Wisdom of the crowds: Decentralized knowledge construction in Wikipedia
ABSTRACT Recently, Nature published an article comparing the quality of Wikipedia articles to those of Encyclopedia Britannica (Giles 2005). The article, which gained much public attention, provides evidence for Wikipedia quality, but does not provide an explanation of the underlying source of that quality. Wikipedia, and wikis in general, aggregate information from a large and diverse author-base, where authors are free to modify any article. Building upon Surowiecki's (2005) Wisdom of Crowds, we develop a model of the factors that determine wiki content quality. In an empirical study of Wikipedia, we find strong support for our model. Our results indicate that increasing size and diversity of the author-base improves content quality. We conclude by highlighting implications for system design and suggesting avenues for future research.
- SourceAvailable from: Benedikt Fecher
Conference Paper: Open Science: One Term, Five Schools of Thought[Show abstract] [Hide abstract]
ABSTRACT: Open Science is an umbrella term that encompasses a multitude of assumptions about the future of knowledge creation and dissemination. Based on a literature review, this paper aims at structuring the overall discourse by proposing five Open Science schools of thought: The infrastructure school (which is concerned with the technological architecture), the public school (which is concerned with the accessibility of knowledge creation), the measurement school (which is concerned with alternative impact measurement), the democratic school (which is concerned with access to knowledge) and the pragmatic school (which is concerned with collaborative research). Open Science, assessment and review, science 2.0, open access, open data, citizen science, science communication, altmetrics1st International Conference on Internet Science; 04/2013
Article: A Verification Method for MASOES.[Show abstract] [Hide abstract]
ABSTRACT: MASOES is a 3agent architecture for designing and modeling self-organizing and emergent systems. This architecture describes the elements, relationships, and mechanisms, both at the individual and the collective levels, that favor the analysis of the self-organizing and emergent phenomenon without mathematically modeling the system. In this paper, a method is proposed for verifying MASOES from the point of view of design in order to study the self-organizing and emergent behaviors of the modeled systems. The verification criteria are set according to what is proposed in MASOES for modeling self-organizing and emerging systems and the principles of the wisdom of crowd paradigm and the fuzzy cognitive map (FCM) theory. The verification method for MASOES has been implemented in a tool called FCM Designer and has been tested to model a community of free software developers that works under the bazaar style as well as a Wikipedia community in order to study their behavior and determine their self-organizing and emergent capacities.IEEE transactions on systems, man, and cybernetics. Part B, Cybernetics: a publication of the IEEE Systems, Man, and Cybernetics Society 06/2012; · 3.01 Impact Factor
- [Show abstract] [Hide abstract]
ABSTRACT: Wikis were originally intended for knowledge work in the open Internet environment, and there seems to be an inherent tension between wikis' affordances and the nature knowledge work in organizations. The objective of this paper is to investigate how tailoring wikis to corporate settings would impact users' wiki activity. We begin by synthesizing prior works on wikis' design principles; identifying several areas where we anticipate high tension between wikis' affordances and organizational work practices. We put forward five propositions regarding how changes in corporate wikis deployment procedures may impact users' wiki activity. An empirical study in one multi-national organization tested users' perceptions towards these propositions, revealing that in some cases there may be a need for modifying wiki's design, while in other cases corporations may wish to change their knowledge work practices to align with wikis' affordances.01/2012;
Wisdom of the Crowds: Decentralized Knowledge Construction in Wikipedia
Ofer Arazy, Wayne Morgan, and Raymond A. Patterson
University of Alberta, School of Business
Recently, Nature published an article comparing the quality of Wikipedia articles to
those of Encyclopedia Britannica (Giles 2005). The article, which gained much public
attention, provides evidence for Wikipedia quality, but does not provide an explanation of the
underlying source of that quality. Wikipedia, and wikis in general, aggregate information
from a large and diverse author-base, where authors are free to modify any article. Building
upon Surowiecki's (2005) Wisdom of Crowds, we develop a model of the factors that
determine wiki content quality. In an empirical study of Wikipedia, we find strong support for
our model. Our results indicate that increasing size and diversity of the author-base improves
content quality. We conclude by highlighting implications for system design and suggesting
avenues for future research.
