ␴2/[SST1(1 Ϫ R1
2)], (3.54)
where SST1 is the total variation in x1, and R1
2 is the R-squared from the regression of x1
on x2. Further, a simple modification of the proof in Chapter 2 for two-variable regression
shows that
Var(␤˜
1) ϭ ␴2/SST1. (3.55)
Comparing (3.55) to (3.54) shows that Var(␤˜
1) is always smaller than Var(␤ˆ
1), unless x1
and x2 are uncorrelated in the sample, in which case the two estimators ␤˜
1 and ␤ˆ
1 are the
same. Assuming that x1 and x2 are not uncorrelated, we can draw the following conclusions:
1. When ␤2 0, ␤˜1 is biased, ␤ˆ1 is unbiased, and Var(␤˜1) Ͻ Var(␤ˆ1).
2. When ␤2 ϭ 0, ␤˜1 and ␤ˆ1 are both unbiased, and Var(␤˜1) Ͻ Var(␤ˆ1).
From the second conclusion, it is clear that ␤˜1 is preferred if ␤2 ϭ 0. Intuitively, if x2 does
not have a partial effect on y, then including it in the model can only exacerbate the mul-
ticollinearity problem, which leads to a less efficient estimator of ␤1. A higher variance
for the estimator of ␤1 is the cost of including an irrelevant variable in a model.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 105

The case where ␤2 0 is more difficult. Leaving x2 out of the model results in a biased
estimator of ␤1. Traditionally, econometricians have suggested comparing the likely size
of the bias due to omitting x2 with the reduction in the variance—summarized in the size
of R1
2—to decide whether x2 should be included. However, when ␤2 0, there are two
favorable reasons for including x2 in the model. The most important of these is that any
bias in ␤˜1 does not shrink as the sample size grows; in fact, the bias does not necessar-
ily follow any pattern. Therefore, we can usefully think of the bias as being roughly the
same for any sample size. On the other hand, Var(␤˜1) and Var(␤ˆ1) both shrink to zero as
n gets large, which means that the multicollinearity induced by adding x2 becomes less
important as the sample size grows. In large samples, we would prefer ␤ˆ1.
The other reason for favoring ␤ˆ1 is more subtle. The variance formula in (3.55) is con-
ditional on the values of xi1 and xi2 in the sample, which provides the best scenario for ␤˜1.
When ␤2 0, the variance of ␤˜1 conditional only on x1 is larger than that presented in
(3.55). Intuitively, when ␤2 0 and x2 is excluded from the model, the error variance
increases because the error effectively contains part of x2. But (3.55) ignores the error vari-
ance increase because it treats both regressors as nonrandom. A full discussion of which
independent variables to condition on would lead us too far astray. It is sufficient to say
that (3.55) is too generous when it comes to measuring the precision in ␤˜1.
Estimating s2: Standard Errors
of the OLS Estimators
We now show how to choose an unbiased estimator of ␴2, which then allows us to obtain
unbiased estimators of Var(␤ˆ
j).
Because ␴2 ϭ E(u2), an unbiased “estimator” of ␴2 is the sample average of the squared
errors: n-1
͚
n
iϭ1u2
i. Unfortunately, this is not a true estimator because we do not observe the
ui. Nevertheless, recall that the errors can be written as ui ϭ yi Ϫ ␤0 Ϫ ␤1xi1 Ϫ ␤2xi2 Ϫ . . .
Ϫ ␤kxik, and so the reason we do not observe the ui is that we do not know the ␤j. When
we replace each ␤j with its OLS estimator, we get the OLS residuals:
uˆi ϭ yi Ϫ ␤ˆ0 Ϫ ␤ˆ1xi1 Ϫ ␤ˆ2xi2 Ϫ … Ϫ ␤ˆkxik.
It seems natural to estimate ␴2 by replacing ui with the uˆi. In the simple regression case,
we saw that this leads to a biased estimator. The unbiased estimator of ␴2 in the general
multiple regression case is
␴ˆ 2 ϭ Θ͚
n
iϭ1
uˆ2
iΙ͞(n Ϫ k Ϫ 1) ϭ SSR͞(n Ϫ k Ϫ 1). (3.56)
We already encountered this estimator in the k ϭ 1 case in simple regression.
The term n Ϫ k Ϫ 1 in (3.56) is the degrees of freedom (df) for the general OLS prob-
lem with n observations and k independent variables. Since there are k ϩ 1 parameters in
a regression model with k independent variables and an intercept, we can write
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 106

df ϭ n Ϫ (k ϩ 1)
ϭ (number of observations) Ϫ (number of estimated parameters).
(3.57)
This is the easiest way to compute the degrees of freedom in a particular application: count
the number of parameters, including the intercept, and subtract this amount from the
number of observations. (In the rare case that an intercept is not estimated, the number of
parameters decreases by one.)
Technically, the division by n Ϫ k Ϫ 1 in (3.56) comes from the fact that the expected
value of the sum of squared residuals is E(SSR) ϭ (n Ϫ k Ϫ 1)␴2. Intuitively, we can
figure out why the degrees of freedom adjustment is necessary by returning to
the first order conditions for the OLS estimators. These can be written as ͚
n
iϭ1 ˆui ϭ 0 and
͚
n
iϭ1 xijuˆi ϭ 0, where j ϭ 1, 2, …, k. Thus, in obtaining the OLS estimates, k ϩ 1 restric-
tions are imposed on the OLS residuals. This means that, given n Ϫ (k ϩ 1) of the resid-
uals, the remaining k ϩ 1 residuals are known: there are only n Ϫ (k ϩ 1) degrees of free-
dom in the residuals. (This can be contrasted with the errors ui, which have n degrees of
freedom in the sample.)
For reference, we summarize this discussion with Theorem 3.3. We proved this theo-
rem for the case of simple regression analysis in Chapter 2 (see Theorem 2.3). (A general
proof that requires matrix algebra is provided in Appendix E.)
Theorem 3.3 (Unbiased Estimation of s2)
Under the Gauss-Markov Assumptions MLR.1 through MLR.5, E(␴ˆ 2) ϭ ␴2.
The positive square root of ␴ˆ 2, denoted ␴ˆ, is called the standard error of the regression
(SER). The SER is an estimator of the standard deviation of the error term. This estimate
is usually reported by regression packages, although it is called different things by differ-
ent packages. (In addition to SER, ␴ˆ is also called the standard error of the estimate and
the root mean squared error.)
Note that ␴ˆ can either decrease or increase when another independent variable is added
to a regression (for a given sample). This is because, although SSR must fall when another
explanatory variable is added, the degrees of freedom also falls by one. Because SSR is
in the numerator and df is in the denominator, we cannot tell beforehand which effect will
dominate.
For constructing confidence intervals and conducting tests in Chapter 4, we will need
to estimate the standard deviation of ␤ˆ
j, which is just the square root of the variance:
sd(␤ˆj) ϭ ␴/[SSTj(1 Ϫ Rj
2)]1/2.
Since ␴ is unknown, we replace it with its estimator, ␴ˆ . This gives us the standard
error of ␤ˆ
j:
se(␤ˆ
j) ϭ ␴ˆ /[SSTj(1 Ϫ Rj
2)]1/2. (3.58)
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 107

