statistics 2022

Types 0f variables =
Quantiles

µ
Nominal (Nordet ) :
} Quantile Qlp) of a sample is the value
•
Sex, race , haire airline name Definition :
the P
-

,
CATEGORKAL
.

Ordinal (arder) level ofeduc rating passager class such that at mosta proportion
score ,
p is smaller than Qlp)
• :
,
,

Discrete canon / taire articula values but can be Infinite & at most ( ) of the data
larger than Qlp)

}
: is
/ tp
•

y ,
avant, #µ
.

& canon Guay by fixed amounts siæofpop no of

passagers
: .

,
Or
calcul .

_
) sort the data (
ascendingoider)
qq.nu, µ, ,,
ang.am, aaawn.in,, my ,,, ,, , gmpw.eu
,
an , =p ,n.gg ,
•
un , ,
,
ange
,

point /if not interpolation)

temperature , responsa time , BMI ,
incarne , frelons . 3) Ifrinteger QLP ,
' is A data .
: linear
'
=

Measures of Center

)
Properties ( unimodal distribution)
Mode : mot frequent value (for discrete variables )
highest frequency (if classes of same width) médian
class interval with symmetric distrib mean
• : =
.

relative t'Queney in the class

class Interval with highest density ( Class width ) •

right stewed - " : mode < mediante mean

Médian M QIO 5) left skewed " mode > médian >

:
=
: •
mean
-

n
✗et + ✗n
☒ f-
- - -

Mean : or average : =

,
☒ =
n

Measures of dispersion Frequency table

/ ✗si ( Xs ✗n )
Range )
classtp.r.gg#mf9--l9
• : Max . .

, ✗n -
min ,
. . .
, Frequency f- Rel .

req .

Interquartile IQR Q3 Qd
range
• : = -
+
+
n

•
Variance : s' =
¥1 Et ;
✗i -
☒Î
"

• Standard deviation :
s = v52 Total fr9 .
Total 1
=

Histograms graphical translation of :

a
fsequency table
freqencyorrel.frequescyrel.fr
Area d- each proportion al
rectangle is to each class

Height
"
is lqual to the
density
=

amplitude amplitude width of class

eq
,
where
-
-

Boxplots Q Q plots
-

1) Box between quartiles 01=010,25) and ☒3=010,75) A Q-Q plot (quantile-quantile plot) is a graphical method to compare two
2) Line at the médian M -
-
QU' 5) probability distributions by plotting their quantiles against each other.
The two distribution to be compared can be either theoretical or empirical.
3) Lower bound LB =
A1 -

1.5 ( Q3 -
Q 1) Most often, one is interested to compare an empirical distribution (from the
1.5 (Q3 Q 1) sample) to a theoretical one.
4) Upper bound UB =
Q3 + -

In this case, the ordered sample is plotted against the quantiles

5) Ends howett tatum
'

of F- l 5) fort 1 in
"
whiskers biggest than LB =
: . . -

,
.

smaller than UB where F is the cumulative distribution function of the theoretical distribution.
highest datum If the two distributions being compared are similar, the points in the Q-Q plot
6) Data outside symbol
of Whiskers represented by a .

=
will approximately lie on the 45 degree line.

Sampling schèmes

•
Simple : random sampling with equal probabilités ( Ex age classes: with proportions of individuels
• Stratified wrt / one or more variables : then simple random
sampling within each stratum in each class
,
tatethatpvp.to obtain sample of
n °O° )
Clustered divide
sample the groups ( all the members vept in the sample )
:
•
into then
population groups , we

•
Systematic :
tatie observations at regular spacing
Estimatoire 1 E- (✗ 4) = ✗ -
ELU)

Est imatoref µ ECXI

= ) En ÊXI 2 ENTV) Elu) + ELV) =

Properties : E- (E)
µ = =
EH Ê ) ? ÎE .ËH=±ÊEHil=âlµ+ÂÏ÷ˢ)
.
✗i -
-
En -

µ
=
µ 3 Var (XU) = à Var ( U )

Var =
=

varfn.E.it?H?varl?.xil--1-n.ElvarHiD--n- ITÎËÏTY fini -

-

a Valu + v) Valu) + VWLV ) =

if U and V are independent
Central Limit Theorem
Reminder if ZNNIO 1) then ✗ : ,
=
a + BZ ~ N / a) 62)
let ✗ 1 ,
. . _

,
✗n be iid ~ F. We demonstrated : E- (E) =
µ Val ,
= ¥ ,
with µ
= Et ✗i) and t' = Var ki) < •
,
i -1
-

, . . .