Wiki is a recent technology that gained prominence in the early 2000s. Wiki allows
users to edit a webpage and submit new versions, immediately replacing the previous version
(which is archived) (Leuf and Cunningham. 2001). One of the most prominent uses of wiki
technology is Wikipedia, the online encyclopedia. In five years, Wikipedia has become one of
the most popular websites (Bryant et al. 2005), and achieved quality similar to the leading
print-based encyclopedia, unexpectedly perhaps for a system that allows any user to edit and
overwrite web page content.
The recent interest in wikis goes well beyond the specific technology. Wikis represent
an approach – referred to by Arazy et al. (2005) as “Open Content Systems” - that is poised to
transform the way in which knowledge and knowledge-bases are constructed. Evans & Wolf
(2005) describe the de-centralized and social approach to knowledge accumulation, which
owes much to the open source movement, as a revolutionary approach to the delivery of
One of the most compelling explanations for Wikipedia’ success is, in short, “the
wisdom of the crowds” (WoC). Surowiecki (2005) suggests that the aggregate knowledge of a
large group is superior to the knowledge of one or a few experts. A simple example illustrates
the WoC principle: if many people are asked to estimate a person’s weight, the average of
their estimates very likely will be close to the person’s actual weight, and closer than (or at
least as good as) the estimate an expert would give. However, not all crowds are wise. We
propose that Wikipedia and wiki technology embody and promote the WoC principles, and
attribute Wikipedia’s success to its ability to effectively leverage the wisdom of the crowds.
In this paper we operationalize the WoC framework, and test its validity using
Wikipedia data. Our findings suggest the WoC principles indeed determine Wikipedia article
quality. The main contribution of this paper is providing an explanation of how open content
systems successfully achieve high quality content. Wiki systems developers and
administrators could utilize our findings to direct systems’ design.
The paper continues as follows: in Section 2 we review related work; Section 3
presents our study’s research question; Section 4 describes the empirical design; in Section 5
we report on the results of the study; finally, Section 6 concludes by discussing the
implications of our findings and highlighting areas for future research.
2. Related Work
“If you put together a big enough and diverse enough group of people and ask them to
‘make decisions affecting matters of general interest’, that group’s decision will, over time, be
‘intellectually superior to the isolated individual’, no matter how smart or how well-informed
he is.” (Surowiecki, 2005, p. XVII)
Surowiecki (2005) synthesizes theoretical and empirical results from various fields
into what he refers to as “The Wisdom of the Crowds” (WoC). Surowiecki provides numerous
examples, from an array of research disciplines ranging from biology to behavioral
economics, which demonstrate that the aggregate knowledge of a large and diverse group is
superior to that of one or a few experts. Three principles are essential, according to
Surowiecki (2005) for assuring that the aggregate contributions of a crowd are high quality:
(a) a large number of contributors and opinions (e.g. crowd size), (b) diversity of ideas and
opinions, and (c) appropriate mechanisms for aggregating the opinions (e.g. the participants
are able to express their opinions independent of influence from others, and aggregation
techniques combine the independent opinions). Surowiecki (2005) asserts that wiki
technology is an appropriate mechanism for content aggregation. This study focuses on: (a)
the number of contributors and opinions (e.g. size), and (b) diversity of ideas and opinions.
We discuss these below.
