Transportation

, Volume 37, Issue 1, pp 105–123

An analysis of trip chaining among older London residents

Authors

  • Jan-Dirk Schmöcker
    • Department of Civil and Environmental EngineeringTokyo Institute of Technology
  • Fengming Su
    • Institute of Comprehensive Transportation of NDRC
    • Alan M. Voorhees Transportation Center, Edward J. Bloustein School of Planning and Public PolicyRutgers University
Article

DOI: 10.1007/s11116-009-9222-z

Cite this article as:
Schmöcker, J., Su, F. & Noland, R.B. Transportation (2010) 37: 105. doi:10.1007/s11116-009-9222-z

Abstract

This paper examines the trip chaining complexity of individuals in London. We adopt two definitions of trip chaining. One based on a 30 min dwell time rule and a second based on home-to-home tours. Our focus is on the complexity of the trip chains as measured by the number of stops on a given tour. The analysis uses the London Area Travel Survey and examines the factors associated with trip chaining for people aged over 65. A comparison with those aged under 65 reveals that older people on average make more complex tours when the 30 min dwell time rule is applied as opposed to when the home-to-home definition is applied. It is further shown that the anchor points of the tour are critical for determining tour complexity, suggesting the usefulness of the 30 min definition. Our analysis also suggests that older people reduce total home-to-home tours by combining different trips into single tours. Through descriptive analysis and ordered probit regression models we examine how reported levels of disability affect their trip chaining and we examine household demographic characteristics as well as proxies for accessibility, such as local population density. The analysis shows that disabilities do not necessarily lead to reduced tour complexity except when walking difficulties become so severe that independent travel is not possible. We suggest that tour complexity might further increase in the future, for example as the spread of mobile phone usage appears to have a further positive influence on tour complexity. Implications for land-use and transport planning are discussed.

Keywords

Trip chainingOlder peopleTour definition

Introduction

The phenomena of trip chaining involves the linking together of trips, usually by car, such that activities are carried out at multiple destinations. Recent research, primarily in the US, has examined the growing propensity and increased complexity of trip-chaining behavior (McGuckin et al. 2005; Noland and Thomas 2007). As populations in most developed countries age, there is a question as to whether people will continue to engage in complex trip chaining, but more importantly, how the spatial distribution of activities may make this increasingly necessary. As the share of older people in the population increases, these questions are important to understand, not only for how travel behavior affects individual lifestyles, but to better plan for future travel patterns and the ability of the transport system to deliver the mobility that ageing individuals demand.

Trip-chaining is often seen as a way to reduce the costs of travel, since activities can be more efficiently carried out when linked in sequence. This may be true especially of shopping trips, which are one of the main types of trips engaged in by older people. On the other hand, the complexity of planning and implementing a trip-chain may not be trivial and may be more difficult if one is unable to drive or faces a physical disability.

This paper attempts to examine these issues using the London Area Travel Survey (LATS) from 2001. The metric chosen for examining trip-chaining activity is the number of stops on a specific tour, which serves as a measure of the complexity of the trip taken. The goal is to understand the impact of socio-demographic and trip characteristics on the propensity to chain trips as well as individual physical disabilities, which can be a major factor amongst older people. An additional focus of the paper is on the spatial distribution of population, in this case proxied by population density. Analysis by Noland et al. (2007) with US data showed that less dense areas are associated with more trip chaining and Frank et al. (2008) found that the number of stops in a tour is best explained by urban form but it is unclear if this can also be verified for London.

Our paper is organized as follows: the following section reviews the trip chaining literature and in particular findings regarding older people’s travel behavior and the influence of population density. We then briefly describe the LATS data and assumptions made on what defines a tour and what defines a trip within a tour. This is followed by our descriptive analysis and an explanation of the ordered regression models estimated. We then discuss the results and provide conclusions.

Literature review

The study of trip chaining has a long history in the transportation literature. Early studies were based on understanding the geography of urban areas and the linkages between trips (Hanson 1980; Takahashi 1986), and especially how shopping trips are linked. Much of the literature on trip chaining has focused on how to better model and forecast travel. For example, Kitamura (1985) investigated the possible treatment of interdependent destination choices in a trip chain and found that if the interdependency is not accounted for, estimation results may be biased. Hensher and Reyes (2000) estimated the relationship between mode choice, especially the use of public transport, and trip chaining. They found that as individuals move from a simple tour (such as, home–work–home) to an increasingly more complex tour (e.g., home–school–work–home) the likelihood of using public transport decreases with the increasing number of links in the chain. The result is consistent with Cirillo and Axhausen (2002) and Ye et al. (2006), who found that complex patterns (involving several stops) are preferably performed by car.

Golob (1986) found that the life cycle of a household is the most important variable for determining the sequences of activities in trip chains, followed by age and income. McGuckin et al. (2005) found that gender and life cycle (based on the number of adults and the age of children) most affected trip-chaining behavior. Noland and Thomas (2007), however, find in their multi-variate analysis that estimated coefficient values of age groups and their association with trip complexity (i.e., the number of stops), varies little for those older than 26, but has a slight drop in complexity for those older than 76. They also confirm that higher income households have more complex tours and that the presence of young children increases tour complexity.

Trip chaining research has more recently focused on specific population segments, especially on gender differences. Using the 1995 Nationwide Personal Transportation Survey (US), McGuckin and Murakami (1999) found that women, especially with children in the household, are more likely to chain household sustaining trips in tours to and from work. Handy (1996) investigated the non-work travel of women, who generally face greater constraints on travel than men, due to greater time pressures and greater concerns about personal safety. For example, women are expected to take more responsibilities for looking after children and doing household maintenance trips such as grocery shopping, which puts them under greater time pressure. Ye et al. (2006) found that gender does not significantly influence tour complexity in the case of non-work tours using 2000 Swiss Microcensus travel survey data.

