1 a gender and helping study with a different outcome
TRANSCRIPT
1
A gender and helping study with a different outcome
2
Here is another set of results from the experiment on helping.
3
4
A loglinear model was fitted to the data. Here is a test of its
goodness-of-fit
5
6
Question 1
• Does this chi-square value measure the goodness-of-fit of a saturated model?
7
Answer
• No. When a saturated model is applied, chi-square has no degrees of freedom and has a value of zero.
8
Shortly, I shall show you a table of tests of K-way and Higher Order
Effects
9
Question 2
• Examine the table. Is the opposite-sex dyadic hypothesis supported by these test results?
10
11
Answer
• No. The opposite-sex dyadic hypothesis predicts a three-way interaction of Participant’s Sex, Interviewer’s Sex and Help. The p-value for the three-way interaction (0.514) does not support this expectation.
12
Here is a table of the backward elimination statistics
13
14
Question 3. How many models are described
here?
15
Answer
• This table is difficult to follow. • FOUR models are described: 1. Interviewer*Participant*Help – the saturated
model.2. Int*Part, Int*Help, Part*Help. All two-way
interactions. 3. Int*Part, Int*Help. Part* Help dropped.4. Int*Help, Part. Int * Part dropped.• Opposite each model, there is a chi-square
value with so-many df.
16
Answer …
• Remember that this chi-square refers to the RESIDUALS associated with the terms that have been LEFT OUT.
• Opposite the final model Int*Help, Part, is the chi-square value 2.435, with df = 3. This chi-square measures the sizes of the residuals when the terms Int*Help*Part (df = 1), Help*Part (df = 1) and Part*Int (df =1) have been removed from the model. That’s why it has 3 degrees of freedom.
17
Question 4.
In the final model, where did Participant come from?
18
Answer
• The main effect of Participant has really been there all the time; but now it needs to be mentioned explicitly in the generating class, because all the interactions involving it have now been removed from the model.
19
The generating class
• In the output, we are told that the generating class is
• Interviewer*Help, Participant.
20
Question 5
• Does the final model include a term for the main effect of the Help factor?
21
Answer
• It must do, according to the hierarchical principle. If there is an interaction term, all lower-order effects among the same factors must also be included in the model.
• The presence of the Interviewer*Help term implies the presence in the model of the main effects of Interviewer and Help.
22
Question 6
• Can you write out an equation for the final loglinear model, expressing the terms verbally, rather than in algebraic symbols?
• The generating class of the final model is
Interviewer*Help, Participant
23
The final loglinear model
• There’s always a constant. • The model contains a main effect of Help. • There is an Interviewer × Help interaction.• By the hierarchical principle, there must also be main
effects of Interviewer and Help. • There’s a main effect of Participant.
tParticipan
of
effectmain
ninteractio
Help r Interviewe
Help
of
effect main
rInterviewe
of
effectmain
constant )ln(E