Transcript Data Analysis

Working memory research
Group member : 唐牧辰 武天翊 侯晓林 赵诣 赵阳
BACKGROUND
Alan Baddeley and Graham Hitch proposed a
Model of Working Memory in 1974, in an
attempt to describe a more accurate model
of short-term memory.
BACKGROUND
•Central Executive
The central executive is a flexible system responsible for the control and
regulation of cognitive processes. It has the following functions:
• binding information from a number of sources into coherent episodes
• coordination of the slave systems
• shifting between tasks or retrieval strategies
Central
Executive
• selective attention and inhibition
Phonologic
al Loop
Visualspatial
Sketchpad
Episodic
Buffer
BACKGROUND
•Phonological loop
The phonological loop (or "articulatory loop") as a whole deals with sound or
phonological information. It consists of two parts:
• a short-term phonological store with auditory memory traces that are
subject to rapid decay.
• an articulatory rehearsal component that can revive the memory traces.
Central
Executive
Phonological
Loop
Visualspatial
Sketchpad
Episodic
Buffer
BACKGROUND
•Visual-spatial Sketchpad
The visual spatial sketchpad is assumed to hold information about
what we see. It is used in the temporary storage and manipulation of
spatial and visual information, such as remembering shapes and
colors, or the location or speed of objects in space.
Central
Executive
Phonologic
al Loop
Visualspatial
Sketchpad
Episodic
Buffer
Factors may Influence Memory
Personal Factors
•Gender
• Occupations
• Age
• Education
Held Factors
Central Executive/Episodic Buffer
• Working memory capacity
• Inner logical relationship among materials
1st Experiment
Phonological loop/Visual-spatial Sketchpad
• Information reception channel ( visual / phonological )
2nd Experiment
Testing circumstance control
Temperature controlling
25~30 ℃
Loudness controlling
40~70
Space controlling
Indoor
No Review !
1st Experiment
Response variables:
Memory level (quantity of right numbers remembered)
Testing factors:
Scale: size of materials (series of integer numbers)
Logicality: logicality of materials
Time: time span on memorizing materials
Testing materials:
Number sequence (has covert inner logicality)
Variables determination
Logicality
low: without logicality (logicality unknown), high: without
logicality (logicality known)
Memory scale
Low: 3*6 (divide 18 numbers into three groups with 6 numbers
each)
high: 2*9 (divide 18 numbers into two groups with 9 numbers
each)
Experiment time
low: morning, med: afternoon, high: night
Data collection
Sample collecting time:
morning (9:00~11:00), afternoon(14:00~16:00), night(19：00~21：00)
Exactly two hours after remembering.
Testing methodology
Divide the number sequence factitiously
Using Access to collect the data
Experiment design
• Two and Three level mixed experimental design
• Number of experiment subjects chosen?
Four
•How to eliminate the noise caused by variability of experiment
subjects?
Focusing on college students, apply all the runs for every subject.
•How to quantize memory effects?
correct answer in correct position
• How to eliminate the noise caused by the testing order?
Choosing randomized test order
Experiment design
Design matrix
run
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
scale
-1
1
-1
1
-1
1
-1
1
-1
1
-1
1
-1
1
-1
1
logicality
-1
-1
-1
-1
-1
-1
-1
-1
1
1
1
1
1
1
1
1
XL
-1
-1
-1
-1
1
1
1
1
-1
-1
-1
-1
1
1
1
1
XL
-1
-1
-1
-1
-1
-1
-1
-1
1
1
1
1
1
1
1
1
X stands for the factor “Time”
XQ
1
1
1
1
-1
-1
-1
-1
-1
-1
-1
-1
1
1
1
1
X
LOW
LOW
MED
MED
MED
MED
HIGH
HIGH
LOW
LOW
MED
MED
MED
MED
HIGH
HIGH
Experiment design
Design matrix (cont’d)
• Scale: -1 stands for low scale (divide 18 numbers into 3 groups), 1 stands
for high scale (divide 18 numbers into 2 groups)
• Time:-1 stands for morning, 0 stands for afternoon, 1 stands for night
• Logical: -1 means the quizzee doesn’t know the inner logicality of the 18
serials number, 1 means the quizzee get the inner logicality of the
numbers.
DOE
13
Data analysis
• Main effects and the interaction effects
Re sp ons e 主效应图
数据平均值
scale
logical
16
15
14
平均值
13
12
-1
1
-1
R es po ns e 交互作用图
1
拟合平均值
Time
-1
16
15
16
14
14
13
1
Time
-1
0
1
Time
12
12
-1
0
16
1
14
logical
logical
-1
1
12
scale
-1
1
16
14
scale
12
-1
DOE
0
1
-1
1
14
Data analysis
• ANOVA
The factor “Time” is not
significant, so we reduce
the design into a two
factors full factorial design:
DOE
15
Data analysis
• Estimation of the coefficients and ANOVA:
Both the two factors and
the interaction of them
are significant.
R-Sq = 75.15% means
the model explain
75.15% of the variation,
because it is an
experiment about
human factor, high
enough.
