Status, Quality, and Attention: What's in a (Missing) Name?

MANAGEMENT SCIENCEVol. 57, No. 2, February 2011, pp. 274–290issn 0025-1909 �eissn 1526-5501 �11 �5702 �0274

informs ®

doi 10.1287/mnsc.1100.1270©2011 INFORMS

Status, Quality, and Attention:What’s in a (Missing) Name?

Timothy S. SimcoeBoston University School of Management, Boston, Massachusetts 02215;

and National Bureau of Economic Research, Cambridge, Massachusetts 02138, [email protected]

Dave M. WaguespackManagement and Organization Department, Robert H. Smith School of Business,

University of Maryland, College Park, Maryland 20742, [email protected]

How much are we influenced by an author’s identity when evaluating his or her work? This paper exploits anatural experiment to measure the impact of status signals in the context of open standards development.

For a period of time, e-mails announcing new submissions to the Internet Engineering Task Force would replaceindividual author names with “et al.” if submission volumes were unusually high. We measure the impact ofstatus signals by comparing the effect of obscuring high- versus low-status author names. Our results show thatname-based signals can explain up to three-quarters of the difference in publication rates between high- and low-status authors. The signaling effect disappears for a set of prescreened proposals that receive more scrutiny thana typical submission, suggesting that status signals are more important when attention is scarce (or search costshigh). We also show that submissions from high-status authors receive more attention on electronic discussionboards, which may help high-status authors to develop their ideas and bring them forward to publication.

Key words : status; technology; sociology of scienceHistory : Received May 5, 2010; accepted September 29, 2010, by Jesper Sørensen, organizations. Published

online in Articles in Advance December 10, 2010.

1. IntroductionSociologists studying markets have long emphasizedthat identity—particularly the visible position withina status hierarchy—matters independently of under-lying quality for competitive outcomes. In transac-tions where there is uncertainty about the underlyingquality of the focal producer or the products, con-sumer decisions about whom to transact with andhow much the product is worth are influenced bysocial affiliation cues (Podolny 1993, 1994). Promi-nent individuals and organizations with superior sta-tus signals benefit from these acts of deference (Merton1968, Podolny 1993). Producer identity consequentlymatters, the argument goes, for a wide variety ofdecisions, from whose paper to publish to which jobapplicant to hire, to how much to pay for a particularbottle of wine.Advancing this line of research, however, is severely

hindered by the dynamic relationship between sta-tus, quality, and performance in real-world settings.For instance, Podolny and Phillips (1996) argue thatstatus is partially determined by innate ability andthat high-status producers benefit both from greaterperceived quality and from greater ability to producehigh-quality goods because of the superior resourcesthey receive. The reciprocal link between status and

quality makes it hard to disentangle their effects. Mul-tivariate regression can produce misleading results—even if one includes a perfect control for quality—because the two constructs are jointly determined.The primary contribution of this paper is that itdetails a testing approach that partly circumvents thisproblem by exploiting random variance in status sig-nals, as opposed to observed variation in status itself.Our basic approach will be familiar to anyone who

has participated in the peer review process and isnicely illustrated by the following anecdote about apaper submitted to the British Association for theAdvancement of Science by the third Lord Rayleigh,who went on to win the 1904 Nobel Prize in physics:

His [Rayleigh’s] name was either omitted or acciden-tally detached, and the Committee “turned it down”as the work of one of those curious persons calledparadoxers. However, when the authorship was dis-covered, the paper was found to have merits after all.

(Strutt 1968, p. 228, as quoted in Merton 1968)

In the Rayleigh case, we have solid evidence thatidentity effects are not simply due to unobservedquality, because removing Rayleigh’s name had nobearing on the innate merits of the manuscript. Bar-ring such an accident, it is difficult to separate thestatus signal from actual quality in observational data,

274

Simcoe and Waguespack: Status, Quality, and Attention: What’s in a (Missing) Name?Management Science 57(2), pp. 274–290, © 2011 INFORMS 275

particularly as we expect that high-quality produc-ers will attempt to acquire (and signal) greater sta-tus. The most talented scientists, for instance, maydisproportionately seek education or employment athigh-status institutions, leading to a strong correlationbetween status and latent quality. The Lord Rayleighexample also highlights the link between status andthe allocation of a scarce resource, namely the criti-cal attention of other scientists. If attention is criticalfor developing and improving new ideas, prominentscientists may have a cumulative advantage derivedfrom superior resources, and not just from successiveratcheting up of perceived quality.This study is essentially a large-scale replication

of the “Lord Rayleigh experiment,” which empha-sizes both identity-based signaling and the increasedattention allocated to high-status actors. Our datacome from the Internet Engineering Task Force (IETF),a community of engineers and computer scientistswho develop protocols used to run the Internet. Allnew protocols go through a nonblind peer reviewprocess, which begins with the IETF’s clerical staffposting a newly submitted manuscript to a Webserver and describing it in a short broadcast e-mailannouncement. The connection to the Lord Rayleighexperiment is that beginning in late 1999, the broad-cast e-mails announcing new manuscripts often usedthe generic label “et al.” instead of providing a com-plete list of author names. Both interview evidenceand statistical tests indicate that the use of et al. wasunrelated to the quality of the paper or the socialstanding of the authors. We therefore treat name-removal via the use of et al. as a natural experimentthat allows us to measure the impact of identity-based signals independently of the latent quality of amanuscript or its authors.We begin by examining the relationship between

author names and publication decisions for 4,165 pap-ers submitted to the IETF and at risk of receiving anet al. We report two main findings. First, concealinghigh-status author names makes publication substan-tially less likely, but only among a pool of “gen-eral interest” manuscripts, where submission rates arevery high and the average rate of publication is quitelow. In other words, identity matters independently ofproduct attributes when search costs are high. By con-trast, within a pool of manuscripts that are “pre-screened” as relevant to one of the IETF’s technicalworking groups, the author name signal does notmatter. Second, and just as important, we show thatthe conventional approach to estimating status effectsis highly sensitive to model specification and can pro-duce large statistically significant effects in settingswhere our natural experiment fails to reject the nullhypothesis that audiences are not responding to sta-tus signals.

We next examine the attention focused on authorsprior to the publication decision. During the periodwe study, the IETF received about 80 new or revisedtechnical manuscripts each week, averaging 23 pagesapiece. Given the large volume of new ideas, identity-based signals may provide a simple heuristic for allo-cating attention to better proposals (Merton 1968) ora useful cue for coordination (Morris and Shin 2002).We find that proposals from high-status authors gen-erate more discussion on IETF electronic discussionboards and are more likely to undergo revisions,but only for “general interest” submissions, and onlywhen the high-status author’s name is not obscured.We interpret these results as evidence that status helpsdraw critical attention to a new idea, which is usefulfor developing the concept and bringing it forward topublication (Podolny 2005, p. 26).The remainder of the paper proceeds as follows.

Section 2 reviews the literature on status and perfor-mance, with an emphasis on the role of status in sci-entific communities. Section 3 describes the IETF andthe origins of the et al. natural experiment. Section 4describes our data, measures, and statistical methods.Section 5 discusses the results. Section 6 concludes.

2. Status Signals and Performance:Theory and Evidence

Sociological theories of status typically proceed intwo steps, by arguing that (1) social positions andaffiliations influence beliefs about a focal actor,and (2) beliefs affect the focal actor’s performance(Podolny and Phillips 1996). Thus, Berger et al. (1980,p. 479) define a status-organizing process as “any pro-cess in which evaluations of and beliefs about thecharacteristics of actors become the basis of observ-able inequalities.”This general framework can be applied to a vari-

ety of settings. For example, Merton (1968) describeshow standing within the scientific community affectsthe distribution of resources (e.g., money, credit, orattention), which matter for both output and recog-nition. In market settings, the link between statusand perceived quality can lead to increased demandor bargaining power (Podolny 1993, Benjamin andPodolny 1999), and if legitimacy concerns induce low-status actors to self-select out of some niches, high-status actors may face less competition (Phillips andZuckerman 2001). These papers typically argue thatstatus mechanisms are strongest in settings wherethere is considerable ex ante uncertainty about theunderlying quality of a focal actor (or their products),because in that case others will rely more heavily onsocial affiliations, behavioral cues, and other status sig-nals when deciding how to act (Podolny 1994).1

1 Uncertainty about the true state of the world plays a similarrole in economic models of signaling (Spence 1973) and reputation

Simcoe and Waguespack: Status, Quality, and Attention: What’s in a (Missing) Name?276 Management Science 57(2), pp. 274–290, © 2011 INFORMS

Building on these ideas, a large and rapidly grow-ing literature provides evidence of the link betweenstatus and performance. For instance, Podolny (1993)measures status using a bank’s relative position insecurities underwriting advertisements and showsthat higher-status investment banks have a greaterspread between cost and price for a given offer-ing. Benjamin and Podolny (1999) measure a win-ery’s status in terms of network centrality (based ona matrix of cross-regional affiliations) and find thathigh-status wineries charge more for bottles of equiv-alent quality. Washington and Zajac (2005), measuringstatus as a historical legacy of privileged treatment atthe National Collegiate Athletic Association (NCAA)tournament, find that high-status college basketballteams with marginal records are more likely to receiveat-large invitations to the NCAA postseason bas-ketball tournament. Stuart et al. (1999), measuringprominence based on centrality within networks ofalliances and equity investors, find that biotechnologystart-ups endorsed by high-status organizations haveinitial public offerings faster and have greater ini-tial public valuations. Stewart (2005) measures statusas the number of third-party endorsements given toopen source software developers and finds that high-status developers are more likely to receive a positiveevaluation.Although this literature provides many insights

into the link between status and performance, somebasic questions remain unresolved. For example, in areview of social capital research, Mouw (2006) arguesthat even though theoretical distinctions have becomeincreasingly refined, the field lacks solid evidence thatthere is a causal relationship between social capitalof any form and outcomes, and is not even close tobeing capable of adjudicating among more refinedalternative theories. We posit that finding evidenceof status effects, and hence developing the theoreticalliterature, is hampered by two substantial empiricalchallenges.

