Are Telecommuters’ and Non-Telecommuters’ Daily Time-Use Behaviors Different? An Episode-Level Model for Non-Mandatory Activities

Abstract

Telecommuters (individuals working from home), and non-telecommuters (individuals working at the workplace) might have significantly different daily time-use patterns. For example, telecommuters might be able to engage in more episodes of non-mandatory activities such as recreation and pick-up/drop-off of children, as many telecommuters have a relatively flexible work arrangement in relation to their work location and schedule. This study aims to explore the non-mandatory activity engagement and duration decisions of these two worker profiles at a disaggregated-episode level. To do so, a multiple discrete continuous extreme value with ordered preferences (MDCEV-OP) model is adopted, using the 2018 travel survey data from the Central Okanagan region of British Columbia, Canada. This model accounts for multiple occurrences of each activity type (i.e., episodes) along with their corresponding durations. In doing so, the model adopts a logical ordering ensuring that the jth episode is not predicted without the occurrence of the (j–1)^th episode of an activity type. The model results reveal that telecommuters are more likely to engage in higher episodes of non-mandatory activities, compared with non-telecommuters. For instance, female teleworkers are found to be more likely to participate in higher episodes of activities associated with household-related responsibilities such as shopping. In contrast, female non-telecommuters are associated with participating in more episodes of personal business-related activities. The findings of this study provide important behavioral insights into the activity-time-use patterns of telecommuters and non-telecommuters, which can be utilized to develop more effective and equitable travel demand management plans, policies, and models.

Keywords

telecommuters non-telecommuters activity time-use non-mandatory activities multiple discrete-continuous activity-episodes

Telecommuting—working remotely—has been considered as a potential travel demand management strategy to reduce peak-hour commute travel, roadway congestion, and emissions, among other things ( 1 ). Although discussions around telecommuting is not new, the COVID-19 pandemic brought strengthened interest into the telecommuting subject area as more and more people experienced the benefits of some form of telecommuting as an alternative to standard working arrangements during the pandemic ( 2 ). For example, telecommuting provides individuals more time each day, because of the elimination of commute travel times, and flexible working hours, compared with non-telecommuters (i.e., individuals working at the workplace). This could potentially result in significant differences in the 24 h activity patterns of telecommuters and non-telecommuters. For instance, the savings in commute travel time allow telecommuters to allocate more time to non-mandatory activities. For example, telecommuting was found to be associated with higher participation in non-mandatory activities ( 3 ). This could be triggered by their willingness to “go out” to compensate for increased sedentary tendencies that come with telecommuting ( 4 ). On the other hand, non-telecommuters are likely to engage in non-mandatory activities after work but at a lower frequency and possibly of higher durations ( 5 ). There could be differences in daily time-use behavior between telecommuters and non-telecommuters, which has not been captured to significant extents in current literature. This investigation of daily time-use behavior has two dimensions: (1) what type of activity to participate in? and (2) for how long? Several attempts at modeling have been made to explore activity engagement and duration choices independently as well as jointly ( 6 – 9 ). However, the majority of studies deal with these choices at an aggregated activity level (i.e., by aggregating time allocation of all episodes of an activity type in a day to single activity level) and not a disaggregated-episode level. An “episode” refers to the multiple occurrences of each activity type. For example, an individual allocating non-zero time to shopping activity twice in a day leads to engaging in multiple episode (i.e., two episodes) of that activity type. These time allocations to different episodes can be of different durations, performed at different times of day, and use different travel modes, and might take place at destinations. To account for such granular-level activity-time-use patterns, recent studies have exploited discrete-continuous joint structures ( 10 , 11 ). However, there exists a research gap, particularly answering the following research question: How do the activity engagement patterns, including duration and frequency of non-mandatory activities, at the episode-level differ between telecommuters and non-telecommuters?

The objective of this study is to investigate the daily activity-time-use behavior of telecommuters and non-telecommuters. More specifically, exploring the activity engagement and duration choices at a disaggregated-episode level that accounts for the multiple occurrences of each activity type. Data comes from the 2018 Okanagan Travel Survey (OTS) conducted in the Central Okanagan region of British Columbia, Canada. To account for episode-level daily time-use behavior, this study adopts a recently developed multiple discrete-continuous extreme value with ordered preferences (MDCEV-OP) model. The MDCEV-OP model is a newer extension of traditional MDCEV models, accommodating a logical occurrence of episodes which ensures that higher-numbered episodes are not predicted with non-zero time allocations without allocating certain time to lower-numbered episodes (e.g., the jth episode of an activity type is not predicted without the occurrence of the (j–1)^th episode of that activity type) ( 12 ). This episode-level granularity of the framework provides more detailed disaggregated results that adds the capability to analyze individuals’ travel behavior, such as activity destination and mode choice which likely varies by each activity episode. Further, the effects of various socio-demographic, mobility tool ownership, and built environment attributes, on an individual's activity participation and duration choices, are tested in the study. Lastly, this study has been conducted as an activity generation component of an activity-based travel demand model, currently under development at The University of British Columbia—Okanagan.

Literature Review

Telecommuting is a flexible working arrangement which allows workers to work remotely, either from home (the most common type of telecommuting) or another location that is not the actual workplace (ex. co-working space), through the use of information and communication technology ( 13 ). Although telecommuting is often seen as a means of lessening peak-hour congestion and emissions because of the reduction in travel for mandatory activities like work (1), it might induce an increase in non-mandatory activities such as personal business and recreation activities. For instance, telecommuters do not have to commute, providing more time to be allotted to other activities. Telecommuters could generate more trips since they are unable to perform non-mandatory activities on commuting trips. Furthermore, since telecommuters are likely to be staying at home for the most of the day, when they go out they could be using vehicles and for longer distances resulting in increased congestion and emissions. This indicates a need to understand telecommuters’ daily time-use behavior, particularly for non-mandatory activities, and how they are different than non-telecommuters.

Past literature has largely focused on investigating the frequency of non-mandatory activity participation of telecommuters. For example, Zhu developed a two-stage least square model considering socio-demographic and transportation attributes to investigate telecommuters’ daily total trips, work trips, and non-work trips. They reaffirm that telecommuting has a complementary effect on non-mandatory trips, implying telecommuters have longer and frequently higher activity engagement rates compared with non-telecommuters ( 14 ). In a similar line of investigation, Asgari and Jin adopted a structural equation modeling system revealing causal relationships between non-mandatory activity participation and telecommuting, considering various demographic variables. They found that increased total durations of non-mandatory activities increased with the likelihood of telecommuting ( 15 ). To account for frequency of non-mandatory activity engagement for telecommuters, a Poisson model was adopted. This study supported the theory that telecommuting is associated with more non-mandatory trip generation, regardless of socio-demographic variables, commute distance, and other variables ( 3 ). In addition, Kim et al. investigated whether telecommuting had a substituting or complementary effect on various types of household travel, utilizing unrelated censored regression models. They found that telecommuting had a complementary effect on households’ vehicle travel ( 4 ). These studies all successfully found that telecommuting produces increased demands of out-of-home (OH) non-mandatory activities focusing on just one aspect, either duration, trip count, or vehicle distance travel.

