cdsc_examples_repository/dissertations/nathante_uw_2021/ch5_conclusion.tex

% \documentclass[12pt]{memoir}

% \usepackage{cdsc-memoir}
% % there are two chapter styles: cdsc-article and cdsc-memo
% % memo assumes that you remove the "\\" and the email address from the
% % \author field below as well as that you will comment out the
% % \published tag
% \chapterstyle{cdsc-article}

% \usepackage[utf8]{inputenc}
% \usepackage{wrapfig}
% \usepackage[T1]{fontenc}
% \usepackage{textcomp}
% \usepackage[garamond]{mathdesign}

% \usepackage[letterpaper,left=1in,right=1in,top=1in,bottom=1in]{geometry}

% % packages i use in essentially every document
% \usepackage{graphicx}
% \usepackage{enumerate}

% % packages i use in many documents but leave off by default
% % \usepackage{amsmath, amsthm, amssymb}
% % \usepackage{dcolumn}
% % \usepackage{endfloat}

% % import and customize urls
% \usepackage[usenames,dvipsnames]{color}
% \usepackage[breaklinks]{hyperref}

% \hypersetup{colorlinks=true, linkcolor=Black, citecolor=Black, filecolor=Blue,
%     urlcolor=Blue, unicode=true}

% % list of footnote symbols for \thanks{}
% \makeatletter
% \renewcommand*{\@fnsymbol}[1]{\ensuremath{\ifcase#1\or *\or \dagger\or \ddagger\or
%  \mathsection\or \mathparagraph\or \|\or **\or \dagger\dagger
%   \or \ddagger\ddagger \else\@ctrerr\fi}}
% \makeatother
% \newcommand*\samethanks[1][\value{footnote}]{\footnotemark[#1]}

% % add bibliographic stuff 
% \usepackage[american]{babel}
% \usepackage{csquotes}
% \usepackage[natbib=true, style=apa, backend=biber]{biblatex}
% \addbibresource{ch5_conclusion.bib}
% \DeclareLanguageMapping{american}{american-apa}

% \defbibheading{secbib}[\bibname]{%
%   \section*{#1}%
%   \markboth{#1}{#1}%
%   \baselineskip 14.2pt%
%   \prebibhook}

% \def\citepos#1{\citeauthor{#1}'s (\citeyear{#1})}
% \def\citespos#1{\citeauthor{#1}' (\citeyear{#1})}

% % memoir function to take out of the space out of the whitespace lists
% \firmlists

% % LATEX NOTE: these lines will import vc stuff after running `make vc` which
% % will add version control information to the bottom of each page. This can be
% % useful for keeping track of which version of a document somebody has:
% % \input{vc}
% % \pagestyle{cdsc-page-git}

% % LATEX NOTE: this alternative line will just input a timestamp at the
% % build process, useful for Overleaf
% % \pagestyle{cdsc-page-overleaf}

% \begin{document}

% \setlength{\parskip}{4.5pt}
% % LATEX NOTE: Ideal linespacing is usually said to be between 120-140% the
% % typeface size. So, for 12pt (default in this document, we're looking for
% % somewhere between a 14.4-17.4pt \baselineskip.  Single; 1.5 lines; and Double
% % in MSWord are equivalent to ~117%, 175%, and 233%.

% \baselineskip 16pt

% \title{Future Directions in the Ecology of Online Communities}
% \author{Nathan TeBlunthuis\\
%         \href{nathante@uw.edu}{nathante@uw.edu}}
% \date{September 1, 2021}

% \maketitle

% %  LocalWords:  

% % TO CONCLUSION:

% \section{}

Chapter 1 says that ``the project of this dissertation is to begin reconstructing organizational ecology in the relatively theory-poor but data-rich context of online communities.'' By focusing on understanding the relationships between related online communities in ecological terms of competition and mutualism and in the emic language of members of overlapping communities, the preceding work seeks to build an empirical foundation to build new ecological theory.  It has found qualitative and quantitative evidence that overlapping online communities often fill distinctive niches by providing complementary benefits to their users.  Competitive dynamics also occur, and can be strong, but do not last as long.  Although competition and mutualism play a role in their growth and survival, communities may not adapt to promote mutualism and avoid competition.  Rather it seems likely that the ``principle of competitive exclusion'' takes hold in some other way, perhaps through a selection process in which communities normally must provide complementary benefits to existing ones in order to take off.

