NORMAL DISTRIBUTION AND DISTRIBUTIONS RELATED TO THE NORMAL DISTRIBUTION

Distribution E[X] V ar(X)

1 x−µ 2
NORMAL f (x) = √1 e− 2 ( σ ) ,∞<x<∞ µ σ2
σ 2π
CHI-SQUARE
Pn
with n degrees of freedom X= i=1 Xi2 where Xi ∈ N (0, 1) i = 1, 2, . . . , n independents n 2n
χ2n
t-STUDENT
with n degrees of freedom T = √UV where U ∈ N (0, 1), V ∈ χ2n 0 n
n−2
, n>2
n
tn
F-SNEDECOR U
n2 2n22 (n1 +n2 −2)
with n1 and n2 degrees of freedom X= n1
V where U ∈ χ2n1 , V ∈ χ2n2 independents n2 −2
, n2 > 2 n1 (n2 −4)(n2 −2)2
, n2 > 4
n2
Fn1 ,n2
DISTRIBUTIONS OF SAMPLING STATISTICS FROM NORMAL POPULATIONS

Sampling distribution of the sample mean

Population variance Population variance unknown
known Large sample size, n > 30 Small sample size, n ≤ 30
X̄−µ X̄−µ (n−1)S 2 X̄−µ
Z= √σ
∈ N (0, 1) S
√
→ N (0, 1) Fisher: σ2
∈ χ2n−1 ⇒ S
√
∈ tn−1
n n n

Sampling distribution of the difference between two sample means

Population variances Population variances unknown
known σX = σY = σ σX 6= σY
nX > 30, nY > 30 nX ≤ 30 or nY ≤ 30
√
!
√
r
2 2
σX σY (X̄−Ȳ )−(µX −µY ) nX +nY −2 nX nY Ȳ )−(µX −µY )
(X̄−r Ȳ )−(µX −µY )
(X̄−r
X̄ − Ȳ ∈ N µX − µY , nX
+ nY
T =√ 2 +(n −1)S 2
√
nX +nY
∈ tnX +nY −2 Z = → N (0, 1) ∈ tν
(nX −1)SX Y Y S2 S2 S2 S2
X + nY X + nY
nX Y nX Y
 2
S2 S2
X + nY
nX Y
Ȳ −(µX −µY )
X̄−r
Z= ν∼  2  2
σ2 σ2 S2 S2
X X Y
nX
+ nY nX nY
Y
nX −1
+ nY −1

Sampling distribution of the ratio of two sample variances

Population means known Population means unknown
nX S ∗ 2 (nX −1)S 2
X X
σ2 σ2
X ∗ 2 σ2 X 2 σ2
nX SX nX −1 SX
nY S ∗ 2
= Y
SY∗ 2 σX
2 ∈ FnX ,nY (nY −1)S 2
= Y
SY2 σX
2 ∈ FnX −1,nY −1
Y Y
σ2 σ2
Y Y
nY nY −1
∗2 1 PnX
where SX = nX i=1 (Xi − µX )2

Sampling distribution of the sample proportion and of the difference between two sample proportions
Distribution of the sample proportion Distribution of the difference between two sample proportions
 q   q 
pq pX qX pY qY
p̂ = X̄ → N p, n
p̂X − p̂Y = X̄ − Ȳ → N pX − pY , nX
+ nY
 CONFIDENCE INTERVALS IN NORMAL POPULATIONS

Confidence interval for the population mean

Population variance known Population variance unknown: large sample size, n > 30
Z = X̄−µ
√σ
∈ N (0, 1) X̄−µ
√S → N (0, 1)
n n
h i h i
Iµ = x̄ − z1−α/2 √σn ; x̄ + z1−α/2 √σn Iµ = x̄ − z1−α/2 √Sn ; x̄ + z1−α/2 √Sn
Population variance unknown: small sample size, n ≤ 30
t = X̄−µ
√S ∈ tn−1
n
h i
Iµ = x̄ − tn−1,1−α/2 √sn ; x̄ + tn−1,1−α/2 √sn
Confidence interval for the difference between two population means
Population variances known Population variance unknown √
but equal: σX = σY = σ
√
Ȳ )−(µx −µy ) (X̄−Ȳ )−(µX −µY ) nX +nY −2 nX nY
Z = (X̄−r −→ N (0, 1) T =√ 2 2
√
nX +nY
∈ tnX +nY −2
2 2
σx σ (nX −1)SX +(nY −1)SY
nx
+ ny
y
 r r   r r 
σx2 σy2 σx2 σy2 (nx −1)Sx2 +(ny −1)Sy2 (nx −1)Sx2 +(ny −1)Sy2
q q
nx +ny nx +ny
Iµx −µy = (x̄ − ȳ) − z 1− α
2
· nx
+ ny
; (x̄ − ȳ) + z1− α
2
· nx
+ ny
x̄ − ȳ − t 1− α
2
· nx +ny −2
· nx ny
; x̄ − ȳ + t 1− α
2
· nx +ny −2
· nx ny

