一、单因素
设单因素
A
A
A有
a
a
a个水平
A
1
,
A
2
,
.
.
.
,
A
a
A_1,A_2,...,A_a
A1,A2,...,Aa,在水平
A
i
(
i
=
1
,
2
,
.
.
.
,
a
)
A_i(i=1,2,...,a)
Ai(i=1,2,...,a)下,进行
n
i
(
n
=
∑
i
=
1
a
n
i
)
n_i(n=\sum\limits_{i=1}^an_i)
ni(n=i=1∑ani)次独立试验,得到试验指标的观测值,如下表所示:
1
2
⋯
n
A
1
x
11
x
12
⋯
x
1
n
1
A
2
x
21
x
22
⋯
x
2
n
2
⋮
⋮
⋮
⋯
⋮
A
i
x
i
1
x
i
2
⋯
x
1
n
i
⋮
⋮
⋮
⋯
⋮
A
a
x
a
1
x
a
2
⋯
x
a
n
a
\begin{array}{c|cccc} \hline & 1 & 2 & \cdots & n \\ \hline A_1 & x_{11} & x_{12} & \cdots & x_{1n_1} \\ A_2 & x_{21} & x_{22} & \cdots& x_{2n_2} \\ \vdots & \vdots & \vdots&\cdots&\vdots \\ A_i & x_{i1} & x_{i2} & \cdots & x_{1n_i} \\ \vdots & \vdots & \vdots & \cdots & \vdots \\ A_a & x_{a1} & x_{a2} & \cdots & x_{an_a} \\ \hline \end{array}
A1A2⋮Ai⋮Aa1x11x21⋮xi1⋮xa12x12x22⋮xi2⋮xa2⋯⋯⋯⋯⋯⋯⋯nx1n1x2n2⋮x1ni⋮xana
假设在各个水平
A
i
(
i
=
1
,
2
,
⋯
,
a
)
A_i(i=1,2,\cdots,a)
Ai(i=1,2,⋯,a)下的样本为
X
i
1
,
X
i
2
,
⋯
,
X
i
n
i
X_{i1},X_{i2},\cdots,X_{in_i}
Xi1,Xi2,⋯,Xini,它们来自具有相同方差
σ
2
\sigma^2
σ2、均值为
μ
i
\mu_i
μi的正态总体
X
i
∼
N
(
μ
i
,
σ
2
)
X_i\sim N(\mu_i,\sigma^2)
Xi∼N(μi,σ2),其中
μ
i
,
σ
2
\mu_i,\sigma^2
μi,σ2均为未知,并且不同水平
A
i
A_i
Ai下的样本之间相互独立。
记
x
i
⋅
=
∑
j
=
1
n
i
x
i
j
,
x
⋅
⋅
=
∑
i
=
1
a
∑
j
=
1
n
i
x
i
j
x_{i\cdot}=\sum\limits_{j=1}^{n_i}x_{ij},\ \ \ \ x_{\cdot\cdot}=\sum\limits_{i=1}^a\sum\limits_{j=1}^{n_i}x_{ij}
xi⋅=j=1∑nixij, x⋅⋅=i=1∑aj=1∑nixij
S
T
=
∑
i
=
1
a
∑
j
=
1
n
i
x
i
j
2
−
x
⋅
⋅
2
n
S
A
=
∑
i
=
1
a
x
i
⋅
2
n
i
−
x
⋅
⋅
2
n
S
E
=
S
T
−
S
A
}
\left. \begin{aligned} &S_T=\sum\limits_{i=1}^a\sum\limits_{j=1}^{n_i}x_{ij}^2-\frac{x_{\cdot\cdot}^2}{n} \\ &S_A=\sum\limits_{i=1}^a\frac{x_{i\cdot}^2}{n_i}-\frac{x_{\cdot\cdot}^2}{n} \\ &S_E=S_T-S_A \\ \end{aligned} \right\}
ST=i=1∑aj=1∑nixij2−nx⋅⋅2SA=i=1∑anixi⋅2−nx⋅⋅2SE=ST−SA⎭⎪⎪⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎪⎪⎫
方差分析表
方差来源 | 平方和 | 自由度 | 均方 | F比 |
---|---|---|---|---|
因素 A A A | S A S_A SA | a − 1 a-1 a−1 | M S A = S A a − 1 MS_A=\frac{S_A}{a-1} MSA=a−1SA | F = M S A M S E F=\frac{MS_A}{MS_E} F=MSEMSA |
