Skewness and kurtosis

normal distribution
MesoRustic

Leftokurtic

Platyburtic

Notation

Eni nitz

Rij mint niz ni min theorderinwhichdouble

Summationis interchangble

End m 2 mini

Stastical undefendence
Two R.vn and y are statastical independent
if and only if

finny
f fly
r

of probability
Characteristics distribution

Aprobability distribution can be summarised in terms

ofits moments
distribution used andvariance
ofthe Two most are mean

Marginalproperty
β
The enfected value denoted
by E x nfa
of v.v
The enfected value of constant Eb b
 E axtb aE x b

If and Yare independent R.VE XY E X ECY

Variance

let her.v and EC U

2 2
Var x 0 E x a

Standard deviation

Var x x
affix if is discreate

Variance can also be Expressed

E x U 2M
E x2 E u 2E XM ECD M
Ex U2 2m
E m
E x2 x

properties
ofvariance

zygyny um
var ax Varch
 2
E axtb aeon b

Var xtY Var X Vary

Var axthy a var x b Var y

Covariance

Defination
The covariance between XandY Cov X Y or ox
by
denoted

is givenby
Cov x y E x 4 114 411
coveniy 4m a Y y

computational formula EEXY F XE Y

Cov x x var X
Cov XM Cov Y x
Cov xD O
Cov ax Y Cov x Y

Cov XtY 2 COV N2 Guy 2

9,2 4 t not yetyou

14
Cov

Var XTY Vanel Vancy 2for my

for N OV EVM i takentwoat atime

forcow
Var x y Var X Var 6 andYare independent
if

let be two random variable with meanUxandMy respectively

then cover y E x Mx Y My
 E XY My MY Uxuy

EEY Uxuy

o
If X andY are independent County
originandscale
Changeof

CoV atbx cidy bdcoven y

Corelation co efficient measures degree

of linear association bw 2
W

P Cov my

Fri Cov ng

oxoy

COV NY P oxoy

Simple linear Regression Model

what proportion of variability defendant variable is explained by

independent Variable

Population regression function concept conditionalregressionfunction

States that E Jimi Bot Bixi

 it is a linear regression function

Linear in variable
Model linear in SLRM SimpleLinear
Parameter BOB
yes No regressionmodel
SLRM SLRM
yes
No not SIRM not SIRM

We can express deviation

of an unduldual Yi from itsExpected value
as follows

Mi Yi E Y ni

Ye E Yin tdi

f E T Xi is assumed to be linear in Xi
then Yi E 91mi
stochastic error term
could
Callunobserved variable that
PithxitMe not be captured hymodel

E Yi ni E E Yin t f Vi Xi Elsini flam

hencethevalue aw
and ofniisfinidviaconstant.ve
made use the fact thatthe enfected value
of is that constant
ofconstant itself
alsonotice we have taken conditionalexpectation
condition whon
given X's
 given
E Yin
E Elgin
E a m 0

Sample regression function

Yi Bi Bai

Yi estimator
of E Yini
Bi estimatorof β

empressing the SRF in its Stochastic form

Yi Bit β thingumated
Sample residual
value of sampleerror
conceptually Mi is analogousto Mi and canbe considered as an

estimate
of Mi
Ordinary least Squares Assumptions

residual term E Ui i
mean
ofthe error o

homosiedastraty or constant Variance term

of error
Var Mi E Ui of i
Serial independence
overaperiod
ofthe Ni implies

oftime
Cov Ma U E Min O
Handom
Variable
 non stochasticity of Xi's thatis the series
ofXi is fined
in repeated samples

Exogently the disturbance term Mi and Explanatory Variable

distributed that is
i are indehendtly they are not
co related with each other

Cov Xi Mi 0
B2 unhart of on will not be accurate if
assumhtion is not mentioned
linearityassumption
The model must be linear in parameter as us islinear
estimation method

We also make normality assumption

forthe disturbanceterm
which is actually required to conduct hypothesis test after
estimation model
of
estimated regression line

yi BE Pixi
Mi yi yi
one possible criteria to select BiandBE so as to make EMi

yr BS pixi o

dunding earation by N

FEE BS BIT
 hence BS andBi should be choosen in sucha way
that it passes through x̅ andT

hence we apply the least Saware criteria whichrequires

βi and BS must be so choosen in sucha
way
that

EME is minimized

Yi BI Bixis minimize
he necessary conditions requires
of minimization
SEMI O S 2EUR 0
JBS dB