,
n

# N µ E)
À N / 0,1 ) À ( ( LT States that I benares like Normal random variable
Then µ therefore ifor large sample
a
sites :
-
:
,

oyrn •
× .

for large values of n / K) ,

the number of variables involved in ✗ .

{
1 with prob p
Estimating proportions ×; Bin H , P)
-

=
Xi with n =L trial
prob d- p)
-
o with .

Here µ ELXI) =p 02 Valk) =p( 1-

p)
-
-
=

Î À
" "

=
is the sample proportion of 1 ,
and for large sample sites :

Îappîix N ( p PI ) P'
-

Other estimation for µ exist , for example :

Ethel truncated -
or d- trimned meam : the meam after discording proportion a ✗ of the observations at the
high and low end distribution .

• if the distribution F is
sym metric then
µ
= EN :) = m where mis the population médian .

mediunli) ) )
, ,

À = médian ( Xi ,
. . .

,
✗n ) can also be used as am estimation of µ .
( similar / y ,
other estimation for t'exist : MADHI ,
. ._
Nn) medias (I Xi
-
_
-

.
 Chi sq distribution .
.
:
if Ki , .
. .

,
Xn ~ N /
µ, o
'
) ,
we have
T' Van Xi )
Estimating Ê
"
2 alternative estimations ×
E)
=

/ ~
yn -
r

Êz
i

Ê
-1
-

s'
=p ? , ( Xi It and J' =L / ✗i À Voulx) 2d
nyj ECX) =D
- -

, If ✗ ,
then ,
-
-

=
#[ n ✗ 2- ni ] =
# { ✗ 2- j']
of and 52
'
There is result on the distribution
no
general s

Duyn?
( ) 52
isdistributedlitedozkn.se)
n l
From aboule Yn therefore s'
-

:
~
i with ,
oz
-

similar / : n J2
Ùn and Tis distribuée litre Dorln
y
~
,
p
.

Estimation of the variance 02

Let ✗n , . .
,
✗ v. 6e kiidr.ir .
~ N ( µ , 02)

Propertg:* Ô=Ïç( ×
Youpi ,
→ ECQ~t.li-1
Consequence regarding the 5 estimated
Remind : chi -

square distribution
" Z, -2k iidr.VN N / 0,1)
From the about E( 5) 0 , ,
. . .

Proper tg
: =

Def : a =
ËZÎ is distribue d. according to I distrib with kdegenotlreebn
EIÂ)=E(¥Ê( x-p K
.

s'
Ë ai ☒Y ✗i 1 anti
- -
- -
= -

EIQ.tk
Properties
:
,

ÊEÇÊ ai x-p ) K 1
Var ( Q) -2k -

- -
-
-

v11 EIÊHi ) 02 ( 1- ✗121=0.95

'
I)
Typical ✗ 0,10 and Zi 1.65
-
=
=
= -
an

Ets ) =P
[ # Épi À/
,

✗
=
0,05 ( n d / 21=0.975 and Zr 1.96
02 alz
- =
☒
-

=L - =

Values :

✗
=
0,01 ( 1- ✗ 12 ) :
0.995 and 2- 1- * =
2-58
Confidence Interval
Pf µ ☒ ≤
} 1. Ipe
-

2- ≤ 2- 1- ✗
Confidence Interval for µ Elxi) when O2 known
prob 1- ✗
=

µ with
-

-
-

,
Lovers .
Nrk

CI [☒ ± Ze E €] Xn , X , Xn EdF distrib P { -
Z -

Ê ≤
µ ☒ ≤ Z
Ê} 1- a /+ I
µ
-
=
E
-

=
-

- ..

,
.

INN / µ , E) →
Ê- NNIO 1) ,
PH -

z -

E ≤ µ ≤
x-tz.EE }
-
-

1- ✗
Covers

À =
f.Ê ✗i
P { [ ☒ ± z E ]ȵ } -
=
1- ✗ ☐☆
.

