Advanced experimental design and analysis tes

Advanced
Experimental Design
Topic 6
Chapters 11 & 12
Oneway Analysis of Variance
Multiple Comparisons
1

Agenda
Analysis of Variance (with
computational illustration)
| Comparison of F and t, connection
| Magnitude of effect (η2, ω2)
| Power
| Multiple Comparisons
|

2

Multiple groups (>2)
What if we have a situation where we

have more than two means that we
wish to compare with one another?
| How would we approach assessing
mean differences?
| Situations this might be an issue?
|

3

1

Type I error rates
could use multiple t-tests, but why might that
not be a good idea?
| inflating the probability of making a type I error
| Error rate per experiment
|

z
z


sum the p values for all of the tests
expected rate of type 1 errors = 6 x 0.05 = 0.30.

4

Three or more groups
Common in outcomes research,
where we may wish to evaluate the
relative efficacy of a number of
different treatments
| Few studies we might be interested in
doing have only two groups
| Oneway ANOVA allows comparisons
among two or more sample means
| Examples of multiple group studies
|

5


ANOVA Structural Model

X ij = µ + τ j + ε ij

τ j = (µ j − µ )

i=# of people, j=# of groups, Xij is the
ith person in the jth group
| µ = grand mean
| τ = adjustment reflecting mean of
jth group
| ε = “uniqueness”
|

6

2

ANOVA: ANalysis Of VAriance
|


Assumptions - with a large enough sample
can relax the assumptions
z
z
z

|

normal distribution of dependent variable
equal variances of groups
Independence of observations

Considered to be ROBUST
z

especially insensitive to violations of
normality when equal numbers in each
group and the distributions are about the
same shape

7

ANOVA conceptualization
Variance in a set of observations can be
partitioned into within groups and between
groups variance
| F statistic: basically the ratio of the between
groups variance to within groups variance
| If no differences between the groups means,
the between and within groups variance will be
about the same, resulting in an f ratio of
around one
|

8

ANOVA degrees of freedom
|
|
|

|

F statistic has two degrees of freedom:
between groups df which = ngroups-1
within groups df which = n - ngroups
F (2, 45) = 6.16, p < .01
ANOVA

CORRECT NUMBER OF STEERING CORRECTIONS

Between Groups
Within Groups
Total

Sum of
Squares
1016.667
3711.250
4727.917


df
2
45
47

Mean Square
508.333
82.472

F
6.164

Sig.
.004

9

3

Steps involved in hypothesis

testing w/ANOVA
1.
2.
3.
4.
5.
6.
7.

Identify H0 and H1
Tentatively assume H0
Choose a sampling distribution
Obtain data, calculate sample
statistic
Compare Fobs with Fcrit
Reach conclusion re: status of H0
Tell the story in terms of constructs
10

ANOVA example

hired by a large corporation to
implement a “wellness program.”
| encouraging good dietary habits and
exercise will decrease sick days,
productivity will increase and have
reduced health insurance benefit costs
due to lower utilization of medical
services
| best way to present the program, in
terms of scheduling?
|

11

ANOVA example (cont.)
freedom to schedule any time you wish
| attendance will be mandatory
| want to find out what people will prefer
| more satisfied with the program = receptive and
attentive to the information presented

| group of 30 employees randomly assigned to
each of the three conditions
| at the conclusion of the program (identical for all
3 groups) assess satisfaction with the program
|

12

4

The data
Condition Satisfaction Cond. Satis. Cond.
Satis.
days
5
nights
5 Saturday
5
days
5

nights
4 Saturday
5
days
5
nights
4 Saturday
5
days
5
nights
3 Saturday
4
days
4
nights
3 Saturday
4
days
4
nights
2 Saturday
4
days
4
nights
2 Saturday
3
days
4
nights
2 Saturday
3
days
3
nights
1 Saturday
3
days
2
nights
0 Saturday
2
13

Calculating F
1681 + 676 + 1444
10
2(df )

X days
5
5
5
5
1052
4

4
30 = 6.3
4
4
3
2
41
ΣX
2
1681
(ΣX)
Variance
0.99
Mean
4.1

Average variance
MSw

1.44

f=MSb/MSw=
fcrit(2,27) p