2 Yi BS Bixi 1 0
2 Yi Bi Bixi Xi

Yi NBS β Exi XiYi BSExi B EXE

R y
for of on x

The OLS regression line passes through x̅ ST

dending
first normal ear by N
 hey defination co variance x Mi x x̅ Mi w̅
of
Since It o

E x x̅ Mi

Eximi EMEO
I EY
COV Ri Mi Exilli

2 condition Y Bd Bix o
necessary
JfE
2 Exi M BS Bix

Exit x

XLMEO

Cov xi.me 0 Cov Yi Mi 0 SinceYi inadenian

function of Xi
estimated co efficient BGand B can be sequentially computed
using the following formula
 Shelf space weekly sales

6 526
3 421
6 581
a 630
3 412
9 560
6 434
3 4h3
9 590
6 570
3 346
9 672

What if all Xi's are same

foreg 3 what will beits
effect on estimated regression line Ans Biis undefined then

Yon X

β Exigi
End

Substituting
Bi J Bix̅ in 2ⁿᵈ normal equation we
get
 Xi Yi BSExi B EXE
XiYi J β x̅ Exi BI EXE

EXiYi TEX BIE EXITS EXE

Es Xi x̅F
β EX x̅EXi EXiYi TEX E X XP 2Xix̅
BI EX Exit EXiYi EXiEYi
I
Bi Eni Enigi

alternately using defination

of co variance
β
42 2

ñ C

Bi Cov x Y
var x
paroundmean
4ᵗʰ property Total variation in Yi maybe expressed as the sum
of two components

1 Variation enflained by the estimated

regressionline and
bythe regression line thatis
yi yi
2 Variation unexplained Mi
Yi y w̅
 yi yi mi
I
Eye yitMY
Eye Ey Ed i low y.im 0
TSS RSS
Regression Ess
Sunofsamore error
explained unexplained

yi I 5 5
yield sixi Bi Bit

yi Bini
hence we E Bin
can write RSS
EYE can also bewritten as
RSS βiExc
Rss YES

RSS Eng
m
RSS β Enigi
Q ThreeVariable N D Y all have o means and went unit
Variance a 4ᵗʰ Variable is C NTD in the regressional
Cony the Globeis 0.8 In the regression Conn scope o
In the regression Don I Slote ofwhat is sum
of is on
Squarred in con
of residual theregression
of
There are 21 observation and all moments are capturedusing
asthe dunson
co

hunt C NTD

Var C Varen Vanco 2Cou D

Cov GY COUNY COULAY

CoV CN Nar N Car NID

COV GP OU ND Vara

Mi 4 5 Bis x x̅

a E Di 512 20
n

farce β var D 120

C NTD
Vance Vanen Van 2 COVID
I I 2COV WID
2 vCNID
CO
22
 N DY
COV GY 0.8 COV GY
0.8
Varcy 0.9 COV CIN
0 S 100 LIN OOH IOU DIY
var N
0 M COV P Y 50
venial 1
4
0.8 COUNY COULAY
0 5 Var N Car NID
0 4 IOU DIY

0.4 Cov N Y
005 vain C Y NID
005 COV NIP
COV CID 0.5 1 112
Q Yon X

Y X
3 1
5 2
6 3

Verify
if jJ
if 396 4.67
3
 Y 1.5004 1.667

1.9 1 667
y
42 3 1 667

13 Host 7 607
4.67
5

Me Y Y
µ Y β pixi
Mi EMi nos plexi
J BJ Bix
Mi F 9 Bix̅ Bix
Gio

Mi Yi Yi
EY

Since it 0 x̅ x̅

properties ofThe Estimators

SmallSampleproblems

1 Unbiasedness

an estimator B is saidto be unbiased estimator of B if it

mean or expected value is equal to the true
 population parameter B
This means that if reheated sample of a guien size are
drawn and B is computed
for each sample then
the those values would be eanal to B
average
of
However the expected Value B is not equal to β the
estimatoris said to be biased and the extent biasedness
of
is measured
by
E B β

