Edition

I
DUN/IIMY VARIABLE
REGRESSION N/ODELS

In Chapter 1 we discussed briefly the four types of variables that one gener-
ally encounters in empiricai anaiysis: These are: ratio scale, interval scale,
orii.rul scale, and nominal scale. The types of variables that we have
encountered in the preceding chapters were essentially ratio scale. Rut this
should not give the impreruion that regression models can deal only with
ratio scale vlriables. Regression models can also handie other type: of vari-
ables mentioned previoirsly. In this chapter, we consrder models that may
involve not only ritio scale variables but also nominal scale variabies. Such
variables .r" ulro known as indicator variables, categorical variables,
qualitative variables, or dummy variables'l

9.1 THE NATURE OF DUMMY VARIABLES

In regression analysis the clependent varjable, or regressand, is frequently
influenced not oniy by ratio scale variables (e.g., income, output, prices,
costs, height, temperaiure) but also by variables that are essentially
qualita-
tive, or nJminal scale, in nature, such as Sex, race, color, religion, national-
ity, geographical region, politicarl upheavals, and party affiliation' For exam-
found to earn less
fi",ir"flai"g al1 other-factors constant, female workers arefound to earn less
than their male counterparts or nonwhite workers are
than wh;tes.2 This pattern may resull from sex or rarcial discrimination' but
whatever the reason, qualitative variables such aS sex and race seem to
lWe will discuss ordinal scale variables in Chap' 15'
2Fo;; ;eview tfr. ."ia."ce on this subject, see Bnrce E. Kaufman and Julie L' Hotchkiss'
The'Economics af"fLabor Mat*et,5th ed', Dryden Press, New York, 2000'
 tvlODELS
PART ONE: SINGLE-EOUATION REGRESSION

the explana-
cleariy shor-rlcl be inch-rded among
influence the regressand and
indicate the presence or absence or a
'T;T^3,1:il1l3ffi,."i'T;t'n black or white, Catholic or
,,quality,, or an uttribrt", sucit ,, -rf
"r-i"*ut.,
they are essentially noruinal scale
non-carhoti", o.rrrJ;;;;;; Republican,
;;.! attributes is by constructing
variables. or. *rv * could "'q;;,tfy'; (or
fu*;of i o' 0' 1 indicating the presence
artificial variabteli#*il; that attribute'
of th;;;;rib"re and o i"Ji.rting the absence of
possession)
,rr^, p.t*n is lJemale and 0 may designate a )
For example 1 may indicate ^ )
male;orlmayindicatethata;;;;isacollegegraduate'lnd0thatthe
t.s that assume such 0 and L values
are ;
person is nor, ;;^;;";"" v*iut to clas-
'.' Su"h,oi*t't''t are thus essentially a device
called dummy
""ti"Uf"t rotl-gorie-s such as male ctr feruale'
sify data into nrut*olly exclus.irr in regression models just as easily
Dummy varrables can be i,,"o,|oi,ted
variables. A. mrtter of fact' a regression model may con-
dummy,- or qualitative' in nature'
as qr-rantitative ^
tain regressors ;;;;; all exclusively
of v*irtt." (ANovA) models'"
Such models ur..uir.a Analysis

9.2 ANOVA MODELS

ToillustratetheANOVAmod"els,considerthefollowingexample

EXAMPLE 9.1
REGION
SALARIES BY GEOGRAPHICAL
PUBLIC SCHOOL TEACHERS'
geo-
51 areas are classified into three
Tableg.lgivesdataonaveragesalary(ndollars)ofpublicschoolteachersin50statesand
year'tbes..These
the District ot cotumoiJil;lh" (2) south (17 states in
North c""i*r 1zi ttut"t in all)'
graphicat regions: trl 5[nn""r1lnd about the format of the
all), and (3) West t" it all)' For tne time oeing' do not worry
'i"i"t
ili"-';d'tfi";:Tfil3fllT,ii,ilfr"'1?'i;"n"
":rH'"i:x,gt'fJixiJr:Tfll?l'i""
,"gjJ,;.;'th;;;ilty" lf you take the simple arith-
difters among the three geographicar you will find that
ihree regions'
salaries or tne Lacners in the
metic average of the ur"lag" tJil*" (Northeast and North cen-
regions ur" ,. iiq,'qzq'tq
these averages for tn" *,r"! inlt"nu*n"rs look different'
trat), $22,8e+ tsou*ril r"i
gi-o]i!s.u, tw".ii.
|}:ffi}]

