a blind analysis you are not allowed to peek! prof. donald koetke senior research professor of...

48
A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

Post on 22-Dec-2015

217 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

A Blind AnalysisA Blind Analysis

You are not allowed to peek!You are not allowed to peek!

Prof. Donald Koetke

Senior Research Professor of Physics

Valparaiso University

Page 2: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 2

OutlineOutlineOutlineOutline

1. What is a “blind analysis”?

• What is a “bias”?

• Does “bias” = “systematic error”?

• What is the special (subtle) bias?

2. What is the history of the blind analysis”?

3. What are examples of “blind analysis” in physics, astronomy, astrophysics, and, … ?

4. Some thoughts and reflections --

Page 3: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 3

OutlineOutlineOutlineOutline

1. What is a “blind analysis”?

• What is a “bias”?

• Does “bias” = “systematic error”?

• What is the special (subtle) bias?

2. What is the history of the blind analysis”?

3. What are examples of “blind analysis” in physics, astronomy, astrophysics, and, … ?

4. Some thoughts and reflections --

Page 4: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 4

OutlineOutlineOutlineOutline

1. What is a “blind analysis”?

• What is a “bias”?

• Does “bias” = “systematic error”?

• What is the special (subtle) bias?

2. What is the history of the blind analysis”?

3. What are examples of “blind analysis” in physics, astronomy, astrophysics, and, … ?

4. Some thoughts and reflections --

Page 5: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 5

OutlineOutlineOutlineOutline

1. What is a “blind analysis”?

• What is a “bias”?

• Does “bias” = “systematic error”?

• What is the special (subtle) bias?

2. What is the history of the blind analysis”?

3. What are examples of “blind analysis” in physics, astronomy, astrophysics, and, … ?

4. Some thoughts and reflections --

Page 6: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 6

OutlineOutlineOutlineOutline

1. What is a “blind analysis”?

• What is a “bias”?

• Does “bias” = “systematic error”?

• What is the special (subtle) bias?

2. What is the history of the blind analysis”?

3. What are examples of “blind analysis” in physics, astronomy, astrophysics, and, … ?

4. Some thoughts and reflections --

Page 7: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 7

OutlineOutlineOutlineOutline

1. What is a “blind analysis”?

• What is a “bias”?

• Does “bias” = “systematic error”?

• What is the special (subtle) bias?

2. What is the history of the blind analysis”?

3. What are examples of “blind analysis” in physics, astronomy, astrophysics, and, … ?

4. Some thoughts and reflections --

Page 8: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 8

A Blind AnalysisA Blind AnalysisA Blind AnalysisA Blind Analysis

What is a “bias”?

Does “bias” = “systematic error”?

What is the special (subtle) bias?

“A prejudice in favor of or against” -- may bedue to computer codes, equipment performanceor setting, decisions/selections/cuts imposed, etc.

Yes - if it causes a systematic shift in a result(This is not a “mistake” or “blunder”.)

Experimenter bias in making decisions/choices to achieve a desired answer. Examples …

Page 9: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 9

A Blind AnalysisA Blind AnalysisA Blind AnalysisA Blind Analysis

1. You know what the answer “should be” -i.e., you know what the “accepted” answer is -- but your answer is different -outside of errors.What do you do?

2. You make a measurement and within errorsit agrees with the “accepted” answer -- What do you do?

You do an experiment and…

Asymmetry!!

Page 10: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 10

A Blind AnalysisA Blind AnalysisA Blind AnalysisA Blind Analysis

The goal of a “blind analysis” is to preventthe experimenter from (unconsciously - or consciously) making decisions in the analysis that would affect the result based on:

• The predictions of a model or theory (e.g., the Standard Model for particle physics)

• Previous measurements known to the experimenter

• The experimenter’s intuition or other predisposition

These are all examples of personal bias

Page 11: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 11

A Blind AnalysisA Blind AnalysisA Blind AnalysisA Blind Analysis

1. What is a “blind analysis”?

A “blind analysis” is an analysis of measured data in which the final answer is kept hidden from the experimenters until all of the decisions about the analysis have been made:• Computer codes have been developed and tested

• Decisions about the number of events (trials) needed have been made (cuts have been selected)

• Apriori agreements are reached about what to do when the real answers from the experiment are revealed; no further analysis!

