1 Introduction

Globally, the population aged 65 or older is estimated to increase from 739 million in 2020 (9.4% of the total) to 1.8 billion (18.7% of the total) in 2060; the 85-and-over population will increase by more than four times over the same period (United Nations, Department of Economic and Social Affairs, Population Division, 2022). Particularly in postindustrial Western countries, the aging of the Baby Boom generation (born 1946–1964) promises a greater older population for decades to come. In the United States, the population aged 65 or older will increase dramatically from 56.1 million in 2020 (17% of the total) to 94.7 million (23% of the total) in 2060; the 85-and-over population will increase from 6.7 million in 2020 (2% of the total) to 19 million in 2060 (4.7% of the total) (Vespa et al., 2020).

As people get older, their travel needs and behaviors change due to their different daily activities, such as working status, health care activities, etc. A good place for older adults should have good accessibility for them to meet their travel needs and promote physical activity. It is widely known that physical activity is important to maintain health and has positive effects in the prevention and treatment of many chronic diseases and age-related disabilities (Chudyk et al., 2015). Designing age-friendly neighborhoods with destinations nearby encourages older adults to do more outdoor activities and be physically active (Winters et al., 2015). Walking and transit use are relatively easy ways for older adults to remain physically active (Kemperman & Timmermans, 2009). According to the Behavioral Risk Factor Surveillance System (BRFSS) 2019 survey (Centers for Disease Control and Prevention, 2020), 51.44% of respondents reported that walking was the physical activity or exercise they spent the most time doing during the past month. By contrast, biking was 2.44%, golf was 1.74%, and tennis was 0.36%.

1.1 Older adults’ travel behavior

Research on older adults’ travel behavior has studied trends and patterns in terms of trip frequency, trip distance, trip purpose, and mode choice. Older adults have different socioeconomic characteristics and different travel behaviors, compared to younger adults. According to the 2017 NHTS (Federal Highway Administration, 2022), older adults have smaller households (without children, most are a couple or single) and fewer vehicles, compared with all households. Most of them are unemployed (Samus, 2013), and their travel behaviors are strongly influenced by possession of driver’s licenses, and living with or without a partner (Cheng et al., 2019; Figueroa et al., 2014). Evidence shows that total trip frequencies and mean distance traveled decline as age advances (Boschmann & Brady, 2013; Moniruzzaman et al., 2013). Currie and Delbosc (2010) study shows that older adults have 30% lower trip frequencies overall compared to younger adults in Australia. The same results have been observed in the United States. According to the 2017 NHTS (Federal Highway Administration, 2022), older adult households generate about six trips per household daily, whereas younger adult households generate about 8.5 trips per household daily. The mean travel distance of older adult households is 17% shorter than younger adult households.

Older adults travel for different purposes than those in the labor force. Compared to younger adults, travel purposes for older adults consist of more shopping, recreational activities, and visiting medical offices instead of commuting to and from work (Chudyk et al., 2015; Samus, 2013; Winters et al., 2015). The car is still found to be the main transport option for older adults (Davis et al., 2011; Cao et al., 2010; Zeitler, 2013), though some studies have found the mode of travel shifts away from the car as people age (Boschmann & Brady, 2013; Cao et al., 2010). Many older adults shift from car driver to car passenger and then to public transportation because of loss of driver licenses (Hensher, 2007). Still, transit trips account for only a small proportion of the overall travel among older adults (Vine et al., 2012). Walking is a much more important mode of travel among older adults than transit. Additionally, destinations that facilitate more social interaction generate more walking among older adults (Winters et al., 2015).