gories on1Y.
 MODELS 299
CHAPTER NINE: DUMMY VABIABLE REGRESSION

EXAMPLEg.l (Continued)
BY STATE, 1986
TABLE 9.1 AVERAGE SALARY OF PUBLIC SCHOOLTEACHERS,

Salary Spending Dz De Salary SPending Dz Ds

22,795 3366 0 1
3346 1 0
19,583 2920 0 1
0 21,570
20,263 3114 1
0 1
0 22,080 2980
2A325 3554 1
0 1
0 22,250 3731
26,800 4642 1
0
0 20,940 2853 1

29,474 4669 1
0 1
0 21,800 2533
26,610 4888 1
0
o 22,934 2729 1

30,678 571 0 1
0
'l 0 18,443 2305 1

27,170 5536 0
0 19,538 2642 1

25,853 41 68 1
0 1
0 20,460 3124
24,500 3547 1
0 1
0 21,419 2752
24,274 31 59 1
0
0 25,160 3429 1

27,170 3621 1
0 0
0 22,482 3947
30,168 3782 1
0 0
0 20,969 2509
26,525 4247 1
0 0
0 27,224 5440
27,360 3982 1
0 0
0 25,892 4042
21,690 3568 1
0 0
0 22,644 3402
21,974 31 55 1
0 0
0 24,640 2829
20,816 3059 1
0 0
0 22,341 2297
18,095 2967 1
0 0
0 25,610 2932
20,939 3285 1
0 0
0 26,015 3705
22,644 391 4 1
0 0
25,788 4123
24,624 4517 0 1
0
29,132 3608 0
27,186 4349 0 1
0
41,480 8349 0
33,990 5020 0 1
0
25,845 3766 0
23,382 3594 0 1

20,627 2821 0 1

1 {or states in the Northeast and

North Central; 0 otherwise'
Note: D2 -
1 tor states in the South; 0 otherwise.
by Atbuquerque Tribune' Nov' 7'
Da = 1986'
Source.INational Educational XIJ""i"ii"", "t't"po,t"o

to compare
There are various statistical techniques
statistically different from one another? But the
generallv. g! b; th;;*" of.analysis of variance's
two or more mean values, which analysis'
wittrin treframework of regression
-- objective can n* r-;orpri,irreo
same
model:
To see this, consider the tollowing
ui (9'2'1)
Yi: flt * frzDzi* FsiDsi*

in state i
where yi : (average) salary of public school te-1c!3r Central
or,,:\ i,tthJsiate is in the Northeast orofNorth
: 0 otherwise (i.e', in other regions the country)
Dsi: l if the state is in the South
: of the country)
0 otherwise (i.e', in other regions (Continued)

ffint,See{qET:,App-liedRegyessionAnalysis,LinearModels,andRe-
1997 Chap' 8'
t"t"iiiiii,it, sue" Publications' '
J PART ONE: SINGLE-EOUATION REGRESSION MODELS

EXAMPLE 9.1 (Continued) |

Notethat(9.2.1)islikeanymultipleregressionmodelconsideredpreviously,exoeptthat.
regressors, taking the
qualitative, or dum.my,
instead of quantitative regressors, we have only
particular category and 0 il it does not belong to that
value of 1 if the observation belongs to a
alt dummy variables by the letter D'
category or group. Hereafter, wz snai designate
faUtJ S.1 shows the dummy variables thus constructed'
that the error te.rm satisfies the usual oLS
what does the model (9.2.1) tell us? Assuming
(9'2'1) on both sides' we obtain:
assumptions, on taking expectation of
MeansalaryotpuuticschoolteachersintheNortheastandNorthCentral:
E(YilDzi- 1, Dzi= 0) : fu + fJz t9'2'2)

South:
Mean salary of public school teachers in the

E{YilDzi: A, Dsi: 1) : fu * frs (9'2'3)

guessed
salary of teachers in the west' ll you
You might wonder how we find out the mean
right' for
that thii is equal to li1, you would be absolutely
public school teachers in the West:
Mean salary of
E(Yil D2i: 0' Dsi : 0) : fu (9'2'4)

teachers in the west is given by the inter-

ln other words, the mean salary of public school
"slope" coefficients fi2 and B3 tell by how
cept, B1,in the multipt" i"gr"rrion (s e r 1, and the
and North Central and in the South differ
much the mean salaries ofieachers in the Northeast
But how do we know if these differences are
from the mean salary oi teachers in the west.
question, let us present the results based on
statistically significanti Before we answer this g't, we obtain the following results:
the regression (9.2.1i U.ing the data given in fante

