Predicting Opportunities for Greening and Patterns of Vegetation on Private Urban Lands

Abstract

This paper examines predictors of vegetative cover on private lands in Baltimore, Maryland. Using high-resolution spatial data, we generated two measures: “possible stewardship,” which is the proportion of private land that does not have built structures on it and hence has the possibility of supporting vegetation, and “realized stewardship,” which is the proportion of possible stewardship land upon which vegetation is growing. These measures were calculated at the parcel level and averaged by US Census block group. Realized stewardship was further defined by proportion of tree canopy and grass. Expenditures on yard supplies and services, available by block group, were used to help understand where vegetation condition appears to be the result of current activity, past legacies, or abandonment. PRIZM™ market segmentation data were tested as categorical predictors of possible and realized stewardship and yard expenditures. PRIZM™ segmentations are hierarchically clustered into 5, 15, and 62 categories, which correspond to population density, social stratification (income and education), and lifestyle clusters, respectively. We found that PRIZM 15 best predicted variation in possible stewardship and PRIZM 62 best predicted variation in realized stewardship. These results were further analyzed by regressing each dependent variable against a set of continuous variables reflective of each of the three PRIZM groupings. Housing age, vacancy, and population density were found to be critical determinants of both stewardship metrics. A number of lifestyle factors, such as average family size, marriage rates, and percentage of single-family detached homes, were strongly related to realized stewardship. The percentage of African Americans by block group was positively related to realized stewardship but negatively related to yard expenditures.

Appendices

Appendices

Table A1 Description of PRIZM 15 and 62 classes

Appendix 2: multi-model selection

Burnham and Anderson’s inferential modeling approach relies on the information theoretic method pioneered by Akaike (1973, 1978), which contends that minimization of the Akaike Information Criterion (AIC) can help select the “order” of likelihood of a set of nested or non-nested models. Complexity comes at the tradeoff of parsimony, and therefore it is commonly accepted that a better model is one that achieves a balance between fit and number of parameters (Myung and others 2000, Wagenmakers and Farrell 2004). AIC penalizes model complexity and indicates which model best compromises complexity and fit. The formula for AIC is given by:

$$ {AIC = - 2\log \,L\,\,(M) + 2k} $$

(1)

where k is the number of parameters plus one and logL(M) is the maximized log likelihood for the model.

AIC scores can be compared for models with the same dependent variable. The model with the lowest AIC score out of a set of models is considered to be most likely to be correct. Although the order of AIC scores gives model rankings, this does not reveal how likely it is that a model with the lowest AIC score is the best model. Small differences in AIC scores can lead to a false sense of confidence that one model is superior to another (Wagenmakers and Farrell 2004). Akaike weights (Burnham and Anderson 2002) show the probability of a given model being the correct one out of a set of potential models and are given by the equation:

$$ {w_{i} \,(AIC) = {{e^{{-.5(\Delta_{i} (AIC))}}}\over{{\sum\limits_{k = 1}^K {e^{{-.5(\Delta_{k} AIC)}}}}}}} $$

(2)

where k = the number of models.