Audience and Theory of Change: Who Must Act, What They Must Believe, and Who Reaches Them

An advocacy page usually begins with a platform ("we need a TikTok") and works backward to a purpose. This lesson reverses the order. The research on how ideas and behaviors move through people tells you what a page can and cannot do; the Pew data tells you where the people you need actually are; and a theory of change turns both into a plan you can measure. Lesson 01 covered how feeds allocate attention once a post exists. This one is about the human network the post has to travel through, and who in it you are trying to move.

The two-step flow, then and now

In 1955 Elihu Katz and Paul Lazarsfeld published Personal Influence, reporting a study of women in Decatur, Illinois, about how they made decisions on public affairs, fashion, movies and household purchases. The finding that organized the field for fifty years was that media messages were mediated by informal opinion leaders "who intercept, interpret, and diffuse what they see and hear to the personal networks in which they are embedded." Lazarsfeld had stumbled on the pattern earlier in a 1940 election study; Decatur confirmed it across domains. The opinion leaders were not public figures. They were the neighbor who read the paper and told you what mattered in it, and they were different people for different topics.

The structure has survived the move online, with one change: the opinion leader can now be a stranger with an audience. Pew's America's News Influencers (November 18, 2024) studied 500 accounts with at least 100,000 followers on Facebook, Instagram, TikTok, X or YouTube that regularly post about news, and surveyed 10,658 adults. About one in five adults (21 percent) said they regularly get news from such influencers; among adults 18 to 29 it was 37 percent. Most of the influencers, 77 percent, had no affiliation or background with a news organization, and 85 percent had a presence on X. The two-step flow for a young voter in 2026 often runs from your page, through a regional or topical influencer who found it, to the voter, with no newspaper in the chain.

For a page, two consequences follow. First, your real audience is often smaller and more specific than your follower count: it is the people who pass things along, and you can identify them from your own data (who shares, who comments with context, who tags others). Second, when your post is carried by an opinion leader, it arrives with that person's interpretation attached. You can shape the interpretation by writing posts that are easy to summarize correctly, with the claim and the source in the first lines, as Anatomy of a Post teaches at the beginner level.

Weak ties, strong ties, and what kind of thing you are spreading

Mark Granovetter's 1973 paper The Strength of Weak Ties (American Journal of Sociology) argued that acquaintances, not close friends, are the bridges between otherwise separate social clusters, because close friends mostly know the same people you do. New information (a job opening, in his study) reaches you through weak ties. For a page, weak ties are the mechanism of reach: a reshare by someone whose friends do not overlap yours is worth more for exposure than ten reshares inside your own cluster.

Damon Centola and Michael Macy complicated this in Complex Contagions and the Weakness of Long Ties (2007). Information is a simple contagion: one exposure is enough, so bridges and long ties spread it fast. Behaviors that are costly, risky or socially uncertain are complex contagions: a person adopts only after seeing several contacts adopt, so the long tie that carries a rumor across town cannot carry a commitment, because the person at the other end sees only one adopter. Centola then tested this experimentally in The Spread of Behavior in an Online Social Network Experiment (Science, 2010), building online health communities whose network structure he controlled. Adoption of a health behavior "was much more likely when participants received social reinforcement from multiple neighbors," and the behavior "spread farther and faster across clustered-lattice networks than across corresponding random networks" with the same number of ties.

Almost everything an advocacy page asks for is a complex contagion. Reading a post is simple. Showing up at a Tuesday night planning and zoning hearing, signing a petition with your name on a public record, putting a sign in your yard, giving an email address: each carries cost or exposure, and people do them after they see others like them do them. That is why a page whose followers never see each other (a broadcast with no community) converts poorly, and why a page that shows its own members acting (photos from the last hearing, names of volunteers who consent, a count of signups) converts better. Bond et al., below, found exactly this: the version of a message that showed friends' faces moved votes, and the version that did not moved nothing.

