Ask Ethan: Is the Hubble tension a real problem?

There’s a mystery over how quickly the Universe is expanding, and it doesn’t appear to be going away anytime soon. Back in the early 2000s, after decades of astronomers arguing over whether the expansion rate of the Universe was closer to 50 km/s/Mpc or 100 km/s/Mpc, the Hubble Key Project was completed. Their results — the most precise in history — measured an expansion rate of 72 km/s/Mpc, with an uncertainty of just 10%. For the next decade or so, all the results remained consistent with this value, but then in the 2010s things got more precise. Measurements of the cosmic microwave background, the large-scale structure of the Universe, and of the cosmic distance ladder all improved, and led to a puzzling situation: where more-precise measurements using different methods led to incompatible results.

The groups using an early relic method, where you start at the Big Bang, calculate a phenomenon, and evolve the Universe forward in time until you see what it gives you, measure an expansion rate of 67 km/s/Mpc, with an uncertainty of just ~1%. But if you start nearby and measure progressively more distant objects, using the distance ladder method, you get ~73 km/s/Mpc, with a very small uncertainty. But just a couple of days ago, I published an interview with famed astronomer Wendy Freedman, who claimed those distance ladder uncertainties may not be so small. This leads to an important follow-up question when it comes to the rate of cosmic expansion, exemplified by Michael Greczek, who inquired:

“Can you review the [latest] data and write a short synopsis of the initial conclusions? It appears that the Hubble Tension is still unresolved, but the blizzard of papers on arxiv.org are very difficult reading for the non-expert.”

There’s a lot of ongoing research here, but the important thing, whenever you want to take stock, is to look not at what one group or another says, but the full suite of available evidence. Let’s examine the issues at play.

This comparison plot shows the difference between theoretical predictions and experimental results for the muon’s magnetic moment. Data-driven (green) calculations of the hadronic vacuum polarization, as derived from various experiments, are not all consistent with one another, with CMD-3’s results (and results inferred from the tau lepton) differing from all others, being the only one marginally consistent with the experimental results. However, lattice QCD predictions, having improved dramatically here in the 2020s, lead to a great general agreement with the experimental data.

Before we get into the distance ladder specifically, I want to show you the above graph, which is about a particle physics result: the magnetic moment of the heavy, unstable fundamental particle known as the muon. The red points show experimental data; the blue point shows a theoretical prediction derived from using Lattice QCD calculations to estimate the hadronic vacuum polarization contribution; the green points show a theoretical prediction if — instead of Lattice QCD calculations — the results of a variety of experiments are used to estimate the hadronic vacuum polarization. Note that:

• the results from SND06, CMD-2, KLOE, BESIII, and SND20 are all consistent with each other, giving a low value,
• the results from BaBar are higher, but still disagree with experiment and Lattice QCD,
• the results from tau lepton studies are still higher, but still disagree with experiment,
• while the results from CMD-3 are still higher, and consistent with direct experiment and Lattice QCD.

When you see results like this, it tells you something important: the systematic errors for the green points are being underestimated. If the same type/class of measurements are giving you vastly differing results, outside of the published error/uncertainty bars and disagreeing with one another, that tells you there are systematic errors at play that are not being properly quantified. If some small difference, like a different method, a different set of observations or measurements, a different analysis technique, or a different calibration, can change your results by more than the size if your error bars, then you’ve underestimated your systematic errors, and are artificially reporting a high precision that is not reflected by the actual science.

This graph shows the global average temperature anomalies relative to the 1850-1900 baseline. The red line shows the multi-year moving average of global temperature, while the dotted green line shows a linear fit to the warming from 1974-2022. As recent years show, the warming trend has accelerated, with 2023 and 2024 marking severe (hottest-ever) departures from the late-20th century trend.

On the other hand, sometimes people express an unhealthy skepticism of very well-established (on scientific grounds) facts. This unhealthy skepticism is often called denialism, and rears its head very often in scientific circles in two different ways.

