Which of Our Divisions Are Real

Telling a real divide from a distorted one

The last piece ended with a question I left hanging. I had sorted the reasons a good idea takes so long to catch on into four kinds, and said that telling them apart was the whole of the job. I closed by asking something harder. When we look at everything that seems to split the country in two, how much of it is a real disagreement, and how much of it only looks like one?

Walker Briefing is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.

That is the question this piece takes seriously. Not whether we secretly agree. That is the answer I would find convenient, and I will get to why in a moment. The question is how you tell a real division from a distorted one when you are looking at a whole country, and cannot ask it the way a doctor asks a family in a single room.

I want to tell you something before I go any further, because one of the answers I might reach is one that suits me personally. I sit on the crossbench and I lead a small party. If the big divide between the major parties turned out to be mostly theatre, with the public agreeing more than the system lets on, that finding would suit me: my politics, the book you will hopefully soon be reading, the audience I am building for it, and the re-election I have told you I am working towards. That is the biggest conflict of interest I have carried into anything in this series, and I am putting it on the table now rather than letting you find it later. I am not here to convince you that we secretly agree. I am here to give you a way of telling the real division from the distorted one, and to put my own convenient answer through the same test as everything else.

The case I would love to believe

Let me make the strongest version of the case I am tempted by, as honestly as I can, because an argument you have only half made is an argument you have not really tested.

The case goes like this. Underneath the surface split, there is more agreement than the system shows us. A lot of what looks like a divided public is really just feeling, or tribal identity, or noise pumped up by the machinery that carries our politics to us. And there are real, documented ways that genuine agreement can be made to disappear, so a whole country can quietly agree on something without a single person in it discovering that everyone else agrees too.

Those ways are worth naming properly, because each one is real, and each one has a limit, and the limit matters as much as the mechanism itself.

There is what an economist named Timur Kuran called preference falsification. People say what they think is safe to say rather than what they actually think, so what everyone can see hides what everyone actually believes, until something breaks the dam and the truth comes out all at once. That is roughly what happened across Eastern Europe in 1989, when public opinion appeared to flip overnight. But this mechanism needs real social cost for being honest. It gets weaker wherever speaking your mind is cheap, or anonymous. And it is weakest of all in the arguments people care about most, where they hold their position out of genuine conviction rather than fear of the neighbours. In those arguments, the disagreement you see is more likely to be the disagreement that is actually there.

There is pluralistic ignorance: most people privately think one thing, while each one of them wrongly assumes everybody else thinks the opposite. This was first properly measured in studies of university students, who mostly disapproved of heavy drinking on campus while assuming, wrongly, that their mates were all fine with it. Its limit is that it has been shown for private embarrassments, things people would rather not admit to their friends. It does not automatically carry over to arguments about actual policy.

There is the spiral of silence: people who think they are in the minority go quiet, out of fear of being frozen out, which makes the minority look even smaller than it is. Its limit is that when researchers measured it carefully, the effect turned out to be small, weaker than the theory promised, and it showed up more in places where everybody is already similar than in places where people actually mix.

And there is false polarisation: the well documented habit of imagining the other side as more extreme, and more united, than they really are. Its limit is the biggest one of all, and I will come back to it, because whether fixing that wrong impression actually cools people down is the exact question this whole piece turns on.

I came to this question as a doctor before I came to it as a politician, and the surgery teaches you something worth keeping in the back of your mind. A family sometimes arrives at a bedside seemingly split about what to do for someone they love, each of them privately holding back what they actually think because they have all misread the room. A doctor who listens patiently can sometimes find that the agreement was there the whole time, hidden only by everybody’s reluctance to speak first. That is pluralistic ignorance, playing out in a single room. It is real.

But the same room teaches the opposite lesson just as often. Sometimes a family genuinely disagrees. And a doctor who treats real disagreement as a mere misunderstanding does not uncover a hidden agreement. He manufactures a fake one, and it falls apart at the worst possible moment. The old rule that keeps a doctor honest here is to believe everything the patient tells him, and at the same time to believe none of it, and let the true picture emerge through listening and checking rather than deciding in advance what he is going to find. That is the method for this piece. It is not the hero of the piece, and my own judgement is not what it is about.

