CSO - the Central Signal Office

A dwelling can be in ferocious demand and still hold no resident on census night — if its most profitable use is not housing a household.

The previous post made a narrow claim. The Census vacancy figure is a measurement, and it deserves to be engaged with as one. The long-term measures — GeoDirectory, the electricity count — answer a different question, and treating the narrow measure as if it refutes the broad one is a category error, not a correction. The Central Statistics Office can count empty houses. The thing nobody did was ask what the count was telling us.

This post takes the next step. Ignore the reasons. Start from the assumption that the CSO can count, that its numbers are accurate. The highest-demand parts of Dublin really do have the highest vacancy — not despite being in demand, but whilst being in demand. What would explain that? And is the explanation strange enough to dismiss, or ordinary enough that we should have expected it all along?

Signal and noise

Every dataset contains both. Signal is the part that carries information — the pattern that means something. Noise is the random scatter on top of it: measurement error, outliers, one-off flukes, the points that don’t line up with anything. An essential skill in reading data is telling the two apart.

If you treat noise as signal — you may try and chase a pattern that isn’t there, or build a theory on a coincidence, but ultimately you’ll learn nothing useful. More importantly if you treat signal as noise — you’ll not only learn nothing useful, you’ll miss the opportunity to learn something new, dismissing something as an outlier because it doesn’t fit what you expected.

The Irish vacancy debate has managed to do both at once, and in exactly the wrong combination.

The census recorded a number — a count of empty homes — and a set of reasons those homes were given as empty. The count is the hard data: a national enumeration, trained staff, repeat visits, a consistent method. The reasons are the soft data: a category ticked for a sample, self-reported or a neighbour’s best guess, largely unverifiable. And the consensus took the hard count to be the unreliable thing — too high, must be wrong, the CSO can’t really count empties — while treating the soft reasons as the solid ground to argue from: “look, we know there were not 30k available rental properties, the data is unreliable.” The count was filed as noise. The reasons were promoted to signal. It was precisely backwards.

I will come back to the reasons later, because they reward a closer look than they got. For now, set them aside entirely and stay with the count — the part that was dismissed — because Ireland’s census happens to have unusually good data for testing whether a dismissed number is noise or signal. It reports vacancy not just nationally and by county, but in exceptionally local granular detail: by Local Electoral Area, and below that by Electoral Division. If a number is noise, it scatters as you zoom in. If it is signal, it sharpens. So let us zoom in.

The data by LEA in Dublin

Here are the top ten highest-vacancy Local Electoral Areas in Dublin, on the 2022 census-night measure — the whole-stock snapshot — with the 2021 median sale price alongside.

For context, number 10, Artane-Whitehall, matches the whole Dublin rate at 5.6%, and the lowest LEAs are out in the suburban ring — Palmerstown-Fonthill at 2.4%, Lucan at 2.8%, Firhouse-Bohernabreena at 3.2%.

The first thing that jumps out is price, and it jumps out because of one name. Pembroke LEA — Ballsbridge, Sandymount, leafy, valuable, in demand, the most expensive electoral area in the country at €800,000 — also has the highest vacancy in the county. That is not where the consensus says the empty homes should be. The dearest district in Dublin should, on the standard logic, be the one most fully occupied: every unit precious, nothing left idle. Instead it sits at the top of the vacancy table.

Is Pembroke an outlier — a single weird point, the kind you dismiss to clean up the chart, the noise, as it were? Let us zoom in further.

Pembroke, by Electoral Division

Drop below the LEA into Pembroke’s constituent Electoral Divisions, and the single high number resolves into a field of high numbers.

These are not just among the highest vacancy rates in Dublin. They are among the highest in the country — and they sit in the highest-demand square kilometres in the State. The lowest ED here, Rathmines East C at 6.5%, would still rank near the top of the all-Dublin LEA table. Pembroke is not one vacant pocket dragging up an average. It is high throughout, and it peaks at nearly 16% in the heart of Ballsbridge. Whatever this is, it is not a rounding artefact and it is not confined to a single odd corner.

By contrast, the areas with some of the lowest demand in Dublin by price carry some of the lowest vacancy. The cheaper outer LEAs — Tallaght, Clondalkin, Lucan, Palmerstown — cluster near the bottom of the vacancy table with the lowest vacancy rates in Dublin.

At first glance: high price, high vacancy; low price, low vacancy; the two move together, but inverted to what Economics 101 tells us.

But it is not that simple, and the places where it breaks are the interesting ones. Some genuinely high-demand, high-price areas sit below the Dublin average. Rathfarnham-Templeogue, comfortable middle-class suburbia at €550,000, runs at 4.4%. Glencullen-Sandyford, prosperous and well-served, is at 4.5%. If high vacancy was actually correlated to high price, these would be elevated. They are not.

When you notice the pattern is not following price cleanly — when high-priced suburbs come in low and a mixed north side inner city areas come in high — you are being told that price is a proxy for something else, and not a perfect one. The real correlation is not just price. It is spatial. It is the inner urban core versus the outer suburban ring. Inner city plus amenity rich leafy and coastal areas closest to the city have the highest vacancy.