Page 12: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 12

OutlineOutline

1. What is a “blind analysis”?

• What is a “bias”?

• Does “bias” = “systematic error”?

• What is the special (subtle) bias?

2. What is the history of the blind analysis”?

3. What are examples of “blind analysis” in physics, astronomy, astrophysics, and, … ?

4. Some thoughts and reflections --

Page 13: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 13

HistoryHistoryHistoryHistory

Blind analysis begins in ~1930s with -

Medical research -- blind tests!

Patients don’t know whether they are getting --

a) the medicine/treatment, or,

b) a placebo

Patients are assigned to (a) or (b) ramdomly

Therefore - the patients are “Therefore - the patients are “blindblind” to their treatment” to their treatmentTherefore - the patients are “Therefore - the patients are “blindblind” to their treatment” to their treatment

They may imagine symptoms or cure, etc., but only the researcher/physician knows if these can be due to the medicine/treatment.

Page 14: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 14

HistoryHistoryHistoryHistory

Blind analysis begins in ~1930s with -

Medical research -- blind tests!

• Patients talk to medical researcher

e.g., how are you feeling? is pain less or more? are the other problems? serious or not so serious? before or after you take the medicine? …etc.

• Researcher hears what patient says

• Researcher examines the patient

• Researcher records all this as “data”

Do you see a Do you see a problem here?problem here?

Do you see a Do you see a problem here?problem here?

c.f.,1937 JAMA 26 June 2178/2.

Page 15: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 15

HistoryHistoryHistoryHistory

Blind analysis begins in ~1930s with -

Medical research -- blind tests!

• Therefore, the medical researcher can (and will) unconsciously and unintentionally interpret and record the information (data) with this bias (knowledge).

The patients are “The patients are “blindblind” to their treatment - ” to their treatment - butbut,,

the researcher is the researcher is NOTNOT!!

The patients are “The patients are “blindblind” to their treatment - ” to their treatment - butbut,,

the researcher is the researcher is NOTNOT!!

The solution is…

Page 16: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 16

Medical research -- The double blind test!

HistoryHistoryHistoryHistory

Blind analysis begins in ~1930s with -

• The list of patients and their treatment is prepared and maintained by someone who is NOT participating in the research. The list is sealed in the “black box”.

• The list can be retrieved from the “black box” only after all the analysis is completed.

The patients are “The patients are “blindblind” to their treatment - ” to their treatment - andand,,

the researcher is “the researcher is “blindblind” to who is getting treated!” to who is getting treated!

The patients are “The patients are “blindblind” to their treatment - ” to their treatment - andand,,

the researcher is “the researcher is “blindblind” to who is getting treated!” to who is getting treated!

1948 Am. Heart Jrnl., XXXVI, 529.

1950 Am. Jrnl. Med., IX, 142/1.

Page 17: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 17

Medical research -- The double blind test!

HistoryHistoryHistoryHistory

Blind analysis begins in ~1930s with -

Consider a simple analysis -- “Was the medicine effective?”

• The data on all patients will indicate whether the medication was effective for each patient.

• When this data analysis is complete (and free from mistakes) and any disagreement among the researchers have been settled and any concerns about bias have been removed, i.e., then (and only then) -

• The “black box” can be opened and the list can be retrieved.

• The data on all patients is now grouped into two groups: group (a) and group (b)

Page 18: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 18

Medical research -- The double blind test!

HistoryHistoryHistoryHistory

Blind analysis begins in ~1930s with -

The simple analysis -- “Was the medicine effective?”

• For what fraction of group (a) [fa] and group (b) [fb] was the medicine effective?• Do a statistical analysis to determine whether the difference in these two fractions is statistically significant - or whether it is consistent with a random occurrence. The result is now known!

• The researchers agree to accept the results obtained and no further analysis of the data is permitted - unless there has been a blunder (mistake) - in which case fix the blunder and report both results and the nature of the “fix”.

Page 19: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 19

HistoryHistoryHistoryHistory

Blind analysis has become the standard methodology in clinical trials.

Blind analysis has been used in the physical sciences only in recent years.

Physicists (and astronomers) are, of course ---- Careful-- Quantitative-- Attentive to bias (systemtic errors)

Why should they need a blind analysis?