MinimumVariance Bestness
saidto be minimum variance
An estimator is
of Best
estimator if A variance is less thanthe variance
ofanyother
estimator

Var B Vor P

Efficiency

is an efficient estimator is the following two conditions are

Satisfied together B
is unbiased B is minimum variance hence an

efficient estimatoris also called best unbiased estimator MUVE

TILLETT
BLUE Bestlinear Unbiased estimator

Linearity
An estimator is said tohave linearity property if itis possibleto
express it as adenear combination
of Sample observation
Mean Saward Error MSE

Ny j TR Moore

Mean Savarred error property

Sufhose two estimators β and B are suchthat Bias

B Bias β but

Var β Varp
In Such a situation we are
facing a tradeoff b w two
Characteristics This notion can be expressed in formof
MSF mean Squarred error

MSE B E B BY

E B E B ECB B
E i Eci Eci B 2 B E B EG β

www.tbe
MSE B Varis Bias B

Large Sample properties

These properties relate tothe distribution
of estimator when
Sample size is large n to

1 Asymptotic Unbaisedness

E B p
him

2 Consistency

Bin a consistent estimator if 1 is satisfiedand

linn B

The OLS estimators are BLUE estimators

Yi Bot Bixe hi
estimator
MOLS
where Mi follows
E Mc O

E Mi
E Mi.us O
We knowthat
β Eni Yi 5
22 m

Enif J Eni
Eni

niYi
Exe

it follows from here β WE here

Wizz

1 E Wi o
22 Proof

2
EWiXi I

3 EWi
2

4 Bi Ewik we Botaxi Mi

Wipo We Bite EWIVI

O Bill

B Bit Wlvi

taking expectation bothside

 sure

E wine
E pi E B
wine Ew E µ
estimator
Eli β
umbaised estimator
ofβ
Q In a sample 3 observation 7 4.67 andx̅ 2
of
Xi x̅ Y 9 3
2
x x̅ 2

β 3 2 15 βO 1067

Using the normal cauation show that Po interceptidefined

but Slopen not
if too
GB
that
oftheslope
Demonstrate the least square estimators
co efficient in themodel YEP Mi is β 7

ESS EM y y

84,5 Yi 9 Y Y V2
57 yi

E Citmi 7 Yi 9 A

JC 11
 Yi 5 mi 1 5 A

Gi YI A
Regresssum Sanace
Rss of
Solve with hypotheticdataset of 10 observation

to Stoke co efficient will change

byx
intercept will change
by X2Bo

Minimum Variance

We Shall compute the Variance of Stoke coefficent and checkthat

it is lower compared to variance
ofanyothe estimator Recall
B Bt Willi
β β EW Mi
unbased is assumedto hold
B E B Ew µ E B B

Variance B E B E B EE Win

E WEME 2
E WilliMIMI
 Variance B EWEE Mi
2

Varis Freedstandardengages

Vow to home above var B is min Variance

test us consider another estimator B cigi then BE

Ci Bot Bixi Mi

βo it β Ec i t Ecimi

E BP BOECI β EG Xi

We require weight to be such BP is unbiased estimator

ECI O
E a c Came yes
n X x̅

E BP β

let us commute Variance ofB

G Wit G W
 42 w c we 2W Ci Wi

42 EW a we 2W Ci Wi

Ckdcheck
42 EWF 3 1 wit't 2Ew wi

var BP 242 Sw Ca Ew

Wiff
E WE ECG wa Ef Ime
0

Var Bt Var B a we
vance varied
Regression without interceptor

Say the no intercept regression model is

Yi BXi Mi
yi PXi

β YiXi
XR
 theVariance of B no intercept model is obtained as follows
for
I EX βXi Mi
EXE

β BEX EXIMI
12

B β
T o

E F P unbased estimator

B B EXIMI
XR

ECB EMI E E

Goodness
offit in no intercept model

For the model without interest term the Sum residual Mi

of
does not necessary addto 0 like the model with intercett
term identity TSS RSS ESS is no Longer forno
true in general
interceptmade
Q why Mi doesnot necessary addtool
Infact the conventional R formula in Such model may
produce negative valuesin some cases

forthe intercent present model we write

TSS RSSTESS

ELY 972 Yi 974 y 4

for no intercept model we ochlace I byo as the

regression line has three
origin so that

Y 412 EM

In Such case R or goodness of fit

E
R a
Eff

Enigi p
Eye

Stastical on SLRM
Inference

Examination Statastical requires

of Significance
knowledge about their sampling distribution
 g

It may be noted that which is the bloke is

normally distributed with β and ofB
mean Variance

i n B
Em
This result whenthe variance disturbance
is useful
ofthe
term is known

2
In practice hasto be estimated

I
GEESEofregression
no
ofobservation noof harameter estimated in mode Dof
intercept isalso a parameter