2.1. Unobserved QualityAt the root of theorizing about status is the argumentthat actor identity matters independently of intrinsicquality (Podolny and Stuart 1995). However, it is verydifficult to establish a causal link between producer

(Kreps and Wilson 1982). However, sociological theories of statusdrop the economic assumption that actors’ beliefs after observing asignal are (on average) correct. For example, individuals may valuehigh-status activities or affiliations independently of their informa-tional value (Gould 2002); low-quality actors may have the abilityto mimic high-status traits; and signals that were once functional(the peacock’s tail) may confer status long after changes in theenvironment have rendered the underlying information obsolete(Feldman and March 1981).

status and observed performance. The problem is typ-ically described in terms of unobserved heterogene-ity: some unmeasured dimension of quality mightdrive both status and performance. However, there isalso the issue of reverse causation, where the hypoth-esized link from status to performance runs in theopposite direction, with variation on status simply anoutcome of demonstrated ability (Gould 2002).Archival studies typically address this problem

within a regression framework, by including a largenumber of controls for quality, power, ability, or otherfactors that may have a direct impact on performanceand asking whether measures of status remain sta-tistically significant. Many of these control variablesare clever and reflect careful thinking about the test-ing context. For instance, in estimating the effect ofproducer status on the price of a bottle of wine,Benjamin and Podolny (1999) control for bottle qualityusing scores from blind taste tests. However, if statusand quality are highly correlated, this approach willalways be sensitive to untestable assumptions aboutmeasurement and functional form.Concerns over unobserved quality should be most

salient in situations of great uncertainty—preciselythe situations in which status is hypothesized asmost relevant—because in these settings, the relativecomprehension of the academic researcher is mostdiminished. In other words, in complex settings theacademic researcher is himself a layperson, searchingfor social cues that are correlated with ex ante quality.It remains unknown whether true experts themselvesrely on these social cues or use other correlated indi-cators that were not included in the academic models.Moreover, if one takes the position that both statusand quality are latent constructs, a linear regressionwill produce biased estimates even if one has a per-fect measure of perceived quality, because perceptionsof quality and status are jointly determined.One approach to this dilemma is to directly manip-

ulate the status of a focal actor, holding all else con-stant. As a practical matter, this is often very difficult,particularly in field settings, because it violates mod-ern norms of merit-based prestige. A second approachis to manipulate the signals that a third party mightuse to assess a focal actor’s status. This approachis sometimes feasible, particularly when suppressingcertain signals (e.g., race or gender) would promotemerit-based evaluation (e.g., Blank 1991, Goldin andRouse 2000). In principle, one could isolate the role ofstatus by manipulating a very fine-grained set of sig-nals: Who are an actor’s friends? Where did they goto school? How do they dress? In this study, we focuson author names, as in the Lord Rayleigh example.Names provide a rather coarse signal, which mightconvey a great deal of information about status andother factors (Berger et al. 1980). Thus, although our


experiment will not be able to disentangle all the pos-sibly salient attributes of a name, we are able to mea-sure the causal impact of the identity-based signal.

2.2. Increasing ReturnsMerton (1968) argued that scientific credit is oftenmisallocated, with prominent scientists accruing evergreater recognition, whereas the contributions of less-well-known scientists are undervalued. A variety ofmechanisms might explain why this does not resultin a social system dominated by a single elite actor(Bothner et al. 2010, Gould 2002). Nonetheless, withinthese limiting parameters the implication of the sta-tus and performance literature is clear: greater statusleads to competitive advantages, and hence greaterstatus.We propose that in the literature there are two dis-

tinct mechanisms offered for generating increasingreturns to status. First is direct improvement in thestatus signal. In this conception, each success leadsto better social positioning for the focal actor, inde-pendent of changes in underlying quality. Recognitionafforded to the focal actor becomes a cascade, witheach act of deference increasing the odds that othersimitate that behavior (Rao et al. 2001). For example,Merton (1968) describes how readers approach thework of a preeminent scholar with “special care” andin so doing are apt to get more out of it.The second explanation for increasing returns to

status is improvement in the quality of the focalactor’s products. The source of these endogenousimprovements could be behavioral: Merton (1968)notes that the validation received by a star scientistmay give him the self-confidence required to tackleimportant problems; Phillips and Zuckerman (2001)posit that middle-status actors are constrained bylegitimacy concerns from taking risky actions; andBenjamin and Podolny (1999) find that high-statuswine producers select higher-quality grapes. Endoge-nous improvements in quality may also result fromthe transfer of tangible resources. For instance, aca-demic scientists with prestigious affiliations generallyhave greater resources at their disposal, and Sorensonand Waguespack (2006) find that film distributioncompanies invest more in marketing for films fromhigh-status production teams. A more encompassingview is that actors with better social positions accessmore valuable knowledge, with improved manage-ment and outputs as a consequence (Powell et al.1996, Burt 2004). In this respect, the greater atten-tion directed toward high-status actors may serve toincrease intrinsic quality and not just enhance per-ceived quality.The improved signal explanation and improved

underlying quality explanation for increasing returnsto status are not mutually exclusive—both can oper-ate at the same time. However, one can seek evidence

of the increased underlying quality story by lookingfor differences in the behaviors or resources of high-status actors, particularly before there is any evalua-tion of quality. We implement this idea by examininghow the presence of a status signal influences, prior tothe screening decision, the allocation of attention fromthird parties and the efforts of the focal actors them-selves. Of course, the same concerns with unobservedquality, reverse causation, and simultaneity that hin-der empirical tests of the status-performance relation-ship also apply to these intermediate outcomes. In thefollowing sections we describe an empirical strategythat relies on random variation in status signals toaddress these concerns.

3. Organizational Setting3.1. The IETF Publication ProcessThe IETF is a voluntary nonprofit organization thatcreates and maintains compatibility standards used torun the Internet. Active participants are mostly engi-neers and computer scientists, representing a widevariety of academic, not-for-profit, and commercialorganizations (Simcoe 2008, Waguespack and Fleming2009). IETF standards are used to accomplish a widevariety of core networking functions, such as assign-ing Internet protocol addresses, routing packets, andencrypting data. Firms that adhere to these protocolscan reasonably expect their products to work with therest of the Internet.The IETF creates new standards and other practi-

cal networking knowledge using a bottom-up processof open publication and community review. Figure 1provides an overview. Anyone may propose a newidea to the IETF by creating an appropriately for-matted document and submitting it as an Internetdraft (ID). Each new ID is posted to a public Webserver, where it remains for a period of six months.New IDs are debated and discussed on a series ofe-mail listservs maintained by the IETF and its var-ious working groups and at IETF plenary meetings,which are held three times per year.The IETF distinguishes between two types of IDs:

working group (WG) drafts and individual submis-sions. WG drafts are often “commissioned” as part ofa broader technical agenda that is very likely to pro-duce a standard.2 Although individual submissionsmay become WG drafts, or lead to the formation of anew WG, it is not very common.Within six months of submission, one of three

things will happen to an ID. First, the authors maydecide to revise their proposal, usually in responseto comments or concerns from the IETF community.

2 In our data, working group IDs have a 43% publication rate, com-pared to 7% for individual submissions.


Figure 1 The IETF Publication Process

Internet drafts Request for Comments (RFCs)

IESG/“Last call”

Workinggroup

Standards track

Individual RFC editor

Nonstandards track

Submitting a revision restarts the ID’s six-month pub-lication clock.3 Second, an ID may get published as aRequest for Comments (RFC). And third, if an ID isnot published and its authors do not submit a revi-sion, the draft will expire. Expired drafts are removedfrom the public ID repository.IDs can follow two routes to publication. One pos-

sibility is to become the “consensus” recommendationof an IETF working group. If a WG draft has sup-port from a solid majority of WG participants, the WGchair will forward the ID to the Internet EngineeringSteering Group (IESG), a group of long-time IETF par-ticipants that act as a de facto editorial review board.For all WG drafts, the IESG issues a “last call” forcomments to the entire IETF. If the last call raises seri-ous issues, the ID is sent back to the WG for furtherconsideration. Otherwise, it is published.The second way for an ID to become an RFC is

through the “independent submission” process. Inthis case, authors submit their ID to the RFC editorand ask that it be published as an individual (non-WG) RFC. The RFC editor typically sends these draftsout to subject matter experts for review. If the review-ers and RFC editor agree that an individual ID istechnically sound and of general interest to the IETFcommunity, it is sent to the IESG to ensure that it doesnot conflict with any WG drafts. If the IESG approves,the draft is published as an RFC.On average, working group RFCs are more sig-

nificant than independent submissions. For example,only WG drafts can be formally designated as “IETFstandards.” Individual submissions are published as“nonstandards-track” RFCs, and Simcoe (2008) findsthat this leads to a less contentious review pro-cess. Nevertheless, individual RFCs can be influential.

