By decreasing property values, discouraging private investment, and inviting criminal activities, abandoned houses contribute to neighborhood decline. Since some neighborhoods have more abandoned houses than others, dealing with this problem is important from an equity standpoint. To improve public policy and planning efforts, this study seeks to better understand why neighborhoods differ in their probability that a house will be abandoned. It examines four related questions using data from Youngstown, Ohio and Columbus, Ohio.
First, the study considers what constructs are most salient to understanding abandonment, and how those constructs relate to the probability that a house will be abandoned. A factor analysis revealed that market conditions, gentrification, physical neglect, and socioeconomic conditions underlie abandonment. A multilevel regression model showed that three of the four constructs (market conditions, gentrification, and physical neglect) predict the probability of abandonment.
Second, it asks whether abandonment exhibits spatial dependence at the neighborhood level, and if so, whether the regression model can be improved by taking this relationship into account. A Moran’s I statistic indicated that abandonment clusters in both cities of interest. Adding a spatially lagged abandonment variable to the multilevel regression model showed that the level of abandonment in surrounding neighborhoods influences the probability of abandonment in a neighborhood of interest.
Third, it examines how other conditions in surrounding neighborhoods influence the probability. Adding a spatially lagged version of each factor to the regression model revealed that physical neglect in surrounding neighborhoods does not influence the probability of abandonment in a neighborhood of interest. However, the levels of market conditions and gentrification in surrounding neighborhoods do influence the probability.
Finally, it considers whether the variable effects generalize between the two cities of interest. Adding interaction terms to the multilevel regression model showed that the effects of neighborhood level abandonment and gentrification were the same for both cities, while the effects of market conditions and physical neglect were stronger for Columbus. All of the lagged variable effects generalize, meaning that the effect of surrounding neighborhood conditions on the probability of abandonment is the same for Youngstown as it is for Columbus.
Conclusion: Instead of thinking about housing abandonment in terms of a large number of variables, policy makers can conceptualize it as consisting of a smaller number of constructs. Factor scores, clusters of abandonment, and predicted probabilities can be mapped to suggest where to invest. The final regression model provides guidance for how to spend scarce recourses as well. It suggests that policy makers should enact strategies that decrease physical neglect in the neighborhood itself, and increase housing demand both in the neighborhood and in surrounding neighborhoods. Prevention of abandonment is likely the best strategy for dealing with the problem; and while more research is necessary to confirm, the suggestions proffered in this paper may serve to prevent abandonment. Furthermore, since most of the variables effects generalize between the two cities, it is possible that the results of this study generalize to other, similar cities as well.