In re: High-Tech Employee Antitrust Litigation

Filing 716

Omnibus Declaration of Christina J. Brown in Support of #715 Reply re Joint Motion to Exclude the Expert Testimony of Edward E. Leamer, Ph.D. , #714 Reply to Joint Motion to Strike the Improper Rebuttal Testimony in Dr. Leamer's Reply Expert Report or, in the Alternative, MOTION for Leave to Submit a Reply Report of Dr. Stiroh filed by Apple Inc.. (Attachments: #1 Exhibit A, #2 Exhibit B, #3 Exhibit C, #4 Exhibit D, #5 Exhibit E, #6 Exhibit F, #7 Exhibit G, #8 Exhibit H, #9 Exhibit I, #10 Exhibit J, #11 Exhibit K, #12 Exhibit L, #13 Exhibit M, #14 Exhibit N, #15 Exhibit O, #16 Exhibit P, #17 Exhibit Q)(Related document(s) #715 , #714 ) (Brown, Christina) (Filed on 2/27/2014) Modified text on 2/28/2014 (dhmS, COURT STAFF).

EXHIBIT C

OMNIBUS BROWN DECLARATION

Deposition of Kevin M. Murphy, Ph.D.

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

1

UNITED STATES DISTRICT COURT

2

NORTHERN DISTRICT OF CALIFORNIA

3

SAN JOSE DIVISION

4
5
6

IN RE:

HIGH-TECH EMPLOYEE

7

ANTITRUST LITIGATION

8
9

)
)
)

THIS DOCUMENT RELATES TO:

)

10

ALL ACTIONS.

)

11

No. 11-CV-2509-LHK

______________________________)

12
13
14

CONFIDENTIAL - ATTORNEYS' EYES ONLY

15

VIDEO DEPOSITION OF KEVIN M. MURPHY, Ph.D.

16

December 3, 2012

17
18
19
20

REPORTED BY:

GINA V. CARBONE, CSR NO. 8249, RPR, CCRR

21
22
23
24
25
KRAMM COURT REPORTING

*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

Page: 1

Deposition of Kevin M. Murphy, Ph.D.

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

06:07:28

1

But the -- as long as -- if you are willing to

06:07:30

2

stick to that assumption that it's really conduct by

06:07:33

3

age, then the age variable can help you identify that.

06:07:37

4

But what you are fundamentally doing is you are asking

06:07:41

5

was the age profile different in the conduct years than

06:07:44

6

in the non-conduct years.

06:07:48

7

gets a result that's actually backwards of what he says

06:07:50

8

you should have gotten.

06:07:51

9

It's not surprising that he

He gets a result that says that the impact was

06:07:55 10

greatest on the youngest people and less on the

06:07:58 11

middle-age people, when his theory was it would be

06:08:02 12

exactly the reverse.

06:08:04 13

It's not surprising, given the amount of noise

06:08:06 14

he's got in his estimates.

06:08:13 15

of how poorly this regression actually performs.

06:08:20 16

Q.

Again, it's an illustration

So I'd like to direct your attention to

06:08:22 17

paragraph 128, please.

06:08:34 18

discussed clustering of standard errors.

06:08:40 19

This is the paragraph when you

What I'd like to ask you is towards the middle

06:08:43 20

of the paragraph you make a reference to -- you say

06:08:46 21

that, "This exhibit shows that none of Dr. Leamer's

06:08:49 22

'undercompensation' estimates for any employer or year

06:08:52 23

is statistically significant at conventional levels

06:08:55 24

under the properly computed standard errors."

06:08:59 25
KRAMM COURT REPORTING

What does the phrase "statistically significant
*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

Page: 363

Deposition of Kevin M. Murphy, Ph.D.

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

06:09:01

1

at conventional levels" mean?

06:09:04

2

06:09:07

3

people use is 95 percent or 5 percent level, however you

06:09:10

4

want to think about it.

06:09:12

5

one.

