Understanding Public Engagement With Global Aid Agencies on Twitter: A Technosocial Framework

Twitter and Its Affordances

A decade ago, boyd and Ellison (2008) defined SNSs as

web-based services that allow individuals to construct a public or semi-public profile within a bounded system, articulate a list of other users with whom they share a connection, and view and traverse their list of connections and those made by others within the system. (p. 211)

We broadly agree with this frequently employed definition, but would add two vital features that it excludes. One, SNS users are not just “individuals;” they may also be “institutions” of various types, from news channels and private companies to political parties, interest groups, nongovernmental organizations, and aid agencies (Gil de Zúñiga, Jung, & Valenzuela, 2012). Two, individual and institutional users do not simply “traverse their list of connections” on SNSs but perform a wide range of social actions that engage these connections in varying degrees—indeed, one can say that the purpose of forging social connections on SNSs is to engage them in such actions (Hoffmann, Lutz, & Meckel, 2016).

An affordance is “an action possibility available in the environment to an individual, independent of the individual’s ability to perceive this possibility” (McGrenere & Ho, 2000; see also Gibson, 1979). Each SNS creates its own environment—or what boyd and Ellison called “bounded system”—with its own set of affordances, although a number of affordances are common to many SNSs. Twitter, launched in 2006, has become one of the most popular SNSs worldwide on account of its wide range of affordances (Chaffey, 2016). Twitter users, called tweeters, can publish short posts of up to 280 characters, called tweets, which pop up on the Twitter feeds of other users. Posting tweets with textual or audio-visual content is therefore one of Twitter’s most basic affordances. Tweets can include hashtags signifying keywords ( . . . #keyword . . . ) that allow other users to easily find their tweets, as well as hyperlinks ( . . . https:// . . . ) to other websites. In addition, users can reply to other users (@username . . . ), or they can retweet other users’ tweets (RT @username . . . ).

Posting of tweets, hashtags, hyperlinks, replies, and retweets constitute Twitter’s affordances, or “action possibilities” that exist within the Twitter environment irrespective of users’ knowledge or perception. However, each of these affordances can have multiple interpretations, which are more perceptual in nature. For instance, publishing a tweet can be a means of sharing information—but it can also be a means of forging a bond with one or more users on Twitter, participating in a protest and so on (Bennett & Segerberg, 2013). Similarly, retweeting another user’s tweet can suggest endorsement of the views expressed in the tweet—but it can also imply quite the opposite (Metaxas et al., 2015). Thus, while Twitter provides a number of action possibilities or affordances, these affordances may be used for multiple purposes and can have multiple social meanings (Evans, Pearce, Vitak, & Treem, 2016; Rathnayake & Suthers, 2018). It is therefore apt to consider these affordances as not just technological but “technosocial” in character. We will begin our empirical inquiry by investigating,

Networked Participation

Global aid agencies aim to promote social change, or “the reduction of social inequalities and the provision of opportunities particularly for people who have few choices” (Waisbord, 2015, p. 146). Instead of modular one-size-fits-all solutions for different societies, social change favors the adoption of multiple, culturally sensitive solutions that address a wide range of issues, including poverty, race, gender inequality, health, education, and so on (Pillai & Maleku, 2015). This requires a bottom-up approach—based on listening to local communities and understanding their concerns from within before coming up with strategies to deal with those concerns (Drydyk, 2005).

Public participation is a “political . . . process where the actors involved in decision-making processes are positioned towards each other through power relationships that are (to an extent) egalitarian” (Carpentier, 2012, p. 175). Carpentier adds that the participatory process unfolds through the juxtaposition of access and interaction. Organizations need to offer citizens access by ensuring their presence within the organizational structure, including media production spaces. Interaction, meanwhile, relates to the “social–communicative relationship” that organizations establish with public participants.