Centola, Becker, Brackbill and Baronchelli added a threshold in Experimental evidence for tipping points in social convention (Science, 2018). In ten online groups of 20 to 30 people (194 participants in all), paid to coordinate on a naming convention, a committed minority consistently overturned the established norm once it reached a critical mass, which the paper puts at about 25 percent of the group and which the authors' institution summarized as roughly 25 percent. The setting is artificial and the groups small, so treat the number as a lab result rather than a rule for counties. The shape of the result, a floor below which committed effort fails and above which it flips the group, matches what organizers describe, and it argues for concentrating effort on one community at a time until it tips rather than spreading thin across many.

The 61-million-person experiment

On November 2, 2010, every US Facebook user over 18 who logged in was assigned at random to one of three conditions. Robert Bond and colleagues reported the result in A 61-million-person experiment in social influence and political mobilization (Nature, 2012). The social message group (n = 60,055,176) saw a banner encouraging them to vote, a link to find their polling place, an "I Voted" button, a counter of how many users had clicked it, and up to six profile pictures of friends who had. The informational message group (n = 611,044) saw the same banner, link, button and counter, with no faces. The control group (n = 613,096) saw nothing. The researchers then matched users to public voting records.

Users who saw the social message were 0.39 percentage points more likely to have voted than controls (s.e.m. 0.17, P = 0.02), and 0.39 points more likely than the informational group. The informational message had no effect: turnout in that group was identical to the control (treatment effect 0.00 percent, s.e.m. 0.28, P = 0.98). Everything the message accomplished, it accomplished through the faces. The indirect effect was larger than the direct one. For each close friend (measured by interaction frequency on the site) who received the social message, a user was 0.224 percentage points more likely to vote, and the authors estimated the message "increased turnout directly by about 60,000 voters and indirectly through social contagion by another 280,000 voters," a total of about 340,000 or 0.14 percent of the voting-age population. Nearly all the transmission ran through close friends, which the authors read as evidence that strong ties carry behavior even online.

Jason Jones and colleagues replicated the design in the 2012 presidential election in Social influence and political mobilization: Further evidence (PLOS ONE, 2017), and again found a significant increase in voting among recipients and among their close friends, with the total friend effect likely larger than the direct effect, in a high-salience election where mobilization effects are usually smaller.

What a page should take from a 0.39 point effect. First, it is small per person and large in aggregate, which is the normal shape of mobilization effects and the reason reach still matters even though persuasion is hard (lesson 09). Second, the active ingredient was social proof from close ties, not information. Your "the hearing is Tuesday at 7" post is the informational condition. Your post showing the eleven neighbors who came last time, with their permission, is the social condition. Third, the friends did not see the banner; they saw their friend's "I Voted" and the behavior it signaled. Design your asks so that doing them is visible to the doer's friends.

Influentials, susceptibles and ordinary spreaders

Sinan Aral and Dylan Walker ran a randomized experiment on 1.3 million Facebook users in Identifying Influential and Susceptible Members of Social Networks (Science, 2012) to separate influence (how much my adoption moves you) from susceptibility (how easily I am moved). Younger users were more susceptible than older; men were more influential than women; women influenced men more than other women; married people were least susceptible. The structural finding matters more than the demographics: influential people were less susceptible than others and clustered together, while susceptible people did not cluster. In plain terms, the people most likely to carry your message are the hardest to move, and they travel in packs. Finding one gets you near others.

Duncan Watts and Peter Dodds tested the broader "influentials" idea with network simulations in Influentials, Networks, and Public Opinion Formation (Journal of Consumer Research, 2007). Under most conditions they examined, "large cascades of influence are driven not by influentials but by a critical mass of easily influenced individuals." Eytan Bakshy, Jake Hofman, Winter Mason and Watts then measured real diffusion in Everyone's an Influencer (WSDM, 2011), following 1.6 million Twitter users and 74 million diffusion events over two months in 2009. The largest cascades did tend to start with users who had many followers and a record of past influence. But predicting which specific user or URL would start a large cascade was "relatively unreliable," and under a wide range of cost assumptions the most cost-effective strategy was to target "ordinary influencers," people with average or below-average influence, in large numbers. Goel et al. (2016), covered in lesson 01, closed the loop with a billion events: most popularity comes from one large broadcast, and structural virality is low.

