100% found this document useful (1 vote)
568 views8.essentials of Econometrics Chapter1-8 PDF
Uploaded by
Druthi KolaCopyright
© © All Rights Reserved
Available Formats
Download as PDF or read online on Scribd
100% found this document useful (1 vote)
568 views8.essentials of Econometrics Chapter1-8 PDF
Uploaded by
Druthi KolaCopyright
© © All Rights Reserved
Available Formats
Download as PDF or read online on Scribd
/ 248
ESSENTIALS OF
ECONOMETRICS.
THIRD EDITION
Damodar N. Gujarati
United States Military Academy, West Point
EM McGraw ye 3 3k
ia Tirwin +
Boston Burr Ridge, IL Dubuque, 1A Madison, Wi New York
‘San Francisco St.Louis Bangkok Bogoté Caracas Kuala Lumpur
Lisbon London Madrid MexicoCity Milan Montreal New Delhi
Santiago Seoul Singapore Sydney Taipei Toronto CHAPTER
THE NATURE
AND SCOPE
OF ECONOMETRICS
Research in economics, finance, management, marketing, and related disci-
plines is becoming increasingly quantitative. Beginning students in these fields
are encouraged, if not required, to take a course or two in econometrics—a field
of study that has become quite popular. This chapter gives the beginner an
overview of what econometrics is all about.
1.1 WHAT IS ECONOMETRICS?
Simply stated, econometrics means economic measurement. Although quan-
titative measurement of economic concepts such as the gross domestic prod-
uct (GDP), unemployment, inflation, imports, and exports is very important,
the scope of econometrics is much broader, as can be seen from the following
definitions:
Econometrics may be defined as the social science in which the tools of economic the-
‘ory, mathematics, and statistical inference ase applied to the analysis of economic
phenomena.?
Econometrics, the result of a certain outlook on the role of economics, consists of the
application of mathematical statistics to economic data to lend empirical support to
the models constructed by mathematical economics and to obtain auumerical results
Yarthur 8. Goldberger, Econometric Theory, Wiley, New York, 1964,
°P. A. Samuelson, TC. Koopmans, and J, R.N. Stone, “Report of the Evaluetive Committee for
Etonometrica," Econometrica, vol. 22, 0.2, Apri 1954, pp. 141-146, CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS.
2 WHY STUDY ECONOMETRICS?
maa
2
[As the preceding definitions suggest, econometrics makes use of economic the-
ory, ‘mathematical economics, economic statistics (i., economic data), and
mathesnatical statistics. Yet, it is a subject that deserves to be studied in its own
rright for the following reasons.
Economic theory makes statements or hypotheses that are mostly qualitative
in nature. For example, microeconomic theory states thi, other things remain-
ing the same (the famous ceteris paribus clause of economics), an increase in the
price of a commodity is expected to decrease the quantity demanded of that
commodity. Thus, economic theory postulates a negative or inverse relation-
ship between the price and quantity demanded of a commodity—this is the
widely known law of downward-stoping demand or simply the law of demand.
But the theory itself does not provide any numerical measure of the strength of
the relationship between the two; that is, it does not tell by how much the quan-
tity demanded will go up or down as a result of a certain change in the price of
the commodity. It is the econometrician’s job to provide such numerical esti-
mates. Econometrics gives empirical (ie., based on observation or experiment)
content to most economic theory. If we find in a study or experiment that when
the price of a unit increases by a dollar and the quantity demanded goes down
by, say, 100 units, we have not only confirmed the law of demand, but in the
process we have also provided a numerical estimate of the relationship between
“tio two variables—price and quantity.
"he main concern of mathematical economics is to express economic theory
in mathematical form or equations (or models) without regard to measurability
ov evapitical verification of the theory. Econometrics, as noted earlier is primar-
ily interested in the empirical verification of economic theory. As we will show
shortly, the econometrician often uses mathematical models proposed by the
mathematical economist but puts these models in forms that lend uiemselves to
empirical testing.
Economic statistics is mainly concerned with collecting, processing, and pre-
senting economic data in the form of charts, diagrams, and tables. This is the
economic statistician’s job. He or she collects data on the GDP, employment, un-
employment, prices, etc. This data constitutes the raw data for econometric
work. But the economic statistician does not go any further because he or she is.
not primarily concerned with using the collected data to test economic theories.
Although mathematical statistics provides many of the'tools employed in the
trade, the econometrician often needs special methods because of the unique
nature of most economic data, namely, that the data are not usuelly generated
as the result of a controlled experiment. The econometrician, like the meteorol-
ogist, generally depends on data that cannot be controlled directly. Thus, data
on consumption, income, investments, savings, prices, etc,, which are colllected
by public and private agencies, are nonexperimentai in nature. The econometri-
cian takes these data as given. This creates special problems not normally dealt
with in mathematical statistics. Moreover, such data are likely to contain errors
of measurement, of either omission or commission, and the econometrician CHAPTER ONE: ‘THE NATURE AND-SCOPE OF ECONOMETRICS 3
may be called upon to develop spetial methods of analysis to.deal with such
errors of measurement,
For students majoring in economics and business there is a pragmatic reason
for studying econometrics. After graduation, in their employment, they may be
called upon to forecast sales, interest rates, and money supply or to estimate de-
mand and supply.functions or price elasticities for products. Quite often, econo-
mists appear as expert witnesses before federal and state regulatory agencies on
behalf of their clients or the public at lange. Thus, an economist appearing before
asstate regulatory commission that controls prices of gas and electricity may bere-
quired to assess the impact of a proposed price increase on the quantity de-
marided of electricity before the comunission will approve the price increase, In
situations like this the economist may need to develop a demand function for
electricity for this purpose. Such a demand function may enable the economist to
estimate the price elasticity of demand; thats, the percentage change in the quan-
tity demanded for a percentage change in the price. Knowledge of econometrics
is very helpful in estimating such demand functions,
It is fair to say that econometrics has become an integral part of training in
economics arid business.
1.3. THE METHODOLOGY OF ECONOMETRICS
How does one actually do-an econometric study? Broadly speaking, economet-
sic analysis proceeds along the following fines.
1. Creating a statement of theory or hypothesis,
2, Collecting data.
3. Specifying the mathematical model of theory
4, Specifying the statistical, or econometric, model of theory.
5, Estimating the parameters ofthe chosen econometric model.
6. Checking for model adequacy: Model specification testing.
7. Testing the hypothesis derived from the model.
8, Using the model for prediction or forecasting,
Tp illustrate the methodology, consider this question: Do economic condi-
tions affect people's decisions to enter the labor force, that is, their willingness
to work? As a measure of economic conditions, suppose We use the unemploy-
intent rate (UNR), and asa measure of labor force participation we use the labor
force patticipation rate (LPR). Data on UNR and LEPR are tegularly published
by the government. So to answer the question we proceed as follows.
