Breaking the laws of Health Geography

EPID 594
Spatial Epidemiology
University of Michigan School of Public Health

Jon Zelner
[email protected]


  • Final project timeline
  • Discussing the laws of health geography
  • Hands-on smoothing activity

Final Project

  • Final form of your project is up to you.

  • Please feel free to continue with your 592 topic if that is of interest!

  • Weekly project checkpoints are there to make sure you are able to make progress in the (limited) time we have.

  • Ultimately, I want you to do what is most useful and interesting for you.

The “First Law” of Geography

“Everything is related to everything else. But near things are more related than distant things.”
-Waldo Tobler, 1969

Why is the first law useful?

“[I]magine a world in which [TFL] is not true. In such a world, the full range of conditions could be encountered in every minute portion of the world. Every room, for example, might contain the full observed range of the Earth’s topographic variation, from the bottom of the Marianas Trench to the summit of Mount Everest[.]
-Goodchild, 2004 p. 301

“A world of white noise”

“White noise” has lots of variation with little autocorrelation.

More to the story?

Goodchild argues that:

  • Variation is more fundamental than autocorrelation.
  • Without spatial variation there is only perfect autocorrelation.
  • Maybe variation needs to be the real First Law?
  • Doesn’t discount role of correlation, just puts them in order.

Spatial autocorrelation + variation in the real world

Measles cases in Texarkana, Tex and Texarkana, Ark in 1970

What is smoothing?

Smoothing lets us separate signal from noise.

What are some examples of epidemiological signals?

  • Causal relationship between a risk factor and an outcome.
  • Trend in a time series
  • Disease hotspot in a spatial dataset.

What are some sources of noise we might encounter in epidemiological data?

  • Random variation, i.e. process noise.
  • Observation error.
  • Exposure misclassification.

A hands-on example


Next Time

Characterizing autocorrelation and variation is tricky

Using a measure like Moran’s I this is anti</span- correlated, but we can see qualitatively that there is more going on.