What Television Does to Your Beliefs

Why We Engage · 19 October 2026 · Reviewed by Faisal Zaidi

Walter Lippmann said in 1922 that we act on a picture of the world rather than the world, and cultivation research spent fifty years trying to measure that picture. The correlation is real, small and still has no established direction, and the two tests run under streaming are surveys rather than anything that settles it.

Answer to the question

What does cultivation theory say about television?

Cultivation theory holds that the more television a person watches, the closer their beliefs about the world move toward what television shows. George Gerbner and Larry Gross set it out in 1976 from content coding plus a survey. Pooled across five decades, the relationship is a correlation of about one tenth, positive and stable, and its direction is still unestablished, because the evidence is almost all self-reported.

Key takeaways

The 2021 meta-analysis pooled 3,842 effect sizes from 406 independent samples and found a correlation of about one tenth, which its authors call enduring rather than small.
Anthony Doob and Glenn Macdonald's 1979 Toronto survey reproduced the correlation across neighbourhoods and found nothing detectable within them, so the pattern was partly a map of who lives where.
Michael Hughes entered nine controls at once on General Social Survey data for 1975 and 1977 and found the effects on fear and alienation largely absent.
Experiments that assign people to different viewing are rare and old, running for weeks rather than the years cultivation claims, and the two cultivation tests under streaming are both surveys.

In 1922 Walter Lippmann described a screen between people and the world. Fifty-four years later two researchers at Pennsylvania tried to put a number on it, and cultivation theory was born.

What did Lippmann actually claim in 1922?

Lippmann's argument sits in Part I, chapter I, section 3 of Public Opinion. He describes a pseudo-environment inserted between a person and their world, a representation people have partly made themselves rather than a lie about it. Behaviour answers the representation. The consequences, in his words, operate not in the pseudo-environment where the behaviour was stimulated but in the real environment where action eventuates.

What Lippmann has is a structure and not a result. Nothing in the passage says that more exposure produces more distortion, which is cultivation theory's claim rather than his, and nothing in it settles whether a supplied picture makes people fearful or fearful people seek the picture out. Our earlier piece, We Define First, Then We See, takes Part III of the same book, where one person meets one ambiguous thing.

How did cultivation theory first get measured?

George Gerbner and Larry Gross published their Violence Profile paper in the Journal of Communication in 1976, out of the Annenberg School of Communications. Their Cultural Indicators programme had two halves bolted together, only the first a measurement of television. The claim was assembled in four steps.

Code prime-time and weekend network drama for violence, producing a Violence Index for what was on air.
Ask survey respondents how much television they say they watch.
Ask the same people about the world, including their own chances of meeting violence.
Correlate the two, and read the gap between heavy and light viewers as television's work.

Step two onward is a cross-sectional survey, meaning everything was measured at one moment with nobody followed over time, so it shows two things go together and not which came first. No viewing was manipulated anywhere in the origin paper. That is the design rather than a hostile reading, and every criticism in the next section follows from it.

No viewing was manipulated anywhere in the origin paper. The experiments that do assign viewing are rare, old and mostly aimed at other outcomes.

Four years later the same team, joined by Michael Morgan and Nancy Signorielli, answered the first objections with two amendments. Mainstreaming holds that television pulls heavy viewers from otherwise different social groups toward a common outlook. Resonance holds that the effect is amplified where a viewer's circumstances echo what is on screen. The phrase most people remember, mean world syndrome, is the programme's own shorthand for the pattern it claimed to find, and we cannot say which publication printed it first.

What happened when the critics reanalysed it?

Trouble arrived before the amendments did. Anthony Doob and Glenn Macdonald ran a door-to-door survey across four Toronto areas in 1979, a high-crime and a low-crime district downtown and the same contrast in the suburbs. Respondents reported their media use and their own estimated chances of meeting violence. Pooled across the four areas, the familiar correlation reproduced. Compare people living in the same place, with local crime held constant, and the within-area relationship was nothing the study could detect.

Three challenges arrived within a year, one of them a fresh survey rather than a reanalysis, each taking a different handle on the question.