This paper focuses on older people. The increase in the population of older people will cause challenges for transport policy (e.g. Metz 2000; Rosenbloom 2003; Golob and Hensher 2007) as compared to younger people, older people are more likely to have restricted mobility. Kendig et al. (2000) found that the major impact of illness and pain amongst older people was on how this limited their activity participation, which in turn was related to lowering their well-being. Metz (2000) also notes that quality of life in old age is related to mobility. Rosenbloom (2003) observed that in the US virtually all older people eventually are confronted with medical and other constraints that affect their mobility and consequent activity in community life.

Alsnih and Hensher (2003) evaluate the evidence on the mobility needs and travel patterns of individuals over 64, and make the further breakdown of distinguishing between the ‘‘young’’ elderly (aged 65–75 years) and the ‘‘old’’ elderly (over 75 years). The younger old are found to often have very few health restrictions and are often more mobile than those younger than 64 if they are retired. Only after age 75 does declining health lead to negative mobility effects on quality of life. Our analysis enables us to further disaggregate any effects between these age groups and even older individuals who may be more likely to suffer medical constraints on their mobility.

Trip-chaining research has more recently examined older age groups. Kim (2004) reported that older people are more likely to share a ride with others when chaining trips and are less likely to use public transport for shopping or doing errands. Hensher and Reyes (2000) found that as age increases, the probability of complex car tours during work decreases while complex car tours to and from work increase.

Golob and Hensher (2007) use data from the annual Sydney travel survey to analyze travel behavior of the city’s population with a specific focus on older people. Using multiple correspondence analysis they illustrate in graphs the relationships between key socio-demographic variables and travel and mobility variables. Their analysis is focused on the usage of modes as well as the number of home-to-home tours. Golob and Hensher show that interaction between gender and marital status results in different mode preferences and tour frequencies. Other key results are that income does not have a significant influence on tour numbers and that reduced travel activity is especially pronounced for those aged over 85. They found that trip chaining peaks at ages between 45 and 54 and decreases after age 65.

Goulias et al. (2007) use panel data for 20,000 residents near Seattle, Washington. Their data set offers the possibility to observe how the baby boomer generation is adapting their travel behavior when they reach retirement age. Using cluster analysis they find that many adapt to the freedom of not commuting by having more flexible travel plans and may reduce travel 1 day but increase travel on a subsequent day. Overall, Goulias et al. conclude that those within the baby boomer generation are not changing their level of travel and continue to drive, posing new challenges for transport planning. One key issue is neighborhood design. Cao et al. (2007) found that better access increases walk trips and that distances to grocery stores has a large effect on walking amongst older people.

About 15% of London’s population is aged 60 or over amounting to 1.2 million people. About 250,000 are aged over 80 and these groups are increasing as in other western cities (Greater London Authority 2006). One feature of London is the widespread availability of public transport. Giuliano et al. (2003) reported that this makes it possible for older people to use public transport more than in the US, but that lower incomes and lower car ownership also are a factor. Findings from the UK’s Department for Transport (2001) also suggest that in general older people in town and city centers are generally more reliant on public transport in Britain, compared to some other countries.

With this background in mind, this study focuses on factors influencing trip chaining of older people in London. Of particular interest are age, disabilities and accessibility patterns as proxied by population density. Our data further allow us to investigate the importance of factors associated with “active ageing” such as increasing car availability and the impact of increasing mobile phone usage on travel patterns. We also examine how different definitions of trip chaining can affect the interpretation of results. Our data enable us to apply different tour definitions. We investigate the importance of these and suggest that defining tours differently leads to different results.

Tour definitions and LATS data

The trip-chaining literature contains a multitude of definitions. In order to define trip chaining, several terms describing travel activities need to be clarified. The most elemental unit of travel activity is the stop, also called sojourn, often defined as a place of activity remote from home (Kitamura 1985; Adler and Ben-Akiva 1979). The movement which carries an individual between his home and a stop or between temporally consecutive stops is called a trip (Lee et al. 2002; Kitamura 1985; Adler and Ben-Akiva 1979).

There are, however, at least two definitions for tour: in the majority of the literature a set of consecutive trips that begin and end at an individual’s home or work is called a tour (e.g. Rutherford et al. 1997; Adler and Ben-Akiva 1979; Kitamura 1985; McGuckin and Murakami 1999). If a tour is composed of two or more stops, it is defined as a trip chain or a complex tour. Frank et al. (2008) adopt a method of defining tours based on the primary purpose of the tour.

The US Federal Highway Administration (FHWA) recently has adopted a rather different definition using a 30 min dwell time threshold. Based on both analysis and expert opinion that this cut-off was a reasonable definition (McGuckin et al. 2005, and personal communication with Nancy McGuckin) they defined a tour as a set of trips that was interrupted by a dwell time longer than 30 min and not necessarily chained to a specific starting and ending point, such as home or work. One of the benefits of this standardized definition is that it allows future analyses of these issues to be based on a similar definition for comparison of trends.

The analysis in this paper uses the interim release of the 2001 LATS, made available by Transport for London (TfL). In addition to gathering household data, the survey involved a personal questionnaire, and a one-day interviewer-administered trip recall interview. In addition, respondents were invited to complete a self-completion trip diary for one day during the following week. 67,252 individuals from 29,973 households were interviewed. From the trip recall interview, 176,453 trips were recorded.

