A steady city average can hide neighborhoods moving in opposite directions

Two routes to a changing average

In Spatial Selection and the Multiscale Dynamics of Urban Change, Jordan T. Kemp, Laura Fürsich, Elias Tannira and Luís M. A. Bettencourt examine population and income changes in the Illinois portion of the Chicago metropolitan area between 2014 and 2019. Their arXiv preprint presents an accounting framework, rather than an experiment or a causal model of residential moves.

The central tool is the Price equation, widely used to describe selection in evolutionary biology. Here, the researchers apply it to geographic units, not biological traits passed between generations.

Imagine a community containing two neighborhoods. If the wealthier neighborhood gains population while the poorer one loses residents, the community’s average income can increase without either neighborhood’s own average income changing. The researchers call that a positive selection effect: higher initial income is associated with greater relative population growth.

Negative selection describes the reverse association—poorer areas gaining population, richer areas losing it, or both. Neither label is a judgment about whether the change benefits residents.

Accounting from blocks to the metropolis

The researchers use American Community Survey household and income-distribution estimates, starting with census block groups. They estimate population from household counts and average household size, and approximate incomes using the midpoints of reported income ranges. The open-ended highest income category receives an assumed value of $400,000.

They then assemble a geographic hierarchy: block groups within census tracts, tracts within communities, communities within counties, and counties within the metropolitan study area.

At each level, the equation separates changes within smaller units from changes in those units’ population shares. Because the calculation can be repeated down the hierarchy, it tracks how local contributions reach larger-scale totals.

An important technical distinction limits what “exact” means here. The main decomposition follows a population-weighted average of logarithmic incomes. That is not the same quantity as taking the logarithm of an ordinary average income. The identity is exact for the chosen measure; it does not make different definitions of aggregate income growth interchangeable.

Strong local change, small net contribution

The metropolitan figures look comparatively quiet. The paper reports annual population growth of roughly 0.025%, alongside nominal average-income growth of 3.42%. The latter is not adjusted for inflation.

Neighborhoods tell a less uniform story. Population and income growth vary substantially across nearby areas, and unusually large changes occur more often than a bell-shaped distribution would suggest. Selection effects are also concentrated: about 20% of tracts account for 59% of the total selection magnitude measured at that level.

The direction of sorting can reverse with geographic scale. In some near-north and northwest communities, population shifts favor higher-income tracts but lower-income block groups within those tracts. A broad view could therefore suggest movement toward affluent neighborhoods while missing a different pattern inside them.

The authors interpret combinations of income change, population change and selection as signatures consistent with processes such as income concentration or concentrated poverty. These are hypotheses about neighborhood transformation, not direct observations of individual households’ decisions.

Despite substantial local effects, positive and negative contributions largely offset when aggregated. The paper reports a net selection contribution equivalent to about 2.1% of metropolitan income growth under its decomposition. That is a relative contribution to the growth measure—not an additional 2.1 percentage points of annual growth.

What the accounting cannot reveal

The study does not follow residents through time. It therefore cannot directly distinguish voluntary moves from displacement, identify individual preferences, or establish which mechanisms caused population shares to change. Likewise, income change within a block group need not mean the same residents became richer: turnover can occur below the smallest geographic unit observed.

The analysis also depends on survey estimates, approximate incomes and geographic assignments. Tracts are assigned to communities by their centroids, and the study covers only the Illinois side of the metropolitan area. Its pre-pandemic period avoids the census-boundary changes discussed by the authors, but does not establish whether the same patterns persist afterward.

The useful result is a sharper accounting question: how much of a changing regional income measure reflects changes within places, and how much reflects places gaining or losing population share? A small net effect does not imply little neighborhood change. It may mean substantial, opposing changes have become almost invisible in the total.