500 likes | 600 Views
Spatial Analysis & Vulnerability Studies START 2004 Advanced Institute IIASA, Laxenburg, Austria Colin Polsky May 12, 2004. Graduate School of Geography. International Geographical Union (IGU) Task Force on Vulnerability. Outline. What is spatially integrated social science?
E N D
Spatial Analysis & Vulnerability StudiesSTART 2004 Advanced InstituteIIASA, Laxenburg, AustriaColin PolskyMay 12, 2004 Graduate School of Geography
International Geographical Union (IGU) Task Force on Vulnerability
Outline • What is spatially integrated social science? A. Qualitative dimensions B. Quantitative dimensions i. univariate ii. multivariate • An example: Vulnerability to the Effects of Climate Change in the US Great Plains
Necessary and sufficient conditions to achieve objective of vulnerability studies: • Flexible knowledge base • Multiple, interacting stresses • Prospective & historical • Place-based: local in terms of global • Explores ways to increase adaptive capacity Source: Polsky et al., 2003
What variables cluster in geographic space? How do they cluster? Why do they cluster? Can you imagine any variables that are not clustered?
Criticisms of quantitative social science: • discovering global laws • overly reductionist • place can’t matter • too deductive, sure of assumptions • Localized quantitative analysis: • exploring local variations and global trends • holistic • place can matter • unabashedly inductive, questions assumptions
Spatial analysis (ESDA) is as valuable for hypothesis testing as for hypothesis suggesting…especially in data-sparse environments. ESDA helpsexplain why similar (or dissimilar) values cluster in geographic space: • Social interactions (neighborhood effects) • Spatial externalities • Locational invariance: situation where outcome changes when locations of ‘objects’ change Source: Anselin, 2004
Outline • What is spatially integrated social science? A. Qualitative dimensions B. Quantitative dimensions i. univariate ii. multivariate • An example: Vulnerability to the Effects of Climate Change in the US Great Plains
“Steps” for Exploratory Spatial Data Analysis (ESDA): • Explore global/local univariate spatial effects • Specify & estimate a-spatial (OLS) model • Evaluate OLS spatial diagnostics • Specify & estimate spatial model(s) • Compare & contrast results
Spatial autocorrelation: • Cov[yi,yj] 0, for neighboring i, j • or • “values depend on geographic location” • Is this a problem to be controlled & ignored • or • an opportunityto be modeled & explored?
The “many faces” of spatial autocorrelation: • map pattern, information content, spillover effect, nuisance, missing variable surrogate, diagnostic, … • Spatial regression/econometrics: • spatial autocorrelation reflects process through regression mis-specification
Spatial Weights Matrices & Spatially Lagged Variables Source: Munroe, 2004
What you know, and what you don’t know… What you know y = X + What you don’t know
OLS assumptions: • Var(ei) = 0 • no residual spatial/temporal autocorrelation • errors are normally distributed • no measurement error • linear in parameters • no perfect multicollinearity • E(ei) = 0
Ignoring residual spatial autocorrelation in regression may lead to: • Biased parameter estimates • Inefficient parameter estimates • Biased standard error estimates • Limited insight into process spatiality
Source: Kennedy (1998) bias versus inefficiency
y = X + W + y = Wy + X + y = X + i , i=0,1 y = Xii + i , i=0,1 Null hypothesis: no spatial effects, i.e., y = X + works just fine Alternative hypothesis: there are significant spatial effects Large-scale: • spatial heterogeneity Small-scale: • spatial dependence
Large-scale: • spatial heterogeneity– dissimilar values clustered discrete groups or regions, widely varying size of observation units Small-scale: • spatial dependence– similar values clustered “nuisance” = external to y~x relationship, e.g., one-time flood reduces crop yield, sampling error “substantive” = internal to y~x relationship, e.g., innovation diffusion, “bandwagon” effect
Which Alternative Hypothesis? observationally equivalent
Outline • What is spatially integrated social science? A. Qualitative dimensions B. Quantitative dimensions i. univariate ii. multivariate • An example: Vulnerability to the Effects of Climate Change in the US Great Plains
“Economic Scene: • A Study Says Global Warming May Help U.S. Agriculture” • 8 September 1994
Ricardian Climate Change Impacts Model Agricultural land value = f (climatic, edaphic, social, economic)
Climate Change Impacts: Agricultural Land Values Source: Mendelsohn, et al. (1994:768)
Great Plains wheat yields & seeded land abandoned: 1925-91 Source: Peterson & Cole, 1995:340
Random? Land Value, 1992 ddddddd
spatial lag/GHET model: y = Wy + X + i , i=0,1
Space, Time & Scale: Climate Change Impacts on Agriculture Source: Polsky, 2004