Neighborhood Design and the Accessibility of the Elderly: An Empirical Analysis in Northern California

ABSTRACT

The low mobility of seniors may be due in part to a history of auto-oriented transportation and land use policy decisions. More recently, land use policies that make it possible to drive less show promise of effectiveness for the population as a whole. However, little attention has been paid to the implications of such policies for older people. Using data collected from Northern California in 2003, this study explores the ability of neighborhood design to preserve accessibility for the elderly by enabling a shift from driving to transit and walking, controlling for confounding factors. The results show that overall, older people drive less and use alternative modes more often than younger people. After controlling for attitudes and socio-demographics, neighborhood design has limited effects on driving and transit use, but enhancing accessibility tends to be a promising strategy for promoting walking trips. This enhanced accessibility has a much larger effect on the elderly than on the younger. Therefore, neighborhood design seems to be an important aspect of sustaining the accessibility of older people.

Key Words:

• aging
• built environment
• elderly
• land use
• self-selection
• seniors
• travel behavior

ACKNOWLEDGMENTS

The data collection was funded by the UC Davis-Caltrans Air Quality Project, the University of California Transportation Center, and the Robert Wood Johnson Foundation. Thanks to Ted Buehler, Gustavo Collantes, and Sam Shelton for their work on the implementation of the survey. The analysis was supported by the Small Urban & Rural Transit Center, North Dakota State University. Comments by three anonymous reviewers improved this paper.

Notes

Notes: SR = Santa Rosa, MD = Modesto, SC = Sacramento, HH = household.

Note: The numbers in parentheses are the pattern matrix loadings for the obliquely rotated factors, obtained using principal components analysis.

Source: Handy et al. (2004).

Notes. ^a We also used two-way analysis of variance (ANOVA) to identify the main effects and interaction effects of age group and neighborhood type. However, our data are not balanced in terms of these two variables. For an unbalanced design, a two-way ANOVA is most appropriately represented as a linear regression with a constant term, two dummy variables (one for age and one for neighborhood type), and their interaction (product). However, the interpretation of such a model is somewhat tedious. For example, finding the age variable to be significant does not necessarily mean a significant main effect for age, since the other variable and the interaction term are simultaneously being controlled for; rather, it simply shows that the mean when age = 1 and neighborhood type = 0 is significantly different from the mean when both variables are 0. To more straightforwardly display and analyze the means by each group of interest, therefore, we chose here to present t-tests instead.

^b S = suburban neighborhood, T = Traditional neighborhood, Y = the younger, E = the elderly. All interaction means are presented to allow for informal comparisons, but statistical tests with respect to the interactions are performed only on the difference between the elderly in suburban and traditional neighborhoods.

^c For full sample; sizes will differ somewhat by variable due to missing data on that variable.

^d For bolded entries, the difference between group means is significant at the 0.05 level.

^e For italicized entries, the difference between group means is significant at the 0.1 level.

Notes. Responses fell on a four-point scale, with 1 being least important (preferences) or true (perceptions) and 4 being most important/true.

^a S = suburban neighborhood, T = Traditional neighborhood, Y = the younger, E = the elderly.

^b For bolded entries, the difference between group means is significant at the 0.05 level.

^c For italicized entries, the difference between group means is significant at the 0.1 level.

^a The weekly vehicle-miles driven variable was log-transformed to bring its distribution closer to normality. One mile was added to each total beforehand to prevent taking the log of zero, which is negative infinity.

^b Responses for neighborhood preferences fell on a four-point scale, with 1 being least important and 4 being most important.

^c The marginal effects for vehicle miles driven (VMD) were calculated for an individual whose characteristics are taken at the sample mean.

^a As presented in Table 3, transit frequency for each of the six purposes was measured on a six-point ordinal scale. The frequency used here is a summation of the scales for the six purposes, which ranges from 6 to 36.

^b Responses fell on a four-point scale, with 1 being least important (neighborhood preferences) or true (neighborhood perceptions) and 4 being most important/true.

^c For a linear model without transformed variables, the marginal effects are the coefficients.

^a Responses fell on a four-point scale, with 1 being least important (neighborhood preferences) or true (neighborhood perceptions) and 4 being most important/true.

^b Marginal effects are averaged over individuals in the sample.

^a Responses fell on a four-point scale, with 1 being least true and 4 being most true.

^b Marginal effects are averaged over individuals in the sample.