No. 5 on the table looks at first glance as though it may buck this trend. Cabra-Glasnevin — middling on price, €395,000, on the whole not a gilded district, nor in the heart of the inner city — sits up at 7%, comfortably above the Dublin median.

Then you drop below the LEA, to its constituent Electoral Divisions, and the outlier dissolves into a gradient. The vacancy is not spread evenly across Cabra-Glasnevin. It is concentrated, and it is concentrated in one direction.

The high-vacancy EDs are the inner city, eastern end, topped by Inns Quay A and B. The lowest vacancy EDs, the Cabra West housing estates and outer Ashtown, are below the Dublin average. The split runs almost monotonically from east to west, and it is steep: the top of the LEA is at 11%, the bottom near 4%. These are a couple of kilometres apart — and one end of the LEA is empty at nearly three times the rate of the other.

(One ED earns an asterisk rather than a theory. Phoenix Park, at 9.3%, is not a residential neighbourhood in any ordinary sense — it is the Áras, the US ambassador’s residence, Farmleigh, the official state guest accommodation, various OPW buildings. Its “vacancy” is institutional, and it should be read out of the residential pattern, not into it. It is a genuine outlier, noise as it were! Set it aside and the gradient is, if anything, cleaner.)

For Cabra-Glasnevin a radial “distance from the city centre” story does not quite capture it either, because the real axis is not the GPO. It is Croke Park. The high-vacancy EDs are the streets within walking distance of the largest stadium in the country, that anyone who has ever tried to find a room for an All-Ireland final or a three-night concert run knows intimately. It is not prime residential territory. But it is prime short-let territory: central, walkable to the stadium and the city, on the rail line. The closer an ED sits to that axis, the higher its vacancy. A Dubliner does not need the regression. They need the location named, and then it is obvious.

At the scale of the Electoral Division, the census vacancy is not a puzzle at all. The vacancy is exactly where conditions combine to reward vacancy. The more hyper-local you look at these numbers, the more they make sense — which is the exact opposite of what you would expect from noise. Noise gets less coherent as you zoom in. Signal gets stronger

Economics 101

There is a phrase that gets reached for whenever the vacancy debate surfaces, to make it sound inarguable: Economics 101. High-demand areas have low vacancy. It’s almost tautological — people are more likely to live where demand is high, prices are higher there, so the incentive to leave a property empty is weaker. Prosperous cities have low vacancy; sleepy villages have empty houses.

Notice what the word “tautological” does. If the relationship is a tautology, it is so obvious it is entirely unnecessary to explain it, it cannot be wrong, and anything that contradicts it cannot be data — it can only be noise. A weird outlier. An exception that proves the rule. This approach doesn’t just predict low vacancy in high-demand areas; it pre-commits you to disbelieving any high-demand area that shows high vacancy, before you’ve even looked at it.

The belief only holds because of a buried assumption. “People are more likely to live where demand is high” quietly equates demand with long-term residential occupation. It also quietly assumes two other basic tenets of Economics 101 - (i) “ceteris paribus” or all other things being equal and (ii) human beings will act rationally in their self interest.

But what happens when all other things are not equal? A rational decision depends on real circumstances, not textbook assumptions.

If you take the position that the CSO’s count is more likely to be reliable than your own confirmation bias, the question stops being “where can I find better data?” and becomes “do conditions exist in which a rational owner would leave a high-value property empty?”

If those conditions do exist — and especially if several stack on top of one another — then a pattern of elevated vacancy in exactly the areas the census identified is not noise. It is the signal.

When vacancy is the rational choice

A dwelling can be in ferocious demand and still hold no resident on census night — if its most profitable use is not housing a household. The census counts whether a household was occupied. It does not count whether the property was in high demand. In much of the country those are the same thing. In the most valuable few square kilometres of Dublin, they come apart.

Economics 101 assumes the only way to make money from a dwelling is to put a household in it — sell it or let it to a resident — so an empty home is always money left on the table, and the higher the value, the more money, and therefore the stronger the pull toward occupation. Anything else would be irrational. That logic is sound only if letting or selling is the best available use. If a more profitable (or less risky) use of an empty property exists, the incentive inverts: high value now raises the return to not housing a household.

Changes in the market over the last ten to fifteen years have created conditions with both opportunities and risks that incentivise vacancy.

A rising market rewards waiting. When prices are climbing, an empty property held through the rise captures the capital gain without the frictions, obligations, or sitting-tenant risk that come with letting it. Vacancy is not a cost to be minimised; it is a position being held.

RPZ rules made vacancy a way to reset the rent. Through the period the census measured, under the Rent Pressure Zone regime, permitted rent increases were capped and a below-market tenancy could not simply be repriced. The cap attached to the tenancy, not the dwelling. For an owner sitting on a long-standing under-market let, the route back to market rent ran through a period of vacancy. The regulation designed to protect tenants made an empty property, for some owners, the rational waypoint to a higher return. The incentive to hold a unit empty was written into the legislation.

