DoE复习笔记
Analysis procedure of one-way experiment
factor | n data | sum |
---|---|---|
A 1 A_{1} A1 | T A 1 T_{A_{1}} TA1 | |
A i A_{i} Ai | x i j x_{ij} xij | T A i T_{A_{i}} TAi |
A a A_{a} Aa | T A a T_{A_{a}} TAa | |
Total Sum | T |
- Calculate the correction term
C T = ( S u m o f a l l d a t a ) 2 n u m b e r o f a l l d a t a CT=\frac{(Sum\ of\ all\ data)^2}{number\ of\ all\ data} CT=number of all data(Sum of all data)2 - Compute the total sum of squares.
S T = ∑ ( S q u a r e d o f i n d i v i d u a l d a t a ) − C T f T = ( n u m b e r o f d a t a ) − 1 S_{T}=\sum (Squared\ of\ individual\ data)-CT \\ f_{T}=(number\ of\ data)-1 ST=∑(Squared of individual data)−CTfT=(number of data)−1 - Calculate the variation of the factor.
S A = ∑ ( S u m o f A i ′ s d a t a ) 2 ( N u m e r o f A i ′ s d a t a ) 2 − C T f A = ( n u m b e r o f l e v e l s ) − 1 S_{A}=\sum\frac{(Sum\ of A_{i}'s\ data)^2}{(Numer\ of A_{i}'s\ data)^2}-CT \\ f_{A}=(number\ of\ levels)-1 SA=∑(Numer ofAi′s data)2(Sum ofAi′s data)2−CTfA=(number of levels)−1 - Calculate the variation of the error
S e = S T − S A f e = f T − f A S_{e}=S_{T}-S_{A} \\ f_{e}=f_{T}-f_{A} Se=ST−SAfe=fT−fA - Create an analysis of variance table
a i s t h e n u m b e r o f l e v e l , n i s t h e n u m b e r o f r e p l i c a t i o n a\ is\ the\ number\ of\ level,n\ is\ the\ number\ of\ replication a is the number of level,n is the number of replication
factor | S S S | f f f | V V V | F 0 F_{0} F0 |
---|---|---|---|---|
A A A | S A S_{A} SA | f A = a − 1 f_{A}=a-1 fA=a−1 | V A = S A / f A V_{A}=S_{A}/f_{A} VA=SA/fA | f 0 = V A / V e ( m a r k ) f_{0}=V_{A}/V_{e}(mark) f0=VA/Ve(mark) |
e e e | S e S_{e} Se | f e = f T − f A = a ( n − 1 ) f_{e}=f_{T}-f_{A}=a(n-1) fe=fT−fA=a(n−1) | V e = S e / f e V_{e}=S_{e}/f_{e} Ve=Se/fe | – |
T T T | S T S_{T} ST | f T = a n − 1 f_{T}=an-1 fT=an−1 | – | – |
- Perform the statistical estimation.
- about the mark
- The variance ratio is compared with the value of F table F f e f A ( α ) F_{f_{e}}^{f_{A}}(\alpha ) FfefA(α)
- If the factor becomes significant at a significant level of 5 % ( > F f e f A ( 0.05 ) ) 5\% (>F_{f_{e}}^{f_{A}}(0.05)) 5%(>FfefA(0.05)), mark * is added to the right of the variance ratio.
- If the factor becomes significant at a significant level of 1 % ( > F f e f A ( 0.01 ) ) 1\% (>F_{f_{e}}^{f_{A}}(0.01)) 1%(>FfefA(0.01)), add two marks * to the right of the variance ratio.