Just as the OLS estimates can be obtained for any given sample, so can the standard errors.
Since se(␤ˆ
j) depends on ␴ˆ , the standard error has a sampling distribution, which will play
a role in Chapter 4.
We should emphasize one thing about standard errors. Because (3.58) is obtained directly
from the variance formula in (3.51), and because (3.51) relies on the homoskedasticity
Assumption MLR.5, it follows that the standard error formula in (3.58) is not a valid esti-
mator of sd(␤ˆj) if the errors exhibit heteroskedasticity. Thus, while the presence of het-
eroskedasticity does not cause bias in the ␤ˆj, it does lead to bias in the usual formula for
Var(␤ˆj), which then invalidates the standard errors. This is important because any regression
package computes (3.58) as the default standard error for each coefficient (with a somewhat
different representation for the intercept). If we suspect heteroskedasticity, then the “usual”
OLS standard errors are invalid, and some corrective action should be taken. We will see in
Chapter 8 what methods are available for dealing with heteroskedasticity.
3.5 Efficiency of OLS:
The Gauss-Markov Theorem
In this section, we state and discuss the important Gauss-Markov Theorem, which jus-
tifies the use of the OLS method rather than using a variety of competing estimators. We
know one justification for OLS already: under Assumptions MLR.1 through MLR.4, OLS
is unbiased. However, there are many unbiased estimators of the ␤j under these assump-
tions (for example, see Problem 3.13). Might there be other unbiased estimators with vari-
ances smaller than the OLS estimators?
If we limit the class of competing estimators appropriately, then we can show that OLS
is best within this class. Specifically, we will argue that, under Assumptions MLR.1
through MLR.5, the OLS estimator ␤ˆj for ␤j is the best linear unbiased estimator
(BLUE). In order to state the theorem, we need to understand each component of the
acronym “BLUE.” First, we know what an estimator is: it is a rule that can be applied to
any sample of data to produce an estimate. We also know what an unbiased estimator is:
in the current context, an estimator, say, ␤˜j, of ␤j is an unbiased estimator of ␤j if E(␤˜j) ϭ
␤j for any ␤0, ␤1, …, ␤k.
What about the meaning of the term “linear”? In the current context, an estimator ␤˜j
of ␤j is linear if, and only if, it can be expressed as a linear function of the data on the
dependent variable:
␤˜
j ϭ ͚
n
iϭ1
wijyi, (3.59)
where each wij can be a function of the sample values of all the independent variables. The
OLS estimators are linear, as can be seen from equation (3.22).
Finally, how do we define “best”? For the current theorem, best is defined as smallest
variance. Given two unbiased estimators, it is logical to prefer the one with the smallest
variance (see Appendix C).
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 108

Now, let ␤ˆ0, ␤ˆ1, …, ␤ˆk denote the OLS estimators in model (3.31) under Assumptions
MLR.1 through MLR.5. The Gauss-Markov Theorem says that, for any estimator ␤˜j that
is linear and unbiased, Var(␤ˆj) Յ Var(␤˜j), and the inequality is usually strict. In other
words, in the class of linear unbiased estimators, OLS has the smallest variance (under
the five Gauss-Markov assumptions). Actually, the theorem says more than this. If we
want to estimate any linear function of the ␤j, then the corresponding linear combination
of the OLS estimators achieves the smallest variance among all linear unbiased estima-
tors. We conclude with a theorem, which is proven in Appendix 3A.
Theorem 3.4 (Gauss-Markov Theorem)
Under Assumptions MLR.1 through MLR.5, ␤ˆ
0, ␤ˆ
1, …, ␤ˆ
k are the best linear unbiased estima-
tors (BLUEs) of ␤0, ␤1, …, ␤k, respectively.
It is because of this theorem that Assumptions MLR.1 through MLR.5 are known as the
Gauss-Markov assumptions (for cross-sectional analysis).
The importance of the Gauss-Markov Theorem is that, when the standard set of
assumptions holds, we need not look for alternative unbiased estimators of the form
in (3.59): none will be better than OLS. Equivalently, if we are presented with an
estimator that is both linear and unbiased, then we know that the variance of this estimator
is at least as large as the OLS variance; no additional calculation is needed to show this.
For our purposes, Theorem 3.4 justifies the use of OLS to estimate multiple regression
models. If any of the Gauss-Markov assumptions fail, then this theorem no longer holds.
We already know that failure of the zero conditional mean assumption (Assumption
MLR.4) causes OLS to be biased, so Theorem 3.4 also fails. We also know that het-
eroskedasticity (failure of Assumption MLR.5) does not cause OLS to be biased. However,
OLS no longer has the smallest variance among linear unbiased estimators in the presence
of heteroskedasticity. In Chapter 8, we analyze an estimator that improves upon OLS when
we know the brand of heteroskedasticity.
S U M M A R Y
1. The multiple regression model allows us to effectively hold other factors fixed while
examining the effects of a particular independent variable on the dependent variable. It
explicitly allows the independent variables to be correlated.
2. Although the model is linear in its parameters, it can be used to model nonlinear rela-
tionships by appropriately choosing the dependent and independent variables.
3. The method of ordinary least squares is easily applied to estimate the multiple
regression model. Each slope estimate measures the partial effect of the corresponding
independent variable on the dependent variable, holding all other independent variables
fixed.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 109

