REVIEW 4 major objections 6 minor 1 cited by
Data marketplaces can increase the willingness to share social media data at low prices
T0 review · 4 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash
Pith's one-line read A data marketplace lifts people's willingness to sell their X data by 12 to 25 percentage points compared with donating it, and by about 7 points compared with a one-time purchase offer, with most willing sellers accepting $0.25 to $2.
desk verdict A genuinely new experiment on marketplace vs one-shot selling, but the money pump in the 'BDM' payment rule undercuts the low-price conclusion and the paper overstates the marketplace-specific effect. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central mechanism is the Becker-DeGroot-Marschak (BDM) incentive-compatible elicitation, adapted for a marketplace with a minimum buyer-matching requirement. Participants state the minimum price they would accept; a sale occurs only if a randomly drawn offer exceeds that price and at least 25 buyers are willing to match it. This design is meant to make stated willingness and price choices incentive-compatible, reducing hypothetical bias, while the buyer-matching threshold operationalizes the marketplace's multi-buyer structure.
What would settle it
Run the same offer structure on a real functioning marketplace with actual payment and data transfer, and compare actual take-up and transaction prices to the stated willingness in these surveys; if actual sales fall far below the 62-67% stated willingness, the central claim fails to generalize.
Extended reading notes
Core claim
The central claim is that data marketplaces meaningfully increase individuals' willingness to sell their X (Twitter) data package and do so at low prices. Marketplace-framed offers raised willingness to sell by 12 to 25 percentage points over a donation baseline across treatments, and by 6.8 percentage points over a single purchase offer from researchers. Minimum acceptable prices did not differ significantly between conditions, yet over 64% of marketplace participants set their price within the suggested $0.25 to $2 range. Within the marketplace setting, neither the type of buyer (university researchers vs. small and medium private companies) nor the inclusion of an automatic sensitive-data removal tool significantly affected willingness to sell.
Load-bearing premise
The load-bearing premise is that people's stated willingness to sell in a hypothetical survey scenario will carry over to a real marketplace where they must actually download, clean, and upload their data and where repeated interactions and trust dynamics differ.
Editorial extensions
If this is right
- Marketplace framing is a stronger lever than payment size: raising the suggested price range from $0.25-$1 to $1-$1.75 produced no additional willingness to sell, so the structure, not the amount, drives the effect.
- A marketplace could cut the per-dataset cost of social media research from the $5-$10 typical of one-off offers to under $2, while sellers can still earn more by selling to multiple buyers.
- Privacy-protection tooling and buyer type (researchers vs. private companies) do not significantly move stated willingness, consistent with the privacy paradox, so recruitment design may matter more than stronger privacy guarantees.
- Data donation remains a low-yield path: the 12-25 point gap suggests donation-framed recruitment misses a large pool of users who would sell for modest amounts.
- A marketplace serving both researchers and companies appears viable from the supply side, since seller willingness was similar across those buyer types.
Reading between the lines
- If stated willingness transfers to real transactions, the binding constraint on data marketplaces will be supply-side friction (downloading, cleaning, and uploading data) rather than price or trust, making friction-reduction the priority for implementation.
- The minimum 25-buyer matching threshold may embed an implicit coordination mechanism; varying that threshold in a follow-up could reveal whether perceived liquidity, not the number of buyers, drives the marketplace effect.
- The 64% clustering inside the suggested price range hints that anchors, not underlying valuations, set prices in this market; a design that randomizes the suggested range could separate anchoring from true reservation prices.
- Repeated interactions and reputational dynamics, absent from the one-shot survey, could either raise prices (through trust) or lower them (through competition); testing in a longitudinal marketplace simulation would clarify the direction.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. This paper reports two preregistered online survey experiments (total N = 2,500 U.S. Prolific participants) that examine whether a data marketplace increases individuals' willingness to sell their X (Twitter) data and affects the minimum price they state they would accept. Study 1 compares a single-offer condition and a marketplace condition against a data-donation control at two suggested price ranges; Study 2 varies buyer type (university researchers vs. private companies) and the presence of a privacy-protection tool. The authors report that monetary incentives increase willingness to sell by 12–25 percentage points relative to donation, that the pooled marketplace-vs-single-offer contrast is 6.8 points, and that over 64% of willing sellers stated prices in the suggested $0.25–$2 range. They also report that buyer type and privacy protection have no significant effects.