The practical synthesis is unglamorous. One large account carrying your post (a regional outlet, a popular local page, a well-followed official) does more than any plan to "go viral." Below that, many ordinary people with average networks, each passing the post to a few friends, are the reliable engine, and they are reachable because they are your own followers. The strategy that fails most often is the one in the middle: courting a handful of mid-sized "influencers" and waiting.

Study	Design	Sample	Finding for a page
Katz and Lazarsfeld 1955	Panel survey, Decatur, Illinois	Women in one city (book-length study)	Media reaches most people through opinion leaders in their own circles, different leaders per topic
Granovetter 1973	Theory with survey evidence	Job seekers (AJS article)	Weak ties bridge clusters and carry new information
Centola and Macy 2007	Theory and simulation	n/a	Behaviors needing reinforcement spread poorly over long ties
Centola 2010	Randomized online communities	Artificially structured online health communities (Science, 2010)	Clustered networks beat random ones for behavior adoption
Centola et al. 2018	Lab coordination game	10 groups of 20 to 30 (194 participants)	Committed minority of about 25 percent flipped the norm
Bond et al. 2012	Randomized field experiment, validated votes	61 million Facebook users	+0.39 points turnout from social message; 0.00 from information; about 280,000 indirect votes via close friends
Jones et al. 2017	Replication, 2012 election	Facebook users (PLOS ONE)	Direct and close-friend effects replicated
Aral and Walker 2012	Randomized field experiment	1.3 million Facebook users	Influentials are less susceptible and cluster; susceptibles do not
Watts and Dodds 2007	Simulation	n/a	Cascades driven by many easily influenced people, not a few influentials
Bakshy et al. 2011	Observational, Twitter	1.6 million users, 74 million events	Predicting the next big spreader is unreliable; ordinary influencers most cost-effective
Pew 2024, news influencers	Content analysis plus survey	500 influencers, 10,658 adults	21 percent of adults, 37 percent of 18 to 29s, regularly get news from influencers; 77 percent have no newsroom background

Where the people are: Pew 2025

Pew Research Center's Americans' Social Media Use 2025 (published November 20, 2025) surveyed 5,022 US adults between February 5 and June 18, 2025 by web, mail and phone, with a margin of error of plus or minus 1.9 percentage points for the full sample. The headline shares of adults who ever use each platform: YouTube 84 percent, Facebook 71, Instagram 50, TikTok 37, WhatsApp 32, Reddit 26, Snapchat 25, X 21, Threads 8, Bluesky 4, Truth Social 3. Pinterest (36 percent) and LinkedIn (32 percent) were last asked in 2024, and Nextdoor (13 percent) in 2021. The demographic tables on Pew's Social Media Fact Sheet are the ones an Idaho page should keep open.

Platform (percent who ever use, 2025)	Ages 18-29	30-49	50-64	65+	Urban	Suburban	Rural	Rep/lean	Dem/lean
YouTube	95	92	85	64	85	87	79	84	85
Facebook	68	80	74	57	71	71	71	72	70
Instagram	80	62	40	19	55	54	37	49	53
TikTok	63	44	30	12	42	37	32	33	40
WhatsApp	37	40	30	20	44	34	17	25	38
Reddit	48	35	16	6	29	30	18	22	32
Snapchat	58	31	13	4	25	26	23	26	25
X	33	25	16	10	23	22	17	24	19
Threads	15	10	6	3	11	9	4	6	10
Bluesky	6	5	3	2	4	5	2	1	8
Truth Social	1	3	5	4	2	4	4	6	1

Source: Pew Research Center Social Media Fact Sheet, survey of US adults February 5 to June 18, 2025. Pew notes that not every difference between groups is statistically significant.