The number of people (i.e the number of ideas and opinions) in a crowd directly
impacts the crowd’s aggregate knowledge. The accuracy of judgments of a statistical group is
best explained by reference to the Jury Theorem (Condorcet, 1976). In a group of n people,
making a binary decision, where each person has the same probability p of choosing the right
answer, and people’s choices are independent, the probability that the group will reach the
correct answer is
correct answer is higher than 50%, Pnconverges with n, group size, to 1 (the larger the
group, the quicker the convergence). This effect has been demonstrated by numerous studies
(Surowiecki 2005; Sunstein 2006).
Diversity also impacts the crowd’s wisdom and is expected in large groups. As
Surowiecki (2005) notes, diversity adds perspectives that would otherwise be absent. Group
diversity also reduces some more destructive aspects of group decision making, such as
groupthink and conformity. Following organizational theorist J.G. March, diversity enables
groups to continue to learn as new perspectives and information are introduced (March 1991).
Mannix and Neale (2005), reviewing fifty years of diversity research, point out how
differences in group members’ skills and knowledge are associated with increased group
If wiki systems, and specifically Wikipedia, embody the WoC principles, then we
would expect Wikipedia articles to exhibit high quality. Viegas et al. (2004) demonstrate that
vandalism is quickly corrected. Emigh & Herring (2005) study the format and writing-quality
of Wikipedia articles, and report that Wikipedia content maintains high standards, suggesting
this as evidence for content quality. Stvilia et al. (2005) demonstrate that Wikipedia’s
1!! / !. When probability of an individual’s
decentralized quality control mechanisms indeed identify and correct many quality threats.
Finally, a recent news article in Nature (Giles 2005) directly tests Wikipedia quality, based on
an expert peer review of a sample of articles, and finds Wikipedia achieves a level of accuracy
comparable to Encyclopedia Britannica1.
While these recent empirical findings regarding Wikipedia strengthen our confidence
in the quality of Wikipedia content, they provide little insight on the factors driving that
quality. O’Reilly (2005), in a column on his website, suggests that the WoC phenomenon may
be responsible for wikis’ success. Although this, to date, has not been demonstrated in
research, there are initial indications that WoC principles (size, diversity, and aggregation
mechanisms) may be associated with Wikipedia’s content quality. Some (e.g. Lih 2004,
Stvilia et al. 2005) argue that the number of contributors and edits in an article is associated
with an article’s quality. A large author-base ensures diversity of opinions (Surowiecki 2005),
and Wikipedia contributors exhibit diversity in terms of background and interests (Bryant et
al. 2005). Diversity of opinion is evident in an article’s discussion page (where conflicting
opinions are deliberated), and high-quality pages (i.e., those selected as Featured Articles)
have larger discussion pages (Stvilia et al. 2005).
3. Research Question
To date, there is little empirical evidence that WoC principles indeed determine
Wikipedia article quality. The objective of this study is to asses the extent to which the WoC
principles explain Wikipedia article quality. The aggregation mechanism is an inherent
feature of wiki technology, and thus we do not empirically examine its existence in
Wikipedia. Crowd size and diversity, on the other hand, may vary greatly across wiki articles
and wiki implementation, and thus we pose the following research question: “to what extent
do crowd size and diversity impact Wikipedia article quality?”
4. Research Method
To investigate the determining factors of Wikipedia’s quality and validate the
proposed framework, we conducted an empirical study, utilizing the 42 Wikipedia articles
included in the Nature study (Giles 2005). We choose this sample (using the full-text of these
articles) because it provides us with independent quality assessment of articles, obtained
through an objective process, and published in a highly-respected journal.
