#
"The Apple iPod iTunes Anti-Trust Litigation"

### Filing
753

***ERRONEOUS ENTRY, PLEASE REFER TO DOCUMENT NO. 754 *** EXHIBITS re 752 Opposition/Response to Motion, filed byApple Inc.. (Attachments: # 1 Exhibit 2, # 2 Exhibit 3, # 3 Exhibit 4, # 4 Exhibit 5, # 5 Exhibit 6, # 6 Exhibit 7, # 7 Exhibit 8, # 8 Exhibit 9, # 9 Exhibit 11, # 10 Exhibit 12, # 11 Proposed Order)(Related document(s) 752 ) (Kiernan, David) (Filed on 1/14/2014) Modified on 1/14/2014 (jlmS, COURT STAFF).

Exhibit 6
Roger Noll, Ph.D.
Confidential - Attorneys' Eyes Only
The Apple iPod iTunes Anti-Trust Litigation
Page 1
Page 2
·1· · · · · · · · ·UNITED STATES DISTRICT COURT
·1· · · · · · · · ·UNITED STATES DISTRICT COURT
·2· · · · · · · ·NORTHERN DISTRICT OF CALIFORNIA
·2· · · · · · · ·NORTHERN DISTRICT OF CALIFORNIA
·3· · · · · · · · · · · OAKLAND DIVISION
·3· · · · · · · · · · · OAKLAND DIVISION
·4
·4
·5· ·THE APPLE iPOD iTUNES· · · · · · Lead Case No. C 05-00037
·5· ·THE APPLE iPOD iTUNES· · · · · · Lead Case No. C 05-00037
· · ·ANTI-TRUST LITIGATION
· · ·ANTI-TRUST LITIGATION
·6
·6
·7· ·____________________________
·7· ·____________________________
·8· ·This Document Relates To:
·8· ·This Document Relates To:
·9· ·ALL ACTIONS
·9· ·ALL ACTIONS
10· ·____________________________
10· ·____________________________
11
11
12
12
13
13· · · · · · · ·CONFIDENTIAL - ATTORNEYS' EYES ONLY
14· · · · · · · ·CONFIDENTIAL - ATTORNEYS' EYES ONLY
14
15· · · · · VIDEOTAPED DEPOSITION OF ROGER G. NOLL, PH.D.
15· · · · · ·Videotaped Deposition of ROGER G. NOLL, PH.D.,
16· · · · · · · · ·Wednesday, December 18, 2013
16· ·taken on behalf of the Defendant, at 1755 Embarcadero
17· · · · · · · · · · Palo Alto, California
17· ·Road, Palo Alto, California, beginning at 9:06 a.m. and
18
18· ·ending at 11:54 p.m., on Wednesday, December 18, 2013,
19
20
19· ·before Darcy J. Brokaw, CSR No. 12584.
21
20
22
21
23· ·Reported by:
22
· · ·Darcy J. Brokaw
23
24· ·RPR, CRR, CSR No. 12584
24
25· ·Job No. 10008944
25
Page 3
Page 4
·1· · · · · · · · · · · · ·APPEARANCES
·1· · · · · · · · · · ·INDEX TO EXAMINATION
·2
·2· · · · · · · · · · ·ROGER G. NOLL, PH.D.
·3
·4· ·For the Plaintiffs and the deponent, Dr. Noll:
·3
·5· · · · ROBBINS GELLER RUDMAN & DOWD, LLP
·4
· · · · · BY: ALEXANDRA S. BERNAY, ESQ.
·5· ·EXAMINATION· · · · · · · · · · · · · · · · · · · · ·PAGE
·6· · · · BY: JENNIFER N. CARINGAL, ESQ.
·6· ·BY MR. KIERNAN· · · · · · · · · · · · · · · · · · · · ·7
· · · · · 655 West Broadway, Suite 1900
·7· · · · San Diego, California· 92101
·7
· · · · · (619)231-1058
·8
·8· · · · xanb@rgrdlaw.com
·9
·9
10· ·For the Defendant, Apple Inc.:
10
11· · · · JONES DAY
11
· · · · · BY: DAVID KIERNAN, ESQ.
12
12· · · · BY: AMIR AMIRI, ESQ.
13
· · · · · BY: ROBERT MITTELSTAEDT, ESQ.