One must note that telecommuters and non-telecommuters are likely to depict significantly different activity engagement rates and associated temporal allocations. Therefore, past studies aimed to capture such temporal interactions for the aforementioned two worker profiles by adopting joint discrete-continuous structures to replicate activity engagement and duration decisions simultaneously ( 16 , 17 ). A key illustration of this difference is that non-telecommuters have the opportunity to perform non-mandatory activities during their commutes (conceivably while returning home from work), an opportunity not available to telecommuters. To reaffirm this, Srinivasan and Bhat ran a multiple component model system which included a seemingly unrelated regression model and a joint mixed logit hazard-duration model to investigate factors affecting the activity generation of four different types of household structures ( 18 ). They revealed that non-telecommuters who leave work before 4 p.m. have longer shopping durations then those who leave later ( 18 ). Similarly, Garikapati et al. adopted a multiple-discrete continuous framework to model activity participation and time allocation for daily commuters primarily making home-based work tours. The study reveals a negative effect of higher commute durations on escorting and maintenance activities. However, female workers made more non-mandatory detours, maybe because of their household roles/responsibilities ( 19 ). In the similar line of investigation, Rajagopalan et al. utilized a multiple discrete-continuous nested extreme value (MDCNEV) model to examine non-mandatory activity-timing behavior of non-telecommuters, and found that, for individuals and couples, the most preferred time to perform OH non-mandatory activities was after arriving back home from work ( 5 ).

On the other hand, past literature also attempts to shed light on 24 h time-use patterns of telecommuters. A major confounding factor associated with non-mandatory activity participation for telecommuters is the desire to “get out” and avoid “cabin fever,” as they are often confined to a single location ( 4 ). To illustrate such differences, Pendyala et al., utilized two waves of travel diaries from California, U.S., reaffirming that telecommuters tend to spread their participation in non-mandatory activities over the day, avoiding sedentary lifestyles associated with telecommuting ( 20 ). Furthermore, Srinivasan and Bhat developed an MDC structure to investigate nuclear family couples’ solo- and joint-discretionary activity participation and duration considering socio-demographics, household characteristics, mandatory activity participation, schedule variables (day-of-week and season-of-year), and built environment characteristics. They found that individuals working from home for long durations are less likely to participate in longer durations of OH non-mandatory activities ( 8 ). Similarly, Paleti and Vukovic developed a joint count and MDCEV model to investigate the choice of telecommuting in multi-worker households, and its effect on daily time-use behaviors of household members. They found individuals who telecommute frequently were less likely to participate in eat-out and socializing activities ( 21 ).

Methodologically, activity-time-use choices such as what type of activity to engage in (discrete), and for how long (continuous), have been modeled simultaneously in past literature ( 5 , 21 , 22 ). However, a major limitation of these studies that are based on discrete-continuous structures is the aggregation of time for each activity type over a specific period rather than accounting for time allocation at an episode-level (i.e., multiple occurrences of each activity type). For example, if an individual goes shopping for 3 h, then goes to dinner for 1 h, then performs another 2 h of shopping, conventional MDCEV models will combine the two shopping episodes into one 5 h episode, thus not revealing the multiple occurrences of the shopping activity. Aggregating such activity episodes not only ignores the multiple occurrences of each activity type but makes it difficult to accommodate the effect of other travel-related choices, such as mode or destination location, that may differ by each episode of an activity on a person’s daily time-use patterns. To accommodate this, Palma et al. and Saxena et al. proposed an extended version of the MDCEV model that deals with activity participation and duration choices simultaneously at an episode-level while accounting for multiple occurrences of each activity type ( 10 , 12 ). Further, Khaddar et al. adopted the MDCEV-OP framework prosed by Saxena et al. to investigate time use and land use interactions at an episode level ( 11 ). However, dealing with the episode-level approach brings a challenge of logical consistency ensuring that a higher-numbered episode does not occur without occurrence of a lower-numbered one. For instance, consider an individual taking part in two episodes of shopping activity in a day (i.e., episode 1 and episode 2). From a prediction standpoint, it would be irrational if episode 2 is predicted with non-zero time allocation and episode 1 is predicted with zero time allocations. Therefore, models must prevent such illogical possibilities. To address this, Palma et al. allow the baseline marginal utility’s deterministic component of the higher-numbered episode to be lower than that of the lower-numbered episode by introducing a negative penalty term ( 10 ). However, this approach does not ensure that higher-numbered episodes are not predicted without the occurrence of the lower-numbered one. On the other hand, Saxena’s approach introduces a non-increasing order on the baseline marginal utility parameters within the MDCEV formulation, thus ensuring that irrational episodes (e.g., the second episode is predicted without occurrence of the first episode) are not predicted.

Contribution of This Study

This study aims to investigate the variation in time-use patterns of the two worker profiles over a 24 h period. More specifically, a newly extended MDCEV model is adopted which deals with activity participation and duration choices at a disaggregated-episode level ( 11 ). In other words, this model accounts for multiple occurrences of each activity type which has been ignored in the past literature ( 23 ). The extended MDCEV model imposes a decreasing order across the baseline parameters to ensure that higher-numbered episodes are not predicted without the occurrence of lower ones (e.g., the jth episode of an activity type is not predicted without the occurrence of the (j–1)^th episode of that activity type). Utilizing this model, the study sheds light on the variation in activity engagement and duration allocation of telecommuters and non-telecommuters. Some of the questions this study aims to answer are as follows: Do telecommuters tend to participate in non-mandatory activities at a higher frequency than non-telecommuters? How does the participation vary at the episode level? Which activity type are telecommuters and non-telecommuters allocating majority of their time to? Furthermore, this study extensively tests the effect of socio-demographic, vehicle ownership, and built environment attributes on activity-time-use patterns of the aforementioned two worker profiles.

Data Sources

This study utilizes data from the 2018 OTS which collected 24 h travel logs from residents of the Central Okanagan region of British Columbia, Canada. Travel pattern data was collected for the most recent weekday for each individual in a household who were 5 years of age and above. This includes information about origin and destination locations, arrival and departure time, travel purpose, companionship, mode, and others. Aside from travel attributes, the survey also collected demographic information at individual and household levels. Individual-level information includes age, gender, occupation, employment status, and driver license ownership, among others. Household-level information such as vehicle ownership, bike ownership, household size configuration, dwelling type, and annual household income was collected. This study utilizes land use and transportation network data from open data portals of different cities in the Central Okanagan region, 2016 Canadian Census for neighborhood information, and point of interest data from Desktop Mapping Technologies Inc. (DMTI). A comprehensive discussion of the survey, comparison with the Canadian Census, and descriptive statistics can be found in the Smart Trips Okanagan Travel Survey ( 24 ).

Data Preparation

One of the first steps toward data preparation was considering working individuals based on their employment status (i.e., full-time workers, part-time workers, or both). Further individuals were disaggregated based on their response toward type of work location. This question includes options such as “work from home,”“no fixed workplace/no usual place of work,” and “work away from home.” Respondents working from home with no work-related trips are considered as “telecommuters,” whereas individuals working at an office/in-person setting and making work-related trips were considered as “non-telecommuters.” Further, the activity-time allocation was categorized as in-home (IH) and OH activities, where OH activities were disaggregated into the following non-mandatory activities: escorting (pick-up/drop-off), personal business (visiting bank, visiting doctor, etc.), recreation (sports, swimming, etc.), eat-out (dine-in restaurant), social (meeting friends/relatives, social/civic events, etc.), and shopping (purchasing groceries, supplies, etc.) and mandatory activities: education and work. The final sample size includes 1,684 non-telecommuters and 369 telecommuters. Later, this OH activity information was defined at an episode level in the decreasing order of their temporal allocation and not in the chronology of their occurrence ( 11 ). The maximum number of episodes considered for each activity type was three, since no/very few individuals were observed to have undertaken more than three episodes for an activity in a day. Apart from OTS, supplementary data sources were explored to extract built environment measures such as length of active transportation network, land use attributes (including share of residential, commercial, industrial, and other areas), and accessibility measures (including distance to the nearest urban center, bus stop, and others). The built environment and land use attributes were generated for 250 m buffer areas based on the respondent’s home location, whereas accessibility measures were generated based on the road-network-based distance from their home location. This built environment, land use, and accessibility measures information were extracted using ArcGIS (v.10.6.1).

