The Effects of Peer Review and Reproducibility on Learning · The Effects of Peer Review and...

Post on 22-Mar-2020

7 views 0 download

Transcript of The Effects of Peer Review and Reproducibility on Learning · The Effects of Peer Review and...

The Effects of Peer Review and Reproducibility on Learning:

a Randomized Experiment

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

1

Introduction

● Problem: Stats/Math Education

(rote learning in highschools vs. non-rote learning in college)

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

2

Introduction

● Statistics Education (rote vs. non-rote learning)

● Constructivism (seems promising)

● Peer Review:

● different effects/roles: Reviewee vs. Reviewer (Van Gennip et al. 2009; Lundstrom et al. 2009; Strijbos et al. 2009)

● assumes Reproducibility

● requires technology (Wessa 2009)

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

3

Purpose of the Study

● Attempts to prove the causal effects of Peer Review (based on Reproducible Computing) on:

● Perceived utility of statistics

● Actual behavior (application of statistics)

● Non-rote learning (conceptual understanding)

● Attitude towards risk

● Setting:

● Fully randomized experiment

● Stock market environment/game

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

4

Experiment embedded in stats course

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

5

XSE Trading Screen

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

6

XSE is linked to Stats Software

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

7

The Challenge

● Implement a Market-Neutral Arbitrage Investment Strategy:

● Long Pile: contains stocks which are expected to rise

● Short Pile: contains stocks which are expected to drop

● Neutral Pile: contains stocks for which no prediction can be made. Note: in efficient markets, all stocks should go to the Neutral Pile

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

8

Hypothesis 1

● Utility hypothesis (practical relevance)

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

9

Hypothesis 2

● Behavior hypothesis

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

10

Hypothesis 3

● Non-rote learning hypothesis

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

11

Hypothesis 4

● Attitude hypothesis

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

12

Cohorts / Groups

● Groups:

● Control group: students receive instructor-based

feedback and need to correct mistakes from

previous assignment

● Treatment group: students engage in peer review

(based on solution provided by instructor)

● Cohorts:

● 2 full rounds of peer review

● 4 full rounds of peer review

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

13

2 Groups x 2 Cohorts for each Hypothesis

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

14

Phases of Experiment

● Phase A: preparation period which was needed

to ensure that the stock market’s statistical

properties are perfect to perform a MNAS

● Phase B: period during which the MNAS is

implemented (decisions are made at the

beginning of phase B)

● Phase C: aftermath

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

15

Market Index (phase A, B, C)

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

16

Odds Ratios

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

17

Binomial Effect Size Display

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

18

Conclusions

● Submitting Peer Review messages does cause:

● Students to use statistics more often

● Non-rote learning

● Changed attitudes towards risk

● No evidence that Peer Review changes

perceived practical relevance

(maybe more rounds of peer review are

needed)

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

19

Strengths versus Weaknesses

● Limitations:

● Only 2 cohorts (2 rounds and 4 rounds of PR)

● No evidence for Relevance Hypothesis

● Strengths:

● with the exception of the practical relevance hypothesis, all experimental observations are based on objective measurements which are generated by innovative, educational technology

● the experiment is embedded in a challenging game which has a history of many years and is known to be enjoyable and captivating

● the measured learning outcomes lie outside of the regular curriculum

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

20

Contact

• Ian E. Holliday i.e.holliday@aston.ac.uk

• P. Wessa patrick@wessa.net

• Website:

• http://www.freestatistics.org

CA

L 2

01

1, M

anch

este

r -

p

rese

nte

d b

y Ia

n E

. Ho

llid

ay,

A

sto

n U

niv

ersi

ty

21