3 Within the IETF, each revision is called an ID. For exposition, wecall the entire series of linked publications an ID and use the term“revision” to denote consecutive submissions.

They often propose new uses for IETF technology,describe lessons learned from implementation, andpose new problems for IETF members to work on.

3.2. IETF vs. Academic PublishingPerhaps because of the IETF’s quasi-academic roots(Mowery and Simcoe 2002), the IETF publicationprocess is similar to academic publishing in severalrespects. IDs are formatted like academic papers.Manuscripts are authored by individuals. Publishedmanuscripts typically undergo successive rounds ofrevisions, and an editorial board makes the final deci-sion on whether to publish. However, there are impor-tant differences between academic publishing andopen standards development at the IETF.First, while academic publishing typically uses dou-

ble blind reviewing, the IETF has an open review pro-cess, where all manuscripts are available online, andany individual may choose to offer feedback to theauthors. IETF authors are aware of the identity of oth-ers choosing to support, ignore, or criticize their work.Similar repositories of academic working papers, suchas the Social Science Research Network, are increas-ingly prominent in academic publishing, but do notthemselves make determinations about which sub-missions are suitable for publication. For our research,one major advantage of the IETF’s open review pro-cess is that outsiders can observe failed submissions,and not just successful publications.A second difference between the IETF and academic

publishing involves the disposal of manuscripts. Sub-missions to academic journals are accepted, rejected,invited to revise and resubmit, or withdrawn. At theIETF, the termination point for failed submissions ismurky. According to IETF insiders, submissions rarelyreceive an “official” rejection. Although authors maybe strongly discouraged from pursuing a particular


approach or idea, they always retain the right to con-tinue revising an ID.4

Finally, the incentives and reward structure for IETFpublishing is less clear than for academic publishing.Whereas academics and career scientists must “pub-lish or perish,” IETF contributors are typically soft-ware engineers employed by firms. These engineersmay contribute IDs as a way to advance their careersor as part of their job when firms have a vestedinterest in particular standards. However, we haveno direct evidence on how RFC publication influ-ences individual outcomes, such as compensation orcareer mobility. Moreover, where academic publishingis usually characterized by a clear hierarchy amongjournals within a field, all contributions to the IETFare published as RFCs.5

3.3. The et al. ExperimentEvery ID submitted to the IETF is posted to a publiclyaccessible Web page. With the initial posting, and foreach subsequent revision, an e-mail announcement issent to the “ietf-announce” listserv. The top panel inFigure 2 shows a typical ietf-announce message, con-taining a title, list of authors, file name (which containsinformation aboutWG affiliation and revision history),date, and abstract. The bottom panel in Figure 2 showshow the same information is presented in the actualInternet draft: author names and affiliations appear onthe front page, and detailed contact information is typ-ically available at the back of the document.Initially, every message sent to the ietf-announce

listserv included the name of every ID author. How-ever, beginning in late 1999, some announcementsreplaced individual names with the generic label et al.or dropped them entirely. Figure 3 provides twoexamples.Our empirical strategy exploits the fact that miss-

ing names sometimes belong to prominent membersof the IETF community. In that case, et al. removes astatus signal from the ietf-announce message. Becauseinterested readers could still find a complete list ofnames within the ID, the situation is akin to double-blind refereeing in the age of working papers andGoogle Scholar: et al. introduces a certain amountof ambiguity but does not make it especially hardto obtain the relevant information. In that sense,our paper is related to a literature on the effects of“salience” or attention costs (e.g., DellaVigna 2008,Esteves-Sorenson 2009).

4 It is not unusual for an ID to go through 5 (or even 10) revisionsbefore publication as an RFC. In a few cases, IDs will go through30 or more revisions.5 The IETF does distinguish between standards-track andnonstandards-track RFCs, and among standards with different lev-els of maturity. These distinctions may play a role similar to that ofthe journal hierarchy.

Figure 2 Author Names in ietf-Announce and Internet Drafts

A New Internet-Draft is available from the on-line Internet-Drafts directories.

Title : Basic Network Mobility SupportAuthor(s) : R. Wakikawa, K. Uehara, K. Mitsuya, T. ErnstFilename : draft-wakikawa-nemo-basic-00.txtPages : 21Date : 2003-2-18

This draft proposes a solution for Basic Network Support. It proposesMobile IPv6 extensions as advocated by the NEMO working group.Our solution differs from Prefix Scope Binding Update…

INTERNET DRAFT Ryuji Wakikawa18 Feb 2003 Keisuke Uehara

Koshiro MitsuyaThierry ErnstKeio University and WIDE

Basic Network Mobility Supportdraft-wakikawa-nemo-basic-00.txt

… (BODY TEXT) …

Authors’ Addresses

Ryuji Wakikawa Koshiro MitsuyaKeio University and WIDE Keio University and WIDE5322 Endo Fujisawa Kanagawa 5322 Endo Fujisawa Kanagawa252-8520 JAPAN 252-8520 JAPANPhone: +81-466-49-1394 Phone: +81-466-49-1394Fax: +81-466-49-1395 Fax: +81-466-49-1395EMail: [email protected] EMail: [email protected]

Keisuke Uehara Thierry ErnstKEIO University and WIDE Keio University and WIDE5322 Endo Fujisawa Kanagawa 5322 Endo Fujisawa Kanagawa252-8520 JAPAN 252-8520 JAPANPhone: +81-466-49-1394 Phone: +81-466-49-1394Fax: +81-466-49-1395 Fax: +81-466-49-1395EMail: [email protected] EMail: [email protected]

Source. Reprinted with permission from the Internet Society. © The InternetSociety (2003). All rights reserved.Notes. The top panel shows a typical ietf-announce message listing all authornames. The bottom panel shows the front page and authors’ addresses asthey appear in the Internet draft.

The practice of using et al. was introduced by theIETF Secretariat, an administrative body that managesthe logistics of the ID publication process, to addressa rapidly growing volume of submissions (see Fig-ure 4). Once et al. was allowed, the decision to useit for an individual ID was left to clerical staff, whowould process incoming IDs by typing the relevantinformation into a standardized form. These individ-uals suggested that they tried to include every name,but would often resort to et al. when things becamebusy—typically during the period just before IETFmeetings, when there would be a spike in new pro-posals (Fuller 2006).6

6 For revisions, the clerical staff would typically cut and paste theoriginal message into a new form. Thus, when et al. appears on themessage for an initial ID submission, it almost always remains forthe entire life of the proposal.


Figure 3 Et al. and Missing Author Names


Title : Middlebox Communications (MIDCOM) Protocol Managed Objects

Author(s) : M. Barnes et al.Filename : draft-barnes-midcom-mib-01.txtPages : 16Date : 2003-7-1

This document describes and defines the managed objects for dynamicconfiguration of middleboxes. The scope of the middleboxes to whichthese managed objects apply is limited to NATs and Firewalls…

Internet Draft M. BarnesDocument: draft-barnes-midcom-mib-01.txt Nortel Networks

Wes HardakerSpartaD. HarringtonEnterasys NetworksM. Stiemerling

Category: Standards Track NEC Europe Ltd.Expires: December 2003 June 2003

Middlebox Communications (MIDCOM) Protocol Managed Objects

… (BODY TEXT)…

Authors’ Address

Mary BarnesNortel Networks2380 Performance DriveRichardson, TX 75082USAPhone: 1-972-684-5432Email: [email protected]

…..

David Harrington, Co-chair SNMPv3 WGEnterasys Networks

….


Title : End-to-End VoIP over MPLS Header CompressionAuthor(s) : J. Ash, B. GoodeFilename : draft-ash-e2e-vompls-hdr-compress-01.txtPages : 0Date : 2003-3-6

VoIP over MPLS typically uses the encapsulation voice/RTP/UDP/IP/MPLS.For an MPLS VPN, the packet header is at least 48 bytes, while the voicepayload is typically no more than 30 bytes. VoIP over MPLS header…

Network Working Group Jerry AshInternet Draft Bur Goode<draft-ash-e2e-vompls-hdr-compress-01.txt> Jim HandExpiration Date: October 2003 AT&T

George SwallowCisco Systems, Inc.

March, 2003

End-to-End VoIP over MPLS Header Compression

… (BODY TEXT)…

7. Authors’ Addresses

Jerry AshAT&TRoom MT D5-2A01200 Laurel AvenueMiddletown, NJ 07748, USAPhone: +1732-420-4578Email: [email protected]

….

George SwallowCisco Systems, Inc.250 Apollo Drive Chelmsford, MA 01824Phone: +1 978 497 8143Email: [email protected]…

Source. Reprinted with permission from the Internet Society. © The Internet Society (2003). All rights reserved.Notes. The example on the left shows an ietf-announce message where use of et al. obscures three author names: W. Hardaker, D. Harrington, andM. Stiemerling. The example on the right shows an ietf-announce message that omits two author names: J. Hand and G. Swallow.