Hypothesis Test

We the test by stating that we want to test the

formalise
Validity the null hypothesis that the value null hypothesis
of of
that the value true population harameter Biro
of that is differentfrom
against the alternate hypothesis

Ho β O
HA BIO
 note the null hypothesis is what we going totest and alternate
hypothesis the conclusion
represent is the enherementaltest
indicate that the null hypothesis is
false
To conduct a hypothesis test we calculate the
t Stastic
following
thcalculated
under β 0
off gff methylothesis

E B
FE
The reason behind calculating is as follow

is normally distributed N it is known

as
follows a N distributionmadog
fm
fp
oil
If

now there is atheorem if we have two Variable X X2 which

are independent and have following distribution

in Non
n 22proof
 t using this we can write

Essima

III Eggs
varia
E
it B β

E
level of Significance is the probability
ofrejecting null hypothesiswhen
it is true Also the standard error of estimator is the
standard deviation
of the sampling distribution ofthe estimator

1212 is critical value t with 12 level and n def

of of significance
if Bfalls within the ionfidence interval we donot rejectthe
null hypothesis Alternatly we that when C Finterval
maysay
includes value 0 we accept null hypothesis and makea
descion Is
of stastestially
insignificant
 Goodnessof
We
fit
already

R Regression 55
Total 55

Cov Nina nigi

F f Is standarddeviationof X Y
my

Faire
r C

EBEIF.ee
whining your

ANOVA MSLRM
Analysis ofvariance

we notedthat TSS
ESStRSS
As before a null hypothesis is β 0
follows 212
2
also the regression sum ofSavarr divided by
distribution 242 and a n 2
If Estorsum
assuming that these two X variable are independenttheir
ratio derided
by respective
dof generates the F Statistic
 F 1 11
E n 2

F Embi
Miyn 2
prognosameh

F FI R
n.ie Iyn
I
f Estat care r

for Linear regression

fades sumgaukas
Sources variation
001
i Benisi a RSS 1
Residuals
EME n 2 ESS n 2
n 1

Qi Yi so 2978.5 Xi
629.37
stops
82 0.6149
find Sample size

GAP Todo
Q 15 new result were given varing amount oftraining
between 3 and 12 his Afterthe training their time to
the denotes duration
perform
job were recorded oftraining
in hour and Yis time todojob in minutes Thefollowing
Summary statistics are calculated

7.2 San Xi x̅ 33.1

7 45.6 Syy 160.2
Say 57.2

bo J bin

1.7023 1
Yi 57 857

doesthe data substanciate the claim thatjob time decreases

with hours
more
of Training

iii
Conductsingletailtest

Ans Mi Tss Regss

257 57156
Syy β say

22
54
 2.19
I

Stat 4.48

Critical value at It Significance at Bdo Single tailtest

to 01,13 2 650

Q Estimate mean
jobtime for 9hr oftraining
and construct 951 CF

wedontknow
population 57.85 1
1.70
std.MY
henceusing t
est Xia
Yi 42055

Yi SECY to 025,13
 Say foreg we have two independent Variable n n2 and econometric

Sheafication

Y Pot β it Baxai Mi

Inaddition to all assumptions we considered in SLRM.NO

enact linear relationship between Xi's

Malefleregression
Forexample Wehave two econometric Variable m N2andoureconometric
steafication
Yi BothXii RzxaitMi
In addition toSLRM assumptionwehave

1 No enact linear relationship between theXi's

Sunday
µ isminimised
generating
ommmsationweobtain

p neggy.fm oysgyfr
Yi Not β Xii β2EXzi
EXiiYL BOEXiitB2EXI.tt βzEXiiX2i
X2iYi βo X2i βaEXaE B E XIIXai
Thevarianceof by
stone toefferentgiven

varieil varia
an 1 may
where un is the sample co efferent ofcorelation bwmandn
E
FI
Measuring goodness
offit
R B EMiyigty.ae EnaiYi
R is a nondecreasing thenumberof variable
functionof
independent