- contribution ratio
- f a c t o r c o n t r i b u t i o n r a t i o : ρ f a c t o r = S f a c t o r − f f a c t i o r × V e S T × 100 ( % ) factor\ contribution\ ratio:\rho_{factor}=\frac{S_{factor}-f_{factior}\times V_{e}}{S_{T}}\times 100(\%) factor contribution ratio:ρfactor=STSfactor−ffactior×Ve×100(%)
- e r r o r c o n t r i b u t i o n r a t i o : ρ e = 100 − ∑ ρ error\ contribution\ ratio:\rho_{e}=100 -\sum\rho error contribution ratio:ρe=100−∑ρ
- statistical estimation
- estimation of mean ( m e a n a t t h e l e v e l ) ± t ( f e , 0.05 ) V e n (mean\ at\ the\ level)\pm t(f_{e},0.05)\sqrt{\frac{V_{e}}{n}} (mean at the level)±t(fe,0.05)nVe
- estimation of the difference between means ( t h e d i f f e r e n c e b e t w e e n m e a n s ) ± t ( f e , 0.05 ) 2 V e n ( A ˉ i − A ˉ j ) ± t ( f e , 0.05 ) ( 1 n i + 1 n j ) V e (the\ difference\ between\ means)\pm t(f_{e},0.05)\sqrt{\frac{2V_{e}}{n}}\\ (\bar A_{i}-\bar A_{j}) \pm t(f_{e},0.05)\sqrt{(\frac{1}{n_{i}}+\frac{1}{n_{j}})V_{e}} (the difference between means)±t(fe,0.05)n2Ve(Aˉi−Aˉj)±t(fe,0.05)(ni1+nj1)Ve
- the existence range of individual data ( m e a n a t t h e l e v e l ) (mean\ at\ the\ level ) (mean at the level)
- about the mark
Two-way layout without replication
B 1 . . . . . B j . . . . . B b B_{1}.....B_{j}.....B{b} B1.....Bj.....Bb | sum | |
---|---|---|
A 1 A_{1} A1 | T A 1 T_{A_{1}} TA1 | |
A i A_{i} Ai | x i j x_{ij} xij | T A i T_{A_{i}} TAi |
A a A_{a} Aa | T A a T_{A_{a}} TAa | |
Total Sum | T B 1 . . . . . T B j . . . . . T B b T_{B_{1}}.....T_{B_{j}}.....T_{B_{b}} TB1.....TBj.....TBb | T |
- Calculate the correction term
C T = ( S u m o f a l l d a t a ) 2 n u m b e r o f a l l d a t a CT=\frac{(Sum\ of\ all\ data)^2}{number\ of\ all\ data} CT=number of all data(Sum of all data)2 - Compute the total sum of squares.
S T = ∑ ( S q u a r e d o f i n d i v i d u a l d a t a ) − C T f T = ( n u m b e r o f d a t a ) − 1 S_{T}=\sum (Squared\ of\ individual\ data)-CT \\ f_{T}=(number\ of\ data)-1 ST=∑(Squared of individual data)−CTfT=(number of data)−1 - Calculate the variation of the factor A.
S A = ∑ ( S u m o f A i ′ s d a t a ) 2 ( N u m e r o f A i ′ s d a t a ) − C T f A = ( n u m b e r o f l e v e l s ) − 1 S_{A}=\sum\frac{(Sum\ of A_{i}'s\ data)^2}{(Numer\ of A_{i}'s\ data)}-CT \\ f_{A}=(number\ of\ levels)-1 SA=∑(Numer ofAi′s data)(Sum ofAi′s data)2−CTfA=(number of levels)−1 - Calculate the variation of the factor B.
S B = ∑ ( S u m o f B i ′ s d a t a ) 2 ( N u m e r o f B i ′ s d a t a ) − C T f B = ( n u m b e r o f l e v e l s ) − 1 S_{B}=\sum\frac{(Sum\ of B_{i}'s\ data)^2}{(Numer\ of B_{i}'s\ data)}-CT \\ f_{B}=(number\ of\ levels)-1 SB=∑(Numer ofBi′s data)(Sum ofBi′s data)2−CTfB=(number of levels)−1 - Calculate the variation of the error
S e = S T − S A f e = f T − f A S_{e}=S_{T}-S_{A} \\ f_{e}=f_{T}-f_{A} Se=ST−SAfe=fT−fA - Create an analysis of variance table
a i s t h e n u m b e r o f f a c t o r A , b i s t h e n u m b e r o f f a c t o r B a\ is\ the\ number\ of\ factor\ A,b\ is\ the\ number\ of\ factor\ B a is the number of factor A,b is the number of factor B