I have made the case as strongly as I can because it deserves that. But it is the strongest version of an argument I am not going to stand on. Everything I have just told you says one careful thing: agreement can be hidden. None of it proves the much more convenient thing: that agreement is hidden wherever it would suit me to find it. To tell those two apart, I need a tool.

Four tests for what divides us

In the last piece I sorted the reasons for waiting into four kinds. The tool here is the same, and I want to say plainly that it is the same move rather than pretend otherwise, because it is a deliberate method. Only what I am pointing it at has changed.

But a country’s divisions are not as tidy as a row of separate boxes. A single argument can be a real disagreement about values, hardened into identity, and amplified by the machinery, all at once. So the honest way to use the tool is not to drop each division into one box and be done. It is to put every apparent division through four questions, and see what each one turns up.

The first is the substance test. Are people actually disagreeing about something real: their values, their material interests, the facts, or the means to a goal they share? Disagreement of this kind is not broken and it is not manufactured. It is what remains once the competing claims have been properly understood, and a great deal of our division is exactly this.

The second is the identity test. Has the position stopped being an opinion and become a marker of who you are, a badge of the group you belong to, with the other side cast as bad people and any compromise felt as betrayal? You catch it by asking whether the warmth or the coldness tracks which party someone votes for, even once you strip the actual policy away.

The third is the amplification test. Are the incentives of our media, our platforms, and our elections making a division louder, hotter, or more visible than the underlying substance would on its own? This is the one where my own language gets dangerous, and I will come to why.

The fourth is the specification test. Does an apparent agreement survive being made concrete, when the proposal, the costs, the trade offs, the losers, and the institutional detail are actually named? A warm, general goal that everyone shares can come apart the moment it becomes a specific thing to vote on.

I am going to walk through what these tests turn up, and I am taking the two where the division is usually real before the two where it can be distorted, on purpose. A piece arguing that some of our division is not what it looks like has a standing obligation to first show, in full, how much of it plainly is. Otherwise it becomes exactly the thing it claims to be diagnosing.

The disagreement that is real

Start where the argument is weakest for me, with disagreement that is exactly what it looks like: the substance test, and what it turns up in plain sight.

In October 2023 Australians were asked to write recognition of Indigenous Australians into the Constitution through the Voice, and they said no, sixty per cent to forty. The vote tracked the old, durable fault lines of Australian life: age, gender, education, and where you live. Two in three of the people who voted No gave the reason that the proposal would divide the country. Researchers at the Australian National University who studied the campaign found that the collapse of support that had once crossed party lines was decisive. The result was decisive evidence that Australians did not agree on the specific proposal placed before them. Whatever agreement may have existed around recognition in principle, it did not extend to the institutional form put to the referendum, and that is a real, structured disagreement about something serious.

Energy is the same story. In 2022, fifty nine per cent of Labor voters and eighty per cent of Greens voters said global warming was extremely important, against twenty two per cent of Coalition voters. By 2025 nuclear power split the country nearly down the middle, and split it cleanly along party lines. Underneath the party lines sit older moral disagreements too, abortion being the clearest, which follow lines of personal conviction and religious belief that no party invented and that do not dissolve when you look closer.

It is worth pausing on the temptation here, because in a piece like this it is easy to reach for the word tribalism and file all of it under feeling. Much of this firm disagreement is not tribalism. It is the reasoned, respectable conflict of people who have thought hard about difficult questions and come to different answers, and calling that manipulation is itself disrespectful to the people doing the disagreeing. What the substance test turns up is the beat that keeps the rest of the argument honest, and I am giving it full weight before I give weight to anything else.

The disagreement that is felt

What the identity test turns up is real too, but real in a different way, and describing it precisely is what keeps it from collapsing into a lazy complaint that people just dislike each other.