4. R2 is the proportion of the sample variation in the dependent variable explained by the
independent variables, and it serves as a goodness-of-fit measure. It is important not to
put too much weight on the value of R2 when evaluating econometric models.
5. Under the first four Gauss-Markov assumptions (MLR.1 through MLR.4), the OLS
estimators are unbiased. This implies that including an irrelevant variable in a model has
no effect on the unbiasedness of the intercept and other slope estimators. On the other
hand, omitting a relevant variable causes OLS to be biased. In many circumstances, the
direction of the bias can be determined.
6. Under the five Gauss-Markov assumptions, the variance of an OLS slope estimator is
given by Var(␤ˆ
j) ϭ ␴2/[SSTj(1 Ϫ Rj
2)]. As the error variance ␴2 increases, so does Var(␤ˆ
j),
while Var(␤ˆ
j) decreases as the sample variation in xj, SSTj, increases. The term Rj
2 mea-
sures the amount of collinearity between xj and the other explanatory variables. As Rj
2
approaches one, Var(␤ˆ
j) is unbounded.
7. Adding an irrelevant variable to an equation generally increases the variances of the
remaining OLS estimators because of multicollinearity.
8. Under the Gauss-Markov assumptions (MLR.1 through MLR.5), the OLS estimators
are best linear unbiased estimators (BLUEs).
The Gauss-Markov Assumptions
The following is a summary of the five Gauss-Markov assumptions that we used in this
chapter. Remember, the first four were used to establish unbiasedness of OLS, while the
fifth was added to derive the usual variance formulas and to conclude that OLS is best
linear unbiased.
Assumption MLR.1 (Linear in Parameters)
The model in the population can be written as
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ … ϩ ␤kxk ϩ u,
where ␤0, ␤1, …, ␤k are the unknown parameters (constants) of interest and u is an unob-
servable random error or disturbance term.
Assumption MLR.2 (Random Sampling)
We have a random sample of n observations, {(xi1,xi2, . . . , xik,yi): i ϭ 1,2, . . . , n}, fol-
lowing the population model in Assumption MLR.1.
Assumption MLR.3 (No Perfect Collinearity)
In the sample (and therefore in the population), none of the independent variables is
constant, and there are no exact linear relationships among the independent variables.
Assumption MLR.4 (Zero Conditional Mean)
The error u has an expected value of zero given any values of the independent variables.
In other words,
E(u|x1,x2,…,xk) ϭ 0.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 110

Assumption MLR.5 (Homoskedasticity)
The error u has the same variance given any value of the explanatory variables. In other
words,
Var(u|x1, . . . , xk) ϭ s2.
K E Y T E R M S
Best Linear Unbiased
Estimator (BLUE)
Biased Towards Zero
Ceteris Paribus
Degrees of Freedom (df )
Disturbance
Downward Bias
Endogenous Explanatory
Variable
Error Term
Excluding a Relevant
Variable
Exogenous Explanatory
Variable
Explained Sum of
Squares (SSE)
First Order Conditions
Gauss-Markov
Assumptions
Gauss-Markov Theorem
Inclusion of an Irrelevant
Variable
Intercept
Micronumerosity
Misspecification Analysis
Multicollinearity
Multiple Linear Regression
Model
Multiple Regression
Analysis
OLS Intercept Estimate
OLS Regression Line
OLS Slope Estimate
Omitted Variable Bias
Ordinary Least Squares
Overspecifying the Model
Partial Effect
Perfect Collinearity
Population Model
Residual
Residual Sum of Squares
Sample Regression
Function (SRF)
Slope Parameter
Standard Deviation of ␤ˆ
j
Standard Error of ␤ˆ
j
Standard Error of the
Regression (SER)
Sum of Squared Residuals
(SSR)
Total Sum of Squares
(SST)
True Model
Underspecifying the
Model
Upward Bias
P R O B L E M S
3.1 Using the data in GPA2.RAW on 4,137 college students, the following equation was
estimated by OLS:
colgpa ϭ 1.392 Ϫ .0135 hsperc ϩ .00148 sat
n ϭ 4,137, R2 ϭ .273,
where colgpa is measured on a four-point scale, hsperc is the percentile in the high school
graduating class (defined so that, for example, hsperc ϭ 5 means the top five percent of
the class), and sat is the combined math and verbal scores on the student achievement test.
(i) Why does it make sense for the coefficient on hsperc to be negative?
(ii) What is the predicted college GPA when hsperc ϭ 20 and sat ϭ 1050?
(iii) Suppose that two high school graduates, A and B, graduated in the same
percentile from high school, but Student A’s SAT score was 140 points
higher (about one standard deviation in the sample). What is the predicted
difference in college GPA for these two students? Is the difference large?
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 111

