Summary of a paper. Estiri, H. and Krause, A. (2016). A Cohort Location Model of household sorting in US metropolitan regions. Urban Studies (OnlineFirst), pp. 1-20. (original). The paper is © its rights holder and is not reproduced here: this is a short account of it in our own words, standing in for it in the reading of Berkeley Stat 243.
4. Reading: A Cohort Location Model of household sorting in US metropolitan regions¶
What this covers¶
A household-sorting model for US metropolitan regions: whether the age of a household's householder predicts how far from the city centre that household is likely to live, across the 50 largest US metropolitan areas.
The question¶
Research on residential mobility had long used a household's position in the lifecycle (newly formed, child-rearing, empty-nest, and so on) as a predictor of when and why households move, but that account was increasingly criticised as too rigid: it forces households into discrete stages and does not fit the more varied trajectories of modern households. A companion paper by the same first author had proposed a "phasic" model that grouped households into three lifecycle phases and showed each phase settles at a different distance from the city centre. This paper asks whether that spatial sorting effect survives when the artificial phase boundaries are removed and a household's age is instead treated as a continuous variable — and, if so, what the shape of the sorting pattern actually looks like across the country's largest metropolitan regions.
The approach¶
The authors build a Cohort Location Model (CLM) from two assumptions. First, a household's housing consumption grows over its housing career, approximated using the age of the householder as a proxy for that career stage. Second, in the prevailing US pattern of low-density suburban development, the overall bundle of housing services on offer (space, structure, access to schools and amenities) increases with distance from the city centre or centres. Combining the two predicts a systematic age gradient in residential location: younger households, with lower housing consumption, cluster near the centre, while older households are drawn toward the urban fringe where more housing services are on offer — with middle-aged households expected to be the hardest group to tell apart spatially, since their housing consumption is changing most slowly.
To test this, they combine 2010 Census block-level population data with block geography for the 50 largest US Core Based Statistical Areas (a combined 2010 population of over 166 million, more than half the US total). For each metropolitan area they located the city centre and any subcentres, computed the great-circle distance from every census block to its nearest one, and standardised that distance to a 0–1 scale (dropping blocks more than 60 miles out as effectively rural). Ages were grouped into eight 10-year cohorts, the finest grouping the Census data allow. For each cohort at each distance band they computed a location quotient — the share of that cohort's households found at that distance, divided by the share of all households found there — so that an age group that is simply larger or smaller overall does not distort the comparison. Location quotients were summarised both with a LOESS-smoothed trend and a linear regression trend, each with confidence intervals, pooling across all 50 metropolitan areas.
What it found¶
The predicted gradient appears clearly. As householder age rises, the trend line's slope moves from negative (over-represented near the centre) to positive (over-represented toward the fringe), with the sign change occurring around age 35 — matching the threshold the earlier phasic model had used to separate its first lifecycle phase. The 15-to-24 cohort has by far the highest location quotient at the centre of any group, and is statistically distinguishable there from every other cohort, including the next-youngest (25-to-34). At the metropolitan fringe the order is close to a mirror image, with the 75-to-84 cohort showing the highest location quotient, though the very oldest group (85 and over) is, perhaps surprisingly, somewhat more centrally located than the cohorts just younger than it, while still being less central than the two youngest cohorts.
Away from the two extremes, the picture is less sharp. The four cohorts spanning ages 35 to 64 account for most of the suburban population, but their location-quotient trend lines overlap enough, once confidence intervals are taken into account, that the model cannot statistically tell them apart from one another in the suburbs — even though collectively they clearly dominate that part of the metropolitan region. The authors read this as consistent with their model, which predicts housing consumption (and therefore location) changes slowly across these middle years. Overall they identify four statistically distinct clusters of cohorts near the centre and five near the fringe.
Limits and context¶
The authors are explicit that census data cannot observe why an older household relocates — in particular "downsizing," a residential adjustment they build into their account of the oldest cohorts, which they note is not a well-studied phenomenon and is likely driven by multiple factors the Census does not record. They also distinguish the generality of their two assumptions: housing consumption rising over a household's life span is argued to hold globally, but the second assumption — that housing services increase with distance from the centre — reflects a specifically American pattern of suburban development, and the model's geometry would need to be adjusted for metropolitan regions elsewhere. The cohort resolution is limited to 10-year age bands by the granularity of the Census data used, and blocks more than 60 miles from a centre were excluded as likely rural rather than genuinely metropolitan. The full analysis code and a cleaned version of the underlying dataset were released publicly so the results could be reproduced.
Sources¶
Read from the publisher PDF: sources/berkeley-stat243/stat243-fall-2021/ps/clm.pdf.
Citation¶
Estiri, H. and Krause, A. (2016). "A Cohort Location Model of household sorting in US metropolitan regions." Urban Studies (OnlineFirst), pp. 1–20. The paper's own first page gives only the page range 1–20 and a 2016 copyright date; no volume or issue number is printed on it, consistent with an OnlineFirst publication ahead of print assignment. DOI: https://doi.org/10.1177/0042098016668783
← 3. Regular Expressions and Testing · Contents · 5. Code Review and Homework Habits →