13· · · · 555 California Street, 26th Floor
14
· · · · · San Francisco, California· 94104
15
14· · · · (415)626-3939
16
· · · · · dkiernan@jonesday.com
17
15
16
18
17
19
18· ·Also present:
20
19· · · · Peter Hibdon, Videographer
20
21
21
22
22
23
23
24
24
25
25
Page 1..4
www.aptusCR.com
YVer1f
Page 21
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · · · ·MS. BERNAY:· Objection.· Argumentative.
·BY MR. KIERNAN:
· · · ·Q.· It would be column A divided by 1 plus
·column B times column B; isn't that correct?
· · · · · ·MS. BERNAY:· Objection.· Vague.
· · · · · ·THE WITNESS:· It would be explain -·explain it to me again.
·BY MR. KIERNAN:
· · · ·Q.· It would be -· · · ·A.· The actual overcharge would be -- so
·the -- what the percentage damages is is the
·percentage of the calculation you get from all the
·independent variables, which is an estimate of the
·transaction price you actually have.· And the
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·regression are correlated within a particular group
·and you don't do anything to correct for that, what
·would be the impact on the reported standard errors?
· · · · · ·MS. BERNAY:· Objection.· Vague and
·ambiguous.
· · · · · ·THE WITNESS:· I didn't completely follow
·the question.· Ask it again.
·BY MR. KIERNAN:
· · · ·Q.· If the residual errors in the regression
·are correlated within a particular group and you
·don't do anything to correct for that, what would be
·the impact on the reported standard errors?
· · · · · ·MS. BERNAY:· Same objection.
· · · · · ·THE WITNESS:· It could be either way.· It
·could make them higher or it could make them lower,
·depending on the nature of the correlation.
·BY MR. KIERNAN:
· · · ·Q.· And why would it impact the reported
·standard errors?
· · · ·A.· Well, it's all built up in the -- in the
·nature of the assumptions one makes in doing a
·regression analysis, which is an independence of the
·standard errors.· And if the standard errors -- if
·the -- if the random shock that is -· · · · · ·(Reporter inquires.)
Page 22
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·transaction price is -- has that overcharge of that
·amount.· All right.
· · · · · ·So I'm not sure I understand -· · · ·Q.· I'm focusing on the formula that's in C.
· · · ·A.· Yes.
· · · ·Q.· And the formula in C is taking the
·percentage of the weighted average price.· And my
·question is -· · · ·A.· That is the existing price.· It's not the
·but-for price.
· · · ·Q.· Right.· And what I'm asking is:· Isn't the
·correct formula to determine the price overcharge
·A divided by 1 plus column B times column B -· · · · · ·MS. BERNAY:· Objection -·BY MR. KIERNAN:
· · · ·Q.· -- because column B reflects the change in
·percentage between -· · · ·A.· Yes, you're right -· · · ·Q.· -- the but-for price and -· · · · · ·(Reporter admonishes.)
· · · · · ·THE WITNESS:· Yes, the 2.3.8 is an
·approximation of what the -- what the exactly
·precise calculation would be, yes.
·BY MR. KIERNAN:
· · · ·Q.· Okay.· If the residual errors in the
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · · · ·THE WITNESS:· If the random shock that is
·in the regression equation does not satisfy the
·independence assumption, then the effect on the
·standard errors of the coefficients could be either
·to elevate them or to reduce them, depending on the
·nature of the violation of the independence
·assumption.
·BY MR. KIERNAN:
· · · ·Q.· Okay.· And are there standard statistical
·tests to test whether the residual errors are
·correlated within a particular group?
· · · · · ·MS. BERNAY:· Objection.· Vague.
· · · · · ·THE WITNESS:· There are many such tests
·and many such corrections.· But the effect is -- the
·existence of even statistically significant
·correlations is small unless those correlations are
·high.· All right.
· · · · · ·So the corrections for autocorrelation of
·residuals are not something that actually matters in
·the vast majority of cases because the -- it's
·almost never the case there's no correlation in
·residual errors, but it's almost never the case that
·making a correction for the auto- -- the correlation
·that does exist matters in terms of the regression.
· · · · · ·It's also the case here that we're not
Page 23
Page 24
Roger Noll, Ph.D.
Confidential - Attorneys' Eyes Only
The Apple iPod iTunes Anti-Trust Litigation
Page 25
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·talking about a source of bias in the coefficients.
·We're talking about a source of bias in the
·estimated statistical significance, the -·BY MR. KIERNAN:
· · · ·Q.· The standard errors?
· · · ·A.· Yeah, the values of the -- the expected
·value of the regression coefficients is not
·affected.
· · · ·Q.· The coefficients aren't affected, but the
·calculations of the standard errors are affected?
