Tony Wang8 min readThe Same Query Says This Duopoly Is 36% or 71%. Only the Sort Order Changed.
Change only the sort order and the same dialysis duopoly measures 36% or 71%. Which giant looks dominant flips entirely.
Across this series we have repeatedly labelled sample figures "sample-relative, not national extrapolations." Dialysis is where we set out to measure how much that caveat is actually worth. The answer is: considerably more than we had been treating it as.
A genuinely concentrated industry
First the substantive finding, which is real. Dialysis in the United States is close to a textbook duopoly. DaVita and Fresenius Medical Care between them run roughly 2,600 to 2,800 US centers each, and industry reporting puts the pair at about 80% of facilities and 90% of industry revenue. Behind them sit a handful of mid-sized operators — US Renal Care, Innovative Renal Care, Dialysis Clinic Inc., Satellite Healthcare.
That concentration shows up immediately in the data. A 100-record sample of dialysis_center resolves to 18 unique website domains — the second-most concentrated category in this series, behind only dollar stores at 4 and far ahead of plumbers at 99.
Then the problem
We ran the same 100-record sample three times against the same category, changing only the sort parameter.
| Sort order | Unique domains | DaVita | Fresenius | Measured duopoly |
|---|---|---|---|---|
| updated_at_desc (recency) | 18 | 17.0% | 54.0% | 71.0% |
| rating_desc | 35 | 45.0% | 12.0% | 57.0% |
| review_count_desc | 55 | 23.0% | 13.0% | 36.0% |
Three things move at once, and all three matter:
- The duopoly's measured share ranges from 36.0% to 71.0% — a 35-point spread on the headline number.
- Which company appears dominant reverses entirely. Recency makes Fresenius look 3.2x larger than DaVita; rating makes DaVita look 3.75x larger than Fresenius. Reality is a near-tie.
- Even apparent fragmentation moves. Unique domains run 18, 35, or 55 for the identical category — so the "how consolidated is this industry" reading itself depends on sort.
The correction we had to make
Having found the sort problem, we thought we had the fix. A direct brand query scoped to one state is sort-independent, and Texas looked decisive:
| State | DaVita | Fresenius | Ratio | Reading |
|---|---|---|---|---|
| Texas | 302 | 287 | 1.05x | Near-parity — matches reported reality |
| California | 410 | 127 | 3.23x | Heavy DaVita skew — contradicts it |
Texas alone would have supported a clean conclusion: state-scoping removes the bias and recovers the true near-parity. California disproves it. Whether the difference reflects genuine regional market share, uneven listing coverage, or both, we cannot tell from this data — and that is the point. A single confirming state is not validation; it is one observation that happened to agree.
We are reporting this because we nearly published the Texas-only version. The check that caught it was simply running a second state.
It is not a dialysis quirk
To test whether this is specific to a concentrated healthcare category, we re-ran dollar stores — a category from an earlier study — changing only the sort.
| Sort order | Unique domains | Dollar General | Dollar Tree | Family Dollar |
|---|---|---|---|---|
| updated_at_desc | 4 | 50.0% | 28.0% | 22.0% |
| rating_desc | 10 | 71.0% | 14.0% | 8.0% |
Dollar General's measured share moves 21 points; Dollar Tree's halves; Family Dollar's drops by nearly two-thirds. The three-company total stays high either way — that finding is robust — but the split between them is not.
What this means for everything else we have published
Every domain-classification figure in this series used sort=updated_at_desc, and every one was labelled sample-relative. That labelling was correct but understated. The honest position now:
- Rankings of "who is largest" within a category should be treated as provisional unless corroborated by external company-reported counts.
- Aggregate concentration findings are more robust than individual brand shares. Dollar stores being overwhelmingly three companies survives the sort change; which of the three leads by how much does not.
- The reliable anchor is external data. Where this series cross-validated against company filings — Mars at 5.1% / 6.5% / 6.8% by three independent methods, or Athletico's ~928 listings against a stated 900+ locations — those findings stand on more than one leg.
- A single confirming check is not validation. Two states disagreed here; one state would have misled us.
How we did this, and the caveats
| What | Detail |
|---|---|
| Sort comparison | Three 100-record samples of category=dialysis_center, has_website=true, identical except sort |
| Generalization test | Two 100-record samples of category=dollar_store, identical except sort |
| State-scoped check | Direct brand queries for DaVita and Fresenius in Texas and California |
| Concentration measure | Unique website domains per 100 sampled records, classified identically throughout |
| External figures | DaVita and Fresenius US center counts and duopoly share from industry market reporting |
Caveats worth stating plainly: we tested three sort orders and two categories, which is enough to show the effect is real and not category-specific, but not enough to characterise its size across all categories. We cannot determine from this data which sort order, if any, is least biased — we can only show they disagree. The California/Texas divergence has more than one possible explanation, including genuine regional differences in market share, and we do not claim to distinguish them. The ~80% duopoly figure is externally sourced and not measured here. And as with every study on this dataset: it's live-growing, so a re-pull may shift all of these numbers.
If you want to run your own analysis on this data — including reproducing this sort comparison — that's exactly what Crawlora's Google Maps business API is for.
Query 132M+ Google Maps business listings directly
Crawlora's Google Maps dataset turns business listings into structured, queryable JSON — filter and facet by category, country, state, county, city, rating, and website status. 2,000 free credits a month, no card.
Frequently asked questions
How concentrated is the US dialysis industry?
Highly. A 100-record sample of category=dialysis_center resolves to just 18 unique website domains, the second-most concentrated category in this series behind dollar stores. DaVita and Fresenius Medical Care each operate roughly 2,600-2,800 US centers, and industry reporting puts the pair at about 80% of US facilities and 90% of industry revenue.
How much does sort order change a measured market share?
Dramatically. Holding dataset, category and sample size constant and changing only the sort parameter, the combined DaVita + Fresenius share measured 71.0% sorted by recency, 57.0% by rating, and 36.0% by review count — a 35-point spread. The count of unique domains also moved, from 18 to 35 to 55, so even the apparent level of fragmentation depends on sort.
Can sort order change which company looks dominant?
Yes, completely. Sorted by recency, Fresenius leads DaVita 54.0% to 17.0%. Sorted by rating, DaVita leads Fresenius 45.0% to 12.0%. Same category, same 100-record method, opposite conclusions. Reality is neither — the two companies operate roughly the same number of US centers.
Is the sort-order effect specific to dialysis?
No. Re-running dollar stores with only the sort changed moved Dollar General from 50.0% to 71.0% and halved Dollar Tree from 28.0% to 14.0%. It is a property of sampling this data rather than of any one industry. Aggregate concentration held up better than individual brand shares: the three-company total stayed high under both sorts.
Do state-scoped direct queries avoid the bias?
Not reliably, and we corrected our own conclusion on this. Texas returned DaVita 302 versus Fresenius 287, a 1.05x ratio matching reported near-parity, which suggested state-scoping solved the problem. California returned DaVita 410 versus Fresenius 127, a 3.23x skew. One confirming state was not validation. The most reliable anchor remains externally reported company data.