There is a good body of research now on how a political opinion stops being an opinion and becomes an identity. When someone holds a view as a matter of right and wrong rather than preference, the work of the researcher Linda Skitka finds people become less tolerant of anyone who disagrees, less willing to be near them, and less willing to compromise. When a position becomes what the researcher Philip Tetlock calls a sacred value, people will not trade it away for anything material, and are morally offended by the very offer to trade it. A study of the conflict between Israelis and Palestinians found that offering money to soften a position on a sacred issue actually hardened opposition, while a symbolic gesture from the other side, on their own sacred ground, could soften it. When someone’s personal identity fuses with their group’s identity, work by researcher William Swann and others shows an attack on the group is felt as an attack on the self, and compromise starts to feel like betraying yourself. Add the process researchers call moralisation, where a once ordinary preference turns into a moral cause and stops responding to argument, and you get, in a synthesis a group of researchers put together in 2020, a genuinely toxic mix: treating the other side as alien, feeling averse to them, and turning the disagreement into a moral one, all at once.

The other side of this needs the same care, because a tool carried without its limits overclaims, and this one is tempting to overuse. A lot of what looks sacred is only pretending to be. It gives way the moment people are actually asked to weigh it against something else they value. The same moral conviction that can make people intolerant also drives volunteering, civic courage, and reform, roughly evenly across the political spectrum. And the very ugliest end of this, treating the other side as less than human, is an extreme state, not the ordinary condition of a rank and file voter, and it must not be smuggled in as if it were typical.

For a reading of how hot this actually runs in Australia, the closest thing we have is the like and dislike scores the Australian Election Study collects for party leaders, and they need to be read with caution, because they mix a leader’s personality in with feelings about the party. Held with that caution, they still show something real. In 2022, voters rated their own side warmly and with rough symmetry. Labor voters gave Albanese seven out of ten. Liberal voters gave Morrison just under seven. Their ratings of the other side were cooler, and lopsided. By 2025 the coldness towards the other side had sharpened. Liberal voters rated Albanese barely above two, and Peter Dutton recorded the lowest ratings of any major party leader in the survey’s history, weak even among his own voters. There is real feeling here, and it is not nothing.

But the point I keep coming back to is this. A division that runs through people’s sense of right and wrong, and their sense of who they are, is a real division, not a manufactured illusion. If part of what divides us runs that deep, it will not dissolve because someone shows us a poll. What the identity test finds disciplines the very thesis I am tempted by, and that is exactly why I have put it early.

The disagreement that is amplified

The amplification test is the one where my own language becomes dangerous, and I want to slow right down here, because this is the point where a careful argument slides most easily into a conspiracy theory.

The structure is well documented. Anger and conflict reliably turn into attention, and attention turns into money. One large study tested real headlines on real readers and found that every extra negative word in an average headline lifted the number of people who clicked on it by just over two per cent, while positive words dragged it down, which puts some real evidence behind the old newsroom line about bleeding and leading. A study across seventeen countries found the same bias toward bad news everywhere, not just in one market. On social media, a study of over half a million messages found each extra moral or emotional word raised sharing by around a fifth, and, tellingly, that this spread within groups of people who already agreed rather than between groups who did not, which is the exact signature of division deepening rather than conversation widening. And in elections, the wedge strategy, deliberately splitting your opponent’s coalition by stirring up a divisive issue, is a recognised and perfectly rational move, because winning a handful of persuadable voters can decide a seat.

Almost every word of that paragraph describes an architecture of incentives, not a plan. The one place a deliberate choice does enter is the wedge, and it is worth being exact about it, because it is where this argument could tip into a conspiracy theory if I let it. A campaign that picks a divisive issue to peel off a few voters is making a conscious choice, and I am not pretending otherwise. But that is a local tactic played to win a seat, inside the incentive structure, not a scheme to set the whole public against itself. What I am not claiming, about any outlet, platform, party, or person, is that the broad division we all live inside is somebody’s conscious design. The wedge is deliberate and small. The climate is undirected and large. Keeping those two apart is the discipline the whole series runs on, and it matters most right here. The main book on hidden consensus, the American one this series has been arguing with, reaches, when it explains why agreement stays hidden, for a story about deliberate design: party strategists and mass media setting the public against itself on purpose, with Russian troll factories running both sides of an argument at once as the vivid example. I have read that argument, and I do not adopt it. I am holding it at arm’s length and naming it as its author’s claim, not mine, for two reasons. It is not what the evidence best supports. And it happens to be the version most flattering to my own position, which is exactly the coincidence that should make me most suspicious of it. An architecture of incentives is both more defensible than a conspiracy, and in a way more damning, because there is nobody to remove and no plot to expose. The thing simply keeps running on its own.