From the data set, we extracted records for all persons aged 65 or older. Of this sub-sample 6,406 persons made at least one trip on the day surveyed which sums up to a total of 19,827 trips. Unfortunately, this data set does not include trips made on weekends, so only data about trips on weekdays are available.

LATS includes data on the starting and ending time of each trip as well as on the trip purpose and whether the starting and ending points are the resident’s home, work place or a different location. This allows us to define tours using the “30 min dwell time threshold” definition as well as the “home-to-home” definition of tours. After data cleaning 7650 ‘home-to-home’ and 13971 ‘30 min dwell time’ tours could be used for our analysis.

In order to understand the impact of population density on trip chaining we further use Census 2001 data (Casweb 2006) to obtain population estimates for each “three-digit-postcode area” within London. As can be seen in Fig. 1 the size of postcode areas can vary significantly but typically contain several thousand households. The figure further shows, not unexpectedly, that the lower density areas are at the edge of the city, although the central area also contains one lower density area, but with a high density of employment.
https://static-content.springer.com/image/art%3A10.1007%2Fs11116-009-9222-z/MediaObjects/11116_2009_9222_Fig1_HTML.gif
Fig. 1

Population density by three digit postcodes in London

Descriptive analysis

Noland and Thomas (2007) and Noland et al. (2007) report on stop numbers per tour using US data. The samples from their data are based on the entire US and hence cover extremely rural as well as extremely densely populated areas. Comparing this to the very different urban setting of our data for London shows that average stop numbers are lower in London. The pattern is similar, with those under 65 making less complex tours. This increases for older people, in particular the “younger-old”, aged 65–74. The results are, however, very different if the tour definitions are changed to the home-to-home definition. In this case we find that with increasing age, tour complexity decreases, especially for those aged over 75. This shows the importance of the tour definition and suggests that older people make more complex tours with relatively short stops (Table 1).
Table 1

Average trips per tour by age group

 

US (30 min dwell time)

London (all tours, 30 min dwell time)

London (home-to-home tours)

16–64

1.40

1.19

2.52

“Young–old” (65–74)

1.46

1.29

2.40

“Old–old” (75+)

1.44

1.31

2.36

Entire sample

1.38

1.21

2.50

Table 2 lists the number of trips per tour based on total home-to-home tours made per day. The trips per tour are statistically different by age groups for 1 or 2 tours per day, based on one-way ANOVA tests. The ANOVA test for 3 or more tours per day shows no statistical significance, although t-tests show a significant difference between those aged 16–64 and the older population. (Further t-tests show that further differentiations of more than 3 tours are not significant.) Clearly with more tours per day the number of trips per tour decreases for all age groups. This suggests there can be some increase in tour efficiency by linking more trips into a single tour. It supports the hypothesis that trip chaining is at least to some degree done to reduce overall travel costs. This pattern persists for the older age groups, although at a lower level of total trips per tour (t-tests show that this difference is statistically significant between the youngest group and the two older groups, but the means for the two older groups are statistically indistinguishable).
Table 2

Average trips per tour by total tours per day (home-to-home)

Tours per day

16–64

65–74

75+

Entire sample

1

2.64

2.47

2.38

2.60

2

2.44

2.36

2.28

2.43

3+

2.33

2.27

2.20

2.32

All tours

2.51

2.40

2.36

2.50

In Table 3 we show a similar analysis for tours defined by the 30 min dwell time rule. Recall that this represents short tours, where the chains are defined by any stop of less than 30 min. The pattern here is quite different, with the lowest average of the total tours being for those with 2 tours per day. For those aged 16–64 these are most likely mainly commuting trips with no intermediate stops. Thus we see that those that make only one tour per day make the largest number of stops and for those making 3 or more tours per day, the number of stops decreases. We cannot distinguish a statistical difference in the mean for those that make more than 4 tours, but one-way ANOVA tests show that the others are statistically significant. Across age groups the main difference is that as age increases, more stops occur when only one tour is made, with a significant drop off for 2 or more tours. This suggests that older people tend to make more stops when they only go once out of the house (for shopping, errands or leisure purposes with short dwell times).
Table 3

Average trips per tour by total tours per day (30 min dwell time)

Tours per day

16–64

65–74

75+

Entire sample

1

1.84

2.10

2.24

1.94

2

1.10

1.15

1.11

1.10

3

1.28

1.27

1.23

1.28

4

1.16

1.17

1.10

1.16

5+

1.21

1.18

1.16

1.21

All tours

1.19

1.29

1.31

1.21

Of more interest is the link between total tours per day and total kilometers travelled. We would expect that those who make more tours per day would tend to travel more on average, but at a diminishing rate. This is only to some degree confirmed in the cross-tabulation shown in Table 4. Those making one tour per day have the lowest average total distance per day (with a statistically significant difference). For those making 2 or more tours per day we see a slightly larger amount of total distance travelled, but average mileage per tour per day decreases. Older people travel lower average distances than younger people with the same diminishing distance pattern as number of tours per day increase. Interestingly the total distance per day travelled decreases for those making 3 or more tours, though statistically the difference is not significant. Of particular interest is that the total distance for those aged 65–74 increases for 3 or more tours compared to other age groups. This might be because they have both more discretionary time than younger age groups and fewer physical constraints or fatigue than older age groups. However, statistically these results are weak.
Table 4

Average total distance travelled per day, in km, by total tours per day (home-to-home)