1.2 Older adults’ travel behavior and built environment

The relationship between the built environment and travel behavior is well studied in the literature. Built environments are often characterized in terms of D variables. The Ds—development density, land use diversity, urban design, destination accessibility, and distance to transit—all have effects on travel behavior (Ewing & Cervero, 2010). These D variables have been widely used to explain trip distances, trip frequencies, mode choices, and overall vehicle miles traveled. While early studies have focused on adults in general, recent studies have paid more attention to specific groups like children, older adults, lower-income populations, women, and minority groups. Studies of older adults’ travel behavior have found that the relationships for the general population also apply to older adults (Chudyk et al., 2015; Winters et al., 2015; Vine et al., 2012). However, there are certain built environment variables that are particularly associated with active travel and physical activity in older adults. Land use diversity is one of the most widely reported variables in the literature on older adults. There is a positive relationship between the sum of destinations within walking distance of home and the number of walk trips by older adults (Cao et al., 2010). Older adults who live in diverse use neighborhoods have higher activity levels, instead of staying at home or traveling outside their neighborhoods (Rosso et al., 2013). Street design quality, like the presence and condition of sidewalks, the presence of benches, safe street crossings, etc., are other key feature for older adults (Hanson et al., 2012; Cheng et al., 2019).

Previous studies in literature have presented a good picture of older adults’ travel patterns and the roles of personal characteristics and built environment in older adults’ travel decisions. However, the remaining question is whether built environment can help keep older adults active. Older adults can be deemed more active and mobile if they generally travel as much in total as they age. Older adults also can be deemed more active if they travel more by walking, biking, and public transportation, which is also a positive result for their health (Fig. 1). This is an important question for planners, policymakers, and public health professionals to address the needs of the globally growing population of older adults.

Fig. 1
figure 1

Conceptual framework of older adults being active

Full size image

1.3 The objective of this study

The objective of this study is to answer the question: are older adults living in compact developments more active than those living in sprawling developments? The following two hypotheses will be tested to provide answers to the research question:

  • Hypothesis 1 – older adults living compact developments travel more in total than those living in sprawling developments; and

  • Hypothesis 2 – older adults living compact development travel more by non-auto travel modes (such as walking, transit) than those living in sprawling developments.

These two hypotheses are tested based on a unique dataset of older adults’ travel diaries from 36 diverse regions of the United States. This disaggregated dataset is the largest sample of household travel records ever assembled outside of the National Household Travel Survey (NHTS). And unlike NHTS, this dataset has precise geocodes for home locations and therefore can characterize the built environment of older adults’ neighborhoods precisely. As importantly, the dataset contains the widely used built environment variables for travel research: density, diversity, design, distance to transit, and destination accessibility, defined consistently across all the regions.

The rest of this paper is structured as follows: Section 2 describes the research methodology, including the study area, data collection, variables, analysis methods, and selection of models. Section 3 presents the results and discussions of the findings. Finally, Section 4 concludes the study and provides recommendations.

2 Research methodology

This study employs a cross-sectional research design to understand older adults’ travel behavior and determine the relative influences of household characteristics, built environment, regional factors, and weather conditions on older adults’ travel choices in 36 regions (metropolitan areas) of the United State. Especially, this study aims to test the relationships between built environment and older adults’ total trips and non-auto trips. The sequence of the methodology is shown in Fig. 2.

Fig. 2
figure 2

The sequence of the methodology

Full size image

2.1 Data collection

The most widely used data source to study travel behavior is the household travel diary survey. Household travel survey data are the fundamental input for regional travel demand modeling and forecasting. Many metropolitan planning organizations (MPOs) conduct their own travel surveys to calibrate and validate their travel demand models. In the survey, random households are selected to report their travel diaries on an assigned day. The travel diary records the details of trips made by all members of the household during the day including locations of origin and destination, time of departure and arrival, trip purpose, mode used for the trip, etc. Over the past decade, we have contacted more than 100 MPOs and collected household travel survey data. The main criterion for inclusion of regions in this study is data availability. Regions had to offer regional household travel surveys with XY coordinates, so we could geocode the precise locations of households and trip ends. It is not easy to assemble databases that meet this criterion, as confidentiality concerns often prevent MPOs from sharing XY data. The resulting pooled dataset consists of over 100,000 households in 36 regions (Table 1), from which older adult households could be extracted and travel choices analyzed. It should be acknowledged that the final sample used for the study is to some degree based on convenience sampling.

Table 1 Regions (metropolitan areas) in the dataset
Full size table

Relative to NHTS, this dataset provides much larger samples (particularly of older adults) for individual regions and precise locations of households. NHTS provides geocodes (identifies households) only at the census tract level. This is especially important to study non-auto travel modes, such as walking and transit, since the travel distances of walking are much shorter than driving and distance to transit stop is a key determinant of taking transit.