Of course, these claims are limited by the empirical tools that were used to support them.
Inferences about competition and mutualism are based upon time series models with fundamentally untestable assumptions.
By fitting a far greater number of models than I could carefully specify I have taken an unabashed ``big data'' approach. 
To make confident claims about any particular competitive or mutualistic relationship between two subreddits I would have to conduct a relatively exacting model selection and comparison procedure based on additional contextual knowledge of the communities' histories.  
The large scale of this analysis supports the general findings enumerated above assuming that any model misspecifications have not introduced errors in a systematic and misleading way.
The fact that both the linear and nonlinear time series analyses and the active community members all seem to agree that mutualism is more common than competition provides some reassurance. It seems quite unlikely that all three will mislead in similar ways. 

The reconstruction project is still beginning, but at this stage we can propose preliminary answers to some key theoretical questions: (1) How do people construct systems of overlapping online communities? (2) What types of ``resources'' are most important for mediating ecological interactions? (3) How do ecological interactions relate to broader dynamics such as the growth of a platform or the popularity of a broader topic? and (4) How do barriers between different platforms affect cross-platform ecological relationships?

Question (1) is fundamental to an ecological explanation for the development of online communities. A preliminary answer is that \emph{people construct systems of overlapping online communities as new online communities find distinctive niches in the neighborhood of existing communities relatively early in their development}.  Chapter 4 finds evidence that systems of overlapping online communities are not constructed through an adaptation process and suggests a selection process as an alternative.  It should be noted that selection and adaptation are not mutually exclusive and the systems of overlapping communities may develop through a hybrid process. Chapter 3 suggests that a large majority of active online communities each have a distinctive ecological niche. It seems likely that successful online communities are often quickly find a niche early in their development.   

Prior research and Chapter 2 both see users and topics as related to rival or non-rival resources that make competition and mutualism between online communities possible. However,  Chapter 2 finds that user and topic overlap densities are very weakly correlated with online community growth suggesting that user and content overlaps are not very close analogs for the kinds of resource overlaps considered by organizational ecologists, such as the technological range of a firm's outputs \citep{dobrev_shifting_2003}. Based on findings from Chapter 3, a preliminary answer to question (2) is that \emph{online communities' ecological niches are a product of content categories, audiences, and social capital}. These dimensions of an online community's niche might be difficult to precisely measure, but they can be described in theory. 

Content categories are socially constructed classes such as ``memes,'' ``Q\&A,'' ``news,'' ``commentary,'' ``art,'' ``documentation'' and ``discussion.''  Online communities often specialize in a subset of possible content categories.
Specialization in a set of content categories might be achieved formally through rules or definitions of topical scope or informally, through the community's size, or the preferences and behaviors of its members. 
The empirical work so far considers topics measured through semantic similarity or language models.
Content categories are likely to be correlated with such measures, but the measures are unlikely to faithfully capture important aspects of content categories like differences in medium, genre and form. 

The notion of social capital and audience as distinct aspects of a niche disentangles the concept of ``user.''  Social capital refers to the benefits that come from interpersonal interaction and sense-making with a homophilous or tight-knit community \citep{ackerman_sharing_2013}. Measures of group size or user overlap may be correlated with social capital, but they do little to distinguish a user who comments as a member crowd-like audience from a user who seeks social bonds and interactions with fellow members of their identity group or enthusiasts in their hobby. 

