Testing two population means using two-sample z-procedure

variances known AND (both sample sizes n1 , n2 ≥ 30 OR n1 , n2 < 30 with normally distributed data)
Case 1: One-sided (upper bound) Case 2: Two-sided Case 3: One-sided (lower bound)

(1-α) level Confidence Intervals

 2   σ 12 σ 22   
 − ∞, x − x + z σ 1 + σ 2  σ 12 σ 22  x − x − z σ1 + σ 2 ,∞
2 2 2
x − x − z + , x − x + z +
 1 2 α
n1 n2   1 2 α 2
n n
1 2 α 2
n n2   1 2 α
n1 n2 
  1 2 1  

x1 − x2
Hypothesis Testing: test statistic z0 =
σ 12 σ 22
+
n1 n2
H 0 : µ1 ≥ µ 2 H 1 : µ1 < µ 2 H 0 : µ1 = µ 2 H 1 : µ1 ≠ µ 2 H 0 : µ1 ≤ µ 2 H 1 : µ1 > µ 2
p-value = Φ( z 0 ) p-value = 2 × Φ (− z 0 ) p-value = 1 − Φ( z 0 )

Size α hypothesis tests

accept H 0 reject H 0 accept H 0 reject H 0 accept H 0 reject H 0
z 0 ≥ − zα z 0 < − zα z 0 ≤ zα 2 z 0 > zα 2 z 0 ≤ zα z 0 > zα

Type II error (β), ∆ = µ1 − µ 2 in H 1 , ∆ 0 = µ1 − µ 2 in H 0 , usually ∆ 0 = 0

       
       
 ∆ − ∆0   ∆ − ∆0   ∆ − ∆0   ∆ − ∆0 
β = 1 − Φ − z α −  β = Φ z α 2 −  − Φ − z α 2 −  β = Φ  zα − 
 σ 12 σ 22   σ 12 σ 22   σ 12 σ 22   σ 12 σ 2

 + + + + 2

 n n2  
 n n 
2 

 n n2   n1 n2 
1 1 1
 
Sample size n; given α , β , ∆ 0 , ∆

n=
(z α (
+ z β ) σ 12 + σ 22
2
) (z α 2
2
(
+ z β ) σ 12 + σ 22 )
n= same as Case 1
(∆ − ∆ 0 ) 2
(∆ − ∆ 0 ) 2

Sample size n; given α and E = half-width of CI

2 2
 zα 2 
z 
(
n =  α  σ 12 + σ 22 ) n =  (
 σ 12 + σ 22 ) same as Case 1
E  E 
 Testing two population means using two independent sample t-procedure
variances unknown AND (both sample sizes n1 , n2 ≥ 30 OR n1 , n2 < 30 with normally distributed data)
2
 s12 s 22 
 + 
If the variances are NOT assumed equal, compute the degree of freedom ν =  n1 n2 
s12 n1
2

+
s 22 n2
2
( ) ( )
n1 − 1 n2 − 1
Case 1: One-sided (upper bound) Case 2: Two-sided Case 3: One-sided (lower bound)

(1-α) level Confidence Intervals

 s12 s 22   s12 s 22 s12 s 22   s12 s 22 
 − ∞, x − x + t + x − x −t + , x1 − x 2 + tα 2,ν + x − x −t + ,∞
 1 2 α ,ν
n1 n2   1 2 α 2 ,ν
n1 n2 n1 n2   1 2 α ,ν
n1 n2 
  

x1 − x 2
Hypothesis Testing: test statistic t 0 = ; X ~ tν
s12 s 22
+
n1 n2
H 0 : µ1 ≥ µ 2 H 1 : µ1 < µ 2 H 0 : µ1 = µ 2 H 1 : µ1 ≠ µ 2 H 0 : µ1 ≤ µ 2 H 1 : µ1 > µ 2
p-value = P( X < t 0 ) p-value = 2 × P( X > t 0 ) p-value = P( X > t 0 )

Size α hypothesis tests

accept H 0 reject H 0 accept H 0 reject H 0 accept H 0 reject H 0
t 0 ≥ −tα ,ν t 0 < −tα ,ν t 0 ≤ tα 2,ν t 0 > tα 2,ν t 0 ≤ tα ,ν t 0 > tα ,ν

Type II error (β), ∆ = µ1 − µ 2 in H 1 , ∆ 0 = µ1 − µ 2 in H 0 , usually ∆ 0 = 0

       
       
