A market screen first filters markets by population, income and competitor counts, before individual trade areas within them are evaluated. Store revenue is determined at the trade-area level, and variation within one metro exceeds the gap between metro averages. As a result, the screen excludes strong sites along with weak ones.
Expansion planning uses two distinct stages, each based on different evidence. Stage one ranks candidate markets using a handful of metro-wide numbers. Stage two evaluates specific addresses within the markets that pass that initial screen. It predicts what a store will do, while stage one determines which addresses stage two will ever see.
The sequence has a consequence almost no one measures: once a market is eliminated in stage one, not a single site in it is evaluated. No forecast, store opening, or post-mortem follows, leaving the process with no evidence of whether eliminating it was the right decision.
Market screening runs first, and it runs on metro averages
Before a specific address is reviewed, the stage-one market screen ranks candidate metros. It compares each market using metro-wide measures: total population, population growth, median household income, the number of category competitors already open, driving distance to a distribution center, and a proxy for brand strength. The output is one row per market, a weighted score, and a cut line.
The logic behind this sequence is sound. Clustered stores share one distribution run and one field supervisor. Local marketing costs less per unit once several stores are in one media market. Brokers work by territory, too, so a requirement floated where the team already operates draws more space than one sent out cold.
Capacity, rather than evidence, usually determines the cut line. A team that can open twelve stores next year screens down to the fifteen or twenty markets it can work, since a longer list has no operational meaning. It's a resourcing decision, although by stage two it appears to be analysis. At that point, markets below the line are entirely excluded from the conversation.
Why a metro average cannot predict a trade area
A metro-level average can't predict an individual trade area: the two represent different geographies. Metro population, income, and competitor counts combine data from territory no single store serves. A store's revenue comes from one drive-time catchment within that territory, while catchments in the same metro vary far more than metro averages vary across metros.
This problem has a formal name and a well-documented history. Ecological inference covers drawing conclusions about individuals from statistics calculated for the groups they belong to, and W. S. Robinson showed in 1950 that a correlation measured across aggregates can differ in magnitude, or even reverse sign, from the same correlation measured among the individuals within those aggregates. Geographers had recognized the aggregation effect earlier, when Gehlke and Biehl reported in 1934 that grouping census tracts into larger units inflated the correlations calculated from them. Stan Openshaw later called this broader set of problems the modifiable areal unit problem: using the same underlying data can produce a different answer depending on the unit's size and where its boundaries are drawn.
When metros are ranked using their averages, treating the sites within them as correspondingly good or bad repeats that inference at a geographic scale. Almost nobody tests the decision to use the metro as the unit of analysis.
Published research on what predicts store sales reaches the same conclusion. A 2023 paper in the journal Land examining home-improvement store sales found that measures of the road network around a store predicted sales better than demographic or site-suitability measures, both in regression analysis and in a non-linear machine-learning model. Road network pattern belongs to a corner, not a metro, so no screen based on metro averages captures it.
One limitation needs to be stated plainly. Nobody has published the decomposition this argument would want to cite: the share of retail store performance variance at the market level and the share at the site level. The mechanism is well established, but the size of the gap is based on operator experience, not a measured statistic.
The practical difference is substantial. A single metro-wide income measure can describe one intersection with heavy daytime population and six category competitors within a ten-minute drive, as well as another with a fraction of that demand and no competitor. The gap grows when the catchment is drawn accurately, because a radius circle and a drive-time polygon diverge most across the suburban and small-metro road networks that define secondary markets. In a smaller market, a trade area built from the road network often extends farther and captures a larger share of the households within it than the equivalent catchment in a dense metro.
The store in a skipped market cleared a harder filter
A store in a market the team nearly skipped cleared a tougher screen than its peers, which is the main reason it outperforms them. In an approved market, brokers circulate the requirement and standard sites regularly reach the committee. In a market the screen removed, no one is sourcing at all, so a site appears only when circumstances force one.
Exceptions follow a recognizable pattern. A franchisee who already lives in the market puts forward a corner and won't be persuaded to drop it. A portfolio appears in a bankruptcy auction, and the boxes have to be accepted or rejected as one set. A landlord the team already works with has a vacancy that nobody was considering. In every instance, someone spends internal credibility reopening a decision that was already made.
That standard is a strict screen: no one makes the case for a site that's merely acceptable. As a result, acceptable sites in closed markets don't surface at all. The few that do are compelling enough to warrant the argument, so stores that eventually open in skipped markets sit at the upper end of what was available there. Comparing the average store in an approved market with the average store in a skipped market therefore compares a full distribution with a truncated one.
