1 causal data mining richard scheines dept. of philosophy, machine learning, & human-computer...

16
1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

Upload: jeffrey-peters

Post on 13-Jan-2016

218 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

1

Causal Data Mining

Richard Scheines

Dept. of Philosophy, Machine Learning, &

Human-Computer Interaction

Carnegie Mellon

Page 2: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

2

Causal Graphs

Causal Graph G = {V,E} Each edge X Y represents a direct causal claim:

X is a direct cause of Y relative to V

Exposure Rash

Exposure Infection Rash

Chicken Pox

Page 3: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

3

Causal Bayes Networks

P(S = 0) = .7P(S = 1) = .3

P(YF = 0 | S = 0) = .99 P(LC = 0 | S = 0) = .95P(YF = 1 | S = 0) = .01 P(LC = 1 | S = 0) = .05P(YF = 0 | S = 1) = .20 P(LC = 0 | S = 1) = .80P(YF = 1 | S = 1) = .80 P(LC = 1 | S = 1) = .20

Smoking [0,1]

Lung Cancer[0,1]

Yellow Fingers[0,1]

P(S,YF, LC) = P(S) P(YF | S) P(LC | S)

The Joint Distribution Factors

According to the Causal Graph,

i.e., for all X in V

P(V) = P(X|Immediate Causes of(X))

Page 4: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

4

Structural Equation Models

• Structural Equations: One Equation for each variable V in the graph:

V = f(parents(V), errorV)for SEM (linear regression) f is a linear function

• Statistical Constraints: Joint Distribution over the Error terms

Education

LongevityIncome

Causal Graph

Page 5: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

5

Structural Equation Models

Equations: Education = ed

Income =Educationincome

Longevity =EducationLongevity

Statistical Constraints: (ed, Income,Income ) ~N(0,2)

2diagonal - no variance is zero

Education

LongevityIncome

Causal Graph

Education

Income Longevity

1 2

LongevityIncome

SEM Graph

(path diagram)

Page 6: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

6

Tetrad 4: Demo

www.phil.cmu.edu/projects/tetrad

Page 7: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

7

Causal Datamining in Ed. Research

1. Collect Raw Data

2. Build Meaningful Variables

3. Constrain Model Space with Background Knowledge

4. Search for Models

5. Estimate and Test

6. Interpret

Page 8: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

8

CSR Online

Are Online students learning as much?

What features of online behavior matter?

Page 9: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

9

CSR Online

Are Online students learning as much?

Raw Data : Pitt 2001, 87 students

For everyone: Pre-test, Recitation attendance, final exam

For Online Students: logged: Voluntary question attempts, online quizzes, requests to print modules

Page 10: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

10

CSR Online

Build Meaningful Variables:

1. Online [0,1]

2. Pre-test [%]

3. Recitation Attendance [%]

4. Final Exam [%]

Page 11: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

11

CSR Online

Data: Correlation Matrix (corrs.dat, N=83)

Pre Online Rec Final

Pre 1.0

Online .023 1.0

Rec -.004 -.255 1.0

Final .287 .182 .297 1.0

Page 12: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

12

CSR Online

Background Knowledge:

Temporal Tiers:

1. Online, Pre

2. Rec

3. Final

Page 13: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

13

CSR Online

Model Search:

No latents (patterns – with PC or GES)

- no time order : 729 models

- temporal tiers: 96 models)

With Latents (PAGs – with FCI search)

- no time order : 4,096

- temporal tiers: 2,916

Page 14: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

14

Tetrad Demo

Online vs. Lecture

Data file: corrs.dat

Page 15: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

15

Estimate and Test: Results

• Model fit excellent

• Online students attended 10% fewer recitations

• Each recitation gives an increase of 2% on the final exam

• Online students did 1/2 a Stdev better than lecture students (p = .059)

Final Exam (%)

Recitation Attendance (%)

Pre-test (%)

Online

.22

5.3

.23

-10

Page 16: 1 Causal Data Mining Richard Scheines Dept. of Philosophy, Machine Learning, & Human-Computer Interaction Carnegie Mellon

16

References

• An Introduction to Causal Inference, (1997), R. Scheines, in Causality in Crisis?, V. McKim and S. Turner (eds.), Univ. of Notre Dame Press, pp. 185-200.

• Causation, Prediction, and Search, 2nd Edition, (2000), by P. Spirtes, C. Glymour, and R. Scheines ( MIT Press)

• Causality: Models, Reasoning, and Inference, (2000), Judea Pearl, Cambridge Univ. Press

• “Causal Inference,” (2004), Spirtes, P., Scheines, R.,Glymour, C., Richardson, T., and Meek, C. (2004), in Handbook of Quantitative Methodology in the Social Sciences, ed. David Kaplan, Sage Publications, 447-478

• Computation, Causation, & Discovery (1999), edited by C. Glymour and G. Cooper, MIT Press