Quantitative Methods

1
Quantitative Methods

• Checking the models I independence

2
Checking the models I independence
Assumptions of GLM
3
Checking the models I independence
Assumptions of GLM
BACAFTER BACBEFTREATMNT
(Model Formula)
(Model)
(Fitted Value Equation or Best Fit Equation)
4
Checking the models I independence
Assumptions of GLM
BACAFTER BACBEFTREATMNT
(Model Formula)
(Model)
5
Checking the models I independence
Assumptions of GLM
(Model)
6
Checking the models I independence
Assumptions of GLM
(Model)
Assumptions of GLM Independence Homogeneity of
variance Normality of error Linearity/additivity
7
Checking the models I independence
Assumptions of GLM
(Model)
Assumptions of GLM Independence Homogeneity of
variance Normality of error Linearity/additivity
8
Checking the models I independence
Independence in principle
9
Checking the models I independence
Heterogeneous data
10
Checking the models I independence
Heterogeneous data
11
Checking the models I independence
Heterogeneous data
12
Checking the models I independence
Heterogeneous data
13
Checking the models I independence
Heterogeneous data
14
Checking the models I independence
Heterogeneous data
15
Checking the models I independence
Repeated measures
16
Checking the models I independence
Repeated measures
17
Checking the models I independence
Repeated measures
18
Checking the models I independence
Repeated measures
Single summary approach Multivariate approach Few
summaries approach
19
Checking the models I independence
Repeated measures
name C100 wtg let wtgLOGWT20-LOGWT3 glm
wtgdiet
GLM RATEDIET
LET K33-31/3 ! 31/3 is the
average of LET K88-31/3 ! 3,
8 and 20 LET K2020-31/3 LET K1K32K82K202
LET RATE(K3LOGWT3K8LOGWT8K20LOGWT20)/K1
20
Checking the models I independence
Repeated measures
21
Checking the models I independence
Repeated measures
GLM LOGWT60 RATE DIET MANOVA NOUNIVARIATE.
22
Checking the models I independence
Nested data
23
Checking the models I independence
Nested data
24
Checking the models I independence
Detecting non-independence
In principle would knowing the error for one or
more datapoints help you guess the error for some
other datapoint? Experiments Does the datapoint
correspond to the level of randomisation? Observat
ions Are there groups of datapoints which are
very likely to have similar residuals? Be
suspicious of - Too many datapoints -
Implausible results - Repeated measures
25
Checking the models I independence
Last words

• Independence is a key assumption, and is the most
  problematic in practice
• Always be alert to possible violations
• Know what can be done at the analysis stage
• Realise that mistakes at the design stage are
  often unrecoverable at analysis

Checking the models II the other three
assumptions Read Chapter 9