Econometric Analysis - 6th Edition

Econometric Analysis, 6th Edition

ERRATA and DISCUSSION
Last updated March 14, 2010

Page Number Date Posted Errata

PAGE NUMBER DATE

erratum ... may be repeated for more than one on the page

44 11/30/08

In Table 4.1, Assumption A1., at the end of the expression, b_Kx_iK should be x_iKb_K. (Li Da, University of Macao)

55 06/28/09

In Table 4.3, the t ratio for lnP_G should be -0.485. (Mark Strahan, Sand Hill Econometrics, Palo Alto.)

57 09/16/08

In Example 4.5, the critical value should 2.39, not 2.37. (Professor Pedro Bacao, University of Coimbra, Portugal.)

69 10/12/09

In the second line of the third paragraph of Example 4.7, "elasticites" should be "elasticities." (Professor William Greene, New York University.)

69,70,71 09/16/08

In Table 4.7,
The analysis is based on 51 observations, not 52. Coefficient -0.178397 should be -0.178395 Standard error 0.04377 should be 0.0147327 Standard error 0.07771 should be 0.0550265 Standard error 0.15707 should be 0.0551728 Standard error 0.10330 should be 0.0357706 t Ratio -3.377 should be -3.233
At the top of page 70,
Reported long run price elasticity -0.411361 should be -0.411358. Reported long run income elasticity 0.970527 should be 0.970522. Reported asymptotic variance 0.026368 should be 0.0263692. The critical t value for 51-6 = 45 degrees of freedom 2.013 should be 2.014. The 95% confidence interval for the long run price elasticity should be -0.411358 +/- 2.014(0.152296) = [-0.718098,-.104618]. In matrix C in Example 4.8: 0.0147326 should be 0.0147327, 0.00109461 should be 0.00110072, -0.0396227 should be -0.0396215, 0.0216259 should be 0.0216279.
The remaining results on page 71 in Table 4.8 are dependent on the simulation. The first two columns of the table are the results reported earlier. The coefficient estimates are correct. The two derived estimates, f₂ and f₃ should be as given above and in the table below. The standard errors are as shown earlier and reported again in the table below. The simulation results will be specific to the application. A program that can produce the particular results shown below and can be reproduced (not the results in the text) using NLOGIT or LIMDEP can be found at delta-method-and-simulation.txt
======================================================== Regression Estimate Simulated Values Parameter Estimate Std.Error Mean Std.Dev. ======================================================== BETA2 | -0.069532 0.01473267 -0.070018 0.0147507 BETA3 | -0.164047 0.05502650 0.162375 0.0541288 GAMMA | 0.830971 0.04576345 0.833081 0.0450839 PHI2 | -0.411358 0.152296 -0.462782 0.205537 PHI3 | 0.970522 0.162386 0.966962 0.188948

? Example 4.7 in Detail. ? Create logged variables for the regression sample;1-52$ create;logg=log(gasexp/(gasp*pop)) ; logg1=logg[-1]$ create;logpg=log(gasp);logi=log(income)$ create;logpnc=log(pnc);logpuc=log(puc)$ ? Drop first observation sample;2-52$ regress;lhs=logg;rhs=one,logpg,logi,logpnc,logpuc,logg1; printvc$ ? Pick off specific coefficients calc ; br2=b(2) ; br3=b(3) ; cr=b(6)$ calc; list ; phir2=br2/(1-cr) ; phir3=br3/(1-cr)$ calc; list ; g22=1/(1-cr) ; g26=g22*phir2 ; g33=g22 ; g36=g33*phir3$ matrix ; g=[0,g22,0,0,0,g26 / 0,0,g33,0,0,g36]$ matrix ; list ; ve=g*varb*g'$ calc;list;sef2=sqr(ve(1,1)) ; sef3 = sqr(ve(2,2))$ ? The preceding is automated with Wald ; labels=b1,b2,b3,b4,b5,gama ; start=b ; var=varb ; fn1=b2/(1-gama) ; fn2 = b3/(1-gama) $ calc;list;ct=ttb(.975,45)$ ? Report the confidence interval. calc;list ; lowerf2=phir2-ct*sef2 ; upperf2=phir2+ct*sef2 $ calc;list ; lowerf3=phir3-ct*sef3 ; upperf3=phir3+ct*sef3 $ ? ? Krinsky and Robb Procedure Standard Errors ? matrix;b2b3c=[br2/br3/cr]$ matrix;vb2b3c=[varb(2,2)/varb(2,3),varb(3,3)/ varb(2,6),varb(3,6),varb(6,6)]$ ? Cholesky decomposition of asymptotic covariance matrix. Matrix;c=chol(vb2b3c)$ ? Results can be replicated. Calc;ran(1234567)$ matrix;v=rndm(1000,3) ; vc = v*c'$ sample;1-1000$ crea;beta2=0;beta3=0;gamma=0$ namelist;bsim=beta2,beta3,gamma$ create;bsim=vc$ create;beta2=beta2+b(2) ;beta3=beta3+b(3) ;gamma=gamma+b(6)$ create;phi2=beta2/(1-gamma) ; phi3 = beta3/(1-gamma)$ dstat;rhs=bsim,phi2,phi3$ calc ; list; lowerf2=xbr(phi2)-ct*sdv(phi2) ; upperf2=xbr(phi2)+ct*sdv(phi2)$ calc ; list; lowerf3=xbr(phi3)-ct*sdv(phi3) ; upperf3=xbr(phi3)+ct*sdv(phi3)$ ? ? Use empirical percentiles instead. ? sort;lhs=phi2$ sort;lhs=phi3$ calc;list;phi2(26);phi2(975)$ Result = -0.983866 Result = -0.209776 calc;list;phi3(26);phi3(975)$ Result = 0.539668 Result = 1.321067 ======================================================== Regression Estimate Simulated Values Parameter Estimate Std.Error Mean Std.Dev. ======================================================== BETA2 | -0.069532 0.01473267 -0.070018 0.0147507 BETA3 | -0.164047 0.05502650 0.162375 0.0541288 GAMMA | 0.830971 0.04576345 0.833081 0.0450839 PHI2 | -0.411358 0.152296 -0.462782 0.205537 PHI3 | 0.970522 0.162386 0.966962 0.188948
(Professor Pedro Bacao, University of Coimbra, Portugal.)

86 09/16/08

In Example 5.1 and Table 5.1, reference is made to Equations (5-1) and (5-3). In this edition, the investment equation is not numbered, so the "of equations (5-1) and (5-3) should state "the first and third investment equations in the Introduction." (Professor Pedro Bacao, University of Coimbra, Portugal.)

98 09/16/08

The denominator in the formula for z at the top of the page should be 0.016078. The result is correct. (Professor Pedro Bacao, University of Coimbra, Portugal.)

101 09/16/08

For the forecast made at the top of the page, the estimated coefficient vector using 1990.I to 2000.IV for the estimation is
matrix ; bvec=[-17.6992/0.000500633/0.000339424/2.86348/-0.00460571]$
As given on page 100, the data vector is
matrix ; xvec=[1/4.48/5.26/9.1396/204]$
The estimate s² is 0.000873331. The predicted value is, therefore,
calc;list ; xb=xvec'bvec ; if=exp(xb + ssqrd/2)$ = 1875.746
not 1885.2 as reported in the example. The overestimation is closer to 9.0% rather than 9.54% as reported. (Professor Pedro Bacao, University of Coimbra, Portugal.)