Population variance unknown and different: σX 6= σY

Large sample sizes: nX > 30, nY > 30 Small sample sizes: nX ≤ 30 or nY ≤ 30
Ȳ )−(µX −µY )
(X̄−r Ȳ )−(µX −µY )
(X̄−r
Z= 2 2
→ N (0, 1) 2 2
∈ tν
S S S S
X + nY X + nY
nX Y nX Y
 r r   r r 
Sx2 Sy2 Sx2 Sy2 Sx2 Sy2 Sx2 Sy2
Iµx −µy = (x̄ − ȳ) − z 1− α
2
· nx
+ ny
; (x̄ − ȳ) + z1− α
2
· nx
+ ny
Iµx −µy = (x̄ − ȳ) − t ν,1− α
2
· nx
+ ny
; (x̄ − ȳ) + t ν,1− α
2
· nx
+ ny
 CONFIDENCE INTERVALS IN NORMAL POPULATIONS
Confidence interval for the population variance
Population mean known Population mean unknown
n
(Xi − µ)2
X

i=1 (n−1)S 2
σ2
−→ χ2n σ2
∈ χ2n−1
 n n 
2 2
X X
 (xi − µ) (xi − µ)  " #
 i=1 i=1
 (n−1)s2 (n−1)s2
Iσ2 = 
 χ2n,1− α
; χ2n, α

 Iσ2 = χ2n−1;1− α
; χ2n−1; α
 2 2  2 2

Confidence interval for the ratio of two population variances

Population means known Population means unknown
∗ 2 σ2
SX 2 σ2
SX
Y
SY∗ 2 σX
2 ∈ FnX ,nY Y
SY2 σX
2 ∈ FnX −1,nY −1
   
s∗2 1 s∗2 1 s2x 1 s2x 1
I σx2 = x
s∗2
· Fnx ,ny ;1− α
; x
s∗2
· Fnx ,ny ; α
I σx2 = s2y
· Fnx −1,ny −1;1− α
; s2y
· Fnx −1,ny −1; α
2 y y 2
σy 2 2 σy 2 2

Confidence interval for a population proportion and the difference between two population proportions
Confidence interval for the proportion Confidence interval for the difference in proportions
 q   q 
pq pX qX pY qY
p̂ = X̄ → N p, n
p̂X − p̂Y = X̄ − Ȳ → N pX − pY , nX
+ nY
 q q  h q q i
p̂q̂ p̂q̂ nX +nY nX +nY
Ip = p̂ − z1−α/2 n
; p̂ + z1−α/2 n
IpX −pY = (p̂1 − p̂2 ) − z1−α/2 nX nY
p̂(1 − p̂); (p̂1 − p̂2 ) + z1−α/2 nX nY
p̂(1 − p̂)
PnX Pn Y y
xi + i=1 i
where p̂ = i=1
nX +nY

Confidence intervals in other non-normally distributed populations

Chebyschev’s inequality Large samples. ML estimator Large samples. Central limit theorem
h i  q q  h i
Iµ = X − √σ ; X + √σ Iθ = θ̂ − z1−α/2 V ar(θ̂); θ̂ + z1−α/2 V ar(θ̂) Iµ = X − z1−α/2 √σn ; X + z1−α/2 √σn
nα nα
 PARAMETRIC HYPOTHESIS TESTING

HYPOTHESIS TESTING: TEST FOR THE VARIANCE

Test for the variance (Population mean known)

Test for the variance (Population mean unknown) n
(Xi − µ)2
X
2
Test Statistic: χ2exp= (n−1)S
σ02 ∈ χ2n−1
i=1
Test Statistic: χ2exp = σ02
H0 : σ 2 = σ02 H0 : σ 2 ≤ σ02 H0 : σ 2 ≥ σ02
2
H1 : σ 2 6= σ02 H1 : σ 2 > σ02 H1 : σ 2 < σ02 H0 : σ = σ02 H0 : σ ≤2
σ02 H0 : σ 2 ≥ σ02
H1 : σ 2 6= σ02 H1 : σ 2 > σ02 H1 : σ 2 < σ02
χ2exp < χ2n−1; α
2
χ2exp < χ2n; α
or χ2exp > χ2n−1;1−α χ2exp < χ2n−1;α 2

