How Technology Shapes Our Children

Executive Summary

Executive Summary

Our future is encoded in the heart of our children — but so much is changing in the 21st century, and it is hard to keep up. Most kids have a phone by eighth grade. How is technology shaping the formation process? Is the impact positive or negative? And how do we make sure the systems shaping our children are doing it in ways we actually want?

Children are encodings of our future.

This article presents a new composite index of moral formation for children — combining dozens of longitudinal studies, weighted by relational intensity, to approximate how the moral formation of kids has shifted over long periods of time. The instrument has limits. But it may be the best-resolution picture we have of how the landscape is changing — and what we may be able to do about it.

The Data

The Data

Section 1

The History of Formation

For most of human history, the people who formed children were also the people who shared their lives with them — the same faces at the breakfast table, the same pastor at the local church, the same neighborhood friends at the edge of the baseball field.

Formation was less a structured program than an organic reality — simply the air children breathed. A child born in rural America in the mid-19th century would absorb her entire worldview inside a radius of perhaps ten miles. Every formative voice in her life was known to her family, accountable to her community, and invested in her flourishing. It was not exceptional. It was normal — so normal, for so many thousands of years, that we never thought to name it.

Every culture forms its young through liturgies — small repeated practices that teach a child what to love, what to fear, what to take seriously, and what to ignore. The liturgies have changed across history. What follows is a brief account of three that have done most of the work in the American story: the home, the living room, and the algorithm.

Liturgy of the Home

The early American settlers crossed an ocean to build a culture of their own, and they built it first inside the home. The household was a small church: family devotions in the morning, scripture at the dinner table, the sermon discussed long after the service had ended. Children absorbed it not through instruction but through proximity — working alongside parents, living inside the same values day after day. The first textbook in New England, the New England Primer, did not open with the alphabet. It opened with scripture.

The pattern extended well beyond Puritan New England — to immigrant households, frontier homesteads, Catholic parishes. The content varied. The structure did not: local, relational, embedded in community, mutually accountable. The people who shaped a child were the people the child would still see at forty.

A day in the life — Colonial New England, c. 1720

• Morning. Father reads scripture aloud. Children expected to recite it by week's end.
• Midday. Children work alongside parents or attend the local school.
• Evening. Evening prayer, scripture memory, sermon text at the dinner table.

Liturgy of the Living Room

Then came the broadcast. Radio reached American living rooms in the 1920s; television followed in the late 1940s. For the first time in history, what shaped a child in the living room had not been made in the neighborhood. By 1960, the average American household had a television set on three to four hours a day.

The shift mattered. A child's worldview was now being formed in part by voices no parent could call by name. But the medium remained constrained. The broadcast was one-to-many: every household got the same program, the same broadcast, the same advertisements. Government bodies enforced content standards. The signal could not adapt to its listener, and it could not learn.

A day in the life — Suburban America, c. 1965

• Morning. Children eat breakfast with parents, walk to school.
• Midday. Living room TV comes on after school — three channels, the same signal in every home.
• Evening. Dinner together, homework, bed.

Liturgy of the Algorithm

The information era opened a new kind of connection. The algorithm era is doing something different: it is rebuilding the formative liturgy itself. Facebook launched a news feed in 2006. At first it was chronological — a running list of what your friends had posted. Then the recommender system arrived, and the math turned out to be irresistible: personalization at one-to-one scale was extraordinarily good for engagement, and extraordinarily good for revenue. YouTube followed in 2008, Instagram in 2010, Snapchat and TikTok after.

What is different now is that the medium no longer broadcasts — it listens. Every micro-interaction feeds it: a swipe, a tap, a pause, a scroll. It knows the time of day. It knows what you search for. It is the first liturgy in human history that adapts to its congregant.

A day in the life — America, c. 2023

• Morning. Wakes up and checks phone before getting out of bed.
• Midday. Phone in the empty spaces — between classes, at lunch, any unstructured moment.
• Evening. Homework, dinner, then alone in the room with the device until late at night.

Section 2

What Is an Algorithm?

An algorithm — or more precisely, a recommendation system — is a computer model that matches what it knows about a user to what it knows about a piece of content, scored against a goal the company has chosen. The tool itself is morally neutral. The goal is not.