Descriptive statistics of activity participation and time allocation for telecommuters and non-telecommuters are presented in Table 1. One must note that 100% engagement is observed for IH activity, and average time allocation is calculated only for those individuals who engaged in that activity type. In relation to activity participation rates, telecommuters engage more in personal business, recreation, social, and shopping activities compared with non-telecommuters. However, results were contrary for escorting and eat-out activities. Interestingly, more telecommuters engage in higher episodes of activities in comparison with non-telecommuters. For instance, 9.21% of telecommuters participated in a second episode of shopping activity, whereas the share of non-telecommuters was only 5.46%. On the contrary, a higher percentage of non-telecommuters engage in a second episode of eat-out activity. In relation to activity duration, telecommuters depicted higher time allocation to different activities compared with non-telecommuters. For example, telecommuters spend an average of 64.3 min on shopping activity, which is 28 min higher than that of non-telecommuters. For personal business, telecommuters’ average duration was 138.5 min, whereas for non-telecommuters this was 56.7 min (almost half). Similarly, at an episode level, telecommuters are spending higher durations in all non-mandatory activities compared with non-telecommuters. As expected, escorting has the lowest share of time allocation at episode and activity levels for both worker profiles.

Table 1.

Summary of Activity Participation and Time Allocation of the 2018 Okanagan Travel Survey Sample

	Activity-level participation	Episode-level participation (% [number] of individuals participating in an episode)			Average time allocation (minutes)
Activity purpose	% (number) of individuals participating an activity	1 or more	2 or more	3 or more	Activity level	Episode level
Telecommuters
Escorting	26.56 (98)	26.56 (98)	15.45 (57)	2.71 (10)	33.4	5.6
Personal business	31.71 (117)	31.71 (117)	6.78 (25)	1.63 (6)	138.5	22.9
Recreation	26.29 (97)	26.29 (97)	2.71 (10)	0.27 (1)	106.0	55.7
Eat-out	10.84 (40)	10.84 (40)	0.54 (2)	0.27 (1)	70.4	10.5
Social	19.51 (72)	19.51 (72)	2.44 (9)	0.27 (1)	214.6	48.6
Shopping	39.84 (147)	39.84 (147)	9.21 (34)	1.90 (7)	64.3	17.6
Non-telecommuters
Escorting	31.06 (461)	31.06 (461)	14.69 (218)	1.82 (27)	17.0	4.3
Personal business	22.44 (333)	22.44 (333)	4.65 (69)	0.81 (12)	56.7	12.6
Recreation	18.67 (277)	18.67 (277)	1.15 (17)	0.27 (4)	99.2	25.7
Eat-out	22.64 (336)	22.64 (336)	2.09 (31)	0.20 (3)	51.9	12.6
Social	18.19 (270)	18.19 (270)	2.22 (33)	0.27 (4)	159.0	28.0
Shopping	31.27 (464)	31.27 (464)	5.46 (81)	0.47 (7)	36.3	13.4

Methodology

This study aims to explore the activity-time-use behavior of telecommuters and non-telecommuters. More specifically, this study investigates the activity participation (i.e., discrete choice) at the disaggregated-episode level and corresponding durations (i.e., continuous dimension) while accounting for multiple occurrences of each activity type. This newly proposed model assumes that individuals allocate certain times to different episodes of different activities such that the utility derived by engaging in the various activities and corresponding episodes is maximized. However, this is subject to a budget constraint of 1,440 min (i.e., total time available in a day). An additively separable non-linear utility is utilized in this study which can be formulated as ( 12 , 25 ):

U (t_{p}) = ψ_{p 1} \ln (t_{p 1}) + \sum_{k = 2}^{K} \sum_{j_{k} = 1}^{J_{k}} ψ_{pk, j_{k}} γ_{pk} \ln (\frac{t_{pk, j_{k}}}{γ_{pk}} + 1)

(1)

where

$U (t_{p})$ = the maximum utility individual p tries to acquire by allocating time to each of the $j_{k}$ episodes of the K activities,

$t_{p}$ = the time allocations vector $t_{pk, j_{k}} \forall j_{k} \in J_{k}$ and $k \in K$ , where $t_{pk, j_{k}}$ is the time alloted by individual p to $j_{k}^{th}$ episode of the k^th activity,

$ψ_{p 1}$ = the baseline marginal utility for time alloted to the outside good which is at the aggregate level,

$ψ_{pk, j_{k}}$ = the baseline marginal utility of allocating time to $j_{k}^{th}$ episode of the kth activity controlling discrete and continious decisions of whether to assign time to the corresponding episode of an activity and the amount of time, and

$γ_{pk}$ = the satiation parameter controlling for time allocation to an episode of an activity.

Equation 1 is valid if $ψ_{i 1} > 0, ψ_{ik, j_{k}} > 0,$ and $γ_{ik} > 0 \forall k \in K .$

In addition, this study considers IH activity duration, duration of work/education, and travel duration as outside goods. Here, in this study, the satiation parameter is presumed to be similar across different episodes of an activity type. Further, the baseline parameter and satiation parameter utility can be represented as:

ψ_{pk, j_{k}} = \exp (β' X_{pk j_{k}} + ε_{pk j_{k}})

(2)

γ_{pk} = \exp (ρ' Z_{pk})

(3)

where

$X_{pk j_{k}}$ = the vector of baseline marginal utility parameter,

$Z_{pk}$ = the vector of satiation parameter,

$β'$ = corresponding estimable coefficient,

$ρ'$ = corresponding estimable coefficient, and

$ε_{pk j_{k}}$ = a random error term identifying unobserved factors that influence a person’s choices.

It must be noted that baseline marginal utilities are parameterzied as a function of observed and unobserved varaibles. As this framework deals with episode-level choices, it is important to label epsiodes of an activity, which clearly differentiates them from one another. One of the approaches could be labeling them based on time-of-day or location. However, this does not determine the number of occureneces of each activity type that might occur during that time-of-day or location. Another approach could be labeling them based on the duration without predicting their chronology of occurrence. In other words, the episode with the longest duration is considered as the first episode, the second-longest as the second, and so forth. However, a person can engage in a lower duration of an episode early in the day, and might engage in higher durations later in the day, but this model is not agnostic to such scheduling decisions and just deals with the time-use decisions at an disaggregate level. To do so, a logical ordering is deployed ensuring that lower-numbered episodes occurs first. For example, the second epsiode of an activity is not predicted without the occurrence of the first episode of that activity. This model imposes a decreasing baseline preference across episodes of an activity, while defining episodes in the non-decreasing order of their observed duration:

ψ_{pk, j_{k}} \geq ψ_{pk, (j_{k} + 1)} \forall k \in {2, \dots . K}, j_{k} \in {1, 2, 3 \dots . J_{k} - 1}

(4)