2 Confidence interval for µ=E(✗i ) when ✗Unknown

tu student distribution with ki degree offieedom
by its estimators2-n.IE/i-x-)2
~

Replace
, ← >

CI =/ ☒ ± tn IH E) Fn ] ⇐ exact confidence
'

c- level (ta) where tn is the Student distribution with n -1

degieesoffrædom
- '

µ
-

,
.
, .

If inSleat of tn 2-e- an the CI Stil valid but with level of approximation

we use
,
is ,
ahigher .

3 Confidence interval for a proportion p (Bernoulli distribution Blp)) Eki) =p , Varlxi) =p(tp)
[☒ ± P '^ P) ] P¥P)
-

'
e (I = 2- 1- ✗ if p known unbiased ELI) Var (5)
n
=p
=

p F- ,

CI =
[À ± -21 an ] if pis Unknown : ÙÂ matos I. ( LT :

Ifm »
N (
p , PÈN)
RE
-

✗ Ê density function
,

Confidence interval for ,Xk~iidN ( µ À

ËÏÎ
4 the variance 02 ( normally distribued data :X , . . . ,

[0 HE
1) S2
)
'
J2 E (I (n
] ( 1- d) confidence level (" s' from pwpesty
-

=
,
~
aboie *
: , ( a)
S2 yn
(Knil i ËÊ]
l)
✗i. irai
n
T'
-

C- CI =
-

5 Confidence Interval for the difference of Mears Xn , . . .

,
Xnx Fid Fx ,
Yn , - -

,
Ynx Faty
1) same variances µ ✗ =
ELXI) , peu =
ELY;) Var (Xi) -
_
Va /Yi ) = 02
,

c- CI ( I I) [ ± 2- 1- au
spa %) ] il t'× & Fy N then
replace the Ze alz normal
quantile
= ~
µ× µ , + we
-
-
-

( nx 1) SÎHNY 1) SI with the tnxtny (1-012) quantile ( exact distribution)

sp
-2
- -

where =

nxtny
-
2

2) diff . variances
À NN replacethe 2- nan normal
]
¥ if t'✗ & Fy then we
quantile
CI =

[ ( i 5) ± -

Zeon
nx
+

My with the
tminlnx.im,
,

-
i
) ( 1- N2) quantile .

Theory of estimation we assume ✗ni . . _

An d F
if bias -0 -

theestimator is unbiased .

Bias bias (Tlk , . . .

,
Xn) O) ,
=
EITIX , , . . .

,
✗ n) ) -

O if bias < 0 Dis underestimated .

if bias > ◦ 0 is over estimated

• Bias of I for µ ECXI ) bias / ☒ µ ) =D = :

of s and J2 for T when Xs ✗ Y N/

' '
• Bias
;D µ ,
. .
.
, n ,

•
Elsa )=E( n
! Él Xi -
It ) =
n (Ë f) =
Es ln -

a) =
02 bias (s' p ) Eté) E- 0
'
-
-
-
unbiased (+ consistent)

• E- (5) =
¥02
=D bias (5,04--5-152) - t' =
02-02=-1--02 bias that deweases with n

estimation smaller variance

we also prefer with a

• Val 5) =
Vain? ÊGIXI ,
-

il
'

) =

(m )
^
ator /È / ,
"
f) =µÎ un D= -

n
?, on
and similar by
Un 1)
Var ( J2 ) on
-

• =
h2

"" "
Given that n ? >
nn ( À) .
> À Vol < vols)
 Meam Squared Error MSE (TIK , . .

.gl/n),0)=E(lTlXn...,Xn) -
O ))
MSE MSEIÔ O )
Decomposition Var IÔ) + biasio O) .
=
,

MSEIS F) Var / 5)
'
? T = =
"

"ᵈ
n , ,

( ¥04? n2 ton
« t'
MSE ( 52,02) Var (5) + biasls :o) !
2N n -

"
-

☐ = +
n2

we can check that ¥ > YÉ for V3 talwegs !) therefore MSE / 5,04 > n > , MSE (52,02)