χ2exp > χ2n−1;1− α or χ2exp > χ2n;1−α χ2exp < χ2n;α

2
χ2exp > χ2n;1− α
2

HYPOTHESIS TESTING: TEST FOR THE EQUALITY OF VARIANCES

Test for the equality of variances (Population means known)

nX
Test for the equality of variances (Population means unknown) 1
X
(Xi − µX )2
nX
2
SX i=1
Test Statistic: Fexp = SY2
Test Statistic: Fexp = nY
(Yi − µY )2
X
2 1
H0 : σX = σY2 2
H0 : σX ≤ σY2 2
H0 : σX ≥ σY2 nY
i=1
2
H1 : σX 6= σY2 2
H1 : σX > σY2 2
H1 : σX < σY2 H0 : 2
σX
= σY2 H0 : 2
σX ≤ σY2 2
H0 : σX ≥ σY2
Fexp < FnX −1,nY −1; α2 H1 : 6=2
σX σY2 H1 : 2
σX > σY2 2
H1 : σX < σY2
or Fexp > FnX −1,nY −1;1−α Fexp < FnX −1,nY −1;α Fexp < FnX ,nY ; α2
Fexp > FnX −1,nY −1;1− α2 or Fexp > FnX ,nY ;1−α Fexp < FnX ,nY ;α
Fexp > FnX ,nY ;1− α2
 HYPOTHESIS TESTING: TEST FOR THE MEAN

Test for the population mean (Population variance known) Tests for the population mean (Population variance unknown)
X̄−µ0 X̄−µ0
Test Statistic: Zexp = √σ
∈ N (0, 1) Test Statistic: Texp = S
√
∈ tn−1
n n

H0 : µ = µ0 H0 : µ ≤ µ0 H0 : µ ≥ µ0 H0 : µ = µ0 H0 : µ ≤ µ0 H0 : µ ≥ µ0
H1 : µ 6= µ0 H1 : µ > µ0 H1 : µ < µ0 H1 : µ 6= µ0 H1 : µ > µ0 H1 : µ < µ0
Zexp > z1− α2 Texp > tn−1;1− α2
or Zexp > z1−α Zexp < zα = −z1−α or Texp > tn−1;1−α Texp < tn−1;α = −tn−1;1−α
Zexp < z α2 = −z1− α2 Texp < tn−1; α2 = −tn−1;1− α2

Test for the difference between two populations means (Population variances known)
X̄−Ȳ −d0
Test Statistic: Zexp = r 2 2
∈ N (0, 1)
σ σ
X + nY
nX Y

H0 : µX − µY = d0 H0 : µX − µY ≤ d0 H0 : µX − µY ≥ d0
H1 : µX − µY 6= d0 H1 : µX − µY > d0 H1 : µX − µY < d0
Zexp > z1− α2
or Zexp > z1−α Zexp < zα = −z1−α
Zexp < z α2 = −z1− α2
Test for the difference between two populations means (Population variances unknown but equal)
2 +(n −1)S 2
(nX −1)SX
Test Statistic: Texp = q Ȳ −d0
X̄−
∈ tnX +nY −2 , S 02 = Y Y
1 nX +nY −2
S0 nX
+ n1
Y

H0 : µX − µY = d0 H0 : µX − µY ≤ d0 H0 : µX − µY ≥ d0
H1 : µX − µY 6= d0 H1 : µX − µY > d0 H1 : µX − µY < d0
Texp > tnX +nY −2;1− α2
or Texp < tnX +nY −2;α
Texp > tnX +nY −2;1−α
Texp < tnX +nY −2; α2 = −tnX +nY −2;1−α
= −tnX +nY −2;1− α2
 7

Test for the difference between two populations means (Population variances unknown and different)
 2
s2 s2
X + nY
nX Y
X̄−Ȳ −d0
Test Statistic: Texp = r ∈ tν where ν '  2  2
2S 2 S s2 s2
X X Y
nX
+ nY nX nY
Y
nX −1
+ nY −1

H0 : µX − µY = d0 H0 : µX − µY ≤ d0 H0 : µX − µY ≥ d0
H1 : µX − µY 6= d0 H1 : µX − µY > d0 H1 : µX − µY < d0
Texp > tν;1− α2
or Texp > tν;1−α Texp < tν;α = −tν;1−α
Texp < tν; α2 = −tν;1− α2
 HYPOTHESIS TESTING: TEST FOR PROPORTIONS

Test for a population proportion

Pn
p̂−p0 xi
Test Statistic: Zexp = s ∈ N (0, 1) where p̂ = i=1
n
.
p0 (1 − p0 )
n
H0 : p = p0 H0 : p ≤ p0 H0 : p ≥ p0
H1 : p 6= p0 H1 : p > p0 H1 : p < p0
Zexp > z1− α2
or Zexp > z1−α Zexp < zα = −z1−α
Zexp < z α2 = −z1− α2