Tours per day

16–64

65–74

75+

Entire sample

1

21.0

15.4

8.9

19.7

2

24.6

18.1

14.7

23.5

3+

23.9

19.8

13.8

23.3

All tours

22.0

16.3

10.0

20.8

Table 5 cross tabulates the same relationship but applying the 30 min dwell time rule. Note that the total distance for all tours is slightly lower for all the age groups compared to Table 3 for the home-to-home tours. This probably is because with the home-to-home tour definition, trips that do not return home are excluded. Interestingly those who make 2 tours per day travel in total less distance per day than those making only 1 tour. As explained above, 1 tour means the person is making a single tour for errands or leisure, whereas 2 tours are most likely tours to a single destination at which the person stays for a longer time (such as work) before returning home. It is likely that for the 16–64 age group, this is due to work trips, where the work activity breaks up the day. For those 65 and over, when they make only one tour per day, it may represent several errands with short activities (leading to more kilometres of travel), while two tours suggest they have longer periods spent in shopping or leisure activities (and thus less total travel). For the 75 + age group, the distance for one tour is noticeably less, suggesting perhaps that they have less energy to engage in longer activities and longer distances. Independently of the number of tours per day the total distance travelled is less as age increases, and is confirmed by ANOVA tests across age groups.
Table 5

Average total distance travelled per day, in km, by total tours per day (30 min dwell time)

Tours per day

16–64

65–74

75+

Entire sample

1

26.6

21.8

8.7

23.4

2

18.1

12.8

8.1

17.0

3

24.2

18.5

13.9

23.2

4

24.3

18.9

15.5

23.5

5+

29.8

23.1

16.8

29.1

All tours

22.1

16.6

10.2

20.9

For London we further analyzed the number of stops per tour by start and end point of the tour (Table 6) applying the 30 min dwell time tour definition. We find that the majority of the tours are made without any intermediate stop before reaching the main destination and that this is especially true for work trips. We also find surprisingly many home-to-home tours recorded with zero stops. Interestingly the percentage of these is significantly larger among those aged over 65. Some of these trips might be walks for the sake of exercise or with a dog, but there might also be some data issues as some respondents might have indicated walks to a nearby store as a tour without a stop.
Table 6

Number of stops for tours starting at home

Number of stops

Tour endpoint

Total

Home

Other

Work

(a) By those under 65

0

788

8.96%

26,209

91.76%

19,705

93.78%

46,702

1

6,694

76.15%

1,745

6.11%

1,013

4.82%

9,452

2

943

10.73%

463

1.62%

250

1.19%

1,656

3

272

3.09%

99

0.35%

36

0.17%

407

4

75

0.85%

33

0.12%

7

0.03%

115

5

13

0.15%

10

0.04%

2

0.01%

25

6+

5

0.06%

5

0.02%

0

0.00%

10

Total

8,790

 

28,564

 

21,013

 

58,367

(b) By those over 65

0

47

2.34%

4,607

89.54%

320

91.43%

4,974

1

1,459

72.66%

396

7.70%

20

5.71%

1,875

2

373

18.58%

114

2.22%

8

2.29%

495

3

98

4.88%

17

0.33%

2

0.57%

117

4

25

1.25%

7

0.14%

0

0.00%

32

5

5

0.25%

2

0.04%

0

0.00%

7

6+

1

0.05%

2

0.04%

0

0.00%

3

Total

2,008

 

5,145

 

350

 

7,503

Comparing those under and over 65 we further find a wider distribution of multiple stops among older people, i.e. those over 65 make more tours with multiple stops. For those still travelling to work, there are more stops, compared to those under age 65. It is not clear why this is the case, but perhaps those over 65 and still working have selected jobs with more schedule flexibility.

In order to obtain an initial understanding about the effect of population density, Table 7 groups the tours (based on the home-to-home tour definition) by population density of the home address. For those under 65 the population density does not seem to have a significant effect on trips per tour but for those over 65 there seems to be some weak tendency for more trip chaining if population densities are lower. Comparing the tour and trip lengths of those under and over 65 a significant difference can be observed. Looking at trips only, Schmöcker et al. (2005) found a similar trend of decreasing trip length with advanced age, with the exception of recreational trips. Table 7 further suggests that with lower population density, trips as well as tours of those under 65 become longer. However, this trend is much less clear for those over age 65.
Table 7

Tour length and modal share for different residential population densities (home-to-home tours)

 

Observations

Average total km per tour

Average distance per tour link

Average trips per tour

Average tour km for tours in personal vehicles

Average tour km for tours with other modes

(a) By those under 65

<500 per sq mile

167

2.85

1.14

2.51

3.97

1.37

500–1,000

848

2.33

0.91

2.54

2.53

1.97

1,000–2,000

923

1.93

0.77

2.51

1.82

2.11

2,000–4,000

1,224

1.72

0.74

2.43

1.73

1.84

4,000–10,000

13,865

1.57

0.64

2.49

1.67

1.50

10,000–25,000

32,915

1.34

0.55

2.45

1.55

1.24

25,000 or more

9,811

1.11

0.45

2.52

1.70

0.98

(b) By those 65 and over

<500 per sq mile

41

0.84

0.47

2.63

1.42

0.96

500–1,000

150

1.27

0.63

2.45

2.72

0.60

1,000–2,000

107

0.87

0.39

2.42

1.12

0.72

2,000–4,000

244

0.98

0.58

2.27

1.40

1.21

4,000–10,000

2,142

0.75

0.45

2.39

1.18

0.89

10,000–25,000

4,052

0.72

0.43

2.39

1.58

0.71

25,000 or more

914

0.74

0.49

2.32

2.44

0.87

In general tours made by private cars are much longer than those made by other modes. The difference in average tour length is much more pronounced for those under 65. Further, as one would expect the share of public transport increases in high density areas for those aged under 65. This does not seem to be the case for those aged over 65. Table 7b shows that those who live in more densely populated areas use their car for longer tours whereas Table 7a shows the opposite trend for those aged under 65. Schmöcker et al. (2007) examined the mode choice of older people in London. They also found that living in Inner London does not necessarily lead to a reduction in the usage of private cars. A reason might be the increasing length of leisure trips as discussed in the literature review. The effects shown in Table 7 were also analyzed using the 30 min dwell time definition showing similar trends.

