Non-Parametric Inference
Non-Parametric Inference
Aug 25
I
Quit 2 Oct 27
Prog
.
Group :
Aug 17
Data File
Source
X1 X2 In
In from FE F
, , . . .
.
pth quantile
:
Ep
for a fixed
pe Co 1),
,
Ho :
Sp
=
Epo vs Ha :
Ep2p
p =
Xcm where p =
upeI
[np] + I
Pi =
P(Xi -
Ep) = + if Xi >
Epo
O [
Y
,, c, In my's
parameter
Y
... a re iid Ber with
P (vi =
1) = p(xi) sp)
=
1 -
F(2p)
= 1 -
d undert Ho
Ranking
statistic
wank
(Xi) = Ri
↑ (RI ,
Rz , ...
Rn) = airi
P (Ri =
j) =
in , i = 1 , 2
, ...
n
statistic Exp : Tossing a coin twice
Signed-rank
:
- =
EHH , HT ,
TH , TT3
a fairs coin
=
X : # of heads in a trial
X(w) : 0 12
x" (0 , x] =
0xc
TT OXXL/
TT HT ,
TH , 11412
,
2 ,
232
- = Sa ,
6, c] - =
[0 ,
a
, 56 .
2, 3 23
,
X (a) = 1
, X(b) = 2
, X(2) = 3
- Not RY
X, X2 , . . . Xn iid obs .
With cdf Ex
,
Pof by ,
pdf fx
X (1 = min
[X1 ,
X2 ...
Xn3
X(n) = max
&X1 , X2, , . . .
Xn3
Y = X()
P(X(x > y)
1-P(xi < y)
=
1-El-Fxi (3)3
2
= 1- (1-Fxi(y))
py(y) =
Fy(y) -
Fy(y)
fy(y) =
=y Fy(y) =
n( -
Fxi(y)) fx(l)
X(m)
Exom) = Fo (2) =
P (at least p
,
Xis <x)
=E Pexactly j , Xis
=
-
Fx(n)
! n-c-ugh-6 du
=
IFxcus(r , n -n
+
1) =
etonit
F(a b) , =
a ciuda , as a
P() =
Fx,) -
-notFl
n
Drugh
-
do
1)
-If()(b , n - n +
fx(n)(x) =
+ 4-0
F(us (1-F(a) > ex ()
-nutig ro ,
(Continuous Case)
Heuristic
argument
x) = Fx Exams
Lolth)
=
-
=im p(uXm I
t
h)(b
2
= (Fx()b-1 lim
-
F()
n-m) ath) (1 F(x +
-
+ 0 + ...
Fx(n))2
-
*
=
itnat Ex (n)) (1 -
-x() ,
nei
Date :
05-Aug-24
Joint Disth of Xcm , Xcs) , MLS
F(y =
P(XU , Xy)
=
Plat least v Xi's < x
,
at least -Xxs =
y)
= /exactly (j-i)
i xis in
Can a
(x y]
,
n -
j
(y ,
x))
(F(" (F()-FC))i-1(- F(e))"
-(n is ! may
(n y)
>,
↑ (X(m = x , Xcx = y)
=
P(Xcx =
y)
1
= () (F(e))"(1 F(y))2
-
puf ,
F(u
F-F -x) y)
n,
phy
-
X (n) XCS) ,
k)
x
-
n y
-
Fx(n)
,
+
,
#x(x ,
= P(x(X(m) k)
= n + h y <X(x = +
,
IC!I a
0-1 St I n - S
Cr-1 ! (s-m -
Th-s) :
x
[
F(y)
-
F(n +
+ k) -
F()]
s
F(y + x)]n
-
x
[1
-
(F(y)
-
2
= ,
of
Joint Idf Xcrl X(02) DIA2C crok
...
,
...
, ,
[in us win .
I
en
v2-pr-1
"
F (x1 , 2 ...
xk)
(r -crit T !
[F()] N
IF (2)-F(1)]
x(ru)
= . . .
[i-F(UxYh-O
rx-2x-1-1 +(1) +(2) -(xx)
[F(xx) 3
. . .
....
-
F(xx 1) - ,
LUnLX
- * (n < U21 . . .
Dk
in a
n
i wil
-
= ric
Ei -FC
Cont X1 ,
X2 ,
... Xn
Ind with cdf F ,
par fx
Joint of Xc , X(2) . . . Xant : (Jacobian method)
X(x = g , (X ,, X2 ...