Figures 5 and 6 support the IETF Secretariat’s expla-nation of et al. usage. Specifically, Figure 5 shows asmoothed estimate of the probability that a draft withthree or more authors receives an et al., where thevertical bars correspond to IETF meeting dates. There

Figure 4 New Internet Draft Submissions

0

500

1,000

1,500

New

sub

mis

sion

s

1996 1997 1998 1999 2000 2001 2002 2003 2004

Individual

Working group

Year

is a very strong cyclical pattern, with periodic spikesjust before each meeting. Figure 6 shows that meeting-related deadlines also created large swings in the totalvolume of submissions over relatively short time peri-ods. There is also a longer time trend, with et al. firstappearing in 1999 but used sparingly until a sharpincrease in early 2001.The key assumption behind our empirical tests is

that the IETF Secretariat’s use of et al. is uncorre-lated with the identity of individual authors. Impor-tantly, this assumption is falsifiable. In particular, wecan examine the correlation between et al. and allauthor-level explanatory variables in our analysis. Ifthe author-level variables and other control variablesare not correlated with et al., we should feel more con-fident in the maintained assumption that unobservedauthor characteristics (which, by definition, appear inthe regression error-term) are also uncorrelated withthe treatment condition.To conduct this test of our central empirical as-

sumption, we estimate a series of logit regressionsfor IDs with two or more authors submitted from


Figure 5 Probability of et al.

0

0.2

0.4

0.6

0.8

1.0

Et a

l. pr

obab

ility

Jul98 Jan00 Jul01 Jan03 Jul04

Notes. Fitted values from LOWESS (locally weighted scatterplot smoothing)regression; sample includes IDs with three or more authors. Vertical linesindicate IETF meeting dates.

Figure 6 Et al. and ID Submission Rates

0

0.2

0.4

0.6

0.8

1.0

Et a

l. pr

obab

ility

0

0.0005

0.0010

0.0015

Den

sity

Jul98 Jan00 Jul01 Jan03 Jul04

ID submissionsEt al. probability

Note. Fitted values from LOWESS (locally weighted scatterplot smoothing)regressions; sample includes IDs with three or more authors.

2001 to 2003 (see appendix Table A.1). In each ofthese models, proximity to an IETF meeting, the sub-mission date, and the number of authors on the IDwere extremely significant. We found no evidence ofa correlation between WG chair authorship and et al.usage. Of the other variables, only in column 4 ofTable A.1 do we find a weakly significant relation-ship between Prior RFCs dummies and et al. usage.However, in all cases we could not reject the joint nullhypothesis that all observable author characteristicshad no impact on the probability of receiving an et al.

4. Data and MethodsWe collected data on all Internet drafts submitted tothe IETF between 1992 and 2004 from the websitehttp://www.watersprings.org and from the archivedietf-announce mailing list. The estimation sampleincludes all IDs submitted between 2001, the year in

Table 1 Variable Definitions

Variable name Definition

Published as RFC Indicator: ID is published as an RFCListservs Count of listservs where ID is mentioned at least once

(excluding e-mails listing 10 or more IDs)Messages Weighted count of e-mail messages on all IETF

listservs that mention focal IDRevisions Number of times ID is submitted to IETFRevised Indicator: Revisions> 1 or initial ID published as RFC

Et Al Dummy Indicator: One or more authors not listedin ietf-announce

WG Chair Author Indicator: At least one author is apast or current WG chair

Unlisted WG Chair Indicator: WG Chair Author= 1 andnot listed in ietf-announce

Prior RFC Count Sum of RFCs previously published by all ID authorsPrior RFCs �a� b� Indicators: Prior RFC Count lies in the interval �a� b�

Authors Total number of authors listed on draftLog Pages Log of number of pages in IDIntl Author Indicator: At least one international authorMultisponsor Indicator: Authors affiliated with

two or more organizationsIETF Meeting Indicators: The date of the next meetingDays-to-Meeting Log count of days until next IETF meetingAuthor E-mails Count of e-mail messages sent by an ID author

which et al. use became widespread, and 2003.7 Wealso exclude a small number of proposals with morethan five listed authors, though including them doesnot alter our main results.8

For each ID, we use public sources to obtain thesubmission date, a complete list of author names(from the ID text), the number of revisions, andan outcome: expiration (failure) or publication as anRFC. For each author, we have a complete list ofsubmissions to the IETF, his or her place in the listof draft authors, an e-mail address, and informationabout whether the author ever served as a WG chair.Table 1 provides a list of variables and definitions,and Table 2 provides summary statistics for the 3,129individual IDs and 1,046 working group IDs in ourestimation sample.

7 We drop IDs from 2004 because in that year the IETF adopted aset of procedures for using et al., and we want to avoid trunca-tion of the dependent variable. We drop IDs from 2000 because theet al. treatment rate is markedly lower in that year, and becausethe tests in appendix Table A.1 showed a statistically significantrelationship between Prior RFC counts and the use of et al. inthat year. Specifically, we found that in 2000 (but not after thatpoint), submissions from authors with five or more prior RFCs wereless likely to receive an et al. However, we could never reject thehypothesis that all author-level observables were jointly insignifi-cant, and results that include all proposals from 2000 to 2003 arequite similar to those reported below. We appreciate the effortsof an anonymous Management Science reviewer in pushing us tounderstand this pattern.8 Ninety-four percent of individual submissions and 89% of work-ing group IDs had five authors or fewer.


Table 2 Summary Statistics

Individual IDs Working group IDs

Variable Mean SD Obs. Mean SD Obs.

Published as RFC 0�069 0�254 3�129 0�431 0�495 1�046Listservs 0�957 0�944 3�129 1�194 0�896 1�046Messages 3�448 7�141 3�129 4�913 8�269 1�046Revisions 2�115 2�097 3�129 4�624 3�504 1�046Revised 0�445 0�497 3�129 0�807 0�395 1�046

Et Al Dummy 0�248 0�432 3�129 0�280 0�449 1�046WG Chair Author 0�258 0�438 3�129 0�459 0�499 1�046Unlisted WG Chair 0�042 0�200 3�129 0�076 0�264 1�046Prior RFC Count 3�594 8�171 3�129 7�712 12�885 1�046Prior RFCs �0� 0�577 0�494 3�129 0�303 0�460 1�046Prior RFCs �1� 0�095 0�293 3�129 0�119 0�323 1�046Prior RFCs �2� 0�066 0�247 3�129 0�080 0�272 1�046Prior RFCs �3�4� 0�061 0�240 3�129 0�104 0�306 1�046Prior RFCs �5�10� 0�082 0�274 3�129 0�162 0�368 1�046Prior RFCs �10+� 0�120 0�325 3�129 0�232 0�423 1�046

Authors 2�022 1�178 3�129 2�247 1�284 1�046Log Pages 2�310 0�983 3�129 2�555 1�049 1�046Intl Author 0�416 0�493 3�129 0�349 0�477 1�046Multisponsor 0�314 0�464 3�129 0�509 0�500 1�046Days-to-Meeting 3�740 0�620 3�129 3�755 0�637 1�046Author E-mails 1�113 2�674 3�129 1�249 3�162 1�046

Note. Sample: Nonadministrative IDs submitted from 2001 through 2003with one to five authors.

4.1. MeasurementThe top half of Table 2 focuses on outcome measures.Our primary outcome is the indicator variable Pub-lished as RFC. Unconditional publication rates are 6.9%for individual IDs and 43.1% for working group IDs.We also examine several intermediate attention-

related outcomes. Listservs and Messages are bothderived from an automated search through a reposi-tory of 720,000 electronic messages posted to 382 dis-tinct IETF hosted listservs (i.e., e-mail messageboards) for text strings that match an ID’s uniquefile name. We exclude messages that are administra-tive or computer generated by discarding those thatcontained the file name of 10 or more unique IDs.Listservs is a count of the electronic bulletin boardswhere an ID is mentioned. Messages is a weightedcount of the number of messages mentioning the ID,where weights are inversely proportional to the num-ber of unique ID file names contained in the e-mailmessage. The sample means in Table 2 show thatworking group IDs receive more attention than indi-vidual submissions. In particular, working group IDsare mentioned on 20% more Listservs and in 45% moreMessages.Our last intermediate outcome is Revised, a dummy

variable that equals 1 if the first draft of an ID ispublished as an RFC or if a revision of that ID issubmitted within six months.9 Although we cannot

9 Few drafts are published with no revisions (less than 1% of allindividual submissions and roughly 7% of the working group IDs).

determine precisely why an ID is abandoned, weinterpret Revised as a measure of the authors’ per-sistence at a point in the review process where theymay be especially sensitive to attention. In particu-lar, authors may view individual submissions as trialballoons that they are prepared to abandon quickly ifthere is little feedback from the broader community.10

The remaining rows of Table 2 provide summarystatistics for explanatory and control variables. Theindicator variable Et Al Dummy equals 1 if the ietf-announce message for an ID either lists some authorsas et al., or simply omits their names.11 Twenty-fivepercent of individual submissions and 28% of WGdrafts receive an et al.Our measure of status is the dummy variable WG

Chair Author, which equals 1 if an ID author has everserved as a WG chair. When an ID has a WG chairauthor, but the name of that author is not visibleon the ietf-announce message—so there is no statussignal—we set the dummy variable Unlisted WG Chairto 1. Note that the mean of Unlisted WG Chair (4.2%) issmaller than the product of Et Al Dummy (24.8%) andWG Chair Author (25.8%), since some ietf-announcemessages may contain an et al. that only obscures thename of nonchair authors. Given the small incidenceof Unlisted WG Chair, our main results rely heavilyon outcomes in a subset of 130 proposals where thename of a WG chair was obscured or omitted in theindividual ID sample, and 79 cases with an obscuredWG chair in the working group ID sample.As a proxy for status, WG Chair Author has

strengths and weaknesses. The main strength is thatit is easy to interpret and almost certainly correlatedwith “true” status within the IETF community. Tobecome a chair, individuals must be put forward bya WG and approved by an area director who sits onthe IESG. A chair’s job combines elements of journaleditor and parliamentarian.12 Specifically, this indi-vidual manages the publication process for all IDsassociated with a WG and decides when the group