The above condition imposes that the distribution of baseline marginal utility parameters accounting for each epsidoe of an activity is right-truncated by that of the previous episode. This accomodates the logical ordering confirming that higher-numbered episodes are never choosen without the lower-numbered episodes. While it is difficult to specify such truncations using the assumed stochastic distributions, to ensure the conditions in Equation 4, the likelihood function of the traditional MDCEV model (that is defined at an episode level) is conditioned on the set of conditions in Equation 4. Keeping this discussion in mind, the likelihood function of the traditional MDCEV defined at a disaggregated-episode level is conditioned on the set of constraints mentioned in Equation 4. The resulting conditional likelihood in a closed form can be expressed as:

\begin{matrix} L {t_{1}^{*}, \dots, (t_{m, 1}^{*}, t_{m, 2}^{*}, \dots t_{m, I_{m}}^{*}), . ., (t_{M, 1}^{*}, t_{M, 2}^{*}, \dots t_{M, I_{M}}^{*}), 0, . . 0} \\ = | J | μ^{δ - 1} (δ - 1)! x \frac{Π_{m = 1}^{M} Π_{i_{m = 1}}^{I_{m}} \exp (μ V_{m, i_{m}})}{Π_{k = 1}^{M} Π_{j_{k} = 1}^{I_{k}} (\frac{e^{μ {\bar{V}}_{k, j_{k}}}}{\sum_{s = j_{k}}^{J_{k}} e^{μ {\bar{V}}_{k, s}}}) * {(\sum_{k = 1}^{K} \sum_{j_{k} = 1}^{J_{k}} \exp (μ V_{k, j_{k}}))}^{δ}} \end{matrix}

(5)

where

$t_{1}^{*}$ = the time alloted to the outside good of an activity,

$t_{m, I_{m}}^{*}$ = the time alloted to each episode of an activity,

$m$ = the index for the chosen activity,

$M$ is the index for the set of activity,

$I_{m}$ = the number of episodes of a activity m,

$i_{m}$ = the episode number for that activity,

δ = the total number of chosen episodes across all activities,

$V_{k, j_{k}} = β_{k, j_{k}}^{'} z_{k, j_{k}} - \ln (\frac{t_{k, j_{k}}^{*}}{γ_{k}} + 1)$ , and

$| J |$ = the determinant of the Jacobian matrix, which is of a similar structure as derived by Bhat et al., except defined at a disaggregated-episode level ( 26 ):

| J | = [Π_{m = 1}^{M} Π_{i_{m} = 1}^{I_{m}} (\frac{1}{t_{m, i_{m}}^{*} + γ_{m}})] [\sum_{m = 1}^{M} \sum_{i_{m} = 1}^{I_{m}} (t_{m, i_{m}}^{*} + γ_{m})]

(6)

The final likelihood equation is very similar to traditional disaggregated MDCEV models, scaled by the rank-ordered logit terms of the baseline marginal utilities of the corresponding episodes being in the decreasing order for all chosen activities. This framework also estimates a scale parameter as suggested by Bhat et al. ( 27 ). This model collapses to a traditional MDCEV when observed time allocations across all activities occur in a single episode. For more details of the likelihood derivation refer to Saxena et al. ( 12 ).

Results

The goodness-of-fit measures considered in this study are log-likelihood at convergence, Akaike information criterion (AIC), and Bayesian information criterion (BIC) values (Table 2). A comparison was made based on goodness-of-fit measures between two empirical models: (1) the proposed MDCEV-OP model and (2) the MDCEV model considering alternatives at a disaggregated level. Results clearly reveal that the MDCEV-OP model provides a superior statistical fit to the MDCEV model (Table 2). This study estimates three separate models for telecommuters, non-telecommuters, and for all workers, while accounting for telecommuting as a dummy variable, to empirically demonstrate the variation in time-use patterns of the different worker profiles. The goodness-of-fit measures clearly indicate that separate models must be adopted to address the differences in the 24 h time patterns, as separate models for telecommuters and non-telecommuters outperforms the all-sample model (Table 2). For brevity, further discussion is based on the separate MDCEV-OP models estimated for aforementioned worker profiles (Table 3).

Table 2.

Goodness-of-Fit Measures

	Telecommuters	Non-telecommuters	All sample
Multiple discrete continuous extreme value with ordered preferences model
N	369	1,484	1,853
Log-likelihood at convergence	−5,180.76	−17,466.7	−20,927.30
AIC	10,447.52	35,035.36	41,956.60
BIC	10,615.68	35,305.79	42,283.35
Disaggregated multiple discrete continuous extreme value model
N	369	1,484	1,853
Log-likelihood at convergence	−5,560.32	−18,804.70	−22,147.10
AIC	11,206.64	37,711.40	44,396.20
BIC	11,373.98	37,987.97	44,677.95

Note: AIC = Akaike information criterion; BIC = Bayesian information criterion.

Table 3.

Parameter Estimation Results of the Multiple Discrete Continuous Extreme Value with Ordered Preferences (MDCEV-OP) Model for Non-Mandatory Activity Participation and Duration of Telecommuters and Non-Telecommuters

	Activity level	Second episode	Third episode	Activity level	Second episode	Third episode
	Baseline preference parameters
	Coefficient (t-stat)
	Telecommuters			Non-telecommuters
Baseline preference constants
Escorting	−8.36 (−37.08)	-	-	−8.18 (−64.45)	1.02 (7.01)	-
Personal business	−8.74 (−33.58)	-	-	−7.93 (−95.37)	-	-
Recreation	−8.49 (−26.01)	-	-	−7.90 (−77.36)	-	-
Eat-out	−8.79 (−34.06)	-	-	−7.91 (−110.80)	-	-
Social	−8.11 (−74.67)	-	-	−7.89 (−150.21)	-	-
Shopping	−8.62 (−35.80)	-	-	−7.87 (−112.11)	-	-
Gender: Female
Escorting	-	-	-	0.04 (1.41)	-	0.05 (1.00)
Personal business	-	0.38 (1.93)	-	-	-	0.12 (2.50)
Recreation	-	0.75 (2.65)	-	0.04 (1.26)	-	-
Eat-out	−0.23 (−1.35)	-	-	-	-	-
Social	-	-	-	-	-	-
Shopping	-	-	0.63 (2.98)	-	0.11 (2.60)	-
Age
Escorting	−0.006 (−1.76)	-	-	-	−0.008 (−5.48)	-
Personal business	-	0.01 (2.38)	-	-	0.004 (1.87)	-
Recreation	−0.003 (−1.00)	-	-	-	−0.001 (−1.03)	-
Eat-out	-	-	-	-	−0.004 (−1.00)	-
Social	-	-	-	-	-	-
Shopping	-	0.006 (1.10)	-	-	-	0.008 (4.58)
Annual household income more than $80,000
Escorting	-	-	0.34 (2.23)	-	-	0.08 (1.82)
Personal business	-	-	-	−0.04 (−1.60)	-	-
Recreation	0.71 (3.21)	−1.46 (−1.38)	-	-	0.45 (3.37)	-
Eat-out	-	-	-	-	-	-
Social	-	-	-	-	−0.14(−1.52)	-
Shopping	0.40 (2.11)	-	−0.75 (−1.85)	-	-	-
Full-time workers
Escorting	0.18 (1.92)	-	-	−0.06 (2.00)	-	-
Personal business	-	-	-	-	-	-
Recreation	-	-	-	-	-	-
Eat-out	0.21 (1.15)	-	-	-	0.20 (1.48)	-
Social	−0.14 (−1.22)	-	-	-	-	0.08 (1.01)
Shopping	-	-	-	-	-	−0.08 (−1.22)
Household with more than three children
Escorting	-	-	0.94 (6.57)	-	-	1.01 (8.59)
Personal business	−0.24 (−2.08)	-	-	-	-	-
Recreation	-	-	-	0.13 (2.14)	−0.37 (−1.23)	-
Eat-out	−0.35 (−1.72)	-	-	-	-	-
Social	−0.62 (−3.80)	-	-	-	-	−0.23 (−1.92)
Shopping	-	-	−0.61 (−2.03)	-	-	−0.25 (−1.34)
No. of vehicles in a household
Escorting	-	0.20 (2.29)	-	-	-	−0.10 (−3.83)
Personal business	-	-	0.13 (1.50)	-	-	0.04 (1.90)
Recreation	-	-		-	−0.13 (−1.43)	-
Eat-out	-	-	-	0.04 (2.95)	-	-
Social	-	-	-	-	0.06 (2.16)	-
Shopping	-	0.11 (1.63)	-	-	0.07 (3.95)	−0.09 (−1.92)
Distance from home to nearest urban core less than 2 km
Escorting	-	-	0.16 (1.00)	-	−0.05 (−1.55)	-
Personal business	-	-	-	-	-	-
Recreation	-	0.28 (1.06)	-	-	-	-
Eat-out	−0.29 (−1.13)	-	-	-	-	-
Social	-	-	-	-	-	0.14 (2.09)
Shopping	-	-	-	0.04 (1.71)	-	-
Land use index
Escorting	-	-	-	-	-
Personal business	-	-	1.30 (2.99)	-	-	0.38 (3.48)
Recreation	-	-	-	0.07 (1.00)	-	-
Eat-out	-	-	1.11 (2.11)	-	-	-
Social	-	-	-	-	-	-
Shopping	-	-	1.30 (2.81)	0.08 (1.31)	-	-
Satiation parameters
Constant
Escorting	2.91 (23.14)			3.37 (54.83)
Personal business	4.72 (29.67)			4.93 (64.90)
Recreation	5.28 (19.87)			6.07 (45.78)
Eat-out	4.59 (18.47)			5.01 (72.13)
Social	6.09 (22.72)			6.62 (53.48)
Shopping	4.35 (21.33)			4.73 (64.23)

