J is the number of hypotheses being tested. Click OK. =0 against H. #0 .1. K is thenumber of parameters in the original model (including the intercept) We will use the F-test to test the null hypothesis H . : P. I The F-test 8. SSEu = the sum of squared errors for the model being tested 1.1. Open the workfile hamburger. To compute this statistic we need: i. which we used in Chapter 7 . : P. Method: Sample: - . in which we assume the hypothesis is true . 1 SSER= the SSE for the restricted model. 111.. T is the number of sample observations v.1 . 8. iv. Our task is to use EViews to conduct single and joint coefficient hypothesis tests and to incorporate nonsample information. and click on QuicMEstimate Equation and type in the total revenue equation making sure to include the intercept (C in EViews notation).I Constructing the F-Statistic The F-statistic given in UE/2 equation (8.wfl.3).Chapter 8 Further Inference in the Multiple Regression Model In this chapter we continue working with the hamburger chain data found in Table 7..

07282 17.3231 6. -- -istimation \$ -.kFlMA] . -5 - .Further Inference in the Multiple Regression Model 87 From the stored results.0000 120.0000 0.171520 49.Least .268585 1964. 91.855334 6. Squares (hILS and.83057 2. -.758 -168.546681 Mean dependent var Akaike info criterion Schwarz criterion F-statistic .@ssr ftest(2)=result-7-1 .2137 1. Dependent Variable: TR Method: Least Squares Date: 04R1/00 Time: 1150 Coefficient Std. % e the resulting object "Restrictl".. assuming the null hypothesis is true. we must save the sum of squared errors. coef(6) ftest ftest(1)=result-7-1.@ncoef :i 5i coef vector SSE-U T-K Now estimate the "restricted" model.871312 0. .@regobs-result-7-1 . Error t-Statistic Prob..949066 0.- Method.19372 0.

. p = P[<. Given the results of the unrestricted and restricted regressions. In order to computep-values and critical values for the F-statistic we need to use two EViews functions.168 49. @cfdist(x.95.v2) computes the critical value Fcfrom the F-distribution with vl and v2 degrees of freedom such that the probability to the left of Fcis p.vl. as shown in the next figure. Click on View/Coefficient TestsTWald-CoefficientRestrictions.00000 Note: We have "programmed" the calculations in EViews.I .vl.v2) computes the probability that the F-random variable with vl and v2 degrees of freedom falls to the left of the value x. 8.. 5 x ] @qfdist(p.1.ftest(2)) Freezing the result.2 Using EViews Coefficient Tests EViews makes using the F-test very simple.88 Chapter 8 Save the sum of squared errors and compute the F-statistic.] Using these functions.ftest(2)) ftest(6)=@qfdist(. and editing. Return to the regression output object Result-7-1. That is.. ftest(5)=l -@cfdist(ftest(4). I F. That is. . P[F.1. The advantage of using EViews is that we can compute thep-value and the exact critical value. we have the following: P-value F-critical value SSEU T-K 1805. the F-statistic value could be computed on a calculator and the critical value looked up (approximately) in the F-tables at the end of UE/2..

..... - :2"C[4] 'Li­ 3 The result is shown in the next figure...1 in UE/2.Resi Coyariance Matrir I I A dialog box opens in which you type in the hypotheses you wish to test... The test is stated in terms of the coefficient number..... . Estimation Equation: *I . It is an alternative testing procedure which we will not consider.........Further Inference in the Multiple Regression Model 89 I - Regresentations Estima tio on Outpul amel freeze1 ~stimatef ~orecastl t a t s~ e s i d s l ~ l I Actual.. as shown in the representations view of the output I I ' i: : . commas: ..... TR = C(1) + C(2)*P + C(3)*A Subst~tuted Coefficients: Pf is c(2). Enter the hypothesis and click OK. ILIIIJIla... Ignore the "Chi-square"value.Fitted.--.. which is Table 8.............

= 0. Multiple hypotheses are separated by commas in the Wald test dialog box.... We now jointly test the null hypothesis that the 2ndand 3d coefficients are zero. we can specify the hypothesis as we did in the previous section.331940 Probability 8. We are given the value of the test statistic and its p-value. : P.1 of this manual. We can also create this value from the ANOVA table output in Chapter 7. Alternatively. in the estimation object. . .90 Chapter 8 Equation: RESULT-7-1 4. . In EViews this test is reported automatically as part of the regression output..7. Coefficie i-. the overall test of model significance is the joint test of the null hypothesis H.2 Testing the Significance of the Model The overall significance of the model is evaluated using an F-test. P3= 0.P = 0 against the alternative hypothesis H I :at least one of the Pk is nonzero. In the multiple regression model .. in the lower right-hand corner.

.Further Inference in the Multiple Regression Model The output is the same as reported in the EViews output. -_ I I . To reproduce the results in Result 8. i tr c p a aAZ i 1 i I I i Your results are as follows: . Note the EViews exponentiation character is '"'.1).3 An Extended Model The worlcfile chap8-3 contains 78 weeks of observations on tr. OR llKe I Eguation S ~ t e a t l c . including the advertising squared term AA2. ~. 91 : Equation: RESULT-7-1 Null Hypothesis: C(2)=0 8.4. [ Dependent variable an expl~citeqirat~un r =ql]+c(2]'% lollowd by list of mdudmg ARM* and PDL terms.~. p and a.5 Click on QuicMEstimate Equation from the EViews main menu and specify equation (8. Click on OK.