: za,tsa.6z - 1794.473D2i- 3264'615Dsi

9',
(1499'615)
,
se : (1128'523) (1435'953)

'illilll" ' ,llllll" ' ,lllill. R2:00e01

(e.2.5)

* indicates the P values'

where
Astheseregressionresultsshow,themeansalaryofteachersintheWestisabout and
and North Central is lower by about $1734,
$26,158, that of teachers in the Northeast mean salaries in the last two
The actual
that of teachers in the South is lower by about $3265' the mean salary of
differential salaries to
regions can be easily obtained by adding these we will find that the
(9.2.3) and (9,2'4)' Doin^g this,
teachers in the West, as shown in Eqs,
mean salaries in the iatter two regions are about $24'424 and $22'894'
are statistically different from the mean
But how do we know that these mean salaries All we have to
the comparison category? That is easy enough'
salary of teachers in the west,,,slope"
(s.z.s) is statistically significant' As can be
do is to find out if each of the coefficients in
coefficient for Northeast and North central is
seen from this regression, the estimated slope south is statistically
23'percent, whereas that of the
not statistically significant, as its p value is
p uul,* is only about 3,5 percent' Therefore, th,e overall conclusion is that
significant, as the
teachers in the West and the Northeast and
statistically tne mean saiari"s of public school
salary of teachers in the south is statistically
North central are about the same but the mean
the situation is shown in Figure 9'1'
signi{icantly lower by about $3265. Diagrammatically,
The dummy variables w-ill simply
A caution is in order in interpreting these differences' for the difterences'
not suggest the reasons
point out the differences, ii they exist, 6ut they do
| (Continued)
 CHAPTER NINE: DUMMY VARIABLE REGRESSION MODELS 301

EXAMPLE9.l (Continued)
I

B, = $26,158

924,424 (Fr* Br1

$22,894 Gr* Br1

West Northeast and South

North Central

FlGURE 9.1
Average salary (in dollars) of public schoolteachers in
three regions.

Differences in educational levels, in cost of living indexes, in gender and race may all have
some effect on the observed differences. Therelore, unless we take into account all the other
variables that may affect a teacher's salary, we will not be able to pin down the cause(s) of
the differences.
From the preceding discussion, it is clear that all one has to do is see if the coefficients
attached to the various dummy variables are individually statistically significant' This exam-
ple also shows how easy it is to incorporate qualitative, or dummy, regressors in the regres-
sion models.

Caution in the Use of Dummy Variables

Although they are easy to incorporate in the regression models, one mLlst Llse
the dummy variables careftilly. In par:ticular, consider the following aspects:

1. In Example 9.1, to distinguish the three regions, we used only two

dummy variables, D2 andDr. Why did we not use three dummies to distin-
guish the three regions? Suppose we do that and write the model (9.2.1) as:

Yi -- u * gtDti * frzDzi * fuD:i * tti (9'2'6)

where D1i takes a value of I for states in the West and 0 otherwise" Thus, we
now have a dummy variable for each of the three geographical regions'
Using the data in Table 9. 1, if you were to run the regression (9.2'6), the com-
putei will "refuse" to run the regression (try it).6 Why? The reason is that in