MSEI 52,02) ÉE where à ÎÉ (

: J E- ÊË Ai -
x-p =
-
-

E
bi / J' v2 )
as El J ) O2 E ( ÊQ)
=
¥ EIÔ) Ê( ) F k
'
-
= -

O
'
= - = K -
i - =
-

Ê " '
"
Var ( EE) (Ê ) UK 1) ¥
"
var / 54 vor O
= = -
=
=
-

MSEI 52,04 Val 5) tbias452.si) -

2¥:"
1- E) 2k¥02 =
o
"
+ =

'"
É )
"°
Efficiency of estimation Twrt / u :
eff (IU) =

MSELT
( if both unbiaxd :
elf ( Tir) =

Var (T )
)
Bydel : most efficient estimation is the one with the Smallest MSE

If effltiu ) > 1 Sag that Tlk , Nn) is more efficient

,
we . . -
than UH ,
. . .
,Xn)
If eff(TN ) < 1 we
Sag that Tlk , Nn) is less efficient
,
. . -
than UH ,
. . .
,Xn)
Consistence :
if HE > O PC / TCK , An) il > E) ¥0 :
. . _
-

( Media is more robust than meant

/
constructions estimations • µ
-
-
EN =
% ✗ flxldx Maximum Likelihood
Method d- Moments ÈH)
• E-
www..jo?lx-peYfHldx--ElX ) -

1) we farm the livelihood function as a

produit
µ Et ✗
"
/ =
§ ✗
"
f-(Hdx initial moment of Kader LI ) =

Ëf f. (Xj) ,
where f. is the
probability density function .

( phi FNÈXÎ
A

) ( EIXIÎFCXIDX
-
oo

E- (X EIN)
"
centered moment ,Koidu

{Ênifo
pu x
-
- - -

-
as
1/01/11 , ✗n) =
Hit folk ) .

folk) .

.. .
-

folxn) : Xi continuous
ta :p ,
. . .

Poki ) Polk -
-
✗i
-
-
-
-
)
H :
'

Pdki ✗ ) n : xi discret
empirical sample of Koidu
. .

and .
- '
moment

I
Compute the moments needed of random variable 2) We talk the
log of the function
p a

First moment : EH
µ
= lol Xp / Xn ) log ( V01 Xp ✗ ))
. . . = . . .

,
n =

{ ËÊ log
second moment : EUH Vaut = + EÎX) = J' + pi = log fohi ) ( ) ✗i continuous
2) In the (Pol ✗i )) discrete
equations aboue replace ✗i
-

xi
-

, ,

• EH) by ÎÇÎ , Xj 3) We desir ate and =D ,

( solve the first oder conditions)
ÎÎÉXI J log folk )
n

{÷
ELXY by I
Ë XÎ
'

• • =
Âmn Sl / 01×1 , ✗ a) € so ✗i continuous
,
◦
. . .

= =
,

µ by ÛMM ÷:* . :*. .. .. "" " " .

Jo
•

6g âµm
a
so
,
• T' 68 Ômm Illustration Xn , . . .
,
✗ n Ea
Exp (b) . f- (X) =
dexptbx) (PDF)
◦ Hain Âum =
ÊËÉXJ log( f-( x )) =
log ( d) -
DX

Fin =

I. Ë Xi ¥ ÊX;) -
?
=
ÊÏÉÇ ;
I)
-
'
l (d) ✗ii. →
✗n) =Ë( hogld) -
Hi ) nlogld)
= -
d. Ê ×:

În
.

Î Ê
,

.""station :
Doivative : -
✗i
=
°
,
=

ê÷ ¥ =

Letti , .in?iaExp(d) density ftp.jexpfyx)

,
-
. . with : ×≥◦ i =)
,
Here ad ,
p -1
Hypothesis testing
galper) ¥
.

We have that
µn Elxi)
= : ± , therefore On = D= =

Binomial test proportion

That impies that Inn =

¥ =
¥
:

Ho :p -0.5
☒ we test the null
Hypothesis
-

on
against the alternative hypothesis Ha :p > 0.5
• The test statistic used is X .
The distribution of ✗ unde Ho is
Bin / n'-25 p -0.5 ) ( or NI 12.512.52 )
approximation by a
-

• The
sample value of the test statistic is v. dos .

Es The p value is completed vndet Ho

-
:

)
,

( 5) Pto /
✗
[¥ ≥
> v.dos -12s
PHO / × ≥ v. 06s) =P µ Z>
.
=

.
≈ . . .
. . .

2.

2- NN / 0.1 )
Type I and Type I error • une conclusion : .
. .