In other words, public participation is the process by which organizations transform from institutions to networks. An institution “refers to macro-political forms of power that primarily rely on vertical organization and concentrate decision making and ideological production in particular positions” (Servaes & Hoyng, 2017, p. 260). As a result, institutions typically have low levels of public participation. In contrast, networks are “micro-political forms of power that primarily rely on lateral communication and interaction” (p. 260). Public participation is higher in networks. But network topology varies: some may be more centralized and others more dispersed. Our study will therefore examine,

Affordances and Participation

Yang and Ott (2016) identify two “governing rules for participation”: the market norm, under which organizations offer people monetary incentives to participate, and the social norm, under which people participate for “maintaining communal relationships with others” on grounds such as altruism, politeness, and reciprocity (p. 833). Their experimental study argues that individuals can be just as motivated by the social norm as they are by monetary compensation. However, “unless people feel properly recognized and acknowledged (by the organization), they will not take actions to participate purely on account of altruism or reciprocity” (Yang & Ott, 2016, p. 839). These findings substantiate long-standing views in public relations about “recognition” and “appreciation” being the key for people to participate in communal activities (Clark & Mills, 1979).

Studies of organizations built around digital technologies, such as Wikipedia and benevolent hacker movements, also indicate that organizational practices and communication strategies directly influence public participation (Fieseler & Fleck, 2013). Organizations that are more open, transparent, and willing to delegate decision-making power to the public are able to generate higher levels of participation (Bandura, 1977). Bowen et al. (2010) describe these approaches as marking a shift “from one-way information sharing, through two-way dialogue and collaboration, to community leadership or empowerment” (p. 304).

This shift corresponds with Grunig and Hunt’s (1984) classic typology of public relations models. The typology includes two models of “one-way communication”: the press agentry model that relies on propaganda by any means through mass media and other channels, and the public information model that typically provides accurate information but also holds back negative information from the public. These models are typically not very effective. As organizations realized this, they turned to dialogic or “two-way communication,” which is also of two types: the asymmetric model, where organizations use dialogic strategies instrumentally to change public behavior for their own benefit, and the symmetric model, in which organizations communicate with the public in order to develop shared meanings for a mutually beneficial relationship. Essentially, the shift from one-way communication to asymmetric and finally symmetric communication reflects the transformation of an organization from an institution into a network through the devolution and decentralization of power.

Several scholars have employed Grunig and Hunt’s (1984) typology to understand the use of SNSs like Facebook and Twitter by nonprofit organizations. Cho, Schweickart, and Haase (2014) content analyzed the Facebook posts of 36 U.S.-based nonprofits. They found “high levels of engagement with organizational messages when two-way symmetrical communication is used, compared to public information or two-way asymmetrical models” (Cho et al., 2014, p. 565). But despite the effectiveness of two-way symmetric communication, most studies suggest that nonprofilts typically employ the one-way models or the asymmetric model. For instance, in Lovejoy, Waters, and Saxton’s (2012) analysis of tweets posted by 73 nonprofits, only 20% of tweets demonstrated conversations with the public and only 16% attempted to make indirect connections with particular Twitter users. They concluded that “nonprofits are primarily using the site to relay information using one-way communication” (Lovejoy et al., 2012, p. 316). Similarly, Waters and Jamal’s (2011) study found that “[r]ather than capitalizing on the interactive nature and dialogic capabilities of the social media service, nonprofit organizations are primarily using Twitter as a means of sharing information instead of relationship building” (p. 323).

This body of research makes it obvious that the manner in which an aid agency uses an SNS like Twitter can shape the level and quality of public participation with the agency. In this study, we examine this relationship at the level of affordances so as to better understand how each affordance influences public participation. This enables us to develop a multidimensional technosocial framework that (a) specifies the relationship between different technosocial affordances of Twitter and public participation and (b) helps strategize how each affordance may be used by aid agencies to improve the level and quality of public participation. Therefore, we finally ask,

Research Design and Data Sampling

This study uses a “most different systems” design of comparative inquiry (Przeworski & Teune, 1970). Systems are understood as spatially or temporally defined networks of social relationships. The specific systems—or cases—examined in such a study are of the same type (e.g., nation–states, social networks) but differ substantially in terms of their sizes, structural features, and other properties. The purpose of the empirical analysis is to match the variation across systems in the values of a particular property with a corresponding variation across systems in the values of another property. Formally speaking, in a study of three most different systems, say “1,” “2,” and “3,” with properties “a,” “b,” “c,” . . . “z” that have very different values in each system, if y₁ > y₂ > y₃ and x₁ > x₂ > x₃, then “x” is likely to be the explanation for “y” as there are no other parallel relationships. In other words, the vast differences across the three systems serve as a “control” for other possible explanations of “y.” It means that any of these systems could increase the value of “y” by raising the value of “x.” Moreover, the relationship between “x” and “y” is expected to be generalizable to other systems. Of course, the relationship should be theoretically and logically explicable, or else it would simply be an example of spurious correlation.

