Computer Output of EViews, MINITAB, Excel, and STATA

In this appendix we show the computer output of EViews, MINITAB, Excel, and STATA,
which are some of the popularly used statistical packages for regression and related statistical
routines. We use the data given in Table E.1 from the textbook website to illustrate the output
of these packages. Table E.1 gives data on the civilian labour force participation rate
(CLFPR), the civilian unemployment rate (CUNR), and real average hourly earnings in 1982
dollars (AHE82) for the U.S. economy for the period 1980 to 2002.
Although in many respects the basic regression output is similar in all these packages, there
are differences in how they present their results. Some packages give results to several digits,
whereas some others approximate them to four or five digits. Some packages give analysis of
variance (ANOVA) tables directly, whereas for some other packages they need to be derived.
There are also differences in some of the summary statistics presented by the various
packages. It is beyond the scope of this appendix to enumerate all the differences in these
statistical packages. You can consult the websites of these packages for further information.
E.1 EViews
Using Version 6 of EViews, we regressed CLFPR on CUNR and AHE82 and obtained the
results shown in Figure E.1.
This is the standard format in which EViews results are presented. The first part of this figure
gives the regression coefficients, their estimated standard errors, the t values under the null
hypothesis that the corresponding population values of these coefficients are zero, and the p
values of these t values. This is followed by R2 and adjusted R2. The other summary output in
the first part relates to the standard error of the regression, residual sum of squares (RSS), and
the F value to test the hypothesis that the (true) values of all the slope coefficients are
simultaneously equal to zero. Akaike information and Schwartz criteria are often used to
choose between competing models. The lower the value of these criteria, the better the model
is. The method of maximum likelihood (ML) is an alternative to the method of least squares.
Just as in OLS we find those estimators that minimise the error sum of squares, in ML we try
to find those estimators that maximize the possibility of observing the sample at hand. Under
the normality assumption of the error term, OLS and ML give identical estimates of the
regression coefficients. The Durbin–Watson statistic is used to find out if there is first-order
serial correlation in the error terms.
The second part of the EViews output gives the actual and fitted values of the dependent
variable and the difference between the two, which represent the residuals. These residuals
are plotted alongside this output with a vertical line denoting zero. Points to the right of the
vertical line are positive residuals and those to the left represent negative residuals.
The third part of the output gives the histogram of the residuals along with their summary
statistics. It gives the Jarque–Bera (JB) statistic to test for the normality of the error terms and
also gives the probability of obtaining the stated statistics. The higher the probability of
obtaining the observed JB statistic, the greater is the evidence in favour of the null hypothesis
that the error terms are normally distributed.
Note that EViews does not give directly the analysis-of-variance (ANOVA) table, but it can
be constructed easily from the data on the residual sum of squares, the total sum of squares

Page 1 of 6
(which will have to be derived from the standard deviation of the dependent variable), and
their associated degrees of freedom. The F value given from this exercise should be equal to
the F value reported in the first part of the table.
FIGURE E.1: EViews output of civilian labour force participation
regression.
Dependent Variable: CLFPR
Method: Least Squares
Sample: 1980–2002
Included observations: 23
Variable Coefficient Std. error t-Statistic Prob.
C 80.90133 4.756195 17.00967 0
CUNR 20.671348 0.08272 28.115928 0
AHE82 21.404244 0.608615 22.307278 0.0319

R-squared 0.77277 Mean dependent var 65.8957

Adjusted R-squared 0.75004 S.D. dependent var 1.16871
S.E. of regression 0.58431 Akaike info criterion 1.88433
Sum squared resid 6.82831 Schwarz criterion 2.03244
Log likelihood -18.67 F-statistic 34.0073
Durbin–Watson stat 0.78763 Prob(F-statistic) 0

Obs. Actual Fitted Residuals

1980 63.8 65.2097 -1.40974
1981 63.9 65.0004 -1.10044
1982 640. 63.6047 0.39535
1983 64.0 63.5173 0.48268
1984 64.4 64.9131 -0.51311
1985 64.8 65.1566 -0.35664
1986 65.3 65.2347 0.06526
1987 65.9 65.8842 -0.28416
1988 65.5 66.4103 -0.51027
1989 66.5 66.6148 -0.11476
1990 66.5 66.5819 -0.08186
1991 66.2 65.8745 0.32546
1992 66.4 65.4608 0.93923
1993 66.3 65.8917 0.40834
1994 66.6 66.4147 0.18530
1995 66.6 66.7644 -0.16441
1996 66.8 66.8425 -0.04251
1997 67.1 67.0097 0.09032
1998 67.1 66.9974 0.10263
1999 67.2 67.0443 0.05569
2000 67.2 67.1364 0.06355
2001 56.9 66.4589 0.44105

Page 2 of 6
 2002 66.6 65.577 1.02304

E.2 MINITAB
Using Version 15 of MINITAB, and using the same data, we obtained the regression results
shown in Figure E.2.
MINITAB first reports the estimated multiple regression. This is followed by a list of
predictors (i.e., explanatory) variables, the estimated regression coefficients, their standard
errors, the T (= t) values, and the p values. In this output S represents the standard error of the
estimate, and R2 and adjusted R2 values are given in percent form.
This is followed by the usual ANOVA table. One characteristic feature of the ANOVA table is
that it breaks down the regression, or explained, sum of squares among predictors. Thus, of
the total regression, sum of squares of 23.226, the share of CUNR is 21.404 and that of
AHE82 is 1.822, suggesting that relatively, CUNR has more impact on CLFPR than AHE82.
A unique feature of the MINITAB regression output is that it reports “unusual” observations;
that is, observations that are somehow different from the rest of the observations in the
sample. We have a hint of this in the residual graph given in the EViews output, for it shows
that the observations 1 and 23 are substantially away from the zero line shown there.