误差 E E E | S E S_E SE | n − a n-a n−a | M S E = S E n − a MS_E=\frac{S_E}{n-a} MSE=n−aSE | |
总和 T T T | S A S_A SA | n − 1 n-1 n−1 |
判断:
(1)若
F
>
F
α
(
a
−
1
,
n
−
a
)
F>F_\alpha(a-1,n-a)
F>Fα(a−1,n−a),则拒绝
H
0
H_0
H0,接受
H
1
H_1
H1,因素
A
A
A有显著影响。
(2)若
F
<
F
α
(
a
−
1
,
n
−
a
)
F<F_\alpha(a-1,n-a)
F<Fα(a−1,n−a),则接受
H
0
H_0
H0,因素
A
A
A无显著影响。
二、双因素
- 无交互作用
设两因素 A 、 B A、B A、B, A A A有 a a a个水平: A 1 , A 2 , ⋯ , A a A_1,A_2,\cdots,A_a A1,A2,⋯,Aa, B B B有 b b b个水平: B 1 , B 2 , ⋯ , B b B_1,B_2,\cdots,B_b B1,B2,⋯,Bb,在每一个组合水平 ( A i , B j ) (A_i,B_j) (Ai,Bj)下,做一次试验(无重复试验)得出试验指标的观察值,如下表所示:
因 素 B ( j ) 因 素 A ( i ) B 1 B 2 ⋯ B j ⋯ B b x i ⋅ A 1 x 11 x 12 ⋯ x 1 j ⋯ x 1 b x 1 ⋅ A 2 x 21 x 22 ⋯ x 2 j ⋯ x 2 b x 2 ⋅ ⋮ ⋮ ⋮ ⋯ ⋮ ⋯ ⋮ ⋮ A i x i 1 x i 2 ⋯ x i j ⋯ x i b x i ⋅ ⋮ ⋮ ⋮ ⋯ ⋮ ⋯ ⋮ ⋮ A a x a 1 x a 2 ⋯ x a j ⋯ x a b x a ⋅ x ⋅ j x ⋅ 1 x ⋅ 2 ⋯ x ⋅ j ⋯ x ⋅ b x ⋅ ⋅ \begin{array}{c|cccccc|c} \hline &因素B(j)\\因素A(i) & B_1 & B_2 & \cdots & B_j & \cdots & B_b & x_{i\cdot} \\ \hline A_1 & x_{11} & x_{12} & \cdots & x_{1j} & \cdots & x_{1b} & x_{1\cdot} \\ A_2 & x_{21} & x_{22} & \cdots & x_{2j} & \cdots & x_{2b} & x_{2\cdot} \\ \vdots & \vdots & \vdots & \cdots & \vdots & \cdots & \vdots & \vdots \\ A_i & x_{i1} & x_{i2} & \cdots & x_{ij} & \cdots & x_{ib} & x_{i\cdot} \\ \vdots & \vdots & \vdots & \cdots & \vdots & \cdots & \vdots & \vdots \\ A_a & x_{a1} & x_{a2} & \cdots & x_{aj} & \cdots & x_{ab} & x_{a\cdot} \\ \hline x_{\cdot j} & x_{\cdot 1} & x_{\cdot 2} & \cdots & x_{\cdot j} & \cdots & x_{\cdot b} & x_{\cdot \cdot} \\ \hline \end{array} 因素A(i)A1A2⋮Ai⋮Aax⋅j因素B(j)B1x11x21⋮xi1⋮xa1x⋅1B2x12x22⋮xi2⋮xa2x⋅2⋯⋯⋯⋯⋯⋯⋯⋯Bjx1jx2j⋮xij⋮xajx⋅j⋯⋯⋯⋯⋯⋯⋯⋯Bbx1bx2b⋮xib⋮xabx⋅bxi⋅x1⋅x2⋅⋮xi⋅⋮xa⋅x⋅⋅
设 x i j ∼ N ( μ i , σ 2 ) x_{ij}\sim N(\mu_i,\sigma^2) xij∼N(μi,σ2),各 x i j x_{ij} xij相互独立, i = 1 , 2 , ⋯ , a , j = 1 , 2 , ⋯ , b . i=1,2,\cdots,a, \ \ \ j=1,2,\cdots,b. i=1,2,⋯,a, j=1,2,⋯,b.