The notion of information quality is multi-faceted, and the most referenced quality
dimensions are: accuracy, completeness, relevance, and timeliness (Kim et al 2005). Giles
(2005) reports the number of errors per article, and these errors include inaccuracies and
missing data. Since we can assume relevance (the articles were judged by experts on the
topic) and timeliness (a recent version of the articles was used), Giles’ measure captures an
1 It is worth mentioning that the article was followed by a debate between Britannica and Nature, where
Britannica challenged the validity of the study (see
http://www.nature.com/nature/britannica/eb_advert_response_final.pdf), and Nature defended
their methodology and findings (see www.nature.com/nature/britannica/index.html ).
article’s (lack of) quality2. To transform the error counts into a quality measure we use the
inverse function, thus Quality=1
We operationalize the Size and Diversity WoC principles as follows. Size is estimated
through two proxies (1) the number of unique contributors to an article (number of authors),
and (2) the total number of contributions (number of edits). Diversity would ideally be
measured through the variance of background, networks, education, and skills for an article’s
authors (Mannix & Neale 2005); however, such data is not available in Wikipedia. Instead,
we employ two proxies: (1) the number of words in an article’s discussion page, and (2) the
number of edit wars (where Wikipedia’s defines an edit war as three edits by a particular user
within 24 hours, with edits from other users in between). The greater the variance in opinions
is, the more likely the authors are to engage in edit wars and argue their opinions on the
discussion page. In order to rule out the explanation that the number of errors was affected by
an article’s length, as proposed by (Stvilia et al. 2005), we test for this effect.
The complete Nature sample includes 42 articles3, 4 of which are reported to have no
inaccuracies. Since our measure of quality is 1/Number of Errors, we removed articles
without errors from our sample. To calculate the measures of Size and Diversity, we
harvested Wikipedia articles, relying heavily on articles’ history logs. We conducted a
thorough data cleansing process, and removed all non-human authors (i.e. software bots) and
edits made by these bots. In addition, since Wikipedia articles change content continually, it
was important that the data collected corresponds to the date of quality assessments. We
obtained the exact cutoff dates (mid-October 2005) of the articles from the Nature article
author through personal correspondence.
We analyzed our data using partial least squares (PLS), since this statistical method
tests for causal relations and allows collapsing of several measures into one latent construct.
The results support our model. First, the items grouped into a latent construct do represent one
concept (the number of authors and the number of edits are highly correlated, 0.94; similarly,
discussion page length and the number of edit wars are highly correlated, 0.78; both
correlations are significant at 0.01). Second, a causal relationship between Diversity and the
dependent variable (Quality) is established, and is statistically significant. Finally, size has a
significant effect on Diversity. Figure 1 below graphically presents the PLS results. Size has a
positive effect on Diversity, and Diversity has a positive effect on Quality (path coefficients
are 0.43 and 0.61 respectfully, both significant at 0.05). Overall, the proposed model explains
30% of the variance in Article Quality (R2 = 0.302).
2 Although not all errors represent the same threat to quality and some errors are more serious than
others, judging error seriousness is highly subjective and requires domain expertise in a wide array of fields. We,
thus, assigned equal weight to all errors.
3 For a complete list of the articles assessed by Nature please refer to (Giles 2005a).
It is interesting to note that the coefficient for the path linking Size to Quality has a
negative sign. This result could be explained by the conflicting effects of diversity. Diversity
is a multi-dimensional construct, where some diversity dimensions (e.g. skills and knowledge)
have a positive effect on group decision-making, while other dimensions (e.g. age, gender,
status) have negative effects on group performance (Mannix & Neale, 2005). Group size
increases both types of diversity, and the correlations between Size and Quality is very close
to zero. In the PLS model, the positive effect of group size on quality is represented through
the link to opinion diversity, and the remaining negative effect is evident in the direct
relationship between size and quality.
6. Discussion and Conclusions
The success of Wikipedia has attracted much public attention recently, since it seems
counter-intuitive that a large and diverse author-base could generate high-quality content in a
seemingly unorganized and uncontrolled manner. Several alternative explanations for this
phenomenon are possible. Some see the key to Wikipedia’s success in the cohesiveness of the
Wikipedia community and authors’ motivations, while others focus on the specific
decentralized control mechanisms (e.g. WatchLists) that Wikipedia employs. In this paper, we
proposed an alternative explanation that is based on the Wisdom of the Crowds framework,
and demonstrated that two factors – crowd size and diversity – explain Wikipedia content
quality. Our findings suggest that a simple model can account for a relatively high percentage
of the variance in article quality. The main contributions of this work are in (1) proposing a
model of the WoC phenomena, based on Surowiecki’s (2005) framework, and (2) in
demonstrating empirically that this model indeed explains Wikipedia’s article quality.
