Lecture 11 - Issues involving spatial data Flashcards
Spatial autocorrelation
- main issue in spatial analysis
- statistically prove that things are related in space (clustering / dispersion)
- be careful if finding two things are correlated (move in the same or opposite directions) but aren’t actually variables that are related (example: people drowning and divorces in Kentucky)
Modifiable Areal Unit Problem
- same data with different results when aggregated differently
- affects correlation & regression
- applies where data aggregated at different areal units
- in NZ - territorial authorities, areal units, meshblocks
- typically larger areas better model, smaller areas lower model performance
MAUP - Scale effects
Analytical differences depending on size of units used (correlations are more pronounced for bigger units)
MAUP - Zonation effects
Major differences depending on how study are is divided up, even at the same scale
Can result in change to strength and nature of relationship.
Gerrymandering
- example of MAUP
- political manipulation of map boundaries so that citizens divided up into districts limiting ability to represent themselves via voting
Ecological Fallacy
-inferences about the nature of individuals deduced from inferences about the group those individuals belong to
- might observe a strong relationship between income and crime at the areal unit level, with lower-income associated with higher crime rates
- our (not correct) conclusion?
1. lower-income persons more likely to commit a crime
2. lower-income areas associated with higher crime
3. higher-income areas tend to experience lower crime rates
-one variable at individual level and one at group level
Ecological Correlation Problem
-correlation between 2 variables at the group level, in contrast to correlations between 2 variables at the individual level
-example: correlation between per capita alcohol consumption and death rates of males 55-64
and
number of drinks per day compared against mortality rates per 1,000 people
-people who drank modestly had lower mortality rates than those who didn’t drink at all, but among higher levels of individual consumption there was a striking linear increase in mortality
Ecological Fallacy Solutions
- limit inferences from ecological studies to determinates of aggregate population health
- if ecological study design is used to investigate determinates of individual health:
- choose homogeneous (small) units
- control for determinates of background risk
- be aware of limitations of causal inference
Nonuniformity of space/edge effect problems
- space is not uniform
- gaps and clusters no unexpected but arise resulting from the non-uniformity of urban space (might have higher crime just because there’s more population there, low crime because its a water body)
- artificial boundaries are imposed on a study area (often just) to keep it manageable
- something might be explained by something happening just outside the study area (people move across artificial boundaries)
Spatial Questions
- What I need to find out?
- Who’s my audience?
- Define hypothesis to be tested
- What do I already know about the spatial patterning of data that will help in exploring the problem?
- Where is the phenomenon coming from? (regression)
- What local initiatives have been targeted in past to area of interest? (things that could be the cause of changes over time)
- What spatial data do I need?
- What are the limits of data used in spatial analysis? (accuracy)