Adjusted 62
 dgus

the relevance additional enplanatry variatile

Testing
of
may introduce new variable Inclusionof
Sometimes
we
Such variable canbe justified it increases enplanatrysum
ifrelation
ofdeviseanonet
squares RSS significantly in tothe ESS Wemay
test procedure built on ANOVA technique

oldmodel Yi BotBihithi
newmodel Yi Bo BXii Baxai Minew

sumigsamht.in iian1.4kmanae9s'infamous
gression
Source Dof Means's
Tahoe of Sumof
variation Sawares
RSSdetox β Exiyi I
2
RssduetoXSX2BENiifitBERaig Rssdue.to
addition
B2 n2iyi
of X2 N 3
ESSunthenew Tssnew Rssnew
mode
TSS Tssnw Eye n l
 To asses the indurdual contribution
significance of ofX2after
allowing contribution X we compute thefollowing
forthe of
F Value

F Ssnew RSSold no new explanatoryVariable

of

Essnew n3

Rnw fold
C1 R2new In 3

We add new variable model

in aregression when inclusion
ofvariable
increases the explained sum in relation to residual sum
of squares
also in such a situation the value of increases improves

Fincreases
whenthe absolute value of computed stat the
added variable becomes
of
newly 71

the Valadity Restriction

Testing
of Linear Equality

Yi Bo LBK
restrictedmodel
pun
Logyi dogbo Pilogh Balogkt Mi
 this is an example
of linear equality restriction Bitra 1 in the
content aboveproduction function
of
Ho β B2 1
HA BitBa 1

There are two approach to test the same

1 t testapproach
the computed t stat

nd
fcommuted Bi Bi β B2
SE Bi B2

SEG BIBI Situationist

another aphroach is the F test aphroach

β 1 B Substituting in modelspecification

logYi 1 B logh Balogkt Mi

logYi Logh α Balogkle Mi

 Restrictedmodel

109 41L Balogy µ

This procedure is known as the restricted least squares we

affly the
following F test procedure let Mine Residual 55 in the Unrestricted
mode and MER Residual ss in the restricted model
let me no linear restriction R be the noof parameters
of
inthe unrestricted model and n observation
no
of
Fiomputed EMER 2MEUR
M Fm n r

MEUR n k

If I the structure
can re write
of
dependent

the
Variable remains thesame then
formulationas follow duding neemanddenominator

by TSS

EMER EMEUR
M
MEUR n k

Rur R r M
irr n k
a
The following model is built to analyse the determinant factorofwage

ln wage BitBzeducation Bzexperience Butenure Bs age

M
The research planning to exclude tenure and co related
age highly
with experience

HN BH BS O
HA Ho nottrue

the restricted mode corresponding toHN

O 0

ln wage Biteseducation Bzenteriance Pfenne t

prestricted
In wage 6.157 0.0457 educationt 0.0121 experiance
In wage 6.746 0.065education 0.0267Exh 0.0094tenure
0.0209age
Genome sum g.gg
N 53
residual sum 6.250

EMER EMER test at St

µ
MEUR n k
 Frommeted 1.19 ftp2mg 100kt as one tailonly
so check it atro
ts
3 183

Yi 2.20 0.104 it 3.48 Xatt 0 34 37

3m 0.005 2.2 0015
SE

Explained 55 112 5
Residual ss 19.5

Which
ofslope co efficent stat different at St

me 6H

use Estat
F t R SIRI

1 am
F
54
Company ABC
dollar spent on Shecial incentive has higher incidence onsales
than dollar spent on advertisement