06:09:17

6

about statistically significant and they don't say at

06:09:19

7

the 1 percent level, at the 5 percent level or whatever,

06:09:22

8

I think the shorthand economist typically uses 5 percent

06:09:25

9

level.

A.

I think the most commonly used level that

I think that's the most common

If people talk -- in economics, when people talk

06:09:26 10

Q.

Is that a requirement of economic analysis?

06:09:28 11

A.

No, it's not a firm requirement.

I'm just

06:09:31 12

saying, you know, that's the conventional level that

06:09:33 13

people use.

06:09:34 14

Q.

Okay.

Is that -- if I wanted to sort of look

06:09:36 15

that up somewhere, would I be able to look it up

06:09:39 16

anywhere?

06:09:40 17

A.

Yeah.

Probably econometric textbook would talk

06:09:45 18

about that.

06:09:51 19

significance at various levels of significance.

But generally people talk about

06:09:55 20

(Reporter clarification.)

06:09:55 21

THE WITNESS:

I'm just telling you the common

06:09:57 22

shorthand in economics is 5 percent, just talking about

06:10:01 23

statistically significant with no modifier.

06:10:03 24
06:10:06 25

MR. GLACKIN:

Q.

So what does statistical

significance mean?

KRAMM COURT REPORTING

*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

Page: 364

Deposition of Kevin M. Murphy, Ph.D.

A.

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

06:10:08

1

It means in a classical statistical problem, it

06:10:12

2

means I achieved a result in terms of my estimate that

06:10:19

3

is typically, say, large relative to what I would expect

06:10:22

4

to happen just by chance.

06:10:26

5

06:10:28

6

no true effect, or no true difference, for example, in a

06:10:32

7

given sample, you are going to find a difference.

06:10:35

8

if the true -- say I had two populations and I was

06:10:38

9

comparing population A and population B, and I had

So in other words, in a world where there were

06:10:41 10

samples from each population, and I was going to

06:10:43 11

Even

calculate the average height from my samples.

06:10:46 12

Even if the true average height in both

06:10:49 13

populations is the same, in my sample there is going to

06:10:52 14

be a difference in the average height of the sample from

06:10:55 15

population A and the average height from the sample of

06:10:59 16

population B.

06:11:00 17

The test of statistical significance is did I

06:11:02 18

get a difference in heights across those two populations

06:11:07 19

that was too big to happen just by chance.

06:11:12 20

we quantify that is to say, did I get a difference in

06:11:16 21

heights that would happen less than 5 percent of the

06:11:19 22

time just by chance.

06:11:22 23

statistical significance.

06:11:24 24
06:11:31 25

Q.

Okay.

And the way

That's really the idea of

Do you agree that this is a

description -- that statistical significance is a

KRAMM COURT REPORTING

*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

Page: 365

Deposition of Kevin M. Murphy, Ph.D.

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

06:11:33

1

description of how certain a statistical result is?

06:11:40

2

06:11:45

3

how precisely I can estimate something, yeah.

06:11:50

4

of a description.

06:11:54

5

about significance and not talk about the components

06:11:56

6

that go into it, then you might say it's -- it could be

06:12:00

7

described in terms of certainty.

06:12:05

8

06:12:10

9

A.

Yeah.

Q.

It's not just -- it's a description of
Somewhat

I mean, if you are just going to talk

Is there any authority for -- well, is it your

opinion -- now, again, I don't want to invite you to

06:12:13 10

launch into -- excuse me.

06:12:16 11

a discursive answer of your reviews about Dr. Leamer's

06:12:20 12

regression.

06:12:22 13

question.

06:12:24 14

I don't want to invite you to

I'd really like to stick to answers to the

Is it your opinion that in order for a

06:12:26 15

statistical analysis to be reliable, it must produce a

06:12:30 16

statistically significant result?

06:12:32 17

A.

Not necessarily.

06:12:36 18

Q.

So --

06:12:38 19

A.

But statistical significance is one thing you

06:12:39 20

do look at.

06:12:44 21

P values, for example, that show up in the table.

06:12:49 22
06:12:55 23

Q.

That doesn't have to be true.