The result has two sources. Some of the effect comes from the sourcing process, not the markets themselves, so a team that decided to open additional small-market stores wouldn't recreate it. The remaining portion is economic and can be measured.
What changes in a smaller market's store economics
Smaller markets differ from flagship markets in three measurable respects: occupancy cost relative to sales, how much new competing space is being built, and the direction in which people are moving. None of these factors appears in a screen ranking metros by population and median income.
Occupancy cost is the clearest measure. The National Restaurant Association reported in September 2025 that limited-service restaurants in urban locations and city centers allocated a median of 6.0% of sales to occupancy in 2024, versus 5.0% in suburban locations and 3.2% in small communities and rural areas. A difference of 2.8 points of revenue separates a comfortable unit from a marginal one. The comparison has a limit, however: for full-service restaurants, the gap nearly disappears, with 6.0% in urban locations against 5.4% in small communities and rural areas. This finding applies to limited-service formats, not restaurants generally.
Next is competing supply. Retail construction completions totaled 4.7 million square feet in the first quarter of 2026, which CBRE described as the lowest since it began tracking the figure in 2005, compared with a quarterly peak above 25 million square feet in late 2015. CBRE's second-quarter report still described completions as historically low, with availability at 4.9%. New space is worth more in a market with four category competitors than in one with forty.
Where people are relocating is the third factor, and broad growth figures can obscure it. Census Bureau estimates released in March 2026 report metropolitan areas collectively lost 119,205 people through net domestic migration from July 2024 to June 2025. Their overall growth instead came from international migration and natural increase. Micropolitan areas gained 67,419 through net domestic migration over that period. A separate Census release in May 2026 found average growth of cities with 5,000 to 49,999 people was 0.7% from 2024 to 2025, versus 0.3% among cities with 250,000 or more.
The direction is more important than the level here. Metro areas still had stronger overall growth, at 0.6% versus 0.2% for micropolitan areas, because births and arrivals from abroad more than offset the domestic outflow. A market screen should focus on that domestic split, since it shows where households already in the country are choosing to live.
The tier label cannot predict the result by itself. For grocery-anchored centers, JLL's June 2026 retail outlook reports year-over-year rent growth of 4.3% in secondary and tertiary markets, versus 3.7% in primary markets. That's a meaningful difference, but not a wide one. The category also includes Denver and San Diego, showing how broadly the tiers are defined. Local supply and demand determine the outcome, not the label.
Where smaller markets genuinely underperform
The case for a smaller market has to account for where performance predictably falls short; a pitch focused only on upside wouldn't survive the first committee meeting. Four constraints matter most: a lower ceiling for absolute volume, less depth for a second or third unit, a thinner labor pool, and a harder exit if the store fails.
The volume ceiling drives most approval decisions. A store may exceed every forecast in a smaller market and still generate less total revenue than a mediocre unit in a dense market. If the committee orders candidates by projected sales rather than return relative to the capital and rent committed, the small-market site loses on the number presented in the room. That is a scoring issue, not a real estate issue.
Site quality and market depth are separate judgments, and treating them as one creates errors in both directions. A market may support exactly one profitable unit. In that case, the screen may correctly reject a four-store cluster while incorrectly rejecting one very good store. A second unit in a small market also takes sales from the first faster than it would in a dense one.
Labor and supervision expense is tangible as well. A thin hiring pool increases the chance that a strong site is held back by a general manager the team can't replace quickly; a store far from its nearest cluster also adds distribution mileage and a supervisor's travel day. Confidence in the forecast narrows, too, since fewer portfolio analogs mean the site analysis behind a small-market unit carries a wider band. That belongs in the packet, rather than being smoothed away.
The screen never corrects itself, because a cut market produces no data
The market screen never corrects itself because its two possible errors produce very different evidence. Approve a weak site, and it opens, underperforms, and appears in every portfolio review for the full lease term. Cut a strong market, and there's no store, no sales data, and nothing for anyone to review.
A process that responds only to visible errors moves in one direction over time. Each underperforming store tightens the screen somewhere, by raising the population floor, increasing the income threshold, or requiring a wider gap from the nearest competitor. Nothing prompts a loosening, since markets that would have justified it never produced a store to point to. Across planning cycles, the cut line rises on evidence arriving from one side only.
