The relations between working from home and travel behaviour: a panel analysis

doi:10.21203/rs.3.rs-1985669/v1

Download PDF

Research Article

The relations between working from home and travel behaviour: a panel analysis

https://doi.org/10.21203/rs.3.rs-1985669/v1

This work is licensed under a CC BY 4.0 License

Journal Publication

published 23 Jun, 2023

Read the published version in Transportation →

You are reading this latest preprint version

Policies to increase the amount of time people spend working from home were widely used during the COVID-19 pandemic. Since research suggests that the resulting increase in working from home will outlast these policies themselves, policy makers want to know what the expected effects of working from home on travel behaviour will be in a post-pandemic world. To answer this question it is important to understand the relations between working from home and travel behaviour. To improve the understanding of the relations between working from home and travel behaviour we apply longitudinal modelling techniques using panel data from the Netherlands Mobility Panel. This allows us to separate between-persons and within-persons relations and effects. We find a negative effect of working from home on commute travel time both before and during the pandemic and a positive effect on leisure travel time only during the pandemic. The sizes of these effects remained roughly similar during the pandemic, although the extent to which working from home affects commute travel time increased during the pandemic. The net effect of working from home on travel time as a whole is negative, indicating that working from home policies could be used to reduce travel time. The results show that part of the relationships between working from home and travel behaviour has changed during the pandemic. As a result, policy makers and transport operators should be careful when estimating future travel demand based on extrapolations of relationships found only before or during the pandemic.

Travel Behavior

Working from Home

COVID-19

Netherlands Mobility Panel

Structural Equations Modeling

Ongoing developments in information and communication technologies have enabled more people to effectively work from home in the last few decades. This trend is not new, and visions of empty highways because of drastically reduced commuting demand were discussed as early as the 1970s and 1980s (Kraut, 1989; Olson, 1983). These visions however did not come to fruition for decades, until the COVID-19 pandemic. In the wake of the COVID-19 pandemic, many governments enforced a work from home mandate in some form (Shibayama et al., 2021). This policy led to, or coincided with, a sharp decline in travel demand (Beck & Hensher, 2020a; de Haas et al., 2020; Molloy et al., 2021).

In many countries, including the Netherlands, the policies intended to reduce the spread of COVID-19 have been scaled down in the spring and summer of 2022. Subsequently, total travel demand increased again. However, the increase in working from home seems to outlast the original measures from the government (Jain et al., 2022), as a substantial share of employees and employers alike have discovered its benefits (Shortall et al., 2021). For policy makers and transport planners, who are interested in a prediction of travel demand in the future, this raises the question what the effects of working from home on travel demand will be in the post-pandemic future. To answer this question, the relationship between working from home and travel behaviour needs to be understood well.

The literature on this relationship shows that people who work from home effectively substitute their travel commute by digital communication technologies, which reduces travel demand for the commute purpose. However, it is not clear what effect this has on the time spent travelling for other purposes. Past research has indicated that people who work from home spend more time on non-commute travel, a phenomenon referred to as complementarity (Elldér, 2020). In addition, working from home may have the longer-term effect that people decide to accept longer, but less frequent trips, for example by relocating or by changing jobs (Mokhtarian et al., 2004).

However, these general findings do not completely answer questions relating to the relationship between working from home and travel behaviour. For one, there is no clear consensus about the relative size of the substitution and the complementary effects: in essence, whether an increase of working from home results in an increase or decrease of total travel demand. In general, earlier studies find a net substitution effect (Mokhtarian et al., 1995), whereas more recent studies find a more balanced or even net complementary effect (Elldér, 2020). A possible reason for this discrepancy is the different study designs: earlier studies often used relatively small samples, based on specific projects or pilots. More recent studies use larger, more general travel survey data which they typically analyse using cross-sectional models. The upside of this approach is the greater power of the larger samples to generalize to the population, but the downside is that the cross-sectional data makes it difficult to disentangle causal effects – and the direction of these effects – from associations. We aim to combine the best of both worlds by using a larger longitudinal sample, the Netherlands Mobility Panel (MPN). This allows us to disentangle between-person and within-person relations between working from home and travel behaviour.

Second, the results from pre-pandemic studies cannot be extrapolated to a post-pandemic world, for two reasons: 1) a new cohort of people started working from home during the pandemic and 2) working from home levels during the pandemic were much higher than they were before the pandemic. Furthermore, results from studies during the pandemic can also not provide a full answer to the question, as they are inevitably intertwined with the broader effects of the pandemic on travel behaviour. Given that post-pandemic conditions have not been fully met as of the time of writing this article, the best approach to provide some estimates for the relations between working from home and travel behaviour post-pandemic is then to combine insights using data pertaining to both the situation before and the situation during the pandemic.