1. One way is when a new, high-quality set of evidence comes in and strongly points to one conclusion. Unhealthy skeptics will reject that data, attempt to poke holes in it, and make flimsy claims to the contrary, with no regard for the actual, factually-supported scientific truth. Fred Hoyle and Geoffrey Burbidge did this in the late 20th century to argue (unsuccessfully) against the hot Big Bang; today, people like Pavel Kroupa do this to argue against dark matter. There is no way to “convince” a denialist with evidence; you must simply continue telling the truth despite the noise they generate.
2. The other way, however, is when someone has convinced themselves that there are real problems within a field, and only they, with their brilliant and unbiased analysis, are capable of conducting this type of science scrupulously. It is an incredibly unhealthy form of egotism, exemplified by Berkeley physicist Richard Mueller, most famous for being a climate change skeptic who only changed his mind when he performed the science himself, basically confirming the results of every other major group or organization (NASA, HadCRU, NOAA, etc.) that had worked on the problem.

The non-expert, to be frank, has no way to tell these scenarios apart.

• Are there legitimate reasons to believe there are systematics that even experts are overlooking?
• Is this a case of someone ignoring the evidence and promoting their own spurious claims with flimsy evidence?
• Or is this a case where someone is devaluing the work of everyone else in the field, convinced that they and only they are capable of getting it right?

To decide, we need to take an expert-level look at the claims being made with all of the latest data (including papers that were published even more recently than when the interview with Wendy Freedman was conducted), and evaluate them on scientific merits alone.

A 2023-era analysis of the various measurements for the expansion rate using distance ladder methods, dependent on which sample, which analysis, and which set of indicators are used. Note that the CCHP group, the only one to obtain a “low” value of the expansion rate, is only reporting statistical uncertainties, and does not quantify their systematic uncertainties at present. There is overwhelming consensus agreement that the expansion rate is around 73 km/s/Mpc using a wide variety of distance ladder methods.

We have to start where everyone agrees: that’s the foundation for moving forward in science. That foundation includes:

• That the early relic methods, including the CMB from Planck (but also from ACT, the South Pole Telescope, etc.) and from baryon acoustic oscillations (including DESI), support a low expansion rate of ~67 km/s/Mpc, with overall a low uncertainty of ~1%.
• That dark energy is a real component of our Universe, making up about ~70% of the energy density, and so is dark matter (at about ~25%) and normal matter (at about ~5%), with a tiny bit of neutrinos and photons thrown in as well.
• That the age of the Universe is very close to around 13.8 billion years since the start of the hot Big Bang, with again an uncertainty of only around ~1%.
• And that the concept of the distance ladder is sound: to start nearby by measuring stars in the Milky Way, then finding similar stars in nearby galaxies and measuring them there while also measuring a different property of those galaxies, and then going out into the deeper Universe to measure those galactic properties all across cosmic history.

Above is a graph from a 2023 review paper, highlighting several different distance ladder techniques and showing the inferred value of the expansion rate of the Universe derived from them. Note the difficulty of reconciling all but the CCHP measurements (perhaps except for those with the largest errors/uncertainties) with the low values obtained by the early relic method. The existence of large numbers of very imprecise, additional methods — something shown in the plot below that is a more modern, unpublished extension of earlier work published here — does nothing to change the precision of the methods that have the smallest errors on them.

The red points, shown with error bars, reflect the highest-precision values of the expansion rate published in the literature, leveraging Cepheid variable stars with the dwindling error bars over time reflecting improvements in data and our understanding of the underlying systematics. Blue points are shown without error bars, and represent an enormous variety of methods, precisions, and assumptions on the expansion rate found in the literature.