Creating a Statement of Theory or Hypothesis
‘The starting point isto find out what economic theory has to say on the subject
you want to study In labor economics, there are two rival hypotheses about the
effect of economic conditions on people's willingness to work, The discouraged
worker hypothesis (effect) states that when economic conditions worsen, as CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS
oliecting Data
reflected in a higher unemployment rate, many unemployed workers give up
hope of finding a job ard drop out of the labor force. On the other hand, the
added-worker hypothesis (effect) maintains that when economic conditions
worsen, inany secondary workers who are not currently in the labor market
(eg, mothers with children) may decide to join the labor force if the main
breadwinner in the family loses his or her job. Even if the jobs these secondary
workers get are low paying, the earnings will make up some of the loss in in-
come suffered by the primary breadwinner.
Whether, on balance, the labor force participation rate will increase or decrease
will depend on the relative strengths of the added-worker and discouraged-
worker effects, If the added-worker effect dominates, LFPR will increase even
when thé unemployment rate shigh. Contratily ifthe discouraged-worker effect
dominates, LFPR will decrease. How do we find this out? This now becomes our
empirical question,
For empirical purposes, therefore, we need quantitative information on the {wo
variables. There are three types of data that are generally available for empirical
analysis,
4. Time series,
2. Cross-sectional.
3. Pooled (a combinatic:: of time series and cross-sectional).
‘Times series data are colle-ted over a period of time, such as the data on
GDP, employment, unemployment, money supply, or government deficits.
Such data may be collected at regular intervals—daily (e.g,, stock prices),
weekly (e.g, money supply); monthly (e.g,, the unemployment rate), quarterly
(eg, GDP), or annually (e.g,, government budget). These data may be quanti-
tative in nature (e.g, prices, income, money supply) or qualitative (e.g., male or
female, employed ot unemployed, married or unmarried, white or black). As
wwe will show, qualitative variabies, also called dummy or categorical variables,
can be every bit as important as quantitative variables,
Cross-sectional data are data on one or more variables collected at one point
in time, such as the census of population conducted by the U.S. Census Bureau
every 10 years (the most recent was.on April 1, 2000); the surveys of consumer
expenditures conducted by the University of Michigan; and the opinion polls
suich.as those conducted by Gallup, Haris, and other polling organizations.
In pooled data we have elements of both time series and cross-sectional data.
For example, if we were to collect data on the unemployment rate for 10 coun-
tries for a period of 20 years, the data will constitute an example of pooled data—
data on the unemployment rate for each country for the 20-year period will form.
“time series data, whereas data on the unemployment rate for the 10 countries for
any single year will be cross-sectional data, In pooled data we will have 200
observations—20 annual observations for each of the 10 countries. CHAPTER ONE: THE NATURE AND SCOPE OF ECONOWETRICS 5
‘There is a special type of pooled data, panel data, also called longitudinal or
micropanel data, in which the same cross-sectional unit, say, a family or firm, is
surveyed over time, For example, the US. Department of Commerce conducts
a census of housing at periodic intervals. At each periodic survey the same
household (or the people living at the seme address) is interviewed to find out
if there has been any change in the housing and financial conditions of that
household since the last survey. The panel data that results from repeatedly in-
terviewing the same household at periodic intervals provide very useful infor-
mation on the dynamics of household behavior.
Sources of Data A word is in order regarding data sources, The success of
“any econometric study hinges on the’ quality, as well as the quantity, of data.
Fortunately, the Internet has opened up a veritable wealth of data. In Appendix A1
we give addresses of several Web sites that have all kinds of microeconomicand
macroeconomic data. Students should be familiar with such sources of data, as
well as how to access or download them. Of course, these data aré continually
updated so the reader may find the latest available data.
For our analysis, we obtained the time series data shown in Table 1-1. This
table gives data on the civilian labor force participation rate (CLEPR) and the
civilian unemployment rate (CUNR), defined as the number of civilians unem-
ployed as a percentage of the civilian labor force, for the United States for the
period 1980-20025
Unlike physical sciences, most data collected in economics (e.g., GDP, money
supply, DowJones index, car sales) are nonexperimental in that the data-
collecting agency (e.g,, government) may not have any direct control over the
data, Thus, the data on labor force participation and unentployment are based on
the information provided to the government by participants in the labor market.
Ina sense, the government is a passive collector of these data and may not be
aware of the added- or discouraged-worker hypotheses, or any other hypothesis,
for that matter. Therefore, the collected data may be the result of several factors
affecting the labor force participation decision made by the individual person.
Thatis, the same data may be compatible with more than one theory.
Specttying the Mathematical Model of Labor Force Participation
‘To see how CLEPR behaves in relation to CUNR, the first thing we should do is
plot the data for these variables in a scatter diagram, or scattergram, as shown.
in Figure 1-1.
‘The scattergram shows that CLEPR and CUNR are inversely related, perhaps
suggesting that, on balance, the discouraged-worker effect is stronger than the
added-worker effect! As a first approximation, we can draw. a straight line
2We consider here only the aggregate CLFPR and CUNR, but data ane available by age, sex and
ethnic composition.
Qn this, see Shelly Lundberg, “The Added Worker Effect,” Journal of Labor Economics, vol. 3,
January 1985, pp. 1-37 CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS
TABLE 1-1 U.S. CIVILIAN LABOR FORCE PARTICIPATION RATE
(CLFPR), CIVILIAN UNEMPLOYMENT RATE (CUNR),
‘AND REAL AVERAGE HOURLY EARNINGS (AHEB2)*
FOR THE YEARS 1980-2002. .
Year CLFPR (%) CUNR(%) —-AHEB2 ($)
1980 638 7A 778
198 63.9 76 7.69
1982 64.0 a7 7.68
1983 64.0 96 779
1984 e44 75 7.80
1985 648. 72 77
1986 65.3 70 781
1987 658 62 773
1988, 659 55 7.69
1989 665 53 764
1990 665 56 752
1991 662 68 745
1992 664 75 7a
1998 663 69 7.39
1994 666 64 7.40
1995 668 56 7.49
1996 66.8 54 7.43,
1997 era 49 7.58
1998 era 45 795
1999 674 42 7.88
+2000 672 49 7.29
2001 669 48 7.99
2002 668 58 cary
“AHEB2 represents averdge houty eamings in 1982 dolars.
‘Source: Economie Repor ofthe Proident, 2002, CLEPR tom
Table 6-40, CUNA fom Table B49, and AHES? fom Table B47,
Data for 2002 are from more recent govemment publications.
Filed Line Plot
os
670
665
Roo
ge:
dao
os
640
65
Cr rT)
6 7
CUNR (%)
FIGURE 1-1 Regression plot for chill labor force participation CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS 7
through the scatter points and write.the relationship between CLFPR and
CUNR by the following simple mathematical model:
CLFPR = B, + B,CUNR (1.1)
Equation (1.1) states that CLFPR is linearly related to CUNR. By and By are
known as the parametets of the linear function B is also known as the inter-
cept it gives the value of CLEPR when CUINR is zero And B; is known as the
slope. The slope measures the rate of change in CLFPR fora unit change in CUNR, or
‘more generally the rate of change in the value of the variable on the left-hand
side of the equation for a unit change in the value of the variable on the right-
hand side, The slope coefficient By can be positive (if the added-worker effect
dominates the discouraged-worker effect) or negative (if the discouraged
worker effect dominates the added-worker effect), Figure 1-1 suggests that in
the present case it is negative.