The main contribution of this study is a deeper insight into the bidirectional relationship between working from home and travel behaviour at a time where both constituent parts of this relationship have changed drastically. We use panel data to separate between-person and within-person relations, which is achieved by applying a random intercept cross-lagged panel model (RI-CLPM; Hamaker et al. (2015)) on the MPN, a large panel containing data collected both before and during the pandemic from the same set of respondents. We use data pertaining to the years 2017 through 2021. The availability of panel data pertaining to both the situation before the COVID-19 pandemic and that contains multiple years during the pandemic also allows us to test whether the pandemic resulted in a change in these within-persons relations.

The insights of this study help provide policy makers and public transport operators with better predictions for the effects of working from home on total travel demand. These predictions affect both relatively short-term decisions, such as public transport operational schedules, and longer-term decisions such as road infrastructure investments. More insights into the relations between working from home and travel behaviour thus are necessary both during the more short-term transition out of the pandemic into the post-pandemic ‘normal’ and the long-term predictions of how working from home will shape travel demand in the more distant future. Furthermore, it may help policy makers and employers in evaluating the effects of working-from-home related policies on travel behaviour.

In Section 2, an overview of the literature on working from home and travel behaviour is provided, culminating in a conceptual model. In Section 3, the methods and data are introduced and described. Section 4 contains the results from the estimated model, and the conclusion and discussion can be found in Section 5.

This section contains an overview of the literature on the relations between working from home and travel behaviour. First, an overview of different conceptual relations between working from home and travel behaviour is shown, followed by a more in-depth look at the effects of working from home on travel behaviour. Then we summarise the literature on the effects of COVID-19 on working from home and travel behaviour, and we conclude by presenting a conceptual model that forms the base of the further analyses.

2.1. Conceptualising the relationship between working from home and travel behaviour

The relationship between working from home and travel behaviour has been studied since at least the early 1970’s, when Nilles (1976) proposed it as a solution to reduce commute travel in the face of the oil crisis (He & Hu, 2015). With the rapid increase in information and communication technologies, working from home has become a more feasible solution for a greater part of the population in the decades since (Harpaz, 2002; Siha & Monroe, 2006). Despite seeing an (albeit relatively small) increase in working from home in developed nations in the decades between the 1970’s and 2019, total time spent travelling remained relatively constant during the same time as well (Kasraian et al., 2018; Susilo & Maat, 2007). It appears thus that until the COVID-19 pandemic, the effect of working from home on travel behaviour was relatively small, at least for the population as a whole.

This small or non-existing effect can partly be explained by the relatively low rate of adoption of working from home, but also partly by the complex relationship between working from home and travel behaviour. For one, literature suggests that there is a bi-directional relationship between working from home and travel behaviour (Mokhtarian et al., 2004). Part of this relationship is due to self-selection effects, as people who choose to work from home, and thus have the option to do so, have different travel behaviour habits than people who do not choose to do so (He & Hu, 2015). Another part is due to a relationship between commute distance and working from home, where people with a longer commute distance seemingly choose to work from home more often (de Vos et al., 2018; Zhu, 2012). These issues make it impossible to estimate unbiased effects of working from home on travel behaviour without controlling for endogeneity (He & Hu, 2015). One of the best ways of doing so is using panel studies, which however are rare in the literature. One application of a panel by de Vos et al. (2018) finds that people who work from home accept a longer commute.

Further complicating the matter is the fact that the effect of working from home on travel behaviour is not as straightforward as it may seem at first glance (Mokhtarian, 1990; Salomon, 1986; Zhu, 2012). Certainly, by replacing a physical (travel) commute for a digital telecommute working from home causes less commute travel trips. This effect is referred to as the substitution effect, where the use of ICT substitutes for physical travel (Andreev et al., 2010). However, the lack of a physical commute saves time, which can be used for travelling for other purposes (Elldér, 2020). If this is the case, working from home also generates new trips, an effect referred to as complementarity (Andreev et al., 2010). A theoretical explanation for this complementary effect can be found in the theory of constant travel time budgets (Ahmed & Stopher, 2014; Milakis & van Wee, 2018), which states that people like to spend, on average, roughly the same amount of time (per day) for travelling. On a societal level, this would mean that people balance travel time spent on commuting and travel time spent on leisure^[1]. If the theory holds true, then if one goes down, the other will go up.