The regions included in this household travel survey sample were, in addition, able to supply GIS data layers for streets and transit stops, population and employment for traffic analysis zones, and travel times between zones by different modes for the same or close enough years as the household travel surveys. Point, line, and polygon data from the different sources were joined with buffers to obtain raw data, such as the number of intersections within buffers. These were then used to compute refined built environmental measures such as intersection density, which is simply the number of intersections divided by land area within the buffer.

What spatial extent of the built environment is most relevant to older adults’ travel decisions? Theoretically, buffers (distances from household locations) could be wide or narrow. Even a determinant as straightforward as walking distance could be anywhere from ¼ mile to 1 mile or more. In this study, the ½ mile buffer around household geocode locations was chosen.

The units of analysis for this study are older adult households, which are defined as households made up entirely of older adults (65-year-old or older). Originally, we identified households with any older adults, but we chose to focus on the households with only older adults. Accessibility and mobility become more serious issues when older adults live alone rather than as part of multi-generational households. For older adults living with younger adults, travel behaviors might be affected by those of younger adults.

2.2 Variables

The final dataset contained 122,134 trips made by 20,018 older adult households in 36 regions. To maintain a full complement of independent variables for subsequent analysis, trips were dropped for lack of travel mode and households were dropped for missing any of the household sociodemographic variables. The greatest loss of cases was due to unknown household income. As is often the case in travel surveys, household income went unreported by a large number of respondents. We could have dropped household income to increase the sample size, but household income is too important from a theoretical perspective to be omitted from the mode-specific analysis.

The dependent variables are numbers of trips made by older adult households using different travel modes, and for walking and transit use specifically, whether any trips were taken (Table 2). Bike trips were not included due to the small sample size with bike trips. There were only 207 households with bike trips even in this large dataset.

Table 2 Variable description
Full size table

Independent variables consist of socioeconomic characteristics and the built environmental variables that have been reported as important factors in travel choice by different studies in the literature. These variables cover all of the Ds, from density to demographics. Vehicle ownership is not included since it is an endogenous variable. Households owning fewer cars tend to drive less and use more of other travel modes, like walking and transit. This study also includes variables at the regional level: population measuring the size of a metropolitan area, average gas price, plus variables measuring temperatures/climate. The temperature variables were collected from Climate Data Online at the National Centers for Environmental Information in the same years as the household travel survey for each region. With different measures, a total of 18 independent variables are available to explain older adult travel behavior in this study. All variables are consistently defined from region to region.

2.3 Compactness

To answer the question as to whether older adults living in compact neighborhoods are more active than those living in sprawling neighborhoods, analysis of variance (ANOVA) and regressions were employed. The dependent variables were the travel outcomes of each older adult household, including number of total trips, number of walk trips, and number of transit trips.

Whether the place was compact or sprawling was defined based on the D variables (Ewing et al., 2014; Hamidi et al., 2015; Tian et al., 2019). In this study, D variables were measured within a ½ mile buffer of the older adults’ homes (in Table 3). First, a principal component analysis (PCA) was applied to measure neighborhood compactness. Density was represented by activity density, diversity by land use entropy, design by intersection density and percentage of 4-way intersections, destination accessibility by percentage of regional employment within 30 minutes by car and transit, and distance to transit by transit stop density. They were combined into a single principal component as the compactness index of the built environment, which is a linear combination of these D variables. Job-population balance, intersection density, and percentage of regional employment within 30 minutes by car and transit were dropped due to low factor loadings.

Table 3 Factor loadings for the compactness index of built environment
Full size table

The Kaiser-Meyer-Olkin (KMO) test statistic is 0.662, which exceeds the threshold value of 0.5. The test measures sampling adequacy for each variable in the model and for the complete model. Bartlett’s Test of Sphericity compares an observed correlation matrix to the identity matrix. Essentially it checks to see if there is a certain redundancy between the variables that we can summarize with a few factors. The test statistic is significant at the 0.001 level or beyond, which means PCA is appropriate.