(iv) Holding hsperc fixed, what difference in SAT scores leads to a predicted
colgpa difference of .50, or one-half of a grade point? Comment on your
answer.
3.2 The data in WAGE2.RAW on working men was used to estimate the following
equation:
educ ϭ 10.36 Ϫ .094 sibs ϩ .131 meduc ϩ .210 feduc
n ϭ 722, R2 ϭ .214,
where educ is years of schooling, sibs is number of siblings, meduc is mother’s years of
schooling, and feduc is father’s years of schooling.
(i) Does sibs have the expected effect? Explain. Holding meduc and feduc
fixed, by how much does sibs have to increase to reduce predicted years
of education by one year? (A noninteger answer is acceptable here.)
(ii) Discuss the interpretation of the coefficient on meduc.
(iii) Suppose that Man A has no siblings, and his mother and father each have
12 years of education. Man B has no siblings, and his mother and father
each have 16 years of education. What is the predicted difference in years
of education between B and A?
3.3 The following model is a simplified version of the multiple regression model used
by Biddle and Hamermesh (1990) to study the tradeoff between time spent sleeping and
working and to look at other factors affecting sleep:
sleep ϭ ␤0 ϩ ␤1totwrk ϩ ␤2educ ϩ ␤3age ϩ u,
where sleep and totwrk (total work) are measured in minutes per week and educ and age
are measured in years. (See also Computer Exercise C2.3.)
(i) If adults trade off sleep for work, what is the sign of ␤1?
(ii) What signs do you think ␤2 and ␤3 will have?
(iii) Using the data in SLEEP75.RAW, the estimated equation is
sleep ϭ 3638.25 Ϫ .148 totwrk Ϫ 11.13 educ ϩ 2.20 age
n ϭ 706, R2 ϭ .113.
If someone works five more hours per week, by how many minutes is sleep
predicted to fall? Is this a large tradeoff?
(iv) Discuss the sign and magnitude of the estimated coefficient on educ.
(v) Would you say totwrk, educ, and age explain much of the variation in
sleep? What other factors might affect the time spent sleeping? Are these
likely to be correlated with totwrk?
3.4 The median starting salary for new law school graduates is determined by
log(salary) ϭ ␤0 ϩ ␤1LSAT ϩ ␤2GPA ϩ ␤3log(libvol) ϩ ␤4log(cost)
ϩ ␤5rank ϩ u,
where LSAT is the median LSAT score for the graduating class, GPA is the median col-
lege GPA for the class, libvol is the number of volumes in the law school library, cost is
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 112

the annual cost of attending law school, and rank is a law school ranking (with rank ϭ 1
being the best).
(i) Explain why we expect ␤5 Յ 0.
(ii) What signs do you expect for the other slope parameters? Justify your
answers.
(iii) Using the data in LAWSCH85.RAW, the estimated equation is
log(salary) ϭ 8.34 ϩ .0047 LSAT ϩ .248 GPA ϩ .095 log(libvol)
ϩ .038 log(cost) Ϫ .0033 rank
n ϭ 136, R2 ϭ .842.
What is the predicted ceteris paribus difference in salary for schools with a
median GPA different by one point? (Report your answer as a percentage.)
(iv) Interpret the coefficient on the variable log(libvol).
(v) Would you say it is better to attend a higher ranked law school?
How much is a difference in ranking of 20 worth in terms of predicted
starting salary?
3.5 In a study relating college grade point average to time spent in various activities, you
distribute a survey to several students. The students are asked how many hours they spend
each week in four activities: studying, sleeping, working, and leisure. Any activity is put
into one of the four categories, so that for each student, the sum of hours in the four
activities must be 168.
(i) In the model
GPA ϭ ␤0 ϩ ␤1study ϩ ␤2sleep ϩ ␤3work ϩ ␤4leisure ϩ u,
does it make sense to hold sleep, work, and leisure fixed, while changing
study?
(ii) Explain why this model violates Assumption MLR.3.
(iii) How could you reformulate the model so that its parameters have a useful
interpretation and it satisfies Assumption MLR.3?
3.6 Consider the multiple regression model containing three independent variables, under
Assumptions MLR.1 through MLR.4:
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ ␤3x3 ϩ u.
You are interested in estimating the sum of the parameters on x1 and x2; call this ␪1 ϭ ␤1
ϩ ␤2.
(i) Show that ␪ˆ
1 ϭ ␤ˆ
1 ϩ ␤ˆ
2 is an unbiased estimator of ␪1.
(ii) Find Var(␪ˆ
1) in terms of Var(␤ˆ
1), Var(␤ˆ
2), and Corr(␤ˆ
1,␤ˆ
2).
3.7 Which of the following can cause OLS estimators to be biased?
(i) Heteroskedasticity.
(ii) Omitting an important variable.
(iii) A sample correlation coefficient of .95 between two independent variables
both included in the model.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 113

3.8 Suppose that average worker productivity at manufacturing firms (avgprod)
depends on two factors, average hours of training (avgtrain) and average worker
ability (avgabil):
avgprod ϭ ␤0 ϩ ␤1avgtrain ϩ ␤2avgabil ϩ u.
Assume that this equation satisfies the Gauss-Markov assumptions. If grants have been
given to firms whose workers have less than average ability, so that avgtrain and avgabil
are negatively correlated, what is the likely bias in ␤˜
1 obtained from the simple regression
of avgprod on avgtrain?
3.9 The following equation describes the median housing price in a community in terms
of amount of pollution (nox for nitrous oxide) and the average number of rooms in houses
in the community (rooms):
log(price) ϭ ␤0 ϩ ␤1log(nox) ϩ ␤2rooms ϩ u.
(i) What are the probable signs of ␤1 and ␤2? What is the interpretation of
␤1? Explain.
(ii) Why might nox [or more precisely, log(nox)] and rooms be negatively cor-
related? If this is the case, does the simple regression of log(price) on
log(nox) produce an upward or a downward biased estimator of ␤1?
(iii) Using the data in HPRICE2.RAW, the following equations were estimated:
log(price) ϭ 11.71 Ϫ 1.043 log(nox), n ϭ 506, R2 ϭ .264.
log(price) ϭ 9.23 Ϫ .718 log(nox) ϩ .306 rooms, n ϭ 506, R2 ϭ .514.
Is the relationship between the simple and multiple regression estimates of the
elasticity of price with respect to nox what you would have predicted, given
your answer in part (ii)? Does this mean that Ϫ.718 is definitely closer to the
true elasticity than Ϫ1.043?
3.10 Suppose that you are interested in estimating the ceteris paribus relationship
between y and x1. For this purpose, you can collect data on two control variables, x2 and
x3. (For concreteness, you might think of y as final exam score, x1 as class attendance, x2
as GPA up through the previous semester, and x3 as SAT or ACT score.) Let ␤˜
1 be the
simple regression estimate from y on x1 and let ␤ˆ1 be the multiple regression estimate
from y on x1,x2,x3.
(i) If x1 is highly correlated with x2 and x3 in the sample, and x2 and x3 have
large partial effects on y, would you expect ␤˜1 and ␤ˆ1 to be similar or very
different? Explain.
(ii) If x1 is almost uncorrelated with x2 and x3, but x2 and x3 are highly cor-
related, will ␤˜1 and ␤ˆ1 tend to be similar or very different? Explain.
(iii) If x1 is highly correlated with x2 and x3, and x2 and x3 have small par-
tial effects on y, would you expect se(␤˜1) or se(␤ˆ1) to be smaller?
Explain.
(iv) If x1 is almost uncorrelated with x2 and x3, x2 and x3 have large partial
effects on y, and x2 and x3 are highly correlated, would you expect se(␤˜
1)
or se(␤ˆ
1) to be smaller? Explain.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 114