Question (3) is an important part of an ecological explanation for the rise and decline of platforms in terms of the communities they host. A preliminary answer is that \emph{ecological interactions and the rise or decline of a topical area drive one another in feedback process}. Chapter 2 suggests that growing platforms may be more likely to have mutualistic dynamics as they have an increasing number of potential niches for online communities of varying sizes and scopes. At the same time, mutualistic interactions among overlapping communities are likely to drive the rise of a platform as mutualists enrich niches in their neighborhoods. In a similar way, competition and the decline of topical area might reinforce each other if out-migration of users interested in the topic induces competition over the remaining users and this accelerates the communities' declines prompting further out-migration.

Question (4) considers the ecological consequences of how different social media platforms divide related online communities such as the Wikis and subreddits about the same topic. A preliminary answer is that \emph{barriers between platforms limit both mutualistic and competitive dynamics} because of how they limit the sharing of users or content across platforms. However, when non-rival resources such as information and community building know-how are transferable across platforms communities on platforms designed to provide different types of benefits are likely to be mutualists.
For example, subreddits and Wikis about similar topics are probably mutualists because wikis are designed primarily for developing and sharing encyclopedic information and subreddits often focus on socialization and discussion.

Now I will sketch several possible directions for near-future work in this research program.  Some of these potential projects seek to develop more complete answers for the key theoretical questions and others will bridge ecological analysis to specific practical problems. 
My hope is that empirical support and theoretical development will soon be sufficiently advanced to inform the design of present and future online community ecosystems and to understand the successes and limitations of peer production.     

\section{Ecological Relationships Between Platforms}

A significant limitation of my empirical studies has been that they focus only on interactions among communities within a single large platform.  However, online communities often overlap across platforms \citep{kiene_technological_2019} and cross-platform interactions are likely to be important \citep{vincent_examining_2018}.  For example, Reddit's growth enormously increased in 2010 when users of rival site Digg.com migrated \emph{en mass}, suggesting that during this period subreddits and Digg sections were in competition \citep{noauthor_digg_2021}.  In chapter 6 of \emph{Building Successful Online Communities}, \citet{resnick_starting_2012} recommend that new online communities ``carve out a useful and defendable (sic) niche in the ecology of competing communities.'' They base this recommendation upon virtually no evidence taken from studies of online communities or organizational ecology but rather by following intuitions drawn from economics and assuming that online communities may find themselves in ``winner-take-all'' situations.  Although they recommend specialization as a strategy for avoiding competition, they also suggest ``lock-in'' features like having different user interfaces and making it so identities cannot be shared between communities. 

At issue is how \citet{resnick_starting_2012} attempt to simultaneously adopt the perspectives of two different types of actors whose interests are often unaligned.
Commercial platforms need to generate private revenues and seem to better fit the classical models of organizational ecology that have niche overlaps as highly correlated with competition.
A commercial platform may find mutualism between cross-platform communities a nuisance and may find the ``lock-in'' features unequivocally beneficial.
However, building a successful online community is not the same as building a platform that hosts online communities.
My ecological studies of relationships between communities suggest that mutualism is widespread among actually existing online communities within a platform.
In my conversations with members of overlapping communities, I learned that they often benefit from overlapping communities on different platforms. 
Therefore it seems likely that that communities on commercial platforms that are both sufficiently ``open'' and sufficiently differentiated  will also be mutualistic, even if the platforms compete with each other over revenues. If so, this points to the promise of designs that support resource sharing across such platforms.

Knowledge about inter-platform ecological dynamics is only beginning to be created.
\citet{nagaraj_how_2021} have found that open source knowledge projects like open street map are hurt by competition with proprietary alternatives.
Cross-platform studies of digital traces face difficulties because it is not generally possible to associate user accounts on different platforms.  
However, the time-series models I have used only depend on finding related communities and therefore enable studying ecological interactions without tracking users across platforms.
I am developing a new dataset of related subreddits, Fandom.com wikis, and Wikipedia articles to investigate ecological interactions between related communities on different platforms.