Specifying the Statistical, or Econometric,
Model of Labor Force Participation
The purely mathematical model of the relationship between CLFPR and CUNR
given in Eq, (1.1), although of prime interest to the mathematical economist, is
of limited appeal to the econometrician, for such a model assumes an exact, or
deterministic, relationship between the two variables; that is, fox a given CUNR,
there is a unique value of CLFPR. In reality, one tarely finds such neat relation-
ships between economic variables. Most often, the relationships are inexact, or
statistical, in nature.
‘This is seen clearly from the scattergram given in Figure 1-1. Although the two
variables are inversely related, the relationship between them is not perfectly or
exactly linear, for if we draw a straight line through the 23 data points, not all the
data points will lie exactly on that straight line. Recall that to draw astraight line
‘weneed only two points.’ Why don’t the 23 data points lie exactly on the straight
line specified by the mathematical model, Eq, (1.1)? Remember that our data on
labor force and unemployment are nonexperimentally collected. Therefore, as
noted earlier, besides the added- and discouraged-worker hypotheses, there
may be other forces affecting labor force participation decisions, As a result, the
observed relationship betweeri CLFPR and CUNR is likely to be imprecise.
Let us allow for the influence of all other variables affecting CLFPR in a
catchall variable u and write Eq, (1.1) as follows:
CLEPR = By + ByCUNR+u (1.2)
SBroadly speeking, a paremeter is an unknown quantity that may vary over a certain set of val-
ues. In statistics a probability distribution function (PDE) of a random variable is often character-
ied by its parameters, such as its mean and variance. This topic is discussed in greater detail in
ters and 3,
in. Chapter 6 we give a more precise interpretation of the intercept in the context of regression
a
"he even ted to ft parabola othe ster points given in Fig 1, but the ress were not CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS
where u represents the random error term, or simply the error term. We let #
represent all those forces (besides CLINR) that affect CLEPR but are not explic-
itly introduced in the model, as well as purely random forces. As we will see in
Paxt Ul, the ertor term distinguishes econometrics from purely mathematical
economics.
Equation (1.2) is an example of a statistical, or empirical or econometric, mode
More precisely, itis an example of what is known asa linear regression model,
| which isa prime subject of thisbook. In such a model, the variable appearing on
the left-hand side of the equation is called the dependent variable, and the vari-
able on the right-hand side is called the indépendent, or explanatory, variable.
In linear regression analysis oisr primary objective is to explain the behavior of
one variable (the dependent variable) in relation to the behavior of one or more
other variables (the explanatory variables), allowing for the fact that the rela-
tionship between them is inexact.
Notice that the econometric model, Eq. (1.2), is derived from the mathemati-
cal model, Eq. (1.1), which shows that mathemtatical economics and economet-
rics are mutually complementary disciplines. This is clearly reflected in the de-
finition of econometrics given at the outset.
Before proceeding further, a warning regarding causation is in order. In the
regression model, Eq, (1.2), we have stated that CLFPR is the dependent vari-
able and CUNRis theindependent, or explanatory, variable, Does that mean that
the two variables are causally related; that is, is CUINR the cause and CLEPR the
effect? In other words, does regression imply causatioa? Not xecessarily. As
Kendall and Stuart note, “A statistical relationship, however strong and however
suggestive, can never establish causal connection: out ideas of c:::sation must
come from outside statistics, ultimately from some theory or other.”8 In our ex-
ample, itis up to economic theory (e.g,, the discouraged-worker hypothesis) to
establish the cause-and-effect relationship, if any, between the dependent and
explanatory variables. If causality cannot be established, itis better to call the re-
lationship, Eq. (1.2), predictive relationship: Given CUINR, can we predict CLFPR?
stimating the Parameters of the Chosen Econometric Model
Given the data on CLEPR and CUNR, such as that in Table 1-1, how do we esti-
mate the parameters of the model, Eq, (1.2), namely, By and By? That is, how do
we find the numerical values (ie,, estimates) of these parameters? This will be
the focus of our attention in Part I, where we develop the appropriate methods
of computation, especially the method of ordinary least spares (OLS). Using OLS
and the data given in Table 1.1, we obtairied the following results:
CLEDR = 69,9963 ~ 0.6513CUNR 3)
‘In statistical lingo, the random error term is known as the stochastic error term.
°M. G. Kendall and A, Stuart, The Advanced Theory of Sttistcs, Charles Grifin Publishers, New
York, 1961, vol.2, Chap. 26,p. 279 CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS 9
Note that we have put the symbol «on CLFPR (read as “CLFPR hat”) to remind
us that Eq. (1.3) is an estimate of Bq, (1.2). The estimated regression line is shown
in Figure 1-1, along with the actual data points,
‘ASEq, (1.3) shows, the estimated value of By is ~ 70 and that of By is * - 0.65,
where the symbol © means approximately. Thus, if the unemployment rate
goes up by one unit (ie., one percentage point, ceteris paribus, CLEPR is ex-
pected to decrease on the average by about 0.65 percentage points; that is, as eco-
nomic conditions worsen, on average, there is a net decrease in the labor force
participation rate of about 0.65 percentage points, perhaps suggesting that the
discouraged-warker effect dominates. We say “on the average” because the
presence of the error term w, as noted eatlier, is likely to make the relationship
somewhat imprecise. This is vividly seen in Figure 1-1 where the points not on
the estimated regression line are the actual participation rates and the (vertical)
distance between them and the points on the regression line are the estimated
1's. As we will see in Chapter 6, the estimated n’s are called residuals. In short,
the estimated regression line, Eq: (1.3), gives the relationship between average
CLEPR and CUNK; thatis, on average how CLFPR responds to a unit change in
CUNR. The value of about 69.9963 suggests that the average value of CLFPR
will be about 69.99 percent ifthe CUNR were zer0; that is, about 69,99 percent
of the civilian working-age population would participate in the labor force if
there were full employment (i.e,, zeto unemployment).
Checking for Model Adequacy: Model Specification Testing
How adequate is out model, Eq, (1.3)? Itis true thata person will take into account
labor market conditions as measured by, say, the unemployment rate before
entering the labor market, For example, in 1982 (a recession year) the civilian un-
employment rate was about 9.7 percent. Compared to that, in 2001 it was only
4.8 percent. A persons more likely to be discouraged from entering the labor mat
ket when the unemployment rate is9 percent than when itis 5 percent. But there
are other factors that also enter into labor force participation decisions. For exam-
ple, hourly wages, or earnings, prevailing in the labor market also will be an im-
portant decision variable. In the short-run at least, a higher wage may attractmore
workers to the labor market, other things remaining the same (ceteris paribus). to
see its importance, in Table 1-1 we have also given data on real average hourly
‘earnings (AHE82), where real earnings are measured in 1982 dollars. To take into
o account the influence of AHE82, we now consider the following model:
CLFPR = B; + B)CUNR + BsAHE82 + u (1.4)
Equation (1.4) is an example of a multiple linear regression model, in contrast to
Eq. (1.2), which is an example of asinipe (txvo-variable or bivariate) linear regression
This is, however, a mechanical interpretation of the intercept. We will see in Chapter 6 how to
interpret the intercept term meaningfully ina given context. 0° CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS
‘model: In the two-variable model there is a single explanatory variable, whereas
in a multiple regression there are several, or multiple, explanatory variables.