Note: “-” = insignificant.

Table 3 represents the parameter estimation results for the MDCEV-OP model. It must be noted that constants are estimated in the baseline preference for each episode of the activity considering “outside good” as the reference. The activity-level constants correspond to the first episode of an activity as it enters the baseline utility of all the episodes of that activity. In addition, the activity-level constants depict the aggregate influence of unobserved factors on the propensity to engage in one or more episodes of an activity. On the other hand, episode-specific constants capture the differential effect for that episode with respect to the activity level constants. For example, in case of non-telecommuters, the baseline preference constant for the second episode of “escorting” activity is −7.16 (because it is the sum of associated activity level constant and the second episode-specific constant, that is, (−8.18 + [1.02] = −7.16). In the adopted MDCEV-OP framework, the baseline utility parameters across different episodes of an activity are always in non-increasing order. Because of this inherent property, overall baseline preferences are in decreasing order even if episode-specific parameters are positive or zero. This ordering in the baseline utility circumvents the need to estimate each episode-specific parameter in the model, and they turned out to be insignificant and were dropped from the final model. In addition to this, a reason for insignificant episode-specific parameters is the activity engagement rates at an episode level. The percentage of individuals participating in a higher number of activity episodes is quite low in the data (Table 1). It must be noted that the scale parameter was also estimated for both the models. The estimated scale parameter for telecommuters and non-telecommuters is 0.47 (21.33) and 0.25 (34.94), respectively. The next subsection explores telecommuters’ and non-telecommuters’ activity-time-use behavior by testing the effect of socio-demographic (individual and household level), vehicle ownership, and built environment attributes.

Non-Telecommuters: Baseline Preference and Satiation Parameter Estimates

Among socio-demographic attributes, females working in a standard work arrangement (i.e., commuting to work) are more likely to engage in higher episodes of escorting (i.e., episode-specific parameter for the third episode is 0.05) and personal business activity (i.e., episode-specific parameter for the third episode is 0.12), along with shopping activity (i.e., episode-specific parameter for the second episode is 0.11). This might be because of females taking more household responsibilities and detouring to accomplish multiple tasks, while returning back to home after work. Older workers commuting to work are less likely to engage in higher episodes of pick-up/drop-offs, recreational, and eat-out activities. However, they showed propensity toward engaging in higher episodes of personal business and shopping activities. Individuals residing in high-income households are likely to engage in higher episodes of recreational activity, which might be because of their association with financial stability. Similar results were observed for escorting activity. This suggests that individuals commuting daily might engage in household-related responsibilities such as dropping-off/picking-up children to/from daycare to/from work. However, they revealed lower propensity toward engaging in multiple episodes of social and personal business activity, revealing work-related constraints. Full-time workers commuting daily showed lower propensity toward escorting and multiple episodes of shopping activity. However, they might be sharing this responsibility with other household members who might be working part-time or from home. They revealed a positive relationship with eat-out (i.e., episode-specific parameter for the second episode is 0.20). This might be because of going out for lunch/dinner with colleagues, family members, or both. Lastly, similar findings were depicted for social activity (i.e., episode-specific parameter for the third episode is 0.08), which might mainly occur after working hours.

In relation to household-level attributes, the presence of more children revealed a negative relationship with higher episodes of social and shopping activity, as individuals tend to allocate the majority of their time to child-care responsibilities after returning from work. However, they tend to engage in multiple episodes of escorting activity, as they might be required to drop-off/pick-up children to/from school/childcare. Interestingly, such individuals are likely to engage in recreational activity but limit themselves to one episode per day. This finding shows the need to take children to parks or recreational areas, but also the time constraints associated with work, preventing them from engaging in recreational activity more than once in a day. The presence of a higher number of vehicles in a household shows a negative relationship with escorting activity, as expected. However, a positive relationship was retained for personal business, eat-out, social, and shopping activities. This indicates that the travel freedom gained through increased vehicle ownership encourages individuals to engage in more non-mandatory OH activities after work. This might also reflect the budgetary flexibility of individuals owning more vehicles.

Among the built environment attributes, workers residing closer to urban cores are less likely to engage in escorting activities. This might be because of the presence of shared mobility services and good public transit accessibility circumventing pick-up/drop-offs. As expected, such residents are likely to engage in social and shopping activity because of closer proximity to destination locations (e.g., shopping centers, cinemas). Similarly, workers residing in mixed land use showed propensity toward engaging in personal business, recreational, and shopping activity. This finding shows the utility gained from urban characteristics allowing them to engage in multiple episodes in a day. Lastly, the magnitudes of satiation parameters controlling the duration of engagement in different OH activities are consistent with corresponding durations presented in Table 1. For example, social activity yields the highest parameter estimate, reaffirming that non-telecommuters spend majority of their time socializing after work. On the other hand, the lowest magnitude was observed for escorting activity, as expected, and is consistent with the average duration of the respondents (Table 1).

Telecommuters: Baseline Preference and Satiation Parameter Estimates

Among the socio-demographic attributes, model results suggest that female telecommuters are likely to engage in higher episodes of personal business activity (i.e., episode-specific parameter for the second episode is 0.38) and shopping activity (i.e., episode-specific parameter for the third episode is 0.63). This finding reaffirms the understanding that females tend to perform a higher share of household responsibilities, such as buying household-related supplies or visiting banks, while working from home. Moreover, a positive relationship was revealed for the second episode of recreational activity. Older telecommuters are less likely to engage in higher episodes of escorting activity, as they tend to share such household-related responsibilities with their younger counterparts. In addition, they showed preference toward personal business and shopping activities. On the other hand, results were not the same for recreational activity. Telecommuters residing in higher-income households showed propensity toward higher episodes of escorting activity (i.e., episode-specific parameter for the third episode is 0.34). This finding demonstrates their work schedule flexibility allowing them to perform responsibilities such as picking-up or dropping-off children from/to day-care. Interestingly, such individuals tend to engage in recreational and shopping activity utilizing their expenditure capabilities to go out to compensate for the increased sedentary lifestyle associated with telecommuting. However, this is restricted to one episode only. Full-time workers working from home revealed a negative relationship with activities such as social activities. They are likely to engage in escorting and eat-out activities, exploiting the flexibility associated with working from home.