\section{Selecting Niche Width}

Choosing a scope is an important design decision for organizations and for online communities. As I found in Chapter 3, broad and narrow scopes are associated with trade-offs in the types of benefits that a community can provide. The choice of scope, or the choice of how a community will specialize, may also have implications for the community's short and long run survival.  According to theories of organizational ecology, the choice of scope may affect a community's competitive and mutualistic dynamics and its ability to weather changes in a turbulent environment.  

Resource partitioning theory, discussed briefly in sections of Chapters 2, 3 and 4, provides a framework for understanding how specialization relates to competition.  It proposes that larger generalists can coexist with specialists because large generalists are not optimally efficient at all of their activities, leaving opportunities for specialists to out-compete them in narrow niches \citep{carroll_concentration_1985}.  Findings from Chapter 2 suggest that one prediction of resource partitioning theory seems to obtain in groups of overlapping online communities. This is that they often have a ``main'' community which is a large generalist and people participate in the specialist communities in order to obtain distinctive benefits not easily obtained in the main community \citep{baum_ecological_2006}.  

A related theory fragment of organizational ecology, niche width theory \citep{dobrev_dynamics_2001, freeman_niche_1983}, proposes that specialists are less able to survive during periods of rapid change. Large generalists may have advantages in changing environments because their diversity of interests which spreads out risk, their experience transferring knowledge between different parts of their organization, and their slack resources can all help them absorb negative outcomes \citep{dobrev_shifting_2003}. 
As discussed in Chapter 4, online communities may inhabit unstable environments where sudden events, ongoing trends, and abrupt policy changes can all affect participation \citep{ratkiewicz_characterizing_2010}.

An example illustrates how environmental change can threaten the success of specialists. During the Trump administration, a number of anti-trump subreddits were organized around specific controversies (e.g., \texttt{r\Slash the\_meuller, r\Slash marchagainsttrump, r\Slash keep\_track, r\Slash russialago}).
\texttt{r\Slash the\_mueller} was a subreddit about the Special Counsel's investigation into Russian election interference. 
% As shown in \ref{fig:the.meuller}, 
the number of posts in these subreddits declined following the end of the investigation. However, this subreddit has survived by successfully adapted and now has several posts a day critical of Trump but not specifically about the Meuller investigation. Yet a similar subreddit, \texttt{r\Slash russialago} has declined to a much lower activity level (a few posts a week) but remains focused on Russian interference. By comparison, the number of posts in the generalist (but still left-leaning) \texttt{r\Slash politics} has remained relatively stable.  
 Niche width theory would predict that shifting to more general types of anti-Trump content may expose \texttt{r\Slash the\_mueller} to greater competition with other political subreddits. However, if it had not adapted it might have little reason to exist after the end of Mueller's investigation.

Theories of online community specialization can be empirically testable with better quantification of the ways that overlapping communities are different from one another.  These include features of content like choice of medium (text, images, video, links), content sources (what websites are they linking to?), types of participants with varying roles and styles of participation, and structures like policies, size and moderation.  Niche width theory additionally requires measuring environmental changes that may threaten the survival of communities. Observable events corresponding to interesting environmental variation may include crisis events, elections and the release or cancellation of entertainment products. Comparing the growth, performance, and ecological dynamics of overlapping communities during times of high or low change can test these theories and point toward design principles for online community scoping that account for the trade-offs in different types of specialization.


% Other studies in organizational ecology, and in biological ecology more generally, resource partitioning refers to how different groups specialize to minimize niche overlaps and avoid competition.   