Reanalysis	What it changed	What happened to the relationship
Anthony Doob and Glenn Macdonald, 1979	Compared people living in the same neighbourhood	Held across pooled areas, not detectable within them
Paul Hirsch, 1980	Entered demographic controls together rather than singly	No linear relationship between viewing and television answers
Michael Hughes, 1980	Nine controls at once, on national survey data	Largely absent, including on items the originators had not reported

Paul Hirsch's reanalysis in Communication Research, also 1980, is the severe one, and worth quoting rather than paraphrasing.

> In Part I, I point out discrepancies in the items, samples, and coding categories employed, and show that when controls are applied simultaneously (rather than singly) there is no linear relationship between amount of viewing and the provision of 'television answers.'

The clause in the middle is the transferable idea: the original analyses controlled for one demographic at a time, so a relationship can survive each test singly and disappear once the tests are run together. Hirsch also noticed that people reporting no television at all sat nowhere near the theory's own gradient.

Michael Hughes reached a similar place independently, and on better data. He reanalysed General Social Survey items on alienation and on fear of walking near home at night, for 1975 and 1977, with nine controls entered at once, among them age, income, education and town size. The effects of television watching were largely absent. Then he ran the same procedure on items the original authors had left out, and those results again failed to support them.

That second move is about which outcomes get reported rather than about statistics, and it is the harder one to answer. The design invites a further worry no single paper in front of us owns, so we state it as ours.

W. James Potter's 2014 essay named a problem of conceptual hygiene rather than arithmetic. He argues that the literature has let Gerbner's macrosystem explanation blur into a set of practices hunting for relationships between exposure and many indicators. He proposes judging the work on heuristic value, empirical support and precision, and what he concluded we have not read.

The originators replied at length, in five items across Communication Research and Public Opinion Quarterly between 1980 and 1981. None of their contents was retrievable here, so we can tell you the exchange happened and not what was argued.

Did anyone ever manipulate the viewing?

People have been assigned to watch different television, but rarely, long ago and at nothing like cultivation's time scale. Jennings Bryant and two colleagues did it in 1981, in the Journal of Communication, controlling what their participants watched and measuring the anxiety that followed. The scholarly record cites it for cultivation effects observed in experimental settings and for crime programming with an unjust resolution producing the strongest effects. No abstract was ever deposited, so its sample, its length and the exact contrast it ran we cannot check and will not print.

In 2019 Sarah Erickson and colleagues assigned students either to binge a series or to watch it an episode at a time, measuring parasocial relationships and narrative transportation rather than any cultivation outcome.

What nobody has done is run that kind of study for years rather than weeks, the span cultivation actually claims. For the long run there are only natural experiments. Kenneth Hennigan and colleagues in 1982 used a freeze on broadcast licences to compare American communities that got television early with those that got it late, and the Notel study followed three Canadian towns before and after one received it.

L. J. Shrum and colleagues published two studies in the Journal of Consumer Research in 1998 using a priming procedure, nudging which examples came to mind just before a judgement. Both found the ordinary correlational result, that heavy viewers' beliefs about social reality sit closer to what television shows. The priming supported a mechanism: people fail to discount television-derived examples when building a belief, and television's influence rides on that failure. What was manipulated was retrieval.

Seven years later Shrum's team found something stranger, and more useful to anyone selling anything. From a survey and an experiment, they reported that television cultivates materialism through a process running during viewing itself, strengthened by active rather than heuristic processing. Judgements about how affluent society is behave the opposite way, weakened by active processing at the moment of judgement. Attention helps a value take hold and hinders a prevalence estimate.

Longitudinal work, which follows the same people over time and would bear on direction, does exist. Michael Morgan published one on adolescents and sex role stereotypes in 1982. We could not read it, and it comes from inside the originating group, so we will not describe what it found.

What does five decades of pooling show?

In 2021 Erik Hermann and colleagues pooled the whole literature in a meta-analysis, a study combining many studies into one estimate: 3,842 effect sizes from 406 independent samples. The pooled correlation came out at about one tenth, varying only narrowly across the things that might have moved it. Michael Morgan, in this literature since the 1980 amendments, co-authored both modern pooled analyses.