6Actually you will get a message saying that the data matrix is singular.
 tilr

302 PART oNE: sINGLE-EouATloN REGRESSIoN MoDELS

) the setup of (g.2.6) where you have a dummy variable for each category or
grolrp und ulro an intercept, yog have a case of perfect collinearity, that is,
I*u"i linear relationshipi among the variables. Why? Refer to Table 9'1'
Imagine that now *. .id the D1 column, taking the valtre of 1 whenever a
statJis in the West and 0 other:wise. Now if you add rhe three D coh-rmns hor-
izontally, you will obtain a column that has 51 ones in it' But since the va|-te
of the intercept a is (irnplicitly) 1 for each obserwation, you will have a
column that aiso contains 51 ones. In otherwords, the sum of the three D
columns will simply reproduce the intercept column, thus leading to
perfect
collinearity. In this case, estimation of the model (9.2,6) is impossible'
The message here is: If a qualitative variable has m categories,-intro-
duce only (zr - 1) dummy variables. In our: example, since the qr-ralitative
rrariable 'ir"gion" has three categories, we intt^oduced only two dummies'
If
you do not f;llow this r^r-rle, you will fall into what is called the dummy vari-
.bl" tr.p, that is, the situation of perfect collinearity or perfect multi-
collinearity, if there is more than one exact relationship among the- vari-
ables. This rule also applies if we have more than one qualitative variable
in
the model, an example of which is presented later. Thr-rs we should restate
the preceding rule Lr: For each qualitative- regressor the number of
dummy variibles introduced muit be one less than the categories of
that vaiiable. Thus, if in Example 9.1 we had information about the gender
of the teacher, we would use an additional dummy variable (but not two)
taking a value of 1 for female and 0 for male or vice versa'
Z.*fne category for which no dummy variable is assigned is known as
the base, b"r"il*rrk, control, comparison, reference, or omitted
cate-
gory. And all comparisons are made in relation to the benchmark category'
3, The intercept value (Fr) represents the mean value of the benchmark
category. In Example 9.1, the benchmark category is the western region'
Henie,ln the regrlssion (g.2.5) the intercept value of about 26,159 repre-
sents the mean salary of teachers in the western states.
4. The coefficienis attached to the dummy variables in (9.2 '1 ) are known
as the differential intercept coefficients because they tell bV h"Y
much
the value of the intercept that receives the value of 1 differs from the
intep
in (9'2'5), the value
cept coefficient of the benchmark category. For-example,
of about -1734 te11s us that the mean salary of teachers in the Nor-theast or
North Central is snraller by about $1734 than the mean salary of about
$26,159 for the benchtlark category, the West'
5. If a qualitative variable his more than one category as in our illus-
trative exaripl", ttre choice of the benchmark category is strictly up to the
par-
researcher. Sometirnes the choice of the benchmark is dictated by the
chosen
ticular problem at hand. In our illustrative exampie, we could have
the south as the benchmark category. ln that case the regression results
given in (9.2.5) will change, because now all comparisons are made in
rela-
tion to the South. Of cotir:se, this will not change the overail conclusion of
our example (why?). In this case, the intercept value will be about $22'894'
which is the mean salary of teachers in the South'
 CHAPTER NINE: DUMMY VARIABLE FTEGRESSION MODELS 303

way to
6. We warned above about the dr-rmmy variable trap. There is a num-
variables as the
circumvent this traf by introdr-rcing as many dummy
the intercept
ber of categories of tnut variable, ltrottid-ed we do not introduce
(9"2"6), and con-
in such a mod.el. ifr*, if we dr:op'the intercept term from
sider the following model,

Yi: * frzDzi * fuDti *wi (9'2'7)

frtDti
longer perfect
we do not fall into the dummy variable trap, as there is no
collineari ty. But ruake sure thai tuhen you rri this regression,
you use the no'
intercept optiort in your regression packag,e'
of
How do we interpret iegression (g.i'.7)? If you take the expectation
(9.2.7), you will find that:

flt : mean salary of teachers in the West

mean salary of teachers in the Northeast and Noflh
central'
frz :
f: : mean salary of teachers in the South'
va'riable
In other worcls, with tlrc intercept suppressed, and allowing a dwnmry
,rruon values of- the various categories'
lor each category, we obtain dirictty ihu as follows:
The results of (9.2.7) for or-rr ilh-rstrative example are
i4 :26,158.62Dti + 24,424.14D2i + 22,894D2i
se: (t128.s23) (8 s7 'st7}) (986'864s) (9'2'8')

t-
L- (23.1795\.
\LJ'Lt'r/ Q7.5072)"
\- 123.1987)''
Rz:o.o9ol

where " indicates that the p values of these / ratios are

very small'
mean (salary) val-
As you can see, tir; J;;*y coefficients give directly
the
and South'
ues in the three ,egionr, WLst, Northeasiand North Central,
T. Which i, u-f,"tt". method of introducing a dummy variable:(2) intro-
(1)
term-or include
duce a dummy for each category and omit the intercept
the intercept term and introduce only (m- 1) dummies, where
m is the
number of iategories of the dummy variable? As Kennedy notes:
convenient because it
Most researchers find the equation with an intercept more
in which they usually have the
erllows them to uj,lr,"r, rnore easily the questions
a dif{terence' and
most interest, namely whether or not the categorization tnakes ,
how muc'h is
if so, by how rtruch. if tt " categorization does make a dilference, by
estimates' Testing whether
measured directly by the dummy variable coefficient
running a r test of a durnmy
or not th" cat*go, irltion is relevant can be clone by
F test on the appro-
variable .,r"ffi.i.ri against zero (or, to be more general, an
priate set of dummy iariable coelficient estimates)'7