113 06/28/09

In equation (6-5), in the denominator after the first equals sign, x should be X. (Mark Strahan, Sand Hill Econometrics, Palo Alto.)

116 09/16/08

To avoid a possible confusion, it might be noted that, unlike the rest of the book, all of Example 6.3 uses base 10 logs. This is because the results reported are Nerlove's, not mine, and he used base 10 in his computations. (Professor Pedro Bacao, University of Coimbra, Portugal.)

119 09/16/08

In Example 6.4, Table 6.5, the standard errors for the MLEs, 2.403 should be 2.345 and 0.663 should be 0.794. These are computed using the negative inverse of the actual Hessian. Full details are shown below for page 505. (Professor Pedro Bacao, University of Coimbra, Portugal.)

124 05/19/09

At the end of the page, the degrees of freedom of the three F statistics should be [7,180], [7,19] and [7,150], respectively, and the critical values should be 2.061, 2.543 and 2.071, respecrtively. (Ola Lotherington Vestad, Ragnar Frisch Institute, University of Oslo.)

125 09/16/08

In Example 6.7, the critical value for F[6,40] is 2.336, not 2.467 as reported in the text at the end of the page. (Professor Pedro Bacao, University of Coimbra, Portugal.)

125 07/25/08

In table 6.6, in all three regressions reported, the coefficients on Govt.Eff. and Democracy should be interchanged. I.e., for "All Countries," the coefficient on Govt.Eff. should be 0.748 and that on Democracy should be 1.629. (Markus Hertrich, University of Basel.)

126 09/05/08

In Table 6.7, in the Preshock column, 9.8482 should be 0.8482. It should be emphasized that the time trend is computed as year-1952, so it takes values 1,2,...,52. (Professor William Greene, New York University (on 10/06/07).)

In the same example, in the last paragraph on page 126, the chi squared reported there, 158.753, was computed by comparing the "Preshock" coefficient vector to the "1953-2004" coefficient vector, which is the wrong test. If the statistic is computed to compare the "Preshock" to the "Postshock," the value is 502.34 with the same conclusion. (Professor Noel Roy, Department of Economics, Memorial University of Newfoundland)

142 09/06/08

In example 7.3, two occurrences of e₀'e₁ should be e₁'e₁. In addition, the b in the second line of the example, in (7-15) and in the line after (7-15) should all be bold and the e in the line before (7-15) should also be bold. (Professor Noel Roy, Department of Economics, Memorial University of Newfoundland)

164 09/16/08

In example 8.2, the falue .776 in the F statistic should be .756. Then, the F ratio becomes 7.976, not 7.771. If one uses 1 - .243578, then the F ratio will be 7.956. The 7.771 appears at two places in Example 8.2. (Professor Pedro Bacao, University of Coimbra, Portugal.)

164 09/16/08

In Table 8.1, the t ratio for the constant term should be -1.19, not -1.10. (Professor Pedro Bacao, University of Coimbra, Portugal.)

165 09/26/08

In Table 11.1, in the third column, the last standard error, 95.632, should be 95.672. (Professor Uli Kleinwechter Agricultural and Food Policy Group, University of Hohenheim, Stuttgart.)

167 09/16/08

In Example 8.3, the expression given for the normality test is incorrect. The correct expression is
c²[2]=n[(1/6)(m₃/s³)² +(1/24)((m₄/s⁴)-3)²]. The value given of 482.12 should be 497.35. (Professor Pedro Bacao, University of Coimbra, Portugal.)

171 09/16/08

In Example 8.4, the value of 2.946 given 4 lines below the equation should be 2.959. The 2.946 reported results if LF is erroneously included in the regression used to compute the residuals. (Professor Pedro Bacao, University of Coimbra, Portugal.)

173 09/16/08

For formulating the LM statistic defined in (8-37), which is 1/2 the regression sum of squares in the regression of g on Z, a considerable convenience is obtained by replacing the overall constant term prescribed for Z just before (8-28) with the remaining dummy variable. The regression sum of squares in this regression will of course be identical because the Z obtained with an overall constant and G-1 of the dummy variables has the same column space as the Z built up from the G dummy variables. With this substitution, simply multiplying out the expression in (8-28) becomes simple, and it results in (8-37), except the summation must be from 1 to G rather than from 2 to G. Thus, the correction here is simply to change the starting index for the summation from 2 to 1.
There is a notational inconsistency in the text. In (8-28), Z includes the constant term. In the discussion on page 173, that overall constant term has fallen out of Z. To resolve the inconsistency, a way to proceed is to change Z to D in the 4th and 6th lines of the paragraph before (8-37), and then Z= [i,D]. In the rest of the paragraph, all occurrences of G-1 are changed to G (3 times), and as noted above, the summation in (8-37) runs from 1 to G not from 2 to G. (Professor Pedro Bacao, University of Coimbra, Portugal.)

173,174 09/28/08

The groupwise heteroscedastic models in both cases are computed by applying ordinary least squares to the transformed data x*(i,t) = [1/s(i)]x(i,t) where s(i) is the groupwise specific standard deviation of the pooled OLS residuals. This suggests two ways to compute the standard errors for the FGLS estimator. The first (asymptotic) is to use (X'W^-1X)^-1 where the diagonal matrix W is formed from the original group specific residual variances. The second is to use the standard OLS results for the transformed (weighted) data. Asymptotically these give the same answer. They will differ in a finite sample. The results in the text use the latter method. (Professor Pedro Bacao, University of Coimbra, Portugal.)

174 03/11/08

In Example 8.5, the value for the LM statistic of 160.576 reported at the bottom of the page is incorrect. It should be 277.443, as shown by the NLOGIT results below.
crea;d1=i=1;d2=i=2;d3=i=3;d4=i=4;d5=i=5;d6=i=6 ;d7=i=7;d8=i=8;d9=i=9;d10=i=10;d11=i=11;d12=i=12 ;d13=i=13;d14=i=14;d15=i=15;d16=i=16;d17=i=17 ;d18=i=18$ name;D=d2,d3,d4,d5,d6,d7,d8,d9,d10,d11,d12, d13,d14,d15,d16,d17,d18$ name;Z=d1,d$ regr;lhs=lgaspcar ;rhs=one,lincomep,lrpmg,D;res=e$ calc;ee=e'e$ crea;g=e*e/(sumsqdev/n)-1$ matrix;list;lm=.5*g'Z*ginv(Z'Z)*Z'g$ Matrix LM has 1 rows and 1 columns. 1 +-------------- 1| 277.44275
(Fernando Martel, Department of Politics, New York University and Professor Pedro Bacao, University of Coimbra, Portugal.)

193 09/16/08

In Table 9.3, the labels DALE and COMP are interchanged. The reference to DALE in the third to last line of Example 9.3 should be to COMP. (Professor Pedro Bacao, University of Coimbra, Portugal.)

202 10/14/09

In footnote 13, "weighing" should be "weighting." (Professor William Greene, New York University.)