In the case of household-level attributes, households with a higher number of children are more likely to engage in higher episodes of escorting activity in a day. Again, this indicates the multiple child-care responsibilities which prompt more escorting episodes ( 19 ). This indicates that telecommuters might be taking the lead in pick-up/drop-off activities often of higher episodes. As they spend the majority of time at work and multiple childcare responsibilities, the propensity to engage in other activities (e.g., personal business, eat-out, social, and shopping) is low, as expected. In relation to vehicle ownership, households with a higher number of vehicles revealed a positive relationship with a second episode of escorting activity. This indicates that, despite multiple vehicle ownership in a household, telecommuters tend to share household-related responsibilities and perform pick-ups and drop-offs for their household members.

In the case of built environment attributes, telecommuters residing closer to regional-urban cores are more likely to engage in higher episodes of escorting activity. This finding indicates that the presence of urban characteristics such as active transportation networks, accessibility to bus stops, and work-related flexibility encourages telecommuters to go out and engage in household-related responsibilities such as pick-up/drop-off children, compared with their suburban and rural counterparts. Similar findings were observed for recreational activity, showing the effect of nearby destinations such as cinemas or parks. Furthermore, individuals living in mixed land use areas revealed propensity toward multiple episodes (up to the third episode) of personal business, eat-out, and shopping activities. This finding shows that individuals with greater accessibility to nearby destinations and work-related flexibility engage in multiple non-mandatory activities in a day. Lastly, the magnitudes of satiation parameters controlling the duration of engagement in different OH activities are consistent with corresponding durations presented in Table 1. For example, social activity revealed the highest satiation parameter (6.09) which is consistent with the longest time invested in social activity among the sample population of telecommuters. Interestingly, recreational activity yields the second highest satiation parameter estimate of 5.28, reaffirming that telecommuters allocate significant amounts of time to recreational activities too. Meanwhile, escorting activity had the lowest satiation parameter (2.91) with an average duration of 33.4 min (Table 1).

Forecasting Analysis

The estimated MDCEV-OP model was further applied to detect how the proposed model could replicate the observed activity engagement and duration rates at an episode level. The forecasting procedure adopted in this study was outlined by Saxena et al. ( 12 ). The forecasting algorithm was applied to the hold out sample of 200 non-telecommuters, and prediction was performed on its own estimation sample for telecommuters, because of the low sample size. In this study 50 sets of error draws were simulated to cover the distribution of error terms for each individual. For comparison purposes, in addition to the developed MDCEV-OP model, a disaggregated MDCEV model has been deployed for the telecommuters and non-telecommuters. The forecasting accuracy of the models have been compared based on weighted root mean square error (RMSE) value as follows:

weighted RMSE = \sqrt{(\frac{\sum_{k} \sum_{j_{k}} o_{k, j_{k}} * r_{k, j_{k}}^{2}}{\sum_{k} \sum_{j_{k}} o_{k, j_{k}}})}

where

$r_{k, j_{k}}$ = the relative error in predictions and can calculated as $r_{k, j_{k}} = \frac{(p_{k, j_{k}} - o_{k, j_{k}})}{o_{k, j_{k}}}$ . Here $p_{k, j_{k}}$ and $o_{k, j_{k}}$ are the predicted and observed activity engagement (duration) rates for the activity episode ( $k, j_{k}$ ). The goodness-of-fit measure suggests that the MDCEV-OP model outperforms the disaggregated MDCEV model and does well in predicting the observed activity participation and duration rates (Table 4).

Table 4.

Fit Measures for the Predictions

		Non-telecommuters		Telecommuters
		MDCEV-OP	Disaggregated MDCEV	MDCEV-OP	Disaggregated MDCEV
Weighted root mean squared error	Activity participation	0.12	0.94	0.07	0.98
Weighted root mean squared error	Activity duration	0.10	0.51	0.05	0.10

Note: MDCEV = multiple discrete continuous extreme value; MDCEV-OP = multiple discrete continuous extreme value with ordered preferences.

The observed and predicted average activity participation (i.e., episode level) and duration rates for the MDCEV-OP model are represented in Table 5. The results suggest that the model accurately predict the general trend in higher participation rates among telecommuters in personal business, recreation, social, and shopping activities compared with non-telecommuters. However, the model predicts a lower participation rate in escorting and eat-out activities among the telecommuters. In the case of accuracy, prediction is within a few percentage points of the observed values. For example, in the case of activity participation rates, 29.5% of non-telecommuters participated in the first episode of escorting activity, whereas the model’s prediction is 29.2% for the same. Similarly, the difference in average observed and predicted participation rate for the second episode of eat-out activity for telecommuters is 0.27%. Lastly, in the case of activity duration predictions, non-telecommuters engage in second episode of shopping activity for an average of 18.5 min, whereas the predicted value is 18.0 min. The difference in observed and predicted average durations for telecommuters for the first episode of social activity is ∼3%.

Table 5.

Observed and Predicted Average Time Allocations (Participation Rates) from the Proposed Multiple Discrete Continuous Extreme Value with Ordered Preferences (MDCEV-OP) Model

Activity purpose		Duration of first episode (minutes) (participation in %)	Duration of second episode (minutes) (participation in %)	Duration of third episode (minutes) (participation in %)
Non-telecommuters
Shopping	Observed	36.9 (30.5)	18.5 (6.5)	1.0 (0.5)
	Predicted	40.8 (27.8)	18.0 (5.8)	9.4 (0.6)
Eat-out	Observed	55.5 (21.0)	6.3 (2.0)	5.0 (0.5)
	Predicted	48.0 (19.2)	22.3 (1.9)	4.4 (0.1)
Social	Observed	173.3 (22.5)	51.3 (2.0)	1.0 (0.5)
	Predicted	146.3 (18.1)	44.6 (1.1)	0.3 (0.01)
Personal business	Observed	51.5 (26.0)	27.6 (5.0)	15.0 (0.5)
	Predicted	47.9 (21.7)	20.2 (3.4)	7.0 (0.3)
Recreation	Observed	101.8 (20.0)	65.0 (1.0)	1.0 (0.5)
	Predicted	107.4 (16.8)	32.3 (1.1)	3.8 (0.1)
Escorting	Observed	14.8 (29.5)	3.0 (11.0)	5.0 (1.0)
	Predicted	12.2 (29.2)	5.8 (10.0)	4.3 (2.6)
Duration at activity level (participation in %)
Outside good	Observed	1,334.5 (100)
	Predicted	1,354.9 (100)
Telecommuters
Shopping	Observed	64.3 (39.84)	23.7 (9.21)	11.6 (1.90)
	Predicted	73.0 (35.88)	27.1 (10.12)	18.7 (1.49)
Eat-out	Observed	70.4 (10.84)	20.0 (0.54)	1.0 (0.27)
	Predicted	77.8 (10.19)	26.0 (0.81)	5.4 (0.07)
Social	Observed	214.6 (19.51)	82.2 (2.44)	15.0 (0.27)
	Predicted	221.3 (18.98)	73.2 (1.50)	4.0 (0.04)
Personal business	Observed	138.6 (31.71)	24.3 (6.78)	21.7 (1.63)
	Predicted	93.0 (29.27)	33.7 (7.13)	17.9 (0.76)
Recreation	Observed	106.0 (26.29)	61.5 (2.71)	50.0 (0.27)
	Predicted	138.8 (23.62)	44.0 (2.95)	11.2 (1.19)
Escorting	Observed	33.5 (26.56)	7.3 (15.45)	3.9 (2.71)
	Predicted	20.4 (28.97)	8.1 (9.39)	4.9 (3.00)
Duration at activity level (participation in %)
Outside good	Observed	1,274.6 (100)
	Predicted	1,288.8 (100)