Significance. The topic is timely and policy-relevant for post-API social media data access, and the study has notable strengths: it is preregistered, uses a large sample, provides the full survey instruments, and implements a randomized design with a clear donation control. If the headline effects were trustworthy, the finding that a marketplace framing—rather than higher payments—drives willingness to share would be an important and actionable result. However, the credibility of the central 'low prices' claim is undermined by the fact that the stated elicitation is not a Becker-DeGroot-Marschak mechanism, and the abstract overstates the marketplace-specific effect. The willingness-to-sell contrasts are still informative as stated-intention results, but the price estimates and the causal attribution to the marketplace need substantial revision.
major comments (4)
- [§4.2 and Appendix S1 T1/T2 examples] The described elicitation is not a BDM mechanism. In the single-offer condition, participants are told that the computer draws a price and that exchange occurs if the participant's stated minimum is below that draw, but the examples (e.g., Example 2: computer draws $0.8, participant states $0.6, payment is $0.6) show that participants are paid their own stated minimum, not the random offer. Under this payment rule, truthful reporting of the minimum acceptable price is not a dominant strategy. For a risk-neutral seller with true valuation v and an offer uniformly distributed on [$0.25, $1], expected payment from reporting m is m × (1 − m)/0.75, which is maximized at m = $0.5 for any v below that point. The marketplace condition has the same feature because payment is m times the number of matching buyers, so the report m directly determines the per-buyer price. Consequently, the 'minimum acceptable prices' and the claim that over 64% of participants set prices in the suggested range (Abstract, §2.1) reflect optimization or anchoring under a non-incentive-compatible rule, not willingness-to-accept values. The central 'low prices' conclusion in the title and abstract therefore needs to be removed or thoroughly re-cast as stated prices under a non-incentivized scenario.
- [Abstract and §2.1] The abstract attributes the 12–25 percentage-point increase to the data marketplace, but the text of Study 1 states that this is the effect of monetary incentives 'regardless of whether offered individually or via a marketplace.' The marketplace-specific contrast, pooled across treatment groups, is the 6.8-point estimate in Table 1. As written, the headline overstates the marketplace's causal role and conflates the effect of payment with the effect of the marketplace mechanism. This is load-bearing for the paper's main claim and should be corrected in the abstract, introduction, and discussion.
- [Table 1] Table 1 reports a coefficient for 'marketplace' of -0.68 with a t-value of 2.110 and p = 0.035, while the text says participants in the marketplace condition are 6.8 percentage points more likely to sell. A negative coefficient cannot produce a positive 6.8-point effect, and the t-value is inconsistent with the printed estimate. This sign error and the odd intercept (-0.69, t = -0.299) undermine confidence in the reported regression results. The authors need to re-run and correct this analysis, because the pooled marketplace-vs-single-offer effect is a key quantity for the paper's claims.
- [Appendix Debriefing] The debriefing states that no marketplace was actually created, that participants would not be asked to download or send their X data, and that the platform was mentioned only 'to help create a realistic experience.' The Discussion acknowledges external-validity concerns, but the paper's abstract and conclusion treat the stated-willingness outcomes as evidence that data marketplaces 'can address researchers' data access challenges.' Given that no real transaction could occur, the outcome is hypothetical stated willingness, and the conclusion should be prominently qualified as such rather than only noted as a limitation.
minor comments (6)
- [Table 2 caption] The caption reads 'Effect of selling on marketplace compared to a single purchase offer,' but Study 2 compares buyer type and privacy protection within marketplace conditions, not a marketplace versus a single purchase offer. The caption should be corrected.
- [§2.2 and Table 2] The text reports p = 0.088 for the researcher-versus-company comparison, while Table 2 reports p = 0.089; please make the values consistent.
- [Abstract] The phrase 'over 64 percentage of participants' should read 'over 64% of participants.'
- [§2.1, Table 3, Table 4, §4.2] Price ranges are inconsistent: some places list a '$1-$2' condition, while the survey items and §4.2 specify '$1-$1.75'. Please standardize the range labels throughout the text, figures, and tables.
- [§4.1 and Data availability] The paper says preregistrations are available on aspredicted.com and code/data are available in an OSF repository, but no URLs are provided. Please include direct links for verifiability.
- [Figure 1 and Tables 4/6] The text says price effects are shown in standard deviations in Figure 1, while the appendix tables report dollar amounts. Please clarify the exact outcome definition and units used in each analysis.