209, 210 09/16/08

In Example 9.8, the test statistic has 9 degrees of freedom. The critical value is 16.919, not 14.07 as reported in the text after the equation in the example on 209 and again in Example 9.8 on page 210. The value for the test statistic given there is incorrect. The 219.17 should be 3193.69. Note that the claim is made that the version of the test statistic in (9-42) and that in Example 9.8 will be numerically identical. This is true in theory, but will not be in practice. The difference will arise because the two treatments use different empirical estimates of the variance parameters that make different degrees of freedom corrections. Absent this, the two forms are, indeed, identical. Note, for example, that the derivation of the result in Baltagi (2005, pp. 65-73) uses the theoretical variance parameters, not the empirical values based on sums of squares with degrees of freedom corections. NLOGIT commands that will produce the two values of the statistic are given below.
name;x=exp,expsq,wks,occ,ind,south,smsa,ms,union$ samp;all$ regr;lhs=lwage;rhs=one,x;pds=7;panel$ regr;lhs=lwage;rhs=one,x$ regr;lhs=lwage;rhs=one,x;pds=7;panel;fixed$$ matr;bf=b;vf=varb$ regr;lhs=lwage;rhs=one,x;pds=7;panel;means$ matr;bm=b(1:9);vm=varb(1:9,1:9)$ matr;d=bf-bm;vd=vf+vm$ matr;list;h=d'*ginv(vd)*d$ crea;expb=groupmean(exp,pds=7)$ crea;expsqb=groupmean(expsq,pds=7)$ crea;wksb=groupmean(wks,pds=7)$ crea;occb=groupmean(occ,pds=7)$ crea;indb=groupmean(ind,pds=7)$ crea;southb=groupmean(south,pds=7)$ crea;smsab=groupmean(smsa,pds=7)$ crea;msb=groupmean(ms,pds=7)$ crea;unionb=groupmean(union,pds=7)$ name;means=expb,expsqb,wksb,occb,indb,southb,smsab,msb,unionb$ regr;lhs=lwage;rhs=means,x,one;panel;pds=7;random$ matr;c=b(1:9);vc=varb(1:9,1:9)$ matr;list;hp=c'*ginv(vc)*c$
(Professor Pedro Bacao, University of Coimbra, Portugal and William Greene, New York University.)

217 09/16/08

In Table 9.6, the value of 0.985422 should be 0.085422.
The robust standard errors reported in parentheses in the second column of values were computed by making a degrees of freedom correction to (9-4); the resulting matrix has been multiplied by (nT)/(nt-K) = 816/810. (Professor Pedro Bacao, University of Coimbra, Portugal.)

216,224,254 11/23/07

In the specification of the production function at the beginning of Example 9.9, the term lnp_cap_jkt should be lnpc_jkt. That is, the model is estimated using the log of private capital and the three separate components of public capital. In the data,
p_cap = hwy + water + util
This same correction must be made at the beginning of Example 9.12 on page 224 and at the beginning of Example 10.1 on page 254. The numerical results given in the examples are based on the correct specification, that is, using private capital, not aggregate public capital.
In the first line of Example 9.9, "Munell" should be "Munnell." In the listing of state abbreviations for Tornado Alley, "AK"should be "AR" and "MS" should be "MO." These latter two state abbreviations should be also be changed likewise in the definition of "Central" (which is the same as Tornado Alley) on page 254. (Professor Randall Campbell, Mississippi State University.)

226 4/03/08

In equation (9-54), at the end of the equation, the ")²]" should be ")]²." (Carter Hill, Louisiana State University)

259 12/06/07

In equation (10-12), the "^" over both appearances of W should be deleted. (Arne Henningsen, University of Kiel )

259-265 11/23/07

Numerous corrections are needed in Examples 10.2, 10.3 and 10.4.
The Breusch-Pagan LM statistic based on the correlation matrix shown on page 259 in the text is 119.242, not 215.88 as reported in the text. However, this is not the correct correlation matrix for the OLS residuals based on the data used in the example. The correct correlation matrix is
GF MW MA MT NE SO SW CN WC GF 1.0000 MW .1734 1.0000 MA .3989 .0633 1.0000 MT .4467 .6969 -.0158 1.0000 NE -.5473 -.2896 .1915 -.5372 1.0000 SO .5253 .4893 .2328 .3433 -.2411 1.0000 SW .4247 .1320 .6514 .1301 -.3220 .2594 1.0000 CN .7509 .3370 .3904 .4957 -.2979 .8050 .3465 1.0000 WC .1672 .5654 .2116 .5736 -.0576 .2693 -.0374 .3818 1.0000
The chi-squared statistic based on this matrix is 102.336. This is also the value that should be given in Example 10.4 on page 265, where the 215.879 is repeated. The 48 state value of 2455.71 also given in Example 10.4 is incorrect (it is based on p_cap instead of pc. The correct value is 2915.564

In Table 10.1, the OLS results for GF and for the Pooled data are incorrect. The correct results are as follows:
GF 11.5669 0.002124 -2.0285 0.1005 1.3577 0.8049 -0.006562 0.01097 0.9970 (4.8502) (0.3012) (1.14883) (0.1243) (-0.8655) (0.1592) (0.003643) Pooled 2.5159 0.2685 -0.1549 0.3123 0.2438 0.3122 -0.01900 0.04171 0.9959 (0.1159) (0.01759) (0.02805) (0.03334) (0.03189) (0.03052) (0.002285)
The remaining OLS results require the following corrections:
For MW, 1.7834 should be 1.7835 For MT, 0.2948 should be 0.2949 and -0.008143 should be -0.008144 For NE, 6.3783 should be 6.3784 For SO, -0.02040 should be -0.02041 and 0.9851 should be 0.9852 For CN, 0.9936 should be 0.9937 For WC, 9.1108 should be -9.1108
Because of the incorrect covariance (correlation) matrix, the FGLS results must all be replaced. The correct values (based on the OLS residuals, using no iteration, and no degrees of freedom correction in the covariance matrix of the residuals) are
GF 11.8735 -0.16721 -1.8161 0.1706 1.1440 0.9349 -0.003974 0.008413 0.9970 (2.02941) (0.1270) (0.4618) (0.05565) (0.3296) (0.08108) (0.001951) MW 4.1692 0.06698 -0.1397 -0.1332 0.5323 0.5372 -0.01565 0.007625 0.9984 (1.00323) (0.08627) (0.1285) (0.06115) (0.1121) (0.08544) (0.002502) MA -8.4614 0.3278 1.6885 0.5173 -0.3140 -0.3093 -0.02906 0.007975 0.9950 (2.1246) (0.1056 (0.3301) (0.06105) (0.1270) (0.1839) (0.004445) MT 2.5230 0.1767 0.1227 -0.3669 0.03931 1.09375 -0.005851 0.01295 0.9940 (3.4487) (0.09170) (0.1142) (0.1733) (0.4629) (0.1704) (0.004203) NE 3.1392 -0.09790 0.2099 0.2446 -0.3088 1.04987 -0.001048 0.006597 0.9986 (1.3723) (0.04677) (0.1704) (0.04875) (0.1202) (0.09656) (0.002644) SO -13.1441 0.09109 0.9961 -0.5823 -0.2988 2.5935 0.02130 0.01719 0.9852 (7.6199) (0.1293) (0.7830) (0.13609) (0.2405) (0.4672) (0.008097) SW -19.1347 -0.3251 3.1643 -0.09668 -1.7129 2.1875 0.01712 0.009914 0.9864 (2.8916) (0.09034) (0.9215) (0.2055) (0.5817) (0.4932) (0.008129) CN 2.7917 -0.03071 -0.5043 -0.02158 0.1591 1.5380 0.006153 0.01345 0.9937 (0.8510) (0.09576) (0.1866) (0.1436) (0.1674) (0.1789) (0.004281) WC -10.2819 0.04750 1.8131 0.6651 -0.4305 0.004467 -0.02914 0.008320 0.9895 (2.4115) (0.1088) (0.4492) (0.09973) (0.1909) (0.1798) (0.003666) Pooled 2.2802 0.2365 -0.03934 0.1226 0.3124 0.3653 -0.01494 NA 0.9995 (0.02538) (0.002373) (0.004915) (0.004646) (0.005230) (0.007060) (0.0003400) FGLS 2.2312 0.2991 -0.1320 0.2687 0.2467 0.3177 -0.01967 NA 0.9967 Het. (0.03643) (0.004196) (0.006546) (0.007369) (0.007622) (0.008832) (0.0005227)
In example 10.3, using the FGLS results above, the correct chi squared statistic is 6,429.85, rather than 10,554.77 given in the example. The chi squared value for the constant returns to scale test should be changed from 148.418 to 136.390. The 9 item discrepancy vector shown on page 203 should be
(-0.73381, -0.13633, 0.91025, 0.06581, 0.09767, 1.79957, 2.21725, 0.14053, 1.09964)
(Professors Randall Campbell, Mississippi State University and William Greene, New York University.)