And it does not run unopposed, which is the counterweight I owe you, and owe you plainly, because it cuts against the very case I would prefer to make. The biggest experiment run on this question so far, a collaboration with Facebook around the 2020 United States election that changed the feeds of around twenty thousand consenting users, found that switching people from a feed ranked by engagement to a simple chronological one changed what they saw, but did not measurably shift their political hostility or their views over three months. That is strong evidence against one narrow but real claim: that swapping an engagement ranked feed for a plain chronological one, over a few months of an election, would by itself cool a person’s politics. It is not evidence that the feeds do nothing over years, and it carries its own caveats, which I am carrying too. Three months in a late campaign might miss slower, longer running effects. Only some users were changed. The platform was changing other things at the same time. A later experiment cut the other way, finding that deliberately reranking feeds to show less hostile content did lower how much people disliked the other side, which shows the algorithm is not just standing there doing nothing, even if it is not the villain either. And beside all of that sits the plainest counterweight of all. A great deal of our division is not the amplification of nothing. It is the visible edge of real and rising conflict over money, stagnant wages, insecure work, and the sense that the ground is shifting under people’s feet. Where inequality gets worse, research by Stewart and colleagues finds, division tends to rise as real conflict finds whatever lines are available to run along.

So I am not going to tell you how much of our visible division is amplification and how much is substance. I do not know. The honest reading of the evidence does not settle it, and it is very likely to differ from one issue to the next. That the amplifying incentives exist is established. How much they weigh against real substance is not, and I am leaving that open rather than inventing a verdict I cannot defend.

The agreement that is only a word

The specification test is the one that turns the tool back on itself, and I will state it here and let its clearest example land a little later, where it can do double duty.

This is agreement that looks real but is not: broad support for a goal when it is put in warm, general terms, which vanishes the moment the costs, the trade offs, and the losers are actually named. Researchers named Zaller and Feldman described the mechanism decades ago. Most of us do not carry a fixed answer to a policy question ready in our head. We build an answer on the spot out of whatever is uppermost in our mind at the time, and since most people hold conflicting thoughts on most real issues, a warm phrasing that calls up only the agreeable ones will manufacture agreement that a costed phrasing, calling up the objections, will just as quickly dissolve. There is a second trap hidden beside it. If real opinion actually splits into two separate camps, the statistical average of those two camps lands somewhere in the empty middle where hardly anyone actually stands, so a single figure claiming a moderate consensus can be nothing more than an accident of folding two opposed groups together and calling the fold agreement.

The last piece left one figure in reserve for this one: the broad support Australians show for expanding renewable energy. Here is the discipline it has to be read through. That support is real as far as it goes, but it is support for a warm, general goal, and it only earns the word consensus if it survives being costed and argued over, not simply because a headline number is large. I will not present broad support for something that sounds good as a hidden national agreement finally revealed. That is precisely the trap the specification test exists to catch, and the cleanest Australian example of it belongs in the next section, where the whole argument has to prove it can keep its feet on our own ground rather than borrowed American ground.

The Australian test

Everything I have said so far carries a standing obligation, the one it would be easiest, and most damaging, to skip. Almost all of the popular writing on hidden consensus, and almost all of that book, is American. It is a long list of American bipartisan majorities on roads, healthcare and money in politics, and not one line of it can be carried across the Pacific without our own evidence, because Australia is not the United States. Our politics are built on different bones: compulsory voting, preferential voting, and a proportional Senate with a durable crossbench that channels frustration a different way.

So take the Australian record on its own terms. Trust in government here sits at or near its lowest recorded level. The share of Australians who think people in government can be trusted fell to twenty five per cent in 2019, the lowest since this survey began in the 1960s, when it stood at around half, and has only recovered to thirty two per cent by 2025. The researchers who run the survey put even that small recovery down to the ordinary bounce that follows a change of government, not to any real return of confidence, and the number of people who think government is run for a few big interests has not improved at all. In 2025, willingness to vote if voting were not compulsory hit a record low, concentrated among the young. This is a picture of a public switching off, not of two armed camps squaring up.