In this study, systems are defined as temporally bound Twitter networks formed around the handles of particular aid agencies. Each system comprises two components: tweets posted from that agency’s Twitter handle (outbound tweets) as well as tweets posted by other Twitter users in which that agency’s handle is “mentioned” (inbound tweets). By comparing the agencies’ use of various Twitter affordances in their outbound tweets (Research Question 1) and then comparing public engagement with these agencies in terms of various network variables in their inbound tweets (Research Question 2), our analysis aims to explain how differences in the use of Twitter affordances matched with differences in the level and degree of public participation (Research Question 3).

We chose USAID, SIDA, and ICRC as the three cases for our study. While each of them is a major aid agency engaged with projects of global social change, they are also quite different from each other. USAID is a U.S. government agency that administers civilian foreign aid. Created in 1961, it works closely with various divisions of the U.S. government and counts “the promotion of U.S. foreign policy interests” as a key objective. As of 2016, it had a funding of $22.7 billion and staff working in more than 100 countries around the world (USAID, n.d.). SIDA was created in 1965 and works under the Swedish ministry for foreign affairs. It focuses on 33 partner countries around the world and is funded by a “guarantee portfolio” worth ~$368 million (SIDA, 2015). ICRC is a Geneva-based international organization that works with multiple governments and nongovernmental organizations worldwide and has staff in more than 80 countries. Created in 1863, its 2015 budget was $1.85 billion (ICRC, n.d.).

Tweets were collected over 3 months (from July to September, 2016) through the social data mining platform Netlytic.org, which uses the Twitter REST API v1.1 (Gruzd, 2018). The API has a cap of 1,000 tweets per 15 minutes, which is well above the frequency at which the tweets were posted (Gruzd, Mai, & Kampen, 2016). The corpus comprises six data sets. Three of these are outbound data sets of tweets posted by USAID, SIDA, and ICRC’s primary Twitter accounts—@usaid, @sida, and @icrc. These include 756, 466, and 513 tweets, respectively. The other three are inbound data sets of public tweets that mentioned any of these three Twitter handles in their texts. These include 40,100 (mentioning @usaid), 5,072 (mentioning @sida), and 50,568 (mentioning @icrc) tweets. The two SIDA data sets (outbound and inbound) had a high proportion of tweets in Swedish, which were translated into English using Google Translate’s machine translation system. Although the accuracy of the translation system varies by language pairs, the English←→Swedish translation, as tested against manual translation by native speakers in a comprehensive study, is the third highest of all language pairs (Aiken & Balan, 2011).

Analysis of Outbound Tweets

The outbound data sets were relatively small in size and so were studied inductively using Glaser and Strauss’s (1967) method of constant comparison with the help of the software NVivo. Our analysis aimed to understand if the agencies were using Twitter’s technosocial affordances—content, replies, retweets, hashtags, and hyperlinks—to transmit information about themselves or to meaningfully engage with communities and build participatory social networks. As each tweet would typically focus on a particular issue, special attention was paid to the frequency of use of keywords and the specific contexts in which they were used. One of the authors carried out the initial coding of all outbound tweets and collected descriptive information regarding hashtags, hyperlinks, replies, retweets, as well as the topic of the tweets. Both authors then discussed the coding and agreed to organize the analysis along three dimensions that form the technosocial framework developed in this study.

Analysis of Inbound Tweets

While each agency’s Twitter handle is the hub of its social network, the network itself is constituted of inbound tweets that mention the agency’s Twitter handle. Two different methods were used to analyze the three inbound data sets.

Aid Agencies’ Use of Twitter’s Affordances

The textual analysis of outbound data sets, which represented how the aid agencies used Twitter’s technosocial affordances (Research Question 1), revealed three broad dimensions across which they could be compared.