This incentive is strong whether you intend to let it in the future or sell it on the open market. In all the areas of Dublin that have high vacancy rates, properties that have no RPZ restrictions will carry a premium.

And meanwhile, there is an alternative income that beats the rent. This is the one that has been measured, and the figures are not subtle. The ESRI’s 2025 profiling of short-term lets across Ireland — Profiling Short-term Let Usage Across Ireland, Devane, Kelly and Slaymaker — set the daily short-let price against the daily private-rental price and estimated how many nights a property would need to be let to match a full month’s rent. In Dublin City, the answer is eight to ten nights a month. Let a property for roughly a third of the month on the short-term market and you match what a full-month tenancy would pay. The report is careful that these may not be like-for-like properties, but it draws the obvious conclusion itself: the figures highlight the attractiveness of the short-let option. An owner clearing the equivalent of a month’s rent in eight nights with no legislative obligations has little incentive to take a long-term tenant — and every reason to leave the property empty, between guests.

It is unlikely to be a coincidence that the highest vacancy rates in Dublin are in exactly the same areas in which you’d expect to find the highest demand for short term lets eg North and South Inner City, Ballsbridge and Croke Park. Rathfarnham and Stepaside have comparatively low vacancy rates. They are expensive areas, hugely popular with affluent middle-class families - less so with airbnbers.

None of these conditions is contested. The rising market is observable. The RPZ rules are on the statute book. The short-let economics are an ESRI table. The only question is what they add up to.

The vacancy data shows what some of the most-wanted stock in those neighbourhoods is actually doing — which is sitting empty on census night, because in the highest-value, highest-amenity locations, the alternatives to long-term occupation are most profitable precisely where demand is most intense. Demand does not guarantee occupation. In specific locations, it competes with it.

Back to the reasons

Now the reasons can be picked back up, because zooming in has clarified what they mean.

The census count was dismissed in large part because the reasons didn’t smell right — among them, 33,653 properties recorded as vacant while enumerated as “for rent.” That looked absurd. There weren’t thirty thousand empty rentals sitting on the open market. That is undisputed. So the count must be inflated. The reasons discredited the number.

But turn it around. The count is the signal; the reasons are the noise. And once you know the conditions are present, the reasons start to look less like proof the count is wrong and more like proof the reason field is. Consider what each condition looks like through the census form. A short-let is, by definition, enumerated as a rental — it is a property let to people, just not for long. A unit deliberately held idle while the rent resets before a new tenancy is, by definition, enumerated as a rental. A property kept empty until it can be sold free of RPZ legislation is, by definition, enumerated as “for sale.” A property whose owner intends to sell but feels it is rational in a rising market to wait until next year, is by definition, enumerated as “for sale.” Every one of those is a rational reason to be empty on census night, and every one of them lands in a reason category — “for rent,” “for sale” — that the discourse then pointed to as evidence the home wasn’t really vacant.

By getting caught up in the reasons, we dismissed the signal as noise, and promoted the noise to signal. The count was telling us how many homes were empty. It was right. The reasons were never built to tell us why with any reliability, and they didn’t — but they were the part everyone reached for.

Signal or noise

So here is the choice, laid out plainly, because the whole thing comes down to it.

On one side: the high-vacancy areas are outliers. Weird districts. Exceptions to a rule that is basically tautological. The data has a general shape nationally — high demand, low vacancy — Co Dublin compared to Co Leitrim confirms this. Anything that contradicts this are the scattered points that don’t fit, explained away one at a time and not worth a theory.

On the other side: the conditions that would make vacancy rational in high-value central Dublin areas are all present, documented, and uncontested. A rising market that rewards holding. A rent-cap regime that made vacancy the route back to market rent. A short-let market that matches a month’s rent in eight to ten nights, concentrated in the central LEAs, skewed toward the scarcest small units. Stack those up, and a pattern of elevated vacancy in exactly those areas is not the exception to the rule. It is exactly what you would expect.

Ask which is more likely. That several of the most reliably measured, most desirable, highest-value districts in the country happen, coincidentally and individually, to be statistical flukes — flukes that line up along a stadium and a coastline and a rail line, and that get more coherent the closer you look. And that this coincidence scanned over the period of two censuses.

Or that they are showing you, accurately, rational human beings making rational decisions based on local market circumstances.

The consensus says the first. It says the count must be misleading, because Economics 101, and the areas that contradict must be noise. But all other things are not equal. Look at the conditions, look at the data, zoom in as the signal gets stronger and stronger. The second answer is not just available. It is the more economical and plausible one.

Two consecutive censuses told us something. The last post argued we should believe the count. This one argues that once you do, the explanation is not exotic.

Please stop the obsession with electricity meters, and let the Central Statistics Office get on with what it does best - providing us with reliable and accurate data.

It’s not their fault if those who interpret the data struggle to separate the noise from the signal.