Recommender systems are not unique to social media. They appear across everyday life — e-commerce, streaming, search, news aggregators. The same core technology runs inside nearly every consumer application you use: Spotify, Netflix, YouTube, TikTok, Twitter, Facebook, Instagram, Snapchat, Google Maps, Amazon.

In each of those cases, the recommender is trying to serve you something useful. A social media recommender has a different goal — not to help you connect or grow, but to keep you engaged and coming back. The distinction matters more than it sounds, especially when the user is a child.

How Recommenders Work

Two approaches dominate, usually used together.

Content-based filtering analyzes the attributes of content you have already engaged with, then surfaces more content with the same attributes. If a child watches several videos about soccer, the system extracts what those videos have in common — topic, tone, length, creator style — and serves more like them. The logic is: you liked this; here is more of this.

Collaborative filtering takes a different approach. It does not analyze the content; it analyzes the audience. If thousands of users with viewing histories like your child's all watched a particular video next, the system concludes that the video is probably a good recommendation — regardless of what it is actually about. The logic is: people like you went here next.

The Self-Reinforcing Problem

Recommender systems are designed to engage, and they will amplify whatever drives engagement — indifferently. They are not moral agents. They are computer systems: heartless by design, indifferent by default.

The compounding risk for children is straightforward. A child drawn toward destabilizing content generates more engagement, which the system reads as a positive signal, which produces more of the same in the next session. The system does not know the difference between a child thriving and a child spiraling. It only knows what kept her watching. A recommender can therefore pull a young user progressively toward content that is less helpful, less self-actualizing, and further from what her community would have chosen for her.

The algorithm cannot tell the difference between a child finding herself and a child losing herself. And it cannot ask.

Section 3

The Generational Gap and the Problem of Resistance

For most of human history, what shaped a child at eighteen looked consistent across generations — parents, church, teachers, peers, in varying proportions, all local and relational. The stability held through broadcast television, which brought a new voice into the living room but could not personalize or adapt to its listener. The algorithmic platforms broke the pattern. For the first time in recorded history, a non-human, non-accountable system is competing directly with parents, faith communities, and schools for the largest share of formative influence on a child's life.

What Shaped Each Generation at Age 18

Three patterns stand out. Family has been more resilient than expected — even in the Gen Z snapshot, it is still the single largest channel, just no longer dominant. The channel that collapsed most dramatically was not television but church and religion, which fell from roughly 30% of the formation budget in the early 20th century to under 3% today. And the Gen Z column marks a first in American history: a single non-human, non-accountable channel — algorithmic feeds — commanding more formative share than family, school, and church combined.

The Problem of Resistance

What makes the shift hard to address is that children themselves are not asking for help. In a 2023 Pew Research survey, 54% of U.S. teenagers said they would find it very hard to give up social media — and a similar share described their own use as a problem they felt unable to control. We are placing some of the most engineered persuasion systems ever built into the hands of developing minds that are not yet equipped to recognize what they are looking at.

This is not a failure of character. The platforms are designed by some of the most sophisticated engineers and behavioral scientists in the world, optimizing systems that have been tested billions of times against a single target: keep the user on the platform. A thirteen-year-old does not stand a fair chance. Neither, usually, does a parent who did not grow up with these tools, does not understand how they work, and cannot see what their child is being served.

It is not a fair fight. And the fight is over the child's formation.

Section 4

Negative Impacts

The downstream signal is now large enough to measure. Teen mental health deteriorated sharply beginning around 2012 — the year smartphones reached saturation among American adolescents — and has not recovered since.

Teen depression and anxiety rates were stable from 1900 through about 2010 — depression at 4–6%, anxiety at 5–8%. Beginning around 2012, both began rising sharply. By 2024, the figures sat at roughly 4–5× the pre-2010 baseline: depression at 20%, anxiety at 24%.

Beyond Mental Health: Formation Outcomes

The mental health data is the most legible signal. It is not the only one. Several other population-level trends correlate with the algorithmic transition and plausibly share a causal pathway.

Attention and self-regulation. Research on heavy social media use consistently finds associations with reduced capacity for sustained attention and higher rates of impulsivity (Lissak 2018; Wilmer et al. 2017). The mechanism is plausible: thousands of times per day, the feed teaches the brain to expect novelty on a two-to-fifteen-second cycle and to disengage from anything that does not immediately reward.