Circularity Check
No significant circularity: the paper's willingness-to-sell and price estimates are estimated from randomized treatments, not derived from or defined in terms of the quantities they predict.
full rationale
The paper is a survey-experimental study, not a derivation. The central quantities—willingness to sell and minimum acceptable price—are outcome measures elicited from participants and compared across randomized conditions (marketplace vs. single offer vs. donation control; buyer type; privacy tool). No outcome is defined as a function of a fitted parameter, and no treatment effect is forced by construction: the donation control is an independent baseline, and the 12–25 pp and 6.8 pp comparisons are empirical contrasts. The suggested price anchors ($0.25–$1, $1–$1.75) and the 25-buyer threshold are experimental design parameters, not fitted inputs disguised as predictions. The paper's self-citations (e.g., Refs. [3], [6], [18]) are contextual and do not carry the load-bearing argument; no uniqueness theorem or prior result by the same authors is invoked to rule out alternatives. The debriefing explicitly states the marketplace does not exist and no data would be requested, which is an external-validity limitation honestly disclosed in the Discussion, not a circular step. The skeptic's point that the implemented BDM variant pays the participant's stated minimum rather than the randomly drawn offer concerns incentive compatibility of the elicitation and thus the validity of the price and willingness estimates; it does not make any result equivalent to an input by definition or by fitted-parameter construction. Circularity is therefore not present.
Assumptions & free parameters
free parameters (2)
- Suggested price ranges in BDM scenarios =
$0.25-$1 and $1-$1.75
- Minimum buyer-matching threshold =
25 buyers
assumptions (3)
- domain assumption Participants' stated willingness to sell under the hypothetical BDM scenario is a valid proxy for actual selling behavior.
- domain assumption The BDM mechanism, including the 25-buyer matching rule, is incentive-compatible and understood by participants.
- domain assumption Privacy concerns measured post-experiment are unaffected by treatment assignment and can be used as moderators.
Cite this review
Pith. "Pith review of Data marketplaces can increase the willingness to share social media data at low prices." pith.science (2026). https://pith.science/paper/GIMFXFBM
@misc{pith2026250616618,
author = {Pith},
title = {Pith review of: Data marketplaces can increase the willingness to share social media data at low prices},
year = {2026},
howpublished = {\url{https://pith.science/paper/GIMFXFBM}},
note = {Machine review of arXiv:2506.16618}
}
read the original abstract
Living in the Post API age, researchers face unprecedented challenges in obtaining social media data, while users are concerned about how big tech companies use their data. Data donation offers a promising alternative, however, its scalability is limited by low participation and high dropout rates. Research suggests that data marketplaces could be a solution, but its realization remains challenging due to theoretical gaps in treating data as an asset. This paper examines whether data marketplaces can increase individuals willingness to sell their X (Twitter) data package and the minimum price they would accept. It also explores how privacy protections and the type of data buyer may affect these decisions. Results from two preregistered online survey experiments show that a data marketplace increases participants' willingness to sell their X data by 12 to 25 percentage points compared to data donation (depending on treatments), and by 6.8 points compared to onetime purchase offers. Although difference in minimum acceptable prices are not statistically significant, over 64 percentage of participants set their price within the marketplace's suggested range (0.25 to 2), substantially lower than the amounts offered in prior onetime purchase studies. Finally, in the marketplace setting, neither the type of buyer nor the inclusion of a privacy safeguard significantly influenced participants willingness to sell.
Figures
Forward citations
Cited by 1 Pith paper
-
Web-Browsing LLMs Can Access Social Media Profiles and Infer User Demographics
Web-browsing LLMs can retrieve X profile content and infer demographics with above-chance accuracy in some cases, but the study's evidence is partly confounded by training-data memorization and a heavily reduced synth...
