Planning Examples

Smart Growth Policies and Automobile Dependence

Portland is well known for its smart growth policies. One of the policy objectives of smart growth is the reduction of automobile dependence. Jun (2008) examined

Reading SPSS Data and Running a Logistic Regression

Figure 13.26 Reading SPSS Data and Running a Logistic Regression: R Script and Outputs

whether or not smart growth policies are effective in reducing automobile dependence. This study modelled the effects of smart growth policies on commuters’ choice to drive alone with emphasis on four types of smart growth policies implemented in Portland: the urban growth boundary (UGB), public transit such as the MAX light rail system and bus senice, mixed land use, and transit-oriented development (TOD).

Figure 13.28 shows changes in density and mode share within Portland’s UGB over the 1990-2000 period. Preliminary results from inter-metropolitan comparisons and analysis of density and mode share changes within Portland’s UGB might support the smart growth proponents’ argument that smart growth policies contributed to a reduction in automobile use and encouraged a mode shift to alternatives to automobile travel. However, such a conclusion would be premature because there is no control over other factors that might affect commuter mode choice, such as

Running a Multinomial Logistic Regression

Figure 13.27 Running a Multinomial Logistic Regression: R Script and Outputs

socio-demographic, location, and transportation-related variables. All of these factors have been considered in the model of this study.

The effect of Portland’s smart growth strategies on the reduction of commuter auto dependence was then assessed using logistic regression. The model was based on the probability that a commuter chooses to drive alone. The outcome is a binary variable of “driving alone" and “any other travel mode except driving alone." The dependent variable was presented as Log (Pi/ [ 1 - Pi]), which is just the log odds. P, refers to the probability of driving alone for zone i, while (1 - Pj refers to the

Gross Housing Density and Commuter Mode Share Within the Urban Growth Boundary, 1990-2000

Figure 13.28 Gross Housing Density and Commuter Mode Share Within the Urban Growth Boundary, 1990-2000

Source: Jun (2008)

probability of choosing alternative modes to driving alone. The logit transformation of the observed probability was used as the dependent variable.

Socioeconomic, land use, location, and transportation-related variables were selected as independent variables. For residential analyses, logistic regression models were built with nine predictor variables for 1990 and 13 variables for 2000. For workplace analyses, models were estimated with seven predictor variables for 1990 and 11 variables for 2000.

Figures 13.29 and 13.30 show the results. Diversified land use in neighborhoods, more extensive provision of public transit service, and decreasing accessibility to freeway interchanges are associated with lower probability of driving alone, while making settlements compact via an urban growth boundary and transit-oriented development had no clear relationship to the probability of driving alone. The analyses also suggest that provision of public transit service and mixed land use implemented in residential zones (origins) were more effective in reducing automobile dependence than those implemented in places of work (destinations).

Mobility Disability and the Urban Built Environment

The second example explores the effects of the built environment in the pathway from impairment to disability (Clarke, Alishire, Bader, Morenoff, & House, 2008). This research examined the effects of built environment characteristics on mobility

Mode Choice by Residence Block Group (Origin)

Figure 13.29 Mode Choice by Residence Block Group (Origin)

Source: Jun (2008)

disability—i.e., using a wheelchair, walker, or cane—among adults aged 45 or more years (n = 1,195) according to their level of lower extremity physical impairment. The authors used multinomial logistic regression as their method.

The data source for this research was face-to-face interviews with a sample of 1,195 adults aged 45 or more years, and stratified into 343 neighborhood clusters in the city of Chicago. To measure built environment characteristics, trained survey raters collected observational data for each block surrounding the respondent’s residential address under the premise that “broken curbs and streets in disrepair are likely to be associated with more obstacles (e.g., rubble, uneven pavement) for pedestrians navigating along sidewalks and crossing streets” (Clarke et al., 2008, pp. 507—508).

A set of control variables for the neighborhood effect was added to this analysis to better isolate the effects of street quality on mobility disability. Neighborhood “social and physical disorder" and “residential security” were two constructs operationalized in the analysis. To control for individual characteristics, key socio-demographic and health factors were added to the models, including age, gender, marital status, race/ ethnicity, and socioeconomic position.

The outcome variable in this study was a nominal variable which includes three levels of outdoor mobility disability: no difficulty, some difficulty, severe difficulty walking two to three blocks. You can now see why multinomial logistic regression was used

Mode Choice by Workplace Tract (Destination)

Figure 13.30 Mode Choice by Workplace Tract (Destination)

Source: Jun (2008)

to examine the effects of individual and built environment characteristics on these three categories of outcomes.

Figures 13.31 and 13.32 report the results of the multinomial logistic regression analyses (no disability is the reference group). The tables present the logistic regression coefficients and odds ratios for the independent variables. Figure 13.31 relates to the outcome of “some difficulty walking two to three blocks" and Figure 13.32 relates to the outcome of “severe difficulty walking two to three blocks." The results show that older individuals with a greater number of health problems, and cigarette smoking, are more likely to experience problems of mobility. Being male and African American increase the odds of mobility disability compared with females and Caucasians. Look at the Figures 13.31 and 13.32 to try to find other independent variables that have a significant effect on each outcome.

