Steps in the model
building process
4
Population multiple
linear regression
model
Assumptions of the
multiple linear
10 model
1 5 Multiple correlation
and regression
Step 2.
Step 3.
Step 4.
Step 5.
Step 6.
Step 7.
Steps in the multiple linear regression model-building process
Identify the variables to be included in building the multiple reares-
“sion model. : ;
Collect sample data.
Specify the relationship that exists between the dependent and the
independent variables, and between the independent variables.
Estimate the parameters of the model specified.
Determine whether or not the assumptions of the multiple linear
regression model have been met.
Statistically test for the usefulness of the mode! developed and for the
individual terms in the model.
Use the model developed for prediction and estimation.
‘The population multiple linear regression model
Po + Bix t Bria t+ + Bate 72 1,2...,N
Y= ith dependent random variable corresponding to xi.15 4125 ++» 5 Xie
Bo. Bry Ba» Beare (k + 1) parameters in the model.
ij = ith level of the jth independent variable, j= 1,2,... ,
«; = Random error term.
Assumptions of the multiple linear regression model
expected value of the error component & is 0 (E(e;) = 0}, and the
1. The
variance of the error component ¢; is 7 [V(e)) = a7). fori= 1,2... . vn.
2. The error components are uncorrelated,
3. Bos Bis. =< Beare (k #1) parameters, and .s;..i2.. - - sxj,are known.
constants.
And, for purposes of hypothesis testing and confidence interval esti
4, The
error component ¢; is normally distributed.| Hypothesis test for
Bi
Conditional proba-
bility distribution
fees point
estimate
‘Two-tailed hypothesis test of regression coefficients in
multiple regression model
Model: Y= fy + Aix; + Barz +++ ++ Bim te
Hypothesis: Hy: 8; = 0
Ha: B #0
Test statistic value: ¢ = 2
Soy
Decision rule: Reject if: > tun,
Do not reject if:
ney) OF 1S ~harnteene
~tainntke) SES fatintks
where
n= Number of observations,
k= Number of independent variables,
Point estimate of conditional probability distribution
variance—multiple linear regression model
2 i
ak I
SSE
ea)
(15.1)
SVrrttoonth
coef
X12,
| Point estimation of the coefficient of multiple
determination, p?
The coefficient of multiple determination is defined similarly to the coeffi-
cient of determination in the simple linear regression case. That is, the
dent variables. The value of r? is computed using equation 15,2.
Coefficient of
multiple determina
Sample coefficient of multiple determination, 1?
ficient of multiple determination is the proportion of variability in the
random variable Y accounted for or explained by the independent variables
+t, This sample statistic measures the relative strength of the
linear relationship between the depéndent random variable and the indepen-
iNormal plot test |
‘The normality plot test for the normality
assumption concerning the residuals
Hypotheses: Hy: Error terms are normally distributed,
Hq: Error terms are not normally distributed,
Test statistic value: r= Correlation between the standardized residuals and
the expected standardized residuals (NSCORES)..
Decision rule: Reject Hy ifr < ry
‘ Do not reject Hy if r = r.
where re is given i Table B.21, Appendix B with a = 0.01, 0.05, and 0.10.
‘The Durbin-Watson test for first-order
autocorrelation of the error terms
Hypotheses: Ho: p =0
Ha: p>0
where p is the first-order autocorrelation of the error terms.
Test statistic value:
Seek
>
where ¢; is the ith residual, and n is the sample size.
Decision rule:
If DW > d,, do not reject Ho: p
If DW < dj, reject Ha: p = 0.
If d, = DW = d,, the test is inconclusive
The critical values d, and d, are given in Table B.20, Appendix B for
and a = 0.01
0.05
The Durbin-Watson test is a test for the first order autocorrelation, p. of the
error terms. The first-order autocorrelation p measures the correlation be-
tween consecutive pairs of error terms: (€1. €2), (€3. €1), (€1, ea),
Although the error terms may be correlated in more complex ways (2.2..(€1.
£:), (€2. €4), (x. es}... . ), usually when they are correlated, it is through
the first-order correlution of consecutive pairs. The DW test statistic has a
minimum value of 0 and a maximum value of 4. If the null hypothesis is true
(p = 0), then the expected value of the test statistic is 2. Hence a value of
|