Jump to content


Mapping uncertainty

- - - - -

  • Please log in to reply
1 reply to this topic

Richard Lycan

Richard Lycan


  • New Member
  • Pip
  • 1 posts
  • United States

I'm making a presentation to a GIS meeting in September and plan to talk about the shift from the use of the US Decenial Census to the American Community Survey (ACS). I work with an applied demography program (the Oregon Population Research Center) that is the the Census Bureau affiliate in Oregon so I have good access to ACS data and organization.

One topic that I wanted to cover in my presentation is mapping of uncertainty. The data in the ACS are an approximately 3% sample repeated over five years, providing a rough equivalent sampling density to that from the long form questionnaire in prior censuses. Sampling error has always been a problem with census sample data, particularly for small areas such as block groups and census tracts, and it will continue to be an issue with the ACS data. However, the ACS is somewhat more explicit in presenting standard errors than was the case for the 1990 and 2000 censuses where one needed to do some calculations. For those who have worked with tract and block group data on topics such as poverty most are aware of the often large error variances in these data, particularly as one drills down into more detailed tables. However when when map these data we often fail to inform the viewer that the map represents one instance of many maps that could result from a different sample set.

My recollection is that I have seen papers presented at NACIS, or maybe elsewhere, that address the problem of representation of sampling variability in such maps. If anyone could steer me to some sources on this it would be helpful to me. I would like to review some of what has been done in this area in my paper. I'm not interested in squaring the circle at this point in time.

Richard Lycan
Population Research Center, Portland, Oregon

Dennis McClendon

Dennis McClendon

    Hall of Fame

  • Validated Member
  • PipPipPipPipPipPipPip
  • 1,076 posts
  • Gender:Male
  • Location:Chicago
  • Interests:map design, large-scale maps of cities
  • United States

I did a paper year before last on "uncertain boundaries," but that was primarily about ways to show when location is approximate.

In the GIS world, I think there's an unhealthy reliance on choropleth mapping, which implies a false precision and obscures multivariate phenomena. The example often used is that if Snow had used choropleth (rather than dot) mapping, we'd still be looking for the origin of that London cholera epidemic.

Perhaps these two companion maps are relevant examples:

Choropleth map of ethnicity
Dot map of ethnicity

Useful techniques to alert the reader of data limitations:
  • Legend wording. Make it say "approximately $10,000 to $15,000 per year" instead of "$10,000 to $14,999 per annum."
  • Dot-density instead of choropleth mapping, so that data outliers are not obscured, as discussed above.
  • "Retracing" into a bubble diagram or floating labels, to break the impression that "within this hard-edged polygon, median per capita income is precisely $12,560." A floating label "middle-class area" or overlapping bubbles labeled "comfortably middle-class neighborhood" and "working class neighborhood" prevents the reader from focusing on a false precision not underpinned by the data.

Dennis McClendon, Chicago CartoGraphics

0 user(s) are reading this topic

0 members, 0 guests, 0 anonymous users