TPeter Kenn edy, A Guide to Econometrics, 4th ed., MIT Press, Cambridge, Mass', 1998,p' 223
 J4 PART ONE: SINGLE-EQUATION REGRESSION MODELS

I VARIABLES
9.3 ANOVA MODELS WITH TWO QUALITATIVE
IntheprevioussectionweconsideredanANoVAmodelwithonequalitative
we consider another ANOVA
variable with thre" ;;g;;i.t. In this -section
and bring out some additional
model, but with ffi" ;;;ir"ii". "uriables,
Points about dummY variables'

Obviously'
EXAMPLE 9.2 Which is the benchmark category here?
it is unmarried, non-South residence' ln other words'
South are the
HOURLY WAGES IN RELATION TO MARITAL
STATUS unmarrieO persons who do not live in the
Therefore, all comparisons are made
AND REGION OF FIESIDENCE o*itt"O
"ui"gory.
in relation to this group. The mean hourly wage in this
the follow- this' the
From a sample of 528 persons in May 1985' benchmark is about $8'81' Compared with
ing regression results were obtainedu: married is higher
average hourly wage of those who are
uy-uolrt $t.io, Ior an actual average wage of $9'91
Y,: 8.8148 * 1'O997Dzi- 1'6729Dsi i:aaf + 1.10)' By contrast, for those who live in
the

sB: (0.4015) (0.4642') (0'4854) loutn, the average hourly wage is lower by about $1'67'
for an actual average hourly wage of $7'14'
1: (2'1.9528) (2.3688) (-3'44621 (9'3'1)
Are the preceding average hourly wages.statistically
They are' lor
(0.oo0o). (o'0182). (0'0006). different compared to the base category?
utttr,"differentialinterceptsarestatisticallysignilicant,
R2:0.0322 as their P values are quite low'
The point to note about this example is this:
Once
where Y: hourly wage ($) you go beyond one quatitative variable'
you have to pay
Dz : :
married stitus, 1 malried, 0 = otherwise 'ctosle
aftention to the category that is treated
as the base
D]: : :
region of residence; 1 South' 0 otherwise category, since alt comparisons are made
in relation to
when you
inat"raiegory. This is especiatly important
and " denotes the P values' with several
have several quatitative regressors' each
ln this example we have two qualitative regressors'
iategories. But the mechanics of introducing
several
assigned a
each with two categories. Hence we have
qual'rtative variables should be clear by now"
single dummy variable for each category'

g.4REGRESSIoNWITHAMIXTUREoFQUANTITATIVEAND
MODELS
OUALITATIVE REGRESSORS: THE ANCOVA
preceding two sections' aI-
ANOVA mod.els of the type discussed in the
psychology, education, and
though common i, fi"1d, such as sociology, Typically' in most eco-
market research, are not that common in economics'
explanatory variabies
nomic research a regression model contains some
Regression models con-
that are quantitaii r" Ird some that are qualitative' variables are called
taining an admixture of quantit;tir" urd qLlalitative models are an exten-
ANCOVA
analysis of rrr.riance (niqCOVA) models.
a method of statistically con-
sion of the ANOVA mod"is in that tirey provide
covariates or control
trolling the effects of quantitative reg.",'o'u, called

tflr; data are obtained lrom the data d

these data
4c.s, Haward University Press, Camb.iag", frf,,tt.,
i9'8' W" Lut'" alieady co,sidered
in Chap. 2.
 305
CHAPTER NINE: DUMMY VARIABLE REGRESSION MODELS

..l

quantitative and qualitative' or

I variables, in a rnodel tfiat includes both
ANCOVA models'
durnmy, regressors' We now illustrate the
Example 9.1 bymaintaining
To motivate the analysis, let us reconsider
that the average saiary of public school teachers
*^y t'ot be different in tl-re
variables that caunot be stan-
three regions if we take into account any
dardized across the regionr. corrria"t t'* public education is primarilv a
example, the variable expendi-
ture on public ,"nat U7 i*"t outioriti".s,
as
the to'", we develop the following
local and state n|!!;;; To see if this is
model:
* gqXi * t'L' (9'4'1)
Y : frt * frzDzi * FzDti