2.2. The effects of working from home on travel behaviour

The complex conceptual relations between working from home and travel behaviour make it difficult to adequately assess the relative strengths of the effects that working from home has on travel behaviour. Earlier review studies show that in total, working from home reduces travel demand and thus that the substitution effect is stronger than the complementary effect (Andreev et al., 2010; Mokhtarian, 1991).

Several more recent studies indicate that people who work from home travel either a greater total distance or make more trips than people who do not work from home (de Abreu e Silva & Melo, 2018; He & Hu, 2015; Zhu, 2012; Zhu et al., 2018). Elldér (2020) found that people who work a full day from home travel a smaller distance, but that people who work from home for part of the day travel more often and for longer distances. (Caldarola & Sorrell, 2022) found a similar effect: working from home is associated with greater distances travelled up to a certain point (working from home 3 or more days a week), where they travel less. Wöhner (2022) finds that people who telework travel a similar distance as people who do not. All the above studies find that working from home is related to making fewer commute trips, but more non-commute trips. Summarising then, many recent studies find that people who work from home travel more than people who do not, suggesting that the complementary effect is at least similar in size to the substitution effect.

This raises the question what is causing the difference between the general findings of the earlier and later studies: research design or a real underlying change in the effects over time. There is a striking difference in research design between many earlier and many later studies. Earlier studies typically used data collected from smaller sample sizes, collected specifically around one or multiple projects where people increased the amount of time they would work from home. Later studies more often use data from general travel surveys, generally consisting of larger sets of respondents that are meant to be representative for a region or country (Elldér, 2020). Earlier studies often collected longitudinal data (from before and after people started working from home), whereas later studies used cross-sectional data.

2.3. The effects of the COVID-19 pandemic on working from home & travel behaviour

The COVID-19 pandemic has had a profound impact both on working from home and travel behaviour. This impact is the result of two related effects: the fear of the pandemic causes people to travel less often, and government mandates aiming to curb the spread of the virus have severely affected people’s ability to employ activities outside of their home.

Research shows that the amount of time people worked from home increased during the COVID-19 pandemic in many countries, all over the world (Beck & Hensher, 2020a; de Haas et al., 2020; Downey et al., 2021; Kolarova et al., 2021; Rafiq et al., 2022; Yilmazkuday, 2020; N. Zhang et al., 2021). This is the result of policies by many national governments, which aimed to inhibit the transmission of the virus by reducing contacts between people both at work and whilst travelling to work. This sudden development contrasts the gradual growth of working from home over the past decades (Mokhtarian, 2020; Stiles & Smart, 2021). Many people started working from home during the pandemic and discovered its benefits and drawbacks for the first time. This means that people were generally not able to self-select into work circumstances that allowed them to work from home, but rather were either forced to do so or suddenly found themselves able to. The effects of working from home of these new groups of people can be vastly different to those found before the pandemic, based on a much narrower group. Finally, studies find that a substantial number of people that started working from home have been content with this change in their life and expect to work from home post-pandemic as well (Jain et al., 2022; Shortall et al., 2021). This means that increased levels of working from home will have a lasting impact on working from home, even after the pandemic and the associated measurements are no longer a driving force behind working from home adoption.

Simultaneously, activity-travel patterns changed drastically during the pandemic as well. In general, commuting travel decreased (Beck et al., 2020; Beck & Hensher, 2020b; de Haas et al., 2020; Shamshiripour et al., 2020). This decline was most substantial for public transport (Beck et al., 2020; de Haas et al., 2020). Some studies report a shift towards travel for leisure purposes, especially those made on foot or on a bicycle (de Haas et al., 2020; J. Zhang et al., 2021). These changes can only partially be explained by the increase in working from home: other contact-reducing measures (up to and including lockdowns) and the general fear of being infected with the disease of course played a much larger role as well.

2.4. Conceptual Model

The relationships discussed in Section 2.1 are graphically summarized in the conceptual model for this study, which is given in Figure 1. We discussed that there are two distinct effects of working from home on travel behaviour: a negative effect on commute travel, referred to as substitution, and a positive effect on leisure travel, referred to as complementarity. One possible mechanism behind these substitution and complementary effects is that of a constant travel time budget, which should then also be observed for other reasons that are not related to working from home.

It is important to point out here that the above conceptual model is an overview of the relationships between working from home and travel behaviour that are expected to exist, based on the overview of the literature. As will be clarified later in the discussion of the research method and the operational model, we empirically test all possible relationships between the concepts in this conceptual model. The conceptual model thus served as a theoretically underpinned base for the further analyses – both how they are set up and how they are interpreted.