If you want to argue against a mainstream position, you owe it to yourself to address the strongest point that supports such a position. In this case, the strongest point supporting a high expansion rate (of ~73 km/s/Mpc and not ~67 km/s/Mpc) is the most accurate, low-error version of the distance ladder:

• Cepheid variable stars measured nearby in the Milky Way and its satellites,
• then those same types of Cepheids in nearby galaxies,
• many of which also have hosted at least one type Ia supernova within them,
• and then to measure type Ia supernovae all across the Universe, including at great cosmic distances.

It was this version of the distance ladder that claimed to achieve a precision of just ~1% for the very first time, elevating the Hubble tension to a full-blown 5-sigma statistical significance: the “gold standard” for discovery in astrophysics. These two figures, below, from the paper that claimed to achieve that significance for the first time, show how both the various sources of error were reduced since 2001 (when the Hubble Space Telescope Key Project’s results came out) and also what the robustness of “5-sigma” looks like, from the paper that first made that claim.

Back in 2001, there were many different sources of error that could have biased the best distance ladder measurements of the Hubble constant, and the expansion of the Universe, to substantially higher or lower values. Thanks to the painstaking and careful work of many, that is no longer possible, as errors have been greatly reduced. New JWST work, not shown here, has reduced Cepheid-related and period-luminosity errors even further than is shown here.

The discrepancy between the early relic values, in blue, and the distance ladder values, in green, for the expansion of the Universe have now reached the 5-sigma standard. If the two values have this robust of a mismatch, we must conclude that the resolution is in some sort of new physics, not an error in the data. There is the possibility, raised by other groups, that somewhere around 70 km/s/Mpc, however, these measurements may yet both be plausible. That position will become less defensible as more independent lines of evidence confirm and validate the SH0ES results.

In the interview I conducted, it was asserted that the collaboration behind that result, SH0ES (Supernova H₀ for the Equation of State) didn’t release all of their raw H-band data for objects: especially for those that were deemed outliers. In principle, making your data:

• open,
• available,
• in all bands,
• including both raw and reduced data,
• along with a transparent methodology laid out for your data reduction,

is a cornerstone for how science should be conducted. Indeed, the whole point behind making data public is so that the community can figure out how to reanalyze, recalibrate, and reinterpret it if necessary. But the exclusion of H-band data for low-quality, outlier data doesn’t harm the community; it’s available upon request and hasn’t been requested of the SH0ES collaboration.

On the other hand, the alternative method used by the CCHP collaboration, instead of using Cepheids, is to use giant stars: evolving stars at the tip of the red giant branch and stars on the asymptotic giant branch. But that data, importantly, hasn’t been made public: not even the color-magnitude diagrams for the tip-of-the-red-giant-branch stars are publicly available. And that’s a problem, because even though the underlying physics behind those stars is known (stars evolve on a specific curve until helium fusion ignites in the core, and then evolve along a different curve), the observable signatures that those stars generate (even according to Wendy’s own source, Lars Bildsten) is much more ambiguous.

The evolution of a solar-mass star on the Hertzsprung-Russell (color-magnitude) diagram from its pre-main-sequence phase to the end of fusion and its eventual transformation into a white dwarf. The “helium in core ignites” caption corresponds to the tip of the red giant branch, whose physics is robustly understood but whose observable signatures are not as “clean” and universal as we’d desire for an ideal distance indicator.

It’s easy to complain that old samples of data — whether type Ia supernovae, Cepheids, or any other distance indicator — have all required recalibration, but this is one of those cases that is truly a feature, not a bug. It’s important to note that our understanding of many things in science:

• how stars, supernovae, and different sets of stellar cataclysms and phenomena, work,
• what observable signatures appear at different sets of wavelengths,
• what variations in those signatures are natural between objects within the same class,
• what distances to objects used in calibration are,

and many others, grow and improve over time. For example, consider a distance estimate to a certain star that’s useful in calibrating your distance ladder, and imagine that star is initially determined to be at a distance of 770 parsecs, with an uncertainty of just ±10 parsecs.

