Why Data Dividends Feel So Intuitive
The pitch is simple. Platforms extract billions from your behavioral data. You get nothing. Pay people their fair share and the problem is solved.
It reads as economic justice. It has the surface logic of labor compensation. You produce something of value, someone profits from it, you receive a cut. Senator proposals in California, academic papers on data as labor, and startup pitches from data-marketplace platforms have all recycled this frame for years.
The intuition is not stupid. Data collection is extractive. The power asymmetry between individuals and platform corporations is real and measurable. The anger behind data dividend proposals is legitimate.
The mechanism is not.
Dr. Patrick Fisher's work on the Personal Data Asset Origination System (PDAOS) starts from the same diagnosis that dividend advocates start from: data subjects have no structural power. Where PDAOS diverges is in the remedy. Micropayment schemes do not transfer power. They launder extraction in the language of compensation.
To understand why, you have to look at the actual economics and the game theory. Neither is forgiving to the dividend model.
The Valuation Problem No One Wants to Discuss
Before any dividend can be paid, someone has to establish what a unit of personal data is worth. This turns out to be nearly impossible to do honestly at the individual level.
Platform revenue from data is a product of aggregation at scale. A single user's browsing history is worth, in terms of direct advertising revenue, somewhere between fractions of a cent and a few dollars per year depending on the study you reference and the vertical. The Financial Times ran their own data calculator in prior years that illustrated this starkly for general readers. Academic estimates from researchers working on privacy economics consistently land in the low single-dollar range for most users annually.
That number sounds damning. It is actually the core problem for dividend advocates.
If the honest market value of an individual's data to a platform is two to five dollars per year, then a politically credible data dividend has to either fabricate a larger number or deliver a payment so small it is insulting. Neither outcome advances user empowerment.
The fabrication route is what most proposals take. They cite platform market capitalizations, divide by user counts and declare a notional per-user value that has nothing to do with the marginal revenue a single user's data actually generates. Market cap is not a data valuation. It reflects network effects, software infrastructure, brand value, monopoly rents and investor expectations. Distributing market cap as a data dividend is conceptually equivalent to paying factory workers a share of the real estate the factory sits on. The numbers are real. The causal relationship is invented.
Pricing individual data accurately would require solving what economists call the attribution problem. When a platform's model produces a prediction that generates ad revenue, how much of that revenue is attributable to your specific data versus the model architecture, the training corpus, the feature engineering and the network effects of ten million other users' data? Attribution at that granularity is not a solved problem. It is an active research area in algorithmic fairness and Shapley value computation, and at commercial scale it is prohibitively expensive even as a calculation.
The W3C Data Privacy Vocabularies and Controls Community Group (DPVCG) has developed consent and data provenance ontologies that are a precondition to even beginning to answer attribution questions. Without machine-readable provenance at the point of data generation, dividend calculations are arithmetic performed on guesses.
Mechanism-Design Failures Under Basic Game Theory
Set the valuation problem aside. Assume you could price data fairly. The mechanism still breaks.
Paying users for their data creates a market. Markets for information goods have well-documented pathologies. The most lethal one for data dividends is adverse selection.
Adverse selection occurs when the payment signal changes the behavior of the party being paid in a way that destroys the value of the thing being paid for. In data markets, once users know they are being compensated for their behavioral data they have incentive to maximize the quantity and apparent quality of what they share while minimizing the actual cost of sharing. In practice this means sharing fabricated or curated data, deliberately browsing to signal high-value demographic attributes, or using automation to generate behavioral traces that look valuable but contain no authentic signal.
This is not speculation about human nature. It is basic mechanism design. The Revelation Principle in mechanism design theory, formalized in work tracing back to Roger Myerson's foundational research, tells us that incentive-compatible mechanisms require that telling the truth be the dominant strategy for participants. Data dividend schemes as currently proposed do not satisfy incentive compatibility. Lying about your data or gaming your behavioral signal is strictly dominant once payment is introduced.
The result is a dataset that is systematically less useful than the passively collected equivalent. Platforms that ran data-for-rewards experiments in prior years found exactly this pattern. Paid behavioral datasets required significantly higher validation overhead and showed measurable distributional drift compared to passive collection baselines. The payment mechanism had poisoned the data supply.
There is a second mechanism failure at the market structure level. Data is not a rivalrous good. When you sell your location history, you still have it. You can sell it again. This creates a leakage problem that commodity markets handle through scarcity but data markets cannot. A data dividend paid once does not prevent re-sale, re-use or secondary market distribution. Without cryptographic enforcement of use limitations (which is technically feasible but not part of any dividend proposal to date), paying for data does not constrain what happens to it after purchase.
The Non-Excludability Problem and Collective Value
The deepest economic problem with data dividends is not valuation or mechanism design. It is the nature of data value itself.