A correlation of about one tenth is small by any ordinary reading, and that reading is ours rather than theirs. The authors do not call it small. They call it enduring, stable over decades despite immense changes in television's structure and technology, and read that stability as support for cultivation theory.

They report the honest complications themselves. The stronger effect of overall viewing weakens over time, which they say gives partial support to the reformist camp. Effects came out larger in smaller samples and in questionnaire data, the pattern that usually signals an estimate biased upward. One tenth is pooled across hundreds of samples and dozens of outcomes, so it cannot be turned into a statement about you.

The originators' case at its strongest is that the measured relationship between television exposure and beliefs about the world has turned up in hundreds of independent samples over five decades, through cable and then the internet, and has not gone away. The critics' case at its strongest is that it is small, drawn almost entirely from self-report at a single moment, and shrinks or vanishes when the obvious confounds are entered together rather than singly.

Has anyone tested cultivation theory under streaming?

It has, at least twice, and both tests are surveys. Sarah Krongard and Mina Tsay-Vogel coded the violence in the five most binged Netflix originals, fifty-two and a quarter hours carrying close to six violent acts an hour, then asked three hundred and sixty-six undergraduates what they watched and how the world looked to them. Heavier viewing of those programmes went with stronger mean world perceptions and weaker kind world perceptions.

So the tests exist, they are few, and they repeat the origin paper's design: exposure reported by the viewer, correlated with a belief reported by the viewer, at one moment. Neither settles direction, and neither uses what the platforms themselves know about who watched what.

The question still matters, because the theory assumed a shared, repetitive, mainstream diet arriving from a handful of channels on somebody else's schedule. Prime time was the mechanism. A viewer picking from a catalogue of a hundred thousand hours is not having the experience cultivation theory was built to describe, and two surveys leave that open.

The theory assumed a shared, repetitive diet from a handful of channels on somebody else's schedule. Two surveys have now tested it on a catalogue of a hundred thousand hours.

A second meta-analysis from the same group in 2023, on social media, where there is no shared schedule at all, pooled to about one eighth, the same order as the television number, so the correlational pattern does travel into a fragmented medium. One of its moderators cuts against the original theory: effects were higher for specific types of content than for general social media use, and cultivation theory's point was that the aggregate diet mattered while the particular programme did not.

This is a different question from the two we have already asked about screens. What Technology Does to Who We Are is about a person and a device, argued between two essayists rather than inside an empirical literature. Both Sides of the Screen-Time Argument is about dose, how much screen time, and whether harm from quantity holds up. This piece is about content and belief, and all of them rest on self-reported exposure correlated with a self-reported outcome.

Wherever viewing and belief are measured together, which is almost everywhere here, an anxious person who watches more is indistinguishable from a watcher made anxious.

Where this evidence runs out

Several things here are weaker than they look. Sample sizes are missing for the founding papers, because few could be verified at source, and a number we cannot check is a number we will not print. Neither the 1976 paper nor the 1981 experiment ever had an abstract deposited, so both designs come from the published record rather than a full reading, and the 1981 specifics stay out. The phrase mean world syndrome has no locus we could establish.

Which leaves a reader of this literature with less than they wanted and more than nothing.

So the test we would apply, the next time something on a screen seems to describe the world outside, is a modest one: ask whether the belief you are about to act on came from a place you could go and check, and if it did not, ask who chose the picture you have instead and what they wanted from you while you were looking at it.

We think the critics have the better of it on causation and the originators on persistence, which leaves cultivation theory as a real correlation nobody has yet shown to run in the direction its name implies, measured in an era of three channels and never properly retested in one where the diet is chosen rather than received.

Common questions

What does cultivation theory say about television?

Cultivation theory holds that the more television a person watches, the closer their beliefs about the world move toward what television shows. George Gerbner and Larry Gross set it out in 1976, from content coding of drama combined with a survey. Pooled across five decades, the relationship is about one tenth, positive and stable, and its direction is still unestablished.

Does watching television make people afraid of crime?

