What You Actually Pay for Free Apps

Why We Engage · 18 October 2026 · Reviewed by Faisal Zaidi

Ask people what a free service is worth and the answer depends on which direction you ask from, whether they have recently gone without it, and which part of the bargain you point at. One side of the trade prices its goods to the cent, and the other side cannot produce a number at all.

Answer to the question

What do free apps actually cost you?

No stable figure exists. In a 2019 survey of 2,416 Americans by Angela Winegar and Cass Sunstein, the median respondent would pay five dollars a month to keep data private but demanded eighty dollars to allow access to the same data, a one-to-sixteen gap. Field experiments priced a month of Facebook at about 48 dollars and 100 dollars by different methods.

Key takeaways

In a 2019 survey of 2,416 Americans, the median respondent would pay five dollars a month for data privacy but demanded eighty dollars to give the same data up, a one-to-sixteen gap the authors call a superendowment effect.
In a 2013 field experiment, 52.1 per cent of those endowed with a ten-dollar anonymous gift card refused two dollars to switch, while only 9.7 per cent of those endowed with a twelve-dollar identified card paid two dollars to switch back.
A four-week deactivation study led by Hunt Allcott put the median price of giving up Facebook at 100 dollars, and found that figure fell by up to 14 per cent after people had been away.
A one-week restriction study led by Roberto Mosquera priced the service at 67 dollars and found its value rose after the break, the opposite sign to the four-week result.

Free apps have a price and nobody, including the person paying it, can state what it is. That is not a complaint about hidden terms.

It is what happens every time researchers try to measure the figure: it moves by a factor of sixteen depending on which direction you ask from, it moves again depending on whether the person has recently gone without the service, and it moves a third time depending on which part of the bargain the instrument is pointed at. One side of the market in free apps prices its goods to the cent. The other side cannot produce a number at all.

The questions worth asking before you install

Four questions do most of the useful work, and the rest of this piece is the evidence that each is justified.

What would you pay per month for this service with no advertising and no data collection, and what would you accept to give it up for a month? If the two numbers are far apart, neither is your price.
Which part are you pricing: the service, the data behind it, or the minutes of attention? Only the first two have ever been measured.
Have you actually gone a week without it? Your answer changes afterwards, and the best field experiments disagree about which way.
Would you take a paid version at the figure you just named? Almost nobody has been offered that trade at a stated price.

The questions feel odd because the standard account of free apps hides a price inside it. In The Attention Merchants (2016) Tim Wu names the arrangement a grand bargain, free content and convenience in exchange for attention, and describes attention as "widely recognized as a commodity, like wheat, pork bellies, or crude oil" (p. 8). The history behind it is archival and starts in 1833, with Benjamin Day's New York Sun selling a penny paper against six-cent rivals and reselling the circulation to advertisers. Readers, in the book's account, may have thought themselves his customers but were in fact his product.

The grand bargain is the book's framing device rather than a finding, and its stated purpose is modest: to get the reader "to see the terms plainly" and then "demand bargains that reflect the life you want to live" (p. 9). A commodity like wheat has a price, a quantity and a seller who knows both. The book offers no evidence that this seller can produce either number.

Acquisti and colleagues on which way you ask

In 2013 Alessandro Acquisti and colleagues ran the cleanest version of the question in The Journal of Legal Studies, with real money, in a shopping mall. Every subject chose between the same two gift cards: a ten-dollar card whose purchases would not be linked to their name, and a twelve-dollar card whose purchase data would go to the researchers. What varied was which card they were handed first, and the order the two were described in.

Among the 349 women who took part, 52.1 per cent kept the ten-dollar anonymous card when they had been endowed with it, refusing two dollars to switch. Only 9.7 per cent of those endowed with the twelve-dollar identified card paid the same two dollars to switch the other way and protect exactly the same data. The authors' summary is that subjects "were five times more likely to choose privacy in one condition over the other, even though all subjects faced exactly the same choice". Order mattered too, with the identified card chosen by 73.3 per cent when described first and 57.8 per cent when described second.

Two further conditions make the paper strong rather than merely striking. When the researchers swapped the privacy feature for a convenience feature, a card usable online against one usable only inside the mall, the asymmetry vanished and most people in both conditions took the more valuable card. Cut the price gap to twenty-five cents and 47.8 per cent gave up the privacy-enhanced card, against 19.5 per cent when the gap was fifteen dollars.

The authors are careful about what this cannot show. Their subjects volunteered for a survey while wandering a mall, which "may signal limited privacy concerns", so the valuations "may not necessarily generalize to other populations". Their conclusion is the line to keep: privacy valuations are "significantly subject to subtle framing effects and are anchored around extreme focal points".

Allcott and Mosquera disagree on the sign

Two field experiments paid people to stop using one of the best-known free apps and asked what the interruption did to its price. They came back with opposite answers.

The larger is a 2020 randomised evaluation led by Hunt Allcott in the American Economic Review, over the four weeks before the November 2018 US midterm elections. The team recruited 2,743 Facebook users through display advertisements, asked each what they would accept to deactivate, and randomly assigned the 61 per cent whose figure came in under 102 dollars to a paid-deactivation group or a control group, leaving 1,661 people in the impact evaluation. Weighted to resemble the average US Facebook user, the median figure to give up four weeks was 100 dollars and the mean 180.

