Behavioral Goal Setting Models
for Operations Management
Workshop Stochastic models for warehousing systems
October 29, 2009
José Antonio Larco,
Kees Jan Roodbergen, René de Koster, Jan Dul
Rotterdam School of Management
Erasmus University Rotterdam
Contact: [email protected]
Introduction
Propositions
Experiment
Results
Conclusions
Goals are interesting for OM…
• Current assumptions of OM models:
– People are predictable, work in a stationary way and are
unaffected by external factors (Boudreau, 2003)
• Challenging goals have a positive effect on performance
– Meta-analysis 8-16% performance increase over “do your
best”” strategies; (Locke and Latham, 1990)
– Well studied: >239 lab experiments, > 156 field studies (Locke
and Latham, 1990)
Introduction
Propositions
Experiment
Results
Conclusions
–
–
–
–
Linear? (Locke and Latham, 1968)
Levels-off? (Locke & Latham, 1982)
Decreases? (See et al, 2006)
Effects of varying skill level?
2. How do workers regulate
their work pace?
–
–
–
Acceleration towards goal
(Hull, 1932) or deadline?
A steady state pattern or
irregular?
Effect of varying goals & skill
level?
600
500
400
?
300
200
100
0
0
0.2
0.4
0.6
0.8
Goal difficulty (probability of success)
1
9
8
Workspeed (picks/min)
1. How is performance related
to goal difficulty?
Cummulative performance s(D)
Two main questions for OM
Goal achieved
7
?
6
5
4
3
2
Deadline
1
0
0
2
All this in OM contexts where workers have a fixed time to work.
4
Time (min)
6
8
Introduction
Propositions
Experiment
Results
Conclusions
Two-fold approach
1. Proposition generation: workers as decision makers
–
Objective: maximize utility/preference
•
Utility derived from work pace itself
•
Utility derived from evaluation w.r.t goal
–
Decision: what work-pace to select? (effort to exert)
–
Behavioral Economic decision models:
–
•
Myopic: Individuals focus only on the “near future”.
•
Planner: Individuals take into account the utility for the whole period.
Derivation of properties from model
2. Propositions Testing: experimental setting
–
Total performance
–
Work pace measurement
Introduction
Propositions
Experiment
Results
Conclusions
Scope
• Work is repetitive; i.e. work content known; cycle times short
• Workers are experienced
• Feedback is provided
.
• Goal G (units processed) to be achieved before deadline D
• Target G serves as reference for evaluating performance
• Workers committed to the goal (Locke and Latham, 1990)
• Cumulative work s(t), work pace, ṡ=ds/dt
Introduction
Propositions
Experiment
Results
Conclusions
Work pace preference (Yerkes-Dodson Law (1908)
• Relates (Hancock &
Warm,1989) :
Utility rate rhythm function R(ṡ)
3
– Stressor
• Defines:
– Maximum desirability
– Range of tolerance
• Properties:
– Convex function
Utility rate (utility units/minute)
– Adaptability/desirability
Natural speed
2
1
0
15
20
25
30
-1
-2
-3
-4
-5
-6
Work speed (picks/minute) ṡ=ds/dt
35
Introduction
Propositions
Experiment
Results
Conclusions
Goal induced preference (Kahneman & Tversky 1979)
• Properties:
Progress utility curve P(s)
– Strictly increasing
0.3
P' ( s (t ))  0; t  [0, D]
P(G   )  P(G   );   0;
– Diminishing sensitivity
P ' ' ( s (t ))  0; s  G
P ' ' ( s (t ))  0; s  G
Utility (utility units) P(s)
– Loss aversion
Target T
0.2
0.1
0
0
100
200
300
-0.1
-0.2
-0.3
-0.4
-0.5
-0.6
• Usage in goal theory
Heath et al.,1999; Steel & Koning, 2006
and Wu, et al. 2008
Total work (picks) s(t)
400
500
Introduction
Propositions
Experiment
Results
Conclusions
1. Myopic Conjecture
max s R( s)  Q( s, s)
Initial conditions :
s(0)  0
Progress utility rate (apply chain rule) - unit consistenc y :
Q( s, s)  P( s ) s
First order conditions :
R( s)   P( s )
Introduction
Propositions
Experiment
Results
Conclusions
1. Myopic Conjecture (Work Pace Propositions)
>Consistent with goal gradient hypothesis (Hull, 1932)
40
Work speed (picks/minute)
35
30
25
20
15
Target 100
Maximum speed achieved at target
10
Target 150
Target 200
5
Target 250
0
0
2
4
6
Time (minutes)
8
10
Introduction
Propositions
Experiment
Results
Conclusions
2. Planning Conjecture
D
max s  ( R ( s)  P( s ))dt
0
Boundary condition :
s(0)  0
D
Recognize that  P( s ) sdt  P ( s ( D))  P ( s (0))
0
Applying euler formula :
s (t )  ct
<Constant work pace!