\section{Ecological Implications for Production and Performance}

So far, the ecology of online communities has focused on understanding competition and mutualism among overlapping online communities.  An important limitation of this work has been to conceptualize competition and mutualism as dynamics related to the growth of online communities.  This follows biologists and organizational ecologists, but not all online communities have to grow in order to provide their intended benefits \citep{foote_starting_2017}.  An important step forward this research program will be to relate interdependence between online communities to outcomes besides growth that may be more directly connected to the value of the public goods that communities produce.

Quantifying the value of public information goods produced by online communities is a major methodological and theoretical challenge.  Much of the field of economics depends on the assumption that the utility of a good can be measured by its price.  Price is a valuable measure of value in economic theory because it is set by market mechanisms that align supply and demand.   Online communities are thought to be able to produce public goods because they can lower transaction costs \citep{benkler_coases_2002}. Negotiating a price in these settings is simply not worth it. A price will reintroduce transaction costs and undercut the pro-social motivations people have for contributing.  

Of course, this does not mean the public goods online communities produce are worthless. Estimates of the cost of replacing by paying editors a market rate placed its value between 6 and 10 billion dollars in 2013 \citep{band_wikipedias_2013}.  However, without a price mechanism, supply and demand may become ``misaligned.'' The quality of Wikipedia articles is uneven and the most popular content is often not the highest quality \citep{warncke-wang_misalignment_2015, gorbatai_exploring_2011}.
In classical economic theories, goods will be produced to meet the demand, but in peer production the size of an audience seems only weakly related to the level of production.
Explaining when online communities will produce high quality public goods like Wikipedia articles \citep{arazy_evolutionary_2019,arazy_determinants_2010,asthana_few_2018} or open source software \citep{champion_underproduction_2021} is thus  important to understanding the successes and failures of peer production.

% Critical mass theory can potentially explain how supply and demand can be linked in public goods production and can also be synthesized with ecology \citep{marwell_critical_1993}.  The central mechanism of the theory the notion of a ``production function,''  which maps a quantity of contributor input to a level of good produced. The theory proposes that the shape of the production function is determined by the collective action problem that a group must overcome to produce the good and determines the level of the good that will be produced by rational actors. Some prior research applies this theory to online communities, but does not operationalize its central propositions related to production functions \citep{solomon_critical_2014}. 

Critical mass theory offers to explain the conditions for successful collective action in public goods production and can also be synthesized with ecology \citep{marwell_critical_1993}. Many CSCW systems appear to require a critical mass of users to start or sustain their usefulness \citep{ackerman_intellectual_2000}.  The most important device in the theory is the \emph{production function}, which maps an individual's contributions to the value they get from contributing. The theory proposes that the shape of the production function is determined by the collective action problem that a group faces in producing the good.  If a production function is \emph{accelerating} (\emph{decelerating}) then a contribution increases (decreases) the payoff of the next contribution.
The rational actors in a group each have their own production function and together these determine the level of the good that they will produced. Some prior research applies this theory to Wikipedia, but does not attempt to measure value of contribution  or operationalize the theory's propositions about the relationship between production functions and collective action \citep{raban_empirical_2010, solomon_critical_2014}. Analyzing critical mass theory in the context of communal public goods production can also be an important theoretical contribution to communication theory \citep{fulk_connective_1996}. 

To illustrate, consider a hypothetical example of the construction of an online community for building a collaborative knowledge base, such as Wikidata. 
This can be cast as a collective action problem because the project can provide a wide range of benefits to a potentially large group of people, but no individual can provide the full range benefits alone \citep{marwell_critical_1993, fulk_connective_1996}.
Say a single individual, the community's founder who is an expert engineer and researcher, attempts to bootstrap the community by providing an initial design and implementation for the novel system, a small number of entries and by making efforts to publicize the community.
The founder hopes that others to join and contribute to constructing a valuable resource.