· · · ·A.· Right, the calculations of the standard
·errors are affected, but the -- but the estimated
·effect of the independent variable is the same, the
·expected estimated effect.
· · · ·Q.· And if the residual errors are correlated
·within a particular group, the standard errors could
·either be overstated or understated?
· · · ·A.· Yes.
· · · ·Q.· Without a correction?
· · · ·A.· They could be.· Although, again, the -·it's not -- it's not a dichotomous issue.· They -·A, they may be affected, and B, the magnitude of the
·effect depends on the exact conditions.
· · · ·Q.· And to know the magnitude of effect, you'd
·have to test it, you'd have to run one of the
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·that -- that's a good way to see if there's positive
·error correlation, but it's not a good way to see if
·there's negative error correlation.
· · · · · ·And the second point is that the nature of
·the error correlation may be that it's dependent on
·particular combinations of variables; and that one,
·the standard tests wouldn't even tell you that it
·exists.
· · · ·Q.· In this case, did you do anything to check
·whether the residual errors in your regression set
·forth in Exhibits 3A and 3B to Noll 10 are
·correlated with any particular group?
· · · · · ·MS. BERNAY:· Objection.· Vague and
·ambiguous.
· · · · · ·THE WITNESS:· What do you mean by "group"?
·BY MR. KIERNAN:
· · · ·Q.· Within any group.
· · · ·A.· What do you mean, "a group"?· I don't
·understand what you mean by a group.
· · · ·Q.· We've been using group for the last ten
·minutes.
· · · · · ·MS. BERNAY:· Objection.· Argumentative.
·BY MR. KIERNAN:
· · · ·Q.· Same group that you've -- the same group
·that you've been referring to.
Page 26
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·standard statistical tests?
· · · · · ·MS. BERNAY:· Objection.· Calls for
·speculation.
· · · · · ·THE WITNESS:· Well, actually, that's not
·what most -- what typically -·BY MR. KIERNAN:
· · · ·Q.· Can you just eyeball it?
· · · ·A.· -- happens.
· · · · · ·(Reporter inquires.)
·BY MR. KIERNAN:
· · · ·Q· ·Can you just eyeball it?
· · · · · ·MS. BERNAY:· Objection.· Vague.
· · · · · ·THE WITNESS:· Can I finish my first answer
·before I answer the next question?
·BY MR. KIERNAN:
· · · ·Q.· Yes.
· · · ·A.· Okay.· It is the case that if you plot the
·errors, you will know from experience if you
·actually have a problem that is causing the
·regression equation to be unreliable.· But so
·"eyeball" is sort of a bizarre word.
· · · · · ·What you actually do is you look at the
·actual scatter plot of points around the regression
·line and see if there is a clustering of
·observations above and below it.· The problem with
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · ·A.· I didn't refer to a group.· I don't know
·what you're talking about.· I know I fully
·intended -· · · ·Q.· You used the term "cluster" -· · · · · ·(Reporter admonishes.)
·BY MR. KIERNAN:
· · · ·Q· ·You used the word cluster, within a
·cluster.
· · · ·A.· I don't agree that there are any clusters
·here.
· · · · · ·MS. BERNAY:· Objection.
·BY MR. KIERNAN:
· · · ·Q.· That's not my question, Dr. Noll.· I asked
·you, did you do anything to check whether the
·residual errors in your regressions set forth in
·Exhibit 3A and 3B are correlated within any cluster
·or group?
· · · · · ·MS. BERNAY:· Objection.· Asked and
·answered.
· · · · · ·THE WITNESS:· I don't know what you mean
·by a group.· And you used the word "or," and I don't
·believe there are any clusters.· So how can I test
·for something when I don't -- I think it either
·doesn't exist or I don't understand what you're
·asking?
Page 27
Page 28
Page 25..28
www.aptusCR.com
YVer1f
Roger Noll, Ph.D.
Confidential - Attorneys' Eyes Only
The Apple iPod iTunes Anti-Trust Litigation
Page 29
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · · · ·What is it you're asking?· Can't you just
·give me an example of what you mean by a group, and
·then we won't have to discuss it?
·BY MR. KIERNAN:
· · · ·Q.· So you don't understand the question?
· · · ·A.· I don't understand what you mean by a
·group, no.· I don't know what you have in mind.
· · · ·Q.· And you don't know what I mean by cluster?
· · · · · ·MS. BERNAY:· Objection -· · · · · ·THE WITNESS:· I know what you mean by a
·cluster, and there aren't any in this particular
·regression.