Replies and Retweets

Just over 6% of USAID’s tweets were replies to other users (see Table 1). Replies mostly addressed other official accounts, such as USAID administrator @GayleSmith, rather than members of the general public—although the exact proportion was difficult to determine as all users do not include such information in their profiles. In addition, over 26% of USAID posts were retweets—again mostly of tweets originally posted by official accounts such as @GayleSmith, @USAIDAfrica, @USAIDMiddleEast, @USAIDEducation, @WhiteHouse, @POTUS, @StateDept, and so on. For instance, a retweet posted on July 6 read, “RT @WhiteHouse: We can end global poverty and hunger within our lifetimes. —@POTUS on Congress’s vote to support @FeedtheFuture.”

OPEN IN VIEWER

Table 1.

Comparison of Replies, Retweets, Hashtags, and Tweet Topics in Outbound Tweets.

Agency	Number of tweets	Replies, %	Retweets, %	Most-used hashtags	Tweet topics
USAID	756	6.2	26.6	#globaldevelopment (139) #letgirlslearn (52)	girls/women (164) food security (43)
SIDA	466	8.6	72.7	#Almedalen (86) #sverigeivarlden (49)	water (66) sustainability (37)
ICRC	513	0.1	34.3	#Syria (43) #Iraq (36)	refugees/displaced (72) conflict/war (61)

Note. USAID = U.S. Agency for International Development; SIDA = Swedish International Development Cooperation Agency; ICRC = International Committee of the Red Cross.

SIDA replied more often (8.6%), and its replies also appeared to address members of the public more frequently than USAID. For instance, a reply posted on September 15 read, “@robinalm Det vore kul om det här nätverket spred sig till andra länder . . . ” (@robinalm It would be great if this network spread to other countries . . . ). The tweeter replied to, @robinalm (Kung Robin), is a Stockholm-based former music producer according to the profile information. SIDA also retweeted more than USAID—indeed, nearly three fourths of its posts were retweets (339 of 466, or 72.7%). The original tweets came from both official and public accounts.

ICRC’s outbound data set was the weakest in terms of replies (0.1%). Its retweets (34.3%) were proportionally higher than USAID, but the original tweets almost always came from other ICRC accounts (e.g., @ICRC_Africa, @ICRC_NYC, @ICRC_uk, @ICRC_ye). For instance, a July 10 post read, “RT @ICRC_Africa: 3/3 Hospitals, humanitarian workers must be protected from attacks by all parties. #SouthSudan.” Another retweet, on September 30, said, “RT @ICRC_ye: More people can learn how to walk again now after our completion of the new training area at Sanaa’s Physical Rehabilitation Center . . . ”

Public Engagement With Aid Agencies

Social Network Analysis

The network graphs indicate that the inbound data set of tweets mentioning each aid agency is of a different type—an inference supported by the quantitative properties of these networks. USAID’s network has a number of small groups, represented by different colors in the network graph (Figure 1). It also has the highest modularity of the three data sets, followed by SIDA and ICRC (Table 2). This implies a number of conversations are taking place around it, constituting multiple online communities. Participants tweet not only to USAID but also to other participants within each of these communities, although not quite as much to members of other communities in the network.

OPEN IN VIEWER

Figure 1.

USAID’s inbound social network.

OPEN IN VIEWER

Table 2.

Comparison of Social Networks in Inbound Tweets.

Agency	Diameter	Centralization	Reciprocity	Density	Modularity
USAID	99	0.412	0.049	0.000137	0.490
SIDA	20	0.387	0.128	0.00115	0.469
ICRC	311	0.464	0.0155	0.0000757	0.442

Note. USAID = U.S. Agency for International Development; SIDA = Swedish International Development Cooperation Agency; ICRC = International Committee of the Red Cross.

In SIDA’s network graph, all participants are closely wound around its Twitter handle and there are only a few isolated nodes (Figure 2). It also has a relatively higher density and reciprocity (Table 2). This suggests that SIDA’s network is highly interconnected, with most participants tweeting not only to SIDA but also to other participants across the network and engaging in two-way conversations with each other. It indicates that SIDA’s network is highly participatory.