3.11 Suppose that the population model determining y is
y ϭ ␤0 ϩ ␤1x1 ϩ ␤2x2 ϩ ␤3x3 ϩ u,
and this model satisifies Assumptions MLR.1 through MLR.4. However, we estimate the
model that omits x3. Let ␤˜
0, ␤˜
1, and ␤˜
2 be the OLS estimators from the regression of y on
x1 and x2. Show that the expected value of ␤˜
1 (given the values of the independent vari-
ables in the sample) is
E(␤˜
1) ϭ ␤1 ϩ ␤3 ,
where the rî1 are the OLS residuals from the regression of x1 on x2. [Hint: The formula for
␤˜
1 comes from equation (3.22). Plug yi ϭ ␤0 ϩ ␤1xi1 ϩ ␤2xi2 ϩ ␤3xi3 ϩ ui into this equa-
tion. After some algebra, take the expectation treating xi3 and rî1 as nonrandom.]
3.12 The following equation represents the effects of tax revenue mix on subsequent
employment growth for the population of counties in the United States:
growth ϭ ␤0 ϩ ␤1shareP ϩ ␤2shareI ϩ ␤3shareS ϩ other factors,
where growth is the percentage change in employment from 1980 to 1990, shareP is the
share of property taxes in total tax revenue, shareI is the share of income tax revenues,
and shareS is the share of sales tax revenues. All of these variables are measured in 1980.
The omitted share, shareF, includes fees and miscellaneous taxes. By definition, the four
shares add up to one. Other factors would include expenditures on education, infrastruc-
ture, and so on (all measured in 1980).
(i) Why must we omit one of the tax share variables from the equation?
(ii) Give a careful interpretation of ␤1.
3.13 (i) Consider the simple regression model y ϭ ␤0 ϩ ␤1x ϩ u under the first
four Gauss-Markov assumptions. For some function g(x), for example g(x)
ϭ x2 or g(x) ϭ log(1 ϩ x2), define zi ϭ g(xi). Define a slope estimator as
␤˜
1 ϭ Θ͚
n
iϭ1
(zi Ϫ z¯)yiΙ͞Θ͚
n
iϭ1
(zi Ϫ z¯)xiΙ.
Show that ␤˜
1 is linear and unbiased. Remember, because E(u͉x) ϭ 0, you
can treat both xi and zi as nonrandom in your derivation.
(ii) Add the homoskedasticity assumption, MLR.5. Show that
Var(␤˜1) ϭ ␴2
Θ͚
n
iϭ1
(zi Ϫ z¯)2
Ι͞Θ͚
n
iϭ1
(zi Ϫ z¯)xiΙ
2
.
(iii) Show directly that, under the Gauss-Markov assumptions, Var(␤ˆ
1) Յ
Var(␤˜
1), where ␤ˆ
1 is the OLS estimator. [Hint: The Cauchy-Schwartz
inequality in Appendix B implies that
͚
n
iϭ1
rî1xi3
͚
n
iϭ1
rˆ2
i1
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 115

ΘnϪ1
͚
n
iϭ1
(zi Ϫ z¯)(xi Ϫ x¯)Ι
2
Յ ΘnϪ1
͚
n
iϭ1
(zi Ϫ z¯)2
ΙΘnϪ1
͚
n
iϭ1
(xi Ϫ x¯)2
Ι;
notice that we can drop x¯ from the sample covariance.]
C O M P U T E R E X E R C I S E S
C3.1 A problem of interest to health officials (and others) is to determine the effects of
smoking during pregnancy on infant health. One measure of infant health is birth weight;
a birth weight that is too low can put an infant at risk for contracting various illnesses.
Since factors other than cigarette smoking that affect birth weight are likely to be corre-
lated with smoking, we should take those factors into account. For example, higher income
generally results in access to better prenatal care, as well as better nutrition for the mother.
An equation that recognizes this is
bwght ϭ ␤0 ϩ ␤1cigs ϩ ␤2 faminc ϩ u.
(i) What is the most likely sign for ␤2?
(ii) Do you think cigs and faminc are likely to be correlated? Explain why
the correlation might be positive or negative.
(iii) Now, estimate the equation with and without faminc, using the data in
BWGHT.RAW. Report the results in equation form, including the sam-
ple size and R-squared. Discuss your results, focusing on whether adding
faminc substantially changes the estimated effect of cigs on bwght.
C3.2 Use the data in HPRICE1.RAW to estimate the model
price ϭ ␤0 ϩ ␤1sqrft ϩ ␤2bdrms ϩ u,
where price is the house price measured in thousands of dollars.
(i) Write out the results in equation form.
(ii) What is the estimated increase in price for a house with one more bed-
room, holding square footage constant?
(iii) What is the estimated increase in price for a house with an additional
bedroom that is 140 square feet in size? Compare this to your answer in
part (ii).
(iv) What percentage of the variation in price is explained by square footage
and number of bedrooms?
(v) The first house in the sample has sqrft ϭ 2,438 and bdrms ϭ 4. Find the
predicted selling price for this house from the OLS regression line.
(vi) The actual selling price of the first house in the sample was $300,000
(so price ϭ 300). Find the residual for this house. Does it suggest that
the buyer underpaid or overpaid for the house?
C3.3 The file CEOSAL2.RAW contains data on 177 chief executive officers and can be
used to examine the effects of firm performance on CEO salary.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 116