·BY MR. KIERNAN:
· · · ·Q.· How do you know?
· · · ·A.· Because I know what cluster analysis is,
·and it doesn't apply to this regression because this
·isn't a sample.
· · · ·Q.· What did you do to determine if there were
·clusters?· What statistical tests did you apply?
· · · · · ·MS. BERNAY:· Objection.
· · · · · ·THE WITNESS:· I looked at the definition
·of a cluster, and it doesn't apply to anything in
·this regression.· I know -- I know what cluster
·analysis is, and it doesn't apply to this
·regression, notwithstanding what many of your
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·There is -·BY MR. KIERNAN:
· · · ·Q.· I'm just trying to understand what you did
·other than reading some books to determine if there
·are clusters in the case.
· · · · · ·MS. BERNAY:· Objection.· Argumentative.
· · · · · ·THE WITNESS:· There is no such thing as a
·test for whether you ought to use cluster analysis
·in a regression that doesn't satisfy the conditions
·for clustering.
·BY MR. KIERNAN:
· · · ·Q.· Okay.· That's what you teach your
·students?
· · · · · ·MS. BERNAY:· Objection.· Argumentative.
· · · · · ·THE WITNESS:· Of course it is.
·BY MR. KIERNAN:
· · · ·Q.· On page 34 of Noll 10 -- let me know when
·you get there.
· · · ·A.· I'm there.
· · · ·Q.· The first paragraph, the last third, you
·state that "Professors Murphy and Topel do not test
·whether the mean residual errors from this procedure
·are statistically significantly different from zero,
·which would have to be the case if the errors within
·a cluster are correlated."
Page 30
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·experts have said.· They're just not right.
·BY MR. KIERNAN:
· · · ·Q.· Anything else other than looking at a
·definition?
· · · · · ·MS. BERNAY:· Objection.· Argumentative.
· · · · · ·THE WITNESS:· I know -- the report, about
·a third of this report is about what cluster
·analysis is and what kinds of problems you apply to
·it and why this isn't a cluster sample problem.· All
·right.
· · · · · ·So, yes, there it is.· I've cited articles
·in the professional literature of which I not only
·have read, but I actually know what they do.· I have
·taught this stuff.· So I know what I'm talking
·about.· And there's references here.· It's not that
·I just read a definition and decided that something
·didn't apply.
· · · · · ·But I know, just from knowing what cluster
·analysis is, that it doesn't apply here.
·BY MR. KIERNAN:
· · · ·Q.· You just know it when you see it?
· · · · · ·MS. BERNAY:· Objection.· Argumentative,
·misstates his prior testimony.
· · · · · ·Come on, David.
· · · · · ·THE WITNESS:· That's complete nonsense.
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · ·A.· Yes.
· · · ·Q.· Did you perform that analysis?
· · · ·A.· No, because I don't believe there are
·clusters.· The premise of that paragraph is if you
·assume a cluster analysis is appropriate, here's
·something you do.· And they didn't do it.· But I
·don't think you should even do that because it's not
·a cluster sample problem.
· · · ·Q.· If it turns out that within a group,
·within a cluster -- we can use the one defined by
·Professors Murphy and Topel -- the mean residual
·errors are statistically significantly different
·from zero, what would that tell you?
· · · ·A.· Nothing.
· · · ·Q.· Why not?
· · · ·A.· Because as I said before, you only get
·that far if you have a cluster sampling problem, and
·we don't have a cluster sampling problem.· So
·there's no point in testing for cluster, the
·presence of clustering effects if you don't have a
·cluster to begin with.
· · · · · ·This is a paragraph written on if there -·if it were a sample -- if the way I had done the
·analysis was to sample some transactions according
·to a subset of the models of iPods that were out
Page 31
Page 32
Page 29..32
www.aptusCR.com
YVer1f
Roger Noll, Ph.D.
Confidential - Attorneys' Eyes Only
The Apple iPod iTunes Anti-Trust Litigation
Page 33
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·there, so instead of having 100-odd iPod models, I
·only had 20, and within those 20, I had just drawn a
·sample of transactions instead of looking at the
·entire universe, then, in principle, there might be
·a clustering problem.· But when you don't have a
·sample of either the models or the transactions,
·it's not a cluster problem.
· · · · · ·So testing for cluster effects is a
·non sequitur.· It's inappropriate, because you don't
·have cluster samples.
· · · ·Q.· Okay.· And other than that basis that
·there's not a clustering problem because it's not a
·sample from a population, any other reason, any
·other basis for your opinion that there's not a
·clustering issue?