Page 3 of 6
MINITAB also produces a residual graph similar to the EViews residual graph. The St Resid
in this output is the standardised residuals; that is, residuals divided by S, the standard error of
the estimate.
Like EViews, MINITAB also reports the Durbin–Watson statistic and gives the histogram of
residuals. The histogram is a visual picture. If its shape resembles the normal distribution, the
residuals are perhaps normally distributed. The normal probability plot accomplishes the
same purpose. If the estimated residuals lie approximately on a straight line, we can say that
they are normally distributed. The Anderson–Darling (AD) statistic, an adjunct of the normal
probability plot, tests the hypothesis that the variable under consideration (here residuals) is
normally distributed. If the p value of the calculated AD statistic is reasonably high, say in
excess of 0.10, we can conclude that the variable is normally distributed. In our example the
AD statistic has a value of 0.481 with a p value of about 0.21 or 21 percent. So, we can
conclude that the residuals obtained from the regression model are normally distributed.
Regression Analysis: CLFPR versus CUNR, AHE82
The regression equation is
CLFPR = 81.0 - 0.672 CUNR - 1.41 AHE82
Predictor Coef SE Coef T P
Constant 80.951 4.770 16.97 0.0
CUNR - 0.67163 0.08270 - 8.12 0.0
AHE82 - 1.4104 0.6103 - 2.31 0.032

S = 0.584117 R-Sq = 77.3% R-Sq(adj) = 75.0%

Analysis of Variance

Source DF SS MS F P
Regression 2 23.226 11.613 34.04 0.0
Residual Error 20 6.824 0.341
Total 22 30.050

Source DF Seq SS
CUNR 1 21.404
AHE82 1 1.822

Unusual Observations
Obs CUNR CLFPR Fit SE Fit Residual St Resid
2 7.10 63.80 65.209 0.155 -1.409 -2.50R
23 5.80 66.60 65.575 0.307 1.025 2.06R

R = an observation with a large standardised residual.

Durbin–Watson statistic 5 0.787065

Page 4 of 6
E3. Excel
Using Microsoft Excel, we obtained the regression output shown in Table E.2.
Excel first presents summary statistics, such as R2, multiple R, which is the (positive) square
root of R, adjusted R2, and the standard error of the estimate. Then it presents the ANOVA
table. After that it presents the estimated coefficients, their standard errors, the t values of the
estimated coefficients and their p values. It also gives the actual and estimated values of the
dependent variable and the residual graph as well as the normal probability plot.
A unique feature of Excel is that it gives the 95 percent (or any specified percent) confidence
interval for the true values of the estimated coefficients. Thus, the estimated value of the
coefficient of CUNR is -0.671631 and the confidence interval for the true value of CUNR
coefficient is (-0.84415 to -0.499112). This information is very valuable for hypothesis
testing.

TABLE E.2: Excel Output of Civilian Labor Force Participation Rate

ANOVA
df SS MS F Significance F
Regression 2 23.2257 11.6129 34.0361 3.65E-07
Residual 20 6.82385 0.34119
Total 22 30.0496

Summary Output
Regression Statistics
Multiple R 0.879155
R Square 0.772914
Adjusted R 0.750205
Standard E 0.584117
Observation 23

Coefficient Standard Err t Stat p-value Lower 95% Upper 95%

Intercept 80.95122 4.770337 16.96971 2.42E-13 71.00047 90.90196
CUNR -0.671631 0.082705 0.082705 9.24E-08 -0.84415 -0.499112
AHE82 -1.410432 0.610348 -2.310867 0.031626 -2.683594 -0.13727

Page 5 of 6
E.4 STATA
Using STATA, we obtained the regression results shown in Table E.3.
Stata first presents the analysis of variance table along with the summary statistics such as R2,
adjusted R2, and the root mean-squared-error (MSE), which is just the standard error of the
regression.
Then it gives the values of the estimated coefficients, their standard errors, their t values, the
p values of the t statistics, and the 95 percent confidence interval for each of the regression
coefficients, which is similar to the Excel output.
E.5 Concluding Comments
We have given just the basic output of these packages for our example. But it may be noted
that packages such as EViews and STATA are very comprehensive and contain many of the
econometric techniques discussed in this text. Once you know how to access these packages,
running various subroutines is a matter of practice. If you wish to pursue econometrics
further, you may want to buy one or more of these packages.

Page 6 of 6