(i) Estimate a model relating annual salary to firm sales and market value.
Make the model of the constant elasticity variety for both independent
variables. Write the results out in equation form.
(ii) Add profits to the model from part (i). Why can this variable not be
included in logarithmic form? Would you say that these firm perfor-
mance variables explain most of the variation in CEO salaries?
(iii) Add the variable ceoten to the model in part (ii). What is the estimated per-
centage return for another year of CEO tenure, holding other factors fixed?
(iv) Find the sample correlation coefficient between the variables log(mktval)
and profits. Are these variables highly correlated? What does this say
about the OLS estimators?
C3.4 Use the data in ATTEND.RAW for this exercise.
(i) Obtain the minimum, maximum, and average values for the variables
atndrte, priGPA, and ACT.
(ii) Estimate the model
atndrte ϭ ␤0 ϩ ␤1priGPA ϩ ␤2ACT ϩ u,
and write the results in equation form. Interpret the intercept. Does it
have a useful meaning?
(iii) Discuss the estimated slope coefficients. Are there any surprises?
(iv) What is the predicted atndrte if priGPA ϭ 3.65 and ACT ϭ 20? What
do you make of this result? Are there any students in the sample with
these values of the explanatory variables?
(v) If Student A has priGPA ϭ 3.1 and ACT ϭ 21 and Student B has priGPA
ϭ 2.1 and ACT ϭ 26, what is the predicted difference in their attendance
rates?
C3.5 Confirm the partialling out interpretation of the OLS estimates by explicitly doing
the partialling out for Example 3.2. This first requires regressing educ on exper and tenure
and saving the residuals, rˆ1. Then, regress log(wage) on rˆ1. Compare the coefficient on rˆ1
with the coefficient on educ in the regression of log(wage) on educ, exper, and tenure.
C3.6 Use the data set in WAGE2.RAW for this problem. As usual, be sure all of the
following regressions contain an intercept.
(i) Run a simple regression of IQ on educ to obtain the slope coefficient,
say, ␦˜
1.
(ii) Run the simple regression of log(wage) on educ, and obtain the slope
coefficient, ␤˜
1.
(iii) Run the multiple regression of log(wage) on educ and IQ, and obtain the
slope coefficients, ␤ˆ
1 and ␤ˆ
2, respectively.
(iv) Verify that ␤˜
1 ϭ ␤ˆ
1 ϩ ␤ˆ
2␦˜
1
C3.7 Use the data in MEAP93.RAW to answer this question.
(i) Estimate the model
math10 ϭ ␤0 ϩ ␤1 log(expend) ϩ ␤2lnchprg ϩ u,
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 117

and report the equation in the usual form, including the sample size and
R-squared. Are the signs of the slope coefficients what you expected?
Explain.
(ii) What do you make of the intercept you estimated in part (i)? In partic-
ular, does it make sense to set the two explanatory variables to zero?
[Hint: Recall that log(1)ϭ0.]
(iii) Now run the simple regression of math10 on log(expend), and compare
the slope coefficient with the estimate obtained in part (i). Is the esti-
mated spending effect now larger or smaller than in part (i)?
(iv) Find the correlation between lexpend ϭ log(expend) and lnchprg. Does
its sign make sense to you?
(v) Use part (iv) to explain your findings in part (iii).
C3.8 Use the data in DISCRIM.RAW to answer this question. These are zip
code–level data on prices for various items at fast-food restaurants, along with charac-
teristics of the zip code population, in New Jersey and Pennsylvania. The idea is to see
whether fast-food restaurants charge higher prices in areas with a larger concentration
of blacks.
(i) Find the average values of prpblck and income in the sample, along with
their standard deviations. What are the units of measurement of prpblck
and income?
(ii) Consider a model to explain the price of soda, psoda, in terms of the
proportion of the population that is black and median income:
psoda ϭ ␤0 ϩ ␤1prpblck ϩ ␤2income ϩ u.
Estimate this model by OLS and report the results in equation form,
including the sample size and R-squared. (Do not use scientific notation
when reporting the estimates.) Interpret the coefficient on prpblck. Do
you think it is economically large?
(iii) Compare the estimate from part (ii) with the simple regression estimate
from psoda on prpblck. Is the discrimination effect larger or smaller
when you control for income?
(iv) A model with a constant price elasticity with respect to income may be
more appropriate. Report estimates of the model
log(psoda) ϭ ␤0 ϩ ␤1prpblck ϩ ␤2income ϩ u.
If prpblck increases by .20 (20 percentage points), what is the estimated
percentage change in psoda? (Hint: The answer is 2.xx, where you fill
in the “xx.”)
(v) Now add the variable prppov to the regression in part (iv). What happens
to ␤ˆ
prpblck?
(vi) Find the correlation between log(income) and prppov. Is it roughly what
you expected?
(vii) Evaluate the following statement: “Because log(income) and prppov
are so highly correlated, they have no business being in the same
regression.”
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 118