· · · ·A.· Only the fact it doesn't satisfy the
·conditions for doing cluster analysis?
· · · ·Q.· The one that you just described.
· · · ·A.· Yes.· That's why it isn't a cluster
·problem, is because it's not a cluster sample.· And
·cluster sampling is a procedure you use when you are
·sampling on both groups and people within a group.
· · · · · ·If you have a population instead of a
·sample, there's no cluster issue, by definition.
· · · ·Q.· And so if the mean residual errors within
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·other two are versions of omitted variable problems.
· · · · · ·So the issue is, is there a sampling issue
·here?· The answer is no.
· · · · · ·Are there omitted variables?· I'm not
·aware of any that would add statistical significance
·to the regression equation without being so highly
·multicollinear that they would destroy the
·coefficient estimates.
· · · · · ·So there can't -- there isn't any -- none
·of the three reasons why you might have a problem
·exist.· So I don't care what the test is, because
·it's testing for something that, in principle, can't
·exist as a problem in the regression.
·BY MR. KIERNAN:
· · · ·Q.· So if you run a test on a particular group
·of transactions and the test shows that the mean
·residual errors are statistically significantly
·different from zero, your opinion is it has no
·impact on the calculation of the standard errors?
· · · · · ·MS. BERNAY:· Objection.· Vague and
·ambiguous.· Misstates prior testimony as well.
·BY MR. KIERNAN:
· · · ·Q.· Let me put it differently.· It does not
·overstate or understate the standard errors that
·you're calculating?
Page 34
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·certain particular groups in the transaction data at
·issue in this case are correlated, that is, they are
·statistically significantly different from zero,
·your opinion is it has no impact on the calculation
·of the standard errors in the case?
· · · ·A.· That's not what I said.
· · · · · ·MS. BERNAY:· Objection.· Misstates his
·prior testimony.
·BY MR. KIERNAN:
· · · ·Q.· What was wrong with -- what do you
·disagree with in the question I just asked?
· · · · · ·MS. BERNAY:· Objection.· Vague.
· · · · · ·THE WITNESS:· First of all, if you look
·within a -- if you define the group as a particular
·model of an iPod, and you look at the errors in
·predicting that, and you find they're correlated, it
·may be -- it's perfectly explained if you took into
·account all the values of all the other independent
·variables.
· · · · · ·So that test in and of itself doesn't
·prove anything.· All right.· The only way it proves
·something -- again, let's go back to the reasons
·cluster sampling can be a problem.· And as stated in
·the report, there's three reasons why it can be a
·problem.· One is a sample bias problem, and the
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · · · ·MS. BERNAY:· Same objection.
· · · · · ·THE WITNESS:· It may or may not.· You
·haven't -- there's not enough information in your
·question to make a prediction about the effect on
·the calculation of the standard errors.
·BY MR. KIERNAN:
· · · ·Q.· What additional information do you need?
· · · ·A.· You have to understand what is the source
·of what you're measuring.· All right.· You have
·to -· · · ·Q.· The source of the observations?
· · · ·A.· No.
· · · · · ·THE REPORTER:· What's the question?
· · · · · ·You guys are cutting each other off.
· · · · · ·THE WITNESS:· Yeah, he does do this,
·doesn't he?
· · · · · ·The very first step is precisely what
·residual errors are you correlating, what actually
·is it.· All right.· And I don't know the answer to
·that.
· · · · · ·All you're telling me is that within a
·model of iPods, the mean residual error isn't
·zero.· That's all you're telling me.· You're not
·telling me anything else about why it might be
·different from zero.
Page 35
Page 36
Page 33..36
www.aptusCR.com
YVer1f
Roger Noll, Ph.D.
Confidential - Attorneys' Eyes Only
The Apple iPod iTunes Anti-Trust Litigation
Page 37
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · · · ·In fact, purely as a statistical matter, I
·would expect it not to be the case that they would
·all be zero, all right, just purely from random
·sample or random -- the random shocks in the
·regression.
· · · · · ·So we have to know why.· Before we can get
·to the question "is that going to affect the
·calculation of the standard errors of the regression
·coefficients," we have to understand why the
·residual errors don't sum to zero.
·BY MR. KIERNAN:
· · · ·Q.· And did you explore any of the why the
·mean residual errors in your regression are
·statistically significantly different from zero for
·certain groups of transactions of iPods?
· · · · · ·MS. BERNAY:· Objection.· Vague and
·ambiguous.· Again, mischaracterizes the prior
·testimony.