Then they asked again. After four weeks away, what people would accept to stay off for the next four weeks fell by up to 14 per cent. The authors use that to refuse their own arithmetic: a standard consumer surplus calculation across the estimated 172 million US users gives 31 billion dollars for four weeks, and they immediately add that such metrics overstate the true welfare gains.

They report the tension straight.

> On the one hand, Facebook deactivation increased subjective well-being, and 80 percent of the Treatment group reported that deactivation was good for them. On the other hand, participants were unwilling to give up Facebook unless offered fairly large amounts of money, even after they had deactivated for four weeks, which should have allowed at least some learning or 'detox' from addiction.

Their sample is relatively young and well-educated compared with the user population, and the figures above come from the accepted manuscript rather than the journal page of record.

The 2020 Experimental Economics paper led by Roberto Mosquera ran a field experiment in 2017 over 1,765 individuals, who first revealed what a week of the service was worth to them and were then randomly assigned to a verified restriction or to normal use. One week of Facebook came out at 67 dollars. Those restricted consumed less news, recognised politically skewed stories less often, reported being less depressed and did healthier things. Afterwards the service's value went up, which the authors read as consistent with information loss or with the use being habit-forming.

So in one experiment going without pushed the price down, and in the other it pushed the price up. Both are randomised, both used real money, and they differ in duration, sample and recruitment. We do not think the evidence supports picking a winner. A quantity that reverses sign when you change the length of the interruption is not behaving like a price.

A quantity that reverses sign when you change the length of the interruption is not behaving like a price.

A third method gives a third number. A 2019 paper in Proceedings of the National Academy of Sciences by Erik Brynjolfsson and colleagues paid participants only if they genuinely forwent the service, and put the median user at about 48 dollars to give up Facebook for a month.

Winegar and Sunstein on an unreliable guide

Asking the same person both directions makes the problem plain. A 2019 survey of 2,416 Americans by Angela Winegar and Cass Sunstein, in the Journal of Consumer Policy, found the median respondent would pay five dollars a month to keep data private, and would demand eighty dollars to allow access to the same data. They call that one-to-sixteen gap a "superendowment effect", against the roughly one-to-two ratio usually found between what people will pay and what they will accept. Priming respondents that the data included health information pushed the demanded figure significantly higher than priming them with demographic data.

Their conclusion goes further than their finding, and it is the one we agree with. Because of missing information and behavioural biases, they write, both measures "are highly unreliable guides to the welfare effects of retaining or giving up data privacy", and policymakers should give them little or no attention until those deficits are addressed. The paper calls itself a preliminary investigation, hypothetical and in one country, so the ratio is what two elicitation methods produce when aimed at the same object.

A standing objection threatens the whole mechanism, and the 2013 paper's own authors name it. In 2005 Charles Plott and Kathryn Zeiler, in the American Economic Review, argued that gaps between what people will pay and what they will accept may reflect subjects misunderstanding the task rather than any asymmetry in preferences, which is the usual reading of the endowment effect. With a modified procedure and extensive training beforehand, on lotteries and on mugs, they observed no gap at all, and their own summary of the field was that there is no consensus on whether these gaps are real or reliable.

The 2013 authors answer on one ground: a one-off binary choice with real money is what a real privacy decision looks like, since people are typically given take-it-or-leave-it options rather than a chance to negotiate. Either way the objection strengthens the case here. If the number moves when the procedure changes, it was never a price.

If the number moves when the procedure changes, it was never a price.

Lin separates the taste from the loss

Tesary Lin's 2022 paper in Marketing Science explains why the published figures disagree with each other. She separates a privacy preference into an intrinsic component, a taste for privacy as such, and an instrumental one, the economic loss a person expects from the firm holding the information, then measures each by combining an experiment with a structural model.

The intrinsic values are small on average, from fourteen cents to 2.37 dollars per demographic variable, and highly varied, with consumers at the right tail valuing their data above what the firm does. Her results also cut against the familiar story that the people with most to hide are the ones who hide. So the figures scattered across free apps research, five dollars, 48, 67, 80 and 100, are not rival estimates of one quantity. Each mixes a taste and an expected loss in proportions set by whichever instrument did the measuring.

What this evidence cannot settle

The gap we most wanted to close is still open. We looked for a randomised study that holds a product constant and compares an advertising-funded version against a paid version of the same thing, and could not find one in the published literature. The nearest field experiment, led by Jingcun Cao in 2023, tested paywall design on an app carrying no advertising at all, and its outcome was subscription conversion rather than use. For free apps, the comparison every reader wants has not been run in public.

Two further limits. Every valuation above prices a service or the data behind it, and none prices attention as such, so the minutes are the part of the bargain nobody has costed. And the mall study's levels do not transfer to a population, even where its internal comparisons hold. On the distance between what people say about privacy and what they do, we have written separately.