Most of the value platforms extract from personal data is collective and non-excludable. A recommendation model trained on the behavioral data of fifty million users is not a sum of fifty million individual contributions. It is an emergent property of the aggregate. The model's accuracy at predicting your preferences depends on what it learned from people who are similar to you, not only from your direct data inputs.
This is an example of what economists call a public good problem in production. Your data contributes to a shared epistemic resource, the trained model, and once that model exists it is non-excludable: excluding any single contributor from the model's benefits (or from the revenue it generates) does not restore anything to that contributor because the contribution was absorbed into the aggregate.
Attempts to price individual contributions to a collective intelligence using Shapley values from cooperative game theory are theoretically correct in framing but computationally intractable at platform scale. A 2019 paper from researchers at Stanford, Ghorbani and Zou, published in arXiv:1904.02868, explored data valuation using Shapley values and found the approach illuminating for small datasets and prohibitive for production-scale systems. Nothing published since has changed that fundamental tradeoff.
This non-excludability problem is precisely why collective governance structures are more coherent than individual payment schemes. If data value is produced collectively, then collective negotiating structures such as data trusts or data unions are the appropriate mechanism for capturing and redistributing that value. Paying individuals for their marginal contributions to a non-excludable aggregate is structurally incoherent.
Political Economy and Why Bad Ideas Persist
If data dividends fail so clearly on the economics, why do they keep reappearing in policy proposals?
The answer is political economy, not economics. Data dividends are legible. Explaining Shapley values, consent architecture and data fiduciary governance to a legislative committee takes three hours and produces no headlines. Explaining that every Californian should receive a check from Google takes thirty seconds and produces a campaign ad.
Legibility has political power independent of correctness. This is not unique to data policy. Medicare drug pricing reform, carbon dividend schemes and financial transaction taxes all have analogs: the intuitive version survives politically long after economists have explained why the mechanism is broken.
There is also a constituency problem. The populations that would benefit most from robust data governance structures such as low-income users with high behavioral data density but little political capital, are the same populations most attracted to the promise of a data dividend check. Policy advocates working in good faith on their behalf have reason to use the dividend frame even knowing its limitations, because it mobilizes attention.
The cost of this is real. Every legislative cycle spent on dividend proposals is a cycle not spent on data fiduciary standards, consent receipt infrastructure or cryptographic portability rights that would generate durable structural power for data subjects.
The Invisible Data, Volume 6 of The Invisible Series, frames this as the legibility trap: the interventions that are easiest to explain are rarely the ones that restructure power. Building structural data sovereignty requires building infrastructure that most people will never directly see.
What Structural Data Ownership Actually Requires
The failure of data dividends does not mean individuals should accept zero compensation or zero power over their data. It means the compensation mechanism has to match the actual structure of data value.
The Personal Data Asset Origination System (PDAOS) developed by Own Your Data Inc approaches this differently. PDAOS treats personal data as a structured asset with cryptographic provenance from the point of origination. This matters because provenance is what makes data valuable in verifiable contexts. Medical records, financial history, behavioral credentialing: all of these gain value from authenticity guarantees that are currently impossible to produce without trusted intermediaries who themselves become points of extraction.
With PDAOS, individuals hold cryptographically anchored claims about their own data rather than handing raw behavioral streams to platforms. The W3C Verifiable Credentials specification (VC-DATA-MODEL) and Decentralized Identifier standards (DID-CORE) provide the technical substrate for this kind of architecture. When data is structured as an asset with provenance, selective disclosure becomes possible: you can share a proof about your behavior without sharing the underlying data. Zero-knowledge proofs make this computationally feasible for an expanding range of claim types.
This architecture does not pay you for your data. It gives you the structural capacity to decide who accesses it, under what terms and with what cryptographic guarantees about use limitation. That is a categorically different power relationship than a two-dollar annual dividend.
MyDataKey, the consumer-facing implementation of PDAOS at mydatakey.org, is designed around this principle. The product does not promise to make data subjects rich. It promises to make them legible to themselves and illegible to extractors by default. That is a harder political message than a dividend check. It is also the one that actually restructures the underlying dynamic.
Collective structures matter here too. Data trusts and data unions operating over PDAOS-compatible data assets can negotiate on behalf of groups of data subjects with verifiable proof of the data they represent. The Open Data Institute has published working frameworks for data trust governance that are worth studying as policy scaffolding. The combination of cryptographic individual sovereignty and collective bargaining governance is the realistic path to meaningful data compensation. Micropayment dividends are not.
The economics are clear. Individual data points are nearly worthless at the margin. Aggregate data value is collectively produced and non-excludable. Payment mechanisms create adverse selection. Attribution is computationally intractable at scale. Any serious proposal for data compensation has to reckon with all four of these constraints simultaneously, and data dividends as currently formulated reckon with none of them.
That is not a design flaw. It is a structural incompatibility between the proposed mechanism and the actual problem.