Nor is a single revision typically sufficient; for published IDs themedian number of revisions is five for individual submissions andsix for WG submissions.10 A previous version of the paper focused on the total number ofrevisions, as opposed to the decision to revise the initial submis-sion. The results are qualitatively similar.11 In robustness tests, we find little difference between omittingnames and obscuring them with et al.12 Though WG chairs cannot submit drafts to their own group,the political nature of their position highlights the role of name-removal in our analysis. Although the variable WG Chair Authorwill capture any effect of high-status individuals exercising politicalpower (as well as unobserved quality, or other differences betweenchairs and nonchairs), there is little reason to think that politicalinfluence is hindered by receiving an et al.: if anything, we expectthat chicanery is aided by secrecy.


has reached consensus on substantive issues. Chairshave high visibility within the IETF, and most havedone some piece of work that is viewed as significant.The main weakness of WG Chair Author is that it isa very crude measure of status. As a dummy vari-able, it is not capable of capturing subtle variationsin the structure of the IETF’s social hierarchy. Stud-ies that attempt to measure status effects directly mayrequire more refined measures, such as those basedon positioning within a social network. Because weare measuring a signaling effect, and not the impactof a change in status, a single crude but salient proxyfor individual reputation is arguably more suitable.Our main control for author experience or reputa-

tion is a series of categorical variables indicating thenumber of Prior RFCs published by all authors whocontribute to the ID. Although there is no mechan-ical link between publishing RFCs and becoming aWG chair, Fleming and Waguespack (2007) show thatthere is a strong correlation between the two mea-sures, and we find a similar relationship in our dataat the ID level (see appendix Table A.2). On pro-posals where the authors collectively have zero PriorRFCs, roughly 6% of manuscripts possess a WG Chairauthor, whereas for proposals where the authors have10 or more Prior RFCs the WG chair rate rises to 85%.This extremely strong correlation highlights a limita-tion of our analysis. Although the et al. experimentcan be used to isolate the effects of name-based signal-ing, or what Merton (1968) called the Matthew Effectin the communication system, it will not reveal howreaders interpret these signals. Thus, a high-statusname may signal the expected quality of the proposal,the future importance of the idea (or technology), thepolitical clout of the author, or the likelihood that oth-ers will pay attention to the ID.The remaining rows in Table 2 provide descriptive

statistics for additional control variables. An averageID in our estimation sample has between two andthree authors (Authors) and is roughly 9–11 pages long(Log Pages). The other controls include a dummy forauthors from outside the United States (Intl Author)and a dummy for IDs whose authors have more thanone primary affiliation (Multisponsor). IETF Meetingcaptures longer term time trends via a set of dummyvariables indicating the date of the next meeting. Weuse Days-to-Meeting to control for “congestion effects”around meeting dates, as described by the IETF cleri-cal staff. Finally, Author E-mails measures the numberof e-mail messages discussing an ID sent by an IDauthor and is used in the attention analysis to controlfor the authors’ own efforts in prompting discussionof a draft.

4.2. MethodsOur statistical methods examine the change in pub-lication rates when a high-status author’s name is

removed from the ietf-announce message advertis-ing a new ID. A simple way to measure the impactof status signals would be to ignore IDs from low-status authors and to use a simple t-test to com-pare the average publication rates when a WG chair’sname is obscured and when it is not. However, thissimple approach would overestimate the signalingeffect if other factors produce a negative correlationbetween et al. and publication. We are particularlyconcerned that meeting-related “congestion effects”(see Figures 5 and 6) may cause both an increase inet al. usage and a decline in publication rates. Forexample, proposals submitted during the premeet-ing rush may receive less attention or be less “pol-ished” than comparable proposals submitted at anearlier date.To control for omitted variables correlated with

et al., we compare the change in publication rateswhen et al. obscures high- versus low-status authornames. Put differently, we use IDs where WG ChairAuthor equals 0 to estimate the baseline impact ofhaving an et al. on the ietf-announce message, andsubtract this baseline from the change in publicationrates when et al. obscures a high-status name. In thisapproach, the status-signaling effect is estimated bya difference in differences: the change in publicationrate associated with removing a high-status author’sname minus the the change in publication rate asso-ciated with removing a low-status author’s name.We implement this idea using the following linearregression:

RFCi = Unlisted Chairi�1 +WG Chair Authori�2

+EtAli�3 + �t + Xi� + �i� (1)

where i indexes the Internet drafts in our estima-tion sample; Xi is a vector of control variables; and�t is a vector of triannual IETF meeting dummiesthat control for time trends in the underlying prob-ability of publication. Though our main outcome isbounded at 0 and 1, we estimate a linear probabil-ity model for two reasons. First, it is easy to inter-pret coefficients as a change in probabilities. Second,and perhaps more important, in nonlinear models, theinteraction term (�1) will not equal the cross-partialderivative of the expected probability of publication(Ai and Norton 2003).In Equation (1), �1 measures the signaling effect.

The change in publication probability associated withhaving a high-status author (�2) may reflect a varietyof factors, such as latent proposal quality or authorexperience. Ideally, we would let �1 vary accordingto the position of the WG chair in the list of authors,because an author’s place may be determined by con-tribution to the ID, which may in turn influence qual-ity. In practice, we do not have enough data to gener-ate precise estimates in such a model, though we do


explore this extension in the appendix (see appendixTable A.3).After estimating the impact of name-based signals

on publication outcomes, we use the same model,based on Equation (1), to examine several inter-mediate outcomes, which we interpret as measuresof attention and its impact on authors’ persistence.Given the rapid increase in total proposals (see Fig-ure 4), attention from the IETF community is likelyto have been a scarce resource during our sampleperiod. A draft that receives very little attention mightbe abandoned for several reasons. For instance, atten-tion is needed for a WG to become interested in aproposal; authors may interpret lack of interest as anassessment of quality; and the quality of drafts thatdo receive early feedback may improve as a result.Conditional on having an interesting idea, there maybe a large difference between receiving a little atten-tion or none at all, thus helping to explain how the“weak treatment” of removing names from an e-mailannouncement, but not the actual submission, canproduce a measurable impact on publication rates.

5. ResultsTables 3 and 4 present our main results. Table 3 showsthat identity-based signals influence publication out-comes for individual submissions, but not workinggroup IDs. Table 4 focuses on intermediate outcomes,and suggests that status signals influence publicationoutcomes for individual IDs by drawing more atten-tion to proposals during the review process.

5.1. Identity as a SignalTable 3 shows that name-based signaling has a signif-icant impact on individual ID publication rates, butno measurable impact on the publication of workinggroup IDs. All of the results are based on the linearprobability model described by Equation (1).The first column in Table 3 focuses on the sample of

individual IDs and presents results from our simplestspecification, which includes a constant, WG ChairAuthor, Et Al Dummy, and the key status-signalingvariable Unlisted Chair.13 The results show that indi-vidual IDs are 9.4% more likely to be published whenthe name of a WG chair appears on the ietf-announcemessage. However, this chair-author effect declinesby 7.2 percentage points when the author’s nameis obscured or omitted. Thus, name-based signalingaccounts for roughly 77% of the total benefits of WGchair authorship. Both the main effect of having achair author and the status-signaling effect are largein comparison to the 5.3% baseline publication rate

13 In this specification, there is no possibility that predicted valuesfall outside the unit interval, even for a linear probability model.

Table 3 Identity as a Signal

Linear probability models of ID publicationUnit of observation= Internet draft

Dependent variable= Published as RFC

Sample: Individual IDs Working group IDs

WG Chair Author 0�094∗∗ 0�054∗∗ 0�086∗∗ 0�029�0�01� �0�02� �0�03� �0�04�

Unlisted WG Chair −0�072∗∗ −0�070∗∗ −0�012 0�008�0�03� �0�03� �0�07� �0�07�

Et Al Dummy −0�021∗ −0�007 0�055 0�069�0�01� �0�02� �0�04� �0�05�

Prior RFCs [1] 0�043∗ 0�053�0�02� �0�05�

Prior RFCs [2] 0�010 0�021�0�02� �0�06�

Prior RFCs �3�4� 0�058∗ 0�150�0�03� �0�06�∗

Prior RFCs �5�10� 0�057∗ 0�022�0�02� �0�05�

Prior RFCs �10+� 0�073∗∗ 0�085�0�02� �0�06�

Log Pages 0�005 0�021�0�00� �0�01�

Intl Author −0�014 −0�036�0�01� �0�03�

Multisponsor 0�004 0�067�0�01� �0�06�

Days-to-Meeting 0�045∗∗ 0�051+

�0�01� �0�03�Constant 0�053∗∗ −0�114∗∗ 0�377∗∗ 0�056

�0�01� �0�03� �0�02� �0�11�

Author count effects N Y N YIETF meeting effects N Y N Y

Observations 3,129 3,129 1,046 1,046R-squared 0�025 0�052 0�010 0�036Mean of dependent variable 0�069 0�069 0�431 0�431

Notes. Appendix Table A.3 presents robustness checks, including a modelthat controls for Unlisted Chair’s position in list of authors. Robust standarderrors are in parentheses.