That leaves a real choice rather than a lesson. If these valuations are unreliable because people lack information and are swayed by how the question is put, then the terms of free apps cannot be set by asking users what they are worth, and somebody else has to set them. If they are unreliable because the thing differs this much between people, central terms will price most people wrongly, and the remedy is to sell the paid version at a stated figure beside the free one. Nothing above picks between those worlds, and almost nobody is offering both.

Common questions

What do free apps actually cost you?

More than most people guess, though the figure depends entirely on how they are asked. Researchers led by Hunt Allcott found the median US Facebook user needed about 100 dollars to deactivate for four weeks. A separate experiment led by Roberto Mosquera priced one week at 67 dollars, and choice experiments reported by Erik Brynjolfsson and colleagues put a month at about 48 dollars.

Why do the figures disagree so much?

Because they measure different things by different means. Angela Winegar and Cass Sunstein found people would pay five dollars a month for data privacy but demand eighty to give it up. Tesary Lin's 2022 decomposition shows why: each headline figure mixes a taste for privacy with an expected economic loss, in proportions set by the elicitation method.

Does going without a free app teach you its true value?

The two best field experiments disagree. In the four-week deactivation study led by Hunt Allcott, what people would accept to stay away fell by up to 14 per cent afterwards. In the one-week restriction led by Roberto Mosquera, the service's value rose. Different durations, different samples, opposite signs, so going without reveals no stable figure.

Are paid versions better than advertising-funded ones?

Nobody has published the test. We looked for a randomised study holding a product constant and comparing an advertising-funded version with a paid version, and found none. The nearest study, led by Jingcun Cao in 2023, concerns an app with no advertising and measures subscription conversion rather than how much the app gets used.

Sources

Wu, T. (2016). The Attention Merchants: The Epic Struggle to Get Inside Our Heads. Knopf, New York; Atlantic Books paperback 2017, ISBN 978-1-78239-485-3. Full read-through pp. 1 to 353 recorded in the vault, reliability pass 1 September 2026; graded as history, with the grand bargain recorded as the author's own framing rather than a finding.

Acquisti, A., John, L. K., & Loewenstein, G. (2013). 'What Is Privacy Worth?' The Journal of Legal Studies, 42(2), 249-274. doi:10.1086/671754. Full text read in the authors' manuscript hosted by the California Attorney General's office at oag.ca.gov, against the OpenAlex and Crossref records for the published version.

Brynjolfsson, E., Collis, A., & Eggers, F. (2019). 'Using massive online choice experiments to measure changes in well-being.' Proceedings of the National Academy of Sciences, 116(15), 7250-7255. doi:10.1073/pnas.1815663116. Abstract read in full, reconstructed from the OpenAlex inverted index. The earlier NBER Working Paper 24514 (2018, doi:10.3386/w24514) carries the author order Brynjolfsson, Eggers, Gannamaneni; Avinash Gannamaneni and Avinash Collis resolve to one OpenAlex author record.

Allcott, H., Braghieri, L., Eichmeyer, S., & Gentzkow, M. (2020). 'The Welfare Effects of Social Media.' American Economic Review, 110(3), 629-676. doi:10.1257/aer.20190658. Abstract read at OpenAlex and full text read in the accepted manuscript posted by the last author at web.stanford.edu; the figures used here come from that accepted version rather than the journal page of record.

Mosquera, R., Odunowo, M., McNamara, T., Guo, X., & Petrie, R. (2020). 'The economic effects of Facebook.' Experimental Economics, 23(2), 575-602. doi:10.1007/s10683-019-09625-y. Abstract read in full, reconstructed from the OpenAlex inverted index; the abstract does not say whether the 67-dollar figure is a mean or a median, so it is not called either here.

Winegar, A. G., & Sunstein, C. R. (2019). 'How Much Is Data Privacy Worth? A Preliminary Investigation.' Journal of Consumer Policy, 42(3), 425-440. doi:10.1007/s10603-019-09419-y. Abstract read in full at the Harvard Law School faculty bibliography, metadata confirmed at OpenAlex.

Plott, C. R., & Zeiler, K. (2005). 'The Willingness to Pay-Willingness to Accept Gap, the "Endowment Effect," Subject Misconceptions, and Experimental Procedures for Eliciting Valuations.' American Economic Review, 95(3), 530-545. doi:10.1257/0002828054201387. Abstract read in full, reconstructed from the OpenAlex inverted index.

Lin, T. (2022). 'Valuing Intrinsic and Instrumental Preferences for Privacy.' Marketing Science, 41(4), 663-681. doi:10.1287/mksc.2022.1368. Abstract read in full at the Crossref item record; earlier SSRN version 2019, doi:10.2139/ssrn.3406412. Sample size and recruitment are not in the abstract and are not stated here.

Cao, J., Chintagunta, P. K., & Li, S. (2023). 'From Free to Paid: Monetizing a Non-Advertising-Based App.' Journal of Marketing Research, 60(4), 707-727. doi:10.1177/00222437221131562. Abstract read in full, reconstructed from the OpenAlex inverted index; the indexed abstract is garbled in one sentence, which is therefore not quoted. The app carries no advertising and the outcome is subscription conversion, not usage.