using the fact that s(t )  constant :
R' ( s) D   P' ( Ds)
<Independent of work pace!
Introduction
Propositions
Experiment
Results
Conclusions
2. Planning Conjecture (Work Pace Propositions)
>Consistent with Planning Conjecture (Parkinson, 1955)
Work speed ds/dt (picks/minute)
28.0
27.5
27.0
26.5
Target 400
Target 350
Target 300
26.0
Target 200
Target 150
Natural pace
25.5
25.0
24.5
0
1
2
3
4
5
6
Time t (minutes)
7
8
9
10
Introduction
Propositions
Experiment
Results
Conclusions
Goals and Performance: Contrasting propositions
295
290
Performance (picks)
285
280
275
270
265
260
Myopic
255
Planner
250
245
0
50
100
150
200
250
Goal (picks)
300
350
400
450
Introduction
Propositions
Experiment
Results
Conclusions
Experiment Design
• Simple order picking task, short cycled (<10 sec)
• Control for learning effects (previous picking rounds)
• Within subject design (3x“4”):
– Pilot study “Do your best” control (n=36 subjects)
– 3 randomized goal levels (10, 50, 90th percentile) per subject
(n=81 subjects)
– Skill proxy: Average work pace of 10 best work pace used - 4
categories constructed
– Credit assigned to all subjects regardless of their performance
• Process view: work pace measured using time stamps
Introduction
Propositions
Experiment
Experiment : Task description
Results
Discussion
Introduction
Propositions
Experiment
Results
Conclusions
Results 1: Goal difficulty-performance
Total performance (number of picks)
> Level-off (Locke and Latham (1982)) (Goal/skill Interaction F (1,77)=7.4; p<0.01)
> Consistent with planner results
80
∆=14.3%
75
70
65
∆=11.9%
Skill Level 1
∆=12.6%
Skill Level 2
Skill Level 3
60
∆=7.1%
55
50
47
59
Goal Level (picks)
66
Skill Level 4
Introduction
Propositions
Experiment
Results
Discussion
Results 2: Work speed-stationary behavior
Work pace picks/min
10
2. Work pace
selection
9.5
1. Stable pattern
9
3. Acceleration
towards deadline
8.5
8
7.5
Goal 59 picks
7
Goal 66 picks
6.5
Goal 47 picks
6
0
1
2
3
4
5
Time (min)
6
7
8
Introduction
Propositions
Experiment
Results
Conclusions
Results 3: Multi-level analysis (pick<round<subject)
DV: Work pace: picks/min
Factor
Intercept
Time
Goal Level 47 Picks
Goal Level 59 Picks
Skill Level 2
Skill Level 3
Skill Level 4
Time x Goal Level 47 Picks
Time x Goal Level 59 Picks
Time x Skill Level 2
Time x Skill Level 3
Time x Skill Level 4
Time x Goal Level 47 Picks x Skill Level 2
Time x Goal Level 47 Picks x Skill Level 3
Time x Goal Level 47 Picks x Skill Level 4
Time x Goal Level 59 Picks x Skill Level 2
Time x Goal Level 59 Picks x Skill Level 3
Time x Goal Level 59 Picks x Skill Level 4
R2
AIC
BIC
MLM Model (picks=3631,
rounds=3, subjects=81)
Coeff
S.E.
