Riki Conrey, Audience Research

Make One Friend

What real data tells us about who heals America's social network

scroll

Most of us know by now that American's social connection is declining. We've been hearing about this since the turn of this century (Bowling Alone), and there has been a surge in urgent reporting about an "epidemic of loneliness." Many groups are affected: older people, younger people, noncollege men, and more.

A decrease in our connections, especially in the weak or casual connections that create job referrals, dates, and a perception of shared humanity, is a really big deal. Social scientists will tell you that it's not access to education or resources that creates mobility from poverty; it's access to rich people. It's not reading instructions on decreasing our implicit bias that decreases racism; it's casual, positive connections across races. And no amount of listening to podcasts about "how the other side thinks" will solve our polarization; it's going to take actually knowing people in the opposite party.

This is a big deal, and it's a growing problem. So far we all agree.

And I have seen a surge of interest in solutions. Funders, organizers, and advocates wonder "what should we invest in that will help?" Who should we teach to make new friends? I wanted to know too, so I used real data and some simulations to find out who should get out there and make new friends.

Friends
—
avg, all groups
Bridge Recovery
—
of damage healed
The Network

I used real data and some simulations to build this network. In 2008, Pew ran a survey asking people to list up to 10 people with whom they discussed important topics.

These 6 circles represent men and women in three age bands. Bigger circles are groups with more friends on average. Thick lines represent lots of cross-linkage between groups.

The Damage

Over many years, the links have been breaking. The General Social Survey asked the same questions in 1985, and the rates of decline are real and measurable. We can carry them forward to today.

Watch the nodes shrink and the lines thin out.

Young Men

So who do we invest in? When I talk to funders and advocates, the first answer I hear is young men. They've lost the most. Let's help them make new friends.

The important thing to know is that people don't just meet each other; they are introduced. You meet people through the people you already know—friends of friends.

Young men's friend count goes up. But look at the lines. When young men get introduced through their friends, they meet…more young men. That's clumping, not healing.

The stat card in the corner tracks bridge recovery—how much of the damage between groups gets healed. That's the number that matters, because cross-group connections are where we find job referrals, depolarization, and what plenty of young men are looking for: dates with young women.

Middle-Aged Women

Now watch what happens when we invest in middle-aged women instead.

Her friends know young people, older people, men and women. When she makes one new connection, the introduction ripples across the whole network, not just back into her own group—and that means bridge recovery is greater: cross-group connections are restored.

The nodes get bigger and the lines between groups get thicker. That's the difference. Young men added friends for themselves. Middle-aged women are healing the connections between everyone.

Watch the bridge recovery number. It's not close.

Married Women

Why women? Because in America right now, middle-aged women are allowed to talk to everyone. A married woman chatting with a stranger at a school event or a neighbor's kid at a barbecue is normal. A young man doing the same thing is not. I hope this changes, but it's true today.

Marriage makes this even stronger. It's social permission plus a built-in bridge to her husband's network. Here's what happens when we invest in married women of all ages.

More bridges light up because the institution reaches across age groups—younger married women, middle-aged married women, older married women all have this same structural advantage. Middle-aged women still win overall because they simply have more connections, but marriage as an institution is a powerful multiplier.

Everyone

I tested every group. No group of men turned out to be uniquely important (you can try them all below). But there was one investment that healed the network almost as well as middle-aged women: everyone.

No one group singled out. Just everyone making one new friend through someone they already know.

Make One Friend Today

That's it. That's the magic solution to the problem: everyone needs to make one friend. Here, it's important for me to specify the kind of friendship we're talking about. The friends we all need to make don't need to be our closest; we don't need to be vulnerable with them. They need to be available when we need something like a job referral, a recipe, or an example of a Republican who actually tells really good jokes.

These new friends also don't need to be of any specific type. Rich people don't need to seek out poor friends. Men don't need to seek out women. The network itself takes care of all that. If we all meet someone who knows someone we know, the whole network gets tighter and closer for everyone.

A Final Note about Institutions

I know that the reason so many of my influential friends are eager to invest in other groups' social skills; it's that making small talk with strangers is hard and scary, and we don't want to do it!

We don't have to. Society made a solution to this, and that solution is institutions. I said that we only meet through introduction, and that's true, but the introducing entity is often an institution: a set of rules which guide shared work. Marriage is an institution. School is an institution; the gym is an institution; my kids' swim team volunteer structure is an institution.