During this period in the community's development, the \emph{critical mass} consists of just the founder, who is motivated and capable of  in the hopes that others will see these contributions and subsequently make their own.  The founder has a large and unique set of resources enabling them pay the \emph{start-up costs} involved in founding the community when no one else would.  After these start-up costs are paid, others can make much more granular contributions like adding entries to the knowledge base. The founder hopes that others will perceive expected benefits from contributing that exceed the costs of contributing. 
% This might not happen and if time goes by and noone else contributes, the founder, all alone and discouraged, might conclude that it is not longer work making their own contributions. 
In theoretical terms, the founder hopes that the others' production functions are accelerating and paying the start up costs will move the others' production functions into a favorable region where they will contribute. 

% If some time goes by and noone else contributes, the founder, all alone and discouraged, might conclude that it is no longer worth making their own contributions.  Now the community has failed to hold on to a critical mass and becomes inactive. But say that the founder's early contributions have been useful to somone else (member 2) who chooses to make their own contributions because they expect to benefit from the ``warm glow'' of reciprocity, through social interaction with the founder, or from the future contributions that their own contributions might attract. 

Ecology has important implications for critical mass theory because important aspects of the collective action problem that influence the production function are related to the composition of the group and prior work suggests that individuals with varying experiences are important to online community growth \citep{kairam_life_2012}.
Heterogeneous groups are thought to be conducive to collective action because they are more likely to contain individuals who can contribute different things like start up costs or rare pieces of information \cite{fulk_connective_1996}.  
% This makes it easier to form a critical mass of individuals who can make start up contributions \citep{marwell_critical_1993}.

Returning to the example of a collaborative knowledge base, it is important to recognize that many contributions will involve \emph{articulation work} activities like documenting, answering questions, naming, and interpreting that are required to make the knowledge base work in practice \citep{schmidt_taking_1992, suchman_supporting_1996}.
Even though contributions of articulation work might not directly add new features or data to the knowledge base, they can be important to accelerating community members' production functions.
A heterogeneous community may be more likely to include members who are skilled at articulation work that benefits other members.
On the other hand, If different subgroups of a large community have sufficiently different application areas some articulation work might be specific to each subgroup. 
For example, biologists might make and document biology-specific norms for the collaborative knowledge base, but this would not be useful to physicists.
Thus individuals' production functions might depend most strongly on the other members of their subgroup when subgroup-specific articulation work is a limiting factor.

% At the same time, the utility of a collaborative knowledge overall often depends on linking to knowledge outside of one's domain of expertiseso the .






% elaborate on what a collective action problem is.
% develop the example of a distributed database and why it might be hard to do collective production of it at different stages or phases
% at different phases of developement the distributed database the critical mass needed to maintain the collective action dynamic is differnet in composition or in form.  
% Ecology is related to critical mass theory because important aspects of the collective action problem that influence the production function are related to the composition of the group. Heterogenous groups are thought to be condusive to production because they are more likely to contain individuals who will contribute very much and therby make it easier to form a ``critical mass'' of individuals who can overcome the start-up costs common in collective action problems \citep{marwell_critical_1993}. For example, a start-up cost for an open source database might take the form of an initial design for a novel system that can only be provided by expert engineers or researchers. But once the initial system is developed, additional features, bug fixes, and documentation can be added by a much broader group of developers who wish to use the system in their applications.  Therefore, open source community's ability to biuld a valuable system depends on including both database experts and application developers.  Prior work suggests that individual with varying experiences are important to online community growth \citep{T}.

I am starting work to find out how production functions help explain when online communities achieve critical mass and produce quality outputs and if relationships among communities influence the shape of production functions in ways that make collective action easier or more difficult in different conditions.
Measuring production functions requires the ability to precisely quantify the quality or value of individual contributions.
As a step in this direction, I have developed an improved measurement of Wikipedia article quality in research accepted for publication and included in Appendix A.
Prior article quality measures have been based on machine learning models that do not provide a continuous measure amenable to statistical analysis and that were miscalibrated for units of analysis like articles or projects.
Research using these measures has got around these problems by adopting an assumption that article quality levels on Wikipedia are ``evenly spaced'' from one another.
I use a method that relaxes this assumption, provides evidence that it is unfounded, and improves the accuracy of the models.