Social comparison and body image. Instagram and TikTok expose adolescents to highly curated, increasingly AI-enhanced images of peers and influencers. The social comparison literature (Festinger 1954) predicts that upward comparison produces dissatisfaction; the platforms serve upward comparison relentlessly. Body image dissatisfaction among teenage girls rose sharply in the early 2010s across multiple national surveys.

Civic and epistemic fragmentation. Recommendation systems optimizing for engagement will tend to serve content that confirms existing beliefs and produces emotional activation. The pattern accelerates ideological sorting and erodes shared factual ground. The political and epistemic polarization of the 2010s is not reducible to social media, but the algorithmic amplification of outrage-producing content is a documented contributing mechanism (Bail et al. 2018).

Caveats. None of these correlations establishes causation. Pre-2010 mental health baselines rely on retrospective estimates. Other contributing factors — opioid-era family stress, economic precarity, declining religious community — are real and concurrent. The signal is strong. The precise causal weight of algorithmic exposure within that signal is contested, and may never be fully resolvable.

Section 5

Conclusion and What We Can Do

Technology will keep advancing. That is not the question. The question is whether the next generation will be formed by people who know their names — or by systems that know only their click patterns.

Every prior era of formation, however imperfect, was conducted by parties a parent could ask: what are you teaching my child? The algorithm cannot answer that. It does not know what it is teaching. It is only optimizing.

This is not a reason for despair. It is a reason for attention. The first generation raised entirely inside the algorithm is finishing high school now. The second is starting. We still have time to decide what we want them shaped by — but the window is the window of their childhood, and childhood does not wait.

Appendix A

Methodology

The sections above draw on a composite index model built from 125 years of channel-specific time-use data. This appendix documents the model's construction so that its assumptions can be examined, challenged, and adjusted.

The question this model tries to answer

"What shapes children" is not a question with a clean empirical answer. There is no instrument that has continuously sampled the formative influences on a representative cohort of American children for 125 years. The most rigorous longitudinal studies (PSID Child Development Supplement, National Study of Youth and Religion, NLSY) reach back only decades, cover only some channels, and measure outcomes rather than channel-share.

What we have instead is a sparse archive of channel-specific measurements: school enrollment series from NCES going back to 1869; Gallup church attendance from 1939; broadcast television household penetration from 1948; parent-child time-use diaries from Robinson 1965, Bianchi from the same year, and Hofferth from 1981; Kaiser Family Foundation media surveys at three points; Common Sense Media's Census of teen media use at four points; Pew Internet & American Life Project from 1995; Pew Research's annual teen-and-technology surveys from 2014.

The model's argument is that something useful can be said about the relative weight of these channels — but only if the answer is structured as a transparent composite index with explicit assumptions, not as a measurement claim.

The composite index formula

A composite index combines multiple measurements into a single estimated quantity using a documented construction rule. The model treats the share of formative influence as a composite index of two quantities, multiplied:

Where T is hours per week of contact with the channel for a typical U.S. child aged 5–17, and W is the per-hour relational-primacy weight of that channel. Each row of the dataset (one five-year period) sums to 100%.

The 100% normalization is a heuristic that enables relative-share comparisons across time; it is not a claim that the channels are perfectly substitutable or that formation is zero-sum.

A.1 T (time exposure) values

T values for each channel and period are drawn from the empirical sources catalogued in Appendix B and the project's Empirical_Anchors_RAW.csv file. Post-1965 cells are anchored in direct measurements; pre-1965 cells are reconstructed from adjacent proxies — school-enrollment series (NCES), broadcast penetration (FCC), female labor-force participation (BLS), and religious affiliation (Finke & Stark; Gallup post-1939) — with proportionally wider error bars.

Three substantive calibration choices affect the T values used in the model:

• Church T is broadened from service-attendance hours to the full religious framework — services + Sunday school + family devotions + parochial schooling + religious cultural context. This matches how religious-socialization researchers (Smith & Denton, King & Roeser) actually measure religious exposure.
• School T counts effective formative time for enrolled students rather than population-averaged hours, on the principle that for an enrolled child the school day is a real formative experience regardless of national enrollment rates.
• Peer T pre-1960 reflects multi-age unsupervised play, sibling-dense households, walking-to-school, and one-room schoolhouse peer time — substantially higher than today's supervised, structured peer time. The original chart's pre-1980 "peers = 0" reading is not empirically supportable.

