A spreadsheet contains the median rent for every neighborhood. Averaging those medians produces one convenient citywide number. The calculation is easy, but the result is not generally the median rent of the city's households. Each neighborhood median is a summary that has already discarded much of its underlying distribution.
Census guidance notes that derived estimates such as medians cannot simply be aggregated. This guide explains the issue with original fictional rent lists. These small lists are complete invented populations, so their arithmetic does not involve sampling uncertainty. Real survey work also requires attention to universes, weights, geography, and margins of error. The examples isolate the aggregation problem before those additional considerations enter.
Start with two small neighborhoods
Imagine fictional Oak has monthly rents of $900, $1,000, and $1,100. Its median is $1,000. Fictional Pine has rents of $1,200, $1,300, $1,400, $1,500, and $1,600. Its median is $1,400. The simple average of the two neighborhood medians is $1,200.
Combining the eight rents and ordering them gives $900, $1,000, $1,100, $1,200, $1,300, $1,400, $1,500, and $1,600. Under the ordinary median convention for this complete list, the middle two values are $1,200 and $1,300, giving a combined median of $1,250. The average of neighborhood medians misses the combined median by $50.
Nothing went wrong with either neighborhood median. The problem is the operation used to combine them. A median identifies a position within a distribution. Averaging two positional summaries does not generally identify the middle position of the combined observations.
Weighting the medians by population does not solve it
An analyst might try weighting each median by its number of rental observations. In the fictional example, that gives three times $1,000 plus five times $1,400, divided by eight. The result is $1,250, which happens to match the combined median in this particular arrangement.
That coincidence is not a general method. Change Pine's five rents to $1,050, $1,100, $1,400, $1,500, and $1,600. Pine's median remains $1,400, and Oak's median remains $1,000. The population weighted average of medians remains $1,250. But the combined middle values are now $1,100 and $1,100, so the combined median is $1,100.
The summaries and group sizes are unchanged, yet the citywide median changes. That is decisive evidence that medians plus group sizes do not contain enough information to recover the combined median in general. A weighting formula cannot restore observations that the summaries no longer reveal.
The distribution near the middle matters
A median depends on where the combined ordering crosses the middle of the population. Values within each neighborhood above and below its median affect that crossing. Knowing only the median tells us little about how tightly those values cluster or how they overlap with another neighborhood's rents.
In the revised Pine example, more values sit near Oak's upper end. That shifts the combined middle downward even though Pine's own median stays fixed. An average of neighborhood medians cannot detect this rearrangement because its inputs do not change. The lost information concerns the shape and overlap of the distributions.
This is why two cities could share the same list of neighborhood medians and neighborhood sizes while having different citywide medians. The summary table is not a compressed file from which every larger area statistic can be reconstructed. Different statistics preserve different information.
An average of medians can still be named honestly
There may be a reason to calculate the mean of neighborhood medians. For example, a researcher might want a descriptive measure that gives every neighborhood equal importance regardless of its number of rental units. That is a distinct indicator about neighborhood summaries. It should be labeled as such.
The label average neighborhood median rent is much closer to the calculation than city median rent. The methods should explain whether neighborhoods are equally weighted or weighted by a specified quantity. Readers should also understand that changing the neighborhood partition can change the indicator, even without changing any individual rent.
A fictional city split into four administrative neighborhoods may produce a different average of medians than the same city split into eight. That sensitivity does not automatically make the indicator useless, but it matters to its interpretation. A household based city median and an average of geographic summaries answer different questions.
Use the direct city estimate when it matches the question
If an official citywide median is available for the desired universe and period, it is usually the appropriate starting point for a citywide median question. Check that it covers the same rent concept, geography, and survey product as the analysis. Do not replace it with an average of smaller area medians simply because the neighborhood file is already open.
Suppose a fictional report has tract medians but wants a city median. The city may contain partial tracts or tracts extending beyond the boundary. Even before the median aggregation problem, the geographic pieces may not form the city exactly. A direct city table can avoid both the invalid median combination and some geography reconstruction problems.
If the direct estimate is unavailable, state that limitation. The absence of the desired statistic does not justify labeling a different calculation as equivalent. A report can show neighborhood medians individually or describe another carefully defined measure without inventing an exact city median.
Distribution tables can support different work
A table of counts across rent ranges preserves more distribution information than a single median. Combining compatible counts across nonoverlapping areas can help describe a broader distribution. However, grouped ranges still do not reveal every exact rent within a range. Any estimated median derived from grouped data needs an appropriate method and clear labeling.
Imagine that the middle of a combined fictional distribution falls inside a range from $1,000 to $1,499. The grouped counts identify the interval containing the middle but not necessarily the exact middle rent. Assigning $1,250 without explanation imposes an assumption about values within that interval.
A rigorous analysis should follow the source's methods or a defensible documented estimation approach. It should not imply that a midpoint is an observed median. More detailed input can improve what is possible, but the method still must match the level of information the data actually contain.
Never add overlapping geographies
Even for statistics that can be summed, overlapping areas create another problem. A city total and its neighborhoods should not be combined as if they were separate populations. Neither should two neighborhood systems that overlap. The same units can be counted twice or more.
For a fictional regional analysis, suppose one table includes the whole city and another includes its northern district. Adding their distributions gives extra weight to the north because its units already appear in the city total. A median calculated from that duplicated distribution would describe an artificial population.
Verify that component geographies are mutually exclusive and cover the intended area before any aggregation. Keep a list of included identifiers and boundary vintages. Geographic completeness and statistical aggregability are separate checks; passing one does not guarantee the other.
Matching rent concepts is essential
A neighborhood figure based on gross rent should not be combined casually with one based on advertised asking rent. Those measures can cover different costs, populations, and time frames. A common dollar sign does not make them compatible. The same warning applies to different bedroom categories or occupied unit universes.
Suppose fictional Oak's median describes all renter occupied units while Pine's describes only newly advertised two bedroom apartments. Averaging them produces a number with no clear common universe. Weighting cannot repair the mismatch because the inputs do not summarize the same concept.
Before worrying about the combination formula, create a short definition record for each source. Include the cost measure, eligible units, period, geography, and uncertainty. If those attributes differ materially, use separate comparisons or choose a consistent source rather than smoothing over the differences with one aggregate figure.
Keep uncertainty attached to the statistic you actually use
Real survey medians have margins of error. Averaging their margins does not produce a valid margin for an imagined city median. A derived indicator needs an uncertainty method appropriate to that indicator, and the direct city median has its own supplied uncertainty where available.
This is another reason to use the published city estimate when it fits the question. A custom shortcut creates both a questionable point estimate and a new uncertainty problem. More elaborate formatting cannot compensate for either. A chart with a precise looking city number may be less reliable than a clearly labeled set of neighborhood estimates.
If the publication reports an average of neighborhood medians for a legitimate purpose, identify it as a publisher calculation. Do not borrow the city median's name or margin of error. The title, arithmetic, and uncertainty statement should all refer to the same statistic.
A final editorial test
Ask a reader what they think the phrase city median rent means. Most will understand it as a middle value for the relevant citywide rental population, not an average of neighborhood middle values. If the calculation does not match that interpretation, change the method or change the label.
The fictional Oak and Pine lists show why this matters without any complicated mathematics. Identical neighborhood medians can coexist with different combined medians. Once that fact is visible, the safe habit follows naturally: preserve the distinction between describing neighborhoods and describing the combined housing population, and use data that support the particular summary the article promises.