You are on page 1of 26

Anova

Arun Kumar, Ravindra Gokhale, and Nagarajan


Krishnamurthy
Quantitative Techniques-I, Term I, 2012
Indian Institute of Management Indore

Moto-Tech Manufacturing Company


Moto-tech produces wafers (thin slice of semiconductor
material) that is required for the fabrication of chips.

Moto-Tech Manufacturing Company


Moto-tech produces wafers (thin slice of semiconductor
material) that is required for the fabrication of chips.
Company wants to improve the quality of wafers
produced.

Moto-Tech Manufacturing Company


Moto-tech produces wafers (thin slice of semiconductor
material) that is required for the fabrication of chips.
Company wants to improve the quality of wafers
produced.
There are three potential factors that may affect the
quality of wafers; temperature, time, and supplier of the
raw material.

Moto-Tech Manufacturing Company


Moto-tech produces wafers (thin slice of semiconductor
material) that is required for the fabrication of chips.
Company wants to improve the quality of wafers
produced.
There are three potential factors that may affect the
quality of wafers; temperature, time, and supplier of the
raw material.
Engineers were, however, convinced that time does not
have any effect on the quality of wafers.

Moto-Tech Manufacturing Company


Moto-tech produces wafers (thin slice of semiconductor
material) that is required for the fabrication of chips.
Company wants to improve the quality of wafers
produced.
There are three potential factors that may affect the
quality of wafers; temperature, time, and supplier of the
raw material.
Engineers were, however, convinced that time does not
have any effect on the quality of wafers.
To test whether, supplier and temperature has any effect,
they collected data.

Engineering Question

What is the engineering question?

Engineering Question

What is the engineering question?

Average length of wafers for different temperature level is the


same or not.

Statistical Problem

Null Hypothesis
H0 : h = l = m

(1)

Alternative Hypothesis
HA : At least one is different from all the other s.

(2)

Shall We Do Three Different t-tests?

What will happen if we test for h = l , h = m , and


l = m separately?

Bonferroni Inequality
Let A1 denotes the event that hypothesis 1 is rejected
correctly and A2 denotes the event that hypothesis 2 is
rejected correctly. If we are conducting the tests at 5%
significance level then P(A1 ) = 0.95 and P(A2 ) = 0.95. We
can prove that P(A1 A2 ) 0.90.

Proof

How Anova Helps?

We want to ensure that P(A1 A2 ) 0.95. Anova ensures


that.

How Anova Helps?

We want to ensure that P(A1 A2 ) 0.95. Anova ensures


that.

Anova Output for the Moto-tech Data


ANOVA
Angstroms

Between Groups
Within Groups
Total

Sum of
Squares
.718
2151.076
2151.794

df
2
132
134

Mean Square
.359
16.296

F
.022

Sig.
.978

Calculating Sum of Squares


Sum of squares for within groups is also known as SSE (Sum
of squares for error)

SSE =

ni
n X
X

(yij yi. )2 ,

i=1 j=1

where n is the number of treatments and yi. is the mean


response for the i th treatment.

(3)

Calculating Sum of Squares


Sum of squares total is also known as SST (Sum of squares
total)

ni
n X
X
SST =
(yij y.. )2 ,

(4)

i=1 j=1

where n is the number of treatments and y.. is the mean of all


the responses for all the treatments.

Calculating Sum of Squares

Sum of squares for between group is also known as SSR (Sum


of squares for regression)

SSR = SST SSE

(5)

Degrees of Freedom

Degrees of Freedom for SSR is n-1, where n is the total


number of treatments.
P
Degrees of Freedom for SST is ni=1 ni 1, where ni is the
total number of responses for the treatment i.

Degrees of Freedom for SSE is dfSST dfSSR .

MSR (mean square for regression) and MSE (mean


square for error)

MSR = SSR .
dfSSR

MSE = SSE .
dfSSE

F-distribution

F-statistic=MSR/MSE follows an F-distribution with


numerator degrees of freedom dfSSR and denominator degrees
of freedom dfSSE .

Interpret the Results


ANOVA
Angstroms

Between Groups
Within Groups
Total

Sum of
Squares
.718
2151.076
2151.794

df
2
132
134

Mean Square
.359
16.296

F
.022

Sig.
.978

p-value

If F-statistic is F , numerator degrees of freedom is n 1, and


denominator degrees of freedom is m 1 then
p-value=P(F(n1,m1) > F ).

Conclusion

There is no difference in the mean length of wafers produced


at different temperatures.

Assumptions

Anova is very sensitive to violation of the following


assumptions.
Each Population is normally distributed.
Variances of the populations are the same.

How Do We Detect Violation of Assumptions?

Normal probability (quantile) plot (p-p plot, q-q plot) to


detect departure from normality.
F-test, Levenes test, Bartletts tests are used to test for
the equality of the variances.

Descriptive Measures for the Moto-tech Data


Descriptives
Angstroms

N
.00
1.00
2.00
Total

45
45
45
135

Mean
3010.0490
3010.2070
3010.2001
3010.1521

Std. Deviation
4.67569
1.84706
4.85946
4.00726

Std. Error
.69701
.27534
.72441
.34489

Descriptives
Angstroms

.00
1.00
2.00
Total

Minimum
3001.72
3006.75
3001.84
3001.72

Maximum
3019.43
3014.20
3019.76
3019.76

95% Confidence Interval for


Mean
Lower Bound
Upper Bound
3008.6443
3011.4538
3009.6521
3010.7619
3008.7402
3011.6601
3009.4699
3010.8342

You might also like