Not on the evidence here. The correlation shows up in pooled data, but Anthony Doob and Glenn Macdonald found nothing detectable once neighbours were compared, and a reanalysis found the effects largely absent under nine controls at once. Experiments assigning people to different viewing do exist, from 1981 and 2019, but neither runs at the time scale cultivation describes and do not run at the time scale the theory describes.

Has cultivation theory been tested on streaming?

Yes, twice at least, and both are surveys. Sarah Krongard and Mina Tsay-Vogel coded the violence in the five most binged Netflix originals, then surveyed three hundred and sixty-six undergraduates, finding heavier viewing related to stronger mean world and weaker kind world perceptions. Siraj Degnah and Yazid Almahraj found a comparable pattern among Saudi viewers. Both rely on self-reported exposure, so neither settles direction.

Does cultivation theory apply to what people buy?

It has been extended there. L. J. Shrum and colleagues reported in 2005, from a survey and an experiment, that television cultivates materialism, with active rather than heuristic processing strengthening the relationship. What their experiments manipulated was retrieval rather than viewing, so this extends the correlational claim rather than demonstrating direction.

Sources

Gerbner, G., & Gross, L. (1976). 'Living with Television: The Violence Profile.' Journal of Communication, 26(2), 172-199. doi:10.1111/j.1460-2466.1976.tb01397.x [Bibliographic record verified at OpenAlex and PubMed PMID 932235. Abstract NOT READ: none was ever deposited at Crossref or OpenAlex. Closed access, no open-access location. No sample size, sampling method or coefficient obtained.]

Gerbner, G., Gross, L., Morgan, M., & Signorielli, N. (1980). 'The "Mainstreaming" of America: Violence Profile No. 11.' Journal of Communication, 30(3), 10-29. doi:10.1111/j.1460-2466.1980.tb01987.x [Bibliographic record verified at OpenAlex. Abstract NOT DEPOSITED. The locus of the phrase mean world syndrome was not verified in any publication.]

Doob, A. N., & Macdonald, G. E. (1979). 'Television viewing and fear of victimization: Is the relationship causal?' Journal of Personality and Social Psychology, 37(2), 170-179. doi:10.1037/0022-3514.37.2.170 [Abstract read in full. Author order confirmed at OpenAlex. Sample size, response rate, exact fear item and all coefficients NOT VERIFIED.]

Hirsch, P. M. (1980). 'The "Scary World" of the Nonviewer and Other Anomalies: A Reanalysis of Gerbner et al.'s Findings on Cultivation Analysis, Part I.' Communication Research, 7(4), 403-456. doi:10.1177/009365028000700401 [Abstract read in full. Part II, Communication Research 8(1), 3-37, 1981, was NOT retrieved and is not characterised.]

Hughes, M. (1980). 'The Fruits of Cultivation Analysis: A Reexamination of Some Effects of Television Watching.' Public Opinion Quarterly, 44(3), 287-302. doi:10.1086/268597 [Abstract read in full. Subsample sizes and all coefficients NOT VERIFIED. Crossref deposits first page 287 only; the range is secondary.]

Gerbner, G., Gross, L., Morgan, M., & Signorielli, N. (1981). 'A Curious Journey into the Scary World of Paul Hirsch.' Communication Research, 8(1), 39-72. doi:10.1177/009365028100800102 [Bibliographic record verified at OpenAlex. Abstract NOT DEPOSITED, contents NOT READ. Four further reply items are confirmed to exist from Michael Morgan's curriculum vitae at umass.edu; none of their contents was read.]

Potter, W. J. (2014). 'A Critical Analysis of Cultivation Theory.' Journal of Communication, 64(6), 1015-1036. doi:10.1111/jcom.12128 [Abstract read in full. A critical essay, not a study. His criteria are verified; his verdict on each of them is NOT VERIFIED and no verdict is attributed to him.]

Hermann, E., Morgan, M., & Shanahan, J. (2021). 'Television, Continuity, and Change: A Meta-Analysis of Five Decades of Cultivation Research.' Journal of Communication, 71(4), 515-544. doi:10.1093/joc/jqab014 [Abstract read in full. Three-level meta-analyses of 3,842 effect sizes from 406 independent samples; overall effect size .107. The authors describe the relationship as enduring and do not call it small; the reading of one tenth as small is ours.]