 ∆ − ∆0   ∆ − ∆0   ∆ − ∆0   ∆ − ∆0 
β = 1 − P X < −tα ,ν −  β = P X < tα 2,ν −  − P X < −tα 2,ν −  β = P X < tα ,ν − 
 s12 s 22   s 2
s 2
2   s12 s 22   s12 s 22 
 + 1
+ + +
 n 1 n2  
 n 1 n 2



 n 1 n2  
 n 1 n2 

Sample size n; given α , β , ∆ 0 , ∆ ; OR Sample size n; given α and E = half-width of CI

This requires the use of OC curve, or Statistical software, e.g. Minitab
 Testing two population means using two independent sample t-procedure
variances unknown AND (both sample sizes n1 , n2 ≥ 30 OR n1 , n2 < 30 with normally distributed data)
(n1 − 1) s12 + (n2 − 1) s 22
If the variances are assumed equal, compute the pool variance s = 2
p ; degree of freedom ν = n1 + n2 − 2
n1 + n2 − 2
Case 1: One-sided (upper bound) Case 2: Two-sided Case 3: One-sided (lower bound)

(1-α) level Confidence Intervals

     
 − ∞, x1 − x 2 + tα ,ν s p 1 + 1   x1 − x 2 − tα 2,ν s p 1 + 1 , x1 − x 2 + tα 2,ν s p 1 + 1   x1 − x 2 − tα ,ν s p 1 + 1 , ∞ 
 n1 n2   n1 n2 n1 n2   n1 n2 
  

x1 − x 2
Hypothesis Testing: test statistic t 0 = ; X ~ tν
1 1
sp +
n1 n2
H 0 : µ1 ≥ µ 2 H 1 : µ1 < µ 2 H 0 : µ1 = µ 2 H 1 : µ1 ≠ µ 2 H 0 : µ1 ≤ µ 2 H 1 : µ1 > µ 2
p-value = P( X < t 0 ) p-value = 2 × P( X > t 0 ) p-value = P( X > t 0 )

Size α hypothesis tests

accept H 0 reject H 0 accept H 0 reject H 0 accept H 0 reject H 0
t 0 ≥ −tα ,ν t 0 < −tα ,ν t 0 ≤ tα 2,ν t 0 > tα 2,ν t 0 ≤ tα ,ν t 0 > tα ,ν

Type II error (β), ∆ = µ1 − µ 2 in H 1 , ∆ 0 = µ1 − µ 2 in H 0 , usually ∆ 0 = 0

       
       
 ∆ − ∆0   ∆ − ∆0   ∆ − ∆0   ∆ − ∆0 
β = 1 − P X < −tα ,ν −  β = P X < tα 2,ν −  − P X < −tα 2,ν −  β = P X < tα ,ν − 
 1 1   1 1   1 1   1 1 
 s p +   s p +   s p +   s p + 
 n1 n2   n1 n2   n1 n2   n1 n2 

Sample size n; given α , β , ∆ 0 , ∆ ; OR Sample size n; given α and E = half-width of CI

This requires the use of OC curve, or Statistical software, e.g. Minitab
 Testing two population variances using two-sample F-procedure
Data must be normally distributed
Case 1: One-sided (upper bound) Case 2: Two-sided Case 3: One-sided (lower bound)

(1-α) level Confidence Intervals

 s12   s12 s2   s12 
 0, 2 fα , n2 −1, n1 −1   2 f1−α 2, n2 −1, n1 −1 , 12 fα 2, n2 −1, n1 −1   2 f1−α , n2 −1, n1 −1 , ∞ 
 s2   s2 s2   s2 

s12
Hypothesis Testing: test statistic F0 = 2 ; X ~ f n1 −1,n2 −1
s2
H 0 : σ 12 ≥ σ 22 H 1 : σ 12 < σ 22 H 0 : σ 12 = σ 22 H 1 : σ 12 ≠ σ 22 H 0 : σ 12 ≤ σ 22 H 1 : σ 12 > σ 22
p-value = 2 × P( X > F0 ) if F0 > 1
p-value = P( X < F0 ) p-value = P( X > F0 )
p-value = 2 × P( X < F0 ) if F0 < 1