记
x i ⋅ = ∑ j = 1 b x i j , x ⋅ j = ∑ i = 1 a x i j x_{i\cdot}=\sum\limits_{j=1}^{b}x_{ij},\ \ \ \ x_{\cdot j}=\sum\limits_{i=1}^ax_{ij} xi⋅=j=1∑bxij, x⋅j=i=1∑axij
S T = ∑ i = 1 a ∑ j = 1 b x i j 2 − x ⋅ ⋅ 2 a b S A = ∑ i = 1 a x i ⋅ 2 b − x ⋅ ⋅ 2 a b S B = ∑ j = 1 b x ⋅ j 2 a − x ⋅ ⋅ 2 a b S E = S T − S A − S B } \left. \begin{aligned} &S_T=\sum\limits_{i=1}^a\sum\limits_{j=1}^bx_{ij}^2-\frac{x_{\cdot\cdot}^2}{ab} \\ &S_A=\sum\limits_{i=1}^a\frac{x_{i\cdot}^2}{b}-\frac{x_{\cdot\cdot}^2}{ab} \\ &S_B=\sum\limits_{j=1}^b\frac{x_{\cdot j}^2}{a}-\frac{x_{\cdot\cdot}^2}{ab} \\ &S_E=S_T-S_A-S_B \\ \end{aligned} \right\} ST=i=1∑aj=1∑bxij2−abx⋅⋅2SA=i=1∑abxi⋅2−abx⋅⋅2SB=j=1∑bax⋅j2−abx⋅⋅2SE=ST−SA−SB⎭⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎫
方差分析表
方差来源 | 平方和 | 自由度 | 均方 | F比 |
---|---|---|---|---|
因素 A A A | S A S_A SA | a − 1 a-1 a−1 | M S A = S A a − 1 MS_A=\frac{S_A}{a-1} MSA=a−1SA | F 1 = M S A M S E F_1=\frac{MS_A}{MS_E} F1=MSEMSA |
因素 B B B | S B S_B SB | b − 1 b-1 b−1 | M S B = S B b − 1 MS_B=\frac{S_B}{b-1} MSB=b−1SB | F 2 = M S B M S E F_2=\frac{MS_B}{MS_E} F2=MSEMSB |
误差 E E E | S E S_E SE | ( a − 1 ) ( b − 1 ) (a-1)(b-1) (a−1)(b−1) | M S E = S E ( a − 1 ) ( b − 1 ) MS_E=\frac{S_E}{(a-1)(b-1)} MSE=(a−1)(b−1)SE | |
总和 T T T | S A S_A SA | a b − 1 ab-1 ab−1 |
判断:
(1)若
F
1
>
F
α
(
a
−
1
,
(
a
−
1
)
(
b
−
1
)
)
F_1>F_\alpha(a-1,(a-1)(b-1))
F1>Fα(a−1,(a−1)(b−1)),则拒绝
H
A
0
H_{A0}
HA0,接受
H
A
1
H_{A1}
HA1,因素
A
A
A有显著影响;若
F
1
<
F
α
(
a
−
1
,
(
a
−
1
)
(
b
−
1
)
)
F_1<F_\alpha(a-1,(a-1)(b-1))
F1<Fα(a−1,(a−1)(b−1)),则接受
H
A
0
H_{A0}
HA0,因素
A
A
A无显著影响。
(2)若
F
2
>
F
α
(
b
−
1
,
(
a
−
1
)
(
b
−
1
)
)
F_2>F_\alpha(b-1,(a-1)(b-1))
F2>Fα(b−1,(a−1)(b−1)),则拒绝
H
B
0
H_{B0}
HB0,接受
H
B
1
H_{B1}
HB1,因素
B
B
B有显著影响;若
F
2
<
F
α
(
b
−
1
,
(
a
−
1
)
(
b
−
1
)
)
F_2<F_\alpha(b-1,(a-1)(b-1))
F2<Fα(b−1,(a−1)(b−1)),则接受
H
B
0
H_{B0}
HB0,因素
B
B
B无显著影响。
- 有交互作用