And particularly here, you can look at the

Okay.

So where are you directing me to?

Are

you on your report or Dr. Leamer's report?

06:12:57 24

A.

In my report.

06:13:11 25

Q.

Is this appendix 22B or Exhibit 22B?

KRAMM COURT REPORTING

So you look at table, say, 22B.

*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

Page: 366

Deposition of Kevin M. Murphy, Ph.D.

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

06:13:14

1

A.

Exhibit 22B or Exhibit 22A.

Either one.

We

06:13:17

2

06:13:20

3

Q.

Uh-huh.

06:13:22

4

A.

So these would be the P values, which is the

06:13:25

5

probability that that you get a number at least that big

06:13:28

6

just by chance.

06:13:34

7

there -- these are from his estimates that restrict the

06:13:37

8

coefficients across.

06:13:42

9

percent, which means it's a number -- I'm going to get a

can go with A, it's the first one.
Okay.

And you can see for lots of these,

You get a lot of these P values 50

06:13:45 10

number that size half the time just by chance.

06:13:49 11

Kind of

what those numbers mean.

06:13:51 12

Q.

You say there is a lot that are 50 percent?

06:13:53 13

A.

I'm saying there is ones that are 50 percent,

06:13:55 14

30 percent, 40 percent.

There is a few that are

06:13:58 15

smaller.

06:14:03 16

know, 30 percent or higher.

06:14:06 17

time I'm going to get a number like that just by chance.

But, you know, the majority of them are, you
That means a third of the

06:14:20 18

Q.

So --

06:14:27 19

A.

And remember, this is just looking for an

06:14:29 20

average effect, let alone asking the question whether

06:14:32 21

there is a common effect.

06:14:35 22

Q.

So if I wanted to look at some authority for

06:14:38 23

the proposition that these P values are a basis to

06:14:44 24

reject Dr. Leamer's regression analysis, what authority

06:14:48 25

should I look at?

KRAMM COURT REPORTING

*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

Page: 367

Deposition of Kevin M. Murphy, Ph.D.

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

06:14:51

1

A.

You could look at any basic econometrics

06:14:55

2

06:14:56

3

Q.

06:14:57

4

if I --

06:14:59

5

A.

06:15:01

6

book that we cite in here.

06:15:07

7

textbooks out there that would talk about these things.

06:15:11

8

06:15:13

9

textbook.

Q.

Should be easy for you to identify one, then,

You can look at Green, you could look at the
There is tons of econometric

And they will say a regression with P values in

that range ought to be rejected?

06:15:15 10

A.

No.

They would say P values in that range are

06:15:17 11

not something that you would say provides really

06:15:21 12

substantial evidence of the hypothesis.

06:15:25 13
06:15:28 14

Q.

Why don't you just give me one textbook that

you are certain includes this proposition.

06:15:30 15

A.

You know, look, I last looked at textbooks 30

06:15:34 16

years ago when I was in school.

People -- we don't rely

06:15:37 17

on textbooks for what we do.

06:15:41 18

done in research and papers and journals and all those

06:15:45 19

things.

06:15:51 20

could look at Green, I guess, would be a textbook that

06:15:54 21

would have it.

We -- you know, it's all

I mean, you know, you could -- you could -- you

You could look at, you know --

06:15:56 22

Q.

Is Green one that you cited in here?

06:15:58 23

A.

Yeah, we cited Green and we cited one other

06:16:02 24

one.

06:16:05 25
KRAMM COURT REPORTING

The book we cited on clustering.
Q.

So the Angrist and Pischke?
*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

Page: 368

Deposition of Kevin M. Murphy, Ph.D.

A.

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

06:16:06

1

06:16:10

2

be a useful one to look at.

06:16:17

3

mean, he'll tell you.

06:16:18

4

06:16:20

5

06:16:23

6

06:16:25

7

But he should be able to tell you that a P value of .5

06:16:30

8

isn't something that you would write home about.

06:16:32

9

Q.

No, Angrist and Pischke is -- yeah, that would
You could just ask Ed.