HYPOTHESIS TESTING: TEST FOR THE DIFFERENCE BETWEEN TWO POPULATION PROPORTIONS

Test for difference between two population proportions

PnX Y y
Pn
xi + i=1
Test Statistic: Zexp = q p̂X −p̂Y ∈ N (0, 1) where p̂ = i=1 i
nX +nY nX +nY
nX nY
p̂(1−p̂)

H0 : pX = pY H0 : pX ≤ pY H0 : pX ≥ pY
H1 : pX 6= pY H1 : pX > pY H1 : pX < pY
Zexp > z1− α2
or Zexp > z1−α Zexp < zα = −z1−α
Zexp < z α2 = −z1− α2
 NON-PARAMETRIC HYPOTHESIS TESTING

Test Statistic Rejection Region

Rexp > R1− α2
H0 : Sample is random
RUNS Rexp ≡ total number of runs or
H1 : Sample is not random
Rexp < R α2
Zexp > z1− α2
RUNS H0 : Sample is random 2n n
Rexp −( 1 2 +1)
Zexp = q 2n n (2nn n −n) ∈ N (0, 1) or
(n1 > 10 and n2 > 10) H1 : Sample is not random 1 2 1 2
n2 (n+1)
Zexp < z α2 = −z1− α2

+
 Texp > K1− 2
 
 α
 H0 : Me = m or
H1 : Me 6= m

 +


 Texp < K α2
+
WILCOXON H0 : Me ≤ m  Texp ≡ Sum of the ranks of the absolute values
+
SIGNED-RANK H1 : Me > m  with positive differences (Di > 0) Texp > K1−α

 H0 : Me ≥ m
+
 H1 : Me < m Texp < Kα

 

 Zexp > z1− α2
 H0 : Me = m or
 H1 : Me 6= m

Zexp < z α2 = −z1− α2



WILCOXON H0 : Me ≤ m  T +
−
n(n+1)
Zexp = qexp 4
∈ N (0, 1)
SIGNED-RANK (n > 15) H1 : Me > m  n(n+1)(2n+1)
24
Zexp > z1−α

 H0 : Me ≥ m
 H1 : Me < m Zexp < zα = −z1−α
10

Test Statistic Rejection Region


 

 Uexp > Ux,1− α2
 H0 : µX = µY or
 H1 : µX 6= µY
UX ≡ difference between maximum 

 possible rank sum for sample X

Uexp < Ux, α2
WILCOXON H0 : µX ≤ µY  and the observed rank sum for sample X
MANN-WITNEY H1 : µX > µY  n1 (n1 +1) Uexp < Ux,α
 UX = n1 n2 + 2
− WX
 H0 : µX ≥ µY
(WX ≡ sum of the observed ranks in sample X) Uexp > Ux,1−α
 H1 : µX < µY

 

 Zexp > z1− α2
 H0 : µX = µY or
 H1 : µX 6= µY

Zexp < z α2 = −z1− α2


WILCOXON 
H0 : µX ≤ µY  U −
n1 n2
MANN-WITNEY Zexp = q n nX (n 2
∈ N (0, 1)
H1 : µX > µY  1 2 1 +n2 +1) Zexp < zα = −z1−α
(n1 > 10 and n2 > 10)  12
 H0 : µX ≥ µY
 H1 : µX < µY Zexp > z1−α

k
H0 : µ1 = µ2 = . . . = µk 12
X Ri2
KRUSKAL-WALLIS Hexp = n(n+1)
− 3(n + 1) Hexp > hα
H1 : ∃ i 6= j, µi = µj i=1 ni
|R̄i − r
R̄j | ≥ cij where
 
n(n+1 1 1
H0 : µi = µj cij = zp 12 ni
+ nj
,
DUNN |R̄i − R̄j |
H1 : µi 6= µj P [Z ≥ zp ] = p,
α
p = k(k−1)
2
χ2exp = ki=1 (ni −np i)
P
GOODNESS H0 : F (x) = F0 (x) npi χ2exp ≥ χ21−α
(Pearson’s χ2 test) H1 : F (x) 6= F0 (x) ∈ χ2k−h−1
H0 : X and Y are ni. n.j 2
Pr Ps (nij − )
INDEPENDENCE independent χ2exp = i=1 j=1
n
ni. n.j
n χ2exp ≥ χ21−α
AND HOMOGENEITY H0 : All r samples ∈ χ2(r−1)(s−1)
are homogeneous