26/05/2011

Tests of Comparison

MSc. Marcos Montes de Oca

Tests of comparison
• There are many
engineering situations
where the qualities and
the uniformities of two
products are to be
compared from the
knowledge of the test
data. The appropriate
tests are designed, and
the test data are
collected and analyzed.

1
 26/05/2011

Test of comparison
• Three distributions
are considered here:
–Normal
–Weibull
–Binomial

Tests of comparison
• Normal distribution
– The qualities of two products are
compared in terms of the differences
between their means
• When their variances are known, and
• When they are unknown but equal.
– The uniformities of two products are
compared in terms of the ratio of their
variances

2
 26/05/2011

Test of comparison
• Weibull distribution • If the test data does not
– The qualities of two follow any known
products are compared distribution the
in terms of the ratio of nonparametric methods
their medians or B10
of comparison can be
lives
used
• Binomial distribution
– The comparison is made
in terms of the
probability of occurrence

TEST DATA WHICH FOLLOW

NORMAL DISTRIBUTION

3
 26/05/2011

Comparison of the qualities of two

products when their variances are known
• Where
– x, y are sample means, and μx, μy are population means
– σ2x, σ2y are population variances
– nx, ny are sample sizes
– Z1-α/2 is the specific value of z from Table 1

1 1
 σ x2 σ y2  2  σ x2 σ y2  2
(x − y ) −  +  Z1−α ; (µ x − µ y ); (x − y ) +  +  Z1−α
 
 nx n y  2
 nx n y  2

x−y
z=
σ x2 σ y2
+
nx ny Eq.01
Standard normal distribution

Table 01

4
 26/05/2011

Example 01
• Two vibratory feeders are set to give equal flow rates
of powder. The operator of one feeder makes five
random checks on his feeder and finds that the average
flow rate for five observations is 150 g/s. The operator
of the other feeder does a similar test and finds that
for 20 observations the average is 120 g/s.
– It is known from past experience that the variance of the
first machine is 100 and the variance of the second
machine is 400
• Determine whether the average flow rate of one
feeder is significantly different from the flow rate of
the other. What is the true difference between the two
flow-rate means at 95% confidence level?

Solution 01-A

Z1−α = 1,96
2 σ x2 σ x2
x − y = 30
+ = 40 = 6,31 g s
nx nx
σ x2 = 20
nx 30
z= = 4,76
σ y2 6,31
= 20
ny

5
 26/05/2011

Analysis 01-A
• Since 4.76 > 1.96 and
the value of z is
positive, it can be
concluded (even in the
presence of the
variations in the flow-
rate data) that the flow
rate of feeder x is
significantly larger than
the flow rate of feeder y
at 95% confidence level

Solution 01-B
Z1−α = 1,96
2

x − y = 30

17,6 g s
30 ± 6,31x1,96 =  
42,4 g s

(17,6 g s < µ x − µ y < 42,4 g s )

6
 26/05/2011

Analysis 01-B
• Hence, it can be concluded
with 95 percent confidence
that the true difference in
the average flow rate of the
two machines can be as low
as 17.6 g/s and not more
than 42.4 g/s

Comparison of the qualities of two products

when their variances are unknown but equal

(x − y ) − Atα 2;v ; (µ x − µ y ); (x − y ) + Atα 2;v

Eq. 02

1
 (n X − 1) s x2 + (n y − 1) s y2 nx + n y  2

A= 
 nx + n y − 2 nx n y 
 Eq. 03

x−y
t2 =
(nX − 1) s x2 + (n y − 1) s 2y 1 1
+
nx + n y − 2 nx n y Eq. 04

7
 26/05/2011

Comparison of the qualities of two products

when their variances are unknown but equal
• Where,
– x, y are sample means, and μx, μy are population
means
– s2x, s2y are sample variances
– nx, ny are sample sizes
– t1-α/2;v is the specific value of t from Table 02
t distribution

Table 02

8
 26/05/2011

Example 02
• In order to determine the relative merits of two materials x and y,
10 specimens of x and 14 of y were tested for wear, with the
following results:

• Based on the above test data, at what confidence level can one
state that one material is better than the other (material with less
amount of wear is better)? Determine also the true difference in
average wear of the populations of material x and material y at 90,
95, and 99 percent confidence.
– The two materials x and y were fabricated on the same
machines, and the variances of the two populations, therefore,
can be assumed to be the same.

Solution 02

9
 26/05/2011

Solution 02

1-α α/2 tα/2;22

0,9 0,05 1,717
0,95 0,025 2,074
0,99 0,005 2,819

Solution 02-A
• Using Eq.03

• By means of Eq.04

− 4,95 x10 −3
t2 = = −5,35
0,925 x10 −3

10
 26/05/2011

Analysis 02-A
• Since t2 = 5.35 is
considerably higher than
the value tα/2; 22 = 2.819
for 99% confidence
• It can be concluded with
more than 99%
confidence that the
average wear rate x is
significantly less than y
(since t2 is negative)
• Hence material x is better
than y

Solución 02-B
• The next step is to determine the true
difference between the two wear rates using
Eq. 02 for (1 – α) confidence

(x − y ) − Atα 2;v ; (µ x − µ y ); (x − y ) + Atα 2;v

11
 26/05/2011

Solution 02-B
• Hence, it can be
• For 90 percent confidence: concluded with 90
percent confidence
− 6,54 x10 −3  that the true
− 4,95 ± 0,925 x1,717 =  
− 3,36 x10 −3  difference in average
wear of materials x
and y can be as little as
3.36 x 10-3 in and as
much as 6.54 x 10-3 in.

(− 6,54 x10 ; µ −3
x − µ y ;−3,36 x10 −3 in)
(3,36 x10 < µ −3
y − µ x < 6,54 x10 −3 )in

Solution 02-B
• Similarly, for higher confidences:
– For 95% confidence

(3,03x10 −3
< µ y − µ x < 6,87 x10 −3 in )
– For 99% confidence

(2,34 x10 −3
< µ y − µ x < 7,56 x10 −3 in )

12