I have also done some methodological work on the ``demand side'' to understand how audiences use Wikipedia content.  Most prior work has been limited to measuring page views.  In Appendix B, I study the amount of time spent reading articles by Wikipedia visitors and find that readers in the Global South remain on pages for longer, especially in the last page view in a session. 
Although the measure used in that study may not be available for use in the future,  this work has prepared me for the time when better reading time data is available.
It will be interesting to see if the audience for an article relates to critical mass dynamics.

\section{Ecology and the Diffusion of Technologies for Community Governance}

Future ecological research can also look at the role of ecological dynamics in the emergence and diffusion of novel artifacts, technologies, information and ideas. Overlapping technology use in particular is a potential mechanism for specialization and mutualism.  I have previously suggested that sharing a host platform may not be sufficient for defining an organizational form because communities have considerable flexibility in making their own rules and configuring their own custom technology.  If sufficiently strong patterns are found in the sets of rules or technologies that communities adopt, these might justify treating communities sharing such structures as organizational forms or at least a potentially important kind of niche overlap. 

When online communities share technologies, this can create important forms of interdependence and collaborative innovation on tools is potentially an important type of mutualism.
For example \citet{chandrasekharan_crossmod:_2019} developed a system called ``Cross Mod'' for subreddits to collaborate on customizable machine learning models for monitoring misbehavior.
Smaller communities pooling data about rule violations can potentially build more accurate models than single communities can.   
Technologies like Cross Mod allow communities to select which other communities they wish to import data from and therefore are most useful when communities are institutionally compatible.
This suggests that sharing governance technologies may be a good proxy for an organizational form. 

However, as I found in Appendix C, my study of algorithmic flagging tools on Wikipedia, machine learning tools for predicting misbehavior may reproduce the biases of community moderators.
They can also improve the fairness of moderator judgments if moderators use the models instead of other biased social signals to find potential misbehavior. 
Additional risks may arise when algorithmic tools are shared by overlapping communities.
The learned norms and and standards of behavior from one community may not be appropriate  in other communities.
If shared flagging algorithms can more easily implement norms that are more widely held, the diffusion of an algorithm that makes regulating behavior easier and more predictable might mediate the diffusion of the norm.

The method I developed for the study in Appendix C provides a way to assess the consequences of a machine learning classifier without intervening in a community.
Future work at the intersection of ecology and online community governance might use this method in a study of the relationships between the performance of algorithms for enforcing different rules, the diffusion of the rules, and the growth and survival of communities having the rules.


\section{Microfoundations for Ecological Macrodynamics}

% Good chance this micro-macro stuff heads to the conclusion. Let's keep trying to make it work for now. These 3 paragraphs seem like a  good argument for a study that links individual behavior or user flows to competition/mutualism or density.

Predominant approaches in HCI and social computing and popular conceptions of social media platforms most often emphasize the role of managers of platforms in building online communities.
However, platforms have only a limited control over the ways that users build communities.
Furthermore, platforms struggle to maintain participants who may migrate to competing platforms.
Communities and their organizers can engage in collective action to protest platform's governance and design decisions \citep{matias_going_2016}.  
Online communities also form intermediate structures over which platforms have limited influence such as the widespread clusters of highly overlapping communities I identify in Chapter 2. 
An important goal of the ecology of online communities is to understand how patterns of action within individual communities are co-constitutive with the cultures and institutions of platforms.  
%Overlapping online communities exist because individuals participate in them, but individuals cannot participate in communities that don't exist. 
This goal faces a key type of puzzle in social science: to account for how  ``micro-level'' individual actors give rise to ``macro-level'' organizations, institutions, online communities, and cultures even as individuals are situated within these very structures.