A P P E N D I X 3 A
3A.1 Derivation of the First Order Conditions in Equation (3.13)
The analysis is very similar to the simple regression case. We must characterize the solu-
tions to the problem
͚
n
iϭ1
(yi Ϫ b0 Ϫ b1xi1 Ϫ … Ϫ bkxik)2.
Taking the partial derivatives with respect to each of the bj (see Appendix A), evaluating
them at the solutions, and setting them equal to zero gives
Ϫ2 ͚
n
iϭ1
(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1Ϫ … Ϫ ␤ˆ
kxik) ϭ 0
Ϫ2 ͚
n
iϭ1
xij(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1 Ϫ … Ϫ ␤ˆ
kxik) ϭ 0, for all j ϭ 1, …, k.
Canceling the Ϫ2 gives the first order conditions in (3.13).
3A.2 Derivation of Equation (3.22)
To derive (3.22), write xi1 in terms of its fitted value and its residual from the regression
of x1 on x2, …, xk: xi1 ϭ xî1 ϩ rî1, for all i ϭ 1, …, n. Now, plug this into the second equa-
tion in (3.13):
͚
n
iϭ1
(xî1 ϩ rî1)(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1 Ϫ … Ϫ ␤ˆ
kxik) ϭ 0. (3.60)
By the definition of the OLS residual uî, since xî1 is just a linear function of the explana-
tory variables xi2, …, xik, it follows that ͚
n
iϭ1 xî1uî ϭ 0. Therefore, equation (3.60) can be
expressed as
͚
n
iϭ1
rî1(yi Ϫ ␤ˆ
0 Ϫ ␤ˆ
1xi1 Ϫ … Ϫ ␤ˆ
kxik) ϭ 0. (3.61)
Since the rî1 are the residuals from regressing x1 on x2, …, xk, ͚
n
iϭ1 xijrî1 ϭ 0, for all j ϭ
2, …, k. Therefore, (3.61) is equivalent to ͚
n
iϭ1 rî1(yi Ϫ ␤ˆ
1xi1) ϭ 0. Finally, we use the
fact that ͚
n
iϭ1 xî1rî1 ϭ 0, which means that ␤ˆ
1 solves
͚
n
iϭ1
rî1(yi Ϫ ␤ˆ
1rî1) ϭ 0.
Now, straightforward algebra gives (3.22), provided, of course, that ͚
n
iϭ1 rî
2
1 Ͼ 0; this is
ensured by Assumption MLR.3.
min
b0, b1, …, bk
89782_03_c03_p073-122.qxd 6/14/05 5:24 PM Page 119

3A.3 Proof of Theorem 3.1
We prove Theorem 3.1 for ␤ˆ
1; the proof for the other slope parameters is virtually identi-
cal. (See Appendix E for a more succinct proof using matrices.) Under Assumption
MLR.3, the OLS estimators exist, and we can write ␤ˆ1 as in (3.22). Under Assumption
MLR.1, we can write yi as in (3.32); substitute this for yi in (3.22). Then, using
͚
n
iϭ1 rî1 ϭ 0, ͚
n
iϭ1 xijrî1 ϭ 0, for all j ϭ 2, …, k, and ͚
n
iϭ1 xi1rî1 ϭ ͚
n
iϭ1 rî
2
1, we have
␤ˆ
1 ϭ ␤1 ϩ Θ͚
n
iϭ1
rî1uiΙ͞Θ͚
n
iϭ1
rî
2
1Ι. (3.62)
Now, under Assumptions MLR.2 and MLR.4, the expected value of each ui, given all inde-
pendent variables in the sample, is zero. Since the rî1 are just functions of the sample inde-
pendent variables, it follows that
E(␤ˆ
1͉X) ϭ ␤1 ϩ Θ͚
n
iϭ1
rî1E(ui͉X)Ι͞Θ͚
n
iϭ1
rî
2
1Ι
ϭ ␤1 ϩ Θ͚
n
iϭ1
rî1и0Ι͞Θ͚
n
iϭ1
rî
2
1Ιϭ ␤1,
where X denotes the data on all independent variables and E(␤ˆ
1͉X) is the expected value
of ␤ˆ
1, given xi1, …, xik, for all i ϭ 1, …, n. This completes the proof.
3A.4 General Omitted Variable Bias
We can derive the omitted variable bias in the general model in equation (3.31) under the
first four Gauss-Markov assumptions. In particular, let the ␤ˆ
j, j ϭ 0, 1, ..., k be the OLS esti-
mators from the regression using the full set of explanatory variables. Let the ␤˜
j, j ϭ 0,1,
..., k Ϫ 1 be the OLS estimators from the regression that leaves out xk. Let ␦˜
j, j ϭ 1, ...,
k Ϫ 1 be the slope coefficient on xj in the auxiliary regression of xik on xi1, xi2, ... xi,kϪ1, i ϭ
1, ..., n. A useful fact is that
␤˜
j ϭ ␤ˆ
j ϩ ␤ˆ
k ␦˜
j. (3.63)
This shows explicitly that, when we do not control for xk in the regression, the estimated
partial effect of xj equals the partial effect when we include xk plus the partial effect of xk
on yˆ times the partial relationship between the omitted variable, xk, and xj, j Ͻ k. Condi-
tional on the entire set of explanatory variables, X, we know that the ␤ˆ
j are all unbiased
for the corresponding ␤j, j ϭ 1, ..., k. Further, since ␦˜
j is just a function of X, we have
E(␤˜
j|X) ϭ E(␤ˆ
j|X) ϩ E(␤ˆ
k|X)␦˜
j
ϭ ␤j ϩ ␤k␦˜
j.
(3.64)
Equation (3.64) shows that ␤˜
j is biased for ␤j unless ␤k ϭ 0—in which case xk has no
partial effect in the population—or ␦˜
j equals zero, which means that xik and xij are
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 120