Conclusions

Recent advances in information and communication technology, along with a pandemic, have triggered a strengthened interest in telecommuting. Telecommuters could have a different daily time-use behavior than non-telecommuters. Given this context, the present study investigates the activity-time-use patterns of two types of worker profiles—telecommuters and non-telecommuters—focusing on their engagement in non-mandatory activities. A newly proposed MDCEV-OP model has been developed using travel survey data from the Central Okanagan region of British Columbia, Canada. This modified MDCEV model deals with activity participation and duration decisions at a disaggregated episode-level accounting for multiple occurrences of each activity type. A logical ordering of the episodes has been imposed to ensure that lower-numbered episodes occur first (i.e., the j^th episode of an activity type is not predicted without the occurrence of the (j–1)^th episode of that activity type). Furthermore, this study tests the effect of socio-demographic attributes, vehicle ownership, and built environment attributes on people’s activity-time-use patterns.

The findings of this study confirm that there is a significant difference in activity-time-use patterns of telecommuters and non-telecommuters over a 24 h period. For instance, in the case of activity participation, female non-telecommuters are more likely to take household-related responsibilities such as pick-up/drop-off of children to school/day-care, as well as running daily errands. They might be undertaking these activities while commuting. In contrast, female telecommuters showed higher propensity toward personal business, recreation, and shopping activities, conceivably because of extra time allocation from not having to commute. The higher propensity toward shopping activities likely comes from household-related responsibilities such as grocery shopping. Interestingly, full-time non-telecommuters are less likely to engage in non-mandatory activities including escorting and shopping, which might be because of time constraints imposed by longer working hours. Although, initially, this negative relationship for escorting is somewhat counterintuitive, as full-time workers could pick-up/drop-off household members on their way to work. However, after further investigation, and considering the positive relationship for full-time telecommuters, it is possible that non-telecommuters might be sharing this escorting responsibility with other household members with more flexible work arrangements. For example, full-time telecommuters are engaging in escorting activity, further implying flexibility associated with working from home and participating in household-related responsibilities.

The presence of a higher number of children in the household indicates a lower likelihood to engage in non-mandatory activities for both telecommuters and non-telecommuters, which might be because of the tendency to spend more time with family after work, mostly at home. Both groups showed positive relationships toward escorting activities, indicating the presence of multiple childcare responsibilities which might stimulate the need to engage in more escorting episodes ( 19 ). In the case of built environment attributes, both non-telecommuters and telecommuters residing closer to urban centers are more likely to engage in various non-mandatory activities. This finding shows that accessibility derived from urban settings motivates both the worker profiles to engage in non-mandatory activities. In a similar line of investigation, this study also deals with the activity duration controlled by satiation parameters. The parameter estimation results clearly reflect that both worker profiles allocate the most time to social activity, followed by recreational activity. As expected, escorting activity shares the lowest satiation parameter estimates, confirming minimal time allocation to pick-up/drop-offs by workers.

Overall, the findings of this study reveal that telecommuters are more likely to perform multiple episodes of an activity compared with non-telecommuters. The findings of this study demonstrate the need to accommodate the daily time-use difference between telecommuters and non-telecommuters while developing travel demand management plans, policies, and models. For instance, one of the interesting findings is that telecommuters are more likely to participate in higher episodes of activities related to household-related responsibilities such as escorting and shopping. Therefore, as employers and government agencies adopt policies and develop incentive programs to promote telecommuting, it is important to recognize the possible implications for telecommuters to be likely burdened with more household-related activities compared with their non-telecommuter counterparts. The results also indicate certain market segments could benefit substantially through adopting telecommuting. For example, the tendency of high-income workers (telecommuters and non-telecommuters) to engage in more non-mandatory activities compared with their lower-income counterparts indicates a need to evaluate how more affordable and equitable access to such activities for marginalized groups can be created. Lastly, results of built environment variables reveal that more mixed land use development closer to urban cores might increase the non-mandatory activity participation of residents (i.e., telecommuter and non-telecommuters). These findings provide the motivation to update our peak-hour four-stage travel demand forecasting model to a 24 h episode-level activity-based model and accommodate telecommuters’ and non-telecommuters’ daily time-use behavior separately for improved, consistent, and accurate prediction of travel demand.

Although the findings on the 24 h time-use analysis are promising, there is scope for improvements to this study, for example, better data collection. This study does not accommodate hybrid work arrangements, as many individuals might prefer telecommuting some days of the week and some days traveling to work, and this is considered as a limitation of the study and is largely attributable to the unavailability of such week-long travel data. Future research should focus on collecting travel data for a longer period—such as over a whole week. This might introduce significant burdens on the respondents if the method of data collection continues to be self-reporting. More research is required on how to collect week-long travel data of better quality while imposing less survey burden. Furthermore, one of the important aspects is time-of-day, particularly when the focus is telecommuters and non-telecommuters, since telecommuters might have the flexibility to participate in activities during off-peak hours. Incorporating such interactions in the form of time-of-day within the developed modeling framework will substantially increase the complexity of the model. For example, time-of-day often leads to sequence/chronology of occurrence of an activity episode; addressing this requires an additional scheduling algorithm within the proposed framework. This could be accommodated in future research while analyzing the activity scheduling procedure. In addition, the adopted modeling framework does not account for correlation among the alternatives, which is a limitation of this study. Although a nested extension of the aggregate-level MDCEV model was developed by Pinjari and Bhat, one of the future research agendas could be extending this episode-level MDCEV framework to accommodate such correlation among the disaggregated-episode-level alternatives ( 28 ). Overall, this study provides insights on how socio-demographic attributes (individual and household level), vehicle ownership, and built environment attributes effect the activity-time-use patterns of telecommuters and non-telecommuters. In addition, the proposed model not only obviates the need of additional modeling techniques to deal with activity generation components at an episode level, but also relaxes the difficulty of considering the effect of accessibility measures such as mode or destination choices that vary by each activity episode ( 11 ). Lastly, this model has been developed to be included as the activity generation component of an activity-based travel forecasting tool—currently under development at the University of British Columbia, Okanagan. The enhanced behavioral representation of this model is expected to improve the forecasting accuracy of the travel model.

Footnotes

Acknowledgements

The authors would like to thank the City of Kelowna for providing the data for analysis. The authors would also like to thank Shobhit Saxena for discussion and help with forecasting.

Authors Contributions

The authors confirm contribution to the paper as follows: study conception and design: S. Khaddar, M. Fatmi; data collection: M. Fatmi; analysis and interpretation of results: S. Khaddar, M. Fatmi; draft manuscript preparation: S. Khaddar, M. Fatmi, T. Nikodym. All authors reviewed the results and approved the final version of the manuscript.

Declaration of Conflicting Interests

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: The authors would like to thank the Natural Sciences and Engineering Research Council (NSERC)–Discovery Grant for their financial support.

ORCID iDs

Shivam Khaddar

Mahmudur Rahman Fatmi

Trevor Nikodym

References

Ryuichi

Nilles

J. M.

Conroy

Fleming

D. M.

Telecommuting as a Transportation Planning Measure: Initial Results of California Pilot Project. Trasportation Research Record, 1990. 1285: 98–104.

