Weeks

Joint probability P x
y PIXztg.Yhm.EE Eetr

PH 0 1
Conditional
probability P X 014 1 41,4

Conditional variance Var Y 0 E Y4X 0 ECY 4 012

Pa o
Efik
FLY NOTconstant on butdepend on X

Law of iterated expectations E Y ELELY X

0.40.6

E VN ELE UNIV ELV.ECNIV

E V O ECW V O ECV I ECW V1
0.0.4 4 1.06 7 4.2

Elatbx 4 atb.EC ItElu atb.ECX tELECUlX

542 3 ECO
67
ElatbxtcYI atb.EC7tC.ELY Var if 47172bc Could s
Var atbx cY b Varix c Var 4 ZbcCorex Y
Coulatb b Coux s c Cover Y

E.g 6 Could Y 5 Cover Y

S atbx b 52
satbxtcy bs 5xtc.SI
Zbc.Sxb.SxYtc.Svy
SatbxtcUiY Cov Xi Y
6 5 4 5 Sry 3 Var Xi Cor Xi hi
OF P Var Xi

E unbiased consistent
E 9 D ECG 0
e.g ECT MY
so Y is unbiased
Var f 0 ash co

Var 8 E O 012
E SxY Oxy unbiased
Sxy Oxy consistent
 Week 6
is regressor independent

Y is regressant dependent
constant intercept
Po is
is slope
β
ELY IX
ftp.xi
41 130 13 Xi Ui

E YIX Ui
LS estimatorEti 71713 Y B
β
Fitted line 7 f 13 Xi

First order condition 8 2 Yi 130 β 0

2X Yi β 131 17 0
STATAFx

1131 7130 Yi β 13 x

Prove I
1
B pix 135
57 7 157 13 T
Y 9 Y 4 4 7 0
S n O

Sta 5,3 13.44 13 5 4 0

Prove TSS ESSTSSR

use 594 0 KTSSHA.GS

a 54514 54 54
S4 a 54 5 2
Yi 5 4 F 214 IT
Due to 5 5,5 0

Yi 5 4 Y 24
Yi 5 472 25
TSS ESSTSSR

ANOVA Table
sum Mean square
ofsquare dg
Explained ESS 1 ESS

Residuals SSR n 2

Total TSS n t 9
2
STATATE

3
ñ
n t
RootMSE SER
 0

22 AHEKE I unitless
fly 2 042241
22 0 ESS o not fit at all
R 1 ESS TSS Perfect fit
SSR EAT CUT Fi UP
SEA 214 5172 Iii 2
Standard error
of the regression
STATAFx
22
SEA

22 how close the points They measure the strength of relation

how steep the line is
β
Roles 7 8H47 FAKE YEAR 41 V z MY
LSA Elni Xi E UI O

Xi Yi for Xi Xn are i i d
Large outliers of Xi Yi are rare

I LSAI ECUINI O
ELY Xi E ftp.xituilxi
p β Xi tE nilXi Elul o
p 131 1
Equal width is not the requirement

Xi and Ui must be uncorrelated

Xi is biased
If Elui 0

Prove Cor Xi Ui 0 Hence corrlxi.nl

CovlXi.ui
ElXiui E Xi E Ui
E Xiu E Xi ECUI Xi E XI E Ui
ECX.DE UI E XI E Ui O

Yi 130 13 Xi with EcuilXi Cto

41 130 2 131 1 ni c

Eln CIX Elni Xi C O

 F LSA 2 whendata are recorded over time

Nontiid

F LSA 3 Rule out large outlier ATB

strongly influence the results

E B1 is unbiased
ui
Var f Vary
Var f 0 as n as is consistent

unbiased consistent
β
IIti Prove is unbiased

ECB.IN Xn ECE Xi Xn El IX
4
E
I 1 1 Xn EC 714 11 lxin.tn

IN Ntu
EE In Xn p

E B Xi E Xi El I IX
E EX III Y
β B IX
ICN E4ilXi 13
AH.EE
β unbiased
Prove is consistent Firstly 0 8
Because Oxy Oxftp.xitu

i ieI
Next
β 5 p
IE β 8
OXY 0 β
1 13144 β Oxx COV Xi Vi β of
5 β
β 8 β Hence consistent

β and130 are bivariate and linear

Why do LLNand CLTapply to and

5
I En
2
When n 00 SE 1 6 2 0
7
i En 1 is i id
sothatLLNandCLTapply

Regression ONLY for linear relation

 Week 7
Makes more accurate Bigger samplesize n

Largervariance of X
Smaller variability
oferror Ui
Leverage weighted error vi Xi Mx ni
The slope estimator is moresensitive to Ui at extreme Xi

For heteroskedasticity robust

1 Model Y 13 311 4

E UI Xi O

Xi Yi for Xi Xn are iii d

Large outliers of Xi Y are rare

Large n

2 Ho 13 13110
141
β 13110

3 Bs
E it dy n 2

4 t tent and tarit 47T dy

5 Tobs

6 7
 Ht Testscore 698.9 2.28 STR
70.36 0.52
SEC SEL

Determine
p value T.EE
k P
value 2 PCT7Itobsl

2 P Z tobst because he 420 is large

2 PLZ 4.38 0

X1 E
X O 4 130 13,0 4

β TD I TDs ELY D ELY D 0

70 1 1301131 FD 0 130 4 β

Es
131 657.4 650 7.4
SEE 1.8
 LSA 4 for Homoskedasticity Var villi is constant
When n is large use heteroskedasticity

When var milk is unequal SE is heteroskedasticity

When var villi is constant SE Sp is homoskedasticity only

f
Ni DSi no 1752
Sp Msthi 2
Fooled variance estimator

Theerrors are homoskedasticity

According to LSA 4 constant VarluilXi

Gauss Markov theorem

is BLUE thesmallest variance

Var Xi Xn Vart Xi Xn
Modern inference large n

Classical inference small n

GET AREAFEEIn 1

FIKE for homo

SE B 8 in se SE Bo V8 in
I for heteroskedasticity ASE B SE Bo