Ordered probit model

Following from Noland and Thomas (2007) we use a multivariate model that examines the complexity of travel patterns and the association between age while medical condition and controlling for socioeconomic factors and neighborhood characteristics.

The modeling approach is to estimate the number of stops in a tour (including zero stops) as a function of the various independent variables. As the stop data is a count, we specify an ordered probit model. This allows for ordinal differences in the dependent variable but does not assume cardinality between preferences (i.e., that the difference between 1 stop and 2 stops is not necessarily equivalent to that between 3 and 4 stops). The ordered probit model has the following general structure:
$$ y^{ * } = X\beta + \varepsilon $$
(1)
where y* is a latent variable measuring the number of stops or tours in our models. As an example, cut points can be defined as follows:
$$ y = \left\{ {\begin{array}{*{20}c} {\begin{array}{*{20}c} 1 & {\text{if}} & { - \infty \le y^{ * } \le \mu_{1} } \\ \end{array} } \hfill \\ {\begin{array}{*{20}c} 2 & {\text{if}} & {\mu_{1} \le y^{ * } \le \mu_{2} } \\ \end{array} } \hfill \\ {\begin{array}{*{20}c} {} & {} & {} & \vdots \\ \end{array} } \hfill \\ {\begin{array}{*{20}c} {} & {} & {} & \vdots \\ \end{array} } \hfill \\ {\begin{array}{*{20}c} m & {\text{if}} & {\mu_{\text{m - 1}} \le y^{ * } \le \infty } \\ \end{array} } \hfill \\ \end{array} } \right. $$
(2)
The μi are unknown parameters to be estimated and β is the partial change in y* with respect to X which means that for a unit change in X, y* is expected to change by β units, holding all other variables constant. Maximum-likelihood estimation is used to estimate the coefficients (\( \hat{\beta } \)) and the cut points (μi). The constant term is absorbed into the cut points. As expected, home-to-home tours have in general more stops than tours based on the 30 min dwell time definition. Nevertheless we adopted three cut points for both our models, firstly to allow better consistency and secondly as our results show that all cut points are significant even though the sample size for 0 stops for home-to-home tours (“walk the dog” trips) and category 3 + stops for the 30 min dwell time model are a small proportion of the sample (Table 8).
Table 8

Cut points in ordered probit models

 

Home-to-home tours

30 min dwell time tours

Cut 1 (μ1)

0+1 stops (73.8%)

0 stops (76.7%)

Cut 2 (μ2)

2 stops (18.1%)

1 stop (17.3%)

Cut 3 (μ3)

3 stops (5.7%)

2 stops (4.6%)

 

4+ stops (2.3%)

3+ stops (1.5%)

The predicted probabilities of the number of stops per tour m for given X and estimated coefficients \( \hat{\beta } \) is
$$ {\hat{\text{P}}\text{r}}\left( {y = m\left| X \right.} \right) = F\left( {\hat{\mu }_{\text{m}} - X\beta } \right) - F\left( {\hat{\mu }_{\text{m - 1}} - X\beta } \right) $$
(3)
whereas the parameter estimates \( \hat{\beta } \) give a good indication as to whether an independent variable has a positive or negative effect on the tour complexity, a direct interpretation of the parameter estimates on the increase in stop numbers is not possible, given the probit transformation of the dependent variable. Therefore calculation of the probabilities in Eq. (3) and the marginal changes compared to a reference case allows a better interpretation of the relative effectiveness of marginal changes in the independent variables. In the next section the results of the estimates \( \hat{\beta } \) as well as the marginal effects are discussed.1

Results

We report results both with tours defined as a home-to-home tour and with the 30 min dwell time definition (Table 9). About 7,650 home-to-home tours equate to 13,971 tours when the 30 min dwell time rule is applied. These tours are made by 5,835 individuals. For non-tour specific variables Table 9 further shows the percentage of individuals associated with each independent variable.
Table 9

Ordered Probit models for those aged over 65 (Italic numbers indicate significance at the 90% level, bold numbers at the 95% level)

 

Home–Home

30 min dwell time

30 min dwell time, h–h tours only

30 min dwell time, not h–h tours only

Coeff

z-stat

Coeff

z-stat

Coeff

z-stat

Coeff

z-stat

Age and medical condition

    Age 65–69 (33.5%)

Reference

Reference

Reference

Reference

    Age 70–74 (27.4%)

−0.04

−0.95

−0.02

−0.63

0.12

1.71

−0.05

−1.21

    Age 75–79 (20.9%)

−0.07

−1.68

−0.10

−2.83

−0.01

−0.16

−0.13

−2.86

    Age 80–84 (12.2%)

0.04

0.78

−0.04

−0.99

0.09

0.87

−0.07

−1.21

    Age 85+ (6.1%)

0.32

4.14

0.18

2.80

−0.04

−0.27

0.24

2.84

    Has no walking disability (81.0%)

Reference

Reference

Reference

Reference

    Has some walking disab. (16.3%)