partially uncorrelated in the sample. The key to obtaining equation (3.64) is equation
(3.63). To show equation (3.63), we can use equation (3.22) a couple of times. For sim-
plicity, we look at j ϭ 1. Now, ␤˜
1 is the slope coefficient in the simple regression of yi
on ˜ri1, i ϭ 1 ..., n, where the ˜ri1 are the OLS residuals from the regression of xi1 on xi2,
xi3, ..., xi,kϪ1. Consider the numerator of the expression for ␤˜
1: ͚
n
iϭ1 ˜ri1yi. But for each i,we
can write yi ϭ ␤ˆ
0 ϩ ␤ˆ
1xi1 ϩ ... ϩ ␤ˆ
kxik ϩ uî and plug in for yi. Now, by properties of the
OLS residuals, the ˜ri1 have zero sample average and are uncorrelated with xi2, xi3, ..., xi,kϪ1
in the sample. Similarly, the uî have zero sample average and zero sample correlation with
xi1, xi2, ..., xik. It follows that the ˜ri1 and uî are uncorrelated in the sample (since the ˜ri1 are
just linear combinations of xi1, xi2, ..., xi,kϪ1). So
͚
n
iϭ1
˜ri1yi ϭ ␤ˆ
1 ΂͚
n
iϭ1
˜ri1xi1΃ϩ ␤ˆ
k ΂͚
n
iϭ1
˜ri1xik΃. (3.65)
Now, ͚
n
iϭ1 ˜ri1xi1 ϭ ͚
n
iϭ1 ˜ri1
2, which is also the denominator of ␤˜
1. Therefore, we have
shown that
␤˜
1 ϭ ␤ˆ
1 ϩ ␤ˆ
k ΂͚
n
iϭ1
˜ri1xik΃͞΂͚
n
iϭ1
˜ri1
2,
΃
ϭ ␤ˆ
1 + ␤ˆ
k ␦˜
1.
This is the relationship we wanted to show.
Again, we prove this for j ϭ 1. Write ␤ˆ
1 as in equation (3.62). Now, under MLR.5,
Var(ui͉X) ϭ ␴2, for all i ϭ 1, …, n. Under random sampling, the ui are independent, even
conditional on X, and the rî1 are nonrandom conditional on X. Therefore,
Var(␤ˆ
1͉X) ϭ Θ͚
n
iϭ1
rî
2
1 Var(ui͉X)Ι͞Θ͚
n
iϭ1
rî
2
1Ι
2
ϭ Θ͚
n
iϭ1
rî
2
1␴2
Ι͞Θ͚
n
iϭ1
rî
2
1Ι
2
ϭ ␴2
͞Θ͚
n
iϭ1
rî
2
1Ι.
Now, since ͚
n
iϭ1 rî
2
1 is the sum of squared residuals from regressing x1 on x2, …, xk,
͚
n
iϭ1 rî
2
1 ϭ SST1(1 Ϫ R1
2). This completes the proof.
We show that, for any other linear unbiased estimator ␤˜
1 of ␤1, Var(␤˜
1) Ն Var(␤ˆ
1), where
␤ˆ
1 is the OLS estimator. The focus on j ϭ 1 is without loss of generality.
For ␤˜
1 as in equation (3.59), we can plug in for yi to obtain
␤˜
1 ϭ ␤0 ͚
n
iϭ1
wi1 ϩ ␤1 ͚
n
iϭ1
wi1xi1 ϩ ␤2 ͚
n
iϭ1
wi1xi2 ϩ … ϩ ␤k ͚
n
iϭ1
wi1xik ϩ ͚
n
iϭ1
wi1ui.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 121

Now, since the wi1 are functions of the xij,
E(␤˜
1͉X) ϭ ␤0 ͚
n
iϭ1
wi1 ϩ ␤1 ͚
n
iϭ1
wi1xi1 ϩ ␤2 ͚
n
iϭ1
n
iϭ1
wi1xik ϩ ͚
n
iϭ1
wi1E(ui͉X)
ϭ ␤0 ͚
n
iϭ1
wi1 ϩ ␤1 ͚
n
iϭ1
wi1xi1 ϩ ␤2 ͚
n
iϭ1
n
iϭ1
wi1xik
because E(ui͉X) ϭ 0, for all i ϭ 1, …, n under MLR.2 and MLR.4. Therefore, for E(␤˜
1͉X)
to equal ␤1 for any values of the parameters, we must have
͚
n
iϭ1
wi1 ϭ 0, ͚
n
iϭ1
wi1xi1 ϭ 1, ͚
n
iϭ1
wi1xij ϭ 0, j ϭ 2, …, k. (3.66)
Now, let rî1 be the residuals from the regression of xi1 on xi2, …, xik. Then, from (3.66), it
follows that
͚
n
iϭ1
wi1rî1 ϭ 1 (3.67)
because xi1 ϭ ˆxi1 ϩ rî1 and ͚
n
iϭ1 wi1
ˆxi1 ϭ 0. Now, consider the difference between
Var(␤˜
1͉X) and Var(␤ˆ
1͉X) under MLR.1 through MLR.5:
␴2
͚
n
iϭ1
wi
2
1 Ϫ ␴2
͞Θ͚
n
iϭ1
rî
2
1Ι. (3.68)
Because of (3.67), we can write the difference in (3.68), without ␴2, as
͚
n
iϭ1
wi
2
1 Ϫ Θ͚
n
iϭ1
wi1rî1Ι
2
͞Θ͚
n
iϭ1
rî
2
1Ι. (3.69)
But (3.69) is simply
͚
n
iϭ1
(wi1 Ϫ ␥ˆ1rî1)2, (3.70)
where ␥ˆ1 ϭ Θ͚
n
iϭ1 wi1rî1Ι͞Θ͚
n
iϭ1 rî
2
1Ι, as can be seen by squaring each term in (3.70),
summing, and then canceling terms. Because (3.70) is just the sum of squared residu-
als from the simple regression of wi1 on rî1—remember that the sample average of rî1 is
zero—(3.70) must be nonnegative. This completes the proof.
89782_03_c03_p073-122.qxd 5/26/05 11:46 AM Page 122

Biometry regression

More Related Content

What's hot (19)

Similar to Biometry regression (20)

More from musadoto (20)

Recently uploaded (20)

Biometry regression