Consumer lending faced the same structural issue and developed a remedy. A credit model trained solely on approved applicants never observes how rejected applicants would have repaid, so it learns to repeat the judgment embedded in the original cutoff. Lenders call the correction reject inference, a set of techniques for estimating how declined applicants would have performed. In personnel selection, the underlying problem, range restriction, has been formally studied for decades. Retail expansion planning encounters the same issue at stage one but has no comparable practice.
The imbalance is clear in how teams discuss their models. Nearly every operator remembers a site that fell short of its target, while very few can identify a market they wrongly rejected. Comparing forecasts against actuals is routine by stage two but effectively impossible in stage one, so stage one keeps its assumptions in place far longer than it should.
How to screen markets without discarding sites
To screen markets without discarding sites, limit what the first stage can decide. The screen remains necessary because teams can't assess every address, and clustering has operating logic behind it. Instead of returning a binary in-or-out result, it creates a ranked queue, with the best trade area for each market attached.
Screen on available category demand, not on population. Focus on how much unmet demand for your category the market contains, rather than on its population. A metro of three hundred thousand with no competitor in the category can have more reachable demand than one of three million with fourteen. Market saturation distinguishes those markets, which population can't reveal.
Attach one real trade area to every market before you cut. In each candidate market, including those you expect to fail, map a drive-time catchment around the strongest corner. Rank markets by what their best site can do, not by the metro average. That was once impractical because an analyst needed a day for each market. It isn't anymore.
Keep the cut list, with reasons and a date. For every market removed, document why it was cut and what would need to change for it to return. Rents move, competitors close, and interchanges open. A cut list with those reasons remains a live document; without them, the decision can't be revisited.
Say plainly where the cut line came from. When the team has capacity for twelve stores, state that the market list extends to twelve. Capacity limits are valid and defensible. Presenting one as an analytical finding is what makes changing that line later so difficult.
Track the overrides, because they are the only feedback stage one gets. Record every site in a cut market that reaches approval. When overrides are frequent and those stores perform, the cut line is wrong, and the people making them see something the screen can't.
How GrowthFactor screens markets
GrowthFactor brings trade-area evidence into stage one rather than waiting for stage two. For each market, the Agent maps drive-time catchments around candidate corners, collects daytime population, competitor spacing, foot traffic and demographics for each one, then ranks markets by what the strongest reachable site within them can support, not by metro population and income.
Every score is organized into five lenses, with the source of each input visible. The operator sets the weights, so a brand whose shoppers will drive twenty minutes for a planned visit can assess its catchment differently from one that depends on passing traffic. Forecasts are delivered as ranges rather than single figures. Across customers, teams report about 80% fewer underperforming locations after the GrowthFactor workflow is in place (JAN2026 customer survey). They can also review roughly five times as many candidate sites as they could by hand, making it practical, not theoretical, to screen every market for its best corner.
It works with the tools a team already has in place. Most operators first use it to prioritize markets alongside their existing stack, then pass the markets that remain into the expansion plan and the sites they pursue.
Frequently Asked Questions about secondary market expansion
What is a secondary market in retail site selection?
A secondary market is a metro outside the largest tier of US markets, with tertiary and micropolitan markets below that. The tiers are conventions rather than official definitions, and operators draw the lines in different places. What matters for site selection is that the label describes a whole metro, while store revenue is set inside one drive-time trade area covering a small part of it.
Do stores in secondary markets actually outperform stores in primary markets?
Some do, for two reasons that run together. Smaller markets often carry a lower occupancy cost ratio and far less category competition inside a given drive time. Separately, a site in a market the screen removed only gets evaluated when someone pushes hard for it, so the small group of stores that come from those markets is selected for quality before anyone opens a spreadsheet.
Why do expansion teams screen markets before individual sites?
Because nobody can evaluate every address in the country, and the constraints behind market-level sequencing are real: distribution radius, field supervision, labor pools and broker coverage all work better in clusters. The screen itself is not the problem. The problem is that it decides in or out on metro averages, then hands stage two a list that can no longer be questioned.
What is the ecological fallacy, and how does it apply to site selection?
The ecological fallacy is drawing a conclusion about individuals from statistics computed on the groups they belong to. W. S. Robinson showed in 1950 that a correlation measured across aggregates can differ in size, and even in sign, from the same correlation measured across individuals. Ranking metros on metro averages and inferring that the sites inside them are good or bad is the same inference, made across geography.
How should a retailer decide when a smaller market is worth entering?
Judge the market on the best trade area available in it rather than on its average, and separate the site question from the depth question. One strong corner with no category competitor inside the drive time can justify a single unit even where the market will never support four. Decide on return against capital committed rather than absolute volume.