Connected-Task Cities Win

A new Journal of Regional Science paper (ungated here) has a fascinating thesis: what makes US cities big and growing lately is not computers, education, creativity, or socializing. Instead it is task connectivity.

Authors Kok and Ter Weel have data on 140K workers in the 168 biggest US cities. Each worker has one of 326 jobs, and each job has weights for 41 different kinds of tasks (listed in table 2). From this they create a measure of what fraction of time workers of each city spend on each task.

They then look at correlations between tasks of these city times. Two tasks that are highly correlated across cities, so that when a city does one task more it usually also does the other task more, are said to be “connected.” It is presumably useful to co-locate connected tasks. If, for a focal task, one adds up all the correlations between that focal task and all the other tasks, one gets a “task connectivity” for that focal task. “Info input” and “work output” type tasks are less connected, and have declined over time, while “mental process” and “interact with others” type tasks are more connected and have increased.

Averaging the connectivity of tasks done in a city, one gets the task connectivity of that city. Kok and Ter Weel find:

Cities with a relatively highly connected task structure seem to be larger, less specialized, and more skilled than cities with lower levels task connectivity. These cities also seem to employ workers for which social skills are relatively more important.

The correlation with city size is pretty strong:


Looking at employment growth of cities from 1990 to 2009, Kok and Ter Weel find that cities with less task connectivity grew less. Other bad signs for city growth are being big, having high rent, being specialized (like Hollywood and silicon valley), being in the Midwest and not in the West, and being cold in July. After controlling for these features, however, these other features were not growth signs: worker education, computer use, use of social skills, doing routine tasks, and local workers well matched to local jobs.

This paints a plausible picture, but one quite different than we usually see. If you want to be a big growing city, forget all that stuff you usually hear about recruiting educated “creative” workers, getting into computers and automation, promoting social interactions, or specializing in a particular industry. Instead have a nice climate, try to attract industries and jobs that do connected tasks, and get your rents down by increasing your building supply.

This also implies that which cities will win is pretty predictable. If the real estate market hasn’t yet recognized this, then do the calc, and invest in the good cities, and drop the bad ones.

Added noon: A similar result is found at the national level. HT Michael Hendrix.

GD Star Rating
Tagged as: ,
Trackback URL:
  • Stephen Diamond

    By itself, the correlation seems useless. Is it market failure when cities show low connectivity? If so, what causes it? If not, leave well enough alone.

  • brendan_r

    A table on page 34 shows some of the fastest and slowest growing MSA’s.

    Fastest in absolute terms: Phoenix, Atlanta, Houston, DC and Vegas.

    Slowest in absolute terms: Detroit, New Orleans, San Jose, Dayton-Springfield, Newark.

    (fastest and slowest in % terms are small cities I know nothing about)

    Maybe I’ve been reading too much Steve Sailer, but I see simpler patterns than task connectivity.

    -Hispanic immigration surged in 3 of the 5 fastest growers; 3 out of 4 excluding DC. (Cheap rents + proximity to the border.)

    -Chaos has been chasing the productive from Detroit, New Orleans and Newark.

    -Detroit, Newark and Dayton each depended heavily on industry that’s proved increasingly easy to outsource since 1990.

    “This also implies that which cities will win is pretty predictable.”

    Sure, but white flight is reasonably predictable. And immigrant flows tend to cluster spatially in predictable ways, too. Surely some real-estate pros made money on Detroit’s collapse and Phoenix’s boom, but do you really think they based that on their respective degree of Task Connectivities?

    “Instead have a nice climate, try to attract industries and jobs that do connected tasks, and get your rents down by increasing your building supply.”

    What about: attract the kinds of people likely to find work, and likely to create an atmosphere conducive to business? The authors control for education, but that can’t discriminate the typical Mexican immigrant from the typical Detroit citizen.

    My point, in this rambling mess, is it’s crazy not to control, at all, for racial demographics. Which the authors don’t. So how do we know ‘task connectivity’ isn’t some weird proxy for underlying demographic shifts, since demographics correlate will all sorts of stuff.

    • stevesailer

      Thanks for the examples.

      To make a meta-point: I’ve been reading Sister Y recently and doing some thinking about what differentiates patterns of thought among people who lean more toward Less Wrong and people who lean more toward HBD.

      There is a tremendous amount of overlap, of course, but I think one distinguishing marker of style of thought is whether you start with abstractions (e.g., “task connectivity”) and eventually get around to giving examples (or, as in this posting, not giving any positive examples, just two negative examples: not Silicon Valley and not Hollywood) versus a tendency to start with examples and move toward abstractions.

      • brendan_r

        Steve, ask your readers their SAT vs. GPA split; the SAT/GPA ratio. I suspect the LessWrongers have a higher ratio than is normal, the HBD’ers even higher (both in contrast to the 550/550, 4.3 split common among female libarts majors). HBD’ers are a bit dumber than the LWers on average, but also less likely to have paid enough attention in school (or maybe I’m projecting?) to unlearn how to notice some simple truths.

        Again I may be projecting, but I get the sense many HBD’ers (LWers too) are interested in Tim Ferriss style self improvement, which I think is popular among smart slackers (LWers are not slackers) who, in their mid 20’s as the testosterone dies down, get really interested in ideas for the first time, and “self-improvement” really means: time to play some catch-up and take things other than whey protein and deadlifts intellectually seriously. Having paid little attention for 25 years, there’s little risk of falling prey to Keep It Complicated Stupid.

        But again, the commonalities dwarf the differences; the differences are starkest among the less bright members (like myself) who sometimes fall prey to hammer-nail syndrome, trying to whack everything w/ either Daniel Kahneman or Arthur Jensen. Anyway, I’m grateful all you guys are around; Robin and Sailer and Eliezer and Gwern all compliment beautifully.

      • stevesailer


      • stevesailer

        I suspect there’s a higher IQ barrier to entry to the Less Wrong-style community, maybe 130 or so.

      • Stephen Diamond

        What’s HBD?

      • brendan_r

        Human biodiversity; sexual, racial, within race, etc.

    • RobinHanson

      “Hispanic immigration surged in 3 of the 5 fastest growers; 3 out of 4 excluding DC. (Cheap rents + proximity to the border.” The analysis already includes rents and geographic locations, and I mentioned those in the post. The task connection effect is in addition to those effects.

      • brendan_r

        OK, fair enough, muddled thinking on my part, let me clarify.

        “We have defined four regions: the North-east, the Midwest, the South and the West.”

        It’s conventional to segment America this way, but if we’re trying to explain growth in city size since 1990- a period in which the US hispanic population more than doubled- rather than lumping Arizona and CA with Wyoming and Montana, wouldn’t it make sense to make the Southwest its own category?

        Since there’s a fixed amount of variation to explain, a specification including SW as a factor would probably reduce the sig of most other factors. Maybe including task connectivity.

        Same w/ black share of the population: if black share, which was omitted, explains what’s up with Detroit, New Orleans and Newark, plus some other cities, then including this factor reduces the sig of the others.

        Is it really reasonable to pay this little attention to demographics?

      • RobinHanson

        When one makes dummy variables for regions, usually one leaves out one region as the reference region, where the relative parameter is implicitly zero. In this case the SouthWest looks like that left out region.

      • brendan_r

        No, I don’t think so in this case. They don’t mention it. And anyway, no one segments the country that way; anytime SW is included then SE is too, and S is eliminated.

  • Pingback: Assorted links

  • Pingback: Assorted links | Homines Economici