The 2.7% Democracy: Why "Are We Still a Democracy?" Is the Wrong Question

Every election cycle, a certain class of politician — usually running for re-election — poses the question with theatrical gravity: "Are we still a democracy?"

The question sounds profound. It's designed to. But it conflates at least four different issues into a single emotional payload: the integrity of elections, the influence of money, the erosion of norms, and the responsiveness of government to citizens. By mashing them together, the question becomes unfalsifiable — and therefore politically useful. It lets the asker gesture at crisis without identifying a single fixable problem.

Here's what they will never say: America was never a democracy in the sense they're implying. The Constitution was explicitly designed to filter the popular will, not transmit it. The Founders said so — in writing, repeatedly. Madison's entire argument in Federalist No. 10 is that direct expression of citizen preferences is dangerous, and that the republic's purpose is to "refine and enlarge the public views" through elected representatives. The honest question isn't whether we're "still" a democracy. It's: how much of what citizens actually want makes it through the system? And now, for the first time, we can measure that precisely.

In 1948, Claude Shannon published "A Mathematical Theory of Communication" — the paper that founded information theory and made the entire digital age possible. Shannon showed that every communication channel has a capacity: a maximum rate at which information can be reliably transmitted. And he showed how to measure how much of that capacity is actually being used.

A recent paper, "The Democratic Channel" (Lopin, 2026), applies this framework to American democracy for the first time. The model treats citizen preferences as the source signal, legislative institutions as the channel through which that signal travels, and policy outcomes as the output. Campaign finance, interest-group lobbying, and institutional friction act as noise. The fundamental question Shannon's framework answers is simple: how much does knowing what citizens want reduce your uncertainty about what policy will actually be?

The answer, computed from 1,836 policy issues spanning 1981–2002 using the landmark Gilens and Page dataset, is devastating.

The citizen-to-policy channel operates at 2.7% of Shannon capacity.

That means 97.3% of the information about what citizens want is lost — destroyed by noise — before it becomes policy. If a telecommunications engineer saw a channel operating at 2.7% efficiency, they would call it non-functional. No consequential data would be routed through it. You wouldn't send a text message through a channel this bad, let alone the preferences of 330 million people.

The elite channel — what the top 10% of earners want — operates at 4.0%. The interest-group channel operates at 2.1%. No channel operates above 5% of capacity. This reframes the entire debate. The standard interpretation of the Gilens-Page finding — that "elites control policy" — misses the point. Nobody's measured preferences control policy. The elite channel gets 1.7 times more through than the citizen channel, but 1.7 times nearly-nothing is still nearly-nothing. The system isn't captured by elites. It's a fundamentally degraded channel that transmits almost no signal from any measured source.

This is where the "are we still a democracy?" crowd gets uncomfortable. The low channel efficiency isn't a bug introduced by Citizens United or gerrymandering or social media. It's a feature of the original architecture. But — and this is crucial — the Founders had good reasons.

Consider the world of 1787. There was no telegraph, no telephone, no internet. A message from Georgia to Massachusetts took weeks by horseback. The idea that citizens could continuously express preferences on specific policy issues was technologically absurd. Elections every two to four years weren't a design choice born of contempt for the people — they were the highest sampling rate the available technology could support. You can't poll the nation when it takes a month to count the votes. (For more on the constraints the Founders faced, see our article on The Constitutional Convention of 1787.)

Plurality voting — compressing a citizen's rich, multidimensional preferences across hundreds of issues into a single binary choice between two candidates — was similarly a technological necessity. There was no mechanism for issue-by-issue input. The best available encoding was to choose a representative who seemed to share your general outlook and trust them to fill in the details. The information destroyed at the encoding stage was enormous, but there was no alternative encoder available.

And the deliberate friction — bicameralism, the presidential veto, the slow machinery of committee and debate — served a genuine purpose in a world where information traveled slowly and errors were hard to reverse. Status quo bias is a reasonable default when the cost of a bad policy change is high and the speed of correction is low.

The system was built to filter, not to transmit. In 1787, that was the right engineering decision. The problem is that the technology changed and the architecture didn't. As our article on Why We Built This explains, 535 people sitting in a building in Washington is an 18th-century workaround for a problem that technology solved decades ago.

There's another reason the system felt more responsive in its early decades, and it has nothing to do with the Founders being wiser or more virtuous. The source signal was dramatically simpler.

In 1790, the electorate was a narrow, homogeneous slice of the population: white, male, Protestant, overwhelmingly of British descent, and largely propertied. Women did not vote. Black Americans did not vote. Indigenous peoples did not vote. The non-landowning poor mostly did not vote. The "citizens" whose preferences the channel transmitted were an ethnically, culturally, and economically coherent group whose interests were far more aligned with each other — and with the elites who ran the system — than any modern electorate could be.