The extracted principal component has an eigenvalue of 1.91, meaning that this one component explains as much variance in the dataset as 1.91 of the original variables. Factor loadings (correlations between the principal component and component variables) range from 0.642 for land use entropy to 0.756 for activity density, as shown in Table 3. Second, using factor score coefficients for the first principal component, the compactness index of the built environment was categorized into three groups – compact (the factor score was 0.4 above the average) with 25.6% households, sprawling (the factor score was − 0.4 below the average) with 36.8% households, and average (the factor score was between − 0.4 and 0.4) with 37.6% households. This categorical built environment variable became one of the independent variables for the ANOVA test.

When studying the effect of built environment on travel, sociodemographic characteristics should always be controlled. Household income was used here as the representative of socioeconomic status. Income was also categorized into three groups – low-income group (< $35,000) with 39.7% households, medium-income group ($35,000 – $70,000) with 33.8% households, and high-income group (>$70,000) with 26.5% households. All were in 2012 dollars.

2.4 Multilevel modeling

With the household travel survey data from 36 regions, our data structure is hierarchical, with older adult households nested within regions. Regardless of individual household characteristics, households located within the same region share some region-level characteristics, such as the size of the region, transportation network, weather, public policies, etc. The best statistical method to deal with nested data is hierarchical modeling (HLM), also called multilevel modeling (MLM). The essence of HLM is to isolate the variance associated with each data level. Specifically in this study, HLM partitions variance between the older adult households (Level 1) and regions (Level 2) and then seeks to explain the variance at each level.

The dependent variables we were most interested in, total trips, walk trips, and transit trips, are count variables, with nonnegative integer values, many small values, and few large ones. Generally, Poisson or negative binomial regression is appropriate for this kind of data distribution. The selection between Poisson and negative binomial depends on the dispersion coefficients. If the count variable is over-dispersed – the variance of the variable is greater than the mean, negative binomial regression is more appropriate. Otherwise, Poisson regression should be used. In addition, both walk trips and transit trips are zero inflated, meaning each has excessive numbers of zero values relative to Poisson and negative binomial distributions. A two-stage hurdle modeling approach (Ewing et al., 2015; Tian & Ewing, 2017) is more appropriate for this kind of data distribution. In the two stages, stage 1 categorizes older adult households as having at least one walk/transit trip or not, and uses binomial logistic regression to distinguish these two states. The stage 2 model estimates the number of walk/transit trips generated by older adult households with any (positive) walk/transit trips. At stage 2, either Poisson regression or negative binomial regression can be used depending on the dispersion. With these criteria, a multilevel negative binomial regression was employed to analyze older adult household total trips and multilevel two-stage hurdle models (multilevel logistic regression for stage 1 and multilevel negative binomial regression for stage 2) were employed to analyze walk trips and transit trips.

The models were estimated with HLM 7, Hierarchical Linear and Nonlinear Modeling software (Raudenbush et al., 2010). Different Ds may emerge as significant in different models, so trial and error was used to arrive at the best-fit models for the travel outcomes of interest. Variables were substituted into models to see if they were statistically significant and improved goodness-of-fit. For each dependent variable, we were looking for the model with the most significant t-statistics and the greatest log-likelihood.

3 Results and discussion

3.1 Descriptive statistics

The descriptive statistics in Table 4 show the differences between older adult households and younger adult households in terms of their socioeconomic and travel characteristics. Specifically, the average household income for older adults is about two thirds that of younger adults ($54,960 vs. $83,704). When people get older and start to retire, their income decreases significantly. The vehicle ownership of older adult households is higher (0.97 vs. 0.85), with about one vehicle per person. The household size is smaller (1.58 vs. 2.70), with 1.58 persons per household, which indicates that there are about half of older adults living alone. Fewer members in older adult households are employed (0.40 vs. 1.42). Older adult households also generate fewer total trips compared to younger households (6.10 vs. 10.38).

Table 4 Younger adults vs. older adults
Full size table

Overall, older adults travel less than younger adults. They are more auto-dependent compared to younger adults. They travel more by auto, yet less by all other modes (such as walk, bike, transit, etc.). Their trips are also shorter, in terms of both travel time and distance. About two thirds of older adult’s trips are home-based trips (either start or end at home), which is not much different from younger adults. The big difference is home-based-work trips. As people get older and start to retire, home-based-work trips drop down from 14.6% for younger adults to 5.8% for older adults. Looking deeper into the trip purposes, school- and work-related trips only make up 5.6% for older adult households, where they are 18.7% for younger adult households. At the same time, shopping trips increase to 26.2% for older adult households, where they are 15.6% for younger adult households.