280 4/30/08

In Table 10.5, the four values in the last row of the substitution matrix should be 0.38512, 0.58205, 0.34994, -0.31536, respectively. (Brian Gould, University of Wisconsin)

287 10/14/09

In equation (11-3), to make the function f(y_i|x_i,b) consistent with the right hand side, we should make the substitution of y_i - h(x_i,b) for e_i. (Professor William Greene, New York University.)

321 03/28/09

At the end of Example 12.3, two lines up from the end, "have might" should be "might have." (Dr. Nada Wasi, University of the Thai Chamber of Commerce, Bangkok.)

325 09/22/08

In Example 12.4, in the second point, about the Wu test, the denominator degrees of freedom for the F statistic is 200, not 201 and the critical value is 3.888, not 4.15. The reported statistic, 8.809 = 2.968² is correct. (Professor Pedro Bacao, University of Coimbra, Portugal.)

388 09/28/08

In Table 13.4, the three entries for TR² using the two stage least squares results should be 8.77151, 1.81497, 12.4952, respectively. (Professor Pedro Bacao, University of Coimbra, Portugal.)

394 09/28/08

In Figure 13.2, the first two spending multipliers should be about +1.9. The first tax multiplier is about -0.3, not -1.3. In the text, the balanced budget multiplier should be 2.332-0.5446=1.7874. (Professor Pedro Bacao, University of Coimbra, Portugal.)

400 3/03/08

In the first line on the page, "theory for of" should be "theory for" (Ryan Polansky, Arizona State University)

402 10/14/09

In the first and second equations in Example 14.2, the boldface symbols F should not be bold. (William Greene, New York University)

436 09/22/08

In the continuation of Example 15.5, at the top of page 436, the matrix F should be multiplied by n=20, producing F=[500.68,14.31/14.31,0.47746].

In the listing of the second derivatives, the first 1/n before the brackets should be deleted. The result is [-H]^-1=... In the actual values, 0.51203 should be -.51243; 0.01637 should be 0.01638. (Professor Pedro Bacao, University of Coimbra, Portugal.)

450 3/03/08

In the 5th line below (15-7), q_n(q) should be q_n(q₀). (Ryan Polansky, Arizona State University)

460 09/22/08

In the second equation on the page, in the asymptotic variance, the matrix in the second line should be multiplied by n. (Professor Pedro Bacao, University of Coimbra, Portugal.)

477 05/16/08

After the second equation, "E[Z_i'u_i]=0", in the statement u_i=[u_i,1987,...], the order of the subscripts should be reversed. They should run 1983, 1984, ..., 1987, not the reverse. (Fernando Martel, Department of Politics, New York University.)

478,479 09/22/08

To be consistent with the surrounding notation, the "q=..." at the center of the page should be "nq=..." The same correction would apply to the header at the bottom of the first column of Table 15.5. The values in the table, save for the correction below, are correct. (Professor Pedro Bacao, University of Coimbra, Portugal.)

479 09/22/08

In Table 15.5, under Revenue, in the m=3 column, -2.7152 should be 2.7152. (Professor Pedro Bacao, University of Coimbra, Portugal.)

495 3/03/08

In the numerator of the definition of g^_i, the function should be f(y_i|x_i,q^) (Ryan Polansky, Arizona State University)

497 3/03/08

At midpage, in the line before "Newey and McFadden..," "advance" should be "advanced." (Ryan Polansky, Arizona State University)

