Philosophy and idea of the person

The philosophy and the idea of
the person-oriented analysis
Jari-Erik Nurmi
Department of Psychology
University of Jyväskylä
Background: a variable-oriented
research
• Since early 20th century (1905 or so), quantitave
behavioral sciences (psychology, sociology, etc.)
have used ”variable-oriented methods” in
analyzing data:
– Measurement of phenomena by operationalizing
them as (usually) continuous variables
– Examining research questions as statistical
associations between variables (correlations, beta
coefficients, exploratory factor analyses, etc.)
– More recently using more sophisticated analysis like
SEM and LGM
– Analysis have focused on inter-individual variance and
covariance between variables
Background: a variable-oriented
research
• More recently the analysis of inter-individual
variance have been complemented by
– Research on intra-individual variance
– And research on nested environments
But something has been missing: the
person-oriented approach
A holistic theory of personality by
David Magnusson
• Magnusson started by using a variable-oriented
psychometrics
• At certain point he begin to rethink the research
on personality and human development
•
”a holistic theory of personality”
– Personality does not consist of (orthogonal) traits but
rather a unique constellation/ combination of traits
– In order to study such unique constellations one
needs a new methodological framework a personoriented approach
Lars Bergman, David Magnusson and
Bassam El’Khouri: A person-oriented
analysis
• The aim is to identify unique combinations of
individual characteristics
• This can be done by using many statistical
methods: creating categorical variables, cluster
analysis, mixture modeling, etc.
• You can do this either by using
– Cross-sectional data or
– Longitudinal data or
– Even diary data
Lars Bergman, David Magnusson and
El’Khouri: A person-oriented analysis
• Key idea:
– Identify groups of individuals who are similar in the
constellation of variable values
– But who differ from other groups of individuals in
these constellations
• Benefits of using the approach:
– Identify typical groups with different pattern of
chracteristics
– Identify the percentage of people showing these
patterns
– Identify developmental trajectories of changing
patterns over time
Many ways to use the person-oriented
approach and many statistical
methods available
1. Cross-sectional research that can be
interesting
• Just when you are interested in different patterns/
constellations at certain time-point
X1
X2
X3
X4
X5
Cluster 1
Cluster 2
Cluster 3
2a. Clustering in longitudinal data
using different clustering criteria over
time
• How do patterns of some individual characteristics
change over time
• How do people change from one pattern to another
time1
time2
Cluster 1
Cluster 1
Cluster 2
Cluster 2
Cluster 3
Cluster 3
Cluster 4
Odd ratios
2b. Clustering longitudinal data using
identical criteria in different time
points: ISOA procedure
• The benefit that clusters are formed on the
basis of identical criteria
– The concept of stability (of cluster membership)
becomes clear
– The concept of change becomes clear
– The role of other predictors become clear
• Should be preceded by 2a type of analysis
The trick is simple but clever
t1
S1 y11 y21…
S2 y11 y21…
S3 y11 y21…
S4 y11 y21…
S5 y11 y21…
S1 y1 y2…
S2 y1 y2…
S3 y1 y2…
S4 y1 y2…
S5 y1 y2…
t2
S1 y12 y22…
S2 y12 y22…
S3 y12 y22…
S4 y12 y22…
S5 y12 y22…
S1 y11 y21CL2…
S2 y11 y21CL1…
S3 y11 y21CL2…
S4 y11 y21CL3…
S5 y11 y21CL2…
S1 y12 y22CL2…
S2 y12 y22CL1…
S3 y12 y22CL3…
S4 y12 y22CL3…
S5 y12 y22CL2…
Rearrange
data
Rearrange
Data again including
cluster membership
Conduct clustering
and save cluster membership
Now you can study …
• Stability: Frequency table and observed vs.
expected frequencies
• Predictors: Multinomial regression
Study example
JEPS-study
• Data 1lk-4lk (kevät)
• Measures
- Motivation
- Achievement beliefs: e.g. failure
expectations
- Task-avoidance
ISOA clustering by cases
2,5
2,0
1,5
1,0
task_avoidance
efficacy
motivation
failure_expectation
social_support
reading_motivation
math_motivation
0,5
0,0
-0,5
-1,0
-1,5
-2,0
-2,5
failure-expectation
low-motivation
autonomic
adaptive-social
100,0 %
65***
80,0 %
n = 80
(40.8 %)
30***
0 **
4 ***
n = 123
(62.8 %)
n = 48
(27.0 %)
3*
60,0 %
n = 64
(36.0 %)
n = 66
(33.7 %)
40,0 %
17*
n = 32
(16.3 %)
20,0 %
n = 35
(19.7 %)
13***
n = 24
(12.2 %)
8***
n = 19
(9.7 %)
n = 22
(11.2 %)
8*
9 **
0**
5*
10 **
7**
n = 26
(13.3 %)
n = 31
(17.4 %)
15***
0,0 %
1st
2nd
failure-expectation
low-motivation
4th
autonomic
adaptive-social
2c. Person-oriented analysis on
developmental changes over time
• Clustering is conducted by using variables
measured over several time-points
• The benefit of the approach is that the change
patterns over time becomes very clear
Development of Reading Skills
among Preschool and Primary
School Pupils
Ulla Leppänen, Pekka Niemi,
Kaisa Aunola & Jari-Erik Nurmi
Reading Research Quarterly.
197.00
70
198.00
60
199.00
200.00
50
201.00
40
202.00
30
203.00
204.00
20
205.00
10
206.00
0
Mean lukut1: lukutai
Mean lukut3: lukutai
Mean lukut2: lukutai
Mean lukut4: lukutai
207.00
Mean LUKUT5
.17
.09
Reading skill
Time 1
1
1
.09
Reading skill
Time 2
1
1
0
Reading skill
Time 3
1
1
.26**
0
-.71***
-.82***
Reading skill
Time 4
0 0
1
Trend 1
Level
.43
1
Trend 2
40
30
Reading Skill Scores
20
Cluster Groups
10
1 N=71
2 N=113
0
Time 1
3 N=11
Time 2
Time 3
Time 4
But … whatever procedure you use
there is always many statistical tools to
conduct the clustering:
– Clustering by cases (e.g., SPSS)
– Mixture modelling (M-plus, AMOS)