Size α hypothesis tests

accept H 0 reject H 0 accept H 0 reject H 0 accept H 0 reject H 0
F0 ≥ f 1−α ,n1 −1,n2 −1 F0 < f 1−α ,n1 −1,n2 −1 f 1−α 2,n1 −1,n2 −1 ≤ F0 ≤ f α 2, n1 −1,n2 −1 F0 > f α 2,n1 −1,n2 −1 or F0 < f1−α 2,n1 −1,n2 −1 F0 ≤ f α ,n1 −1,n2 −1 F0 > f α , n1 −1,n2 −1

Type II error and sample size n are usually not of interest

 Testing two population proportions using two-sample z-procedure
Data are from binomial distribution, X1 ~ Bin(n1, p1), X2 ~ Bin(n2, p2)
Case 3: One-sided (lower
Case 1: One-sided (upper bound) Case 2: Two-sided
bound)

x1 x x + x2
First of all, compute these estimates pˆ 1 = , qˆ1 = 1 − pˆ 1 ; pˆ 2 = 2 , qˆ 2 = 1 − pˆ 2 , pˆ = 1
n1 n2 n1 + n2

(1-α) level Confidence Intervals

 pˆ 1 qˆ1 pˆ 2 qˆ 2   pˆ 1 qˆ1 pˆ 2 qˆ 2 pˆ 1 qˆ1 pˆ 2 qˆ 2   pˆ 1 qˆ1 pˆ 2 qˆ 2 
 − 1, pˆ 1 − pˆ 2 + zα +   pˆ 1 − pˆ 2 − zα 2 + , pˆ 1 − pˆ 2 + zα 2 +   pˆ 1 − pˆ 2 − zα + ,1
 n1 n2   n1 n2 n1 n2   n1 n2 
    

pˆ 1 − pˆ 2
Hypothesis Testing: test statistic z 0 =
1 1 
pˆ (1 − pˆ ) + 
 n1 n2 
H 0 : p1 ≥ p 2 H 1 : p1 < p 2 H 0 : p1 = p 2 H 1 : p1 ≠ p 2 H 0 : p1 ≤ p 2 H 1 : p1 > p 2
p-value = Φ( z 0 ) p-value = 2 × Φ (− z 0 ) p-value = 1 − Φ( z 0 )

Size α hypothesis tests

accept H 0 reject H 0 accept H 0 reject H 0 accept H 0 reject H 0
z 0 ≥ − zα z 0 < − zα z 0 ≤ zα 2 z 0 > zα 2 z 0 ≤ zα z 0 > zα
 Testing two population proportions using two-sample z-procedure
Data are from binomial distribution, X1 ~ Bin(n1, p1), X2 ~ Bin(n2, p2)
Case 1: One-sided (upper bound) Case 2: Two-sided Case 3: One-sided (lower bound)

Type II error β,
H 0 : p1 ≥ p 2 H 1 : p1 < p 2 H 0 : p1 = p 2 H 1 : p1 ≠ p 2 H 0 : p1 ≤ p 2 H 1 : p1 > p 2
 1 1   1 1   1 1 
−z pˆ (1 − pˆ ) +  − ( p1 − p2 )  z pˆ (1 − pˆ ) +  − ( p1 − p2 )  z pˆ (1 − pˆ ) +  − ( p1 − p2 ) 
 α  n1 n2    α 2
 n1 n2    α  n1 n2  
β = 1 − Φ  β = Φ  β = Φ 
 p1q1 p2 q2   p1q1 p2 q2   p1q1 p2 q2 
 +   +   + 
n1 n2 n1 n2 n1 n2
     
 1 1 
−z ˆ
p (1 − ˆ
p ) +  − ( p − p ) 
 α2 n n  1 2

 1 2 
− Φ 
 p1q1 p2 q2 
 + 
n1 n2
 

Sample size n; given α , β , d = p1 − p 2 from H 1

2 2
 ( pˆ 1 + pˆ 2 )(qˆ1 + qˆ 2 )   ( pˆ 1 + pˆ 2 )(qˆ1 + qˆ 2 ) 
 zα + zβ pˆ 1 qˆ1 + pˆ 2 qˆ 2   zα 2 + zβ pˆ 1 qˆ1 + pˆ 2 qˆ 2 

n= 2  
n= 2  same as case 1
d  d 
   
   

Sample size n; given α and E = half-width of CI

zα2 ( pˆ 1 qˆ1 + pˆ 2 qˆ 2 ) zα2 2 ( pˆ 1 qˆ1 + pˆ 2 qˆ 2 )
n= n= same as case 1
E2 E2