Page 20: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 20

HistoryHistoryHistoryHistory

Ernest Rutherford (1934) --

“It seems to me that in some way it is regrettable that we had a theory of the positive electron before the beginning of the experiments. Blackett* did everything possible not to be influenced by the theory, but the way of anticipating results must inevitably be influenced to some extent by the theory. I would have liked it better if the theory had arrived after the experimental facts had been established.” Ernest Rutherford, Proc. Solvay Conference,

(Gauthier-Villars, Paris 1934), p 177.* Nobel prize in physics for discovering the positron (1948)

Page 21: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 21

HistoryHistoryHistoryHistory

Gregor Mendel (1865) --

The classic case is Gregor Mendel’s work on inheritance.

84 degrees of freedom, giving a probability that hewould have measured this well to be only 7x10-5 !!

In every case the data agreed with the theoretical ratios within less than the standard errors.Taking the whole together, 2 was 41.6 on

Page 22: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 22

HistoryHistoryHistoryHistory

Modern examples --

From nuclear and subnuclear physics ---

Reason: It’s the field in which the blind analysis techniques have been widely used - and -

It is the field with which I am most familiar.

Page 23: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 23

HistoryHistoryHistoryHistory

Page 24: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 24

HistoryHistoryHistoryHistory

Page 25: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 25

B-meson lifetime ratio

The mean ratio has a 2 of 4.5 for 13 degrees of freedom; P=0.985

8 Major experiments

Over 2000 physicists

Bottom line:

The agreement appears to be

too good!!

HistoryHistoryHistoryHistory

Page 26: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 26

OutlineOutlineOutlineOutline

1. What is a “blind analysis”?

• What is a “bias”?

• Does “bias” = “systematic error”?

• What is the special (subtle) bias?

2. What is the history of the blind analysis”?

3. What are examples of “blind analysis” in physics, astronomy, astrophysics, and, … ?

4. Some thoughts and reflections --

Page 27: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 27

An example of a “search” for an uncommon occurrence

A reaction that violates the Standard Modelconservation of lepton number

If you do not find the reaction, you have not shown the Standard Model to be incorrect

If you do find the reaction, you had better be very sure that you have got it right!!!

Page 28: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 28

MEGA

Search for e

Question: When and How are the values of determined?

Signal: Ee = E = 52.8 MeV

Boxes represent 2 boundaries

ExamplesExamplesExamplesExamples

Page 29: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 29

MEGA

Search for e

Blind Analysis: The boundaries aredetermined before the analysis of

(most of) the data!

Signal: Ee = E = 52.8 MeV

Boxes represent 2 boundaries

ExamplesExamplesExamplesExamples

Page 30: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 30

Page 31: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 31

TWIST

High precision ee

ExamplesExamplesExamplesExamples

A search for physics beyond (not included in) the Standard Model.

p = 29.8 MeV/c

Measure: pe()

Page 32: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 32

Nu

mb

er

X = E/Emax

N(x) = a + bxa,b from theory

ExamplesExamplesExamplesExamples

Simulation = “Monte Carlo”

Simulation ofthe experiment

using a,b

Nexp(x) = a’ + b’x

Analyzed expimental

data

Find a’ & b’ bycomparison of

real data with

simulateddata

Find a’ & b’ bycomparison of

real data with

simulateddata

The detector system distorts the distribution

Page 33: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 33

Muon decay spectrum

Current SM = -0.007 ± 0.013 0 = 0.7518 ± 0.0026 3/4 = 0.7486 ± 0.0026 ± 0.0028 3/4P = 1.0027 ± 0.0079 ± 0.0030 1

)1(3)34(3

2)33(

)(cos0

2

2

xx

xxx

dxdx

d−+−+−∝

Γ

⎥⎦⎤

⎢⎣⎡ −+−+ )34(

3

2)1(cos xxP δθξμ

ExamplesExamplesExamplesExamples

Page 34: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 34

EnergyCos()

ExamplesExamplesExamplesExamples

Muon decay spectrum from Standard Model

Page 35: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 35

EnergyCos()

TWIST measured spectrum

ExamplesExamplesExamplesExamples

EnergyCos()

Standard Model spectrum

Monte Carlo computer code

,,,

Page 36: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 36

EnergyCos()

TWIST simulated spectrum

,,,

EnergyCos()

TWIST measured spectrum

ExamplesExamplesExamplesExamples

Problem!