Notice that in the multiple regression, Eq. (1.4), we also have included the error
term, u, for no matter how many explanatory variables one introduces in the
model, one cannot fully explain the behavior of the dependent variable, How
many variables one introduces in the multiple regression is a decision that the
researcher will have to make in a given situation. Of course, the underlying
economic theory will often tell what these variables might be. However, keep in
mind the warning given earlier that regression does not mean causation, the
relevant theory must determine whether one or more explanatory variables are
causally related to the dependent variable.
How do we estimate the parameters of the multiple regression, Eq, (1.4)? We
‘over this topic in Chapter 8, after we discuss the two-variable model in
Chapters 6 and 7. We consider the two-variable case first because it is the build-
ing block of the multiple regression model. As we shall see in Chapter 8, the
multiple regression model is in many ways a stcaightforward extension of the
two-variable model.
For our illustrative example, the empirical counterpart of Eq, (1.4) is as fol
lows (these resulls are based on OLS):
CLEPR 80.9013 — 0.6713CUNR — 1.4042AHE82, (15)
‘These results are interesting because both the slope coefficients are negative
‘The negative coefficient of CLINR suggests that, ceteris paribils (ie, holding the
influence of AHES82 constant), a one-percentage-point increase in the unem-
ployment rate leads, on average, to about a 0.67-percentage-point decrease in
CLFPR, pethaps once again supporting the discouraged-worker hypothesis. On
the other hand, holding the influence of CLINR constant, an increase in real av-
erage hourly earnings of one dollar, on average, leads to about a 1.40 petcentage-
point decline in CLEPR." Does the negative coefficient for AHE82 make eco-
nomic sense? Would one not expecta positive coefficient—the higher the hourly
earnings, the higher the attraction of the labor market? However, one could jus-
tify the negative coefficient by recalling the twin concepts of microeconomics,
namely, the income effect and the substitution effect."
Which model do we choose, Bq. (1.3) or Eq, (1.5)? Since Eq, (1.5) encompasses
Eq. (1.3) and since it adds an additional dimension (easnings) to the analysis, we
aay choose Eq, (1.5). After all, Eq. (1.2) was based implicitly on the assumption
that variables other than the unemployment rate were held constant. But where
“TAs we will discuss in Chapter 8, the coefficients of CUNR and AHES2 given in Eq. (1.5) are
known as partial regression cogficents. in that chapter we will discuss the precise meaning of parti
regression coelficents.
Consult any standard textbook on microeconomics, One intuitive justification of this result is
2s follows. Suppose both spouses are in the labor force and the earings of one spouse rises sub-
stantially. This may prompt the other spouse to withdraw from the labor force without substantially
alfecting the family income. CHAPTER ONE: THE NATURE AND SOOPE OF ECONOMETRICS 11.
do.we stop? For example, labor force patticipation may also depend on family
wealth, number of children under age 6 (this is especially critical for married
‘women thinking of oiting the labor market), availability of day-care centers for
young children, religious beliefs, availability of welfare benefits, unemploy-
ment insurance, and so on. Even if data on these variables are available, we may
not want to introduce them all in the model because the purpose of developing
an econometric model is not to capture total reality, but just its salient features,
If we decide to include every conceivable variable in the regression model, the
model will be so unwieldy that it will be of little practical use. The model ulti-
mately chosen should be a reasonably good replica of the underlying reality. In
Chapter 11 we will discuss this question further and find out how one can go
about developing a model.
Testing the Hypothesis Derived from the Model
Having finally settled on a model, we may want to perform hypothesis testing.
‘That is, we may want to find out whether the estimated model makes economic
sense and whether the results obtained conform with the underlying economic
theory. For example, the discouraged-worker hypothesis postulates a negative
relationship between labor force participation and the unemployment rate. Is this
hypothesis borne out by our results? Our statistical results seem to be in confor-
mity with this hypothesis because the estimated coefficient of CUNR is negative.
However, hypothesis testing can be complicated. In our illustrative example,
suppose someone told us that in a prior study the coefficient of CUNR was
found to be about -1. Are our results in agreement? If we'rely on the model,
Bq. (1.3}, we might get one answer; but if we rely on Eq, (1.5), wemightget another
answer. How do we resolve this question? Although we will develop the neoes-
sary tools to answer such questions, we should keep in mind that the answer toa
particulas hypothesis may depend on the model we finally choose.
‘The point worth remembering is that in repression analysis we may be inter-
‘ested not only in estimating the parameters of the regression model but also in
testing certain hypotheses suggested by economic theory and/or prior empiri-
cal experience.
Using the Model for Prediction or Forecasting
Having gone through this multistage procedure, you can legitimately ask the
question: What do we do with the estimated model, such as Eq, (1.5)? Quitenat-
urally, we would like to use it for prediction, or forecasting. For instance, sup-
Bose we have 2004 data on the CLINR and AHE82, Assume these values are 6.0
and 10, respectively. If we put these values in Eq, (1.5), we obtain 62.8315 per-~
cent as the predicted value of CLFPR for 2004. That is, ifthe unemployment
rate in 2004 were 6.0 percent and the real hourly earnings were $10, the civilian
labor force participation rate for 2004 would be about 63 percent. Of course,
when data on CLFPR for 2004 actually become available, we can compare the 12 CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS
predicted value with the actual value, The discrepancy between the two will
represent the prediction errr. Naturally, we would like to keep the prediction
error as small as possible. Whether this is always possible is a question that we
will answer in Chapters 6 and 7.
Let us now summarize the steps involved in econometric analysis.
‘Step Example
4. Statement of theory ‘The Added- /Dsoouraged-Worker Hypothests
2. Collection of Data Table 1-1
‘3, Mathematical Model of Theory: CLFPR = B; + &:CUNA
4 Egonometlc Model of Theory: CLEPR = B, + B,CUNA + u
5. Parameter Estimation: CLFPR = 69,9963 — 0.6513CUNR
8. Model Adequacy Check: CLFPR = 80.9 ~ 0.671CUNR — 1.4AHE@2
7. Hypothesis Test: &<00rB,>0
8. Prediction: ‘What is CLFPR, given values of CUNA and AHEB2?
Although we examined econometric methodology using an example from
labor economics, we should point out that a similar procedure can be employed
to analyze quantitative relationships between variables in any field of knowl-
edge. As a matter of fact, regression analysis has been used in politics, interna-
tional relations, psychology, sociology, meteorology, and many other disciplines.
1.4 THE ROAD AHEAD
Now that we have provided a glimpse at the nature and scope of econometrics,
let us see what lies aliead. The book is divided into four paris.
Partl, consisting of Chapters 2,3,4,and5, reviews the basics of probability and
statistics for the benefit of those readers whose knowledge of statistics has become
rusty. Reader should have some previous background in introductory stati
Part II introduces the reader to the bread-and-butter tool of econometrics,
namely, the classical linear regression model (CLRM). A thorough understanding
of CLRM is a must in order to follow research in the general areas of economics
and business.
Part Ill considers the practical aspects of regression analysis and discusses a
variety of problems that the practitioner will have to tackle when one ot more
assumptions of the CLRM do not hold.