Yi: average annual salary of public

schgglleachers in state ($)
where ($)
{ - spend"ing on public school per pupil
Dzi : f , ifif"itate is in the Nortireast or North Central
: 0, otherwise
Dti : 1, if the state is in the South
: 0, otherwise

The data on X are given in Table 9.1 . Keep

in mind that we are treating the
that besides the two qualitative
West as the benchmark category' Also, note
which in the context of the
regressors, we h;;; . quantitaiiu" ,o.iuble, X
ANCOVAmodelsisknownasacovariate'asnotedearlier'

EXAMPLE 9.3
AND
TEACHER'S SALARY IN RELATION TOfiEGION
;PENiitNG oN pueuc scHool PER PUPIL

FromthedatainTableg'l,theresultsofthemodel(9'4'1)areasfollows:
y,: 13,269.11 - 1673'514Dzi- 1144'157Dsi * 3.2889Xi
(0.3176)
se: (1395.056) (801'1703) (861'1182) (e.4.2)
(10.353er
t- (9.51 5). (-2.0889)- (- '3286)..
1
1

R2 : A.7266

5 percent'
and "* indicates p values greater than
where
*
indicates p values less than 5 percent,
AstheseresultsSuggest,ceterisparibus:aspublicexpenditure-goesupbyadollar,on
go". ,p by about ry 29 Controlling for spending
average, a public scno6iteacner,s salary significant for the North-
on education, we now the ditferential intercept coefficient is
"""ir,ut in"'" from those of
results are different
for the South'
east and North-Central region, but not
(9.2.5).Butthisshould*tU"surprising'forin(9'2'5)wedidnotaccountforthecovariate'
differencesinperpupilpub|icspendingoneducation.Diagrammatically,wehavethesitua.
tion shown in Figure 9'2. r^-.h^ +r-.r^6 r6^i^nc ctatisti
regression lines for.the three regions' statisti-
Note that although we have shown three that the three
for the viest and the South' Also note
cally the regression lines are the same
regiession lin"s are drawn parallel (why?)' (Continued)
 IO PART oNE: SINGLE.EQUATIoN HEGHESSIoN MoDELS

9.9 PANEL DATA REGRESSION MODELS

/
Recall that in Chapter 1 rve discussed a variety of data that are available fbr
enrpirical analvsis, such as cross-section, time series, pctoled- (combination
of time series and cross-section data), and, panel daia. The technique of
dummv vatlable can be easily extendecl to pooled and panel data. Since the
use of panel data is becoming increasingly common in applied work, we will
consider this topic in some detail in Chapter 16.

9.10 SOME TECHNICAL ASPECTS OF

THE DUMMY VARIABLE TECHNIQUE

The lnterpretation of Dummy Variables

in Semilogarithmic Regressions
In Chapter 6 we discussed the log-lin models, where the regressand is loga-
rithmic and the re-qressors are linear In such a model, the slope coetficients
of t]re regressors give the semielasticity, that is, the percentage change in the
regressand for a unit change in the regressor. This is only so tf the irgr"rro,
is quantitative. What happens if a regressor is a dummy variable? To Le spe-
cific, consider the following model:

ln{: flt*B2Diys, (e.10.1)

where Y: hourly wage rate ($) and D : 1 for female and 0 for male.
How do we interpret such a model? Assuming E(ta):0, we obtain:
Wage function fbr male workers:

E(InY I Di :0) : frr (9.10.2)

Wage function for female workers:

E(lnYlDt:1): fu* flz (9.10.3)

Therefore, the intercept fi gives the mean log hourty earnings and the
"slope" coefficient gives the difference in the mlan log hourly Iarnings
of
male and females. This is a rather awkwar"d way of staiing things. But if we
take the antilog of flt, what we obtain is not thsmean hourly *ig", of male
workers, but their median wages. As you know ntecln, *iAion, and. mod.e
are the three measlrres of central tendency of a random variable. And if
we take the antilog of (f r * fr), we obtain the median hourly wages of
femaie workers.
Structural break
In econometrics and statistics, a structural break is an unexpected change over
time in the parameters of regression models, which can lead to
huge forecasting errors and unreliability of the model in general. This issue was
popularized by David Hendry, who argued that lack of stability of coefficients
frequently caused forecast failure, and therefore we must routinely test for
structural stability. Structural stability − i.e., the time-invariance of regression
coefficients − is a central issue in all applications of linear regression models.