Nayak

Pandit

Potential of Telecommuting for Different Employees in the Indian Context beyond COVID-19 Lockdown. Transport Policy, Vol. 111, 2021, pp. 98–110. https://doi.org/10.1016/j.tranpol.2021.07.010.

S. Y.

Telecommuting, Income, and out-of-Home Activities. Travel Behaviour and Society, Vol. 2, No. 3, 2015, pp. 131–147. https://doi.org/10.1016/j.tbs.2014.12.003.

Kim

S. N.

Choo

Mokhtarian

P. L.

Home-Based Telecommuting and Intra-Household Interactions in Work and Non-Work Travel: A Seemingly Unrelated Censored Regression Approach. Transportation Research Part A: Policy and Practice, Vol. 80, 2015, pp. 197–214. https://doi.org/10.1016/j.tra.2015.07.018.

Rajagopalan

B. S.

Pinjari

A. R.

Bhat

C. R.

Comprehensive Model of Worker Nonwork-Activity Time Use and Timing Behavior. Transportation Research Record, 2009. 2134: 51–62. https://doi.org/10.3141/2134-07.

Shabanpour

Golshani

Tayarani

Auld

Mohammadian

A. (Kouros)

. Analysis of Telecommuting Behavior and Impacts on Travel Demand and the Environment. Transportation Research Part D: Transport and Environment, Vol. 62, 2018, pp. 563–576. https://doi.org/10.1016/j.trd.2018.04.003.

Budnitz

Tranos

Chapman

Telecommuting and Other Trips: An English Case Study. Journal of Transport Geography, Vol. 85, 2020, p. 102713. https://doi.org/10.1016/j.jtrangeo.2020.102713.

Srinivasan

Bhat

C. R.

A Multiple Discrete-Continuous Model for Independent- and Joint-Discretionary-Activity Participation Decisions. Transportation, Vol. 33, No. 5, 2006, pp. 497–515. https://doi.org/10.1007/s11116-006-8078-8.

Losa Rovira

Faghih Imani

Sivakumar

Pawlak

Do In-Home and Virtual Activities Impact out-of-Home Activity Participation? Investigating End-User Activity Behaviour and Time Use for Residential Energy Applications. Energy and Buildings, Vol. 257, 2022, p. 111764. https://doi.org/10.1016/j.enbuild.2021.111764.

10.

Palma

Enam

Hess

Calastri

Crastes dit Sourd

Modelling Multiple Occurrences of Activities during a Day: An Extension of the MDCEV Model. Transportmetrica B, Vol. 9, No. 1, 2021, pp. 456–478. https://doi.org/10.1080/21680566.2021.1900755.

11.

Khaddar

Saxena

Fatmi

M. R.

An Episode-Level Joint Model for Activity Engagement, Duration, and Destination Location Choice Decisions. Transportation Letters, 2022, pp. 1–11. https://doi.org/10.1080/19427867.2022.2102347.

12.

Saxena

Pinjari

A. R.

Paleti

A Multiple Discrete-Continuous Extreme Value Model with Ordered Preferences (MDCEV-OP): Modelling Framework for Episode-Level Activity Participation and Time-Use Analysis. Transportation Research Part B: Methodological, Vol. 166, 2022, pp. 259–283. https://doi.org/10.1016/j.trb.2022.09.008.

13.

Mokhtarian

P. L.

Defining Telecommuting. Transportation Research Record, 1991. 1305(14): 273–281.

14.

Zhu

Are Telecommuting and Personal Travel Complements or Substitutes?

Annals of Regional Science, Vol. 48, No. 2, 2012, pp. 619–639. https://doi.org/10.1007/s00168-011-0460-6.

15.

Asgari

Jin

Impacts of Telecommuting on Nonmandatory Activity Participation Role of Endogeneity. Transportation Research Record, 2017. 266: 47–57. https://doi.org/10.3141/2666-06.

16.

Chu

Y. L.

Modeling Workers’ Daily out-of-Home Maintenance Activity Participation and Duration. International Journal of Transportation Science and Technology, Vol. 9, No. 2, 2020, pp. 137–147. https://doi.org/10.1016/j.ijtst.2019.12.002.

17.

Calastri

Hess

Daly

Carrasco

J. A.

Does the Social Context Help with Understanding and Predicting the Choice of Activity Type and Duration? An Application of the Multiple Discrete-Continuous Nested Extreme Value Model to Activity Diary Data. Transportation Research Part A: Policy and Practice, Vol. 104, 2017, pp. 1–20. https://doi.org/10.1016/j.tra.2017.07.003.

18.

Srinivasan

Bhat

C. R.

Modeling Household Interactions in Daily In-Home and out-of-Home Maintenance Activity Participation. Transportation, Vol. 32, No. 5, 2005, pp. 523–544. https://doi.org/10.1007/s11116-005-5329-z.

19.

Garikapati

V. M.

You

Pendyala

R. M.

Vovsha

P. S.

Livshits

Jeon

Multiple Discrete-Continuous Model of Activity Participation and Time Allocation for Home-Based Work Tours. Transportation Research Record, 2014. 2429: 90–98. https://doi.org/10.3141/2429-10.

20.

Pendyala

R. M.

Goulias

K. G.

Kitamura

Impact of Telecommuting on Spatial and Temporal Patterns of Household Travel. Transportation, Vol. 18, 1991, pp. 383–409. https://doi.org/10.1007/BF00186566.

21.

Paleti

Vukovic

Telecommuting and Its Impact on Activity-Time Use Patterns of Dual-Earner Households. Transportation Research Record, 2017. 2658: 17–25. https://doi.org/10.3141/2658-03.

22.

Nurul Habib

K. M.

Day

Miller

E. J.

An Investigation of Commuting Trip Timing and Mode Choice in the Greater Toronto Area: Application of a Joint Discrete-Continuous Model. Transportation Research Part A: Policy and Practice, Vol. 43, No. 7, 2009, pp. 639–653. https://doi.org/10.1016/j.tra.2009.05.001.

23.

Eluru

Pinjari

A. R.

Pendyala

R. M.

Bhat

C. R.

An Econometric Multi-Dimensional Choice Model of Activity-Travel Behavior. Transportation Letters, Vol. 2, No. 4, 2010, pp. 217–230. https://doi.org/10.3328/TL.2010.02.04.217-230.

24.

Smart Trips Okanagan Travel Survey. 2018 Okanagan Travel Survey Report 3: Analysis of Survey Results and Trends. 2020.

25.

Bhat

C. R.

The Multiple Discrete-Continuous Extreme Value (MDCEV) Model: Role of Utility Function Parameters, Identification Considerations, and Model Extensions. Transportation Research Part B: Methodological, Vol. 42, No. 3, 2008, pp. 274–303. https://doi.org/10.1016/j.trb.2007.06.002.

26.

Bhat

C. R.

A Multiple Discrete-Continuous Extreme Value Model: Formulation and Application to Discretionary Time-Use Decisions. Transportation Research Part B: Methodological, Vol. 39, No. 8, 2005, pp. 679–707. https://doi.org/10.1016/j.trb.2004.08.003.

27.

Bhat

C. R.

A New Flexible Multiple Discrete–Continuous Extreme Value (MDCEV) Choice Model. Transportation Research Part B: Methodological, Vol. 110, 2018, pp. 261–279. https://doi.org/10.1016/j.trb.2018.02.011.

28.

Pinjari

A. R.

Bhat

A multiple discrete–continuous nested extreme value (MDCNEV) model: Formulation and application to non-worker activity time-use and timing behavior on weekdays. Transportation Research Part B: Methodological, Vol. 44, No. 4, 2010, pp. 562–583. https://doi.org/10.1016/j.trb.2009.08.001.