0.04

1.02

0.02

0.59

0.02

0.25

0.02

0.46

    Impossible to walk (2.6%)

0.26

2.23

0.18

1.83

0.21

0.90

0.27

2.15

Socio-demographic variables

    Gender: male (47.2%)

0.16

4.77

0.06

2.20

−0.13

−0.06

0.06

1.68

    Race: white (89.6%)

0.22

3.84

0.25

5.11

0.33

2.86

0.27

4.46

    Possess mobile phone (14.0%)

0.16

3.83

0.08

2.26

0.14

1.69

0.08

1.87

Annual income and employment

    Not working (93.2%)

Reference

Reference

Reference

Reference

    Part time working (4.1%)

0.15

2.12

−0.002

0.03

0.05

−0.295

−0.01

0.177

    Full time working (2.7%)

0.13

1.45

−0.02

0.29

−0.14

0.61

−0.02

0.26

    <£5k (16.5%)

Reference

Reference

Reference

Reference

    £5–10k (23.1%)

0.18

3.50

0.09

1.95

0.21

2.13

0.07

1.22

    £10–15k (12.7%)

0.20

3.29

0.11

2.15

0.42

3.76

0.06

0.93

    £15–20k (7.5%)

0.22

3.18

0.15

2.51

0.22

1.64

0.15

2.05

    £20–25k (5.1%)

0.08

1.05

0.01

0.14

−0.06

−0.38

0.00

0.00

    £25–35k (4.4%)

0.30

3.73

0.13

1.94

0.15

0.92

0.14

1.74

    £35–50k (2.6%)

0.03

0.33

−0.01

−0.16

0.28

1.58

−0.06

−0.51

    £50–75k (1.4%)

0.13

1.06

0.06

0.60

0.23

1.00

0.04

0.34

    £75k+ (1.1%)

0.23

1.65

−0.03

−0.25

−0.17

−0.71

0.02

0.15

    Do not know (7.6%)

0.08

1.08

0.09

1.58

0.17

1.31

0.09

1.23

    Refused (18.0%)

0.09

1.64

0.03

0.59

0.15

1.38

0.02

0.27

Household structure

    Single person (39.6%)

Reference

Reference

Reference

Reference

    Married/cohabiting (53.2%)

−0.05

−1.23

−0.01

−0.20

0.00

0.05

0.00

−0.10

    With children (2.0%)

−0.04

−0.33

0.21

2.33

0.27

1.42

0.23

2.14

    Other (5.2%)

0.16

2.06

0.16

2.34

−0.11

−0.75

0.19

2.31

Car availability and freedom pass

    Has driving license (50.7%)

−0.06

−1.39

−0.01

−0.34

−0.08

−0.96

−0.01

−0.22

    Household owns a vehicle (56.1%)

0.01

0.29

0.02

0.50

0.03

0.38

0.02

0.46

    Has freedom pass (88.6%)

0.00

0.07

0.03

0.77

0.00

−0.04

0.03

0.61

Urban form/population density

    Pop. density less than 500 per square mile (0.5%)

0.40

2.06

0.33

2.07

0.08

0.24

0.39

2.08

    500–1,000 (1.8%)

0.21

1.85

0.20

2.07

0.23

1.09

0.21

1.82

    1,000–2,000 (1.4%)

0.10

0.78

0.14

1.26

0.22

0.88

0.14

1.06

    2,000–4,000 (2.9%)

0.20

1.90

−0.13

−1.54

−0.18

−0.87

−0.15

−1.44

    4,000–10,000 (26.8%)

0.09

1.59

0.06

1.19

0.29

2.86

0.02

0.38

    10,000–25,000 (54.2%)

0.11

2.26

0.07

1.60

0.19

2.05

0.06

1.06

    More than 25,000 (12.4%)

Reference

Reference

Reference

Reference

Tour specific

    Using car in tour

0.12

3.03

0.19

5.47

0.18

2.29

0.21

5.15

    Average link speed

0.001

2.50

0.000

0.66

0.020

6.74

0.001

1.30

    Home–home tour

Reference

    Other–other tour

1.79

33.44

Reference

    Other–home tour

2.16

59.20

0.35

6.85

    Home–other tour

2.15

58.92

0.34

6.55

Day of week of tour

    Monday

Reference

Reference

Reference

Reference

    Tuesday

−0.06

−1.23

0.10

2.47

−0.08

−0.96

0.10

2.17

    Wednesday

0.00

−0.06

0.08

1.96

0.08

0.84

0.10

2.14

    Thursday

0.00

0.04

−0.03

−0.74

−0.07

−0.72

−0.02

−0.47

    Friday

−0.04

−0.93

0.15

4.14

0.00

0.06

0.19

4.29

Cut points

    μ1

0.54

3.62

0.81

6.17

1.70

4.95

1.02

6.64

    μ2

1.32

8.81

0.45

3.40

1.12

3.29

1.70

11.02

    μ3

1.92

12.64

1.21

9.11

1.99

5.81

2.32

14.64

Observations/Model fit

    Number of observations

7,650

13,971

2,008

11,963

    Sample size (individuals)

5,835

    Log likelihood (intercept only)

−12,007.0

−19,845.7

−3,248.9

−10,079.6

    Log likelihood (final)

−11,828.0

−14,490.1

−3,124.1

−9,850.2

    Mc Fadden adjusted r2

0.015

0.270

0.038

0.023

In order to illustrate the effects of the tour beginning and end points we report three different 30 min dwell time models. The first one includes all tours, the second only those with start and end points at home and the final one includes only tours where either the beginning or the end point are not the respondents home. The first model is referred to as the “home-home” model and the latter three as “30 min” models. Several model specifications have been tested, in particular on the interaction between age and disability as well as income and employment. However, the simple models reported here without interactions generally gave the best model fit.