Sales Bit Raadversement 33 incentive It

 n 18

Variabe to efficient SE t P
C 396 59 3548.11 0.11 0.91
Add 18.63 8.92 2.08 0.05
incentive 30 69 3 60 8.51 0000

const daddlu Incentive

const 1258905 26674 7101
Ado 26674 79.6M 2.941
Incentive 7101 2 9h1 12.992

AN Bz 82 0 No B B2 0

HA βz β270 HA P3 Baco
Weare not usingthis

An Sincewell willinclude
stat
β β greater

thaneaualt
TItnaci acod.at varcis
ril
trombuted 1295
Significance level 0 10 isdof

tomputed 1.295 2 toritical 1.34

Accept No
Choosing linear and log linear regression

MWD Mackinnon Davidson whik

You
β
Mo Linear Model
HA loglinear
InY onLux
Yj estimatedLinearmode

Log Linear Log log In4 estimated

logmodel
Log Lin
Lin Log

2 Estimate the linear model and estimate Value

estimate the log Linear model andobtain the estimated values
obtain

21 lnig Int

Regress YY on X and 21

reject Ho of the co efficent of 2 is statistically significant

by usual t test

Obtain 22 antilog Int

Regress logs on logofX's and 22 rejectthe alternate hypothesis

if tthe 10 efficient of22is statistically significant bytheUsual

test
UAOTchecknotes
Consider the following model

Y Bo BiXi Mi

Yiis salaryof ith teacher

and Xi 1
if
doctorate

0ᵗʰ aren't doctorate

mean salaryofgrown ofteachers who
E YIN Es
Po if Kio
E YANI Pots'mdn are doctorate
any groupofteacherswho
This showsthat the intercept is the mean salary of thegrownof
non doctorate teachers
B meanSolaray
doitanate mean salary non doctorate

The sloke co efficient isthe difference

of mean salaries ofthedoctorate
and nondoctorate teacher BytestingtheNullhypothesis β1
teacher 0

We cantest weather there exist anystatastically significant

diff
in Salaries the 2groups teachers
for of
Yi Both X it βaXaitMi

teacheris doct
I 1 if
o otherwise

a 1 if teacheris PGradelat
 0 Otherwise
graduate and
doctorate
in meansalary of
is diff
E Yi X 1 2 0 Bo t B 2
Pranddoctorate
β diff
in meansalary of
Yi Xi 0 2 1 βo B2 22

Df Ye x 0 42 0 Bo 23
mean salaryof Graduate

Suppose Y Both 2 hi
0
2
if graduate
2 L postgraduate
if doctorate
2 2
if
E y 2 0 BO
E 91 2 1 Both notgood
f 41 2 21 Bo tag

Interaction Dummy

For example we needto examine variation in clothing enhenditure

of individual oftheir gender and

interms education

Y Pot BiXii Batait Mi

Xi I iffemale X2 1 if doctorate

male
Of 0
if otherwise
effect these dummyvariable on clothing expenditure canbe
of
mulflicative also not only additive

In case of additive model that the differential effect gender dummy

of
is contrast across 2levels education dummy andVica Vera
of
If We Relax this assumption and tryto recognise theinteraction

Yi Both it Baxai Bz Xii Kai Me

E YeXp 1 2 0 BOTBI
B2
E Yi X 0 X2 1 Bot
E 47 1 1 2 1 Bo BI Bath

β diff effect ofbeing afemale

β 2 diff effect of being a doctorate
of β3 diff effect of being both doctorate ffemale
Dummy Variable approach
Structural stability
chowtesk Simple stat athodgkas awhole

done
fora very longtime frame
co efficient are stable or not

Rich Poor s neednotbesame

9
C dotaxit Mi B Bixity
Niobs N2obs
 Ho do Bo and α BI
HA do Bo and α B

Accept Ho Structural stability

RejectHo Structural inability

Steps Use all nitna n observations toestimate the followingmodel

c Xot Xi Xi Mi and compute the Residual Samofsanases ESS
let it be Essn
dof n.tn R n R

teh 2 Run Regression

for each underdually within and's
numbers obs and obtain their respective ESS
of
Essn and 55m

dof n k dof na k

Sep 3 Compute following F Stat

Fr nitna 2k Estat Essn Essm Essna R

Essnit Essn nitna 2k

StekH followthe descionrule

If Fcal Faritical reject null to
 are not the
if reject Ho then the 2 regression same nostructural
stability in the consumptionfunction
Essn is very high unexplained ESS is
veryhigh
limitations

