Hire 💕 8617370543 Kushinagar Call Girls Service Call Girls Agency
Brown bag 2012_fall
1. New Metrics & Measurement
for information search dynamics in decision making
for information search dynamics in decision making
Presenter: Xiaolei Zhou
Advisor: Dr. Joe Johnson
Miami University
JDM Lab
2. Participants make a decision among several options (Rows),
described by several attributes (Columns). For example, they
must predict which movie has the highest receipts based on
several binary features (above). The values of each cell in the
information table are occluded until an eye fixation occurs on the
cell
Stars Budget Rating Original
Movie A + - - +
Movie B - + + +
Movie C + - - -
3. Stars Budget Rating Original
Movie A +
Movie B
Movie C
Participants make a decision among several
options (Rows), described by several
attributes (Columns). For example, they must
predict which movie has the highest receipts
based on several binary features (above). The
values of each cell in the information table are
occluded until an eye fixation occurs on the
4. Stars Budget Rating Original
Movie A
Movie B -
Movie C
Participants make a decision among several
options (Rows), described by several
attributes (Columns). For example, they must
predict which movie has the highest receipts
based on several binary features (above). The
values of each cell in the information table are
occluded until an eye fixation occurs on the
5. Stars Budget Rating Original
Movie A
Movie B
Movie C +
Participants make a decision among several
options (Rows), described by several
attributes (Columns). For example, they must
predict which movie has the highest receipts
based on several binary features (above). The
values of each cell in the information table are
occluded until an eye fixation occurs on the
6. Stars Budget Rating Original
Movie A -
Movie B
Movie C
Participants make a decision among several
options (Rows), described by several
attributes (Columns). For example, they must
predict which movie has the highest receipts
based on several binary features (above). The
values of each cell in the information table are
occluded until an eye fixation occurs on the
7. Stars Budget Rating Original
Movie A
Movie B
Movie C -
Participants make a decision among several
options (Rows), described by several
attributes (Columns). For example, they must
predict which movie has the highest receipts
based on several binary features (above). The
values of each cell in the information table are
occluded until an eye fixation occurs on the
8. Stars Budget Rating Original
Movie A -
Movie B
Movie C
Participants make a decision among several
options (Rows), described by several
attributes (Columns). For example, they must
predict which movie has the highest receipts
based on several binary features (above). The
values of each cell in the information table are
occluded until an eye fixation occurs on the
9. Stars Budget Rating Original
Movie A
Movie B
Movie C -
Participants make a decision among several
options (Rows), described by several
attributes (Columns). For example, they must
predict which movie has the highest receipts
based on several binary features (above). The
values of each cell in the information table are
occluded until an eye fixation occurs on the
10. Stars Budget Rating Original
Movie A +
Movie B
Movie C
Participants make a decision among several
options (Rows), described by several
attributes (Columns). For example, they must
predict which movie has the highest receipts
based on several binary features (above). The
values of each cell in the information table are
occluded until an eye fixation occurs on the
11. Stars Budget Rating Original
Movie A
Movie B
Movie C -
Participants make a decision among several
options (Rows), described by several
attributes (Columns). For example, they must
predict which movie has the highest receipts
based on several binary features (above). The
values of each cell in the information table are
occluded until an eye fixation occurs on the
13. Strategy Used: Weighted Additive ( WADD)
Stars Budget Rating Original
Movie A + - - +
Movie B - + + +
Movie C + - - -
✔
14. Existing Measurements
Transition Matrices1,2
( Frequency of specific
transition types)3
.
Number of Acquisitions
Time per Acquisition
Proportion of Information acquired
Final Choice Made
1. Payne, J.W., Bettman, J.R., & Johnson, E.J. (1993). The adaptive decision maker. Cambridge University Press.
2. Böckenholt, U., & Hynan, L. S. (1994). Caveats on a process-tracing measure and a remedy. Journal of Behavioral Decision Making, 7, 103–117.
3. Ball, C. T. (1997). A comparison of single-step and multiple-step transition analyses of multiattribute decision strategies. Organizational Behavior and
Human Decision Processes, 69, 195-204.
A A A A E E E E I I I I I E E B B B B J J J C C A A E C J J J C C C C C C J J D D K K D D L L D L
A A B B B C C C C C D D D C C D D E E E F F F E E G G G G H H H H I I I J J I J K K L L K L H H
LEX:
WADD:
A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1
A0 2
B0 1
C0 1 2
D0 1 2
E0 1 1 1
F0
G0
H0
I0 1
J0 2 1
K0 1
L0 1
A1 B1 C1 D1 E1 F1 G1 H1 I1 J1 K1 L1
A0
1
B0
1
C0
2
D0
1 1
E0
1 1
F0
1
G0
1
H0
1
I0
2
J0
1 1
K0
2
L0
1 1
LEX:
WADD:
Stars Budget Rating Original
Movie A A
+ B
- C
- D
+
Movie B E
- F
+ G
+ H
+
Movie C I
+ J
- K
- L
-
15. What we missed ?
Experimental paradigms such as eye-tracking
collect high-resolution process data revealing the
information acquired en route to making decisions.
However, the metrics deployed in analyzing these
data have not kept pace, focusing instead on
summary statistics. Analysis of search dynamics has
been severely limited to crude measures such as
relative direction (row- vs. column-wise transitions,
or search “pattern” in the task below)1,2
, or at best
counting the frequency of very specific transition
types3
. We import techniques from other fields to
remedy this shortcoming.
1. Payne, J.W., Bettman, J.R., & Johnson, E.J. (1993). The adaptive decision maker. Cambridge University Press.
2. Böckenholt, U., & Hynan, L. S. (1994). Caveats on a process-tracing measure and a remedy. Journal of Behavioral Decision Making, 7, 103–117.