Part IV discusses two comparatively advanced topic—simultaneous equa-
tion regression models and time series econometrics.
This book keeps the needs of the beginner in mind, The discussion of most top-
ics is straightforward and unencumbered with mathematical proofs, derivations,
etc’ firmly believe that the apparently forbidding subject of econometrics can
be taught to beginners in such a way that they can see the value of the subject
without getting bogged down in mathematical and statistical minutiae. The
"Some ofthe proofs and derivations are presented in my Basic Econometrics, Ath ed, MeGraw-
Hill, New York, 2003. CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS 13
student should keep in mind that an introductory econometrics courses just like
the introductory statistics course he or she has already taken, As in statistics,
econometrics is primarily about estimation and hypothesis testing. What is dif-
ferent, and generally much more interesting and useful, is that the parameters
being estimated or tested are not just means and variances, but relationships be-
tween variables, which is what much of economics and other social sciences are
all about.
A final word. The availability of comparatively inexpensive computer soft-
ware packages has now made econometrics readily accessible to beginners. In
this book we will largely use three software packages: Euiews, Excel, and
Minitab, These packages are readily available and widely used. Once students
get used to using such packages, they will soon realize that learning economet=
tics is really great fun, and they will have a better appreciation of the much
maligned “dismal” science of economics.
KEY TERMS AND CONCEPTS
QUESTIONS
‘The key terms and concepts introduced in this chapter are
Econometrics Random error term (error term)
Mathematical economics Linear regression model:
Discouraged-worker hypothesis dependent variable
(effect) independent (or explanatory)
Added-worker hypothesis (effect) variable
Time series data Deterministic vs. statistical
a) quantitative relationship
b) qualitative Causation
Cross-sectional data Parameter estimates
Pooled data Hypothesis testing
Panel (ot longitudinal or micropanel _ Prediction (forecasting)
data)
Scatter diagram (scattergram)
a) parameters
b) intercept
6) slope
141, Suppose a local government decides to increase the tax sate on sesidential prop-
erties under its jurisdiction. What will be the effect of this on the prices of resi-
dential houses? Follow the eight-step procedure discussed in the text toanswer
this question.
1.2. How do you perceive the role of econometrics in decision making in business
and economics?
13. Suppose you are an economic adviser to the Chairman of the Federal Reserve
Board (the Fed), and he asks you whether it is advisable to increase the money
13 4 CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS
PROBLEMS
TABLE 1-2
ie
supply to bolster the-economy. What factors would you take into account in
your advice? How would you use econometrics in your advice?
14. ‘To reduce the dependence on foreign oil supplies, the government is thinking
of increasing the federal taxes on gasoline. Suppose the Ford Motor Company
has hired you to assess the impact of the tax increase on the demand for its cars.
How would you go about advising the company?
1.5, Suppose the president of the United States is thinking of imposing tariffs on im-
ported steel to protect the interests of the domestic steel industry. As an economic
advisor to the president, what would be your recommendations? How would you
set up an econometric study to assess the consequences of imposing the tariff?
16, Table 1-2 gives data on the Consumer Price Index (CPI), S&P 500 stock index,
and three-month Treasury bill rate for the United States for the years 1980-2001.
a. Plot these data with time on the horizontal axis and the three variables on the
vertical axis. If you prefer, you may use a separate figure for each variable
. What relationships do you expect between the CPI and the S&P index and
between the CPT and the three-month Treasury bill rate? Why?
«. Foreach variable, “eyeball” a regression line from the scattergram.
CONSUMER PRICE INDEX (CPI, 1982-1984 = 100),
‘STANDARD AND POOR'S COMPOSITE INDEX (S8P 500,
{941-1943 = 100), AND THREE MONTH TREASURY BILL
RATE (3-mT BILL, %).
Year orl ‘S&P 500 3-mT bill
1980 824 118.78 1151
1981 90.9. 128.05 14.03
1982 96.5 119.71 10.68
1983 99.6 160.41 8.63
1984 103.9 160.48 9.58
1985 1076 106.84 ow
1986 108.6 23634 8.98
1987 1136 286.83, 5.82
1988 1183 265.79 6.69
1989 124.0 322.84 812
1990 1907 334,59 751
4991 136.2 376.18 5a2
1992 1403 415.74 345
1998 1445 451A 302
1994 1482 490.42 429
1095 1824 54172 551
1996. 156.9 670.50 5.02
1997 1605 873.43 507
1998 163.0 1085.50 481
1999 168.6 1827.33 468
2000 1722 1427.22 585
2001 aA 1194.18 3.45
‘Source: Economic Report ofthe President, 2002, Tabos B-60, CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS 15,
TABLE 4-3 GMS EXCHANGE RATE BETWEEN GERMAN MARK AND
U.S, DOLLAR AND THE OP| IN THE UNITED STATES AND
GERMANY, 1980-1998,
Year ews oPius, CPI Germany
1980 1.8175 824 86.7
1981 2.2692 90.9 922
1982 2.4281 965 974
1983 2.5599 99.6 1003
1984 2.8455 1080 4027,
1985 2.9420 107.6 104.8
1986, 24705, 1096 104.6
1987 4.7981 1136 1049
1988 4.7570 118.3 1063
1989 1,8808 1240 1092
1990 1.6166 130.7 1122
1991 1.8610 1362. 1163
1992 4.5618 1403, 122.2
1998 1.8545 1445 1278
1994 1.6216 1482 1314
1995 1.4321 1824 1935
1996 1.8049 156.9 1855
1997 1.7948 160.5 1978
1998, 4.1879 183.0 1984
‘Source: Ecencmlc Report of te President, 2002. GM trom
‘Table 8-110; OPI (1862-1884 = 100) from Yabo 8-408,
17, Table 1-8 gives you data on the exchange tate between the German mark and
the US. dollar (number of German marks per US. dollar) ag well as the con-
sumer price indexes in the two countries for the period 1980-1998.
4. Plotthe exchange rate (ER) and the two consumer price indexes against time,
measured in years. : .
b. Divide the US, CPIby the German CPi and calli the relative price ratio (RPR).
¢. Plot ER against RPR.
4. Visually sketch a regression line through the scatterpoints,
APPENDIX 1A: Economic Data
on the World Wide Web’
Economic Statistics Briefing Room: An excellent source of data on output, income,
employment, unemployment, earnings, production and business activity, prices
and money, credits and security markets, and intemational statistics,
http//wwwwhitehouse.govifsbr/esbrhtm
4 should be noted that this ist is by no means exhaustive. The sources listed here are up-
dated continually. The best way to get information om the Internet isto search using a key word
(ec. unemployment rate), Don't he surnrised vant e=t a mlthiors af infnematinn on fhe 16 CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS
Federal Reserve System Beige Book: Gives a suntmary of current economic con-
ditions by the Federal Reserve District. There are 12 Federal Reserve Districts.
hitp//wwwhbog.fth fed.us/fome/bb/current
Government Information Sharing Project: Provides regional economic informa-
tion; 1990 population and housing census; 1992 economic census; agriculture
census for 1982, 1987, 1992; data on US. imports and exports 1991-1995; 1990
Equal Employment Opportunity information,
hitp’//govinfo.kerrorst.edu
National Bureau of Economic Research (NBER) Home Page: This highly regarded
private economic research institute has extensive data on asset prices, labor,
‘productivity, money supply, business cycle indicators, etc,-NBER has many
links to other Web sites,
. httpeiwwwanberorg
Panel Study: Provides data on longitudinal survey of representative sample of
US. individuals and families. These data have been collected annually since 1968.
http://www.umich.edu/~psid
Resources for Economists ont the Internet: Very comprehensive source of infor-
mation and data on many economic activities with links to many Web sites. A
very valuable source for academic and nonacademic economists.