Xn)
X(2) = 92(X1 ,
Xz , . . .
xn)
&
X(n) =
gn(X1 ,
Xz , . . .
Xn)
fun =) ,
o
- (r(1) Kens)
(15i// Exc ,M n
2(2) ...
2
.. ·
,
=
X (1) , X (2) .. -
X (n)
all cases of
inverse transformations
=
nix (i)
fX(x)
(m1
,
,
x(2)
nz
. .
...
X(n)
un) =
nix (i)
Moments of order statistics :
moment
Ist order
single :
fr =
In , Urin =
E(Xpin)
fro : n(x)dx
=
(x
b
(x
n -
-
=
Cr . n
F(x)0 (1 -
F()) f()dx
Impin)
/(11fdn)- (14) from (h) du
=
>
Cusing Jensen's
inequality)
=
Coin
(iv) f(u) do
Date :
06-Aug-25 If
ELIX1] exists then ECX) finite .
= (win) E
-
Lumin =
=
E(Xn)
= E(X) = n
paralel converse :
(k)
Urin =
(u" foin(x) do
-x
X- ((0 , 1)
f(x)
= 2
/
MER
,
F(x) =
E +
y +an - x
,
uER
n)zd
-
Ju (Ettant) (E-t
+
+an
(in
urin =
un,n is finite v =
2(1)n -
1 n > 2
,
(k)
Mr ; n
p = k + 1(1)n -
k
Va (Xoin) =
Gbin = or
, nin
rPhin- (evin)2
=
(jk)th order
product moment ,
Crisive Ec"
I n - D
S - x -
Mr , sin
=
E(Xin Xsin) = (F(x-F(n)) C- (e))
+
= Un , sin-Main Usin
Disth about
symmetric
o
In ↓ b
F( u)
-
= 1 -
F(x)
X1 ,
X2 ,
...
Xn And CdfF , sym
- about
0 .
Yi ,i 1() n
= -
Xi =
y X
Fy(m)
=
P(y[x) =
p( -
X = u)
= P(Xx u) -
= 1 -
F(
-
u)
# =
Fx(m)
Xx . . .
Xcn >
y2 . . . .
Y(u)
* (mE Ycr) E X (n + 1)
-
-n
(X( = ( Xx)
, X(x . . .
x(n)) -
X(n , . . . -
=> Mr ; n
=
-Un-n + in
(k)
Mrin = (1)"Un-r + in
(X(0) x(s)) ,
*
( -
X(n -
s+ 1) ,
-
X(n - n +
1)
Mois ; n
= C-1't'en-s + 1
,
n-p + 1 in
U
(j k) ,
=
(-15th in ,
n -n +1 : n
(p ,
S ; n)
-
u = F(m) ,
x = F (u)
= x(u)
I
urin =
Coin
(u(u) un-(1-u)h-du
O
For (Co 1) ,
,
u =
E +tan-y
+
[ (u 1)]
-
tan + -
= x(u)
= ) - Got u = x(u)
eoin =
u =
F(x)
Crossing
u F(y)
ncas
=
S -n -
1n -
S
wo-
urisin =
your (e -
u)( -a)
dud u
X1 , Xz ... Xn V =
F(X)
,
v. = F(X 1) .
. .
.,
Un = F (Xn)
<Vans
conv youp-ra
Up : n
FtUin] =
=
(0
ruyp - du =
Tp = (p -
1) !
1 e-crypt an
=
#
Ine) &, N(0 1)
,
Eg(T) g(u))
-
- (0 (g'(u)(2)
,
Date : 25
09-Aug
.
X1 ,
X2 , . . .
In In 0(0 , 1)
O ou
,
u = F(x)
-
a = n(u) =
F (k) = U
(k)
!ukum-1p-4) d
-
·
upin =
Join k= 1, Unin =
pro
In =
1-pos
up , sin =
= , sin =
1
moments of For cont Disth
Approximating any
a s .
#
. .
X X2 ,
... Xn nd from F
nd
F(x1) F(x2) .. ·
F(yn) v(0 , 1)
u = F(x)
X(x) (X(2)(
(X(n)
. . .