This isn't nostalgia and it isn't prejudice — it's signal theory. A source signal with low entropy (high internal agreement) is easier to transmit through a noisy channel. When the electorate largely agrees on fundamentals — because it has been artificially restricted to a group that shares class, race, gender, and culture — even a terrible channel can appear to "represent the people" tolerably well. The people in the room got roughly what they wanted because they all wanted roughly the same things.

Every expansion of the franchise increased the entropy of the source signal. The 15th Amendment added the preferences of formerly enslaved people. The 19th Amendment added women. The Voting Rights Act of 1965 made Black suffrage real rather than theoretical. Each expansion made America more democratic in the most important sense: more people had a voice. But each also made the signal richer, more complex, more multidimensional — and therefore harder to transmit through a channel that was already operating near zero capacity.

The channel was never upgraded to match. The same two-party, winner-take-all, low-sampling-rate architecture that could passably represent a homogeneous 18th-century gentry is now asked to represent 330 million people spanning every conceivable political, cultural, and economic position. The channel's capacity didn't change. The signal's complexity exploded. The result is what the data shows: 2.7%.

The story of American democracy since 1787 is not — despite what the politicians tell you — a story of democratic erosion. It's a story of channel capacity that never grew while the demands on it compounded relentlessly.

Each decade layered new noise sources onto the original architecture. The rise of political parties in the 1790s created the first encoding distortion: candidates were no longer independent signals but filtered through partisan machinery. The spoils system of the Jacksonian era injected patronage noise. The Gilded Age introduced industrial-scale lobbying. The New Deal added vast federal agencies — each with its own institutional interests, its own bureaucratic friction, its own capacity to attenuate the citizen signal before it reached the policy output.

By the 20th century, the noise stack was formidable: party primaries that pre-filter candidates, campaign finance that amplifies wealthy voices, a permanent lobbying class in Washington, regulatory agencies captured by the industries they oversee, a national security apparatus operating largely outside democratic oversight, and a media ecosystem that compresses complex policy positions into tribal identity markers. Each layer made sense in isolation. Together, they constitute a noise floor so high that the original signal — what do the people actually want? — is nearly inaudible.

The independent validation in the paper confirms this reading. The Stimson-NOMINATE channel — which measures broad ideological mood versus congressional ideology — operates at 29% efficiency. That's the macro-level "vibe." But drill down to specific policies and the number collapses to 3-4%. The channel functions as a low-pass filter: it transmits the broad direction (are we moving left or right?) while destroying the specific content (what do citizens actually want on healthcare, defense, education, the environment?). Politicians read the vibe and ignore the details. The architecture lets them.

The comparison to Rome is not rhetorical — it's structural. The Roman Republic maintained its republican institutions — the Senate, the consuls, the popular assemblies — for centuries after those institutions had ceased to function as designed. The Senate still met. Elections still happened. The forms were preserved. But the substance — the transmission of popular will into governance — had been hollowed out by the accumulation of institutional cruft: patron-client networks, military commands that overrode civilian authority, grain distributions that replaced genuine representation with material appeasement, and a professional political class whose interests had completely diverged from the citizens they nominally served.

Rome didn't fall because someone abolished the Republic. It fell because the Republic became a label attached to an institution that no longer performed its stated function. The channel degraded gradually, over generations, until the gap between the name and the reality was so vast that Augustus could assume autocratic power while technically holding only republican offices. Nobody had to kill the Republic. It had already been dead for decades — a republic in name, not in functional reality.

The American system is not Rome. But the pattern — preserving institutional forms while the underlying transmission mechanism degrades — is identical. The United States still holds elections, still has a Congress, still calls itself a representative democracy. The 2.7% figure suggests that the representational content of those institutions is approaching the Roman endpoint: the forms persist, the function has nearly vanished.

Now we can see the "are we still a democracy?" question for what it is: a performance.

The politicians asking this question are themselves components of the noisy channel. They benefit from the low-pass filter. They campaign on broad ideological mood — "hope," and "change," — precisely because that's the only frequency the system transmits. They never campaign on specific policy positions with binding commitments, because the system isn't designed to hold them to specifics.

When they pose the question, they're catering to an illusion — the idea that there was a golden age of democratic responsiveness that we've fallen from. The data shows there wasn't. The 1981-1990 period shows 3.4% citizen efficiency; 1991-2002 shows 4.1%. The variation is noise-level. There is no fall from grace. They're also deflecting from structural causes. By framing the problem as one of norms or opponents, they avoid discussing the engineering reasons the channel is degraded — encoding loss from plurality voting, adversarial noise from lobbying, absurdly low sampling rates from infrequent elections. These are fixable problems. But fixing them would reduce the politicians' own power.