Read the rural column first if you work in Idaho. Facebook is flat at 71 percent across urban, suburban and rural adults. Every other platform falls off in rural areas: Instagram from 55 to 37, TikTok from 42 to 32, Reddit from 29 to 18, X from 23 to 17, Threads from 11 to 4, WhatsApp from 44 to 17. YouTube holds at 79. If your decision makers are county commissioners elected by rural voters, the platforms where those voters are reachable in numbers are Facebook and YouTube, and the rest are for specific segments. Read the age columns next. Instagram, TikTok, Snapchat and Reddit are platforms of the young; Facebook peaks at 80 percent among 30 to 49 year olds and is the only platform besides YouTube that a majority of adults 65 and older use (57 percent). The party columns are closer than most people expect: X is 24 percent Republican to 19 Democrat, TikTok 33 to 40, Bluesky 1 to 8, Truth Social 6 to 1, and Facebook, YouTube, Instagram and Snapchat are within a few points.

Pew's full report adds daily use, which matters more than ever-use for a page that posts several times a week. The daily-use questions came from a separate Pew survey of 5,123 adults conducted February 24 to March 2, 2025, reported in the same document. Among all adults, 52 percent use Facebook daily, 48 percent YouTube, 24 percent TikTok and 10 percent X. Daily Facebook use is 57 percent in rural areas against 49 percent urban, and 58 percent among 30 to 49 year olds. Daily TikTok use is 47 percent among 18 to 29 year olds and 5 percent among those 65 and older.

News use is a different question from use, and Pew's Social Media and News Fact Sheet (survey August 18 to 24, 2025) answers it. Fifty-three percent of US adults get news on social media at least sometimes and 21 percent often. The share who regularly get news on each platform: Facebook 38 percent, YouTube 35, Instagram 20, TikTok 20, X 12, Reddit 9, Nextdoor 6, WhatsApp 5, Threads 3, Bluesky 2, Truth Social 2, Rumble 2. Women are more likely than men to get news on Facebook, Instagram and TikTok; men are more likely on YouTube, X and Reddit; younger adults dominate Instagram, TikTok, X and Reddit news use. A striking pair: X is used by 21 percent of adults and is a regular news source for 12 percent, a high news ratio; Snapchat is used by 25 percent and was not among the twelve platforms Pew asked about as news sources in 2025.

Platform	Ever use (2025)	Regularly get news there (2025)	Who skews toward it for news (Pew)
Facebook	71 percent	38 percent	Women; adults without a college degree
YouTube	84 percent	35 percent	Men; Black, Hispanic and Asian adults; both parties about equally
Instagram	50 percent	20 percent	Women; younger adults; Democrats more than Republicans
TikTok	37 percent	20 percent	Women; younger adults; no college degree; Democrats more than Republicans
X	21 percent	12 percent	Men; younger adults
Reddit	26 percent	9 percent	Men; younger adults; Democrats more than Republicans
Nextdoor	13 percent (2021)	6 percent	Not broken out
Threads	8 percent	3 percent	Not broken out
Bluesky	4 percent	2 percent	Not broken out; 8 percent of Democrats use it, 1 percent of Republicans

Two cautions. These are national figures; Idaho skews older and more rural than the nation, so lean toward the rural and 50-plus columns when you have no local data. And Pew changed survey modes in 2023 (from phone to web and mail, then web, mail and phone), so compare 2025 figures with 2024 and 2023, not with 2021.

Building a theory of change

The term comes from Carol Weiss, writing for the Aspen Institute Roundtable on Community Change in a 1995 volume on evaluating community initiatives. Her diagnosis, recorded by the Center for Theory of Change, was that programs failed to show results because "the assumptions that inspire them are poorly articulated": nobody had written down how the activities were supposed to produce the outcome, so nobody tracked the intermediate steps. She used "theory of change" for "the set of assumptions that explain both the mini-steps that lead to the long-term goal" and the links between activities and outcomes at each step, and urged planners to map "the sequence of outcomes that are expected to occur" and build evaluation around whether they do. The Center's method is backward mapping: name the long-term goal, then the preconditions that must be in place for it, then the interventions that produce each precondition, then the indicators that show each one happened. The W.K. Kellogg Foundation's Logic Model Development Guide (2004) is the standard nonprofit workbook for the same exercise in inputs, activities, outputs and outcomes.