In this study, authors were specifically interested in the effects of the built environment on mobility disability. As you can see from the resulting tables, the presence of any street in not-so-good (fair or very poor) condition increases the odds of mobility disability, all else being equal.


Logistic regression is the go-to model for categorical response data, being commonly used for a wide variety of planning applications. It is distinguishable from

linear regressions in that (1) the dependent variable is categorical in nature and (2) the model assumes a nonlinear relationship between the outcome and explanatory variables.

As with linear regression, researchers can use logistic regression in order to achieve one of two main objectives: explanation or prediction. In many studies, the focus is on the independent variables, with the goal being to identify the extent to which each one explains the outcome variable. In other studies, the focus is primarily on the outcome and how to predict one of the categories of that outcome variable.

In recent years, there is an increasing number of applied and methodological studies discussing the extension of (or alternative to) the logistic regression models. They include multilevel logistic regression (when the data has a nested structure), classification tree or random forest (when you are more interested in the classification process and/or outcome rather than probability or hypothesis testing), and support vector machines (when you have too many independent variables). As these advanced methods are beyond the scope of this book (except the multilevel modeling in Chapter 7 in Advanced Quantitative Research Methods for Urban Planners), see Agresti (2012); Hastie, Tibshirani, and Friedman (2009); and Izenman (2008) if you want more information.

Works Cited

Agresti, A. (2012). Categorical data analysis (3rd ed.). New York, NY: John Wiley & Sons.

Cheng, J., & Masser, I. (2003). Urban growth pattern modeling: A case study of Wuhan city, PR

China. Landscape and Urban Planning, 62(4), 199-217.

Clarke, P., Alishire, J., Bader, M., Morenoff, J., & House, J. (2008). Mobility disability and the urban built environment. American Journal of Epidemiology, 168(5). 506—513.

Cramer, J. (2002). The origins of logistic regression. Tinbergen Institute discussion Paper. Retrieved from

Ewing, R., Tian, G., Goates,J., Zhang, M., Greenwald, M. J., Joyce, A.....Greene, W. (2015).

Varying influences of the built environment on household travel in 15 diverse regions of the United States. Urban Studies, 52(13), 2330-2348.

Fang, S., Gertner, G., Sun, Z., & Anderson, A. (2005). The impact of interactions in spatial simulation of the dynamics of urban sprawl. Landscape and Urban Planning, 73(4), 294—306.

Field, A. (2009). Discovering statistics using SPSS (3rd ed.). London: Sage Publications.

Forsyth, A., Hearst, M., Oakes, J., & Schmitz, H. (2008). Design and destinations: Factors influencing walking and total physical activity. Urban Studies, 45, 1973-1996.

Frank, L., Schmid, T, Sallis, J,, Chapman, J., & Saelens, B. (2005). Linking objectively measured physical activity with objectively measured urban form. American Journal of Preventive Medicine, 28(2), 117-125.

Greene, W. H. (2012). Econometric analysis. New York, NY: Prentice Hall.

Hastie, T, Tibshirani, R., & Friedman, J. (2009). The elements of statistical learning: Data mining, inference, and prediction (2nd ed.). New York, NY: Springer' Series in Statistics.

Hosmer, D., Hosmer, T, Le Cessie, S., & Lemeshow, S. (1997). A comparison of goodness-of-fit tests for the logistic regression model. Statistics in Medicine, 16, 965-980.

Hosmer, D., & Lemeshow, S. (1989). Applied logistic regression. New York, NY: John Wiley & Sons. Hu, Z., & Lo, C. (2007). Modeling urban growth in Atlanta using logistic regression. Computers, Environment and Urban Systems, 31(6), 667-688.

Izenman, A. J. (2008). Modern multivariate statistical techniques: Regression, classification, and manifold learning. New York, NY: Springer.

Johnson, B.. & Parker, B. (2006). School trips: Effects of urban form and distance on travel mode. Journal of the American Planning Association, 72(3), 337-346.

Jun, M. (2008). Are Portland's smart growth policies related to reduced automobile dependence? Journal of Planning Education and Research, 28, 100—107.

Krizek, K., & Johnson, P. (2006). Proximity to trails and retail: Effects on urban cycling and walking. Journal of the American Planning Association, 72(1), 33—42.

Menard, S. (2010). Logistic regression, from introductory to advanced concepts and applications. Thousand Oaks, CA: Sage Publications.

O’Connell, A. (2006). Logistic regression models for ordinal response variables. London: Sage Publications.

Stoltzfus, J. (2011). Logistic regression: A brief primer. Academic Emergency Medicine, 18, 1099-1104.

Sutton, S.A. (2014). Are BIDs good for business? The impact of BIDs on neighborhood retailers in New York City. Journal of Planning Education and Research, 34(3), 309-324.

Takano, T., Nakamura, K., & Watanabe, W. (2002). Urban residential environments and senior citizens’ longevity in megacity areas: The importance of walkable green spaces. Journal of Epidemiol Community Health, 56, 913—918.

Theobald, D., & Hobbs, T. (1998). Forecasting rural land-use change: A comparison of regression-and spatial transition-based models. Geographical & Environmental Modeling, 2(1), 65-82.

< Prev   CONTENTS   Source   Next >