13
9. Single Factor Experiment 9.6 Multiple comparison methods 1 ANOVA: means are significantly different. Which means differ from the rest? Contrasts Comparison of pairs: o Tukey-Kramer o Least significant difference (LSD) o Dunnet Comparison with control:

9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

  • Upload
    others

  • View
    38

  • Download
    0

Embed Size (px)

Citation preview

Page 1: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6 Multiple comparison methods

1

ANOVA: means are significantly different.

Which means differ from the rest?

• Contrasts

• Comparison of pairs:

o Tukey-Kramer

o Least significant difference (LSD)

o Dunnet

• Comparison with control:

Page 2: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.1 Multiple comparison methods: Contrast

2

We may interested to compare a combination of treatments

(more than two).

A………………………………………………………

In contrast method we may hypothesize any combination of

treatment means.

……………………………………………………………………

……………………………………………………………………

……………………………………………………………………

Page 3: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.1 Multiple comparison methods: Contrast

3

For instance, a two-tailed hypothesis may be stated as:

𝐻0:

𝑖

𝑐𝑖 𝜇𝑖 = 𝜇∗

𝐻1:

𝑖

𝑐𝑖 𝜇𝑖 ≠ 𝜇∗(9.49)

The test statistic is defined as:

……………………………………………………………………

……………………………………………………………………

Page 4: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.2 Simultaneous confidence intervals

4

So far we have applied the concept of confidence interval to

only one particular population statement.

In some cases, however, we wish to determine a confidence

interval for a set of “r” confidence interval all with the equal

confidence level.

1) 𝜇1 , 𝜇2 𝐶. 𝐿. 1 − 𝛼

2) 𝜇1 , 𝜇3 𝐶. 𝐿. 1 − 𝛼

3) 𝜇2 , 𝜇3 𝐶. 𝐿. 1 − 𝛼

1 &(2)& 3 𝐶. 𝐿. ?

Exp.- an experiment with three treatment levels:

Page 5: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.2 Simultaneous confidence intervals

5

For simplicity suppose r=2:

Then the probability that the events are not correct are:

𝑃 𝐴1 = (1 − 𝛼)(9.I)

𝑃 𝐴2 = (1 − 𝛼)

𝑃 𝐴1′ = 𝛼

(9.II)𝑃 𝐴2

′ = 𝛼

Then the probability that either or both are incorrect:

……………………………………………………………………

Page 6: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.2 Simultaneous confidence intervals (cont’d)

6

From the probability of the complementary events we know:

Combining (9.IV) with (9.III):

The last term on the right is …………………………………:

𝑃 𝐴1′ ∪ 𝐴2

′ + 𝑃 𝐴1 ∩ 𝐴2 = 1 (9.IV)

𝑃 𝐴1 ∩ 𝐴2 = 1 − 𝑃 𝐴1′ − 𝑃 𝐴2

′ + 𝑃 𝐴1′ ∩ 𝐴2

′ (9.V)

……………………………………………………………………

Page 7: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.2 Simultaneous confidence intervals (cont’d)

7

In general for a set of “r” comparison:

This is called ……………………………………….

Then to ensure that the simultaneous confidence interval is not

too small:

𝑟 𝑠𝑒𝑡 𝑜𝑓 𝑐𝑜𝑚𝑝𝑎𝑟𝑖𝑠𝑜𝑛: 𝐶. 𝐿. 1 − 𝛼𝑜𝑣𝑒𝑟𝑎𝑙𝑙 𝑡ℎ𝑒𝑛 𝛼𝑜𝑣𝑒𝑟𝑎𝑙𝑙 = 𝑟𝛼𝑖𝑛𝑑𝑖𝑣𝑖𝑑𝑢𝑎𝑙 (9.52)

……………………………………………………………………

𝛼𝑖𝑛𝑑𝑖𝑣𝑖𝑠𝑢𝑎𝑙 =𝛼𝑜𝑣𝑒𝑟𝑎𝑙𝑙

𝑟

Page 8: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.3 Simultaneous confidence intervals: Tukey

8

“a” treatment means are compared simultaneously.

Two-tailed hypothesis:

𝐻0: 𝜇𝑚 = 𝜇𝑙

𝐻1: 𝜇𝑚 ≠ 𝜇𝑙

(9.53)

The test statistic for this method is:

Note: There is no need to apply Benferroni rule in this method.

……………………………………………………………………

Page 9: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.3 Simultaneous confidence intervals: Tukey

9

The test statistic is compared with a critical value that is

obtained from studentized range distribution .

The Corresponding confidence interval is:

For unequal sample sizes the method is called Tukey-Kramer.

𝑦𝑚. − 𝑦𝑚. − 𝐸 ≤ 𝜇𝑚 − 𝜇𝑙 ≤ 𝑦𝑚. − 𝑦𝑚. + 𝐸

𝐸 = 𝑞𝛼(𝑎, 𝑁 − 𝑎) 𝜎𝜀2

2

1

𝑛𝑚+

1

𝑛𝑙

(9.56)

……………………………………………………………………

Page 10: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.4 Simultaneous confidence intervals: LSD

10

The LSD method is based on a modification of the hypothesis

for two population.

The test statistic is defined:

𝑅𝑒𝑗𝑒𝑐𝑡 𝐻0 𝑖𝑓 𝑦𝑚. − 𝑦𝑙. > 𝐿𝑆𝐷 (9.58)

Note: Apply Benferroni rule in this method.

……………………………………………………………………

Page 11: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.5 Simultaneous confidence intervals: Dunnet

11

In some situations, we consider one treatment as control and

compare the rest with the control.

𝐻0: 𝜇𝑖 = 𝜇𝑐 𝑓𝑜𝑟 𝑖 = 1,2, … , 𝑎 − 1

𝐻1: 𝜇𝑖 ≠ 𝜇𝑐

(9.59)

The test statistic is defined as:

𝑅𝑒𝑗𝑒𝑐𝑡 𝐻0 𝑖𝑓 𝑦𝑖. − 𝑦𝑐. > 𝐷𝑡𝑒𝑠𝑡,𝑖,𝑐 (9.61)

𝑛𝑐

𝑛≥ 𝑎 (9.62)

……………………………………………………………………

Page 12: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.7 Choice of sample size

12

Operating Characteristic curve may be used

𝛽 = 𝑃 𝐹𝑎𝑖𝑙 𝑡𝑜 𝑟𝑒𝑗𝑒𝑐𝑡 𝐻0 𝐻0 𝑖𝑠 𝑓𝑎𝑙𝑠𝑒 = 1 − 𝑃 𝐹𝑡𝑒𝑠𝑡 > 𝐹𝑐 𝐻0 𝑖𝑠 𝑓𝑎𝑙𝑠𝑒) (5.4)

Φ2 =𝑛 𝐷2

2𝑎 𝜎𝜀2

(9.63)

The method is based on an iteration approach

Page 13: 9. Single Factor Experiment - gheshlaghi.profcms.um.ac.irgheshlaghi.profcms.um.ac.ir › ... › doe-sec9-sfe-4-st.pdf · 9. Single Factor Experiment 9.6.3 Simultaneous confidence

9. Single Factor Experiment

9.6.7 Choice of sample size (cont’d)

13