7.7878 0.1458 ***
0.0089 0.0202
-0.6757 0.0894 ***
-0.2273 0.0894 *
0.6841 0.1952 **
1.0170 0.1952 ***
1.7094 0.1952 ***
-0.0826 0.0287 **
-0.0229 0.0242
-0.0074 0.0285
-0.0204 0.0285
-0.0132 0.0285
-0.0093 0.0402
0.0040 0.0402
0.0042 0.0402
0.0144 0.0335
-0.0202 0.0335
-0.0666 0.0335 *
OLS Model (n=3631)
Coeff
7.8331
-0.0148
-0.6933
-0.2347
0.6445
0.9799
1.6724
-0.0342
-0.0240
0.0047
-0.0011
0.0369
-0.0451
-0.0354
-0.1020
0.0409
-0.0151
-0.0871
S.E.
0.0678 ***
0.0186
0.0682 ***
0.0683 ***
0.0785 ***
0.0784 ***
0.0784 ***
0.0223
0.0222
0.0238
0.0238
0.0238
0.0246 .
0.0246
0.0246 ***
0.0246 .
0.0245
0.0245 ***
0.3965
6338
6512
6180
***p<0.001, **p<0.01, *p<0.05 Reference values: Goal Level 66 Picks, Skill Level 1
Introduction
Propositions
Experiment
Results
Conclusions
MLM Random effects
Factor
By Pick Round j within Subject I
(Intercept)
By Pick Round j within Subject I
Time
By Subject
(Intercept)
By Subject
Time
By Subject
Time x Goal Level 47 Picks
By Subject
Time x Goal Level 59 Picks
Residual
Variance of
coefficients
S.E. of
coefficients
0.265087
0.514866
0.003444
0.05869
0.282252
0.531274
0.002298
0.047939
0.005494
0.074119
0.000433
0.020806
0.242124
0.492061
Corr
-0.306
-0.195
0.619
Introduction
Propositions
Experiment
Results
Conclusions
Conclusions
•
Confirmation of “leveling off” effect in goal difficultyperformance relationship
•
Challenging goals induce steady state behavior
(explanations for this? -Carver & Shreier, 1998?)
•
Acceleration towards deadline not towards goal
•
General support for planner conjecture model
Introduction
Propositions
Experiment
Results
Conclusions
Implications for OM:
•
Confirmation that challenging goals work
•
Usage of different goals as source of variable
capacity: (demand fluctuations and deadlines)
•
New advantages of challenging goals: steady state
behavior and enhanced predictabilility
•
Verification of steady-state work pace to identify
whether goal is adequately set.
•
Monitor progress towards the goal
Introduction
Propositions
Experiment
Results
Conclusions
Further research…
•
Vary time frames
•
Study feedback effects
•
Prediction of performance
•
Trade-off with other OM goals: quality, fatigue,
safety
•
Replications in “real world” settings
Extra Results
Performance Distribution
100%
Cummulative frequency
90%
80%
70%
60%
Do your best
50%
Goal: 47 picks
40%
Goal: 59 picks
30%
Goal: 66 picks
20%
10%
0%
30
40
50
60
70
Performance: Total Picks in 8 minutes
80
90
Extra Results
Prospect theory for goals!
Satisfaction rating performance
100%
90%
80%
70%
60%
50%
Goal: 47 picks
40%
Goal: 59 picks
30%
Goal: 66 picks
20%
10%
0%
0
10
20
30
40
50
60
Perfromance: Picks in 8 minutes
70
80
90
Extra Results
In Depth: Skill Level 1
10
Work pace picks/min
9.5
9
8.5
Goal 66 picks
8
Goal 59 picks
7.5
Goal 47 picks
7
6.5
6
0
1
2
3
4
Time (min)
5
6
7
8
Extra Results
In Depth: Skill Level 2
10
Work pace picks/min
9.5
9
8.5
Goal 66 picks
8
Goal 59 picks
7.5
Goal 47 picks
7
6.5
6
0
1
2
3
4
Time (min)
5
6
7
8
Extra Results
In Depth: Skill Level 3
10
Work pace picks/min
9.5
9
8.5
Goal 66 picks
8
Goal 59 picks
7.5
Goal 47 picks
7
6.5
6
0
1
2
3
4
Time (min)
5
6
7
8
Extra Results
In Depth: Skill Level 4
10
9.5
Work pace picks/min
9
8.5
Goal 66 picks
8
Goal 59 picks
7.5
Goal 47 picks
7
6.5
6
0
1
2
3
4
5
Time (min)
6
7
8