The trick of the institutional setting is that we engage in parallel, scripted labor; we follow a story together. That makes it possible to interact with very tight guardrails and zero social risk. It's pretty hard to screw up social interactions of the form "that test was hard," "how was class today; I saw there were burpees," or "how do you use these stopwatch buttons."

Those interested in how we're going to get everyone (including you) to make one friend should know that this can't be an advertising campaign; it's an IRL organizing campaign which creates millions of opportunities for zero-risk social interactions in the course of shared work.

Don't believe me?

Ask the Data

Connections

Their group —
Everyone else —

Bridge Recovery

—
of damage healed
Methods & Data — For Serious Nerds ▼

AI helped with this

I used Claude Code to implement the statistical models I designed, build the network matching algorithm, and present these results on this website. I didn't use Claude to design research questions, interpret results, or write copy. Because AI is very bad at knowing what's important and is even worse at writing compelling prose.

Data sources

Pew Personal Networks Survey (2008): N=2,512 adults, each naming up to 10 people they discuss important matters with. Provides alter gender, race, relationship type, and contact frequency. Hampton, Keith N., Lauren Sessions Goulet, Lee Rainie, and Kristen Purcell. "Social Networking Sites and Our Lives." Pew Research Center, Washington, D.C. (June 2011).

General Social Survey (1985): "Important matters" module with age, gender, and kin relationship for each alter. Used to extract age homophily priors and empirical decline rates. Smith, Tom W., Davern, Michael, Freese, Jeremy, and Morgan, Stephen L., General Social Surveys, 1972-2022. NORC ed. Chicago: NORC. The GSS is a project of NORC at the University of Chicago, with principal funding from the National Science Foundation. Network module: Burt, Ronald S. (1984), "Network Items and the General Social Survey," Social Networks 6(4): 293-339; Marsden, Peter V. (1987), "Core Discussion Networks of Americans," American Sociological Review 52(1): 122-131.

Pew Research Center bears no responsibility for the analyses or interpretations of the data presented here. The opinions expressed herein, including any implications for policy, are those of the author and not of Pew Research Center.

Network construction

Six-pass kin-first matching algorithm (Pew_network_homophily.R). Spouse pairs matched first with GSS-derived age weights, then kin reciprocal (parent-child, sibling), kin one-way, friend reciprocal (gender + race + age band), and friend one-way with cascade fallback. N=1,500 persons, 3,992 directed edges, 0 unmatched slots.

Age homophily priors extracted from GSS 1985 non-kin, non-spouse pairs only (n=2,488). Kin ties (n=1,351) excluded to prevent parent-child age gaps from contaminating friend priors. Laplace-smoothed across 5 age bands.

Damage model

Per-cell decline rates (alters/year) computed from GSS 1985 to Pew 2008, both capped at 5 alters. Young men lose connections fastest (0.025/yr); older women lost none (clamped to 0). Kin protection ratios from empirical survival data — kin ties are 1.3x to 2.3x more durable than non-kin, depending on the group.

Three damage patterns tested: uniform linear (null model), empirical linear (per-cell rates), and empirical accelerated (x1.57, matching Pew-to-TESS acceleration). Time horizon: 41 years (1985-2026).

Reconstruction model

Adjacent-possible (friend-of-friend) introductions. Each step: a focal person picks a broker from their existing contacts, the broker introduces them to one of the broker's contacts. Fails if the target is already known or at max alters (10). No random tie formation — all new connections go through existing network structure.

Kin inflow: young isolates receive parent connections at a rate matching cohort turnover (~2.1/step). This is activation-independent.

What "bridge recovery" means

Mean cross-group distance in the network (average shortest path between all pairs of the 6 age-by-gender groups). Bridge recovery is the percentage of damage to this metric that gets healed during reconstruction. Higher means more cross-group connections restored — not just more friends, but more connections between different groups.

The visualization

Six meta-nodes (3 ages x 2 genders). Node size proportional to sqrt(mean degree). Edge thickness and opacity use squared density ratios — strong connections pop, weak ones fade. Force simulation with log-scaled attraction. Single representative replicate; density matrices are stable at N=1,500 with 200+ people per cell.

Reproducibility

Pipeline: extract_age_homophily.R → Pew_network_homophily.R → Pew_investment_sweep.R → Pew_viz_data.R. All scripts available on request.

Make One Friend · Riki Conrey
Data: Pew Personal Networks 2008 (N=2,512) · GSS 1985 (N=1,534) · All results from simulation
© 2026 Riki Conrey. All rights reserved.