+10% significance; ∗5% significance; ∗∗1% significance.

for IDs with no chair and no et al. Finally, our esti-mates show that receiving an et al. is associated with a2.1% drop in publication rates for IDs without a chairauthor. We interpret the negative main effect of et al.as the impact of congestion, produced by the largeinflux of new proposals over the entire sample period,and particularly just before IETF meetings. This nega-tive coefficient on Et Al Dummy implies that includingthe low-status control sample reduces our estimate ofthe signaling effect.The second model in Table 3 adds a variety of

controls to our baseline specification, including acomplete set of meeting-date and author-count fixedeffects. If the et al. process is truly uncorrelatedwith observable features of the ID, this should haveno impact on our estimates of the signaling effect.


Table 4 Identity and Attention

OLS regressionsUnit of observation= Internet draft


Dependent variable: Listservs Messages Revised Listservs Messages Revised

WG Chair Author 0�263∗∗ 0�953∗∗ 0�106∗∗ 0�055 −0�191 0�060∗

�0�04� �0�29� �0�02� �0�06� �0�41� �0�03�Unlisted WG Chair −0�192∗ −1�285∗∗ −0�106∗ −0�135 0�790 −0�000

�0�09� �0�46� �0�05� �0�13� �1�22� �0�05�Et Al Dummy 0�111∗ −0�671+ −0�071∗ 0�006 0�311 −0�038

�0�06� �0�40� �0�03� �0�09� �0�85� �0�04�Log Pages 0�034∗ 0�118 0�020∗ −0�014 0�318 0�037∗∗

�0�01� �0�08� �0�01� �0�02� �0�23� �0�01�Intl Author 0�052 −0�078 0�035+ 0�024 0�253 0�020

�0�03� �0�23� �0�02� �0�06� �0�48� �0�03�Multisponsor −0�028 0�089 0�072∗∗ −0�179 −0�714 −0�016

�0�04� �0�30� �0�03� �0�11� �1�10� �0�04�Days-to-Meeting −0�065∗∗ 0�060 0�049∗∗ −0�127∗∗ 0�505 −0�010

�0�03� �0�17� �0�02� �0�04� �0�31� �0�02�Author E-mails 0�157∗∗ 1�641∗∗ −0�002 0�093∗∗ 1�703∗∗ 0�005

�0�01� �0�13� �0�00� �0�02� �0�19� �0�00�Constant 0�722∗∗ 0�451 0�175∗∗ 1�526∗∗ −0�654 0�607∗∗

�0�11� �0�70� �0�07� �0�17� �1�42� �0�09�

Author count effects Y Y Y Y Y YIETF meeting effects Y Y Y Y Y Y

Observations 3,129 3,129 3,129 1,046 1,046 1,046R-squared 0�211 0�358 0�021 0�116 0�356 0�058Mean of dependent variable 0�957 3�415 0�445 1�194 4�859 0�807

Notes. Appendix Table A.4 presents robustness checks. Robust standard errors are in parentheses.+10% significance; ∗5% significance; ∗∗1% significance.

And indeed, the coefficient on Unlisted WG Chairremains negative, statistically significant, and withinthe 95% confidence interval for our previous esti-mate. The main effect of WG Chair Author dropsby roughly 42%, due to the inclusion of the highlycorrelated Prior RFCs variables. The additional con-trol variables in column 2 are jointly significant, andthe model R-squared is more than twice that of ourfirst model. The main effect of Et Al Dummy dropsto zero once Days-to-Meeting is introduced to controlfor any congestion effects. We find that having aninternational author is associated with a small declinein publication rates. However, neither proposal size(Log Pages) nor author diversity (Multisponsor) has ameasurable impact on outcomes.Because the first two columns of Table 3 report our

main finding, we subject them to a variety of robust-ness checks (see appendix Table A.3). First, we showthat similar results can be obtained as the marginaleffects of a logit specification. We then show thatit makes no difference whether a chair’s name isobscured by et al. or simply omitted from the ietf-announce message. As a third robustness check, weallow the Unlisted Chair coefficient to vary accordingto a chair’s position in the list of authors. That model

strongly rejects the hypothesis that all the unlistedchair coefficients are 0 (p = 0�00), which is reassuring,because one might worry about a correlation betweenUnlisted Chair and latent quality if authors are onlyobscured when they make a smaller contribution tothe proposal. Next, we show that our main results arerobust to dropping single authored drafts from theestimation sample. Finally, we show that the magni-tude of the signaling effect is unchanged if we restrictthe sample to IDs with two authors and no chair inthe lead author position.Returning to Table 3, the last two models examine

the impact of status signals in the sample ofWG drafts,where there is much less uncertainty about proposalquality and a higher chance of publication. Column 3uses the very simple specification used in the first col-umn. Although we find a comparable effect for WGChair Author, the point estimate on the status-signalingparameter is one-sixth that of the individual IDs andstatistically indistinguishable from 0. This is consis-tent with the idea that the impact of status signals ismore pronounced in environments characterized byhigh uncertainty. The results in column 3 also make animportant methodological point. In particular, there isa large, positive, and statistically significant coefficienton our proxy for status, even though we conclude that


there is in fact no status signaling effect. The resultsin column 4 show a similar pattern after introducing afull set of controls. Our point estimate of the status sig-naling remains very close to 0. As with individual IDs,the main effect ofWG Chair Author drops substantiallywhen Prior RFCs are introduced, though the impact ofWG Chair Author and Prior RFCs remain jointly signifi-cant at the 10% level (p = 0�08).We draw two main lessons from Table 3. First,

based on the individual ID results, we conclude thatidentity-based signaling does play a role in the IETFpublication process. Remarkably, our estimates sug-gest that signaling explains roughly three-quarters ofthe total benefit from having a high-status authoron a proposal. Second, comparing the individual IDresults with the those in the WG sample, we concludethat this signaling effect is absent when proposals areviewed as more important and thus have a higherex ante chance of publication. This second lesson sug-gests an interesting twist on the widely held view thatstatus signals are more salient in an environment withhigh uncertainty. In particular, the level of uncertaintymay not be an exogenous factor, but rather a func-tion of how much attention individual IETF partici-pants, and the entire community, pay to a particularset of initiatives. Thus, differences in ex ante uncer-tainty will often reflect collective judgements and theexpected costs of applying alternative search mech-anisms or screening heuristics to learn about moresalient attributes of the actor (or idea) in question.Our next set of results explores this hypothesis moresystematically by focusing on proxies for attention asa intermediate input in the publication process.

5.2. Status and AttentionTable 4 uses the same empirical model, based onthe et al. experiment, to examine a set of interme-diate outcomes—specifically Listservs, Messages, andRevised. We interpret these measures as indicators ofthe amount of attention that an ID receives after itis released. As in Table 3, Table 4 presents parallelanalyses of the individual ID and working group IDsamples.14

The first column in Table 4 focuses on Listservs dis-cussing an Individual ID. The main effect of WG ChairAuthor is highly significant. The average individual IDis mentioned on roughly 0.96 listservs, and a proposalsubmitted by a WG chair author is mentioned on anadditional 0.26 boards. The coefficient on Unlisted WGChair is statistically significant and accounts for nearly

14 Though we present ordinary least squares (OLS) results and omitthe Prior RFC indicators for easy interpretation of the interactioncoefficient, appendix Table A.4 shows that we obtain very similarresults from Poisson QML (quasi-maximum likelihood) models (i.e.,Poisson with robust standard errors to account for overdispersion)and robust logit models that include a full set of Prior RFC effects.

three-quarters of the main WG chair effect. The secondcolumn in Table 4 repeat this exercise using Messagesas the outcome variable. We find that IDs from high-status authors are mentioned in 0.95 more messagesthan an average ID: roughly a 25% increase. In thismodel, the signaling effect—as measured by the coef-ficient on Unlisted WG Chair—is again large and statis-tically significant at the 1% level. The dependent vari-able in the third column is Revised, which captures theauthors’ decision to resubmit the initial manuscript.This outcome is closely linked to final publication,because most unpublished IDs are abandoned after asmall number of revisions, whereas publication oftenrequires several rounds of editing. Once again, wefind a statistically significant increase for submissionswhere WG Chair Author equals 1 and a negative andstatistically significant coefficient on Unlisted Chair,indicating that the decision to revise is influenced byname-based signaling.The last three columns in Table 4 examine Listservs,

Messages, and Revised for working group IDs. As withthe publication outcome, we find no evidence of anidentity-based status signaling effect among propos-als where there is less uncertainty. Although WG ChairAuthor has a statistically significant effect on Revised,only 19% of the submissions in this sample are aban-doned without revision.Overall, the results in Table 4 indicate that among

individual IDs, high-status authors receive more atten-tion and that much of this effect is caused by plac-ing their names on the ietf-announce message. Thisattention effect is plausible—particularly given thelarge volume of IDs and preliminary nature of manyproposals—yet it is nevertheless surprising that a dif-ference of one or two e-mail lists (or messages) can leadto a substantial divergence in publication probabilities.Because status signals also influence authors’ decisionsto revise their initial submissions, we interpret thesefindings as evidence of strong increasing returns toattention in the early stages of the creative process. Itis unclear whether this is driven by unique features ofthe IETF publication process or is a more general fea-ture of creative work. However, we believe this to bean interesting subject for future research.

6. ConclusionMany authors have written about the importance oflabels and identity. Perhaps themost famous statementof the hypothesis that a name does not (or should not)matter belongs to Shakespeare:

What’s in a name? That which we call a rose By anyother name would smell as sweet; So Romeo would,were he not Romeo called, Retain that dear perfectionwhich he owes Without that title.