3.2 Analysis of variance (ANOVA)

With the two independent variables: household income and neighborhood compactness, a two-way ANOVA test was conducted for each of the travel outcomes. The results are presented in Table 5 and Fig. 3. The estimated marginal mean value, instead of descriptive mean or actual mean, is reported because the estimated marginal mean is the mean response for each factor, adjusted for any other variables in the test. The estimated marginal means for income and built environment were adjusted for the covariation between them. This, of course, is the reason for including income in the test – particularly, we want to see if the built environment factor still has an effect, beyond the effect of income.

Table 5 Results of two-way ANOVA for the effects of income and built environment on older adults’ travel behavior
Full size table
Fig. 3
figure 3

Means (of numbers of trips per household per day) plot of the two-way ANOVA test for the effects of income and built environment on older adults’ travel behavior

Full size image

The results show significant variations existed in number of total trips and trips made by different modes across neighborhood types and income levels. The F values show the ratio of the variation between neighborhoods (or income groups) to the variation within neighborhoods (or income groups) – higher ratios suggest a stronger effect, as indicated by the low p-value.

For the number of total trips, both income and built environment have significant effects. However, the income effect is much greater than the built environment effect based on F values. Older adult households with higher incomes generate more trips across compact and sprawling neighborhoods. The difference between compact and sprawling neighborhoods is minor at all income levels. Hypothesis 1 is not confirmed and older adults living in compact neighborhoods do not travel more in total trips than those living in sprawling neighborhoods. This is consistent with the literature that the number of total trips generated by a household is primarily determined by its socioeconomic characteristics (Ewing & Cervero, 2010).

For the number of walk trips, both income and built environment have significant effects. However, the built environment effect is much greater than the income effect based on F values. Older adults living in compact neighborhoods generate about three and five times as many walk trips as in average and sprawling neighborhoods, respectively. This is clear evidence that older adults living in compact neighborhoods are more active than those living in sprawling neighborhoods and that Hypothesis 2 is confirmed. The difference among the three income levels is minor for all types of neighborhoods. It indicates that for the number of walk trips generated by a household, built environment is more important than socioeconomic characteristics.

For the number of transit trips, both income and built environment have significant effects. Besides, the interaction between income and built environment is also significant and the strongest among all tests. Overall, as neighborhoods get more compact, older adult households generate more transit trips. This is also clear evidence that Hypothesis 2 is confirmed. As income gets higher, older adult households generate fewer transit trips. The difference of transit trips among neighborhoods is greater than across income levels, which indicates that the built environment is more important than socioeconomic characteristics for transit trips. Besides, older adult households living in compact neighborhoods with the lowest income level generate the highest number of transit trips.

For the number of auto trips, both income and built environment have significant effects. As neighborhoods get more compact, older adult households generate fewer auto trips, which is a positive result for traffic safety as driving ability declines at advanced ages. As income goes up, older adult households generate more auto trips. This is consistent with what we know from the literature.

In sum, Hypothesis 1 is not confirmed, and Hypothesis 2 is confirmed. Older adults living in compact developments do not travel more in total but travel more by walking and public transportation than those living in sprawling developments. These findings support the literature that the built environment still has a significant effect on older adults’ travel behavior after controlling for sociodemographic characteristics. More importantly, these findings suggest that older adults living in compact neighborhoods are more physically active than those living in sprawling neighborhoods through more non-auto travel. They travel more by walking and public transportation, yet travel less by automobile. It is also worth noting that older adults living in compact neighborhoods with the lowest income level generate the highest number of transit trips.