And most importantly, they're manufacturing urgency to win elections. "Are we still a democracy?" is a fundraising email, not a policy position. It motivates the base to vote for the person asking the question — which changes nothing about the channel's architecture. You're replacing one noisy component with another noisy component. The honest version of their speech would be: "The system transmits 2.7% of what you want. I have no intention of changing that. Vote for me."

The degradation isn't even uniform. When you break down channel efficiency by policy domain using 69 years of Gallup data and federal budget allocation, the topology of the noise becomes visible. Defense spending tracks public concern at 33% efficiency — the highest of any domain. This makes sense: where concentrated interests like defense contractors align with public concern, the channel transmits. Crime policy operates at 13.3%. Education at 12.2%.

Then there's healthcare, at 8.9% — the lowest measured efficiency. Where concentrated interests like pharmaceutical companies actively oppose public preferences, the channel barely functions. The environment is even more revealing: 13.3% efficiency, but the linear correlation between public concern and spending is actually negative. The public cares more; spending doesn't follow. The mutual information framework captures this nonlinear relationship that traditional regression-based measures miss entirely.

This pattern is consistent with an adversarial noise model. Interest groups don't just add random static to the channel. They selectively inject noise precisely on the issues where their preferences diverge from the public's. The channel doesn't degrade uniformly — it degrades where it matters most. The system is not merely inefficient. On the issues where citizens' preferences most conflict with concentrated economic power, it is actively hostile to transmission.

Information theory doesn't just diagnose the problem — it tells you exactly where the losses occur, which means it tells you where to intervene.

The first and most obvious fix is to increase the sampling rate. Elections every two to four years are starvation-level sampling for a signal that evolves continuously. Direct democratic mechanisms, citizens' assemblies, and digital participation platforms increase the rate at which preferences are measured. This is what Constitution.Vote does — continuous polling on specific issues, not once every 1,461 days. The Founders couldn't sample faster because the technology didn't exist. We can. (See Digital Democracy: From Athens to the Internet for how this became possible.)

The second is to improve the encoding. Plurality voting projects a rich, high-dimensional preference signal onto a single binary choice. A citizen enters the booth with nuanced views on hundreds of policy dimensions and the ballot offers two names. This is not compression — it's destruction. Ranked-choice voting, approval voting, quadratic voting, and issue-by-issue polling all encode more of the preference structure per transmission. A ballot with two names on it is information poverty. A platform where you vote on each issue independently is information wealth.

The third is to reduce the noise. Campaign finance, lobbying, and revolving-door employment inject adversarial noise into the channel. Reforms targeting these sources reduce noise power. But it's important to note: even the elite channel only operates at 4%. Noise reduction alone can't fix a channel this degraded — the encoding loss at the input stage is the dominant factor.

The fourth, and most radical, is to build an alternative channel entirely. Sortition — random selection of legislators from the population — eliminates the electoral channel, replacing it with a direct statistical sample of the source. In information-theoretic terms, this swaps a low-bandwidth, high-noise channel for a direct observation. It sounds extreme, but the math suggests the current channel is so degraded that bypassing it may be more efficient than repairing it. Constitution.Vote's liquid delegation system is a step in this direction: delegates emerge algorithmically from voting behavior, not from campaigns and fundraising.

The most important finding in the paper isn't the 2.7% number. It's this: the source entropy decomposition shows that the citizen signal has the same entropy as the elite signal. Citizens' preferences aren't vague, incoherent, or contradictory. They're just as structured, just as precise, just as information-rich as elite preferences. Both source distributions carry approximately 3.1 bits of entropy. The channel simply doesn't transmit the citizen signal. It attenuates it selectively while passing the elite signal at modestly higher fidelity.

This kills the most common excuse for ignoring public opinion: the comfortable belief that "ordinary people don't really know what they want" or that citizen preferences are too diffuse to be actionable. The data says otherwise. The source signal is coherent. The channel is the bottleneck. The information is lost somewhere between the citizen and the statute book — not because citizens lack clarity, but because the pipe was built for a simpler era, never upgraded, and corroded by 250 years of accumulated institutional noise.

So the next time a politician asks "are we still a democracy?", you have a precise answer. We never were, by any information-theoretic standard. The system transmits 2.7% of the citizen signal. It transmitted roughly that much in the 1980s and roughly that much in the 2000s. The question isn't whether to mourn the democracy we've lost. It's whether to finally build one that works — with the technology that now exists to do so.

The full research paper, "The Democratic Channel: An Information-Theoretic Measurement of Preference-Policy Transmission in the United States" (Lopin, 2026), including all data and code, is available at github.com/novalis78/The-Democratic-Channel.