factor | S S S | f f f | V V V | F 0 F_{0} F0 |
---|---|---|---|---|
A A A | S A S_{A} SA | f A = a − 1 f_{A}=a-1 fA=a−1 | V A = S A / f A V_{A}=S_{A}/f_{A} VA=SA/fA | f 0 = V A / V e ( m a r k ) f_{0}=V_{A}/V_{e}(mark) f0=VA/Ve(mark) |
B B B | S B S_{B} SB | f B = b − 1 f_{B}=b-1 fB=b−1 | V A = S B / f B V_{A}=S_{B}/f_{B} VA=SB/fB | f 0 = V B / V e ( m a r k ) f_{0}=V_{B}/V_{e}(mark) f0=VB/Ve(mark) |
e e e | S e S_{e} Se | f e = f T − f A − f B = ( a − 1 ) ( b − 1 ) f_{e}=f_{T}-f_{A}-f_{B} \\= (a-1)(b-1) fe=fT−fA−fB=(a−1)(b−1) | V e = S e / f e V_{e}=S_{e}/f_{e} Ve=Se/fe | – |
T T T | S T S_{T} ST | f T = a b − 1 f_{T}=ab-1 fT=ab−1 | – | – |
8. Perform the statistical estimation
-
Estimation of significant factor effects
-
estimation of mean ( m e a n a t t h e l e v e l ) ± t ( f e , 0.05 ) V e n u m b e r o f d a t a a t t h e l e v e l (mean\ at\ the\ level)\pm t(f_{e},0.05)\sqrt{\frac{V_{e}}{number\ of\ data\ at\ the\ level}} (mean at the level)±t(fe,0.05)number of data at the levelVe
-
estimation of the difference between means ( t h e d i f f e r e n c e b e t w e e n m e a n s ) ± t ( f e , 0.05 ) 2 V e n u m b e r o f d a t a a t t h e l e v e l (the\ difference\ between\ means)\pm t(f_{e},0.05)\sqrt{\frac{2V_{e}}{number\ of\ data\ at\ the\ level}} (the difference between means)±t(fe,0.05)number of data at the level2Ve
-
Selection of optimal conditions and estimation of population mean
1. If only one factor is significant, T A i T_{A_{i}} TAiis the optimal level, the population mean estimation is as follows. ( m e a n a t e a c h l e v e l ) ± t ( f e , 0.05 ) V e n u m b e r o f d a t a a t t h e l e v e l (mean\ at\ each\ level)\pm t(f_{e},0.05)\sqrt{\frac{V_{e}}{number\ of\ data\ at\ the\ level}} (mean at each level)±t(fe,0.05)number of data at the levelVe
2. If both factors are significant, is the optimal level combination, and the population mean estimate is obtained as follows μ ^ ( A i B j ) ± t ( f e , 0.05 ) V e n e = A i ˉ + B j ˉ − T ˉ ± t ( f e , 0.05 ) ( a + b − 1 ) V e a b n e = a b a + b − 1 \hat{\mu}(A_{i}B_{j})\pm t(f_{e},0.05)\sqrt {\frac{V_{e}}{n_{e}}}=\bar{A_{i}}+\bar{B_{j}}-\bar{T}\pm t(f_{e},0.05)\sqrt{\frac{(a+b-1)V_{e}}{ab}} \\ n_{e}=\frac{ab}{a+b-1} μ^(AiBj)±t(fe,0.05)neVe=Aiˉ+Bjˉ−Tˉ±t(fe,0.05)ab(a+b−1)Vene=a+b−1ab -
Estimate the difference of the mean
- T h e d i f f e r e n c e b e t w e e n t h e t w o m e a n s ± t ( f e , 0.05 ) V e n d The\ difference\ between\ the\ two\ means \pm t(f_{e},0.05)\sqrt{\frac{V_{e}}{n_{d}}} The difference between the two means±t(fe,0.05)ndVe
-
n
d
n_{d}
nd is called the effective number of replications of the mean difference
w h e n t h e m e a n d i f f e r e n c e b e t w e e n A i B j a n d A p B q ( i ≠ p , j ≠ q ) n d = a b 2 ( a + b ) ( i ≠ p , j = q ) n d = b 2 ( i = p , j ≠ q ) n d = a 2 when\ the\ mean\ difference\ between\ A_{i}B_{j}\ and\ A_{p}B_{q}\\(i\ne p,j\ne q) \ n_{d}=\frac{ab}{2(a+b)}\\ (i\ne p,j=q) \ n_{d}=\frac{b}{2}\\ (i= p,j\ne q) \ n_{d}=\frac{a}{2} when the mean difference between AiBj and ApBq(i=p,j=q) nd=2(a+b)ab(i=p,j=q) nd=2b(i=p,j=q) nd=2a
- The existence range of individual data