TOTALITES
i p value the
propability under Ho of Observing Somethin
:
-

false positive at least as extreme as what has been observed in the sample
B
The conclusion :

false négative 5%)

☒ small
p value
-

leg < :
evidence
against Ho
we define probabilities associated with each situation
rejet Ho (testis significant )
:

✗ =P / eject Ho / Ho true) =P " ( eject to) ← Worse !

against Ho
◦
bah no evidence
large value :

and
p =P (don't eject Hot Houot tue) PHA ( don't eject to) p
-

don't
eject (test is inconclusive)
 Critica value
2ᵉ
"
if µ > ,
ka :(1- x) -

quantile of distribution of I undotZ '"

the
kx-ipeot-zi.at/rn
rejection region :X > kx :
PH.LK > ka) =P, ↳ ( ÉTÉ > =L -
Y÷ )
• if µ < 2 Ka : X -

quantile .
,

K¥10 )
Zx Mrn
☒ ka PHOLI < ka) =P,"/ < a kx≈µo +

rejection region <

-
_
: :
where 2- ✗ = -
2- 1- a
→
if µ -1-2 , Kin and KI we the # 2) and H N2) -

quantiles
PHOLILKÎ ) -42 -
and Ptidx > kà ) -
-
N2
Relation ship between
p value and ka
-

smaller values of ✗
larger values ofkx :
rejet Ho less often but reduce Type I error
with p value
:

} TMP
-

☆ if p value < ✗ -
then to is rejected -
value is the minimal value d- ✗ which
gives significant test
a .

• it p value ≥ ✗ -

I then to is not rejected

) ] equivalent p value # pltbistrue)
[
{Tas )
ftp.IY
are
and < × -

The statements > ka

Power d- atest la " d- test son to the ' et ' d- the s' aphiwhee points stat
]
-

[
: .

the right ,
left ) Small grows the slowest to
big giowsthefaskst (to the
n=Eû÷:)
n' : :

=p,→(rejette )
-

: ,
n
-

M' Ho
TypeIegor
)
_

Ne fix ✗ Ï
: -

1-01-21
: .

Z
-2ps
- _

, a

:p Ptaldoritejecf +62f
- -
-
x
Mrn
#
Type error =

Ka)
wedefinethepowoof.ae/-estas:power=1-p--1-PHn-ldoritejectHo)-- Ptnlreject Ho) =P HAUT >

level
For
prefer test with
large power
✗
a
given ,
we statistics
Tests
• Test
of location proportion Special case
:

☒ →
1)
°

✗
i~B.in/1.p),p--Iusedtoestinatep.Ho:p--po,TeststatisticT= po -

( 1- po) Ff0 ON ""

'
sonne
tej reg
. here

y
. .

'
B T test
-
: T known n

{
E- µ
2-
qq.Z~NIO.tl rejectionHoregion µ : Z > 2- 1.x ; QCZ , a) 1- x
.
>
=
, µ . -
=

( of )
tecteo : 2- C -
Z, -
✗ or 2- < 2- ✗ ; (Zx ) - X

µ =/ Moi 12-1>12-1 -
€1 ; QLZ , a) _
= 1- ✗ 12

↳ studentt test
for location ( T' Unknown)
-

K, . . .
,
✗ n Fa N ( µ , )
☒ -

Ho

{
1- T~tn.se T > Truc x )
if
:
Ho true µ> peu
=

s / rn ;
5- ¥1 .Ë( ✗i -
x-P Ka quantile of tn distrib , .
rejection region tuto :
T < -
Tn ka) -

str can also be replaced by Fa if Voris gives ( or standard deviation ) ttttlo :/Tl >
Tn, (E)
À Wilcoxon rank test for location Xp Anita . . -
F
,
Funknown , symmetiicaround µ=E( Xi)
Ë Ni

{
1-
Âw×N(◦ 1) itejectionHoregion 2- 2- 1.x ; (2- a)
=
> 1- ✗
' µ
>
Ho : =

nrlnrtykn.tl) ( of )
n
tecteo : 2- C -
Z, -
✗ oe 2- < 2- ✗ ; (Zx ) - X

✗i 127 129 131 132 135 138 µ =/ plo : 12-1>12-1 ¥ / -

; Q (Z , _
a) = 1- ✗ 12
✗i -

pho -8 -
G -
U -
3 0 3
Hi -

Mo / 8 6 4 3 3
tank / Xi Hot 4 1.5 1.5 meanof 1 & 2
5 3 ←
-

signcxi pro ) -
- - -
-

signed tankini -5 -

y -3 -1.5 l -5
is 5 be one is =
H
n,

fric)
{
'
02--002 %
"

Testing >
t
>
'

☐ µ unknown °
, ,
'
'
Ien 1) s
02<0! ;
# X < (1- ×)
-

✗ %
~
' rejection region
040!
; fn.im#cy'n-il1- E)
• 2 sample t test -
Xy . - -
✗ nx
,
%, . . .