Morgan, M., & Shanahan, J. (1996). 'Two Decades of Cultivation Research: An Appraisal and a Meta-Analysis.' In B. Burleson (Ed.), Communication Yearbook 20, 1-45. Thousand Oaks: Sage. [Bibliographic record verified from Michael Morgan's curriculum vitae, read in full at umass.edu. Book chapter, no DOI, no abstract exists, contents NOT READ. The widely circulated average correlation is NOT VERIFIED at source and is not printed.]

Hermann, E., Morgan, M., & Shanahan, J. (2023). 'Cultivation and social media: A meta-analysis.' New Media & Society, 25(9), 2492-2511. doi:10.1177/14614448231180257 [Abstract read in full. Meta-analysis of 460 effect sizes from 66 independent samples; overall effect size .123. Social media, not streaming video. The content-specific exposure moderator cuts against the original theory.]

Morgan, M., Shanahan, J., & Signorielli, N. (2015). 'Yesterday's New Cultivation, Tomorrow.' Mass Communication and Society, 18(5), 674-699. [Bibliographic record verified from Michael Morgan's curriculum vitae only. No DOI retrieved, abstract NOT READ, contents NOT VERIFIED and not characterised.]

Shrum, L. J., Wyer, R. S., Jr., & O'Guinn, T. C. (1998). 'The Effects of Television Consumption on Social Perceptions: The Use of Priming Procedures to Investigate Psychological Processes.' Journal of Consumer Research, 24(4), 447-458. doi:10.1086/209520 [Abstract read in full. The priming was manipulated; viewing was not. Sample sizes, priming manipulations and all effect sizes NOT VERIFIED.]

Shrum, L. J., Burroughs, J. E., & Rindfleisch, A. (2005). 'Television's Cultivation of Material Values.' Journal of Consumer Research, 32(3), 473-479. doi:10.1086/497559 [Abstract read in full. Data from both a survey and an experiment. Sample sizes, the experimental manipulation and all effect sizes NOT VERIFIED.]

Morgan, M. (1982). 'Television and adolescents' sex role stereotypes: A longitudinal study.' Journal of Personality and Social Psychology, 43(5), 947-955. doi:10.1037/0022-3514.43.5.947 [Bibliographic record verified at OpenAlex. Abstract NOT DEPOSITED and NOT READ. Panel size, number of waves, interval, attrition and every result NOT VERIFIED; no number is printed.]

Lippmann, W. (1922). Public Opinion. Part I, chapter I, section 3. [Read in full from Project Gutenberg ebook 6456, which carries no page numbers, so part, chapter and section are cited. Lippmann supplies the frame and no finding.]

Bryant, J., Carveth, R. A., & Brown, D. (1981). 'Television Viewing and Anxiety: An Experimental Examination.' Journal of Communication, 31(1), 106-119. doi:10.1111/j.1460-2466.1981.tb01210.x. [NO ABSTRACT EXISTS ON THE PUBLISHED RECORD: the Journal of Communication carried no abstracts in 1981, Oxford Academic serves only a paywalled first-page image, and Crossref, OpenAlex, PubMed and Semantic Scholar hold none. Citation, author order and page range confirmed at the item record]: a genuine experimental manipulation of assigned television content with anxiety as the outcome, cited twice by Rubin, Perse and Taylor in their review of cultivation methodology, once for cultivation effects observed in experimental settings and once for crime-oriented programming depicting an unjust resolution being most likely to produce them. NOT VERIFIED and not printed: the sample size and composition, the number of weeks, the justice-restored contrast and the exact anxiety and victimisation measures. First author Jennings Bryant, then at the University of Massachusetts Amherst, with Carveth and Brown doctoral candidates there.

Erickson, S. E., Dal Cin, S., & Byl, H. (2019). 'An Experimental Examination of Binge Watching and Narrative Engagement.' Social Sciences, 8(1), 19. doi:10.3390/socsci8010019. [Full text read, open access]: a true experiment manipulating the format of exposure to a television narrative, binge against episodic, using first seasons of Felicity and Everwood, with 77 undergraduates at the University of Michigan, 41 binge and 36 episodic. Outcomes were parasocial relationships with favourite characters and narrative transportation, both increased by binge viewing. It manipulates how much television is watched at a sitting and measures no cultivation outcome, so it is a manipulation of viewing rather than a cultivation test. Erickson first author.