Chow test captures both intercept and coefficient

Asa result dummyvariable athroach

Rich Rich
800
Both poor

Bo
Ix
ouch

poor weestimate thefollowingmodel

2 isdummy
for c Bot Bixit Xia the
rich Poor

In Model 1 C Bo Pixi 12 Mi
21 1
iffamily is rich
21 0
if otherwise

It followsthat rich Bot Bixit Mi 2 1

Poor C βotBiXitMi 21 0

onlydifference in intercept
We are postulating that therich has has higher Subsistence
level of consumption

Wheather such a changein Statastically Significant or not

candee examined
by estimating the

c Bot β Xi XP Me

and testing the Ho 71 0

Stastically Significant
HA X FO

we reject Ho theconsumption
If for are deft

Dummy variable modelto enamine diffin Slope

rejectNull thenthese
C Pot BIX zix Mi emit deffinStoke
co efficient
c Pot BAX Xt Mi Zi l andthereisstructural
Bot Bixt Mi 21 0 instability

3ʳᵈcase
c POTBIX 2x faithi
e Bots BAX X Mi 271
Pot Bix Mi 22 0
chook cost Number
Tyne
1 349000 623
2 537000 6553

3 R 170000 400
4 0 526000 663
5 R 7,001000 563
6 R 28000 236
ORR 1401000 307
8 0 120000 146
9 0 611000 99
10 0 2461000 520
11 0 4137,000 453
12
R 1,501000 350
13 R 140000 423
14 R 130000 352

Assume model
of cost on number
conduct chow test in cost function acrossbschool
forany difference
tyke

1 cost on number
212755162126
RSS
forwhole model
MSS
forError 13037903156
1011
SSE 1 524 54838 ofenteredated

Estate 16 32
fz.ms
2 take all lo's together and R's together dono keosehostly

For Occachational 101

MSS Regression 169630136559
FStat 18.193

F
MISE
ESS 4.661961649 1010

For Regular R

RSS 888973520
7415771422286
TSS
1010
Ess 1326874 077
dofn doÑ data
p
R simplynoofParamet
F Essn Essai Fssnz R estendt 2

Essnit Essna 10

Fomhute
Ferictical.no rejectnull

F 8 062

structural instability in there

 Hetroscadasaty

Consequence
of Hetroscadasticity
Var Mi of

Unbiasedness weknow fromSRM B

β Enimi
nc2
hence Expectedvalue
ofBaro
ECP ECB
MÉEY
ECB β

hence estimator stillremain unbased no of component is involve

2 Bestness property

2
Var B ECP B

EE
E Enini E Mi of

E Enim 2 mini Mills

2

all 1
 2 2
n µ ph roscadastc underhomoscedasdh

Var B ME Ri valid
In
o
2
Eni

ki is avariable with unknown Values

bestness property is not satisfied

β is no Longer a minimum variance and test estimator itisonly

unbiased andnot the best hence it is not efficient
Consistency

To prove this under Hetroscedasticity Suppose

Or i
EISEI
of Oi

Var Bmeiosa Em
ENE

m oi
2
Eni
 Enioi In in
I Eni 4m

Covariance between XEand of which

is refresented as
sniff
is finite number
Egf is also finite

as n s

var B o hence consistency from Ci

satisfied

est to check hetroscedascity

Goldfeld Quandtest
Breusch Pagan fodfrey test

Varch of Saphose thatis variance termis afunction

of error
of non stochastic variables

detect beltroscedasticity
Step 1 Estimate I Both I the byors and obtaintheresidual

5th 2 Runthe Aunitary regression of the Savaresof Residualson the

Control Variable

Regress ME on
M to 1 1 9

Steh 3 construct theNull hypothesis and alternate hypothesis

homoscedasticity

Retro
Stehm Compute LM nR where n is noofobs Used to
estimate the ausatiary Regression model and R is the
determination the aurahary model
co efficient
of of regression

This LM stat
follow ftp.figihfuuendentvanaue used tocentimetres

N ntrictical reject Null and conclude thatthere

if computed

is significant evidence
of hetroscadastudyundata
Goldfeld Quant test

The hypothesis are

2
Ho Variance Me Xi
HA Var Milk of

eh identifythe Vanalle towhich You suspect the varianceof

disturbance termto be related

2 sort the raw data in ascending order

3 cutof some central obs breaking thedata set in 2 distinct

set

H
fit byOLS first and LastStehof
Scharate regression tothe
observation
5 compute residual sum Samares
of oftwo regression
and comfult Fstat