Reference graph
Works this paper leans on
-
[1]
Out-group animosity drives engagement on social media
Steve Rathje, Jay J Van Bavel, and Sander Van Der Linden. Out-group animosity drives engagement on social media. Proceedings of the national academy of sciences , 11 118(26):e2024292118, 2021
work page 2021
-
[2]
Managing extreme ai risks amid rapid progress
Yoshua Bengio, Geoffrey Hinton, Andrew Yao, Dawn Song, Pieter Abbeel, Trevor Darrell, Yuval Noah Harari, Ya-Qin Zhang, Lan Xue, Shai Shalev-Shwartz, et al. Managing extreme ai risks amid rapid progress. Science, 384(6698):842–845, 2024
work page 2024
-
[3]
Existential risk narratives about ai do not dis- tract from its immediate harms
Emma Hoes and Fabrizio Gilardi. Existential risk narratives about ai do not dis- tract from its immediate harms. Proceedings of the National Academy of Sciences , 122(16):e2419055122, 2025
work page 2025
-
[4]
The privacy paradox: Per- sonal information disclosure intentions versus behaviors
Patricia A Norberg, Daniel R Horne, and David A Horne. The privacy paradox: Per- sonal information disclosure intentions versus behaviors. Journal of consumer affairs , 41(1):100–126, 2007
work page 2007
-
[5]
Do we need a social media accelerator? SocArXiv doi, 10, 2023
Christopher A Bail, D Sunshine Hillygus, Alexander Volfovsky, Max Allamong, Fatima Alqabandi, Diana ME Jordan, Graham Tierney, Christina Tucker, Andrew Trexler, and Austin van Loon. Do we need a social media accelerator? SocArXiv doi, 10, 2023
work page 2023
-
[6]
Meysam Alizadeh, Emma Hoes, and Fabrizio Gilardi. Tokenization of social media en- gagements increases the sharing of false (and other) news but penalization moderates it. Scientific Reports, 13(1):13703, 2023
work page 2023
-
[7]
New ai regulation in the eu seeks to reduce risk without assessing public benefit
Barbara Prainsack and Nikolaus Forg´ o. New ai regulation in the eu seeks to reduce risk without assessing public benefit. Nature Medicine, 30(5):1235–1237, 2024
work page 2024
-
[8]
Osd2f: An open-source data donation framework
Theo Araujo, Jef Ausloos, Wouter van Atteveldt, Felicia Loecherbach, Judith Moeller, Jakob Ohme, Damian Trilling, Bob van de Velde, Claes De Vreese, and Kasper Wel- bers. Osd2f: An open-source data donation framework. Computational Communication Research, 4(2):372–387, 2022. 12
work page 2022
Show all 61 references
-
[9]
Leveraging data donations for communication re- search: Exploring drivers behind the willingness to donate
Nico Pffner and Thomas N Friemel. Leveraging data donations for communication re- search: Exploring drivers behind the willingness to donate. communication methods and measures 00, 00 (2023), 1–23, 2023
2023
-
[10]
Citizen science, public policy
Christi J Guerrini, Mary A Majumder, Meaganne J Lewellyn, and Amy L McGuire. Citizen science, public policy. Science, 361(6398):134–136, 2018
2018
-
[11]
The blind spot in data donations: who is (not) willing to donate digital data in social scientific research
Joanna Strycharz, Corine Meppelink, Brahim Zarouali, Theo Araujo, and Hilde Voorveld. The blind spot in data donations: who is (not) willing to donate digital data in social scientific research. Computational Communication Research, 6(2):1, 2024
2024
-
[12]
What managers need to know about data exchanges
Jose Parra-Moyano, Karl Schmedders, and Alex Pentland. What managers need to know about data exchanges. MIT Sloan Management Review , 2020
2020
-
[13]
The social psychology of groups
John W Thibaut. The social psychology of groups . Routledge, 2017
2017
-
[14]
Determinants of willingness to donate data from social media platforms
Zolt´ an Kmetty,´Ad´ am Stefkovics, J´ ulia Sz´ amely, Dongning Deng, Anik´ o Kellner, Edit Paul´ o, Elisa Omodei, and J´ ulia Koltai. Determinants of willingness to donate data from social media platforms. Information, Communication & Society , pages 1–26, 2024
2024
-
[15]
User-centric ap- proaches for collecting facebook data in the ‘post-api age’: Experiences from two stud- ies and recommendations for future research
Johannes Breuer, Zolt´ an Kmetty, Mario Haim, and Sebastian Stier. User-centric ap- proaches for collecting facebook data in the ‘post-api age’: Experiences from two stud- ies and recommendations for future research. Information, Communication & Society , 26(14):2649–2668, 2023
2023
-
[16]
The willingness to sell personal data
Volker Benndorf and Hans-Theo Normann. The willingness to sell personal data. The Scandinavian Journal of Economics , 120(4):1260–1278, 2018
2018
-
[17]
External validity
Michael G Findley, Kyosuke Kikuta, and Michael Denly. External validity. Annual review of political science , 24(1):365–393, 2021. 13
2021
-
[18]
Simple prompt injection attacks can leak personal data observed by llm agents during task execution
Meysam Alizadeh, Zeynab Samei, Daria Stetsenko, and Fabrizio Gilardi. Simple prompt injection attacks can leak personal data observed by llm agents during task execution. arXiv preprint arXiv:2506.01055 , 2025. 14 Full Results Study 1 Table 3: The effect of data marketplaces o...
2025 arXiv
-
[22]
Hawaiian or other Pacific islander
-
[25]
Non-Hispanic White 16
-
[35]
Republican Party SOCIAL MEDIA ACC 1 Do you have a social media profile (e.g., Facebook, X(Twitter), Instagram, Reddit, TikTok)? 17
-
[36]
Yes SOCIAL MEDIA ACC 2 Do you have accounts on any of the following social media sites? Please select all that apply • X(Twitter) • Instagram • Reddit • TikTok • None SOCIAL MEDIA USE On a typical day, how much time would you say you spend on social media (such as Facebook and...