A.2 W (relational primacy) values

W values represent the per-hour formative weight of each channel relative to passive broadcast (≈1.0 baseline). They are not empirical measurements. No published study has assigned a numeric formative weight per hour to any channel. What the underlying literatures establish is qualitative direction and relative magnitude — that parental contact is more formative per hour than passive media, that engagement-optimized feeds are more formative per hour than non-personalized broadcast, that religious ritual carries unusually high per-hour weight for participating children.

The default W values in this model:

The interactive model permits these values to be adjusted live within plausible ranges. The "Literature default" preset above produces the headline allocations shown in the charts above.

A.3 The peer/algorithm split rule

When a teen watches a TikTok her friend sent, the exposure is simultaneously algorithmic (engagement-optimized content selection) and peer-mediated (friend social interaction). The model handles this with an explicit split parameter: at 0%, all algorithmic exposure stays in the Algorithm channel; at higher values, a fraction is reallocated to the Peers channel. The default setting is 0% (counting the medium rather than the relational target), which lands the 2020 Algorithm cell at approximately 38%. A 30% split lowers it to roughly 28%; a 60% split lowers it to roughly 18%.

This is a load-bearing modeling choice. Any number you read off the chart for either Peers or Algorithm in the modern era reflects an implicit position on this split.

A.4 The religious-instruction reallocation

In religious families, a meaningful fraction of parental contact time consists of religious instruction — grace at meals, family devotions, Bible reading, moral framing, choice of parochial schooling. The model treats this as Church formation rather than Parent formation, on the principle that the religious tradition is the formative agent and the family is its bearer. This is how religious traditions describe themselves; it is also why early-twentieth-century Church reads at ~30%+ of formation rather than the ~5–7% pure service-attendance hours would produce.

This is a substantive theological/cultural claim the model embeds explicitly. It is not a measurement. Naming it openly is what allows it to be argued for or against.

Findings: three transitions

The model produces a coherent century-long picture of the shifting weight of seven formative channels. Three transitions stand out.

The pre-broadcast equilibrium (1900–1945)

For the first half of the twentieth century, the model produces a stable distribution in which parents account for roughly 40%, church/religion for 30–35%, in-person peers for 15–17%, and school for 5–11% of the influence budget. Broadcast media enters as a small share (radio) starting in the 1920s, never exceeding ~2% before 1945.

The composition reflects the demographic and institutional realities of the period: large multigenerational households (Census household composition); high religious affiliation (~90%) and active religious framework (Finke & Stark); compulsory schooling spreading state-by-state through 1918 with average attendance still under 130 days/year as late as 1930 (NCES); and sibling-dense, neighborhood-rooted unsupervised peer time as the dominant non-family, non-institutional context (reconstructed from Census and corroborated by retrospective time-use comparisons in Hofferth's 1981 baseline).

The broadcast era (1950–1995)

Television's adoption is the single most rapid channel shift in the series. Household penetration goes from 9% in 1950 to 87% in 1960 (FCC). Children's daily viewing reaches roughly 27 hours per week by 1979 (Nielsen). The model registers broadcast media climbing from ~7% of the influence budget in 1950 to a peak of ~25% in 1980, then plateauing through 1995.

The corresponding declines are distributed across the older channels — parents from ~38% to ~28%, church from ~24% to ~10%, school relatively stable. Peers in this period rise modestly as suburbanization, school consolidation, and youth-cohort identity make age-graded peer groups more salient (Larson & Verma 1999; Csikszentmihalyi & Larson 1984).

The algorithmic transition (2010–2020)

The model's most dramatic transition is recent. Algorithmic feeds register near zero in 2005 (engagement-optimized recommenders for children essentially did not exist), reach roughly 5% in 2010 (Facebook newsfeed, basic YouTube recommendations), 17% in 2015 (Instagram, Snap, Musical.ly), and 38% in 2020 (TikTok For You Page era, accelerated by the COVID pandemic).

This is a phase change rather than a continuation of the broadcast trend. Broadcast television peaked at ~25% of the influence budget and held there for fifteen years before declining. Algorithmic content went from 0% to 38% in fifteen years.