设两因素 A 、 B A、B A、B, A A A有 a a a个水平: A 1 , A 2 , ⋯ , A a A_1,A_2,\cdots,A_a A1,A2,⋯,Aa, B B B有 b b b个水平: B 1 , B 2 , ⋯ , B b B_1,B_2,\cdots,B_b B1,B2,⋯,Bb,为研究交互作用的影响,在每一个组合水平 ( A i , B j ) (A_i,B_j) (Ai,Bj)下重复做 n n n次 ( n ≥ 2 ) (n\geq2) (n≥2)试验,每个观察值记为 x i j k , k = 1 , 2 , ⋯ , n x_{ijk},\ k=1,2,\cdots,n xijk, k=1,2,⋯,n,如下表所示:
B ( j ) A ( i ) B 1 B 2 ⋯ B b A 1 x 111 x 112 ⋯ x 11 n x 121 x 122 ⋯ x 12 n ⋯ x 1 b 1 x 1 b 2 ⋯ x 1 b n A 2 x 211 x 212 ⋯ x 21 n x 221 x 222 ⋯ x 22 n ⋯ x 2 b 1 x 2 b 2 ⋯ x 2 b n ⋮ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ ⋯ A a x a 11 x a 12 ⋯ x a 1 n x a 21 x a 22 ⋯ x a 2 n ⋯ x a b 1 x a b 2 ⋯ x a b n \begin{array}{c|cccc} \hline &B(j)\\A(i) & B_1 & B_2 & \cdots & B_b \\ \hline A_1 & x_{111}\ x_{112}\cdots x_{11n} & x_{121}\ x_{122}\cdots x_{12n} & \cdots & x_{1b1}\ x_{1b2}\cdots x_{1bn} \\ A_2 & x_{211}\ x_{212}\cdots x_{21n} & x_{221}\ x_{222}\cdots x_{22n} & \cdots & x_{2b1}\ x_{2b2}\cdots x_{2bn} \\ \vdots & \cdots\ \cdots\ \cdots & \cdots\ \cdots\ \cdots & \cdots\ \cdots\ \cdots & \cdots \\ A_a & x_{a11}\ x_{a12}\cdots x_{a1n} & x_{a21}\ x_{a22}\cdots x_{a2n} & \cdots & x_{ab1}\ x_{ab2}\cdots x_{abn} \\ \hline \end{array} A(i)A1A2⋮AaB(j)B1x111 x112⋯x11nx211 x212⋯x21n⋯ ⋯ ⋯xa11 xa12⋯xa1nB2x121 x122⋯x12nx221 x222⋯x22n⋯ ⋯ ⋯xa21 xa22⋯xa2n⋯⋯⋯⋯ ⋯ ⋯⋯Bbx1b1 x1b2⋯x1bnx2b1 x2b2⋯x2bn⋯xab1 xab2⋯xabn
设 x i j k ∼ N ( μ i j , σ 2 ) x_{ijk}\sim N(\mu_{ij},\sigma^2) xijk∼N(μij,σ2),各 x i j k x_{ijk} xijk相互独立, i = 1 , 2 , ⋯ , a , j = 1 , 2 , ⋯ , b , k = 1 , 2 , ⋯ , n . i=1,2,\cdots,a, \ \ \ j=1,2,\cdots,b, \ \ \ k=1,2,\cdots,n. i=1,2,⋯,a, j=1,2,⋯,b, k=1,2,⋯,n.
记
x i ⋅ ⋅ = ∑ j = 1 b ∑ k = 1 n x i j k , x ⋅ j ⋅ = ∑ i = 1 a ∑ k = 1 n x i j k , x ⋯ = ∑ i = 1 a ∑ j = 1 b ∑ k = 1 n x i j k . x_{i\cdot\cdot}=\sum\limits_{j=1}^{b}\sum\limits_{k=1}^{n}x_{ijk},\ \ \ x_{\cdot j\cdot}=\sum\limits_{i=1}^a\sum\limits_{k=1}^{n}x_{ijk},\\ x_{\cdots}=\sum\limits_{i=1}^a\sum\limits_{j=1}^{b}\sum\limits_{k=1}^{n}x_{ijk}. xi⋅⋅=j=1∑bk=1∑nxijk, x⋅j⋅=i=1∑ak=1∑nxijk,x⋯=i=1∑aj=1∑bk=1∑nxijk.