3.3 Multilevel modeling

3.3.1 Total trips

The best-fit model for household total trips is presented in Table 6. The elasticities of total trips with respect to the individual variables are also shown in the table. An elasticity is the ratio of the percentage change in one variable associated with the percentage change in another variable (Ewing & Cervero, 2010). The number of older adult household total trips increases with household size and income and decreases with the age of the youngest household member. Larger and younger households with higher incomes generate more trips and are more active. For the built environment, the compactness index is not statistically significant. Again, this supports that Hypothesis 1 is not confirmed and older adults living in compact neighborhoods do not travel more in total than those living in sprawling neighborhoods. At the regional level, the number of household total trips decreases with regional population. The greater the region’s population size, the more destinations can be reached with fewer trips. Again, the result of this model is consistent with the literature and the previous test that the number of total trips generated by a household is primarily determined by its socioeconomic characteristics.

Table 6 Multilevel negative binomial regression for household total trips
Full size table

3.3.2 Walking

The best-fit models for walk trips are presented in Table 7. The age of the youngest household number is the only statistically significant socioeconomic variable. The older the youngest household member is, the less likely the household will take any walk trips. For the built environment, the likelihood of any walk trips increases with the compactness index. This index measures density, diversity, design, and distance to transit. The higher values of these Ds mean more destinations within walking distance of home. At the regional level, the likelihood of any walk trips decreases with regional population. The greater the region’s population size, the lower the percentage of destinations that can be reached by walking. The likelihood of any walk trips increases with regional annual average low temperature and decreases with number of days with temperature greater than 90 °F. This means, generally, warmer temperatures encourage older adults to walk, but not extremely high temperatures.

Table 7 Two-stage multilevel regressions for household walk trips
Full size table

The number of walk trips for the subset of older adult households that make walk trips increases with household size and decreases with the age of the youngest household member. The number of walk trips increases with the compactness index. It also increases with regional gas prices and decreases with the number of days with temperature lower than 32 °F. The higher the regional gas price, the more competitive is walking compared with maintaining vehicle ownership and driving. The relationship of walking to the built environment has already been discussed, as has the relationship of walking to temperature.

The compactness index is statistically significant and has a positive sign in both models, which indicates that Hypothesis 2 is confirmed. That means older adults living in compact developments do travel more by walking than those living in sprawling developments.

3.3.3 Transit

The best-fit models and elasticities for transit trips are presented in Table 8. The likelihood of an older adult household making any transit trips decreases with household size and household income and increases with household workers. The likelihood of any transit trips increases with the compactness index of the built environment. At the regional level, the likelihood of any transit trips increases with regional gas prices. The higher the regional gas price, the more competitive is taking transit compared with maintaining vehicle ownership and driving, especially, for older adult households with decreased income. The likelihood of any transit trips decreases with regional annual average low temperature and the number of days with temperature lower than 32 °F. This is the same with walking – that warmer temperatures encourage older adults to get out and take transit.

Table 8 Two-stage multilevel regressions for household transit trips
Full size table

The number of transit trips for the subset of older adult households that make transit trips increases with household size and decreases with number of workers in the household and household income. The compactness index and all the regional variables are not statistically significant.

The compactness index is statistically significant and has a positive sign only in the multilevel logistic model, which indicates that Hypothesis 2 is partially confirmed. That means older adults living in compact developments are more likely to use transit but not necessarily generate more transit trips (if they use transit at all) than those living in sprawling developments.

The pseudo-R2 of the multilevel models range from 0.09 to 0.66. However, the pseudo-R2 in multilevel regressions is not equivalent to the R2 in linear regressions. It bears some resemblance to the statistic used to test the hypothesis that all coefficients in the model are zero, but it is not a measure of how well the model can predict the outcome variable as in linear regression.

4 Discussion

This study comes with limitations. First, we had a two-level dataset with older adult households (Level 1) nested within regions (Level 2). The sample size at Level 2 is small (36 regions) and has affected the results. Having a larger sample size at Level 2 would have helped achieve more significant associations between explanatory and outcome variables. Still, this study had one of the largest sample sizes, compared to other studies on the same topic in the literature. Future work is needed to expand the sample size at Level 2.

Second, to what extent residential self-selection affects the results of this study is unclear due to the lack of data. Existing studies claim that the strength of built environment-travel behavior associations drops considerably after controlling for the effect of residential self-selection (Zang et al., 2019). While this study found correlations between the built environment and older adults’ walking and transit use, further research is needed to examine these relationships while controlling for older adults’ travel attitudes.