(Romeo and Juliet, II, ii, 43–47)

This paper presents evidence that Juliet was wrong,at least within the context of Internet standards devel-


opment. We exploit a unique natural experiment cre-ated by the fact that the use of et al. obscures thenames of some authors who nevertheless contribute toproposals brought before the IETF. We find that whenet al. obscures the name of a high-status author—specifically a current or former IETF working groupchair—there is a significant drop in the publicationrate relative to the case where et al. obscures the nameof a low-status author. Our method, which relies onmanipulating the status signal as opposed to socialposition itself, shows that status effects are present forindividual ID submissions, where there is high ex anteuncertainty and a scarcity of attention, but not forWG submissions, which typically receive a great dealmore scrutiny. The standard approach of controllingfor observable quality finds evidence of status effectsin both cases.These results provide statistical evidence of Mer-

ton’s Matthew Effect in the novel organizational con-text of open standards development. Our estimatessuggest that under appropriate conditions, name-based signaling has a large impact on publication rates,explaining up to three-quarters of the benefits of hav-ing a high-status author on the proposal. This is espe-cially surprising given the “weak” nature of the treat-ment condition: even when the use of et al. obscures anauthor’s name on the e-mail announcement, it is rela-tively easy to learn the identity of all authors by down-loading the relevant proposal.Our paper also explores the role of attention at early

stages of the IETF publication process. In particular,we show that proposals from high-status authors gen-erate more conversation among IETF participants, butonly when their name appears on the announcementmessage. This suggests that increasing returns to atten-tion may be a mechanism that explains how our rela-tively weak treatment condition leads to a substantialchange in publication rates. More generally, it suggestsa source of positive feedback from an actor’s initial sta-tus or position to the underlying quality of his work.This positive feedback loop between status, atten-

tion, and quality reconciles some of the tensionbetween sociological theories that emphasize increas-ing returns to status and economic models of signal-ing or reputation, where signals are only used whenthey convey information about the sender’s under-lying quality. However, our paper does not addressthe important and contentious question of whetherscreening on author names is based on a (possiblycorrect) belief that status is linked to latent qualityor, alternatively, that it reflects the application of adouble standard that favors privileged actors. Sta-tus is rewarded in either case, but Merton (1968)calls the first mechanism “functional” because it ulti-mately selects for better outcomes. Our results sug-gest that name-based screening at the IETF is func-tional because it only happens when search costs are

high and because we expect endogenous improvementin the quality of these manuscripts. However, statusmight also flow from nonfunctional attributes suchas power (Washington and Zajac 2005) or celebrity(Rindova et al. 2005).One approach for distinguishing between func-

tional and nonfunctional screening involves decom-posing measured status into distinct elements thatcorrespond to perceived quality and social stratifi-cation (e.g., Piskorski 2007, Jensen and Roy 2008).Despite its prevalence in the literature, the decom-position approach is problematic in that it relies onstrong assumptions about the validity of particularmeasures and does not address the empirical problemof unobserved quality highlighted in this paper. Analternative approach utilizes the idea that functionaland nonfunctional mechanisms have very differentimplications for long-run performance. In particular,Becker (1993) suggests that nonfunctional screeningwill lead to worse ex post performance within thefavored group, because the application of a doublestandard privileges the lowest-quality actors within afavored group. This idea has been applied to distin-guish between statistical (functional) and taste-based(nonfunctional) discrimination in a variety of settings(e.g., Ayres and Waldfogel 1994, Knowles et al. 2001)and may provide future researchers with some trac-tion on the difficult problem of distinguishing betweenfunctional and nonfunctional status mechanisms.Regardless of whether identity-based signaling is

functional or discriminatory, our results highlight theenduring importance of Merton’s proposition that, “itis important to consider the social mechanisms thatcurb or facilitate the incorporation of would-be con-tributions into the domain of science” (Merton 1968,p. 61) In particular, we show that high-status actorsgather more attention for their ideas and ultimately aremore likely to get them published. This finding has sig-nificant implications for the dissemination of knowl-edge in an era of massive low-cost distribution tech-nology. Over the last 40 years, a remarkable decline inthe costs of accessing knowledge—often based on tech-nologies developed within the IETF—has producedtremendous growth in the volume of accessible ideas.Our findings highlight the role of status and social net-works within this highly competitive intellectual envi-ronment, where attention is increasingly likely to con-stitute a scarce and valuable resource.

AcknowledgmentsThe authors gratefully acknowledge financial support fromthe NET Institute (http://www.netinst.com) and the Kauff-man Foundation. Lee Fleming was a catalyst for this researchand provided valuable input on several early drafts. Theauthors also thank Ajay Agrawal, Jóhanna K. Birnir, ScottBradner, Brent Goldfarb, Ben Hallen, Olav Sorenson, andDavid Tan for helpful comments.


Appendix. Robustness Checks

Table A.1 Exogeneity Tests for the et al. Process

Marginal effects from logit rgeressionUnit of observation= Internet draftDependent variable= Et Al Dummy

Individual IDs Working group IDs

Sample: Two or more authors Two or more authors

Wald-test: Top �0�89� �0�67� �0�75� �0�16�panel jointly zeroa

WG Chair Author −0�039 0�007 −0�007 0�082�0�04� �0�05� �0�06� �0�07�

Multisponsor 0�014 0�008 0�064 0�053�0�04� �0�04� �0�06� �0�06�

Intl Author 0�002 0�002 0�026 0�021�0�02� �0�02� �0�03� �0�03�

Log Pages 0�015 0�024 −0�045 −0�001�0�04� �0�04� �0�08� �0�08�

Longest Nameb −0�003 −0�003 0�006 0�013�0�01� �0�01� �0�01� �0�01�

Prior RFC Effects a �0�331� �0�062�+

Days-to-Meeting −0�155∗∗ −0�150∗∗ −0�191∗∗ −0�178∗∗

�0�03� �0�03� �0�05� �0�06�Three Authors 0�628∗∗ 0�630∗∗ 0�667∗∗ 0�676∗∗

�0�03� �0�03� �0�05� �0�05�Four Authors 0�732∗∗ 0�732∗∗ 0�780∗∗ 0�790∗∗

�0�02� �0�02� �0�03� �0�03�Five Authors 0�699∗∗ 0�699∗∗ 0�764∗∗ 0�768∗∗

�0�02� �0�02� �0�03� �0�03�

Observations 1,723 1,723 638 638Log-likelihood −637.5 −634.5 −233.8 −228.8

Notes. Specification also includes SubmissionDate and SubmissionDate 2 tocontrol for time trends. Robust standard errors are in parentheses.

aFigures in square brackets are p-values from a Wald test for joint signif-icance. If Et Al Dummy is exogenous, this test should fail to reject the nullhypothesis.

bNumber of digits in the last name of author with the longest surname.+10% significance; ∗5% significance; ∗∗1% significance.

Table A.3 Robustness of Signaling Effect

Models of individual ID publicationUnit of observation= Internet draft


Sample: All IDs All IDs All IDs Two or more authors Two authors and no lead chair

Specification: Logita OLS OLS OLS OLS

WG Chair Author 0�088∗∗ 0�054∗∗ 0�054∗∗ 0�056∗ 0�096∗

�0�01� �0�02� �0�02� �0�02� �0�04�Unlisted Chair −0�031∗ −0�073∗∗ −0�066∗ −0�111∗

�0�01� �0�03� �0�03� �0�05�Et Al Dummy −0�022∗ −0�008 −0�007 −0�009 0�004

�0�01� �0�02� �0�02� �0�02� �0�02�Missing Et Al b 0�008

�0�05�Missing Et Al ∗Chair 0�006

�0�02�

Table A.2 Status and Reputation

OLS regressionsa

Unit of observation= Internet draftDependent variable=WG Chair Author

Individual IDs

Sample: All drafts Single author

No Prior RFCs 0�103∗∗ 0�001�0�02� �0�03�

Prior RFCs [1] 0�228∗∗ 0�160∗∗

�0�03� �0�05�Prior RFCs [2] 0�461∗∗ 0�503∗∗

�0�04� �0�06�Prior RFCs �3�4� 0�569∗∗ 0�550∗∗

�0�04� �0�06�Prior RFCs �5�10� 0�732∗∗ 0�651∗∗

�0�04� �0�05�Prior RFCs �10+� 0�879∗∗ 0�791∗∗

�0�03� �0�04�Two Authors −0�063∗∗

�0�01�Three Authors −0�053∗∗

�0�02�Four Authors −0�016

�0�02�Five Authors −0�044

�0�03�

IETF meeting effects Y YObservations 3,129 1,406R-squared 0�588 0�602Mean of dependent variable 0�258 0�262

Note. Robust standard errors are in parentheses.aModel does not include a constant, so the coefficient on Prior RFC

dummies shows the probability that a sole-authored ID was written by aWG chair.

+10% significance; ∗5% significance; ∗∗1% significance.