-
[38]
This means all X(Twitter) users can get a full copy of their data
3+ hours Study 1 – Treatment 1 – Intro S1 T1 INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP file. This means all X(Twitter) users can get a full copy of their data. Since X(Twitter) has closed its data access API, resear...
-
[45]
Other S1 T1 OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILLING S1 T1 20 Study 1 – Treatment 2 – Intro S1 T2 INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP fil...
-
[52]
Other S1 T2 OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILLING S1 T2 Study 1 – Treatment 3 – Intro S1 T3 INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP file. ...
-
[59]
Other S1 T3 OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILLING S1 T3 24 Study 1 – Treatment 4 – Intro S1 T4 INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP fil...
-
[66]
Other 26 S1 T4 OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILLING S1 T4 Study 1 – Control – Intro S1 CONTROL INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP fi...
-
[73]
Other S1 CONTROL OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILL S1 CONTROL Study 1 – Post-Survey POST-SUR VEY-INTRO In the following questions, we are curious to find out your knowledge on online data privacy and s...
-
[74]
Other AGE Which of the following categories includes your age?
-
[75]
65+ RACE How do you describe yourself? (Please check the one option that best describes you)
-
[76]
American Indian or Alaskan Native
-
[77]
Hawaiian or other Pacific Islander
-
[78]
Asian or Asian American
-
[79]
Black or African American
-
[80]
Other EDUCATION What is the highest level of education you have completed?
-
[81]
None, or grades 1-8 32
-
[82]
High school incomplete (grades 9-11)
-
[83]
High school graduate (grade 12 or GED certificate)
-
[84]
Technical, trade or vocational school AFTER high school
-
[85]
Some college, no 4-year degree (includes associate degree)
-
[86]
College graduate (B.S., B.A., or other 4-year degree)
-
[87]
In general, would you describe yourself as:
Post-graduate training/professional school after college (toward a Master’s degree or Ph.D., Law or Medical school) IDEOLOGY In politics, people sometimes talk about liberal and conservative. In general, would you describe yourself as:
-
[88]
Somewhat conservative
-
[89]
Very conservative PID – FORCED If you absolutely had to choose between only the Democratic and Republican Party, which do you prefer?
-
[90]
Republican Party SOCIAL MEDIA ACC 1 Do you have a social media profile (e.g., Facebook, X(Twitter), Instagram, Reddit, TikTok)?
-
[91]
Yes SOCIAL MEDIA ACC 2 Do you have accounts on any of the following social media sites? Please select all that apply. 33 • X(Twitter) • Instagram • Reddit • TikTok • None SOCIAL MEDIA USE On a typical day, how much time would you say you spend on social media (such as Facebook...
-
[92]
Less than 30 minutes
-
[93]
This means all X(Twitter) users can get a full copy of their data
3+ hours Study 2 – Treatment 1 – Intro S2 T1 INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP file. This means all X(Twitter) users can get a full copy of their data. Since X(Twitter) has closed its data access API, resear...
-
[100]
Other S2 T1 OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILLING S2 T1 Study 2 – Treatment 2 – Intro S2 T2 INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP file. ...
-
[107]
Other S2 T2 OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILLING S2 T2 Study 2 – Treatment 3 – Intro S2 T3 INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP file. ...
-
[114]
Other S2 T3 OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILLING S2 T3 Study 2 – Treatment 4 – Intro S2 T4 INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP file. ...
-
[118]
I don’t trust private companies
-
[119]
I trust private companies but am unsure whether they share data with third parties
-
[121]
Other S2 T4 OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILLING S2 T4 Study 2 – Control – Intro S2 CONTROL INTRO Social media platforms like X(Twitter) now allow users to export and download their data as a ZIP file....
-
[122]
I need to know exactly who will be using my data
-
[123]
I need to understand the specific purpose for which my data is needed
-
[124]
I don’t trust X(Twitter)
-
[125]
I don’t trust researchers
-
[126]
I trust researchers but am unsure whether they share data with third parties
-
[127]
I am unsure about what is included in my X(Twitter) data
-
[128]
Other S2 CONTROL OTHER You have selected ”other”. Please specify: Displayed as open textbox if ”Other” was selected in NOTWILLING S2 CONTROL Study 2 – Post-Survey POST-SUR VEY-INTRO In the following questions, we are curious to find out your knowledge on online data privacy an...
Reviewed August 6, 2026 · model on record in the stance chip above.
Discussion (0). Sign in to comment.