For an advocacy page the chain has a specific shape, because the goal is almost always a decision by a small number of identifiable people.

1. The decision. What, exactly, must happen, by whom, by when? "Three of five commissioners vote no on the rezone at the March 10 hearing." "The library board adopts the policy at its April meeting." "The council puts the bond on the November ballot by the August deadline." A page that cannot name the decision is a newsletter, which is a fine thing to be and a different thing to plan.
2. The belief or action that produces the decision. Decision makers respond to a few things: evidence they can cite, visible constituents, credible threat of electoral or legal consequence, and cover from peers. Which one does this decision need? A rezone vote often turns on the record (traffic, water, comp plan consistency) and on who shows up. A ballot placement turns on whether the council believes it will pass.
3. The audiences that reach the decision maker. Not "the public." The people the commissioner hears from and believes: residents of the affected precincts, the chamber, the irrigation district board, the pastors, the realtors, the newspaper, the two Facebook groups where the county argues. For each, write how they reach the decision maker (testimony, letters, a call from the board chair, a story).
4. The platforms that hold those audiences. Only now do you pick platforms, and you pick them from the Pew tables and your own local data. Rural residents over 50: Facebook and YouTube. Students at the college: Instagram and TikTok. The reporter and the policy crowd: X and, increasingly, Bluesky and Threads. The neighborhood that will testify: Nextdoor and a Facebook group.
5. The behaviors you need from each audience, written as complex or simple contagion. Knowing the hearing date is simple; one post does it. Testifying is complex; it needs social proof, a named organizer, a ride, a script and the sight of others committing.
6. The indicators, one per step, with a number and a date. Email list of affected-precinct residents to 400 by February 15. Fifteen committed testifiers by March 1. Two stories in the regional paper by March 5. Each is a leading indicator for the one after it.

Questions and answers

Our decision maker is a state agency, not an elected board. Does the theory of change still work?

Yes, with a different third block. Agencies respond to the record (comments filed in the docket, data, legal argument), to legislators who oversee their budget, and to press. The audiences that reach them are commenters, the relevant legislative committee members, and reporters who cover the beat. The platforms follow from who those people are, and they are usually X, Bluesky, LinkedIn and email rather than Facebook. The laws and rules course covers the comment and rulemaking process itself.

How do I find my opinion leaders if my page is small?

Export or scroll your shares and comments for the last six months and list every person who shared with their own words added, or who tagged someone else. On a page with 800 followers that list is usually 20 to 40 names. Katz and Lazarsfeld's opinion leaders were ordinary people defined by the role they played in their circle, and yours are too. Thank them privately, give them early access to documents, and ask what their friends ask them.

Is Bond et al.'s 0.39 percentage points worth the effort?

Across 60 million people it was about 340,000 votes including the friend effect. Across a county of 20,000 voters the same rate is about 80 votes directly and perhaps a few hundred through friends, which decides many local elections and most bond measures near the two-thirds line. The effect also came from a single banner on a single day. A sustained effort showing neighbors acting over weeks has no comparable published measurement, and you should not assume it scales linearly, but the direction is clear and the ingredient (visible action by close ties) is cheap to supply.

Should I target the "susceptible" people Aral and Walker describe?

Their finding is that susceptible people adopt easily and do not cluster, while influential people resist and do cluster. For a complex contagion you need both: susceptibles to build the visible early count, and the clustered influentials to make the behavior normal in their group. The practical reading is to seed in clusters (one road, one congregation, one school) rather than scattering, which is also what Centola's clustered-network result recommends.

Pew's numbers are national. How do I get Idaho or county figures?

Pew does not publish state-level platform shares. Your best proxies are the rural and age columns, your own page analytics, and the audience estimates Meta's ad tools give for a geography (which lesson 06 uses for paid reach). The census county profile gives you the age structure to weight the Pew columns by. Say "national rural figure" when you cite it, and do not present it as an Idaho number.