505,506 10/06/07

The results in the application on pages 505 and 506 were not accurately computed. By recomputing them with tighter tolerances, the results change at the second and third decimal places. The values in table 16.1 should be
-4.7185 (2.3445) 15.6027 (6.974) 3.1517 (0.7943) 1.0000 (0.000) -82.91605 -88.43626 0.0000 0.0000 0.0000 7.9145 -0.85570 -0.026941 -7.4592 -32.8987 -2.2420 0.66891
The values in the three matrices should be changed to
V = [5.499/-1.653,0.6309],
V_E = [4.900/-1.473,0.5768],
V_B = [13.37/-4.322,1.537]
In the confidence interval test, 3.1517 should be 1.1509 and [1.5942,4.7085] should be [1.5941,4.7076].
In the likelihood ratio test, 11.0465 should be 11.0404.
In the Wald test, 3.1517 should be 1.1509 and 2.70895 should be 2.73335
In the Lagrange multiplier test, 7.9162 should be 7.9145 twice, the matrix should be [0.00995/0.26776,11.199] and the result of 15.687 should be 15.686. On the next page, the value 5.1182 should be 1.2653.
Further, using the 17 internal digits of the computation with no external rounding, the following shows the results
? Obtain Maximum Likelihood Estimates. create;logy=log(y)$ maximize;labels=beta,ro ; start = -2,2 ;fcn=-ro*log(beta+e)-lgm(ro)+(ro-1)*logy-y/(beta+e)$ +--------+--------------+----------------+--------+--------+ |Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| +--------+--------------+----------------+--------+--------+ |BETA | -4.71850 3.65680241 -1.290 .1969 | |RO | 3.15090** 1.23984807 2.541 .0110 | +--------+-------------------------------------------------+ | Note: ***, **, * = Significance at 1%, 5%, 10% level. | +----------------------------------------------------------+ ? Verify that first derivatives sum to zero calc;bm=b(1);rm=b(2)$ create;bi=1/(bm+e)$ create;gb=-rm*bi+y*bi*bi;gr=log(bi)-psi(rm)+logy$ calc ;list;sum(gb);sum(gr)$ Result = .000000 Result = .000000 ? Compute second derivatives and expected second derivatives. create ;gbb=rm*bi*bi-2*y*bi^3;grr=-psp(rm);grb=-bi$ create;egbb=-rm*bi^2$ namelist;g=gb,gr$ ? Sum second derivatives to obtain Hessian and expected Hessian calc;hbb=-sum(gbb);hrr=-sum(grr);hbr=-sum(grb);ehbb=-sum(egbb)$ +------------------------------------+ | Listed Calculator Results | +------------------------------------+ HBB = -.855704 HRR = -7.459184 HBR = -2.241969 EHBB = -.877926 matrix;list;opg=$ Matrix OPG has 2 rows and 2 columns. 1 2 +---------------------------- 1| 13.37220 -4.32174 2| -4.32174 1.53722 matr;h=[hbb/hbr,hrr];list;h=$ Matrix H has 2 rows and 2 columns. 1 2 +---------------------------- 1| 5.49914 -1.65285 2| -1.65285 .63085 matr;eh=[ehbb/hbr,hrr];list;eh=$ Matrix EH has 2 rows and 2 columns. 1 2 +---------------------------- 1| 4.90032 -1.47286 2| -1.47286 .57675 matrix ; zero=init(2,2,0)$ matrix ; bbb=[b/b/b] $ matrix ; vbbb=[h,zero,zero/zero,eh,zero/zero,zero,opg]$ matrix ; stat(bbb,vbbb)$ +--------+--------------+----------------+--------+--------+ |Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| +--------+--------------+----------------+--------+--------+ |BBB_1 | -4.71850** 2.34502540 -2.012 .0442 | |BBB_2 | 3.15090*** .79426173 3.967 .0001 | |BBB_3 | -4.71850** 2.21366581 -2.132 .0330 | |BBB_4 | 3.15090*** .75944322 4.149 .0000 | |BBB_5 | -4.71850 3.65680241 -1.290 .1969 | |BBB_6 | 3.15090** 1.23984807 2.541 .0110 | +--------+-------------------------------------------------+
When the second coefficient is fixed at 1.00000 the following results are obtained.
maximize;labels=beta,ro ; start = -2,1 ; fix = ro ;fcn=-ro*log(beta+e)-lgm(ro)+(ro-1)*logy-y/(beta+e)$ +--------+--------------+----------------+--------+--------+ |Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| +--------+--------------+----------------+--------+--------+ |BETA | 15.6027 10.0255472 1.556 .1196 | |RO | 1.00000 ......(Fixed Parameter)....... | +--------+-------------------------------------------------+ ? Verify that first derivative sums to zero calc;bm=b(1);rm=b(2)$ create;bi=1/(bm+e)$ create;gb=-rm*bi+y*bi*bi;gr=log(bi)-psi(rm)+logy$ calc ;list;sum(gb);sum(gr)$ Result = .000000 Result = 7.914484 ? Compute second derivatives and expected second derivatives. create ; gbb=rm*bi*bi-2*y*bi^3;grr=-psp(rm);grb=-bi ; egbb=-rm*bi^2 $ namelist;g=gb,gr$ ? Sum second derivatives to obtain Hessian and expected Hessian calc;list ; hbb=-sum(gbb);ehbb=-sum(egbb) ; opg=gb'gb$ HBB = .021662 EHBB = .022597 OPG = .009949 ? Estimates of standard error calc ;list;sqr(1/hbb) ; sqr(1/ehbb) ; sqr(1/opg)$ Result = 6.794363 (Actual second derivative) Result = 6.652414 (Expected second derivative) Result = 10.025547 (Sum of squares of first derivatives.)
(Professor William Greene, New York University and Xu Zhongmin, China.)

520 3/03/08

In the line above (16-42), "it's" should be "its." (Ryan Polansky, Arizona State University)

525 9/22/08

In the equation at the end of the page, the inverse matrix, [(1/2)Z'Z]^-1 should be inverted again. The end result is that the superscript -1 should be deleted. The computation should also be based only on the lower right submatrix of Z'Z. Assembling the parts, the equation at the end of the page should be
l_WALD = a^'{[0,I](2(Z'Z)^-1)[0,I]'}^-1a^ (Brian Gould, University of Wisconsin, Professor Pedro Bacao, University of Coimbra, Portugal.)

526 09/22/08

In the equation at the top of the page, the "2" should be "1/2" and the matrix should not be inverted - the superscript -1 should be deleted. (Professor Pedro Bacao, University of Coimbra, Portugal.)

527 03/28/09

In the equations at the end of the page, in the second, e_t=re_t+u_t should be e_t=re_t-1+u_t. (Dr. Nada Wasi, University of the Thai Chamber of Commerce, Bangkok.)

528 09/22/08

In (16-63), the term "-1/2ln(1-r²)" should be "+1/2ln(1-r²)." In the second line of the equation, in the denominator at the end of the expression, "s_n²" should be "s_u²." In the equation after (16-63), the numerator term "re₁" should be "re₁²." (Professor Pedro Bacao, University of Coimbra, Portugal.)

529 09/22/08

In Example 16.7, note that the autocorrelation coefficient is computed as (1-d/2) where d is the Durbin-Watson statistic in the regression. If one uses the familiar Box-Jenkins estimator for the least squares residuals, as for example on page 728, then the value will be 0.945761. (Professor Pedro Bacao, University of Coimbra, Portugal.)

548 09/22/08

In the equation for lnL_i in the 5th equation, "+T_ilnq" should be "-T_ilnq." (Professor Pedro Bacao, University of Coimbra, Portugal.)

549,550 09/22/08

In Table 16.9, the log likelihood function for the random effects MLE should be 249.25, not 2162.938. On page 550, the chi squared statistic will be 4297.57 rather than 8124.949. (Professor Pedro Bacao, University of Coimbra, Portugal.)

576 3/03/08

In the first line of Section 17.3, "proved" should be "proven." (Ryan Polansky, Arizona State University)

578 1/24/10

In the equation for H₅(37) at the end of the page, the result, 0.448, should be 0.488. (Serena Ng, Columbia University)

583 12/23/09

In footnote 7, also in footnote 44 on page 827 and in the references, on page 1105, The Borsch-Supan and Hajivassiliou citation should be 1993, not 1990. (Joachim Wilde, Department of Econometrics and Statistics, University of Osnabruck, Germany.)

587 03/06/10

Six lines below the equation, "strightforward" should be "straightforward.". (Professor Mary Beth Walker, Georgia State University.)

628 09/28/08

The regression in the middle of the paragraph is based on 202 observations, not 201. (Professor William Greene, New York University.)

636 09/22/08

In the last line of definition 19.1, z_t+k should be z_t+k-1. In the two equations in Definition 19.3, z_t+a should be z_t+a-1 both times and z_t+k+b should be z_t+k+b-1 both times. (Professor Pedro Bacao, University of Coimbra, Portugal.)

646 03/28/09

In the definition of the Q test at the end of the page, "T-2" should be "T+2." (Dr. Nada Wasi, University of the Thai Chamber of Commerce, Bangkok.)

649 03/28/09

Two lines below (16-31), the reference to 16.9.2.d should be to 16.9.2.b. (Dr. Nada Wasi, University of the Thai Chamber of Commerce, Bangkok.)