F E552 dL
Essi dfi

if Fiombut Foric reject Null and conclude Hetrosyd

in Significant

Matiollenealth

Yi Bo Biri Bnaki Mi

III Ii.IE

in
min U'll β xx X'Y
 Variov matrix
Van B E BB BB
x1 x f u'm x1

62 X
CRACK inflationfactor
puanance
Bj 02 WIFI
SST 1 Ri 5
jan

02 E ME
SST
Ef nsi ñi

SST Var ni XR

Ri in there fromthe following equation

KJI Not Tmi Tanai TKNRi G

thereis no terminRHS

Nyi Not Tini TRRRI G

2
Rj 0.9
 Rita

Endogenity problems

ill omittedvariable bias

Wage BotBieduct Paability
wage Bot Bieduct Mt

UP Paability till
violated
educ but this get
5hPMassumption µ
Cov Raability M educ o

Bo BatM
y
B 0

n
Cov n Bot Bath
 B B 104,4
ECB B Since couln.nl

ErrorinVariahhty
Properties
ofthe instrument
Relevance
1 Cov z my to
0
2 Cov ZIM Exogenity

o
EMI
to
yi
BoI
β J FIVE
Both M
g
B1 1
4221
E GE
Y
 21 2 m n

his instrument
ofitself

ECB o β BIV.EC I
nt.YY
Bau Bit
EI
E BTV β E Mi M 2L 2
E 21 2 Nc x̅
β
Since Could 21 0
Weak Instrument couca n 0

Botbin till where coucn.nl to

y
plum BIv Bit
7
43
Asymtoticbias

L t
C thiswill absords this
willalso
 corrcn.nl allaretve
avg.es
weak instrument then itis possible
Ifwe using
OLS Bias in less than

FUBiasoffice
com
2n
 moments
Generalised method
of
Two stage least Savare

y Bot Brit Bana BrenMar1 Breya M

Cov 42 µ to

Let 21122,23 22 are the instruments forgz

Reduced
First stage form fauation

Ya TO am tama t Apr1Mar 1 t AR 2 Artrizot

From stage one we estimate

y2 A aim AIna Ap_ Mr I ER Akitrer

Toufya m to
but
couly M o
 HO AR Akito c

take k associated with 2

IV is special case 2525

a
of

If no
of instrument in 1 then IV and 2525 are the same

Bo Biya M
Proof
y 0
00
92 M

Suphose Zis instrument

First stage
not A 2 E
92
y at Az

Second stage
BotBigit µ
y
Picasis
C0Yf
 couly nitric
V Yi

AT Cov y z

FF var a

c rivance vary
couY
Masn
T

Q y BotBint Banath
ou n 1m27 0
and min unobserved

omitted variable bias

Error in variables

reported
y
y Actual

BotBrant Ranat M
y Actual value
y Reheted value
y
 y Bot Binit Bama M E
covenaM7 cov mm o
yo y e

var M e via tule

POTBIRA M
y e ni nit

y Pot Bilni c M
Bo Biguilt U Ble

Coulmu O
If eou m e no endogentlyproblem

Lassical error in Variable Assumption FV

Blackmony audactualIncome uncorelated
cov np e

cuu ni e e 0

Cov mile oe

lov 24 M Bie covenum β foulane BI couch e Bioe

TÑÑÉÉÉ
cov nie o

coven e of
Ifhaveof
one the explanatory
Variableis related with error term we
endogenity problem

β Bit cov M Ble mn

V m

β
E
β E2om

AE
ne ilimBi Bi
oez

Attenuation Bias

at
e

Plin β OBI

B theestimator will be undervalued

care 1
If is ve

B1 the estimator will be overvalued

If is ve
 1
IFI
Application

ln wage Bo Bleduct Beexterianet Bsemperance tabulity the

ability is unobserved

tests bserve d
rialulityte test2
kablitytez assyinfaT.es

Botheduc Baemperand Benterance AM

y testff
Botheduc Baemperand Benfenance test 1 U αe
y a

Cov test 2 M e

by CFV assumption

Cov ability o
e

cs
e

Cov test 1 e 62

Cov testi M αcou testine

0 does
Run Regression with tests as an instrument oftest 1

Cou testatest 8,22 ability

cou tests le del 0