Structural change
In economics, structural change is a shift or change in the basic ways a market or
economy functions or operates.

Such change can be caused by such factors as economic development, global shifts
in capital and labor, changes in resource availability due to war or natural disaster
or discovery or depletion of natural resources, or a change in political system. For
example, a subsistence economy may be transformed into
a manufacturing economy, or a regulated mixed economy may be liberalized. A
current driver of structural change in the world economy is globalization.
Structural change is possible because of the dynamic nature of the economic
system.

Patterns and changes in sectoral employment drive demand shifts through

the income elasticity. Shifting demand for both locally sourced goods and for
imported products is a fundamental part of development. The structural changes
that move countries through the development process are often viewed in terms of
shifts from primary, to secondary and finally, to tertiary production. Technical
progress is seen as crucial in the process of structural change as it involves
the obsolescence of skills, vocations, and permanent changes in spending and
production resulting in structural unemployment.
Examples
Historically, structural change has not always been strictly for the better.
The division of Korea and the separate paths of development taken by
each state exemplifies this. Korea under Japanese rule was relatively uniform in
economic structure, but after World War II, the two countries underwent
drastically different structural changes due to drastically different political
structures.

South Korea's economy before the 1950s mostly consisted of agriculture. During
the 1960s and 1970s, Korea began to change their structure to IT, micro systems
technology, and also services. More than 50% of the world uses
a Samsung smartphone, whose headquarters are located there. Today, South
Korea's economy is the 15th strongest economy system.

In the Ruhr Area (Ruhrgebiet) in Germany, the economy was mostly marked
by coal and steel industry. During and after the coal crisis began in the 1960s and
1970s, this area started to change its economic structures to services, IT
and logistics. The city Dortmund opened the first technology center named
"Technologiepark Dortmund" in the 1980s. Companies including Signal
Iduna and Wilo are based there.

Structural change can be initiated by policy decisions or permanent changes in

resources, population or the society. The downfall of communism, for example, is
a political change that has had far-reaching economic implications.
Structural changes in employment

Economic structural changes impact also on employment. A developing economy

typically reveals a high share of employment in the primary sector, while the share
of employment in the tertiary sector is high in an advanced/developed economy.
Chow test:
The Chow test, proposed by Econometrician Gregory Chow in 1960, is a test of
whether the true coefficients in two linear regressions on different data sets are
equal. In Econometrics, it is most commonly used in time series analysis to test
for the presence of a structural break at a period which can be assumed to be
known a priori (for instance, a major historical event such as a war). The Chow
test is often used to determine whether the independent variables have different
impacts on different subgroups of the population.
c+\0w
*ft
1A
Cfi0\rr
- -4

t 1 6O

t5
^
+
op -44;-
Y1rl4, 194L
A;-A

\
4
+i Nttk
k .L
*t A
\ U-.IL* t-
,Arqh
.{- t-
L
"^

^
:e'r1$*,.-
4r+
"A
\!u @ W4
{"
\

VA-w, ryAw
OA/U

l"t )
4 pr--|

=^
-f bxry:(X*YT€
$-
 I

?+-_ Vt:;- 4ul

a----

A 1t- br'Xtk +o,Xde

$*

€-
*
; -eytU*zgt(B'f
+t
V b*- -/

A Qz-
+l (O t

+ Qz-
\

H* \

b ;b
+1 c t

+r ft b\ +bL
e\ c
+1 o

{1 fr
e +C-7 t
 W** {i dn l
v Y"* A,_f)
ficryl \

€
a
b
A,;9
a

-a_, Nw-<-
+-('^--
\b^ {-' c'{\

\lA *ic,
t,c-
C fio\r'r
L (+,-r9 V-
/'

G rq L.
t
t1 ?l L tfrz
f LV- ) C"n r-f )0
 &V bp--
YJ)8) A"Ab
b

I t
Y(A
"9>
I J r^-'Lg

A-tr<$
q ,1 . V---
vtul
p*=
r0\
-Yw

a\A-9
k
Kt5
 I
P nr
Wu
? I r-
\ -''
C

lo"r) -P e;
I Lrs
)

o eb
<-

*\

fu*r* q-
g'oE )
Lb c^rt<-

+{ .c
6 /

.ru
 \'(
ce
<-
t.
-/b <- .-\A9) t
-+L /

++ o