Now imagine that we have a new, superior tool for measuring stellar distances: something like the ESA’s Gaia mission. All of a sudden, that distance is now determined exquisitely, and turns out to be at 762.3 parsecs, with a new uncertainty of only ±0.9 parsecs. If you’re going to be responsible, you have to go back to all of the old calibrations that were performed previously — calibrations that relied on the old 770 parsec figure — and recalibrate them. That small difference, even though it’s just a 1% difference in stellar distances, is going to propagate to observables like brightness, inferred galactic distances, and the overall expansion rate as a result. When we gain new information, we have to do a recalibration of older data to be consistent with the new things that we’ve learned; using the old, inferior calibration would be the scientifically unscrupulous thing to do. After all, it’s only with these improvements, made over time, that we can reduce the systematic errors in our analysis.

This screenshot shows a single Cepheid variable star as imaged with the Hubble Space Telescope in galaxy NGC 7250 at a distance of 20 megaparsecs. While JWST may be able to resolve such a Cepheid individually, the Hubble data is much more ambiguous.

It’s also easy to complain that some Cepheids, like the one shown above, could be biasing our results when we attempt to measure their brightnesses over time and infer the distance to the galaxies that host them from those measurements. After all, a slight difference in brightness and/or distance could dramatically affect the cosmic distance ladder, and our inferred rate of expansion as a result.

But do “bad Cepheids,” like this one, truly have a meaningful impact?

Not really, and it’s important to understand why.

• First off, we don’t determine distances to a galaxy through one or even a handful of Cepheids, but through several dozen-to-hundred Cepheids.
• Second, those Cepheids are not all assigned the same weights, but the worse Cepheids are assigned a greater uncertainty (or, if they’re bad enough, excluded entirely from the analysis), which means they contribute less or even not at all to the overall inferred galactic distance.
• And third, both the statistics on Cepheids, including Cepheids-per-galaxy, as well as the uncertainties on those Cepheids, are superior, star-for-star, to either tip-of-the-red-giant-branch or asymptotic giant branch stars.

In other words, yes, you can’t usefully measure every star within a class of distance indicators to the needed precision to have it be a high-quality data point in determining the distance to an extragalactic object. But the existence of less-than-ideal stars doesn’t mean the measurement technique is no good.

In fact, the ideal test should be to use many different classes of distance indicators: not just Cepheids, not just giant stars, but also RR Lyrae stars, Mira variables, water masers, and anything else you can reliably understand and measure. Same thing with type Ia supernovae: use other options like fundamental plane galaxies, Tully-Fisher galaxies, surface brightness fluctuations, etc. If there’s a problem with any one particular method, comparing those results with the other methods will reveal which one, if any, is the outlier.

This graph shows the Hubble diagram inferred using distances derived from the updated surface-brightness-fluctuation zero point calibration from tip of the red giant branch measurements and improved optical color measurements as of February 2025. Although galactic properties like Tully-Fisher, fundamental plane, or SBF may have more intrinsic variation than type Ia supernovae, they have much greater statistics, allowing for comparably precise distance ladder estimates of the expansion rate.

But the biggest problem — at least, as I see it — is that the CCHP group isn’t doing something more cleanly, pristinely, or with less uncertainty than any other group, which is their claim, but they are using only a small subsample of the full suite of objects they could use in their analysis, which is a common logical (and scientific) fallacy known as cherry-picking.

It’s important to address both of these aspects. First off, it’s okay to have a smaller sample size if your sample is demonstrably superior to a larger one: because it’s higher-quality, because it has smaller uncertainties, because it can be measured with higher-precision, etc. But their uncertainties, both statistically and systematically for both the TRGB and the JAGB methods, are greater than the uncertainties associated with using Cepheids.