T = the number of observations in the sample.301 3.918707 2592. In this case.4 The Significance of Advertising Next we generate the F-statistic for the joint test of the hypothesis: Ho: P3 = 0.360999 -0.015887 29. Click OK. C(4) = 0 in the coefficient restrictions field of the Wald Test dialog box.) i 92 Chapter 8 1 Included observations: 78 Coefficient Std. P4 = 0 to determine the significance of advertising in our model.6179 Mean dependent var Akaike info criterion Schwarz criterion 8.026755 0.0000 0.19792 3.741410 1.0964 122.52473 -6. . and K = 4.Coefficient Restrictions from your equation's toolbar and enter the joint null hypothesis restrictions C(3)=0.degrees of test K freedom in the denominator. J = 2. Error t-Statistic Prob.446944 7. where J = the number of coefficient restrictions.604041 0.969966 -1.0000 0. T = 78.581822 0.4641 -10.0000 0. A 110. Click on View/Coefficient TeststWald . The resulting F(J.421708 0.878548 5.T-K)statistic has J degrees of freedom in the numerator and T. and K = the number of estimated coefficients.

Further Inference in the Multiple Regression Model 93 Equat~on. 8. is the restricted equation q r ) ( tr.vl.41 whereas the F(2. RESULT-8-5 Null Hypothesis: C(3)=0 C (4)=0 F-stat~st~c 261. you must have previously estimated and named both the restricted (eq-r) and unrestricted (eq-u) equations. el is the unrestricted equation ( e q u ) + @ssr is the sum of squared errors (SSE in UE/2). Note: To calculate fstat. enter the restriction p3 + 80P4= 1. and click OK.Coefficient Restrictions from your equation's toolbar.95.@ssr)/2)/(eq~u.@ssr/(@regobseq-u. .2. = pl + Pzp.v2) is the upper a percent right tail critical value for the F-distribution with vl . @regobs is the number of regression observations (Tin your textbook) @ncoef is the number of estimated coefficients (K in your textbook) @qfdist(l-a.5 The Optimal Level of Advertising To compute the F statistic regarding the hypothesis that \$40. The foregoing F-test statistic and F-critical value could also be computed with the following commands typed into the EViews command window: scalar fstat = ((eq-r. As your text notes. our conclusion is critical to reject the null hypothesis that advertising has no statistically significant effect on total revenue in favor of the alternative hypothesis that at least one of the advertising coefficients is non-zero.000 per week is the optimal level of advertising Click on ViewICoefficient TestsIWald .@ncoef)) scalar fcrit = @qfdist(.eq~u.@ssr . + e. degrees of freedom in the numerator and v2degrees of freedom in the denominator.4110 Probabil~ty 0 000000 The F-statistic is 261.74) value is 3.74) where tr.12. = pl + Pzpf + P3af+ p4a2.

05 can be computed by typing the following command into the EViews command window: scalar fcrit-ad = @qfdist(. as explained in Chapter 8.4. note that the square root of 0.1. Click on View/Coeficient TestsIWald . = .74) critical value for a = .000. Therefore.95.74) where we name the critical value fcrit-ad so as not to confuse it with the F-critical value calculated earlier. Since the test statistic value 0. 8.0637 is less than the critical value fcrit-ad = 3.4.6 The Optimal Level of Advertising and Price Next we conduct the F test regarding optimal advertising and price presented in section 8.6 in UE/2.063721 is 0. For a single equality hypothesis there is an exact relationship between the F-test and t-test procedures: t2 = F. The F(1. In the optimal advertising case. we fail to reject the null hypothesis that the optimal level of advertising is \$40.Coefficient Restrictions from your equation result-8-5 toolbar and enter the restrictions P3 + 80P4= 1 and + 2P2 + 40P3+ 1600P4 175 and click OK. In ~~iews. Note that this same result is forthcoming from a t-test of the optimal advertising hypothesis.970.252.The test statistic follows the F distribution with degrees of freedom in the numerator equal to the number of restrictions (J= 1 in this case) and denominator degrees of freedom equal to the number of regression observations less the number of estimated coefficients (T ­ K = 74 in this case).2 of UE/2. the t-test statistic reported in result R8.3 of UE/2. either test will produce the same decision.

75 is less than the F-critical value = 3.180375 You can compute the F(2.000. .74) value for a = . total revenue will average \$175.05 by typing the following command into the EViews critical command window: scalar fcrit-ad-price = @qfdist(. and at a price of \$2 per burger. The results of this test are 1-1 ' ~ i e v ~o c~ ~ h j e c t s j Print tdamel ~reeze] ~stirnatel ~r ~ zl ~orecastl t a t 4 RI s 1 Wald Test Equation: RESULT­8-5 Null Hypothesis: C(3)+80*C(4)=1 C[1)+2*C(2)+40*C(3)+1600*C(4)=175 F-statistic 1.752979 Probability 0.2. we fail to reject the null hypothesis that the sample data are consistent with the joint hypothesis that optimal weekly advertising expenditures are \$40.Further Inference in the Multiple Regression Model 95 i Coefficient restnct~oqs sepa!ated b ~ l ccp_mmas.12.95.74) Since the F-statistic = 1.000 per week.