650,651 10/06/07

The results in the first row (OLS) in Table 19.2 are incorrect. The correct results (which also appear in Table 6.7 on page 126) are
+--------+--------------+----------------+ |Variable| Coefficient | Standard Error | +--------+--------------+----------------+ Constant| -26.6786858 2.00029667 LY | 1.62496242 .19517515 LPG | -.05392226 .04216064 LPNC | -.08343220 .17653370 LPUC | -.08467492 .10241732 T | -.01392611 .00477323
The R² for the regression is 0.9649324. The remaining results in the table are correct except for the standard error given for the ML estimate of rho. The value given is computed using T-1=51 rather than T=52 in the denominator. The correct value, to be consistent with (19-32) is 0.02816. The value of 0.861 for the AR(2) model is obtained as follows: (1) The OLS residuals are regressed on two lagged values (no constant). Leading observations are filled with zeros. The two slopes are 0.751941 and -0.022464 as reported. (2) All variables including the constant term are transformed, then the first two observations are dropped. (3) OLS using the 50 transformed observations produces the results in the table. (4) The FGLS coefficient vector is used to recompute the residuals for all 52 observations. (5) The 0.861 is the first autocorrelation coefficient. The new residuals are put in deviation form - call it du2. The lagged value, du2L is computed and the missing value is filled with a zero. The autocorrelation coefficient is du2L'du2/du2'du2. In the 7th line of text below the table, the critical value given for the chi squared distribution, 11.01, should be 11.07.

On page 651, the value 0.78750 given in the third line for the autocorrelation coefficient was computed using 1-d/2 where d is the Durbin-Watson statistic in (19-23), not using (19-19) as stated in the text. The motivation for this computation appears at the top of page 646. (Professor Randall Campbell, Mississippi State University and Professor Pedro Bacao, University of Coimbra, Portugal.)

650,651 10/06/07

The values of 0.292 and 1.416 for the experiment carried out just before Section 19.9.3 are obtained as follows:
sample;1-52$ regress;lhs=log_g;rhs=x;ar1;alg=p;maxit=1$ (Prais-Winsten) crea;et=log_g - x'b$ (Residuals) crea;ut=et-rho*et[-1]$ Transformed residuals) crea;ut1=ut[-1]$ Manually fill in the zeros in ut and ut1. calc;list;ru=ut1'ut/ut'ut;dwru=2*(1-ru)$$
RU = .292161 DWRU = 1.415678 (Professor Pedro Bacao, University of Coimbra, Portugal.)

654 09/28/08

At the end of 19.5, in the second to last line, 1/(1-r)² should be 1/(1-r²). In table 19.3, the variances of e and of u should be recomputed accordingly. For both the FE and RE models, .074687 should be 0.0302657. For the RE model, 0.074380 should be 0.030141. For the nonautocorrelation case, for the RE model, the 0.0875682 should be replaced by the square root of the difference between the OLS and FE estimates of s_e², which would be 0.0771064. Finally, we note, the estimated autocorrelation, 0.717897 is obtained by beginning the summation in the denominator at t=1 rather than t=2, and dividing the numerator of the estimator on page 653 by n(T-1) and the denominator by nT. (Professor Pedro Bacao, University of Coimbra, Portugal and Verena Dexheimer, Johannes-Gutenberg - University of Mainz, Germany.)

665 03/28/09

In the third line of the second full paragraph, the reference to (19-45) should be to (19-42). (Dr. Nada Wasi, University of the Thai Chamber of Commerce, Bangkok.)

679 09/28/08

In Example 2, for the two estimated equations, T=202, not 201. (Professor Pedro Bacao, University of Coimbra, Portugal.)

690 03/28/09

In the full paragraph, in the second and penultimate lines, the references to Chapter 21 should be to Chapter 22. (Dr. Nada Wasi, University of the Thai Chamber of Commerce, Bangkok.)

698 3/03/08

In the first line on the page, "of to" should be "of." (Ryan Polansky, Arizona State University)

709 3/03/08

In the second line on the page, "function" should be "functions." (Ryan Polansky, Arizona State University)

730 9/28/08

In the second line under figure 21.1, "obserations" should be "observations." (William Greene, New York University)

730,731 09/28/08

The two regression coefficients for the AR(2) process should be 1.47701 and -0.51553, not 1.497 and -0.4986. The process is still stationary. But, the PACF must be recomputed. The new values are shown below.
+--------+--------------+----------------+--------+--------+----------+ |Variable| Coefficient | Standard Error |t-ratio |P[|T|>t]| Mean of X| +--------+--------------+----------------+--------+--------+----------+ |DY1 | 1.47701*** .11328577 13.038 .0000 -.0152586| |DY2 | -.51553*** .11276262 -4.572 .0000 -.0099138| +--------+------------------------------------------------------------+ | Note: ***, **, * = Significance at 1%, 5%, 10% level. | +---------------------------------------------------------------------+ Time series identification for E Box-Pierce Statistic = 10.6480 Box-Ljung Statistic = 12.3380 Degrees of freedom = 14 Degrees of freedom = 14 Significance level = .7134 Significance level = .5792 * => |coefficient| > 2/sqrt(N) or > 95% significant. PACF is computed using Yule-Walker equations. ----+-------------------------------+-------+-------------------------------+ Lag | Autocorrelation Function |Box/Prc| Partial Autocorrelations | ----+------------------+------------+-------+------+------------+-----------+ 1 | .063 | |* | .23 | .063 | |* | 2 |-.119 | *| | 1.06 |-.133 | * | | 3 |-.235 | ***| | 4.27 |-.241 | *** | | 4 | .108 | |* | 4.95 | .142 | |** | 5 | .142 | |** | 6.11 | .113 | |* | 6 | .117 | |* | 6.91 | .108 | |* | 7 |-.091 | *| | 7.39 |-.047 | * | | 8 | .058 | |* | 7.58 | .189 | |** | 9 |-.167 | **| | 9.19 |-.321*| **** | | 10 | .034 | |* | 9.25 | .021 | |* | 11 |-.004 | *| | 9.25 | .043 | |* | 12 | .013 | |* | 9.26 |-.072 | * | | 13 |-.134 | *| | 10.31 |-.179 | ** | | 14 |-.076 | *| | 10.65 |-.114 | * | | ----+------------------+------------+-------+------+------------------------+
(Professor Pedro Bacao, University of Coimbra, Portugal.)

745 3/03/08

(1) In the 4th line, "questions" should be "question."
(2) In the first line of section 22.2.4, "of the of the" should be "of the."
(3) At the end of the page, DF_t should be DF_t. (Ryan Polansky, Arizona State University)

751 09/28/08

The second to last equation is missing a term bt. In addition, it follows from the construction that in this now corrected equation, g* equals g-1, the summation should run to p instead of p-1. The "where" and the last equation on the page are superfluous and should be deleted. (Professor Pedro Bacao, University of Coimbra, Portugal.)

753 09/22/08

In the equation for c_j at the top of the page, p must equal the L in the equation for a at the end of the list of equations. Thus, p is redundant and may be changed to L in the equation for c_j. (Professor Pedro Bacao, University of Coimbra, Portugal.)

754 10/13/08

Seven lines from the bottom of the page, "T=201" should be "T=202." The computations there are correct, and are based on T=202, not 201. (Houston Stokes, Department of Economics, University of Illinois, Chicago.)

