Formula Sheet for MATH 236∗,2018−2019

n
P
xi Chebyshev’s Theorem:
i=1 Joint Probability Distributions: 1
Mean: x̄ = For k > 1, P (µ − kσ < X < µ + kσ) ≥ 1 −
n marginal X
pmfs: X k2
Quartiles, Median, and IQR: fX (x) = fX,Y (x, y) and fY (y) = fX,Y (x, y)
Qi = value at 0.25i(n + 1)th ordered position, i = 1, 2, 3 y x Some Discrete Probability Distributions:
marginal Zpdfs: Discrete Uniform: (X ∼ Discrete U [a, a + 1, . . . , b])
x̃ = Q2 and IQR = Q3 − Q1 Z
2
fX (x) = fX,Y (x, y) dy and fY (y) = fX,Y (x, y) dx 1 a+b 2 (b − a + 1) − 1
f (x) = , µX = , σX =
Variance and Standard Deviation: y x b−a+1 2 12
 n
P
2 conditional distributions: Bernoulli: (X ∼ B (1, p))
n n xi fX,Y (x, y) 1−x
(xi − x̄)2 x2i − i=1n fX|Y (x|y) = , provided that fY (y) > 0 f (x) = px (1 − p) , µX = p, σX 2
= p (1 − p)
P P
√ fY (y)
s2 = i=1
= i=1
and s = s2 Binomial: (X ∼ B (n, p))
n−1 n−1 fX,Y (x, y)  
fY |X (y|x) = , provided that fX (x) > 0 n x n−x 2
Covariance: fX (x) f (x) = p (1 − p) , µX = np, σX = np (1 − p)
n
P Pn x
(xi − x̄)(yi − ȳ) xi yi − nx̄ȳ
i=1 i=1
Mean of a Random Variable (R.V.): Hypergeometric:
s N −s


(X ∼ Hypgeo (N, s, n))
sxy = =
X
n−1 n−1 discrete: E (X) = µX = xf (x) ns 2 N −n s  s
f (x) = x Nn−x
, µX = , σX = n 1−
sxy x Z
∞ n
N N −1 N N
Correlation Coefficient: r =
sx sy continuous: E (X) = µX = xf (x)dx Negative Binomial:
 (X ∼ N B (r, p))
−∞ x−1 r x−r r 2 r (1 − p)
f (x) = p (1 − p) , µX = , σX =
Combination and Permutation: Means of FunctionsX of Random Variables (R.V.s): r−1 p p2
Geometric: (X ∼ Geo (p))
 
n n n! n! discrete: E (g (X)) = g (x) f (x)
Cx = = and Pxn = x−1 1 2 1−p
x (n − x)!x! (n − x)! x f (x) = p (1 − p) , µX = , σX =
p p2
XX
n (A) and E [g (X, Y )] = g (x, y) f (x, y)
Probability of an event: P (A) = x y
Poisson: (X ∼ P oisson (λt))
n (S) Z ∞ e−λt (λt)
x
2
continuous: E (g (X)) = g (x) f (x)dx f (x) = , µX = λt, σX = λt
Conditional Probability: x!
P (A ∩ B) Z ∞ Z −∞
∞
P (A|B) = , provided that P (B) > 0 and E [g (X, Y )] = g (x, y) f (x, y) dxdy
P (B) Some Continuous Probability Distributions:
−∞ −∞
The Rule of Total Probability: Continuous Uniform: (X ∼ U [a, b])
Variance and Standard Deviation ofia R.V.:  1 2
P (A) = P (A|E1 ) P (E1 ) + · · · + P (A|EK ) P (EK ) 2
h
2 b−a , a ≤ x ≤ b , µ =
a+b 2 (b − a)
variance: V ar (X) = σX = E (X − µX ) f (x) = X , σ X =
0, elsewhere 2 12
2
or alternatively V ar (X) = E X 2 − [E (X)]


2


Bayes’ Rule: q Normal: (X ∼ N µ, σ )
P (A|Ei ) P (Ei ) 2 1 (x−µ)2
P (Ei |A) = standard deviation: σX = σX f (x) = √ e− 2σ2 , − ∞ < x < ∞, µX = µ, σX 2
= σ2
P (A|E1 ) P (E1 ) + · · · + P (A|EK ) P (EK ) 2πσ
Covariance and Correlation Coefficient of R.V.s: X −µ
Note: Z = ∼ N (0, 1)
Discrete Probability Distributions: covariance: Cov (X, Y ) = σXY = E [(X − µX ) (Y − µY )] σ
pmf: f (x) = P {X = x} for all values of x or alternatively Cov (X, Y ) = E (XY ) − E (X) E (Y )
Gamma:  (X1∼ Gamma α−1 −x/β
(α, β))
X σXY β α Γ(α) x e , x≥0
cdf: F (x) = P {X ≤ x} = f (t) for − ∞ < x < ∞ correlation coefficient: Corr (X, Y ) = ρXY = f (x) = , where α > 0,
σX σY 0, elsewhere
t≤x Z ∞
Continuous Probability Distributions: Means and Variances of Linear Comb.s of R.V.s: β > 0, and Γ (α) = xα−1 e−x dx, for α > 0.
E (aX + b) = aE (X) + b 0
pdf: Pdf (f (x)) of a random variable (r.v.) is constructed µX = αβ, σX 2
= αβ 2
so that the area under its curve bounded by the x axis is E [g (X) ± h (X)] = E [g (X)] ± E [h (X)]
Exponential: (X ∼ Exp (β))
equal to 1 when computed over the range of the r.v. for E [g (X, Y ) ± h (X, Y )] = E [g (X, Y )] ± E [h (X, Y )]  1 −x/β
e , x≥0
which f (x) is defined. X⊥ ⊥ Y ⇒ E (XY ) = E (X) E (Y ) f (x) = β , where β > 0.
x 2
V ar (aX ± bY ± c) = σaX±bY 2 2 2 2 0, elsewhere
±c = a σX + b σY ± 2abσXY
Z
cdf: F (x) = P {X ≤ x} = f (t) dt for − ∞ < x < ∞ 2 µX = β, σX2
= β2
−∞
X⊥ ⊥ Y ⇒ V ar (aX ± bY ) = σaX±bY = a2 σX
2
+ b2 σY2
 n sd sd n n
• d¯ − tn−1,α/2 √ < µd < d¯ + tn−1,α/2 √
P
Xi n
X X
P
(Xi −X̄)2 n n SST = (yi − ȳ)2 = yi2 − nȳ 2
i=1
Important Statistics: X̄ = and S 2 = i=1
n−1 i=1 i=1