/ Yny Eki)=µ× El Yi) ,

-

µ , , same var o? unknown Ho :(1×414 , HA :µx > µ>

~N ~N

Î 0>0
☒ or µ µ, -0
-
-

tn ,

{
1- x)
T
tnxtny
=
-2
tu µ -1
:
> (
rejection region
" no >
,

qing + n1 / , ,

µ
<
psy : TC tnxtng-r.CN
=/ nx 1) SÎ +1ns 1) si
✗

Sf
-
-

nxtny
-
2 µ✗ =/ ftp://T >
tnxmj-zlxc-bc.ua:{ + = ✗

}
SÎ=n± ÎÉCXI l'

dfwekh-salterwhailecorr.sy-nf-IE.LY
,

Êt (Ï

È
À } ;
-

Il
"

now the same ,

but -1-0
,
unknown Ho :µx=µy ap .
+

o : o :

{
I - I µ×> µ >
: T >
tmltl ni ( nx -

1)
+

njcny -

n¥yapptmincnxm.IN
' pej :

µ ✗
<
µ
: 1- <
tms)
+

µ, :/ Tl tmt )
=/ >
µ✗
continued :L
sample t test -

K ,
. . .
,Xnx~N(µ×i # j' ai YnyNN(plus ;h ) . . _ variances are known
I

{
I Z -21 Attia) ✗
N°11)
> 1-
t'× > µ]
- -
:
✗
z
-
-

= ~

rejection region
0¥ ÇI µx< fly Zx qlza) n x
2- : < - -

µ×≠µy : 12-1>12-1 Et Qltn-42fr-42 -

Xn ,
. . .
,Xn×~N(µ×i # j' y . . _

YnyNN(plus ;h ) variances are Unknown Ho :µ×=µz

I I
NN/ 0,1)
-

2- =

¥ Ésaü -
.
.

sÎ=n¥Ê:( ×; -
x-p
same
tej les -
as above

2
m "
Kaboul : mears are known
Sy ¥ ¥ ( y J)
- '
- -

, ,
,

test :
0×2=05 ,µ×,µy unknown K ,
. .

.im/~N(ttxi0x)jYy...YnyNNlpts;On )
Fisher
5×2 ←

{
À > Ty
'
F =

gyz
~
Fm , -1
; ng
-

,
whol SÎ >
si F >
T-nx-yny.IN
Sissy rejection region F4
T-ny-unx.tk )
Nicoya + =

§;
F > Fnmailinmin (E) -
'
Emoi
0×2 # OÙ f- =
F < F 4- E
5min
. "

• 2
comparing groups
:
testot proportions PEPY in and My we
big tensihundreds
,
..
.

{
-
I "" " > :
Z > 2- ta (tra) -
t -
✗
px py
ÊÉ
-2 = "

# rejection region alex)

"
2- < Zx ✗
pu p )
:
pxcpy
=
_
+

Px =/ Py : 12-1>12-1 %) QLZI.sn/z)-- 1- %
Îtngy
-

Â
m'
=

nxtny
← Mann Whitney Wilcoxon Xn , about of the data
rank test Xnxifdt-x.Yy.in> Faty Strong assumption the distribution
'

no
-
- -
- i.

Ho 0=0 Ha :D > 0 CDF :

Fx CDF Fyly)=F×(ytL)
↳ not necessary normal
=µ× µ y
: : -

Étape N ( MEI
nxnylnxtny +1)
)
nx

N' [(Ri -
-
i ) -
_
ËRI -

2
,
12
i -1- l' =\

Riront of Xi . in combine d
sample k×=n¥ˢ+ -211 -
x ) F -

☒ Composing ANOVA ] values in group ( columns)

groups ( lines
than 2. I ) test :
Quality of of
group variances
: each
→ we
more
groups ,

normally the errors

testing
:
Ho :

µ , -_µz= . .
.
-
-

µ,
=

µ Yij =
µ + ai +
Eij i=1 , . . -
I , j' 1 ,
. . .