Table A.3 (Continued)

Models of individual ID publicationUnit of observation= Internet draft


Sample: All IDs All IDs All IDs Two or more authors Two authors and no lead chair

Specification: Logita OLS OLS OLS OLS

Unlisted Chair2 c −0�094∗∗

�0�03�Unlisted Chair3 c −0�052

�0�04�Unlisted Chair4c −0�065

�0�04�Unlisted Chair5 c −0�100∗

�0�04�Log Pages 0�005 0�005 0�005 0�003

�0�00� �0�00� �0�01� �0�01�Intl Author −0�014 −0�014 −0�016 0�003

�0�01� �0�01� �0�01� �0�01�Multisponsor 0�004 0�004 0�006 0�024

�0�01� �0�01� �0�01� �0�02�Days-to-Meeting 0�045∗∗ 0�046∗∗ 0�049∗∗ 0�052∗∗

�0�01� �0�01� �0�01� �0�01�Constant −0�114∗∗ −0�114∗∗ −0�128∗∗ −0�157∗∗

�0�03� �0�03� �0�05� �0�05�

Prior RFC effects N Y Y Y YAuthor count effects N Y Y Y YIETF meeting effects N Y Y Y Y

Observations 3,129 3,129 3,129 1,723 692R-squared 0.052 0.052 0.061 0.063

Note. Robust standard errors are in parentheses.aLogit estimates are reported as marginal effects at sample means.bMissing Et Al = 1 iff the ietf-announce message does not list all authors and does not use et al. to acknowledge missing names.cUnlisted ChairX = 1 iff the X th author is a WG chair and is not listed in the ietf-announce message.+10% significance; ∗5% significance; ∗∗1% significance.

Table A.4 Robustness of Attention Effects

Poisson and logit regressionsUnit of observation= Internet draft


Specification: Poisson Poisson Logita Poisson Poisson Logita


WG Chair Author 0�206∗∗ 0�269∗ 0�022 −0�085 0�158 0�036�0�05� �0�11� �0�03� �0�06� �0�14� �0�03�

Unlisted WG Chair −0�233∗ −0�622∗∗ −0�092+ −0�086 0�212 0�002�0�10� �0�21� �0�05� �0�11� �0�21� �0�06�

Et Al Dummy 0�108+ −0�250∗ −0�071∗ 0�029 0�081 −0�033�0�06� �0�12� �0�03� �0�08� �0�18� �0�04�

Log Pages 0�041∗ 0�057+ 0�021∗ −0�008 0�119∗ 0�030∗∗

�0�02� �0�03� �0�01� �0�02� �0�05� �0�01�Intl Author 0�063+ −0�024 0�044∗ 0�024 0�066 0�025

�0�04� �0�08� �0�02� �0�05� �0�10� �0�03�Multisponsor −0�051 0�010 0�056∗ −0�190∗ −0�168 −0�024

�0�05� �0�09� �0�03� �0�09� �0�19� �0�05�Days-to-Meeting −0�082∗∗ −0�028 0�044∗∗ −0�116∗∗ 0�095 −0�005

�0�03� �0�06� �0�01� �0�03� �0�07� �0�02�Author E-mails 0�079∗∗ 0�133∗∗ −0�001 0�049∗∗ 0�109∗∗ 0�006

�0�01� �0�01� �0�00� �0�01� �0�01� �0�01�Constant −0�249+ 0�599∗ 0�424∗∗ 0�409

�0�13� �0�26� �0�15� �0�34�


Table A.4 (Continued)

Poisson and logit regressionsUnit of observation= Internet draft


Specification: Poisson Poisson Logita Poisson Poisson Logita


F -test: WG Chair and Prior �0�00� �0�00� �0�00� �0�01� �0�16� �0�09�RFCs jointly zerob

Prior RFC effects Y Y Y Y Y YAuthor count effects Y Y Y Y Y YIETF meeting effectsa Y Y Y Y Y Y

Observations 3,129 3,129 3,129 1,046 1,046 1,046Log-likelihood −3,713.4 −1,1282.0 −2,103.5 −1,305.7 −4,229.1 −501.7

Note. Robust standard errors are in parentheses.aMarginal effects of logit model calculated at sample means (which requires replacing IETF Meeting Date dummies with a linear time-trend).bFigures in square brackets are p-values from a Wald test for joint significance of the WG Chair and Prior RFC Count dummies.+10% significance; ∗5% significance; ∗∗1% significance.

ReferencesAi, C., E. C. Norton. 2003. Interaction terms in logit and probit

models. Econom. Lett. 80(1) 123–129.Ayres, I., J. Waldfogel. 1994. A market test for race discrimination

in bail setting. Stanford Law Rev. 46(5) 987–1047.Becker, G. S. 1993. Nobel lecture: The economic way of looking at

behavior. J. Political Econom. 101(3) 385–409.Benjamin, B. A., J. M. Podolny. 1999. Status, quality, and social order

in the california wine industry. Admin. Sci. Quart. 44(3) 563–589.Berger, J., S. Rosenholtz, M. Zelditch. 1980. Status organizing pro-

cesses. Annual Rev. Sociol. 6 479–508.Blank, R. M. 1991. The effects of double-blind versus single-blind

reviewing: Experimental evidence from The American EconomicReview. Amer. Econom. Rev. 81(5) 1041–1067.

Bothner, M. S., R. Haynes, W. Lee, N. Smith. 2010. When doMatthew Effects occur? J. Math Sociol. 34(2) 80–114.

Burt, R. S. 2004. Structural holes and good ideas. Amer. J. Sociol.110(2) 349–399.

DellaVigna, S. 2008. Psychology and economics: Evidence from thefield. J. Econom. Literature 47(2) 315–372.

Esteves-Sorenson, C. 2009. Micro-costs: Inertia in television viewing.Working paper, Yale School of Management, New Haven, CT.

Feldman, M. S., J. G. March. 1981. Information in organizations assignal and symbol. Admin. Sci. Quart. 26(2) 171–186.

Fleming, L., D. M. Waguespack. 2007. Brokerage, boundary span-ning, and leadership in open innovation communities. Organ.Sci. 18(2) 165–180.

Fuller, B. 2006. Phone interview with Dave M. Waguespack,November 14.

Goldin, C., C. Rouse. 2000. Orchestrating impartiality: The impactof “blind” auditions on female musicians. Amer. Econom. Rev.90(4) 715–741.

Gould, R. 2002. The origins of status hierarchies: A formal theoryand empirical test. Amer. J. Sociol. 107(5) 1143–1178.

Jensen, M., A. Roy. 2008. Staging exchange partner choices: When dostatus and reputationmatter?Acad.Management J. 51(3) 495–516.

Knowles, J., N. Persico, P. Todd. 2001. Racial bias in motor vehi-cle searches: Theory and evidence. J. Political Econom. 109(1)203–229.

Kreps, D. M., R. Wilson. 1982. Reputation and imperfect informa-tion. J. Econom. Theory 27(2) 253–279.

Merton, R. K. 1968. The Matthew Effect in science. Science 159(3810)56–63.

Morris, S., H. S. Shin. 2002. Social value of public information. Amer.Econom. Rev. 92(5) 1521–1534.

Mouw, T. 2006. Estimating the causal effect of social capital:A review of recent research. Annual Rev. Sociol. 32(1) 79–102.

Mowery, D., T. Simcoe. 2002. Is the Internet a U.S. invention? Aneconomic and technological history of computer networking.Res. Policy 31(8–9) 1369–1387.

Phillips, D., E. Zuckerman. 2001. Middle-status conformity: Theo-retical restatement and empirical demonstration in two mar-kets. Amer. J. Sociol. 107(2) 379–429.

Piskorski, M. J. 2007. Positions of power and status: Reciprocity inthe venture capital industry. Working paper, Harvard BusinessSchool, Boston.

Podolny, J. M. 1993. A status based model of market competition.Amer. J. Sociol. 98(4) 829–872.

Podolny, J. 1994. Market uncertainty and the social character of eco-nomic exchange. Admin. Sci. Quart. 39(3) 458–483.

Podolny, J. M. 2005. Status Signals: A Sociological Study of MarketCompetition. Princeton University Press, Princeton, NJ.

Podolny, J. M., D. J. Phillips. 1996. The dynamics of organizationalstatus. Indust. Corporate Change 5(2) 453–472.

Podolny, J., T. Stuart. 1995. A role-based ecology of technological-change. Amer. J. Sociol. 100(5) 1224–1260.

Powell, W. W., K. W. Koput, L. Smith-Doerr. 1996. Interorganiza-tional collaboration and the locus of innovation: Networks oflearning in biotechnology. Admin. Sci. Quart. 41(1) 116–145.

Rao, H., H. Greve, G. Davis. 2001. Fool’s gold: Social proof in theinitiation and abandonment of coverage by Wall Street ana-lysts. Admin. Sci. Quart. 46(3) 502–526.

Rindova, V., I. Williamson, A. Petkova, J. Sever. 2005. Being goodor being known: An empirical examination of the dimensions,antecedents, and consequences of organizational reputation.Acad. Management J. 48(6) 1033–1049.

Simcoe, T. 2008. Standard setting committees. Working paper,Boston University, Boston. http://ssrn.com/abstract=899595.

Sorenson, O., D. M. Waguespack. 2006. Social structure andexchange: Self-confirming dynamics in Hollywood. Admin. Sci.Quart. 51(4) 560–589.

Spence, M. 1973. Job market signaling. Quart. J. Econom. 87(3)355–374.

Stewart, D. 2005. Social status in an open-source community. Amer.Sociol. Rev. 70(5) 823–842.

Strutt, R. J. 1968. The Life of John William Strutt: Third Baron Rayleigh.University of Milwaukee Press, Milwaukee.

Stuart, T., H. Hoang, R. Hybels. 1999. Interorganizational endorse-ments and the performance of entrepreneurial ventures. Admin.Sci. Quart. 44(2) 315–349.

Waguespack, D. M., L. Fleming. 2009. Scanning the commons? Evi-dence on the benefits to startups of participation in open stan-dards development. Management Sci. 55(2) 210–223.

Washington, M., E. Zajac. 2005. Status evolution and competition:Theory and evidence. Acad. Management J. 48(2) 282–296.

Documents

Status, Quality, and Attention: What's in a (Missing) Name?