I

Well, if you'll take his word for it, whatever

his answer is, then I'm happy to do that.
A.

I sure hope he's still the same guy I knew.

But it's worse than that.

It's not the P

06:16:34 10

values here.

06:16:38 11

precision that you have for estimating even the average

06:16:41 12

effect.

06:16:49 13

Q.

06:16:55 14

Is there a better way to estimate the effect of

A.

I think if you are going to do it, you would

have to do it a different way.

06:17:06 17
06:17:09 18

It's really problematic, and it's unfortunate.

this conduct than using a regression analysis?

06:17:02 15
06:17:03 16

It's really -- it's really the degree of

Q.

What are some possible ways that are feasible

given the data?

06:17:11 19

A.

First off, I think you wouldn't want -- the

06:17:13 20

theory -- economics tells us that there is going to be

06:17:16 21

differential effects for different people, which I think

06:17:19 22

pushes you away from the regression analysis to begin

06:17:21 23

with.

06:17:27 24

going to give you an average, and that's not going to

06:17:29 25

tell you whether there was class-wide harm.

KRAMM COURT REPORTING

Because the regression analysis, at most, is

*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

I think you
Page: 369

Deposition of Kevin M. Murphy, Ph.D.

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

06:17:31

1

would have to move away from that.

06:17:36

2

regression analysis is going to be useful for that.

06:17:39

3

If you were going to do a regression analysis

06:17:41

4

you would have to have one that does a much better job

06:17:44

5

of controlling for the other determinants of firm-level

06:17:48

6

compensation over time.

06:17:53

7

solve your potential problem.

06:17:58

8

06:18:00

9

Q.

I don't think the

That's the thing that would

What I'm asking is, is there some mechanism

other than a regression analysis by which this can be

06:18:03 10

accomplished?

06:18:05 11

A.

06:18:09 12

But Professor Leamer

hasn't done it.

06:18:09 13
06:18:12 14

There very well could be.

Q.

Can you tell us any mechanisms, other than a

regression analysis, that would account for this --

06:18:16 15

A.

Sure.

You know, if I had some time to work on

06:18:18 16

it, I could come up with something probably.

06:18:21 17

what I was asked to do.

06:18:26 18

regression, the number of flaws it has, cannot be put

06:18:30 19

forward as the answer to this question.

06:18:33 20

can't.

06:18:41 21

Q.

The regression -- I think the

You don't have to be sorry.

time I've heard it, Dr. Murphy.

06:18:46 23

really okay.

06:18:48 24

A.

KRAMM COURT REPORTING

It really

And I'm sorry to say that.

06:18:43 22

06:18:48 25

That's not

It's not the first

Believe me.

It's

I understand.

Anyway....
MR. GLACKIN:

So, look, I have probably, I

*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

Page: 370

Deposition of Kevin M. Murphy, Ph.D.

1

In Re: HIGH-TECH EMPLOYEE ANTITRUST LITIGATION

I, Gina V. Carbone, Certified Shorthand

2

Reporter licensed in the State of California, License

3

No. 8249, hereby certify that the deponent was by me

4

first duly sworn and the foregoing testimony was

5

reported by me and was thereafter transcribed with

6

computer-aided transcription; that the foregoing is a

7

full, complete, and true record of said proceedings.

8
9

I further certify that I am not of counsel or
attorney for either of any of the parties in the

10

foregoing proceeding and caption named or in any way

11

interested in the outcome of the cause in said caption.

12

The dismantling, unsealing, or unbinding of

13

the original transcript will render the reporter's

14

certificates null and void.

15
16

In witness whereof, I have hereunto set my
hand this day:

December 6, 2012.

17

_______ Reading and Signing was requested.

18

_______ Reading and Signing was waived.

19

___X___ Reading and signing was not requested.

20
21
22

_________________________

23

GINA

24

CSR 8249, RPR, CCRR

V. CARBONE

25
KRAMM COURT REPORTING

*CONFIDENTIAL - ATTORNEYS' EYES ONLY*

Page: 385