And where you would expect the trenches to be deepest, you find the opposite. The share of voters who see no real difference between the major parties has risen, while the share who see a good deal of difference has fallen. The combined major party vote has drifted to historic lows, and in 2025, for the first time, more Australians gave their first vote to a minor party or an independent than to the Coalition, while the share of voters who feel aligned to no party at all reached a record quarter of the electorate. This is dealignment and realignment: a public drifting away from both major parties. Whether they are converging on some common centre or dispersing into several different camps, the evidence here does not say. But the drift away from both poles is itself the reverse of the American pattern, where the two poles have hardened.

On the one measure that lets us make a genuine comparison, the gap between warmth towards your own side and coldness towards the other side, the United States grew that gap by roughly seventy per cent between the late 1970s and 2016, while Australia over the same long stretch has trended flat, if anything sliding a little. That single fact is the load bearing piece of evidence in this whole article, and it points the opposite way to the answer I would find convenient. I am holding it firmly, and stating its limit too. The measures used to compare countries are not identical, so the exact size of the gap is contested. But the direction for Australia, flat or sliding, is not seriously in doubt.

It is tempting to give our institutions the credit for this, and there is a reasonable case that compulsory voting pulls parties towards the middle by dragging disengaged people to the ballot box, and that preferential voting rewards a candidate who can reach beyond their own base. But I have to mark that as my own reasonable guess, not proof. The best known study of compulsory voting in Australia shows that it changes who wins elections. It does not show that it cools the temperature between the sides.

Now the figure that does the most work, and the one this whole piece’s method was built to handle properly. In 2022, something like four in five Australians told the Election Study they supported constitutional recognition for Indigenous Australians, with support running between roughly three quarters and six in seven across the states. A year later the actual proposal was defeated sixty to forty. There, in one pair of numbers, is the specification test in its purest form. The four in five was real, as a reading of warmth towards a general good thing. It was not agreement on the specific, argued over, institutionally concrete change to the Constitution that the Voice actually was, and when the specific question arrived, with its trade offs and its opponents and its actual words on the ballot paper, the warmth did not survive contact with it. Broad support for an objective does not establish support for every institutional means proposed to reach it. This is exactly why this piece has to earn the word consensus case by case, rather than assume it, and why a big friendly number is the start of an inquiry, not the end of one. Some Australian agreement is broad and genuine. Some is broad and shallow. Telling the two apart, on our own evidence, is the whole discipline, and the recognition figures are the warning against skipping that work.

What kind of fix this actually points to

Suppose everything I have said so far holds up. Suppose there really is agreement out there, in identifiable places, that the system keeps hidden from us. The natural next thought is that surfacing it, through better polling, or citizens’ juries, or patiently correcting what each side wrongly believes about the other, would itself fix things. I do not think the evidence supports that, and explaining why is what connects this piece back to the rest of the series.

To the extent that real agreement is genuinely kept out of sight, the hiding is best explained by the same machine of incentives this series has spent its whole length taking apart: the media economics, the platform dynamics, the electoral rewards. Being invisible, in other words, is mostly a side effect of causes we have already named, not a fresh problem with a fresh fix of its own.

And the direct evidence that surfacing hidden agreement treats a symptom rather than a cause is fairly strong. The most authoritative summary of the research on citizens’ juries and deliberation reports, as a settled finding of the field, that good deliberation is plural rather than one voice. What it reliably produces is a workable agreement on what to do next, and mutual recognition that the other side has a legitimate point of view, not the revelation of some consensus that was there the whole time. Serious researchers in this field have rarely claimed otherwise. Australia’s own most searching example makes the point. When South Australia put the question of a nuclear waste industry to a jury of several hundred ordinary people in 2016, the jury did not come together. About two thirds came down against it under any circumstances, and about a third in favour under certain conditions, a durable split of roughly two to one, and the minority wrote a formal statement of dissent. The deliberation did not dissolve a disagreement. It surfaced one, and put it on the record properly, which is a genuinely useful thing to do, but not the thing the hopeful version of this story promises.