· · · · · ·THE WITNESS:· First of all, you're
·assuming in the way the question is answered that I
·know which ones are statistically significantly
·different from zero, and I don't.
· · · · · ·Secondly, all I did is examine the reasons
·given by Professors Murphy and Topel as to why these
·things were different from zero, and they're wrong.
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·bootstrapping as a technique.· And describe what
·"bootstrapping" is so I make sure I understand it.
· · · ·A.· Sample -- you have a small number of
·observations and -- this was actually invented by my
·college roommate.
· · · · · ·You have a small number of observations,
·and the idea is if you just ran a single regression
·on the small sample that you have, the end wouldn't
·be large enough to be able to detect an effect, a
·causal effect of one variable on another.
· · · · · ·So what you do is you draw a sample
·from -- a sample with replacement; that is to say,
·you pick an observation, pull it out, count that as
·an observation, and you put it back into the puddle
·of all the observations and you draw another one.
· · · · · ·And you do that several times, run a
·regression.· And then you do it all again and run
·another regression, and then you do it all again and
·run another regression.· And then you use the
·distribution of the coefficients from those
·regressions as a way to estimate what the true
·coefficient is.
· · · ·Q.· Is that something you did in this case?
· · · ·A.· No.· We don't have a small sample.· We
·have a population.
Page 38
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · · · ·So I didn't go beyond that.
· · · · · ·And they're small to begin with.
·Regardless of whether they're statistically
·significant, they're small anyway.
· · · · · ·So there's certainly no proof that the
·answer in the regression equation about what damages
·are is in any way affected by anything in there that
·they discuss with regard to cluster analysis.
·BY MR. KIERNAN:
· · · ·Q.· Dr. Noll, when you say "they're small to
·begin with.· Regardless of whether they're
·statistically significant, they're small anyway,"
·what are you referring to as they are small?
· · · ·A.· Well, there's a -- in the backup stuff to
·the reports, the residual errors, the mean residual
·errors by model are not big numbers.· That's what I
·recall.· I don't remember the precise thing because
·it was months ago.
· · · · · ·But we did in fact examine what the basis
·was for their statements about the mean residual
·error, and there was no -- there was really nothing
·very important there.
· · · ·Q.· In your report, you discuss one technique
·for -- well, strike that.
· · · · · ·You set forth a description of
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · ·Q.· What was the point of discussing that in
·your report?
· · · ·A.· The point of discussing it was the
·mischaracterization of what independence means, that
·the -- Professors Murphy and Topel mischaracterize
·independence as being the same observation.· And in
·bootstrapping, you use the same observation over and
·over and over and over again, and it doesn't violate
·the independence assumption.
· · · ·Q.· What does the independence assumption
·refer to, in your words?· You disagree with
·Professor Murphy and Professor Topel.· Define for me
·what you're referring to as the independence
·assumption.
· · · ·A.· It's that the random component of the
·regression equation -- the distribution of that
·random component is unaffected by the observed
·values of any other component.
· · · · · ·And the reason the independence assumption
·is satisfied in bootstrapping is that you're
·randomly drawing samples.· So before the fact, what
·the next observation is going to be is independent
·of what the previous observation was.
· · · ·Q.· When you were referring to "random
·component," were you referring to the residual?
Page 39
Page 40
Page 37..40
www.aptusCR.com
YVer1f
Roger Noll, Ph.D.
Confidential - Attorneys' Eyes Only
The Apple iPod iTunes Anti-Trust Litigation
Page 41
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · ·A.· The residual is an estimate.· It's the
·unexplained variance.· The independence assumption
·refers to the underlying distribution of the error
·term.· And then the residual error in the regression
·equation is for the entire equation.· By definition,
·a regression analysis has to have mean zero.
· · · · · ·So the issue then about residual errors
·being correlated is you draw some subset of the
·observations and say is that sub- -- does that
·subset have correlations.· And then if it does, that
·means the correlations of the residual errors in
·that group, if on average they're greater than zero,
·that means all the others have to be on average less
·than zero.
· · · · · ·And then the issue is why does one
·subsample have positive residual errors and another
·have negative.· And there's some potential
·explanations for that, one of which is the actual
·way you created the groups, because you may not have
·taken fully into account the effect, the actual
·effect that's already explained in the regression of
·some of the independent variables.
· · · ·Q.· And if I understand your answer, one way
·to test your independence assumption is to look at
·the distribution of the error terms in the
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·dealing with the data, have constructed subsamples
·in a way to get groups that have residual errors
·that are statistically significantly different from
·zero.· That doesn't tell me anything about the
·underlying quality of the regression, the standard
·errors or anything else.· It just means that I've
·cherry-picked.