Defending the model

The model is built to survive hostile questioning from three quarters: developmental psychologists who will challenge the W values, historians who will challenge the pre-1965 reconstructions, and technology researchers who will challenge the algorithmic-feed allocations.

Defensible claims

• This is a composite index built from documented proxies. The construction rule is published. The W settings can be adjusted live and the allocation re-derived.
• Post-1965 T values are anchored in direct time-use measurements — ATUS, Nielsen, Pew, Common Sense Media, Kaiser, Gallup. The shape of each channel's trajectory matches the corresponding independent time series.
• The model does not claim algorithmic feeds cause a 38% share of formative influence. It claims that under a documented allocation rule, they account for ~38% of the influence budget — and the assumptions required to move that figure to 20% or to 50% are explicit and adjustable.
• The model embeds an explicit theological/cultural claim — that religious tradition is the formative agent and the family is the bearer — and surfaces that claim rather than hiding it.

Claims the model does not make

• It does not claim to be measurement or survey data.
• It does not claim single-percentage-point precision.
• It does not extend to populations outside the United States without re-estimation.
• It does not claim that pre-1950 cells are anything other than reconstructions.
• It does not equate exposure-weighted-by-relational-primacy with formation outcomes.

Appendix B

Limitations and Gaps

Five things the model cannot do

1. It cannot measure formation outcomes. No longitudinal cohort raised on engagement-optimized feeds has yet reached adulthood. The link from algorithmic exposure to durable identity formation is hypothesized, not measured.
2. It cannot eliminate self-report bias. ATUS over-counts engaged parent time; CSM teen self-reports underestimate phone use; Gallup church attendance is over-reported by 15–25 points relative to actual attendance counts. All T values inherit these biases.
3. It cannot produce quantitative W values from the literature. No published study assigns numeric formative weight per hour to any channel. The W settings are calibrated judgments anchored to research traditions that establish direction and relative magnitude, not numbers.
4. It cannot model concurrent exposure. A child watching TV with parents present is counted under both channels. Phone multitasking during family meals is treated as additive when it partly displaces family time.
5. It cannot prove causation. Even where exposure and outcomes correlate, the model does not establish that one causes the other.

Channels not separately tracked

• Siblings — distinct relational category, especially in pre-1960 large households where older siblings co-raised younger ones. Currently bundled into Peers.
• Extended family — grandparents, aunts/uncles, multigenerational households. Significant pre-1960; effectively absent from the modern model.
• Books/Literature — substantial formative time pre-television; still shapes vocabulary and worldview today.
• Music — has migrated from familial → broadcast → algorithmic without being treated as a distinct channel.
• Workplace/chores — pre-child-labor-law era, working alongside adults shaped many children. After-school employment still matters today.
• Self/solitude/play — solo reading, daydreaming, unstructured solo play. Developmentally important; not captured.

Stratifications not broken out

• Age. A five-year-old's allocation is very different from a fifteen-year-old's. The model averages across the 5–17 band.
• Race, class, region. Black church attendance is materially higher; immigrant family structures different; rural/urban gap significant.
• Religious tradition. Mormon, Orthodox Jewish, evangelical, mainline Protestant, Catholic, Muslim, and secular households have radically different exposure profiles.
• Geography. United States only. Cross-national application requires re-estimating every series.

Appendix C

References and Source Data

Primary data sources

Every T value in the model is grounded in a documented measurement or explicitly marked as a reconstruction. The full per-cell catalogue lives in Empirical_Anchors_RAW.csv; the data coverage map below shows which years have direct measurements and which require modeling.

