Ecological fallacy

• common statistical problem that arises from aggregation
• several features are combined, they may create seemingly co-occurring phenomena
• these may be interpreted as causal relationships

Example: Treaty of Nöteborg and its impact

• supposedly the 1323 border influences demographic and economic differences between eastern and western Finland
• but in reality:
  • no set border for the most part
  • no direct causal connection
• seeing patterns that look vaguely like the 1323 border is an example of ecological fallacy

MAUP

• modifiable areal unit problem
• subset of ecological fallacy
• areal units can change arbitrarily
• rarely discussed in historical studies

Administrative units

• quantitative historical studies usually use data collected by administrative unit
• common in:
  • economic history/cliometrics
  • historical demography

• administrative units keep changing
  • diachronic change
  • local differences
• Eurostat’s NUTS areas are very different between countries
  • disgusting
  • leads to misleading choropleth maps and statistical analyses

dieghernan, Wikimedia commons

Dealing with spatial issues

• removing problematic areas
  • analyses are weaker
  • can introduce bias (e.g. quickly changing areas can be growing especially fast)
• ignoring problems completely

Fine-grained data

• fine-grained spatial analysis reduces the effects of spatial scale
• problems:
  • not enough ready-made data available
  • solutions to such problems require vast amounts of work

Adjusting areal units

• estimating and correcting for the impact of changing borders
  • used in Historical GIS projects (e.g. GBHGIS)
• imputation and interpolation
  • requires statistical knowledge
  • potentially very deceptive

Finnish and Russian population

• large Russian minority before 1917
• Orthodoxy used as a proxy for Russian nationality

Changing segregation

• town space changes
• so do spatial units
• St. Petersburg suburb shrinks over a long period
  • poll tax records change yearly

Fine-grained analysis

• population located down to plot level
• raster file representing population density
  • 50 by 50 meter resolution
  • continuous population density model
• almost ignores changes in suburb boundaries

Problems

• MAUP
• ecological fallacy
• lack of data

Solutions

• improved data collection
• interpolation techniques
• problems with spatial scale can never be fully solved, as data is inherently full of compromises