If a majority of citizens want a policy, does it become law?
Most people assume "yes, eventually." That's the whole pitch of representative democracy: preferences in, policy out, some friction in the middle.
But we all know that the "how much" varies wildly in practice, and that it changes with the political climate at any given time. I wanted to see if I could measure that, so I built an agent-based model of four democratic legislatures (pure parliamentary, pure presidential, and the two semi-presidential hybrids) and ran each one 200 times across four political scenarios: baseline, fragmented, polarised, small.
The ultimate goal: can running experiments like this help us understand the tradeoffs between different institutions?
In total, I ran 3,200 simulated legislative sessions. The code is open source on GitHub, the live demo is on Streamlit where you can move every dial yourself, and the full paper is on arXiv: Why fragmented parliaments stop passing legislation: Opposition discipline and representation across four democratic institutions (arXiv:2608.24554). This post is the plain-English version.
The existing literature gives us three plausible explanations for legislative failure: coalitions can't form, parties enforce too much discipline, or committees keep proposals from reaching a vote. In a real legislature these mechanisms overlap, which makes them hard to separate from observational data. The model's contribution is to make them independently switchable, so I can ask not just whether fragmentation causes gridlock, but which mechanism is doing the work.
Three things surprised me. The second one changed the paper.
parliaments don't fail gracefully. they collapse.
Under fragmentation (five small parties, no natural majority) a pure parliamentary system passes 0.05% of bills.
Not fewer bills. Not slower bills. Effectively zero. Political scientists have been arguing versions of this about Italy for decades, and my model reproduced it cleanly: seed-averaged data, bootstrap confidence intervals, the works.
The obvious explanation is the textbook one: no coalition can form, and a parliament without a government can't legislate. That was the story in the first draft of this post. It was wrong.
the second ingredient: an opposition that coheres
A hung parliament (one where no party or coalition has a majority) has no governing coalition in the model. So whose whip applies? My model had two ways to answer that question, and running both turned out to be the most interesting experiment in the project:
- Cohesive obstruction: with no coalition in power, every MP counts as opposition and votes against bills on party reflex. Passage: 0.05%.
- Personal vote: no whip in play at all, every MP just votes their own preference. Passage: 46.4%.
A fragmented parliament where nobody gets their way on government formation still legislates at essentially full presidential-system rates (the presidential benchmark in the same scenario is 44.8%). Government formation failure blocks nothing by itself.
The collapse requires an extra ingredient: an opposition that sticks together against everything. Fragmented systems with office-seeking parties can keep passing laws through minority and caretaker cabinets. Belgium ran for more than 500 days without a full government while budgets kept moving. The failure mode here is closer to Weimar: anti-system blocs treating every bill as a battlefield.
An independent check agrees. There's a separate ablation called no_discipline that zeroes out whipping everywhere; run it on the fragmented parliament and passage jumps from 0.05% to 46.7%, nearly identical to the personal-vote number, reached through a completely different manipulation. And the rescue ordering is clean:
| Institution | Passage @ fragmented | Δ when discipline removed |
|---|---|---|
| Parliamentary | 0.0005 | +0.466 |
| Premier-presidential | 0.013 | +0.312 |
| President-parliamentary | 0.086 | +0.228 |
| Republican | 0.448 | −0.242 |
The more an institution depends on forming a majority government, the more it depends on the whip. Presidential systems decouple the executive from legislative coalitions, so they don't need it.
But the sign flips. Under polarisation (two locked factions, extreme bills), enforcing discipline costs the parliament 66 percentage points of passage, because every bill becomes a partisan litmus test. So the whip is a vote-aggregator, full stop: governing coalitions use it to pass, anti-system blocs use it to block. Which side holds it in a hung parliament decides whether legislation happens at all.
Most reform debates quietly assume one world or the other (too many factions, or two locked factions) and prescribe for it. The model says check which world you're in first.
there is no best system, just a dial
I added a metric called the policy representation gap: the L2 distance between the bills that pass and the median preference of the constituencies they're supposed to represent.
Under polarisation:
| Institution | Passage rate | Representation gap |
|---|---|---|
| Parliamentary | 88.2% | +0.037 |
| Premier-presidential | 72.5% | −0.012 |
| President-parliamentary | 73.0% | −0.010 |
| Republican | 21.8% | −0.530 |
The parliamentary system passes four times as many bills, but they drift away from what voters want. The presidential system passes far fewer, and the ones that survive the veto sit much closer to the public's median.
This is the deepest finding in the project. Throughput and fidelity live on the same spectrum. You cannot max both. Every real democracy is a point on this curve, and "more democratic" turns out to be a choice about which axis you're willing to give up.
so, do democracies work?
Depends what "work" means.
- Passes what the majority wants, fast: parliaments, right up until they collapse catastrophically. And the collapse is about opposition cohesion, not gridlock arithmetic.
- Only passes laws that match the public: presidential systems, at the cost of long stretches where nothing passes at all.
- Survives whichever crisis shows up: the hybrids are doing something interesting in the middle. They lose on both metrics individually but lose less when the scenario shifts. There's a real case for designs whose worst case is mediocre rather than catastrophic.
One caveat: the model is generative, in the Kollman-Miller-Page and Laver-Sergenti tradition: a stochastic pipeline of institutional rules, not a living political system with elections and adaptation. It isolates structural tradeoffs, it doesn't forecast what Italy does next year. But the tradeoffs it isolates are sharp and reproducible (every result is seed-deterministic and CI-gated).
The paper is on arXiv at arXiv:2608.24554. Model, sensitivity analyses, and the interactive demo are all here: github.com/tofuadmiral/institutional-representation-abm. Go break a parliament, it's weirdly fun.