Third, we did not have health-related information to include as controls. However, had we controlled for health might be problematic because causality can run in both directions. On one hand, walking and other active travel can contribute to better health. On the other hand, better health allows older adults to travel more by active modes.

Summarizing the results, we found the following patterns regarding older adults’ travel behavior and choice. Compared to younger adults, older adults travel less (fewer total trips, shorter travel time and distance) and are more auto dependent. Their travel needs are more shopping related and less work and school related. These findings are consistent with the literature. Older adults with lower incomes travel less than those with higher incomes. In addition, older adults with lower incomes are more likely to use transit and use transit more frequently, which is indicated by ANOVA tests. This is concerning, as Reina and Aiken’s study (Reina & Aiken, 2022) found that the share of subsidized households headed by older adults grew in a national sample since 2000 and neighborhood amenity measures did not reflect this change.

The statistical analysis (both ANOVA tests and multilevel modeling) consistently shows that Hypothesis 1 is not confirmed but Hypothesis 2 is confirmed. That means that while the number of total trips generated by an older adult household is still primarily determined by sociodemographic characteristics (such as income), modal variables (such as numbers of walking and transit trips) are primarily determined by the built environment in their neighborhoods. Older adults living in compact neighborhoods are more active than those living in sprawling neighborhoods. They do not travel more generally. But rather, they travel more by walking and public transportation, and travel less by automobile. More specifically, the compactness index of the built environment is not statistically significant in the model of total trips but is statistically significant in three of the four non-auto travel models and has large elasticities. Overall, the model results show that the built environment of neighborhoods where older adults live – density, diversity, design, and distance to transit – are important to keep older adults active.

Weather has been reported as an important factor on individual’s travel behavior and travel choice in general (Böcker et al., 2013). This study tests whether this is the case for older adults. The results show that weather conditions, specifically temperature, have a strong impact on older adult’s walking and transit usage behavior. Warmer temperatures encourage older adults to walk and take transit. Extreme temperatures discourage older adults from traveling by walking or transit. We as planners cannot control the weather. Still from a planning perspective, by knowing the influence of weather on older adult’s travel behavior and choice, it benefits planners to address the influence of weather when designing housing or other facilities for older adults. Also, by controlling weather in the modeling process, it helps to uncover the true relationship between the built environment and older adult’s travel behavior.

It is worth noting that the effect of the regional gas price on older adult households’ decision to take transit or not has the largest elasticity among all the variables tested. It makes sense that older adults may be very sensitive to travel costs as household income decreases with aging. This is also supported by the ANOVA test that older adults living in compact neighborhoods with the lowest income levels generate the highest number of transit trips. Household income and transit stop density included in the compactness index are statistically significant in both transit models. All the evidence indicates the importance of public transportation for older adults, especially low-income ones, to support their mobility and keep them active. A recent study (Li et al., 2022) found that looking for a more supportive neighborhood is the second most prevalent reason (30%) for older adults to relocate.

5 Conclusion

Are older adults living in compact development more active? The answer is yes. Older adults living in compact neighborhoods are more active than those living in sprawling neighborhoods, not by traveling more in total but traveling more by walking and transit. Walking and transit can be the alternatives and affordable options to fulfill older adults’ independent mobility needs (Ravensbergen et al., 2022), while also keeping them active. In fact, public transportation is older adults’ main travel choice in some other countries (Hu et al., 2013). Given the sprawling-built environment in most cities in the United States, older adults won’t suddenly shift to walking and transit. To encourage older adults to walk and use transit, it takes efforts from other parties in addition to create highly walkable, mixed-use, and compact neighborhoods. For example, transit authorities should consider older adults’ travel patterns and reform their policies by developing reduced-fare programs, adding more frequent bus stops, and expanding the use of low-floor vehicles (Hess, 2009). In addition, studies found that training programs that support older adults to try mobility alternatives have been very effective (Ravensbergen et al., 2022; Shaheen & Liu, 2010). These programs include obtaining skills, knowledge, and confidence through experience and practice, as well as providing detailed information on walking distance, presence of elevator or ramp, size of stairs, etc.