,
]
meam group random
or ✗ n' ✗ 2ᵉ -
. . _
=✗I= 0 effect effect error Eij~NLO.tt ) ,
Edi =D ( all di -
-
O )
]

Ê
I

( /
numeiato' ( top )
SSB ✗
Yij € § Iii
I -1 %# = '
; ✗** = ,

È FI
,

F =
-

1. Il ] y
-

y 1. y

( 1))
Ssw

Il ]
group meam
general meam

denominator (left )
-

ÉE Y# * [ JËIY } ÈÊ( y
ÎËJÊYÏÎLÎËË IÀ
i ï»"*
i
( Yij -
=
:*
- "
** + ;
sspi →
g.
-

ç
= ,
:p -1
*
qu
.

J I -
J
total variation between inside each
group
IËLÈYÏÎ groups
vor var

Igg yig?
. .

ECJ 1) si
adf.tt] l)
ggw =
trick : ssw =
- -
.
-

÷, , i :|

{
]
FI (N
F >
rejet Ho
SST=Êz yij ËËÊYÏ) df :( 1) Il] rejection region ; :( ] 1)
-
I -

I. + -1)
1) (X) do
÷,
F < FEI ; Il] not
eject to
-

,
I J
-

rejection region
interval
:
if -0.05
d-

[FI-i.us v10.95 )
,

-
ksi-FI-i.IU-yk.SI )
; ta
] therefore wea -0.05
- ! !! ËÎ
F K

Goodness of fit - -

À test K categoriaalvars.lk ,
with
apiobability Pj associated to each Cj / Épi =L , sample size n

Ho observe 0s , 0k items G, La respective/y

categories
:
pi-pi pk we in
-
-
. . . .

/
. .
. . .

Éloi Et
?

☒ Éloi
z
eil ifth.is tue ✗ tends totale high positive values
×?
-

appuyez
~
-

= > ,
Ho
in ei i"
Ê ande ≥ g-
,

rejection region :[ Kaito )

ei-expeckdno.fi/emsincatef01gCiundvHo
,

( otherwisemerge categories
(i 1. K) ↳
XE.ilo.SN
-

"

pi ¥ e.
# with ntotalnoofikns
-
_
=

DE ✗ test of independence 2uategoricalvasAtB-Tcategou.es/orA ] categories for B Cij possible combinations

withpijpwba.Zpii-1Hoipij-pi.pj.tt/;j
, .

sample ) in size n
"
, we observe 011,0m . . .
D= ] items incategoriescn.cn C ,
. .
. ,,

Qi)
? £2 ( Q; ° '

Dii Ei;)
ÉÉ
' -
,

✗
-

jappwtf-T-D.CI
= ~
Oit -1)
i=1j= , Eij i- l
j =/
-

Aj
154.95=2826 ;]
"
E;
;
- n

sumol-allcells.in table
 Econometrics relevant notes
More random Stoff
( ou ( X 4) ,
= E -

END /Y EH . -

]
↳ car ( × , 4) =
E- (XY) -

Ecx) Et )

if X,Y are
independent : cou / ×, 4) =D
Vork + 7) =
ta ✗ + Va y +2calXY )

Distribution density
FI (
→¥ËË ,

f- ✗ ( x) =
x )

Properties
•
f- (x) ≥ 0
OÇ -1-1×1<1 f- × =
P(✗ < x )
• Place ✗ ≤
✗
6) =
) fcxldx
a

•
Flx ) =

-
f fltldt
o

he Method d- Moments Coul Xu) El Xu) =D

-

Elu × } Elu] =D impies

-
-

FÉE À Â FÊTE
{ Efxu
Elu] =
EIY Po A. ×] =D -
-

→ -
-
=
0 a)
] =
EIXIY po FX)] =D -

n'
§
" " ? 0 (2)
pion ? yi p
-
→
xiyi n × =
- -

from (1) À =
g- _

PÎI (3) •
Il ✗ and Y aie positive'y Willard Slope will 6e positive
← sample
:
"

§ Ë
"

Replace Blink) :
↳✓ ( × ,
y, •
If X and Y negativdywielated Slope Will
are : be
négative
fris sample
=
=
cou
ÇÎ Weneed ✗
touareg au" the sample
" "
2
par / × ,
.

between ✗ &Y E) ( ✗i -
← sample •

Ordinary Least Squares ! ( yi f- ÊÛÎ Fn !