· · · · · ·So that's why the answer to questions like
·you've been asking me always have to be "it
·depends."· It depends on how the subsample was
·collected whether any test of whether the residual
·errors are positive or negative even makes sense to
·begin with.
· · · ·Q.· When dealing with -- strike that.
· · · · · ·Are there other cases in which you have
·worked with an entire population of transactions in
·estimating a regression?
· · · ·A.· Yes.
· · · ·Q.· And in those cases, have you done anything
·to test the independence assumption that we've been
·discussing?
· · · · · ·MS. BERNAY:· Objection.· Calls for
·speculation, vague and ambiguous.
· · · · · ·THE WITNESS:· I have -- the only -- first
·of all, the only circumstances in which that even is
Page 42
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·regression?
· · · ·A.· It can be, but it isn't necessarily.· You
·have to -- you have to have an underlying both
·economic theoretic and econometric theoretic reason
·to believe that the groupings you have make sense.
· · · · · ·In other words, I can always construct a
·way to separate a sample into two groups so that one
·has positive residual errors on average and the
·other has negative, but that doesn't mean that
·there's a problem with the regression analysis,
·because I've constructed it to produce that, that
·result.· And that's why I say you'd have to know
·what the reason for it is.
· · · · · ·Just to take a very simple example, I
·could just take the ten observations where the model
·underestimates the true value by the maximum amount,
·all right, the worst possible observations in terms
·of underpredicting the dependent variable.· Then I
·could call that a group, and I say, ah-hah, those
·are statistically significant positive residual
·errors.
· · · · · ·But that doesn't mean there's anything
·wrong with the model.· It doesn't mean there's
·anything funny going on with violation of
·independence.· It just means that me, as the person
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·an interesting issue would be where you had very low
·explanatory power in the regression.· All right.
·Then it's possible that you could have economically
·and econometrically meaningful subgroups that had
·positive or negative residual errors.
· · · · · ·And so if you have extremely high
·R-squares, if your regression is doing a good job
·explaining the data, then it would not be a
·meaningful exercise to do that.· And in most cases,
·I never do, because the R-square, like this one, is
·very high.
·BY MR. KIERNAN:
· · · ·Q.· So if the R-square is very high and you're
·dealing with a -· · · ·A.· Population.
· · · ·Q.· -- population subset, your opinion is
·there's no reason to test the independence
·assumption?
· · · · · ·MS. BERNAY:· Objection.· Mischaracterizes
·the prior testimony.
· · · · · ·THE WITNESS:· Right.· I have normally not
·attempted to test, but there are -- the only
·circumstances in which I would do that is if there
·was -- there was some really big outlier prediction
·errors and they were all the same thing.· And you
Page 43
Page 44
Page 41..44
www.aptusCR.com
YVer1f
Roger Noll, Ph.D.
Confidential - Attorneys' Eyes Only
The Apple iPod iTunes Anti-Trust Litigation
Page 45
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·could get still get a high squared with a very small
·subset getting big prediction errors.
· · · · · ·(Reporter inquires.)
· · · · · ·THE WITNESS:· You can have a high
·R-squared in a regression and still have a group of
·predictions that were -- where the prediction error
·is large.· And then you would -- you would still
·want to address whether that group -- you had some
·omitted variable for that group or something.
· · · · · ·But again, that's not really likely to
·happen if you already have group identifiers.· See,
·again, the -- by definition, if you have group
·identifiers, the residual error within that group is
·going to be zero.· The mean residual error is going
·to be zero, because that's what regression analysis
·does.
· · · · · ·So that's why, for example, the most
·conventional solution to cluster problems is to use
·group identifiers, indicator variables, to get the
·mean of those residual errors for each group to
·zero.
·BY MR. KIERNAN:
· · · ·Q.· In this case, did you perform any
·statistical test to determine or to test your
·independence assumption?
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
·BY MR. KIERNAN:
· · · ·Q.· If you could turn to page 27.
· · · · · ·Let me know when you get there, Dr. Noll.
· · · · · ·MS. BERNAY:· 27?
· · · · · ·MR. KIERNAN:· Yeah, of Noll 10.
·BY MR. KIERNAN:
· · · ·Q.· In the second paragraph, you state that
·"Whereas new iPod owners in late 2006 became more
·locked in to iPods over time..."
· · · · · ·Do you see that sentence?