5.1 Primary sources by channel

5.2 Coverage by period

Primary data sources

• Bureau of Labor Statistics. American Time Use Survey, 2003–present.
• Bianchi, S. M., Robinson, J. P., & Milkie, M. A. (2006). Changing Rhythms of American Family Life. Russell Sage Foundation.
• Hofferth, S. L., & Sandberg, J. F. (2001). Changes in American children's time, 1981–1997. Journal of Marriage and Family, 63.
• Hofferth, S. L. (2009). Changes in American children's time — 1997 to 2003. Electronic International Journal of Time Use Research.
• Robinson, J. P. Americans' Use of Time, 1965–1966. ICPSR 7254.
• Robinson, J. P. Americans' Use of Time, 1985. ICPSR 9875.
• Robinson, J. P., & Godbey, G. (1997). Time for Life: The Surprising Ways Americans Use Their Time. Penn State Press.
• Rideout, V. J., Foehr, U. G., & Roberts, D. F. (2010). Generation M2: Media in the Lives of 8- to 18-Year-Olds. Kaiser Family Foundation.
• Common Sense Media. (2021). The Common Sense Census: Media Use by Tweens and Teens.
• Pew Research Center. (2024). Teens, Social Media and Technology 2024.
• Pew Research Center. Internet/Broadband Fact Sheet, 2000–present.
• Gallup. Historical church attendance series, 1939–present.
• Pew Research Center. Religious Landscape Studies, 2007, 2014, 2024.
• Finke, R., & Stark, R. (2005). The Churching of America, 1776–2005. Rutgers University Press.
• National Center for Education Statistics. Common Core of Data and historical statistics.
• Goldin, C., & Katz, L. (2008). The Race Between Education and Technology. Harvard.
• Television Bureau of Advertising. National TV Household Penetration Trends.
• ICPSR. Introduction of Television to the United States Media Market, 1946–1960.
• Vandewater, E. A., & Lee, S. J. (2009). Measuring children's media use in the digital age. American Behavioral Scientist.

Research traditions backing W

• Bowlby, J. (1969). Attachment and Loss, Vol. 1: Attachment. Basic Books.
• Ainsworth, M. D. S., Blehar, M. C., Waters, E., & Wall, S. (1978). Patterns of Attachment. Erlbaum.
• Bronfenbrenner, U. (1979). The Ecology of Human Development. Harvard University Press.
• Groh, A. M., et al. (2017). Attachment in the early life course: Meta-analytic evidence for its role in socioemotional development. Child Development Perspectives, 11(1).
• Bandura, A. (1977). Social Learning Theory. Prentice-Hall.
• Hovland, C. I., & Weiss, W. (1951). The influence of source credibility on communication effectiveness. Public Opinion Quarterly, 15(4).
• Petty, R. E., & Cacioppo, J. T. (1986). The Elaboration Likelihood Model of persuasion. Advances in Experimental Social Psychology, 19.
• Katz, E., & Lazarsfeld, P. F. (1955). Personal Influence: The Part Played by People in the Flow of Mass Communications. Free Press.
• Csikszentmihalyi, M., & Larson, R. (1984). Being Adolescent: Conflict and Growth in the Teenage Years. Basic Books.
• Larson, R. W., & Verma, S. (1999). How children and adolescents spend time across the world. Psychological Bulletin, 125(6).
• Steinberg, L., & Monahan, K. C. (2007). Age differences in resistance to peer influence. Developmental Psychology, 43(6).
• Brown, B. B. (2004). Adolescents' relationships with peers. In Handbook of Adolescent Psychology (2nd ed.).
• Smith, C., & Denton, M. (2005). Soul Searching: The Religious and Spiritual Lives of American Teenagers. Oxford University Press.
• King, P. E., & Roeser, R. W. (2009). Religion and spirituality in adolescent development. In Handbook of Adolescent Psychology.
• Putnam, R. D. (2000). Bowling Alone: The Collapse and Revival of American Community. Simon & Schuster.
• Hattie, J. (2009). Visible Learning: A Synthesis of Over 800 Meta-Analyses Relating to Achievement. Routledge.
• Gerbner, G., Gross, L., Morgan, M., & Signorielli, N. Cultivation theory series (1976–1998).
• Anderson, D. R., Huston, A. C., Schmitt, K. L., Linebarger, D. L., & Wright, J. C. (2001). Early childhood television viewing and adolescent behavior: The recontact study. Monographs of the Society for Research in Child Development, 66(1).
• Haidt, J. (2024). The Anxious Generation. Penguin.
• Allcott, H., Braghieri, L., Eichmeyer, S., & Gentzkow, M. (2020). The welfare effects of social media. American Economic Review, 110(3).
• Twenge, J. M., Martin, G. N., & Campbell, W. K. (2018). Decreases in psychological well-being among American adolescents after 2012 and links to screen time. Emotion, 18(6).
• Bhargava, V. R., & Velasquez, M. (2021). Ethics of the attention economy: The problem of social media addiction. Business Ethics Quarterly, 31(3).

Appendix D

Project Files

The full project is built across the following files, all in this folder.