We find that the model fit of the home-to-home model is very poor. The pseudo R2 is only 0.015. With the 30 min threshold definition and including dummy variables explaining whether the tour is home–home, home–other or other–other gives a fairly acceptable pseudo R2 value of 0.27. The number of observations is much greater when the 30 min dwell time is used, but both models have ample sample size. From the t-statistics of the tour type variables in the 30 min models and the lower R2 for the 30 min model when only home-to-home tours are considered it becomes clear that tour type is a dominating factor in understanding tour complexity. Therefore it is also not surprising that the final two models with preselected tour types have similar explanatory power as the home–home model.

Clearly 30 min, home–home tours are the most complex tours as can be observed from the tour specific variables. These are tours where the traveler tries to accomplish all tasks of a tour in a relatively short amount of time. Our models suggest that once the traveler has stayed at one particular point for a longer time, then further longer stops are more likely or the traveler will return home without making numerous further stops. Our results further show that among tour types which do not have home as the origin and as the destination anchor the “other–other” tours are most complex. This suggests that in these cases there may be more than one major destination in the tour. Perhaps errands that require a shorter activity time are more often carried out in between the ‘other’ activities instead of at the beginning or the end of a home-to-home tour.

In initial analysis reported in Noland et al. (2007) we found that those who answered they have a “disability limiting their mobility” surprisingly tend to make slightly more complex tours. The reason might be that those with travel disabilities offset their physical impairments by better tour planning thus reducing the need for multiple tours. To better understand the importance of different disability types we tested variables such as seeing, hearing, walking disabilities and difficulties in orientation. Of these only walking disability has a statistically significant effect as reported in Table 9. Furthermore, those who reported “impossible to walk” make less complex tours and having “some walking disability” has no effect, or (if any) a slightly positive effect.

In the “30 min” models differences in age amongst our sample of older persons has slightly less importance. In the home-to-home model it is clearer that those aged in their late sixties make more complex tours, especially compared to those aged 85 or over. One explanation might be that for those aged over 85 as well as for anyone for whom walking becomes impossible, independent travel is reduced and possibly replaced with door-to-door trips with accompanying persons.

Our gender and ethnicity dummy variables also confirm findings reported in the literature review. Men make significantly less complex tours compared to women and non-minorities (whites) are associated with more complex tours. Evaluation of the marginal probabilities suggests that the likelihood of home-to-home tours being more complex (more than 1 intermediate stop, i.e. 4 or 5 trips) decreases by 23.9% for men and 30.2% for non-whites. It could be that ethnic minorities have more of their usual trip destinations in the neighborhood therefore reducing the need to chain trips. However, Schmöcker et al. (2005) also report, using the same dataset, less frequent total trips for ethnic minorities. This does not support the above hypothesis but rather suggests that there is in total less travel demand.

We further included mobile phone possession as an explanatory variable in our model. Interestingly this has a significant positive effect on tour complexity in particular on home-to-home tours. We conclude therefore that the scheduling flexibility gained through mobile phone possession is no longer just enjoyed by younger people but also by those aged over 65. This finding adds to the growing body of literature on the relationship between telecommunications and travel demand. Choo and Mokhtarian (2007) for example show with US time series data that total travel demand and telecommunications have a complementary positive effect on each other. Our results add to this that this increase might be associated with increased tour complexity. As only 14% of our 2001 sample possess a mobile phone one might further expect the percentage to rise over time.

Our income variables indicate that a very low income is associated with less tour complexity, but differences between some middle and upper income groups is low. This finding confirms US (Noland et al. 2007) and Australian (Golob and Hensher 2007) results showing that income does not have a very significant impact. In these results, the pattern is not clear but suggests that amongst older people, middle income groups (£10-35 K) have the most complexity in their tours. As the difference between low and high income groups does not appear to be as significant in the 30 min models, one could speculate that middle income groups combine their complex (shopping) tours more often with other activities (such as dining out) that last longer than 30 min. We further include employment as a separate explanatory variable. Though there is as expected some correlation with some income groups, interaction with income levels does not lead to different results nor improve the model fit. We find that part time employment is positively associated with stop numbers but not full time employment. This result is intuitive as part time workers appear to use their shorter working hours for additional trips. Interestingly part-time work is not significant in the 30 min models suggesting that the additional trips have rather longer dwell times (or perhaps they make their stops on the way to work which occupies less of their day).

Possession of a driving license and having a car in the household has surprisingly no positive effect on trips per tour. An explanation might be that there is less need to chain trips as the car is available to make more frequent trips. In previous research we found that having a car increases trip frequency. The inclusion of tour specific variables in this analysis shows that those tours made with a car are more complex, distinct from ownership per se. Figure 2 also illustrates the growing importance of the car and the reduced share of public transport for more complex tours. Interestingly the percentage of tours made by car does not decrease but stays relatively constant regardless of the stops per tour but only the combination of drive + walk increases with increased stops.
https://static-content.springer.com/image/art%3A10.1007%2Fs11116-009-9222-z/MediaObjects/11116_2009_9222_Fig2_HTML.gif
Fig. 2

Tour complexity (home-to-home) and modes chosen

We also include a dummy variable for whether the individual is a “freedom pass holder”, which allows for free public transport within Greater London after 9.30 a.m. All London citizens older than 60 are entitled to such a pass but not everyone applies for it. Previous results in Schmöcker et al. (2005) showed that those in the possession of such a pass, made fewer total trips, probably suggesting some endogenous effects. There is no statistically significant association with tour complexity for those holding a freedom pass.