This spectrum is the S.M. spectrum -- nothing is

hidden; the experimenter is not “blinded”.

Compare spectra ’,’,’,’

’,’,’,’

Page 37: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 37

EnergyCos()

TWIST measured spectrum

ExamplesExamplesExamplesExamples

EnergyCos()

Standard Model spectrum

Monte Carlo computer code

o,o,o,o

Unknown spectrum

o,o,o,o are generated randomly, are encrypted, stored secretly, and used to generate the simulated data. Nobody knows what the offsets from ,,, are.

o,o,o,o are generated randomly, are encrypted, stored secretly, and used to generate the simulated data. Nobody knows what the offsets from ,,, are.

Page 38: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 38

EnergyCos()

TWIST simulated spectrum

?

EnergyCos()

TWIST measured spectrum

ExamplesExamplesExamplesExamples

Compare spectra to get , , ,

’,’,’,’

= ’ - o

= ’ - o = ’ - o = ’ - o

Page 39: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 39

ExamplesExamplesExamplesExamples

Before you can open the black box:

• Get all computer codes working and tested

• Identify all sources of systematic error and evaluate the size of each one

• Take all of the data you will need including data to help estimate the systematic errors

• Be sure that the Monte Carlo programs accurately simulate your experiment or you will have false values

• Analyze all of the data you intend to use to get the result

o,o,o,o

Page 40: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 40

ExamplesExamplesExamplesExamples

o,o,o,o = ’ - o

= ’ - o = ’ - o = ’ - o

MeasureCalculate the results

Compare with Standard Model predictions

Write the paper

Page 41: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 41

ExamplesExamplesExamplesExamples

Let’s look at one of your experiments Let’s look at one of your experiments

from PHYS-245from PHYS-245

Page 42: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 42

ExamplesExamplesExamplesExamples

Present accepted value: c = 299,792,458 m/sc 3 x 108 m/s

A measurement of the speed of light

12

T

osc

t =θ

2πT ⇒ c =

D

t

D

Page 43: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 43

ExamplesExamplesExamplesExamples

Present accepted value: c = 299,792,458 m/sc 3 x 108 m/s

A measurement of the speed of light

12

T

osc

t =θ

2πT ⇒ c =

D

t

D

To = T +δT

To

co

Blinded

Page 44: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 44

OutlineOutlineOutlineOutline

1. What is a “blind analysis”?

• What is a “bias”?

• Does “bias” = “systematic error”?

• What is the special (subtle) bias?

2. What is the history of the blind analysis”?

3. What are examples of “blind analysis” in physics, astronomy, astrophysics, and, … ?

4. Some thoughts and reflections --

Page 45: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 45

ReflectionsReflectionsReflectionsReflections

1. Guide to a blind analysis: If my answer were to come out to be six standard deviations from the expected result, what would I do? Make the list, and then -do all of that before you look at the answer!

2. A blind analysis is intended to guard against experimenter bias. It will not guard against fraud; that requires integrity and honesty.

3. A blind analysis is likely not the fastest way to an answer.

Page 46: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 46

ReflectionsReflectionsReflectionsReflections

4. It is not always possible to achieve perfect “blindness” -- e.g., drug testing.

5. Don’t need to plan everything in the analysisbefore beginning; just keep the answer hidden.

6. A blind analysis may not work for every experiment -but it is worth investigating before you begin.

7. In a blind analysis you want to hide the answer from anyone else who might want to offer advice that may be based on the answer.

Page 47: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 47

Thankyou

ReflectionsReflectionsReflectionsReflections

8. A blind analysis removes the “comfort factor” - knowing what answer you are getting so you can make changes, do more analysis, repeat measurements, etc., if the answer is not what you expect.

But, that may not be the best science!But, that may not be the best science!But, that may not be the best science!But, that may not be the best science!

Page 48: A Blind Analysis You are not allowed to peek! Prof. Donald Koetke Senior Research Professor of Physics Valparaiso University

3/30/07 VU Colloquium 48