755 4/30/08

In the first line of Example 22.4, the reference to Example 21.6 should be to 22.3. (Olle Olsson, Swedish University of Agricultural Sciences, Uppsala)

764 05/19/09

The cross reference to Section 22.4.2 in the 5th line of Section 22.4.2 should be to Section 22.3.2. (Professor Bruce Dixon, University of Arkansas.)

772 3/03/08

In the second line of the page, "regressionlike" should be "regression-like." (Ryan Polansky, Arizona State University)

774 09/22/08

In the second equation on the page (complementary log log model), "exp[exp" should be "exp[-exp." (Professor Pedro Bacao, University of Coimbra, Portugal.)

782 09/22/08

Under the two equations for PSI, "over the range of GRADE" should be "over the range of GPA." (Professor Pedro Bacao, University of Coimbra, Portugal.)

781,782 09/22/08

In Table 23.1, the values given for the Gumbel model are from a complementary log log model, not from a Gumbel model. Change "Gumbel" to "Complementary log log" in the second line of the last paragraph on page 781 and in the header line of Table 23.1 on page 782. The results for a Gumbel model are as follows:
+--------+--------------+----------------+--------+--------+ |Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| +--------+--------------+----------------+--------+--------+ |Constant| -7.14055** 3.20817669 -2.226 .0260 | |GPA | 1.58449* .89781077 1.765 .0776 | |TUCE | .06023 .12276520 .491 .6237 | |PSI | 1.61623** .81909913 1.973 .0485 | +--------+-------------------------------------------------+ Computation of Partial Effects Using the Delta Method +--------+--------------+----------------+--------+--------+ |Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| +--------+--------------+----------------+--------+--------+ |Scale | .36209*** .02373008 15.259 .0000 | |GPA | .57373* .33289382 1.723 .0848 | |TUCE | .02181 .04393503 .496 .6196 | |PSI | .58522** .28572006 2.048 .0405 | +--------+-------------------------------------------------+
(Professor Pedro Bacao, University of Coimbra, Portugal; recomputed by William Greene, New York University.)

784 09/22/08

In the last summation in the third equation on the page, and in the summation in the 5th equation, the upper index, "N" should be "n." (Professor Pedro Bacao, University of Coimbra, Portugal.)

785 09/22/08

In the first line of the page, the matrix G_i should be the matrix of derivatives of the term in the summation in the second equation on page 784. That is, G_i(b^) is the derivative of f(x_i'b^)b^ with respect to b^'. (Professor Pedro Bacao, University of Coimbra, Portugal.)

785 09/22/08

It might be noted that the values in parentheses in Table 23.3 are computed using the naive formula for the variance of the mean, which is the expression given for the "natural estimator" on the preceding page. (Professor Pedro Bacao, University of Coimbra, Portugal.)

790 09/22/08

Just above the header for Section 23.4.5, "99 percent" should be 95 percent.". (Professor Pedro Bacao, University of Coimbra, Portugal.)

790 09/22/08

In Table 23.4, the marginal effect for age and its standard error in the two models should be as shown below: The estimated probit models are as given in the text.
create;age=wa;age2=wa*wa$ create;kids=(kl6+k618)>0$ create;educ=we$ create;income=faminc/10000$ namelist;x=one,age,age2,income,educ,kids$ probit;quietly;lhs=lfp;rhs=x$ matrix;xb=mean(x)$ create;ab=xbr(age)$ wald;start=b;var=varb;labels=b1,b2,b3,b4,b5,b6 ;fn1=n01(xb'b1)*(b2+2*ab*b3)$ +--------+--------------+----------------+--------+--------+ |Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| +--------+--------------+----------------+--------+--------+ |Fncn(1) | -.00837*** .00282345 -2.966 .0030 | +--------+-------------------------------------------------+ probit;lhs=lfp;rhs=x;het;hfn=kids,income;mar$ calc;kb=xbr(kids);ib=xbr(income)$ wald;labels=b1,b2,b3,b4,b5,b6,c1,c2;start=b;var=varb ;fn1=n01(xb'b1/exp(c1*kb+c2*ib))*(b2+2*b3*ab)$ +--------+--------------+----------------+--------+--------+ |Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| +--------+--------------+----------------+--------+--------+ |Fncn(1) | -.00825 .00648692 -1.272 .2035 | +--------+-------------------------------------------------+
(Professor Pedro Bacao, University of Coimbra, Portugal.)

791 1/30/08

In the 5th line, "... dummy variables" should be "dummy variable" (William Greene (author), New York University)

797 3/03/08

In the third line of the page, delete the comma before "itself." (Ryan Polansky, Arizona State University)

802 03/11/08

In Table 23.5, the results for the Logit F.E.(C) estimator are incorrect. The correct values, in order, with standard errors in parentheses are: -6312.57 for lnL and 0.08384(0.0063820), -0.06521(0.15743), -0.07802(0.066186), -0.12179(0.054660), -0.04847(0.092639) for the coefficients. (Aurora Alonzo Anton, University of the Basque Country, Bilbao, Spain.)

824 11/06/07

In Table 23.13 and also in data Table F23.1 on page 1090, the variable names West and Midwest are interchanged. (The variables are correct in the original Burnett (1997) paper and in Greene (1998) referenced in the text.) (Markus Hertrich, University of Basel.)

827 12/08/08

In the second line below the first equation block at the top of the page, "that from the basis" should be "that form the basis." (William Greene, New York University.)

844 3/06/10

In the line before the second equation, "log-odds ratios" should be "log odds." (Profesor Uwe Jensen, Albrecht University)

845 1/13/08

In the equation for the partial derivative of d_ij with respect to a_l that appears in the middle of the page, there are two errors. First, the a_l in the denominator of the partial derivative should be a_l'. Second, the plus sign at the end of the equation before the last term should be a minus sign. (Professor Daniele Paserman, Department of Economics, Boston University.)

853 07/30/07

On page 853, in Table 23.23, row 5, column 2, the value 41.650 should be 41.657. In Table 23.24, in the first row, the value -0.15501 should be -0.015501. (The two errata were reported for the 5th edition, and were, unfortunately carried over to the 6th.) (Professor Curtis Simon, Clemson University.)

853 09/22/08

In Table 23.24, the standard errors for the model with choice based sampling should be as shown in the results below.
+--------+--------------+----------------+--------+--------+ |Variable| Coefficient | Standard Error |b/St.Er.|P[|Z|>z]| +--------+--------------+----------------+--------+--------+ |BG | -.01333*** .00491747 -2.711 .0067 | |BT | -.13405*** .02569720 -5.216 .0000 | |CH | -.00108 .01106360 -.097 .9225 | |AA | 6.59403*** 1.61807045 4.075 .0000 | |AT | 3.61895*** .83833382 4.317 .0000 | |AB | 3.32181*** .86915976 3.822 .0001 | +--------+-------------------------------------------------+
The Log-likelihood (sample shares) for the model with choice based sample weights should be -218.9929, obtained by fitting the model with only the choice specific constant terms.

In Table 23.25, in the third row, under Bus, "15(4) should be "15(14)," (Professor Pedro Bacao, University of Coimbra, Portugal.)