2
min
1N E)
'
( Ûi
>
pi pixi )- = = -
-

Po ft i =\ ;
, ;

{§ ÊÊÛÎ =
ÇÎ ( yi -

pi pix;)- =D ÉÉTÈS minimise :

find l
"
ado condition :
dérivative :O

i "
solve : same esti matois as MM
→ .

S
Û ÊÛÎ =
% ÇÎXI / yi § PÎX;) =D - -
Assomptions : o ' const , µ ~ Nco , ☐ 2)
SPÎ i = ,
MLE OLS = =
MM

Maximum Livelihood Approach Must seek the value of the parameter function fly / )
rector
p that maximises the livelihood ×

04K ) )
'

If we assume that Y ~ N ( ✗
p , ,
we maximise

Max : ln f- 191×1 =

€ 1- ln (21704×1)
,
-

204k)
^
.
( Yi -

Po p x ;)
-

,
?
Note ifhomoskedasticu &
:
normalityof
MLE =
as = MM

{§ / f- (YIN) =D dim ( )

il-u~NHHY-MLE-l-OLS-MMPe.ir
ln of
no equations =

p
farming a test
In the common case where Ho :
pi
=
O
( Ha usually =/ 0 : twosided ,
unless spécifie
Two -
sided test C :
tn ki ✗%

Ê
•
level
-

rejet to at the a
Rejet toit t stat ≤ -
-
C or t -
stat ≥ c • If we ,
we
Sag is Statistical
y significant .

One sided test level

§; insignifiant at the ✗
•
It fait statistical
-

* we toi eject Ho is
rejet toit wesaj
.

Upper tail p; > 0 : t stat ≥ c

-

,
g
Lower tail p;
< 0 :
rejet Ho if t stat ≤
-
- C
P -
value :
if p
-
value < level of test eject

Confidence interval Norma / E :

µ Voir :O
"

selfs;) ; Ê selfs;)]
"

A- d) % CI :
[ §; -

c- + C- Uniforme :{tatb ) var :

¥16 -

a)
where tn
✗[
✗ 12
C
percentile of distrib
:
1- k 2K
E :k
i
var
-

:
-

If Ho :p , O and O is not in CI rejet

-

Ho stat
significant
-

Poisson E :b Var d
Bernoulli E :P Var :

pls pt
-
-

pq
"
Generic faim forte moment et
Ufa b)
IË ai 6k
- i ,

mk

m ,
=
{ (atb )
Mz
=

§ (à tab +62 )
tab 63 )
'

Ê(
> tab +
m> =
a

)
?
(oîtâbtàb tabs
"

ma + 6
 Power of a test

Example 's

5=22 n=8

0=10
✗= 0.05

Ho :
µ -30
-

HA µ 30 Ha µ -25
: < : -

if Ha :
µ
> 30
1) Hoist rue critica value kx : :
=

flot ZM En =
30-1.645 =
24.20
≤ -0.22 )
-

1- plz
2h20 25 and use -21
/ -0.2261=0.409
✗
-
-

2) under Ha :

10158
=
-0.226 P 2- ≤

Example Effect on Heatment : Ai lpie) Bi (post ) Di Bi

-
-
-

Ai niid N( µ ,
02 )
onesided t test Ho :µ☐=0 (no effect ) H tes > 0 (
positive effet )
:

2=0.05
-
:
,

In O
NN / ) In f- ÊDI Â / Tn
st.ovr.jnlarge-sweappwx.bz
normal
-

µ >, À
:

Tn where =
=

qqn i. |

Tn 1.64 -20.95
Rejection region : > =

Now : suppose Hais true ,

and µ >, =D .
Then ,
the
power
is

Blot Pua ( Tn. > 1.64)

=P Hal > 1.64

)
=

Pna ( In -0+0>1.64)
ÔD / Tn

Pmf 'ËË > 1.64 -

Ë) horloge n' -

Ê• NNIO 1) ,

BIO) ≈ 1- ∅ / 1.64 q%n )-