Our findings have direct implications for wiki system designers and administrators. In
order to achieve high quality content, it is essential that many users participate in authoring
wiki pages. Also, it is important that participation levels be high (i.e. each page has many
edits). To entice participation, organizations using wikis should strive to eliminate barriers
(e.g. allow users to post anonymously) and provide incentives for contributions. It is
important that the author-base be diverse, thus organizations should encourage cross-
departmental collaboration, and perhaps inter-organizational collaboration, in wiki authoring.
Lastly, it is essential that wiki systems incorporate aggregation mechanisms that will ensure
the independence of users’ opinions.
Although our empirical analysis was performed on a wiki system, we believe that our
results generalize to other open content system (e.g. advanced discussion forums, such as
Slashdot) and collaborative software development. In future research we plan to expand on
the current study and address some limitations. We employed indirect measures of crowd size
and diversity, and thus it is suggested that future research explores alternative measures. Our
study used a small sample of Wikipedia articles, and future research should try to generalize
these results to alternative samples of Wikipedia articles and to other wiki systems.
Specifically, we plan to explore these ideas in organizational settings. Lastly, it would be
interesting to apply our model to other types of open content systems, such as append-based
systems (e.g. discussion groups), open source systems, and other types of decentralized
knowledge management systems where information quality is paramount.
Arazy, O. & R. Patterson. 2005. An analysis of open content systems. Proceedings of the
Fifteenth Annual Workshop on Information Technologies and Systems. Las Vegas,
Nevada. December 10-11, 2005.
Bryant, S., A. Forte & A. Bruckman. 2005. Becoming wikipedian. Group05.
Condorcet, 1976. Selected Writings, editor K.M. Baker.
Emigh W. & Herring S., 2005, Collaborative authoring on the web: A genre analysis of online
encyclopedias. 38th International Conference on System Sciences.
Evans, P. & B. Wolf. 2005. Collaboration rules. Harvard Business Review. 83 (7/8): 96-104.
Giles, J. 2005. Internet encyclopedias go head to head. Nature. 438 (15): 900-901.
Giles, J. 2005a. Supplementary information to “Internet encyclopedias go head to head”.
Online at Nature www.nature.com/news/2005/051212/full/438900a.htm
Kim, Y., R. Kishore, & G. L. Sanders. 2005. From DQ to EQ: understanding data quality in
the context of e-business systems. Communications of the ACM. 48 (10): 75-81.
Leuf B. & Cunningham W. 2001, The Wiki Way: quick collaboration on the web. Addison-
Lih, A. 2004. Wikipedia as participatory journalism: reliable sources? Fifth International
Symposium on Online Journalism, April 16-17, 2004. University of Texas at Austin.
March, J. 1991. Exploration and exploitation in organizational learning. Organization
Science. 2: 71-87.
Mannix, E. & M. Neale. 2005. What differences make a difference? Psychological Science
in the Public Interest. 6 (2): 31-55.
O’Reilly, T. 2005. What is web 2.0. Design patters and business models for the next
generation of software. www.oreillynet.com, September 30, 2005.
Stvilia, B., M. Twidale, L. Gasser, & L. Smith. 2005. Information quality discussions in
wikipedia. Working paper. University of Illinois at Urbana-Champaign.
Sunstein C., 2006. Infotopia, Oxford University Press.
Surowiecki, J. 2005. The wisdom of crowds. New York: Random House.
Viegas, F., M. Wattenberg & K. Dave. 2004. Studying cooperation and conflict between
authors with history flow visualizations, CHI 2004.