Aitp://econwpa.wwstl.edu/EconFAQ/EconFaq-html
‘The Federal Web Locator: Provides information on almost every sector of the
federal government; has international links,
http:/erwwdaw.villedu/Fed-Agency/fedwebloc.html
WebEC:WWW Resources in Economics: A most comprehensive library of eco-
nomic facts and figures.
http://wuecon.wustl.edu/~adneteciWebHo/WebEc.himt
‘Ameriéan Stock Exchange: Information on ome 700 ‘companies listed on the
second largest stock market.
httpy/www.amex.com/
Bureau of Economic Analysis (BEA) Home Page: This agency of the US.
Department of Commerce, which publishes the Survey of Current Business, is an
excellent source of data on all kinds of economic activities.
hitp:lwww.bea.dac.gov!
Business Cycle Indicators: You will find data on about 256 economic time series.
Attpswwwglobalexposure.convbei.html
CIA Publication: You will find the World Fact Book (anwial) and Handbook of
International Statistics,
hitp//www.odic.gow/cia/publications/pubs.html
Energy Information Administration (Department of Energy [DOE]): Economic
information and data on each fuel category.
hitpulwwweiadoe.gov/
FRED Database: Federal Reserve Bank of St. Louis publishes historical eco-
nomic and social data, which include interest rates, monetary and business
indicators, exchange rates, etc.
hitpu/wwwestls.frb.org/fred/fred.html (OVO CHAPTER ONE: THE NATURE AND SCOPE OF ECONOMETRICS 17
wag74718
International Trade Administration: Offers many Web links to trade statistics,
cross-country programs, etc.
httpulwwwiita.docgov/
STAT-USA Databases: ‘The National Trade Data Bank provides the most com-
prehensive source of international trade data and export promotion informa-
tion. It also contains extensive data on demographic, political, and sociveco-
nomic conditions for several countries.
http:/www.stat-usa.gov/BEN/databases.html
‘Statistical Resources on the Web/Economics: An excellent source of statistics col-
lated from various federal bureaus, economic indicators, the Federal Reserve
Board, data on consumer prices, and Web links to other sources.
httpy/wwwlib.umich.edu/libhome/Documents.centers/stecon.html
Buren of Labor Statistics: The home page contains data related to various as-
pects of employment, unemployment, and earnings and provides links to other
statistical Web sites.
hitp://stats.bls.gov:80/
U.S. Census Bureau Home Page: Prime source of social, demographic, and
economic data on income, employment, income distribution, and poverty.
hittp:l/www.census.govi
General Social Surcey: Annual personal interview survey data on U.S. house-
holds that began in 1972. More than 35,000 have responded to some 2500 different
questions covering a variety of data.
http//www.iepstumich.edw/GSS/
Institute for Research on Poverty: Data collected by nonpartisan and nonprofit
university-based research center on a variety of questions relating to poverty
and social inequality,
hitpywwwesscawisc.edwirp/
Social Security Administration: The official Web site of the Social Security
Administration with a variety of data,
hitpuhrww.sa.gov!
Federal Deposit Insurance Corporation, Bank Data and Statistics:
http:fwww {dic gov/bank/statisticallindex html
Federal Reserve Board, Economic Research and Data:
http:}www.federalreserve.govima.him
US. Census Bureau, Home Page:
httpywww.census.gov
U.S. Department of Conimerce, Bureau of Economic Analysis:
httpuwww.bea.gov :
US. Department of Energy, Energy Information Administration:
httpy/wwweia.doe.govineichistorichistorichtm
U.S. Department of Health and Human Services, National Center for Health -
Statistics:
Ritpdfeewwede.gowinchs
U.S, Department of Housing and Urban Development, Data Sets:
pulwrwwhudusenorg/datasets/pdrdatas.html
1] 18 CHAPTER ONE: THENATURE AND SCOPE OF ECONOMETRICS
18
ULS. Department of Labor, Bureau of Labor Statistics:
hitpd/wwwobls.gov
USS. Department of Transportation, TranStats:
hitp:/wwwéranstats.bts.gov.
U.S. Department of the Treasury, Internal Revenue Service, Tax Statistics:
hitp//wwwirs.gov/taxstats
Rockefeller Institute of Government, State and Local Fisoal Data:
hitp:/stateandlocalgateway.rockinst.org/fiscal_trends
Arierican Economic Association, Resources for Economists:
hitpshwwwafe.org,
Anterican Statistical Association, Business and Economic Statistics:
hitp//www.econ-datalinks.org
American Statistical Association, Statistics in Sports:
http//www.amstat.org/sections/sis/sports.htm!
European Central Bank, Statistics:
hittp:/wwwecb.int/stats
World Bank, Data and Statistics:
hitpdlwww.worldbankorgidata
International Monetary Fund, Statistical Topics:
httpu/www.mgorg/external/np/sta/indexhtm
International Monetary Fund, World Economic Outlook:
hittp//www.imf.org/external/pubs/ft/weo/2003/02/datalindex.htm.
Penn World Tables:
hitp?//pwrtecon.upenn.edu
Current Population Survey:
httpulwwwbls.census.gov/cps/epsmain.htm
Consumer Expenditure Survey:
hittpv/www.bls.gov/cexfhome.htm
Survey of Consumer Finances:
httpviwww.federalreserve.gov/pubs/oss/oss2/scfindex.html
City and Countty Data Book:
itp/fwww.census.gov/prod/wwwicedb.html
Panel Study of Income Dynamics:
http:/psidonline.isr.umich.edu
‘National Longituidinal Surveys:
hittpyiwwwbls.gov/nlshhome.htm
National Association of Home Builders, Economic and Housing Data:
http:/www.nahb.org/category.aspx?sectionID=113
National Science Foundation, Division of Science Resources Statistics:
hittpywww.nsf govisbelsrsistats htm PART {
BASICS OF PROBABILITY
AND STATISTICS
This part consists of four chapters that review the essentials of statistical theory
that are needed to understand econometric theory and practice discussed in the
remainder of the book.
‘Chapter 2 discusses the fundamental concepts of probability, probability dis-
tributions, and random variables.
Chapter 3 discusses the characteristics of probability distributions such as
the expected value, variance, covariance, correlation, conditional expectation,
conditional variance, skewness, and kurtosis. This chapter shows how these
characteristics are measured in practice.
Chapter 4 discusses four important probability distributions that are used
extensively in practice: (1) the normal distribution, (2) the t distribution, (3) the
chi-square distribution, and (4) the F disttibution. In this chapter the main fea-
tures of these distributions are outlined. With several examples, this chapter
shows how these four probability distributions form the foundation of most
statistical theory and practice.
