instructor: mr. le quoc members : pham thi huong tran thi loan nguyen ngoc linh hoang thanh hai...
TRANSCRIPT
Instructor: Mr. Le Quoc
Members : Pham Thi Huong
Tran Thi Loan
Nguyen Ngoc Linh
Hoang Thanh Hai
ANOVA IN DEPTH – DESIGNS AND EXAMPLES
Table of contents
• Full factorial design
♪ One way ANOVA
♪ Multifactor ANOVA
♪ Repeated measures design
♪ Mixed design
• Non-full factorial (nested) design
♪ Latin square design
♪ Split-plot design
Rationale of ANOVA
• Purpose: test for significant differences between means.
• Why the name analysis of variance? In order to test for statistical significance between means, we
are actually comparing (i.e., analyzing) variances.
• The Partitioning of Sums of Squares Variances can be divided, that is, partitioned. Ex: SS total = SS between-group + SS within-group
• Dependent and independent variables.– Dependent variables: measured = response– Independent variables: controlled = factor
Full factorial designOne way ANOVA
Test for the effect of one factor (independent variable) on the response variable (dependent variable)
Example: Test the effect of 3 books on the participants’ probability to have a lover. 3 levels: Book A (Aggressive Approach), Book M (Moderate Approach) and Book P (Passive Approach). Prepare a sample of participants and assign randomly to each a book. After one month, all participants are tested for results in interactive tests.
Book A Book M Book P
( mark of Love Test of each person)
Full factorial designMultifactor ANOVA
♣ Definition: Test for the effect of multiple factors on the dependent variable simultaneously
♣ Example: Test the effect of both the books (A, M and P) and gender
(Male, Female and Homosexual) on probability to have a lover. Prepare a sample and randomly assign one book to each participant. After one month, all participants are tested for results in interactive tests.
Book A Book M Book P
Gender 1Male
( mean mark of Love Test )
Gender 2 Female
Gender 3Homosexual
Full factorial designMultifactor ANOVA
♣ Advantages: + More realistic
+ More efficient than multiple t-tests
+ Enhance the power ( sensitivity of the test)
♣ Disadvantages - Difficult to be completely randomized
Interaction Effect• Example: Male who read book A and women who read book P tend to have
high score. Conversely, Male reading book P and Female reading book A have the lowest score. Out of people reading book M, homosexual have highest score.
• Note: Interaction effects often override main effects.
0
10
20
30
40
50
60
70
80
90
100
Male Homosexual Female
Mean Score
Book A
Book M
Book P
Full factorial designRepeated measures design
• Definition: An experimental design in which the measurements are taken at two or more points in time on the same set of experimental units.
• Example: Test the effect of 3 books on the participants’ probability
to have a lover measured by the score of a test. 3 levels: Book A, Book M and Book P. Prepare a sample of participants and assign randomly to each a book. After one month, all participants are tested in interactive tests. Then they all must complete a multiple-choice test.
Full factorial designRepeated measures design
• Advantages– Repeated measures required in some research
hypothesis (ex: longitudinal research)– Reduce the error– Economical
• Disadvantages:– Carryover – Progressive effect
Solution: Counter balance
Full factorial designMixed design
• Definition: Combination of both multi-factor and repeated measures design.
• Example: Test the effect of both the books (A, M and P) and gender (Male, Female and Homosexual) on probability to have a lover. Prepare a sample and randomly assign one book to each participants. After one month, all participants are tested in interactive tests. Then they all must complete a multiple-choice test.
• Ad/disadvantage: Same as multi-factor and repeated measures.
Non-full factorial designLatin Square design
• Definition: A Latin Square extends the Randomized Complete Block design to the case in which there are two blocking factors and one treatment. It is used to comparing t treatments in t rows and t columns, where rows and columns represent two blocking factors. The allocation of experimental treatments is such that each treatment occurs exactly once in each row and column.
• Example: Treatment factor: Book (3 levels: A, M, P )
Blocking factor 1: Apperance (3 levels: Beautiful (Handsome), Normal,Ugly)
Blocking factor 2: Gender (3 levels: Male, Female, Homosexual)
Non-full factorial designLatin Square designApperance 1
Beautiful (Handsome)Apperance 2
NormalApperance 3
Ugly
Gender 1Male
Book M Book A Book P
Gender 2 Female
Book A Book P Book M
Gender 3Homosexual
Book P Book M Book A
Non-full factorial designLatin Square design
• When to use : The Latin square design applied when there are repeated
exposures/treatments and two other factors => It’s useful where the experimenter desires to control variation in two different directions
• Advantage: This design avoids the excessive numbers required for full three way
ANOVA => economical
• Disadvantage: The number of levels of blocking factors and treatment factor must
be equal Not reflect the interaction
Non-full factorial designSplit-plot design
• Definition: Some factors of interest may be hard-to-change while the remaining factors are easy-to-vary. the running order of the treatment combination is determined by these “hard-to-change factors”
• Example: Restrict randomization by determining the hard-to-change factor: Location. Choose randomly one of three level of factor “Location”. Within that level, randomly select a participant of whichever gender and randomly assign him/ her a book. Then randomly select another level of “location "and so on. After one month, all participants are tested for results in interactive tests.
Location
Urban
Rural
Mountainous
Male
Female
Homosexual
Book M
Book A
Book P
Non-full factorial designSplit-plot design
Non-full factorial designSplit-plot design
• Advantages: + Increasing precision in estimating certain effects
+ Saving time, money and easy to follow the results
• Disadvantages:
- Sacrificing precision in other effects
THANK YOU