Generally Helpful Tidbits Sampling Distributions Related to the Normal

pmf: p(y) cdf: F(y) = z ≤y p(z) P[a ≤ Y ≤ b] = ba p(y) Assume Y1, · · · , Yn are iid samples from N(µ, σ 2 ). Then
Í Í

Ȳ = [(1/n) i=1 Yi ] ∼ N(µ, σ 2 /n)

Ín
pdf: f (y) = dy F(y) cdf: F(y) = z ≤y f (z)dz
d
∫
∫b
P[a ≤ Y ≤ b] = a f (y)dy = F(b) − F(a) If Z ∼ N(0,1), Z 2 ∼ χ2 (1)
Zi2 ) = ( i=1 [(Yi − µ)/σ]2 ) ∼ χ2 (n)
Ín Ín
P(Y = y|X = x) = P(Y = y ∩ X = x)/P(X = x) ( i=1
E[Y k ] = y y k p(y) or y y k f (y)dy
Í ∫ [(n − 1)S 2 /σ 2 ] ∼ χ2 (n − 1)
√ √
[(Ȳ − µ)/(S/ n)] ∼ t(n − 1) [(Ȳ − µ)/(σ/ n)] ∼ N(0,1)
V[Y ] = σ 2 = E[(Y − E[Y ])2 ] = E[Y 2 ] − (E[Y ])2
If Z ∼ N(0,1) and W ∼ χ2 (ν) are independent r.v.’s, then
Cov(X, Y ) = E[XY ] − E[X]E[Y ] ρ = Cov(X, Y )/(σX , σY )
T = (Z/ W/ν) ∼ t(ν).
p
Indicator function: I[a,b] (y) = 1 if a ≤ y ≤ b; 0 otherwise
If W1 ∼ χ2 (ν1 ) and W2 ∼ χ2 (ν2 ) are independent r.v.’s, then
Binomial – Y ∼ Bin(n, p) F = [(W1 /ν1 )/(W2 /ν2 )] ∼ F(ν1, ν2 )
p(y) = n
y
y p (1
− p)n−y , y ∈ [0, n], p ∈ [0, 1] e.g., F = (S12 /σ12 )/(S22 /σ22 ) ∼ F(n1 − 1, n2 − 1).
n

y = n!/[y!(n − y)!]
Central Limit Theorem
E[Y ] = np; V[Y ] = np(1 − p); m(t) = [pet + (1 − p)]n
Let Y1, · · · , Yn be iid draws from an arbitrary distribution with known
n = 1 ⇒ Bernoulli distribution √
µ and σ 2 (< ∞). If Un = (Ȳ − µ)/(σ/ n), then
Geometric – Y ∼ Geom(p) lim P[Un ≤ u] = Φ(u) ∀ u ,
n→∞
p(y) = (1 − p)y−1 p, y ∈ [1, ∞), p ∈ [0, 1]
For n & 30, Ȳ ∼ N(µ, σ 2 /n) and i=1
Ín
Yi ∼ N(nµ, nσ 2 ), approx.
E[Y ] = 1/p; V[Y ] = (1 − p)/p2 ; m(t) = pet /[1 − (1 − p)et ]
Hypothesis Testing
Poisson – Y ∼ Poi(λ)
λ y −λ Small-sample difference in means (equal population variance):
y ∈ [0, ∞); E[Y ] = V[Y ] = λ; m(t) = eλ(e
t −1)
p(y) = y! e ,
Ho : µ1 − µ2 = µd and Ha : µ1 − µ2 , µd
Uniform – Y ∼ Uniform(a, b) X̄1 − X̄2 − µd
T =  ∼ tn1 +n2 −2
f (y) = (b − y ∈ [a, b]; E[Y ] = (a + b)/2;
a)−1 ,
r 
Sp n1 + n2
2 1 1
V[Y ] = (b − a)2 /12; m(t) = (ebt − e at )/[t(b − a)]
(n1 − 1)S12 + (n2 − 1)S22
Normal – Y ∼ N(µ, σ 2 ) Sp2 =
n1 + n2 − 2
y ∈ (−∞, ∞), µ ∈ R, σ ∈ R+
2 2
f (y) = √ 1 e−(y−µ) /2σ
2πσ 2
E[Y ] = µ; V[Y ] = σ 2 ; m(t) = exp(µt + t 2 σ 2 /2) Levene’s test statistic:
N(µ, σ 2 ),
then Z = (Y − µ)/σ; Z ∼ N(0,1).
Ík 2
If Y ∼ (n − k) i=1 ni ( Z̄i• − Z̄)
y−µ
W=
P[Y ≤ y] = Φ σ = Φ(z) (non-analytic function) Ík Í ni
(k − 1) i=1 ( Z̄i j − Z̄i• )2
j=1
Gamma – Y ∼ Gamma(α, β)
where, e.g., Zi j = |Yi j − Ȳi• | (mean), Zi j = |Yi j − Ỹi• | (median), or
f (y) = y α−1 e−y/β /[βα Γ(α)], y ∈ [0, ∞), α ∈ R+ , β ∈ R+ Zi j = |Yi j − Yi•0 | (10% trimmed mean).
∫∞
Γ(α) = 0 y α−1 e−y dy = (α − 1)Γ(α − 1)
Wilks’ theorem: if λ is the LRT test statistic, then for large n,
If n is a positive integer, Γ(n) = (n − 1)!
−2 log(λ) has approximately a χr−r
2 distribution.
E[Y ] = αβ; V[Y ] = αβ2 ; m(t) = (1 − βt)−α o

α = 1 ⇒ exponential distribution Simple Linear Regression

β = 2, α = ν/2, ν ∈ Z+ ⇒ chi-square distribution
Ín
( i=1 Yi xi ) − n x̄Ȳ
β̂0 = Ȳ − β̂1 x̄ and β̂1 = Ín 2
Moment Generating Functions ( i=1 xi ) − n x̄ 2
mY (t) = E[etY ] = σ2 σ 2 x̄ 2 σ2
y e p(y) = y e f (y)dy
Í ty ∫
ty
V[ β̂0 ] = + Ín 2 and V[ β̂1 ] = Ín 2
mY (t) = 1 + t µ10 + (t 2 /2!)µ20 + · · · n ( i=1 xi ) − n x̄ 2 ( i=1 xi ) − n x̄ 2
µk0 = y y k p(y) = y y k f (y)dy = (d k m/dt k )|t=0
Í ∫
Sample correlation:
E[Y ] = µ10
= µ and = V[Y ] +
E[Y 2 ] (E[Y ])2 = µ20 Ín r
i=1 (Xi − X̄)(Yi − Ȳ ) SXY SX X
r = Ín = √ = β̂
E[e ] = y e p(y) = y ety f (y)dy
∫
tY Í ty
n 1
[( i=1 (Xi − X̄)2 )( i=1 (Yi − Ȳ )2 )]1/2
Í
SX X SYY SYY

Method of Moment Generating Functions

ANOVA
If, e.g., X = a1Y1 + a2Y2 + b, mX (t) = ebt mY1 (a1 t)mY2 (a2 t) Õ ni
k Õ k
Õ
Match mX (t) to known mgf to name the distribution PX . Total SS = (Yi j − Ȳi• )2 + ni (Ȳi• − Ȳ )2 = SSE + SST
i=1 j=1 i=1

t-distribution table: front page of the exam Standard Normal Percentage Points
χ2 -distribution table: back of formula page z0.1 = 1.28 z0.05 = 1.645 z0.025 = 1.96
z0.01 = 2.326 z0.005 = 2.576