There is a second body of evidence pointing the same way, and it is the limit I flagged much earlier. If a large part of our division were simply false polarisation, each side misreading how extreme the other really is, then correcting that wrong impression should cool the hostility down. The evidence that it does is weak. A recent study pulling together sixty seven separate tests found that correcting people’s wrong beliefs about their opponents changes the belief readily enough, but its effect on how much they actually dislike the other side is generally small, inconsistent, and dependent on exactly what you measure. The dislike is stubborner than the misperception, which is hard to square with the idea that the misperception is what causes the dislike. This argument among researchers is live and not settled. Other careful researchers have found that correction does help, and I am presenting that disagreement as a disagreement. But the weakness of the null’s opposite is the most disciplining evidence in this whole piece, because it suggests a meaningful part of our division is real, and unhidden, sitting exactly where no amount of surfacing will ever reach it.

That is the precision I have to carry through to my conclusion. Where real agreement is kept invisible, the invisibility is mostly a downstream effect of an architecture we have already diagnosed. But mostly is not entirely, and a genuine slice of what looks like division is not hidden agreement at all. It is the honest, unhidden shape of a society made up of different people disagreeing about hard things, and no fix should try to make that disappear. One caution on numbers, learned earlier in this series. No single correlation from the citizens’ jury research should ever be dressed up as proof that assemblies reveal some hidden national consensus, and I am not resting any weight on a figure of that kind here.

Who wins if you believe me

I owe you the audit I promised at the start, and I am running it in the open, because the most important discipline in a piece like this one is not a footnote tacked on at the end. It is a test applied in full view.

Ask who benefits if you accept the strong version of this piece’s claim, that underneath the division the system sells us, we already agree. Some of the people who benefit are sympathetic. A public that shares more common ground than it can see would be a public that could act together, if only the ground were made visible, and that possibility is real, and it is the reason this question is worth asking at all. But other beneficiaries are not sympathetic, and honesty means naming them with the same weight. Any movement that wants to speak in the name of a silent, agreeing majority benefits, because claiming to speak for a majority that cannot be seen to disagree with you is one of the oldest tricks in the politics of shutting down disagreement. Anyone who would rather not answer their opponent’s actual argument benefits too, because “we secretly agree and the division is manufactured” is a way of dismissing someone without ever engaging with what they actually said.

And then there is me, the interest I put on the table at the start. A crossbench member who leads a small party has a direct, structural interest in the claim that the split between the big parties is not a real one that the public actually holds. It is not a passing interest, and it is not narrow, the way my interest in one crop was in the last piece. It goes to my standing. The honest test is not whether I feel neutral. It is whether each finding here would survive being read by someone who knows exactly what I stand to gain from it.

The careful claim survives that test. Some division is real and some is distorted. Genuine agreement exists in identifiable places and is partly kept out of sight by an architecture we have already diagnosed. A large part of our division is genuine disagreement about values, interests, or facts, or deeply felt identity, that will not dissolve just because you look at it closely. A reader who knew my interests could check every step of that against the Australian evidence and get there independently.

The convenient claim, that we secretly agree and the division is mostly manufactured, does not survive the test. That is exactly why I refused to rest on it, and why I have spent this whole piece taking apart the version of it I would most like to be true.

Where this leaves us

So the picture that comes out of all this is not the flattering one, and not the despairing one either. It is a country that disagrees, really and lastingly, about some serious things, the Voice and energy among them, and I have named those as the real disagreements they are, rather than explaining them away. It is a country whose bitterness towards the other side, measured as best we can measure it, has not climbed the way America’s has, and has by the honest reading drifted the other direction, and I have named that too, plainly, because the evidence points clearly. And it is a country in which some genuine agreement is real, but held partly out of sight, less by anyone’s design than by the ordinary running of the incentive machine this series has already taken apart.

The one thing I have deliberately not settled is how much of the division that remains is amplification, and how much is substance, because the evidence does not settle it, and I will not pretend it does. That is the judgement I am leaving open, and it is left open only where it belongs.

Which leaves a question standing, the way the last piece left one standing for this one. If some of our division is real and lasting, and some of it is felt, and some is amplified by an architecture we have already taken apart, and some of what looks like agreement is only a warm word that does not survive being made concrete, then the useful question is no longer whether we secretly agree. It is what a public that can run those four tests, and tell a real division from a distorted one, is able to do about the divisions that turn out to be real. That is the ground the next piece walks.

Walker Briefing is a reader-supported publication. To receive new posts and support my work, consider becoming a free or paid subscriber.