Krongard, S., & Tsay-Vogel, M. (2018). 'Online original TV series: Examining portrayals of violence in popular binge-watched programs and social reality perceptions.' Psychology of Popular Media Culture, 9(2), 155-164. doi:10.1037/ppm0000224. [Abstract read in full, with the method and results sections read]: the cultivation test under streaming. Study 1 is a content analysis of the first seasons of the five most frequently binge-watched online original series, all Netflix originals, House of Cards, Unbreakable Kimmy Schmidt, Marco Polo, Bloodline and Daredevil, with 52.25 hours coded, of which 4.5 hours were violence, giving 5.9 violent acts an hour. Study 2 surveyed 366 undergraduates, 78.7 per cent women, aged 19 to 25, at a large university in the north-eastern United States: exposure was positively related to mean world perceptions and negatively related to kind world perceptions. The authors' own scope: self-reported exposure, and correlation is not causation. Published online December 2018 in the 2020 volume, under the journal's former name; a correction notice exists at doi:10.1037/ppm0000239. Krongard first author, Boston University.

Degnah, S., & Almahraj, Y. (2024). 'Patterns of Saudi Viewers' Use of Netflix and Their Awareness of its Impacts on Society: A Quantitative Study in the light of the Cultivation Theory.' Journal of King Abdulaziz University: Communication and Media Studies, 1(1), 159-185. doi:10.4197/comm.1-1.5. [Abstract read in full, in Arabic]: a second cultivation test under on-demand viewing, correlational. A quantitative field survey of a purposive sample of 247 Saudi viewers, reporting a positive relationship between the density of Netflix viewing and awareness of the Western values promoted in Netflix drama, which the authors say matches cultivation theory's central hypothesis. The indexed record lists Degnah first and Almahraj second, which is the order used here; the secondary literature sometimes reverses it. The authors scope the finding to Saudi viewers and note that Netflix content in the Arab world differs from the Western catalogue.

Hennigan, K. M., Del Rosario, M. L., Heath, L., Cook, T. D., Wharton, J. D., & Calder, B. J. (1982). 'Impact of the introduction of television on crime in the United States: Empirical findings and theoretical implications.' Journal of Personality and Social Psychology, 42(3), 461-477. doi:10.1037/0022-3514.42.3.461. [Abstract read]: A NATURAL EXPERIMENT, NOT A MANIPULATION, and described as such in the article. An interrupted time-series design with switching replications exploiting the Federal Communications Commission freeze on new broadcast licences from 1949 to 1952, which gave some American communities television years before others. No consistent effect on violent crime, burglary or car theft; the one robust result was a rise in larceny theft, which the authors attribute to relative deprivation rather than imitation. Hennigan first author.

Williams, T. M. (Ed.). (1986). The Impact of Television: A Natural Experiment in Three Communities. Orlando: Academic Press. [Not verified at a publisher record; catalogue records only]: A NATURAL EXPERIMENT, NOT A MANIPULATION. Three similar Canadian towns of roughly 700 people, pseudonymised as Notel with no television, Unitel with one channel and Multitel with four, studied before and two years after Notel received television, with fieldwork between 1973 and 1975. Attribution risk recorded: catalogue records give the editor as Tannis MacBeth, who also published as Tannis MacBeth Williams, and the article therefore names the study rather than the editor.

A cultivation test using the streaming platforms' own viewing records rather than viewer self-report. [NOT FOUND, and this is the absence the article now claims. Krongard and Tsay-Vogel and Degnah and Almahraj both test cultivation under streaming by survey; neither uses platform data. Wingralek and colleagues 2022 in Current Problems of Psychiatry is a narrative review, Humphries 2018 is an unpublished Hawaii master's thesis, and Ahmed 2021 in Arab Media and Society is framed on third-person effect.]