n (n−1)s2 (n−1)s2 n n n
• χ2n−1,α/2
< σ2 < χ2n−1,1−α/2
X̄ − µ X̄1 − X̄2 − (µ1 − µ2 )
X X X
(yˆi − ȳ) = b21
2
(xi − x̄)2 = b21 x2i − nx̄2


Z= Z= SSR =
√σ
r   
σ12 σ22 s21 1 σ12 s21 i=1 i=1 i=1
n
n1 + n2
• s22 fα/2 (v1 ,v2 )
< σ22
< f
s22 α/2
(v2 , v1 ) n
X n
X n
X
(n−1)S 2 SSE = (yi − ŷi )2 = (yi − (b0 + b1 xi ))2 = e2i
Z= q p̂−p , where p̂ = X
χ2 =
p(1−p) n σ2 i=1 i=1 i=1
n
Prediction r Intervals: r
1 1 SST = SSR + SSE
Estimation: x̄ − zα/2 σ 1 + < x0 < x̄ + zα/2 σ 1 +
  n n SSR SSE
E Θ̂ = θ ⇐⇒ Θ̂ is an unbiased estimator of θ r
1
r
1 R2 = SST =1− SST and R2 = r2
    x̄ − tn−1,α/2 s 1 + < x0 < x̄ + tn−1,α/2 s 1 +
Bias Θ̂ = E Θ̂ − θ n n n
P
e2i
2 2 i=1 SSE
2
σΘ̂ < 2
σΘ̂ ⇒ Θ̂1 is a more efficient estimator of θ than Θ̂2 σ̂ = s = n−2 = n−2
1 2
Tolerance Intervals:
∗
A tolerance interval containing at least 100(1 − γ)% of the Disclosure: This paper is to be distributed during the
Confidence Intervals: The formulae given below are population with confidence 100(1 − α)% is exam. It does not mean that you should not learn these for-
most commonly used ones. You are expected to identify mulae. Rather, you should learn these formulae and their
and use them, including their usage in hypothesis testing. x̄ ± kn, α, γ s usage in various problems. Any (implied or not) missing
formula we studied in the course is of your responsibility.
• x̄ − zα/2 √σn < µ < x̄ + zα/2 √σn
q q Hypothesis Testing:
• p̂ − zα/2 p̂(1− p̂)
< p < p̂ + z p̂(1−p̂)
(traditional) (Check the formula for Confidence Intervals)
n α/2 n
Test statistic is compared with (one sided) zα or tv,α .
q
p̃(1−p̃) Test statistic is compared with (two sided) zα/2 or tv,α/2 .
• ñ = n + 4, p̃ = X+2 ñ , p̃ − zα/2 ñ < p <
q
p̃ + zα/2 p̃(1−ñ
p̃)

Correlation:
• x̄ − tn−1,α/2 √sn < µ < x̄ + tn−1,α/2 √sn n
P
(xi −x̄)(yi −ȳ)
n
P
xi yi −nx̄ȳ
sxy i=1 i=1
q 2
σ σ2
q 2
σ σ2
r= , where sxy = n−1 = n−1 ,
• (x̄− ȳ)−zα/2 nxx + nyy < µX −µY < (x̄− ȳ)+zα/2 nxx + nyy sx sy v !2
s u n
P
q n uP n xi
p̂ (1−p̂ )
• (p̂x − p̂y ) − zα/2 p̂x (1− p̂x ) P t x2i − i=1
+ y ny y < p x − p y < (xi −x̄)2 n
nx
q sx = i=1 n−1 = i=1 n−1 and
p̂ (1−p̂ )
(p̂x − p̂y ) + zα/2 p̂x (1−
nx
p̂x )
+ y ny y (traditional) s
v
u n
P
!2
y
n
t n yi2 − i=1n i
P uP
q (yi −ȳ)2
p̃ (1−p̃ )
• (p̃x − p̃y ) ± zα/2 p̃x (1−
ñx
p̃x )
+ y ñy y sy = i=1
n−1 = i=1
n−1 .
q 2 q 2
s s2 s s2
• (x̄−ȳ)−tv,α/2 nxx + nyy < µX −µY < (x̄−ȳ)+tv,α/2 nxx + nyy
2
s2 Simple Linear Regression:

s2
x y
nx + ny
where v = !2 !2 is the degrees of freedom. Equation of the regression line: ŷ = b0 + b1 x.
s2 s2
y n
x P
nx ny
nx −1 + ny −1
xi yi − nx̄ȳ
sy sxy
(not assuming equal pop. variances) b1 = r = 2 = i=1 n and b0 = ȳ − b1 x̄.
sx sx P 2 2
q 2
s s2
q 2
s s2
xi − nx̄
• (x̄−ȳ)−tv,α/2 npx + npy < µX −µY < (x̄−ȳ)+tv,α/2 npx + npy i=1

where v = nx + ny − 2 is the degrees of freedom and