% This was because taking up inter-organizational dependence as an object of study raised a similar micro-macro puzzle.
Micro-macro puzzles are not only found in the constitution of individual persons and the social structures they inhabit.
Organizational ecology takes up a different kind of micro-macro puzzle at the level of reciprocal dependence between organizations and the organizational fields or industries they comprise.
The performance of an individual organization depends on ecological dynamics in its organizational field, but the organization itself contributes to these very dynamics.
Initial work in organizational ecology avoided this reciprocal causation by minimizing the action of individual organizations.
Structural inertia constrained the agency of organizational actors, and external institutions, competition, and legitimacy constrained organizational performance. 

At first, organizational ecologists did not deny that factors internal to organizations matter to organizational performance.
Yet they argued that \emph{ceteris paribus}, the chances of an organization's survival depend on environmental conditions and on mutualistic and competitive pressures \citep{hannan_organizational_1989}.
Later on, organizational ecology began accounting for rational adaptation and failure of individual organizations \citep{baum_ecological_2006}. 
Recently they have incorporated the role of human cognition and social learning into their conceptualizations \citep{hannan_concepts_2019}, but as far as I am aware, empirical analyses have not stretched all the way from individual persons to inter-organizational dynamics.

Online communities provide a distinctive opportunity to connect individual behaviors to outcomes at the community and ecological levels thanks to the finely grained behavioral data that made possible the analyses in Chapters 2 and 4.
However, all of the measures used in these projects have aggregated the behavior of many individuals into measures of overlap or group size.
I have not shown how the ways that individuals navigate among overlapping online communities give rise to the ecological dynamics I find.
Aware of this limitation, I initially proposed constructing an agent-based model to theorize the micro-mechanisms of ecological dynamics.
Along the way, I found that talking to individuals provided a more valuable micro-level account of how and why people participate in overlapping online communities.

These interviews surfaced a conceptual model of a process by which new communities in a topical area spin-off specialists. 
An important direction for future research will be to operationalize and test this model with data.
This future work should look for inspiration from measures of individual behavior introduced in recent research in HCI and social computing \citep{tan_tracing_2018, tan_all_2015, zhang_understanding_2021}.
Specifically, \citep{tan_tracing_2018} provide a method to associate newly created subreddits with prior subreddits whose users join the new subreddit and measure the language use of individuals to characterize their similarity to the other members of the community. Also, \citet{waller_generalists_2019} quantify users of online communities as generalists and specialists based on their activity styles using embedding methods. 

\section{Focused Case Studies}

% find a better rationale than this?
% Why haven't we done this already? (b/c not as scientific?)
Finally, in order for ecological research in online communities to be useful to publics and practitioners, it will be important to conduct focused case studies of practical and popular interest.
Studies of the ecology of political communities, communities tying to make sense of the pandemic, ``meme stock'' and cryptocurrency communities, and pop culture fandom communities are all promising candidates.
A future project should investigate one or more cases in a mixed-methods study combining carefully constructed time series models for inferring ecological relationships and qualitative data in the form of grounded narrative accounts or interviews.

In conclusion, my research set out to understand interdependence among online communities through the lens of organizational ecology.
It has questioned the how well foundational assumptions of organizational ecology apply to online communities and set out to validate basic assumptions like when online communities will form competitive or mutualistic relationships.
It has provided new methods for studying competition and mutualism among online communities and shown that mutualistic relationships are more common than competitive ones because they last longer.
Although the question of how groups of mutualistic online communities are constructed remains open, selection process theories provide a starting point for future investigation.
Many applications of ecological theories and methods to important questions about the emergence, performance, and design of online communities are promising.
% As I continue my work, I am releasing well-documented code and datasets to support this future work and I hope, other research yet unimagined.

% bibliography here
\setcounter{biburlnumpenalty}{9001}
\printbibliography[title = {References}, heading=secbib]


% \end{document}                  %

%  LocalWords:  
%%% reftex-default-bibliography: ("ch5_conclusion.bib")