3. Ball, C. T. (1997). A comparison of single-step and multiple-step transition analyses of multiattribute decision strategies. Organizational Behavior and
Human Decision Processes, 69, 195-204.
17. Lag Sequential Analysis
Order:
Does search location at time t depend on location at
time (t-1), (t-2),...?
Stationarity:
Is the nature of the search process consistent over the
course of a trial?
Homogeneity:
Is the nature of the search process consistent across
trials or participants?
18. 18
Lag Sequential Analysis
Observed Frequency table
Expected Frequency tables (first order,
second order, ...)
G2
(LRX2
) - likelihood ratio statistic
19. 19
IPF (Iterative Proportional Fitting)
[Deming-Stephan algorithm]
[Deming-Stephan algorithm]
IPF is a computer algorithm used to calculate
expected frequency for each cell by using
margins of every order’s two-way contingency
table. (especially designed for calculate
expected frequency tables of sparse
matrices).
20. 20
Hierarchical log-linear Model
Example: lag 2
Three models:
1. Saturated Model: [012] Three-way associations
2. Reduced Model 1: [01][12][02] homogeneous associations
3.Reduced Model 2: [01][12]
21. 21
Hierarchical log-linear Model Testing
Adjusted Residuals
Lag=1: two models [01] vs. [0][1]
First, use IPF to compute the expected frequency for different models.
Second, test the significance of adjusted residuals between two models:
Lag=2: hierarchically test three models
[012] vs. [01][12][02], if no ordering effect, then [01][12][02] vs.[01][12]
First, use IPF to compute the expected frequency for three models.
second, test the significance of the adjusted residuals between models in
order:
22. 22
Results
“Decreasing”: The higher time
pressure is, subjects’ behaviors
become more random.
“green”- 1st order effect
“dark blue” - complete random effect
“Increasing”: The lower time pressure
is, subjects’ behaviors become more
organized.
“light blue” - 1st order effect
“red” - complete random effect
24. 24
What we can do more?
More than “lags” or homogeneity?
How do experimental conditions affect the
search process?
Does a proposed model describe actual
search behavior?
25. How to differentiate strategies
“Comparing”:
WADD
LEX
“Distance”:
Sting Edit Distance
26. 26
String Edit Distance
[Needleman-Wunch Algorithm]
[Needleman-Wunch Algorithm]
Dynamic programming methods5,6
are used
stepwise to determine whether to insert,
delete, or substitute codes at each position to
minimize cost.
Operations:
5. Day, R. F. (2010). Examining the validity of the Needleman–Wunsch algorithm in identifying decision strategy with eye-movement data. Decision
Support Systems, 49, 396–403.
6. Cristino, F., Mathô t, S., Theeuwes, J., and Gilchrist, I. D. (2010). ScanMatch: A novel method for comparing fixation sequences. Behavior Research
Methods, 42, 692-700.
27. 27
Simple example
Two sequences:
Substitution Matrix:
Gap Penalty (d): -5
Score calculation: (theoretical range [0,1])
S(A,C)+S(G,G)+S(A,A)+(3*d) + S(G,G)+S(T,A)+S(T,C)+S(A,G)+S(C,T)
= -3 + 7 + 10 + (3*-5) + 7 + -4 + 0 + -1 + 0 = 1
A G A C T A G T T A C
: | | | : : : :
C G A - - - G A C G T
A G C T
A 10 -1 -3 -4
G -1 7 -5 -3
C -3 -5 9 0
T -4 -3 0 8
28. 28
Implementation
(ScanMatch Toolbox/Matlab)
(ScanMatch Toolbox/Matlab)A A A A E E E E I I I I I E E B B B B J J J C C A A E C J J J C C C C C C J J D D K K D D L L D L
A A B B B C C C C C D D D C C D D E E E F F F E E G G G G H H H H I I I J J I J K K L L K L H H
LEX:
WADD:
A A A A E E E E B B B B J J J J C C C C K K K K D D D D L L L L D D D D
A A A B B B C C C D D D E E E F F F G G G H H H I I I J J J K K K L L L H H H
LEX_Theoretical :
WADD_Theoretical:
LEX vs. WADD
LEX_Theory vs.
WADD_theory
LEX vs.
LEX_Theory
WADD vs.
WADD_theory
LEX vs.
WADD_Theory
WADD vs.
LEX_theory
Stars Budget Rating Original
Movie A A
+ B
- C
- D
+
Movie B E
- F
+ G
+ H
+
Movie C I
+ J
- K
- L
-
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
First, let me introduce you the experimental task which we used for the current study.
To differentiate different strategies. Needs to insert foot notes.
First, We need to “tool” to describe what really happened during our data collection? Is the data really can reveal the ordering effect? ( We assuming there should has one, and test such “hypothesis”).
Things need to be determined by LSA .
reduced model 1: lag0 and lag2 are associated with each other and lag1 but do not interact with lag1, (homogenous associations). reduced model 2: lag0 and lag2 are independent conditional on lag1.
lag=1 : [01] is the full model, [0][1] is the reduced model.
Redo the graph in SAS if have time
So, now we had a metrics that caught the human searching info with higher order precision, then, next we can use the following algorithm to analyze their eye-movement data.
So, now we had a metrics that caught the human searching info with higher order precision, then, next we can use the following algorithm to analyze their eye-movement data.
First, We need to “tool” to describe what really happened during our data collection? Is the data really can reveal the ordering effect? ( We assuming there should has one, and test such “hypothesis”).