Chapter 5is devoted to a discussion of the two branches of classical statistics—
estimation and hypothesis testing. A firm understanding of these two topics will
considerably make our study of econometrics in subsequent chapters easier.
‘These four chapters are written in a very inforntal yet informative style so
readers can brush up on their knowledge of elementary statistics. Since stu-
dents coming to econometrics may have different statistics backgrounds, these
four chapters provide a fairly self-contained introduction to the subject.
All the concepts introduced in these chapters are well illustrated with several
practical examples,
19 CHAPTER 2
REVIEW OF STATISTICS |:
PROBABILITY AND
PROBABILITY
DISTRIBUTIONS
‘The purpose ofthis and the following three chaptersis to review somefundamen-
tal statistical concepts that are needed to understand Essentials of Econometric.
‘These three chapters will serve as a refresher course for those students who
have had a basic course in statistics and will provide a unified framework for
following discussions of the material in the remaining patts of this book for
those whose knowledge of statistics has become somewhat rusty. Students
who have had very little statistics should supplement these three chapters with
a good satistics book. (Gome references are given at the end of this chapter)
Note that the discussion in Chapters 2 through 5 is nonrigorous and is by no
micans a substitute for a basic course in statistics, It is simply an overview that
is intended as a bridge to econometrics.
2.1 SOME NOTATION
Inthis chapter we come across several mathematical expressions that often can
be expressed more conveniently in shorthand forms.
The Summation Notation
‘The Greck capital letter E(sigma) is used to indicate summation or addition,
Thus,”
Sama Ke tbh
ist 22 PART ONE: BASICS OF PROBABILITY AND STATISTICS
where fis the index of summation and the expréssion on the left-hand sides the
shorthand for “take the sum of the variable X from the first value ({= 1) to the
nth value (i =)"; X stands for the ifh value ofthe X variable.
re oy X)
is often abbreviated as
‘ Ls
' where the upper and lower limits of the sum are known or can be easily deter-
ined or also expressed as
yx
x
which simply means take the sum ofall the relevant values of X. We will use all
these notations interchangeably.
Properties of the Summation Operator
Some important properties of © are as follows:
1, Where kis a constant
Dk=nk
ist
That is, a constant summed n times is m times that constant. Thus,
4
Y3=4x3=12
a
In this example n = 4 and k= 3.
2, Where k is a constant
DRX SED X
That is, a constant can be pulled out of the summation sign and put in
front of it,
3 Lith) =CX+ LY
Thats, the summation of the sum of two variables is the sum of their in-
dividual summations.
4, La + bX) =a +b YX;
wheres and b are constants and where use is made of properties 1,2,and3.
‘We will make extensive use of the summation notation in the remainder of
this chapter and in the rest of the book.
We now discuss several important concepts from probability theory. CHAPTER TWO: REVIEW OF STATISTICS |: PROBABILITY AND PROBABILITY DISTRIBUTIONS "23
2.2 EXPERIMENT, SAMPLE SPACE, SAMPLE POINT, AND EVENTS
Experiment
The first important concept is that of a statistical or random experiment. In sta-
tistics this term generally refers to any process of observation or measurement
that has more than one possible outcome and for which there is uncertainty
about which outcome will actualy materialize,
Example 2.1,
Tossing a coin, throwing a pair of dice, and drawing a card from a deck of
cards are all experiments. It is implicitly assumed that in performing these
experiments certain conditions are fulfilled, for example, that the coin or
the dice are fair (not loaded). The outcomes of such experiments could be
ahead or a tail if a coin is tossed or any one of the numbers 1, 2,3, 4,5, or 6
if a die is thrown, Note that the outcomes are unknown before the experi-
ment is performed. The objectives of such experiments may be to estab-
lish a law (eg,, How many heads are you likely to obtain in a toss of, say,
1000 coins?) or to test the proposition that the coin is loaded (e.g., Would
you regard coin as being loaded if you obtain 70 heads in 10 tosses of
acoin?).
Sample Space or Population
‘The set of all possible outcomes of an experiment is called the population or
sample space, The concept of sample space wss frst introduced by von Mises,
an Austrian mathematician and engineer, in 1931.
Example 2.2,
Consider the experiment of tossing two fair coins. Let H denote a head and
Ta tail. Then we have these outcomes: HH, HT, TH, TT, where HH means a
head on the first toss and a head on the second toss, HT means a head on the
first toss and a tail on the second toss, etc.
In this example the totality of the outcomes, or sample space or popula-
tion, is 4—no other outcomes are logialy possible. (Don’t worry about the
coin landing on its edge.)
Example 23.
‘The New York Mets are scheduled to play a doubleheader, Let O; indicate the
outcome that they win both games, Q, that they win the first game but lose
the second, Os that they lose the first game but win the second, and O, that
they lose both games, Here the sample space consists of fouir owtcomes: Os,
0s. Ox, and Ox 24 PARTONE: BASICS OF PROBABILITY AND STATISTICS
Sample Point
Events
Venn Diagrams
24
Each member, or outcome, of the sample space or population is called a sample
point. In Example 22 each outcome, HH, HT, TH, and TT, ise sample point. In
Example 23 each outcome, 01, Ox, Os, and Oy is a sample point.
An event isa particular collection of outcomes and is thus a subset of the sam-
ple space.
Example 24,
Letevent be the occurrence of one head and one tail in the coin tossing ex-
periment, From Example2.2 we see that only outcontes HT and TH belong to
event A, (Note: HT and TH are a subset of the sample space HH, HT, TH, and
TT). Let B be the event that two heads occur in a toss of two coins. ‘Then, ob-
‘viously, only the outcome HH belongs to event B. (Again, note that HH is a
subset of the sample space HH, HT, TH, and TT).
Events are said to be mutually exclusive if the occurrence of one event pre-
vents the occusrence of another event at the same time. In Example 2.3, if 0; oc-
curs, thatis, the Mets win both the games, itrules out the occurrence of any of the
other three outcomes. Two events are said to be equally likely if we are confident
that one event is as likely to occur as the other event, In 4 single toss of a coin a
head is as likely to appeas as a tail. Events are said to be collectively exhaustive
if they exhaust all possible outcomes of an experiment. In our two coin-tossing
example, since HH, HT, TH, and TT are the only possible outcomes, they are(col-
lectively) exhaustive events. Likewise, in the Mets example, 01, Oz, O3, and Ox
are the only possible outcornes, barring, of course, rain or natural calamities such
as the earthquake that occurred during the 1989 World Series in San Francisco.
‘Asimple graphic device, called the Venn diagram, originally introduced by
‘Venn in his book, Symbolic Logic, published in 1861, can be used to depict sam-
ple point, sample space, events, and related concepts, as shown in Figure 2-1.
In this figure each rectangle represents the sample space § and the two circles rep-
resent two events A and B. If there are more events, we can draw more circles to
represent all those events. The various subfigures in this diagram depict various
situations,
Figure 2-1(@) shows outcomes that belong to A and the outcomes that doniatbe-
long to, which are denoted by thesymbol A, whichis called the complement of A.