854 10/20/07

In Table 23.26, in the first column of figures, the values 0.194e-5, -0.46e-7 should be 0.194e-4 and -0.46e-6, respectively. In the second column, 0.000110, -0.0037 should be 0.000109, -0.0038, respectively. In the 5th column, -0.0000013, -0.000244, -0.000113 should be -0.000013, -0.00244, -0.00113, respectively. (Professor Koji Yamamoto, University of Tokyo.)

877 10/20/07

The reference to "Fin and Schmidt (1984)" should be to "Lin and Schmidt (1984)." The same correction is necessary on page 878 and in the references on page 1114. (Professor Steven Yen, University of Tennessee.)

882 3/03/08

In line five of footnote 18, "four surveys" should be "five surveys." (Ryan Polansky, Arizona State University)

884 10/27/09

In the second line of the 5 line equation at the end of the page, the term -w'g_i should be -w_i'g. (Hongtao Li, Department of Physics, University of Arizona.)

888 2/20/08

In table 24.3, in the last two (least squares) columns, the coefficient on b₅ and the standard error are interchanged. 0.318 (0.449) should be 0.449 (0.318). In the discussion of Example 24.8, in the second paragraph, third line, 0.149 should be 0.480. two lines later, -0.0113 should be -0.11003. (Aurora Alonso Anton, University of the Basque Country, Bilbao, Spain)

890 10/17/07

In equations (24-22), (24-23) and (24-24), in the definitions of the expectation functions, the 5 occurrences of z_i should all be w_i. (Professor Marco Cozzi, Queen's University)

891 3/03/08

In the line under the equation at mid page, delete the comma after "latter." (Ryan Polansky, Arizona State University)

898 07/12/08

The citation to Greene (1992) 4 lines above Section 24.5.8 should be deleted. Greene (1992) used the selection model of Boyes et al. (1989) that is discussed on page 896, not the approach on page 898. (Professor William Greene, New York University)

908 08/07/09

In the 4th line of the second paragraph, the x in the exponential function should have a subscript i. In equation (25-2), the prime on x in the leftmost bracketed term should be deleted. (Hai-Anh Dong, The World Bank.)

914 08/07/09

In the 5th line from the end of the page, the "=" after "lnL_i(NBP)" should be a minus sign. (Hai-Anh Dong, The World Bank.)

915 08/07/09

In the first line of the equation at the end of the page, inside the square brackets, the superscript "1" should be a prime. (Hai-Anh Dong, The World Bank.)

921 03/11/08

In Table 25.2, the nonzero results for the Poisson Fixed Effects Estimator are incorrect. The correct values, in order, with standard errors in parentheses are: -60421.58 for lnL and 0.0235735(0.1397), -0.05344(0.01638), -0.3348(0.04060), 0.006688(0.01530), -.09988(0.02964) for the coefficients. (Aurora Alonzo Anton, University of the Basque Country, Bilbao, Spain.)

922 3/03/08

In the second line after (25-12), "the sum" should be "they sum." (Ryan Polansky, Arizona State University)

959 3/06/10

In the line before the unnumbered expression for the 3x3 determinant, "the following shortcut" should read "the following shortcut due to P. Sarrus" (Profesor Uwe Jensen, Albrecht University)

965 05/13/09

In the matrix at the top of the page, the off diagonal zeros should be boldface. (Those elements are matrices, not scalars.) (Mark Strahan, Sand Hill Econometrics, Palo Alto.)

967 3/06/10

In the line after (A-77), "If c is any solution" should be if c is any nonzero solution" (Profesor Uwe Jensen, Albrecht University)

968 3/06/10

In the 4th line, "of a symmetric matrix" should read "of a symmetric matrix such as X'X" (Profesor Uwe Jensen, Albrecht University)

968 3/06/10

In the matrix expression for L, the third column (containing all zeros) should be deleted. (Profesor Uwe Jensen, Albrecht University)

982 1/28/09

In the 4th equation on the page, C_ji should be A_ij (Muhammed Abdullah Sahin, School of Business, University of Connecticut)

997 08/20/07

In the equation that defines W for the Wishart distribution, the second (x_i-m) should be followed by a prime. (Professor William Greene, author.)

1055, 1056, 1058 05/08/08

In Theorems D.21, D21A and D22, the condition given is insufficient. What is required is that g(z_n) be continuous and continuously differentiable, not involve n,and g'(m) must not be zero. In D.21A and D.22, these must hold for each of the functions. (Professor Stefan Huschens, TU Dresden.)

1064 09/22/08

Five lines below expression (E-1), "Sterling's" should be "Stirling's."
In the expression for the derivatives of the gamma function, lnP should be lnt. (Professor Pedro Bacao, University of Coimbra, Portugal.)

1076 10/28/08

In the second line of Section E3.7, "Example 16.21" should be "Example 16.16." (Lei Xu, School of Business and Economics, University of Exeter, UK.)

1076,1077 03/14/10

In equations (E-28), (E-29) and (E-32), the first large parenthetical expression is missing its exponent, (1-y_it). (William Greene, New York University, Stern School of Business.)

1081 10/06/07

In the header for Table F2.2, 36 should be 52. Note that the data file contains the gasoline expenditure variable. The variable G noted in the table is computed as indicated from the data in the file. (Randall Campbell, Mississippi State University.) In the list of variables it is stated that Pop is the U.S. population in millions. In the data file, the data are in thousands, not millions. (Peter Overland, Department of Economics, University of Texas at Dallas.)

1084 09/22/08

In the footnotes in Table F6.1, The data set is also used in Example 16.6. (Professor Pedro Bacao, University of Coimbra, Portugal.)

1084 09/17/07

In the header for Table F6.2, "901" should be "840.". (Professor Houston Stokes, University of Illinois at Chicago.)

1086 09/22/08

In the footnotes in Table F9.1, The data set is also used in Example 16.11. (Professor Pedro Bacao, University of Coimbra, Portugal.)

1086 09/22/08

In the footnotes in Table F9.1, The data set is also used in Example 16.8. (Professor Pedro Bacao, University of Coimbra, Portugal.)

1087 04/03/08

In the data table 10.1 on this website (the Newlove electricity data), in observation 42, the labor price given of 2.020 should be 2.050." (Professor Houston Stokes, University of Illinois at Chicago.)

1088 09/17/07

In the notes to Table F11.1, in the frequencies listed, 2158 should be 1079 and 987 should be 887. In the list of uses, 17.5 should be 17.6. (Professor Houston Stokes, University of Illinois at Chicago.)

1092 09/17/07

The list of variables in Table FC.1 is missing "E=Education." (Professor Houston Stokes, University of Illinois at Chicago.)

1107 10/7/09

The reference given for Steve Cecchetti's Journal of Econometrics paper on magazine prices is missing the year. It was published in 1986. (William Greene, New York University.)

1123 08/07/09

The following reference is missing:
Hilbe, J., Negative Binomial Regression, Cambridge University Press, Cambridge, 2007. (Hai-Anh Dong, The World Bank.)

1133 01/18/08

The Moulton (1986) reference is missing. The missing citation is Moulton, B. "Random group effects and the precision of regression estimates." Journal of Econometrics, 32, 3, pp. 385-397. (Professor Lonnie Magee, Department of Economics, McMaster University.)

1133 03/14/09

The two McKoskey references on page 1133 should be deleted. (They appear, correctly, on page 1132). (Alberg Zevelev, Federal Reserve Board.)