We Found a Richmond Pricing Pattern in June. A 3x Bigger Sample Couldn't Confirm Half of It.

Last month we published a piece arguing that price-per-square-foot mostly measures your ZIP code and your floor plan, not your deal — built on 176 June sales. Two findings drove it: the same square foot cost 2.4x more in Richmond's priciest ZIP than its cheapest, and bigger homes sold for meaningfully less per foot, even inside the same neighborhood. We said the size effect was 'a real structural feature of how homes are priced.'

This month we pulled a fresh scrape — 558 single-family sales, three times the sample. The location finding held up and got sharper. The size finding effectively vanished. Before you read that as 'the pattern was wrong,' read the next section: the honest answer is we can't tell, because this month's data collection quietly changed the question we were asking.

What held up: the ZIP effect, and it got bigger

Across 558 July sales, median $/sqft is $220. Sorted by ZIP (33 codes, restricted to n≥4 for this comparison), the spread runs from $493/sqft in 23221 — the Museum District, same ZIP that topped June's list — down to $161/sqft in 23234 (North Chesterfield). That's a 3.06x spread, up from June's 2.44x.

That the Museum District tops the list in two independently-drawn, differently-filtered scrapes a month apart is the strongest evidence in this piece: location-driven $/sqft is not a fluke of a small June sample. It reproduces at 3x the sample size, with a different search filter, and the spread didn't narrow — it widened.

What we can't confirm: the size discount

June's second finding was sharper and more surprising: bigger homes sold for less per square foot, and the discount survived inside the same sub-market (six of seven areas showed a smaller-home premium, Pearson -0.21, Spearman -0.31, log-log elasticity 0.80). If that's a real structural feature of home pricing — fixed costs like the lot, kitchen, and systems spread thinner across more square footage — it should still show up in a bigger sample.

The declining staircase from June — $273 down to $185 — is gone. In the size bands July's search actually captured (1,750 sqft and up), $/sqft is close to flat, if anything drifting slightly upward at the largest sizes. The correlation flips sign: Pearson +0.12, Spearman +0.05 (both close to zero, and the reverse direction from June), and the log-log price-to-size elasticity is 1.10 — meaning, in this sample, an extra 1% of square footage comes with slightly more than 1% more price, not less.

Within-area premium: mixed, not clean

We re-ran June's within-area test — splitting each sub-market at its own median size and comparing smaller-half vs larger-half $/sqft — on July's data (areas with n≥10). The result is far less consistent than June's six-of-seven clean sweep.

Five of eight areas still show a small-home premium, but three now show the opposite — most notably Henrico, where larger homes sold for 12% more per foot than smaller ones. Compare that to June, where six of seven areas agreed on direction and the outlier (Glen Allen) was a small, -5% reversal. This is what a confounded re-test looks like: not a clean contradiction, not a clean confirmation — a genuinely mixed result once the sample composition changes underneath you.

Why we're publishing an inconclusive result

It would be easy to either (a) quietly drop the size angle and only publish the ZIP-spread win, or (b) write a punchy 'the pattern breaks!' headline off data that can't actually support that claim. Both would be a worse piece than this one. The useful finding here is procedural: two Zillow map-search scrapes taken a month apart used materially different filters (no sqft floor vs. 1,750+ sqft; 7-day sold window vs. 30-day; no lot-size floor vs. 1,000+ sqft), and that alone was enough to flip a correlation's sign in the aggregate. If you ever see a real estate 'trend' story built on comparing two scraped snapshots, the search parameters behind each pull matter as much as the numbers that come out.

How to actually use this

• •Trust the ZIP-level $/sqft spread. It has now reproduced across two independent scrapes with different filters and a 3x larger sample — that is meaningfully more robust than a single month of data.
• •Don't treat June's size-discount finding as debunked. It's unconfirmed pending a scrape with a consistent sqft floor across months — the honest status is 'pending retest,' not 'wrong.'
• •Be skeptical of any single-month real estate 'trend' piece (ours included) that compares two scraped datasets without checking whether the underlying search filters matched.
• •If you're shopping compact homes (under 1,750 sqft), this month's dataset has nothing to say about your segment — it was filtered out at the source.