One might expect that larger households in general make more complex tours in order to meet the various travel demands of the household members. Our results do not necessarily confirm this. Only 30 min tours are more complex for those living with children which we expect to be grown-up children in our sample. Those who answered household structure as “other”, i.e., neither living alone, nor with a partner or with children, make significantly less complex tours. Though further information is missing one might suspect that these might be people living in care homes in which most of their shopping needs are catered for.

Noland and Thomas (2007) found that decreases in population density are associated with more complex tours using data for the entire US. In London this effect can only be confirmed for those living in the least densely populated areas of London and effects are larger when the home-to-home tour definition is applied. An explanation for the lower significance of the population density data might be that even the less dense neighborhoods of London are not comparable to rural areas or exurban areas in the US. Further the number of respondents living in low density areas is too small to draw definite conclusions.

Our analysis provide more significant results comparing areas with high (10–25 k people per square mile) and extremely high density (> 25 k people per square mile). The effect of population reverses here. The reason might be that in the medium to high density areas a good spread of shopping facilities exists, for example in the form of local shopping streets, leading people to chain trips. Older people living in Central London will have more difficulties accessing such facilities or possibly drive to multi-store complexes at which all demand can be satisfied at a single location also reducing trips per tour.

We further include a variable to control for the day of the week on which the tour is taken. The 30 min models for all tours show that the most complex tours are generally made on Mondays. In particular older people seem to avoid complex tours on Fridays. When we look however at the model with the home-to-home tour definition we cannot confirm these results. An explanation might be that on Monday’s older people do more of their shopping trips with frequent (short) stops. These tours might be offset with less frequent other complex tours so that the overall effect if one looks at home-to-home tours is not significant.

Finally we include “Average link speed” defined as travel time in minutes divided by tour-km. We find a difference depending on our tour definition: With the home-to-home tour definition we find that travel time (and perhaps congestion) is indeed influencing older people’s travel behavior and increases tour complexity. However, only looking at tours with 30 min dwell times this association is not found. In the final “30 min + HH” model we find a negative significance. The marginal effects suggest that a 10% increase in travel time leads to a 12% reduction in the probability of the tour having more than one destination (on home-to-home tours with less than 30 min dwell time at each stop). This shows that on home-to-home tours with only short stops (e.g. shopping in the neighborhood) older travelers consider congestion more in their destination choices than on tours with other starting or ending points. The result appears plausible but requires further investigation.

Conclusions

This paper analyzed trip chaining behavior and possible explanatory variables. Our models are based on two different definitions of what constitutes a tour. The analysis has shown that the definition can lead to different results and consequent interpretation. The home-to-home definition is more widely used in the literature as it probably more easily explains travel behavior and is more consistent with research on activity chains. The “30 min dwell time” definition was adopted by the US Federal Highway Administration to distinguish major stops or primary trip purposes from stops en-route. Our analysis suggests that the “30 min” models provide a better fit to the data for examining tour complexity, at least for this dataset as the anchor points of the origin and destination are an important explanatory variable to understand tour complexity. Home-to-home tours are more complex than tours with other destinations or starting points, an effect that remains hidden with home-to-home models. Our analysis with the two definitions further reveals that increasing age is associated with a reduction in tour complexity for the (total) home-to-home tour but that age has little association with making stops en-route to the main tour destination. Our cross-tabulations of number of trips and total distance also reveal more interesting effects with the 30 min dwell time rule. This demonstrates some of the benefits of basing analysis on the 30 min dwell time rule, as it essentially helps to define more realistic break points in what is a complex tour between multiple primary destinations that could be part of a home-to-home tour.

Further splitting up 30 min tours by distinguishing whether the tour anchor points are at home or not suggests that once people make one longer stop, additional long stops are more likely. Together with our general finding that 30 min tours are more complex for older people this clearly has implications for land-use planning and shopping streets: If it is possible to provide resting facilities for older people they are more likely to add additional trips to their tour.

Only for those with severe mobility problems does trip complexity decrease. This finding also confirms the analyses of other data sets such as the Sydney Travel Survey (Golob and Hensher 2007). The literature suggests that the retired “younger-old” are often at least as mobile as those younger than 65, expressed in terms of number of trips, trip distance and, as our analysis shows, also in terms of tour complexity. One policy implication is that efforts to increase mobility should be focused on those with severe mobility problems. In our ageing society the increasing number of those with some (slight) mobility problems might not require much support but a minority might require more support than currently envisaged.

Previous findings emphasized the future changes in travel patterns when we expect more elderly to have a driving license and being able to afford cars. Our results add another dimension to these findings: Having a car has no association with trip complexity but using the car is associated with more complexity. We further provide evidence that tour complexity will increase in the future as mobile phone use increases. It appears that the flexibility to adjust tours after one has already left home is enjoyed by people of ages. As complex tours are more likely made by private car this might be a concern for transport planners.

Finally, we find that the day of the week influences trip chaining patterns. This is an important finding for transport policy as there appears to be a weekly pattern in trip complexity. Mondays, immediately following the weekend appear to have more complex tours with short-dwell time trips, possibly shopping or other errands sometimes referred to as personal business trips. This type of information could be useful for providing mobility services and clearly requires more investigation.

Footnotes
1

In the interpretation of results we found that the marginal effects do not add much to our understanding. We note differences in the discussion and for brevity, we therefore omit the (rather large) table of marginal results. This is available on request from the authors.

 

Copyright information

© Springer Science+Business Media, LLC. 2009