And secondly, although Wendy claimed in her interview with me that she selected which galaxies to use in her subsample based on distances, that turned out to be demonstrably untrue. The 24 galaxies in the CCHP sample are not the nearest of the 35 galaxies used in the overall Cepheid/SH0ES sample, but instead exclude some of the 24 closest and include some of the 11 most distant. Notably, they even exclude a nearby galaxy that they themselves previously measured and published a paper on: NGC 7814 (which housed supernova SN2021rhu, on the right in dark red in the figure below). Had they included it, it would have raised their inferred expansion rate substantially. They have, indeed, cherry-picked a subsample of galaxies that biases them, artificially, toward lower values of the expansion rate.

This chart shows the 35 possible galaxies to choose from that have resolvable stars (Cepheids, tip of the red giant branch, or JAGB) and also were host to at least one type Ia supernova. The light red galaxies show which galaxies were included in the CCHP results; the dark red were excluded.

It is indeed true that, as Wendy noted in our interview together, it may turn out that the expansion rate of the Universe, as measured via distance ladder methods, will eventually return to being a lower value, maybe even below 70 km/s/Mpc, and fall into line with the measured value of the expansion rate as inferred from the early relic methods. When she says,

“My view is we just need to be a bit more patient, or, you know, collectively impatient. We want to see an answer. But this is an empirical question. We will get to the bottom of it. And I think we are doing that. I think the accuracy — I mean, you know where I came from, right? It was a factor of two. It astounds me now that we’re talking about a couple of percent accuracy. Groups that historically have not agreed are coming to agreement. And then we find where there are other places where we disagree, well, now we’ll look in that direction. And I think we’re going to make, in the next few years, a lot of progress in understanding where these differences have arisen.”

This is true: more and better data, with more careful calibration, superior statistics, and an improved understanding of our sources of error will all lead to us better understanding what drives different groups using different methods or different distance indicators to get different values. But it strikes me that, when I look at the data from all groups that aren’t associated with the CCHP collaboration, there’s a lot more agreement than there is disagreement.

A compilation of distance ladder measurements of H0 in comparison to the Pantheon+SH0ES, where the third rung of the distance ladder is redone using various techniques. The legend shows the different techniques included in constructing this figure.

And this is why I’m extremely hesitant to accept the assertion that we just don’t know what the expansion rate of the Universe is today, and that we just need to wait for even better data than what we already have in hand. Sure, better data is forthcoming:

• from the Vera Rubin Observatory, which is just beginning its science operations,
• from JWST, which continues to probe the Universe as never before,
• from ESA’s Gaia mission, which will still have one final data release to come,
• and from ESA’s Euclid and NASA’s Nancy Roman Telescope, which will survey large areas of the sky, deeply, from space, enabling us to find more supernovae and measure many more distant properties of galaxies to unprecedented precision.

But, as you can see in the figure above, there is already agreement among groups using different methods, with different members doing the analysis, and with different approaches to the distance ladder. While only groups using Cepheids and type Ia supernovae have gotten down to a ~1-2% uncertainty, the other groups get the same consistent values with only slightly larger uncertainties. Most importantly, they’re not consistent with a conclusion of “no Hubble tension;” only with the idea that there may be a slightly less (or more) severe Hubble tension instead.

Imagine what the community would think if the situation were reversed: if all of these other groups had a lower value for the expansion rate while CCHP’s results were higher. Would their conclusions be taken seriously, or completely ignored? It’s my opinion that it’s only because their conclusion is favorable — more in line with the standard model of cosmology than the other groups — that it’s so appealing to think that they’ve got something right that everyone else has missed. In reality, their results are not the best or most precise or most accurate distance ladder results at all; they’re just the ones that give the lowest value for the expansion rate. Until their results can be robustly reproduced, independently and with open, accessible data, the default conclusion should be that the Hubble tension is real, and we should be taking this discrepancy extremely seriously. That conclusion might change, but if it does, it will fall to new, superior data to change it. For now, the Hubble tension remains unresolved.

Send in your Ask Ethan questions to startswithabang at gmail dot com!