· · · ·A.· Yes.
· · · ·Q.· When you're referring to "new iPod
·owners in late 2006," who are you referring to?
· · · ·A.· People who had just bought or were about
·to buy an iPod.
· · · ·Q.· Okay.· And are you referring to consumers
·that purchase an iPod -- only those consumers who
·purchase an iPod with 7.0 implemented?
· · · ·A.· The issue of lock-in effect also depends
·on how locked in you are, of course.· But the -·if -- and it also depends on whether you have bought
·digital downloads from the iTunes Store or not.
· · · · · ·So the degree of lock-in would be affected
·by 7.0, but it's not the only factor affecting
·lock-in.
Page 46
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · · · ·MS. BERNAY:· Objection.· Asked and
·answered.
· · · · · ·THE WITNESS:· I have -- I have not
·performed a test of the independence assumption as
·you've put it in that way, no.· It would be
·unnecessary, because there are no groups with
·outlying residual errors in the R-squared spot.· And
·by definition, the mean residual errors by group are
·going to be zero.
·BY MR. KIERNAN:
· · · ·Q.· And if statistical tests show that mean
·residual errors within groups are correlated, that
·does not affect your analysis or any of your
·opinions in any way?
· · · · · ·MS. BERNAY:· Objection.· Calls for
·speculation.
· · · · · ·THE WITNESS:· It might or it might not,
·depending on what the reason for finding that
·correlation was, that statistically significant
·correlation was.· It would purely depend on the way
·the test was performed and the way the groups were
·created and the way the residual errors were
·calculated.· All right.· That's what it would depend
·on.
·///
·1·
·2·
·3·
·4·
·5·
·6·
·7·
·8·
·9·
10·
11·
12·
13·
14·
15·
16·
17·
18·
19·
20·
21·
22·
23·
24·
25·
· · · · · ·MR. KIERNAN:· Okay.· Move to strike as
·nonresponsive.
·BY MR. KIERNAN:
· · · ·Q.· With respect to your first sentence, you
·state:· "Whereas new iPod owners in late 2006 became
·more locked in to iPods over time..."
· · · · · ·My question is:· The "new iPod owners in
·late 2006," does that refer to purchasers of iPods
·that only included 7.0?
· · · · · ·MS. BERNAY:· Objection.· Asked and
·answered.
· · · · · ·THE WITNESS:· My nonresponsive answer was
·in fact responsive.· It depends on other things.
·All right.· People who bought 7.0, obviously 7.0 -·iPods with 7.0 in them contributed to a lock-in
·effect more than people whose iPods did not have
·7.0.
· · · · · ·But on the other hand, if people,
·regardless of whether they bought 7.0, bought music
·from the iTunes Store in a DRM-protected fashion,
·they would be experiencing lock-in as well.
·BY MR. KIERNAN:
· · · ·Q.· Okay.
· · · ·A.· So that, contrary to your assertion, it
·was completely responsive.· You just didn't
Page 47
Page 48
Page 45..48
www.aptusCR.com
YVer1f
Roger Noll, Ph.D.
Confidential - Attorneys' Eyes Only
The Apple iPod iTunes Anti-Trust Litigation
Page 117
·1· · · · · · · · · · ·REPORTER CERTIFICATION
·2
·3· · · · · · ·I, Darcy J. Brokaw, a Certified Shorthand
·4· ·Reporter, do hereby certify:
·5· · · · · · ·That prior to being examined, the witness in
·6· ·the foregoing proceedings was by me duly sworn to
·7· ·testify to the truth, the whole truth, and nothing but
·8· ·the truth;
·9· · · · · · ·That said proceedings were taken before me at
10· ·the time and place therein set forth and were taken down
11· ·by me in shorthand and thereafter transcribed into
12· ·typewriting under my direction and supervision;
13· · · · · · ·I further certify that I am neither counsel
14· ·for, nor related to, any party to said proceedings, nor
15· ·in any way interested in the outcome thereof.
16· · · · · · ·In witness whereof, I have hereunto subscribed
17· ·my name.
18
19· ·Dated:· December 19, 2013
20
21· ·_____________________________
· · ·Darcy J. Brokaw
22· ·CSR No. 12584, RPR, CRR
23
24
25
Page 117
www.aptusCR.com
**Disclaimer:** Justia Dockets & Filings provides public litigation records from the federal appellate and district courts. These filings and docket sheets should not be considered findings of fact or liability, nor do they necessarily reflect the view of Justia.

**Why Is My Information Online?**