Figure 2-1(b) shows the union (ie., sum) of A and B, that is the event whose
outcomes belong to set A or set B. Using set theory notation, it is often denoted
as AUB (read as A union B), which is the equivalent of A + B. CHAPTER TWO: REVIEW OF STATISTICS I: PROBABILITY AND PROBABILITY DISTRIBUTIONS 25
oe ®
@ @
FIGURE 21 Vern diagram,
‘The shaded area in Figure 2-1(c) denotes events whose outcomes belong to
both set A and set B, which is represented as AN B(read.as A intersects B), and
is the equivalent of the product AB,
Finally, Figure 2-1(d) shows that the two events are mutually exclusive because
they have no outcomes in common. In set notation, this means ANB = 0 (or
that AB = 0).
2.3 RANDOM VARIABLES
Although the outcome(s) of an experiment cant be described verbally, stich as a
head or a tail, or the ace of spades, it would be much simpler if the results of all
experiments could be described numerically, that is, in terms of numbers. As we
will see late, for statistical putposes such representation is very useful,
Example 25.
Reconsider Example 2.2. Instead of describing the outcomes of the experi
ment by HH, HT, TH, and TT, consider the “variable” number of heads in a
toss of two coins. We have the following situation:
Fitstccin —_-Sgcond.¢oin Number of heads
T T 0
a H 1
H i 1
H H 2
os 26 PART ONE: BASICS OF PROBABILITY AND STATISTICS
‘We call the variable “number of heads” a stochastic or random variable (r.v.,
for short). More generally, a variable whose (numerical) value is determined by the
outcome of an experiment is called a random variable. In the preceding example the
xv, number of heads, takes three different values, 0, 1, or 2, depending on
whether no heads, one head, or two heads were obtained in a toss of two coins.
In the Mets example the rv, the number of wins, likewise takes three different
values; 0, 1, or 2.
By convention, random variables are denoted by capital letters, X, Y, Z, etc.,
and the values taken by these variables are often denoted by small letters. Thus,
if X is an rv, x denotes a particular value taken by X.
_An ry. may be either discrete or continuous. A discrete random variable
takes on only a finite number of values or countably infinite number of values
(ce. as many values as there are whole numbers). Thus, the number of heads in
a toss of two coins can take on only three values, 0, 1; or 2. Hence, it is a’discrete
rv. Similarly, the number of wins in a doubleheader is also a discrete rv. since it
can take only three values, 0, 1, or 2 wins. A continuous random variable, on
the other hand, is an tv. that can take on any value in some interval of values.
‘Thus, the height of an individual is a continuous variable—in the range of, say,
60 to 72 inches it can take any value, depending on the precision of measure-
ment. Similar factors such as weight, rainfall, or temperature also can be re-
garded as continuous random variables.
24 PROBABILITY
Having defined experiment, sample space, sample points, events, and random
variables, we now consider the important concept of probability. First, we define
the concept of probability of an event and then extend it to random variables.
Probability of an Event: The Classical or A Priori Definition
If an experiment can result inn mutually exclusive and equally likely outcomes,
and if mz of these outcomes are favorable to event A, then P(A), the probability
that A occurs, is the ratio m/n, That is,
umber of outcomes favorable toA en
total number of outcomes
Note the two features of this definition: The outcomes must be mutually exclusive
(that is, they cannot occur at the same time), and each outcome must have an
equal chance of occurring (for example, ina throw of adie, any one of the six numbers
has an equal chance of appearing).
P(A) =
Example 26.
Ina throw of adie numbered 1 through 6, there ae six possible outcomes: 1,
2,3,4,5,0r 6. These outcomes are mutually exclusive since, ina single throw (CHAPTER TWO: REVIEW OF STATISTICS |: PROBABILITY AND PROBABILITY DISTRIBUTIONS 27
of the die, two or more numbers cannot tun up simultaneously. These six
outcomes are also equally likely. Hence, by the classical definition, the prob-
ability that any of these six numbers will show up is 1/6—there are six total
outcomes and eaich outcome has an equal chance of occurring, Here n = 6
and m=1,
Similarly, the probability of obtaining a head in a sirigle toss of a coin is 1/2
since there are two possible outcomes, Hand T, and each has an equal chance of
coming up. Likewise, in a deck of 52 cards, the probability of drawing any sin-
gle card is 1/52. (Why?) The probability of drawing a spade, however, is 13/52.
(Why?)
The preceding examples show why the classical definition is called an a pri-
ori definition since the probabilities are derived from purely deductive reason-
ing. One doesn’t have to throw a coin to state that the probability of obtaining a
head ora tail is 1/2, since logically, these are the only possible outconies.
But the classical definition has some deficiencies, What happens if the out-
comes of an experiment are not finite or are not equally likely? What, for exam-
le, is the probability that the gross domestic product (GDP) next year will be a
certain amount or what is the probability that there will be a recession next
year? The classical definition is not equipped to answer these questions. A more
widely used definition that can handle such cases is the relative frequency def-
inition of probability, which we will now discuss.
Relative Frequency or Empirical Definition ot Probabitity
To introduce this concept of probability, consider the following example.
Example 2.7. :
Table 2-1 gives the distribution of maéks received by 200 students on a mi-
croeconomics examination. Table 2-1 is an example of a frequency distribu-
tion showing how the rv, marks in the present example are distributed. The
numbers in column 3 of the table are called absolute frequencies, that is, the
number of occurrences of a given event. The numbers in column 4 are called
relative frequencies, that is, the absolute frequencies divided by the total
‘number of occurrences (200 in the present case). Thus, the absolute frequency
of marks between 70 and 79 is 45 but the relative frequency is 0.225, which is
45 divided by 200,
Can we treat the relative frequencies’as probabilities? Intuitively, it seems
reasonable to consider the relative frequencies as probabilities provided the
number of observations on which the relative frequencies are based is reason-
ably Jarge. This is the esséince ofthe enipirial, or relative frequency, definition of
probability 28 PART ONE: BASICS OF PROBABILITY AND STATISTICS
28
TABLE 2-1 THE DISTRIBUTION OF MARKS RECEIVED BY 200
STUDENTS ON A MICROECONOMICS EXAMINATION.
Midpoint of Absolute Relative
“Marks infeval = frequency frequency.
«) @ 6) (4) = (3200
08 5 0 o
10-19 15 0 o
20-29 25 0 0
30-39 5 10 0.050
40-49, 6 20 0.400
50-69 5 8 0175
60-69 65 50 0.250
70-79 6 45 0225
00-89 85 30 0.150
90-99 5 10 0,050
Total: 200 1.000
More formally, if in trials (or observations), m of them are. favorable to
event A, then P(A), the probability of event A, is simply the ratio m/n (te, rela-
tive frequency) provided n, the number of trials, is sufficiently large (technically,
infinite) Notice that, unlike the classical definition, we do not have to insist
that the outcome be mutually exclisive and equally likely. .
Inshort, if the number of trials is sufficiently large, we can treat the relative
frequencies as fairly good measures of true probabilities. In Table 2-1 we can,,
therefore, treat the relative frequencies given in column 4 as probabilities”
Properties of Probabilities ‘The probability of an event as defined earlier
has the following important properties:
1. The probability of an event always lies between O-and 1. Thus, the prob-
ability of event A, P(A), satisfies this relationship:
0