DOE
16
Data analysis
• Residual
R e s p o n s e 残差图
正态概率图
与拟合值
99.9
99
2
残差
百分比
90
50
0
-2
10
1
-4
0.1
-5.0
-2.5
0.0
残差
2.5
5.0
12.0
直方图
13.2
14.4
拟 合值
15.6
16.8
与顺序
16
2
残差
频率
12
8
-2
4
0
0
-4
-3
-2
-1
0
残差
1
2
3
1 5
10 15 20 25 30 35 40 45 50 55 60
观 测 值顺 序
• We cannot obtain any unique pattern on the residual versus
fitted value and the residual versus
observation order.
DOE
17
Data analysis
• Residual
残差2 的概率图
正态 - 95% 置信区间
99.9
均值
0
标准差 1.246
N
64
AD
0.731
P 值
0.054
99
百分比
95
90
80
70
60
50
40
30
20
10
5
1
0.1
-5.0
-2.5
0.0
残差2
2.5
DOE
5.0
18
Data analysis
• Residual versus factors
残差与 sc al e
（响应为 Response）
3
2
The range and the variation of the residuals when the logical
factor is at different levels have distinct difference, this can be
explained that not all of the persons being tested cannot make
good use of the inner logic of the number sequence when he
is trying to remember them.
0
残差与 l og ic al
-1
（响应为 Response）
-2
3
-3
2
1
-4
-1.0
-0.5
0.0
scale
0.5
残差
残差
1
1.0
0
-1
-2
-3
-4
-1.0
DOE
-0.5
0.0
logical
0.5
1.0
19
Conclusion:
• For short time (exactly two hours later) number sequence
memory, the memory index is independent of the “Time” factor.
• Short time number sequence memory index has great
relationship with the scale factor.
• Short time number sequence memory index has significant
relationship with the logicality factor.
• Short time number sequence memory index has significant
relationship with the integration factor of scale and logicality.
DOE
20
2st Experiment
Experiment
Design
Experiment Material
• A sequence of two simple shapes
Consider the capacity of working memory is approximately
between 6 and 10 for most people. And in order to amplify the
experiment result , we choose upper-bound.
Reception Channel
• Demonstrate the image on the PPT slides
• Audio description of the same sequence
• Combine the visual & audio ways
Data Recording
• Excel files written by Visual Basic micros
inside slides
Experiment design
• 3 specimen
• 3 treatments (shape/sound/shape & sound)
• 3 replications
• To eliminate the block factor of test order as possible, we
apply the test as the following sequence
Subject1
Subject2
Subject 3
Replication 1
1-2-3
3-1-2
2-3-1
Replication 2
2-1-3
1-3-2
3-2-1
Replication 3
3-1-2
2-3-1
1-2-3
Data Analysis
• The box plot: the mean responses of the three treatments
are different from each other at a significant level.
R e s p o n s e 的箱线图
10
9
8
Response
7
6
5
4
3
2
1
shape
shape&sound
Acceptance_Format
DOE
sound
23
Data Analysis
• One-way ANOVA:
• P value is smaller than the significant 0.05, which shows that
the factor is significant. And the comparison of the mean under
different treatments (Tukey )
DOE
24
Data Analysis
• One-way ANOVA:
•This indicates that our model is weak in explaining the data,
only 23.58% of the total variation.
DOE
However S = 2.768 R-Sq = 23.58% R-Sq（调整） = 18.95%
25
Data Analysis
存储残差1 的概率图
正态 - 95% 置信区间
99
均值
标准差
N
AD
P 值
95
90
0
2.687
36
1.012
0.010
80
百分比
70
60
50
40
30
20
10
5
1
-10
-5
0
存储残差1
5
10
• P value is too small and the residual variance is normally
distributed.
DOE
26
Data Analysis
残差与 Ac ce pt an ce F or ma t
（响应为 Response）
5.0
2.5
残差
0.0
-2.5
The essential reason is that the individual difference and
individual-5.0
preference!
-7.5
0.0
0.5
1.0
Acceptance Format
1.5
2.0
•Thus we should do ANOVA for every single subject
• The residual variance displays different distribution in
Acceptance Format (0: shape 1:sound 2:shape&sound)
DOE
27
Data Analysis
• Target 1: ANOVA
• The response is significant thus the image memory is the fittest
for the first participant
DOE
28
Data Analysis
• Target 2: ANOVA
• The response is not significant thus we can conclude no optimal
memory approach for the second participant
DOE
29
Data Analysis
• Target 3: ANOVA
• The response is significant thus the optimal memory approach for the
third participant is shape and shape & sound. R-Sq value is a bit small
this indicates that the interpretability is not strong. However,
considering the ergonomics, the result is within the acceptance.
DOE
30
Data Analysis
The residual plot of target 3
存储残差3 的概率图
正态 - 95% 置信区间
99
均值
4.934325E-16
标准差
1.803
N
9
AD
0.414
P 值
0.261
95
90
80
百分比
70
60
50
40
30
20
10
5
1
-5.0
-2.5
0.0
存储残差3
2.5
5.0
7.5
More replications is needed to
eliminate the potential
noise!
DOE
31
Conclusion:
• The memory effects varies a lot highly depending on the
personal factors.
• More replications is needed to precisely estimate the personal
preference on working memory.
Thanks
Q&A