S T = ∑ i = 1 a ∑ j = 1 b ∑ k = 1 n x i j k 2 − x ⋯ 2 a b n S A = ∑ i = 1 a x i ⋅ ⋅ 2 b n − x ⋯ 2 a b n S B = ∑ j = 1 b x ⋅ j ⋅ 2 a n − x ⋯ 2 a b n S A × B = ∑ i = 1 a ∑ j = 1 b x i j ⋅ 2 n − x ⋯ 2 a b n − S A − S B S E = S T − S A − S B − S A × B } \left. \begin{aligned} &S_T=\sum\limits_{i=1}^a\sum\limits_{j=1}^{b}\sum\limits_{k=1}^{n}x_{ijk}^2-\frac{x_{\cdots}^2}{abn} \\ &S_A=\sum\limits_{i=1}^a\frac{x_{i\cdot\cdot}^2}{bn}-\frac{x_{\cdots}^2}{abn} \\ &S_B=\sum\limits_{j=1}^b\frac{x_{\cdot j\cdot}^2}{an}-\frac{x_{\cdots}^2}{abn} \\ &S_{A\times B}=\sum\limits_{i=1}^a\sum\limits_{j=1}^{b}\frac{x_{ij\cdot} ^2}{n}-\frac{x_{\cdots}^2}{abn}-S_A-S_B \\ &S_E=S_T-S_A-S_B-S_{A\times B} \\ \end{aligned} \right\} ST=i=1∑aj=1∑bk=1∑nxijk2−abnx⋯2SA=i=1∑abnxi⋅⋅2−abnx⋯2SB=j=1∑banx⋅j⋅2−abnx⋯2SA×B=i=1∑aj=1∑bnxij⋅2−abnx⋯2−SA−SBSE=ST−SA−SB−SA×B⎭⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎬⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎪⎫
方差分析表
方差来源 | 平方和 | 自由度 | 均方 | F比 |
---|---|---|---|---|
因素 A A A | S A S_A SA | a − 1 a-1 a−1 | M S A = S A a − 1 MS_A=\frac{S_A}{a-1} MSA=a−1SA | F 1 = M S A M S E F_1=\frac{MS_A}{MS_E} F1=MSEMSA |
因素 B B B | S B S_B SB | b − 1 b-1 b−1 | M S B = S B b − 1 MS_B=\frac{S_B}{b-1} MSB=b−1SB | F 2 = M S B M S E F_2=\frac{MS_B}{MS_E} F2=MSEMSB |
交互作用 A × B \\ A\times B A×B | S A × B S_{A\times B} SA×B | ( a − 1 ) ( b − 1 ) (a-1)(b-1) (a−1)(b−1) | M S A × B = S A × B ( a − 1 ) ( b − 1 ) MS_{A\times B}=\frac{S_{A\times B}}{(a-1)(b-1)} MSA×B=(a−1)(b−1)SA×B | F 3 = M S A × B M S E F_3=\frac{MS_{A\times B}}{MS_E} F3=MSEMSA×B |
误差 E E E | S E S_E SE | a b ( n − 1 ) ab(n-1) ab(n−1) | M S E = S E a b ( n − 1 ) MS_E=\frac{S_E}{ab(n-1)} MSE=ab(n−1)SE | |
总和 T T T | S A S_A SA | a b n − 1 abn-1 abn−1 |
判断:
(1)若
F
1
>
F
α
(
a
−
1
,
a
b
(
n
−
1
)
)
F_1>F_\alpha(a-1,ab(n-1))
F1>Fα(a−1,ab(n−1)),则拒绝
H
A
0
H_{A0}
HA0,接受
H
A
1
H_{A1}
HA1,因素
A
A
A有显著影响;若
F
1
<
F
α
(
a
−
1
,
a
b
(
n
−
1
)
)
F_1<F_\alpha(a-1,ab(n-1))
F1<Fα(a−1,ab(n−1)),则接受
H
A
0
H_{A0}
HA0,因素
A
A
A无显著影响。
(2)若
F
2
>
F
α
(
b
−
1
,
a
b
(
n
−
1
)
)
F_2>F_\alpha(b-1,ab(n-1))
F2>Fα(b−1,ab(n−1)),则拒绝
H
B
0
H_{B0}
HB0,接受
H
B
1
H_{B1}
HB1,因素
B
B
B有显著影响;若
F
2
<
F
α
(
b
−
1
,
a
b
(
n
−
1
)
)
F_2<F_\alpha(b-1,ab(n-1))
F2<Fα(b−1,ab(n−1)),则接受
H
B
0
H_{B0}
HB0,因素
B
B
B无显著影响。
(3)若
F
3
>
F
α
(
(
a
−
1
)
(
b
−
1
)
,
a
b
(
n
−
1
)
)
F_3>F_\alpha((a-1)(b-1),ab(n-1))
F3>Fα((a−1)(b−1),ab(n−1)),则拒绝
H
A
B
0
H_{AB0}
HAB0,接受
H
A
B
1
H_{AB1}
HAB1,交互作用
A
×
B
A\times B
A×B有显著影响;若
F
2
<
F
α
(
(
a
−
1
)
(
b
−
1
)
,
a
b
(
n
−
1
)
)
F_2<F_\alpha((a-1)(b-1),ab(n-1))
F2<Fα((a−1)(b−1),ab(n−1)),则接受
H
A
B
0
H_{AB0}
HAB0,交互作用
A
×
B
A\times B
A×B无显著影响。
PS:
利用EXCEL进行方差分析(数据来源:应用概率统计 陈魁 清华大学出版社 习题10.4)