THE THREE-VARIABLE MODEL:

NOTATION AND ASSUMPTIONS

Generalizing the two-variable population
regression function (PRF), we may write
the three-variable PRF as
where Y is the dependent variable, X2
and X3 the explanatory variables
(orregressors), u the stochastic
disturbance term, and i the ith
observation; incase the data are time
series, the subscript t will denote the tth
observation.In Eqβ1 is the intercept term.
As usual, it gives the mean or average
effect on Y of all the variables excluded
from the model, although its mechanical
interpretation is the average value of Y
when X2 and X3 are set equal to zero.
The coefficients β2 and β3 are called the
partial regression coefficients, and their
meaning will be explained shortly.
We continue to operate within the
framework of the classical linear
regression model (CLRM), weassume the
following:
• Zero mean value of ui, or E (ui | X2i, X3i) =
0 for each i
• No serial correlation, or cov (ui, uj) = 0 i j
• Homoscedasticity, orvar (ui) = σ 2
• Zero covariance between ui and each X
variable, orcov (ui, X2i) = cov (ui, X3i) = 0
• No specification bias, orThe model is
correctly specified
• No exact collinearity between the X
variables, orNo exact linear relationship
between X2 and X3
INTERPRETATION OF MULTIPLE
REGRESSION EQUATION
• As mentioned earlier, the regression
Given the assumptions of the classical
regressiol model, it follows that, ontaking
the conditional expectation of Y on both
sides, we obtainE(Yi | X2i, X3i) = β1 +
β2X2i + β3i X3iIn words, gives the
conditional mean or expected value of Y
conditional upon the given or fixed
values of X2 and X3. Therefore, as in
thetwo-variable case, multiple regression
analysis is regression analysis
conditional upon the fixed values of the
regressors, and what we obtain is
theaverage or mean value of Y or the
mean response of Y for the given values
ofthe regressors.
• THE MEANING OF PARTIAL
REGRESSION COEFFICIENTS
coefficients β2 and β3 are known as partial
regression or partial slope coefficients. The
meaning of partial regression coefficient is
as follows: β2 measures the change in the
mean valueof Y, E(Y), per unit change in
X2, holding the value of X3 constant. Put
differently, it gives the “direct” or the “net”
effect of a unit change in X2 on themean
value of Y, net of any effect that X3 may
have on mean Y. Likewise, β3measures the
change in the mean value of Y per unit
change in X3, holdingthe value of X2
constant.4 That is, it gives the “direct” or
“net” effect of a unitchange in X3 on the
mean value of Y, net of any effect that X 2
may have onmean Y.
 OLS AND ML ESTIMATION OF THE PARTIAL REGRESSION COEFFICIENTS
• OLS Estimators
To find the OLS estimators, let us
first write the sample regression
function(SRF) corresponding to the
PRF of as follows: Yi = βˆ1 + βˆ2X2i +
βˆ3X3i + ˆui where uˆi is the residual
term, the sample counterpart of the
stochastic disturbance term ui. As
noted in Chapter 3, the OLS
procedure consists in so choosing the
values of the unknown parameters
that the residual sum of squares
(RSS) is as small as possible.
Symbolically,
• where the expression for the RSS
is obtained by simple algebraic
manipulations. The most
straightforward procedure to
obtain the estimators that
willminimize is to differentiate it
with respect to the unknowns, set
theresulting expressions to zero,
and solve them simultaneously.
As shown inAppendix 7A,
Section 7A.1, this procedure
gives the following normal
equations.
 Where r2 3 is the sample coefficient of correlation between X2 and X3 as defined in
Chapter 3.

Variances and Standard Errors of OLS Estimators

• Having obtained the OLS estimators of the Or, equivalently

partial regression coefficients,we can derive the
variances and standard errors of these estimators
in themanner indicated in Appendix 3A.3. As in
the two-variable case, we need thestandard
errors for two main purposes: to establish
confidence intervals andto test statistical
hypotheses. The relevant formulas are as
follows:
 Properties of OLS Estimators

• The properties of OLS estimators of the multiple regression model parallelthose of

the two-variable model. Specifically:
• The three-variable regression line (surface) passes through the means Y¯, X¯ 2, and
X¯3, which is evident from of the two variable model]. This property holds generally.
Thus in the k-variable linearregression model [a regressand and (k − 1) regressors]:
Yi = β1 + β2X2i + β3X3i +···+ βkXki + ui, we haveβˆ1 = Y¯ − β2X¯ 2 − β3Xˆ 3 −···− βkX¯k.
• The mean value of the estimated Yi ( = Yˆi) is equal to the mean valueof the actual
Yi, which is easy to prove:Yˆi = βˆ1 + βˆ2X2i + βˆ3X3i= (Y¯ − βˆ2X¯2 − βˆ3X¯3) + βˆ2X2i +
βˆ3X3i (Why?)= Y¯ + βˆ2(X2i − X¯2) + βˆ3(X3i − X¯3) = Y¯ + βˆ2x2i + βˆ3x3i. where as usual
small letters indicate values of the variables as deviationsfrom their respective
means.Summing both sides over the sample values and dividingthrough by the
sample size n gives ¯Yˆ = Y¯. (Note: x2i = x3i = 0. Why?) Notice that by virtue, we can
writeyˆi = βˆ2x2i + βˆ3x3i, where yˆi = (Yˆi − Y¯).Therefore, the SRF can be expressed in
the deviation form asyi = ˆyi + ˆui = βˆ2x2i + βˆ3x3i + ˆui.= ¯uˆ = 0, which can be verified
from [Hint: Sum bothsides of over the sample values.]
• The residuals uˆi are uncorrelated with X2i and X3i, that is, == 0 (see Appendix 7A.1 for proof).
• The residuals uˆi are uncorrelated with Yˆi; that is, uˆiYˆi = 0. Why?[Hint: Multiplyon both
sides by uˆi and sum over the sample values.]
• From it is evident that as r2 3, the correlationcoefficient between X2 and X3, increases toward 1,
the variances of βˆ2 andβˆ3 increase for given values of σ 2 and In the limit, whenr2 3 = 1 (i.e.,
perfect collinearity), these variances become infinite. The implications of this will be explored
fully in Chapter 10, but intuitively thereader can see that as r2 3increases it is going to be
increasingly difficult toknow what the true values of β2 and β3 are. [More on this in the next
chapter, but refer to Eq.
• It is also clear from eg that for given values of r2 3and the variances of the OLS estimators are
directly proportional to σ 2; that is, they increase as σ 2increases. Similarly, for given valuesof
σ 2and r2 3, the variance of βˆ2 is inversely proportional to that is, thegreater the variation in the
sample values of X2, the smaller the variance ofβˆ2 and therefore β2 can be estimated more
precisely. A similar statement canbe made about the variance of βˆ3.
• Given the assumptions of the classical linear regression model, which are spelled out in
Section, one can prove that the OLS estimators ofthe partial regression coefficients not only
are linear and unbiased but alsohave minimum variance in the class of all linear unbiased
estimators. Inshort, they are BLUE: Put differently, they satisfy the Gauss-Markov theorem.
(The proof parallels the two-variable case proved in Appendix 3A,Section 3A.6 and will be
presented more compactly using matrix notation inAppendix C.)