W(n) =
F (X (m)
F " (U(r)
X(m
g (Ucr)
=
=
-Pr)g(n)
= g(Pv) +
(Uc-Pr)g(n) +
(E)
=
-
~
[xcm] =
v[Uc](g'Ps)" =
HTz2
=
g(Pn)
=CU val = Pr
- (v = pr
=xm)Iem
-
(X(m X(s)
=
Cov ,
Asymptotic
Disth of
sample quantic
Xin) is the
6th sample quantile
if r = up - I
[nP1 + 1
#No
E(Xv n) =
g(pn) F
+
(ii)
g(U(m)
=
=
X(m) =
(-)
-
= F
/(b) Eb
-
= F =
# =
# XI , X2 ,
. . . Xn
End (u ,
02)
X -
N(u ,
Mn >
-
N(u + ,
52π)
=
N(u ,
=
Efficiency : (Mr ,2)
= 0 . 6366
Mn - N(u ,
i - (n)
=( u)
f(u) =
Efficiency : (Mn x) ,
:
= >
n
of fixed p =(0 1)
of hypothesis Ep . ,
Testing
Ho :
Ep =
Ep (fixed)
vs HA :
Ep Epo
#
Date :
11-Aug-25
'
E
Chapter -
2
Gibbons book
>
PLEXcnEd] =
Excus(6) -
Ex , n(a) pat
= (Y) (Fx(b))"(1 -
(I -
Fx(a))
# U. , ve ... Un
ind v(o ,
1)
R ((n) =
-
Vi
-
Bi (n - 1
, 2)
QR =
V(x) -
V(m) ~B = (s -
v
,
n -
(s -
m) + 1)
Ex -
↑ (X(x =
x) =
Fx(u)
-
Ex-1)
Exi
S -
n -
1
* "
F(mu)
cus =1)
>
-
( (m) ( (F(us) -
-v
-
)!
S
F(us)(2 (s)
-
(1 -
f(xm) +
: ((us))S
n S
(1-F(us))
-
(s)
-
F(un))
Curcus-
n
-
1
(1 -
f(x))
f(x) F(x))2
m
(1
-
F
= #(1) -
n ...
, 2 ...
m !
Date :
13-Aug-25
#
P
T- :
=Fun EJ
ap = xm Jas
p =
np
[np1 + 1
P[X(n) =
Ep =
Xcs)] = # (0 , S ,
n ,
p)
< S
= (picc
=
P[n =
yes -
1/Y-Bin (n ,
b) ]
np z n-o + 1
>
symmetric I
-
=
s =
,
P[n = Y = n -
/ y-Bin (n =
=) ]
-
connection
finite population
[ ↑
(Normal approximation for binomial)
·
= n =
n - 32
m +
n -n + 1 =
3/2
Tolerance
Interval/Coverage -
Given p , UECo , 17 Lu
[ (Ucn Ucss)
W
,
P[/fcudus p] , =
r is the corns
[x,m X(x)]
,
coverage - -
I
=
p[F(u) -
F(z) =
p] = w
mcs ,
P[F(X(s)) -
F(Xcn))] = r
↑ [Ucs-Ucs p] ,
= w
(Ucx)-Ucm - B = (s
-n , n -
(s -)
+
Y
P
=
timnecS ,
useb cusn-c du
Test for pandomness
Run Test : -
A >
-
n1
- n ni + n2
types objects
=
Run : 2 of - - 42
B
of #identical objects
length roun : in a run .
R =
total no Of
. run's of n = n , +he objects .
R2
11
: 11 11 11 "n2 1 1" type I .
# (R1-R2)11 => Ri = R2 ,
R2-1 ,
Reti
Date :
14-Aug-25
of
type I-A Ri
obj
u1 ,
buns
nituz = n
, R = RitR2 ,
IRI-R2lEt
↑ cells st no cells
empty =
(i (
-
↑ (H1 = N ,
Re =
b)
Cr) P(RED) =
2
+
(_ ) (i " )
(i1 ) [(1) +
(ii)]
·
(i) (ut) , v = 1 ,
2 . .
mingue the
(i)
↑(Re mc) =
=
, n = 2 ... min
Enta
(rith2)p(R =
n) = 2
() ( ,
=
-
-
(i) (i) wi = +2 - 1 , nz + 1
·
-(E) (i)
v = 2 (1) 2 min (n1 n2) ,
+ C
o if i even
C = 1 if i odd
in
Ho
arrangement
: pandom
n = 9 ,
n1 = 5, nz = 4 < = 0 .
05
,
W : obs .
REC on obs .
R&C2
v = 2 (1) 2 min (5 4)
,
+ 1
2(1) a
PH. (R = 2) +
Pho (R>a) =
Es = 0 .
024 20 .
05