( m e a n f o r e a c h l e v e l o r c o m b i n a t i o n o f l e v e l s ) ± t ( f e , 0.05 ) ( 1 + 1 n e ) V e (mean\ for\ each\ level\ or\ combination\ of\ levels)\pm t(f_{e},0.05)\sqrt{(1+\frac{1}{n_e})V_e} (mean for each level or combination of levels)±t(fe,0.05)(1+ne1)Ve
Concept of estimation
Parameter | Estimation value |
---|---|
μ \mu μ | T ˉ \bar T Tˉ |
a i a_i ai | A ˉ i − T ˉ \bar A_{i}-\bar T Aˉi−Tˉ |
b i b_i bi | b ˉ i − T ˉ \bar b_{i}-\bar T bˉi−Tˉ |
( a b ) i j (ab)_{ij} (ab)ij | A i B j ˉ − B ˉ j − A ˉ i + T ˉ \bar{A_{i}B_{j}}-\bar B_{j}-\bar A_{i}+\bar T AiBjˉ−Bˉj−Aˉi+Tˉ |
- In the case that only A is significant, μ ^ ( A i B j ) = μ + a i \hat\mu(A_{i}B_{j})=\mu+a_i μ^(AiBj)=μ+ai and n e = b n_e=b ne=b
- In the case that only B is significant, μ ^ ( A i B j ) = μ + a i \hat\mu(A_{i}B_{j})=\mu+a_i μ^(AiBj)=μ+ai and n e = b n_e=b ne=b
- In the case that both A and B are significant , μ ^ ( A i B j ) = μ + a i + b j \hat\mu(A_{i}B_{j})=\mu+a_i+b_j μ^(AiBj)=μ+ai+bj and n e = a b a + b n_e=\frac{ab}{a+b} ne=a+bab
Two-way layout with replication
- Calculation of
T
A
i
B
j
T_{A_{i}B_{j}}
TAiBj
Create a table summing repetition data. - Calculation of correction term( C T CT CT)
- Calculation of sum of squares
S T , f T S_T,f_T ST,fT - Calculation of the variation of the factor A
- Calculation of the variation of the factor B
- Calculation of the variation of the combination of A and B
S A B = ∑ ∑ ( S u m o f A i B j ′ s d a t a ) 2 ( N u m b e r o f A i B j ′ s d a t a ) − C T f a b = a b − 1 S_{AB}=\sum\sum\frac{(Sum\ of\ A_iB_j's\ data)^2}{(Number\ of\ A_iB_j's\ data)}-CT \\ f_{ab}=ab-1 SAB=∑∑(Number of AiBj′s data)(Sum of AiBj′s data)2−CTfab=ab−1 - Calculation of the variation of the interaction of A and B
S A × B = S A B − S A − S B f A × B = f A B − f A − f B = ( a − 1 ) ( b − 1 ) S_{A\times B}=S_{AB}-S_A-S_B\\ f_{A\times B}=f_{AB}-f_A-f_B=(a-1)(b-1) SA×B=SAB−SA−SBfA×B=fAB−fA−fB=(a−1)(b−1) - Calculation of the variation of the error.
S e = S A B − S A − S B − S A × B = S T − S A × B f e = f T − f A B S_{e}=S_{AB}-S_A-S_B-S_{A\times B}=S_{T}-S_{A\times B} \\ f_{e}=f_{T}-f_{AB} Se=SAB−SA−SB−SA×B=ST−SA×Bfe=fT−fAB - Creation of an analysis of variance table
factor | S S S | f f f | V V V | F 0 F_{0} F0 |
---|---|---|---|---|
A A A | S A S_{A} SA | f A = a − 1 f_{A}=a-1 fA=a−1 | V A = S A / f A V_{A}=S_{A}/f_{A} VA=SA/fA | f 0 = V A / V e ( m a r k ) f_{0}=V_{A}/V_{e}(mark) f0=VA/Ve(mark) |
B B B | S B S_{B} SB | f B = b − 1 f_{B}=b-1 fB=b−1 | V A = S B / f B V_{A}=S_{B}/f_{B} VA=SB/fB | f 0 = V B / V e ( m a r k ) f_{0}=V_{B}/V_{e}(mark) f0=VB/Ve(mark) |
A × B A\times B A×B | S A × B S_{A\times B} SA×B | f A × B = ( a − 1 ) ( b − 1 ) f_{A\times B}=(a-1)(b-1) fA×B=(a−1)(b−1) | V A × B = S A × B / f A × B V_{A\times B}=S_{A\times B}/f_{A\times B} VA×B=SA×B/fA×B | f 0 = V A × B / V e ( m a r k ) f_{0}=V_{A\times B}/V_{e}(mark) f0=VA×B/Ve(mark) |
e e e | S e S_{e} Se | f e = f T − f A − f B = ( a − 1 ) ( b − 1 ) f_{e}=f_{T}-f_{A}-f_{B} \\= (a-1)(b-1) fe=fT−fA−fB=(a−1)(b−1) | V e = S e / f e V_{e}=S_{e}/f_{e} Ve=Se/fe | – |
T T T | S T S_{T} ST | f T = a b − 1 f_{T}=ab-1 fT=ab−1 | – | – |