What is the difference between a theory of change and a content calendar?

The calendar says what you will post and when. The theory of change says why posting it should lead, through named audiences and behaviors, to the decision you want, and what you will measure to find out whether it is working. The calendar is derived from the theory of change, not the other way around. Lesson 10 builds a full 12-month calendar from a theory of change for an invented Idaho network.

Our issue is statewide. Do we still start from a decision?

Yes. A statewide issue resolves into specific decisions: a committee chair's choice to hold a hearing, a floor vote, a governor's signature or veto, a ballot measure qualifying and passing. Each has named decision makers and a date. Write a chain for the next one, and let the later chains wait; Weiss's warning about unexamined assumptions applies most to plans that try to cover a whole legislative session at once.

Why does the lesson keep saying "with consent" about showing members?

Because the social proof that moved votes in Bond et al. was friends' faces, and a page that publishes identifiable people at a political event without asking creates risk for them and for the page. Idaho's public records law makes hearing sign-in sheets public, but publishing a neighbor's photo and position on a contested matter is your choice, and the ethics code in lesson 09 sets the rule: ask first, every time.

Hard questions

Isn't a theory of change just consultant jargon for common sense?

The method is simple and the point of writing it down is Weiss's: programs fail because the assumptions linking activity to outcome are never stated and therefore never tested. A page that posts three times a week "to raise awareness" has an unstated theory (awareness leads to action by someone) that the Bond experiment contradicts (information alone moved nobody; social proof moved voters). Writing the chain exposes the weak link before the hearing rather than after. If a one-page chain feels like jargon, drop the term and keep the page.

The diffusion studies are about products, health behaviors and voting banners. Why trust them for advocacy?

Because they are the only randomized evidence on how behavior moves through online networks at scale, and because the mechanism they find, reinforcement from several close contacts, is the one organizers have described for a century. Bond et al. and Jones et al. are directly about political behavior with validated outcomes. Centola's experiments used health behavior but the design isolates network structure, which does not know what it is carrying. The honest caveats are that effect sizes will differ by setting and that none of these studies measured a sustained local campaign. Treat them as the mechanism and your own indicators as the measurement.

If close friends carry behavior, why build a public page at all instead of a private group chat?

You need both, and they do different jobs. The page is the broadcast layer: it reaches weak ties, gets picked up by larger broadcasters (Goel et al.), and gives reporters and decision makers something to find. The close-tie layer, where the complex contagion runs, lives in groups, chats and in person. A page without the second layer informs and does not mobilize; a group chat without the first never reaches beyond the people already in it. Lesson 04 designs the two layers as a network of pages and groups.

Doesn't targeting audiences by age, place and party mean you are manipulating people by segment?

Choosing where to speak is not the same as changing what you say to each group. The Pew tables tell you which platform a 60-year-old rural homeowner is likely to be on; they do not tell you to say something different to her than to a 25-year-old. The lesson's persona method requires the same facts and sources for every segment and records each segment's own words about the problem. Where the ethical line falls, and the law on disclaimers when you pay to reach a segment, is the subject of lesson 09 and the campaign finance course linked from lesson 04.

Pew's platform shares are a year old by the time you read them. Are they useful?

The shares move slowly. Pew's own trend shows TikTok rising from 21 percent of adults in 2021 to 37 in 2025 and Instagram from 40 to 50, while Facebook and YouTube held steady; the rural and age gaps have been stable for years. A year-old figure is a far better basis than a vendor's blog post or a hunch, and the lesson dates every number so you can replace it when Pew publishes the next round, usually in the fall.

The 25 percent tipping point came from 194 people in a naming game. Why mention it?

Because the shape of the finding, not the number, is the lesson: a committed minority below a threshold fails and above it flips the group, so effort concentrated in one community until it tips beats effort spread across ten. The lesson labels the number a lab result and tells you not to apply it to counties. Centola's book-length treatment and his 2010 field experiment make the same argument with more external validity, and your own indicators (how many on one road have committed, not how many across the county) are the version that applies to you.