OPEN IN VIEWER

Figure 2.

SIDA’s inbound social network.

ICRC’s network, on the other hand, has many isolated participants tweeting to ICRC but not to each other (Figure 3). Its high centralization implies very few nodes dominate the conversation—but low modularity means other participants do not constitute subgroups either (Table 2). Low density and low reciprocity means these participants are not interconnected and they are not having reciprocal conversations with each other. Members of the network are simply relaying messages to ICRC, but there is little exchange of ideas among them—or even with ICRC. Such a network is not very participatory.

OPEN IN VIEWER

Figure 3.

ICRC’s inbound social network.

Topic Modeling

While social network analysis helped us understand the types of networks formed by the aid agencies’ inbound tweets, topic modeling explained the semantic structure of inbound tweets—identifying the key issues discussed in their content and how these issues varied in proportion in each network. The USAID data set covered a fairly broad range of issues. Women, children, and climate change emerged as the dominant issues, featuring in the two most prominent topics (Table 3 and Figure 4). Health, nutrition, youth, and education were of secondary importance, while issues such as water, food security, poverty, and disease were marginalized. Region-wise, Africa emerged as an important concern.

OPEN IN VIEWER

Table 3.

Topic Model of USAID’s Inbound Tweets.

Climate change	Global goals	Food/health	Africa	Praise for USAID	Girls/women
worldbank	usaid	hlthpolicyplus	working	people	insightful
usaid	today	usaid	water	work	usaid
amp	opendemocracy	support	part	juanorlandoh	usaidethiopia
zyyuxp	learn	health	humanitarian	access	globaldevelopment
wpd	global	education	nomorentds	announces	potus
usaidgh	amp	youth	change	barefootboomer	women
whitehouse	development	feedthefuture	impact	donate	letgirlslearn
chains	promotes	food	somalia	xaovwrhb	girls
afdb	familyplanning	amp	ntds	challenge	africa
data	globalgoals	statedept	shared	development	gaylesmith
potus	del	helping	endpoverty	proud	summit
communities	story	aid	mosquito	investments	world
join	letgirlslearn	read	usaidafrica	proposal	children
climate	equal	genderequality	cgiar	leadership	partners
million	building	security	devex	usaidasia	families
partnership	nutrition	offer	diseases	amazing	countries
stronger	innovation	project	foodsecurity	launched	hiv
program	great	futures	strengthen	usaid	simple
lives	community	youthday	apoyando	stories	globaldevel

Note. USAID = U.S. Agency for International Development.

OPEN IN VIEWER

Figure 4.

Proportion of use of different topics in USAID’s inbound tweets (%).

SIDA’s inbound data set focused on a much smaller and different range of issues. Water was by far the most important issue, followed by sustainable development and global development (Table 4 and Figure 5). Women and gender issues emerged as a secondary concern. Like the USAID data set, though, Africa was a key area of interest.

OPEN IN VIEWER

Table 4.

Topic Model of SIDA’s Inbound Tweets.

Global goals	Projects/partners	Africa	Water	Sustainable development	Praise for SIDA
gender	miljobloggen	kenya	sida	latest	working
agenda	safe	kizyjs	oecd	amp	securingwater
charlottepetrig	based	sharing	sweden	almedalen	innovations
great	nation	rural	water	mittm	interesting
devtalks	ppl	region	development	charlottepetrig	anders
usaid	generaldirekt	swe	today	page	aids
social	moving	campaign	wwweek	development	lead
chair	sid	companies	sdg	anamgren	inspiring
aid	voices	sid	work	women	conflict
day	independientes	brottsanalys	support	martin	staffan
stockholm	provided	martinsen	global	sdgs	political
utrikesdep	stop	power	https	sustainable	network
choose	wastewater	wateraidsverige	peace	global	play
globalgoals	camillabruckner	education	important	goals	eu
dev	application	textile	swemfa	ehstad	mqos
meeting	response	penalties	africa	learn	discussing
happy	oernerot	note	dac	eainequalities	age
sustainability	words	bangladesh	world	programme	mariaberlekom
fordfoundation	couple	implementation	sverigeivarlden	supporting	ulrikahs

Note. SIDA = Swedish International Development Cooperation Agency.

OPEN IN VIEWER

Figure 5.

Proportion of use of different topics in SIDA’s inbound tweets (%).

Tweets in ICRC’s inbound data set focused on a broader range of issues than the SIDA data set, but not quite as many as the USAID data set (Table 5 and Figure 6). The dominant issues—conflict, refugees, and disappeared people—were quite different from the other two data sets. There was a special emphasis on conflict zones such as Syria, Yemen, Iraq, and Kashmir, as well as on the impact of conflicts on families. There was also some concern over nuclear weapons.

OPEN IN VIEWER

Table 5.

Topic Model of ICRC’s Inbound Tweets.

Kashmir	Conflict/displaced	Nuclear weapons	Attacks on doctors	Syria/Yemen	Impact on families
icrc	icrc	unrest	icrc	ring	bone
family	hospitals	joint	unhumanrights	icrc	reflin
syria	antibiotic	streets	federation	babatamim	icrc
camp	people	situations	supplies	humanitarian	attack
kashmir	amp	work	joint	syria	people
fire	displaced	nuclear	people	amp	children
year	conflict	staff	attacks	aleppo	today
workers	syredcrescent	icrc	devastating	water	health
long	helping	forced	back	syredcrescent	aleppo
care	disappeared	support	msf	usadestroysyemeneconomy	yemen
situations	million	convoy	refugees	call	humanity
inside	crisis	missing	es	statement	aid
kids	iraq	sos	humanity	kids	families
lack	humanitarian	wounded	hiroshima	yemen	lives
separated	syria	days	stop	aid	war
bring	convoy	colleagues	doctor	concerned	flee
piliers	killed	nagasaki	respond	violence	amp
working	years	rlb	suffering	food	week
death	feel	iraq	facilities	world	rmardiniicrc

Note. ICRC = International Committee of the Red Cross.

OPEN IN VIEWER

Figure 6.

Proportion of use of different topics in ICRC’s inbound tweets (%).

Relationship Between Affordances and Public Engagement

The analysis indicates that each agency’s outbound and inbound data sets—or how an agency uses Twitter and what sort of social networks come into being around it—are closely interlinked (Research Question 3). Three broad relationships can be identified. First, the more an aid agency replies to or retweets the public, the more participatory social networks it creates. SIDA’s outbound data set has the highest proportion of replies and retweets. Correspondingly, its inbound data set forms the most dense, decentralized, and reciprocal social network. USAID’s midrange proportion of replies and retweets corresponds to the midrange density, decentralization, and reciprocity of its inbound social network. And ICRC, with a very low proportion of replies and retweets of the public, also has the lowest levels of density, decentralization, and reciprocity in its inbound network. These results imply that an aid agency’s willingness to reply to tweets addressed to it and retweet members of the public can lead to more participatory social networks in which members of the public often mention and reciprocate each other, and a larger proportion of participants drives the conversation.

Second, a similar relationship can be observed between an agency’s proclivity to use publicly oriented hashtags and hyperlinks and the participatory potential of its inbound social network. Once again, SIDA emerges as the agency most likely to use hashtags not related to official programs and hyperlink to nonofficial pages—corresponding with its highly participatory social network. On the other hand, USAID and ICRC often use hashtags of official programs and hyperlinked to official web pages—and their inbound social networks are not as participatory. These relationships indicate that if an aid agency uses Twitter’s structural elements—replies, retweets, hashtags, and hyperlinks—to directly engage in conversations with the public rather than to simply transmit information about itself and its programs, it is likely to build more participatory social networks.

Finally, there is a broad consonance of tweet topics in each agency’s outbound and inbound data sets, suggesting that the specific issues an agency focuses on in its tweets tend to be the same issues that the public tweets to it about. Some of the issues that emerged as important in our data may well have been on account of the time period of data collection—for instance, most ICRC tweets about nuclear weapons related to the 20th anniversary of the release of the International Court of Justice’s advisory opinion on the legality of nuclear weapons, which fell on July 8, 2016. Nonetheless, the topical connection between each agency’s outbound and inbound data sets is apparent.