¹We use a broad definition of leisure travel, encompassing all trips made for non-work and non-education related purposes. Stricter definitions of leisure travel also exclude habitual trips made for maintaining one’s lifestyle, such as groceries. We do include these trips in our definition of leisure travel.

This section describes the data and methods used to investigate the relations visualized in the conceptual model. The data panel is described first, followed by an operationalisation of the concepts into variables. Then descriptive statistics of the data are presented, followed by a description of the research method. Finally, the section culminates in an operational model and the model specification and estimation.

3.1. Data & Sampling

To investigate the relationship between working from home and travel behaviour, we use data from the MPN, a longitudinal household panel that consists of a 3-day travel diary and a set of questionnaires (Hoogendoorn-Lanser et al., 2015). We use five waves of data from the MPN, spanning the years 2017 through 2021. Effectively, this gives us three waves of data collected before the COVID-19 pandemic and two waves collected during the COVID-19 pandemic.

Since we are interested in the relationship between work and travel behaviour, we only use respondents that worked at least 16 hours per week during each wave. Furthermore, we use a pure-stayer sample of respondents who fully participated in all the five waves. The use of a pure-stayer sample allows us to more directly compare the effect sizes found before and during the pandemic, as they are based on data generated by the exact same people for both time periods. The final pure-stayer sample consists of 1100 respondents. A possible drawback of using a pure-stayer sample is the potential of non-random dropout, which could lead to a skewed data set. To test whether this is the case, we give a distribution of socio-demographic characteristics for both the pure-stayer sample and the cross-sectional sample of the 2019 MPN wave in Table 1.

Table 1 Sample distribution of both the pure-stayer sample and participants of the 2019 wave

		Sample (pure stayer, %)	Sample (2019 wave, %)
Gender	Male	52	50
Gender	Female	48	50
Age (Years)	18-29	8	15
	30-39	31	27
	40-49	25	26
	50-59	27	24
	60+	9	8
Education	Low	11	14
	Medium	38	40
	High	52	46
Net Personal Income (€/month)	Low (< €2000)	32	38
	Modal (€2000 to €3000)	46	43
	High (> €3000)	22	19
Urban Density (addresses/m²)	<500	7	9
	500-1000	20	21
	1000-1500	17	18
	1500-2500	33	32
	>2500	23	20
Household type	Single	26	20
	Adult household	39	44
	Kids < 13	29	28
	Kids > 13	7	8

This comparison between the socio-demographics of the pure-stayer sample and the cross-sectional sample, given in Table 1, shows that drop-out is indeed higher in some subgroups, leading to a slightly skewed sample. In particular, younger people (younger than 29 years) seem to drop out more often.

3.2. Operationalisation

The main interest of this research is to a) determine the direction and strength of the (causal) relations between working from home and travel behaviour and b) find whether these effects changed during the COVID pandemic. To find the bidirectional (causal) relations between working from home and travel demand we need to include four variables within each wave.

First, we include the weekly number of hours worked from home and the weekly number of hours worked in total. These two variables allow us to effectively capture the effect of working more from home, whilst controlling for the effects of simply working more. Effectively, we need to separate increases in working from home that result from a simple increase in the total number of hours worked and increases that result from a shift from hours worked in a separate work location (for example an office) to hours worked from home. These variables are measured using a questionnaire, where respondents are prompted to answer how many hours on average they worked in total in the last few weeks. Afterwards, they distribute these hours over the location where they worked, giving us the weekly number of hours spent working from home.

For travel behaviour, we distinguish two types of travel. The first is commute travel, which we define as travel with the purpose of moving to and from the work location. Second, we distinguish leisure travel, which we broadly define as all travel that is not made specifically either for work (= either commuting or business travel) or educational purposes. We operationalise travel behaviour as the time spent traveling for either commuting or leisure purposes during the 3-day observational period of the MPN. We choose to use travel time, as opposed to travel distance, since we assume that people base their decisions on changes in travel time. In other words, we expect that the time saved by commuting less because of working from home is the most tangible benefit felt by people who work from home, and that it is this saved up time they might spend on travelling for other purposes. This is also the main thought behind the theory of constant travel time budgets, which we can tie into our research in this way.

We extend the model beyond these four variables in two ways. First, we include the number of weekend days that are included in a respondent’s travel diary allotment. MPN respondents keep 3-day travel diaries, where the allotted weekdays are kept stable across waves. Some respondents then have 1 or 2 weekend days in these allotted days, whereas others have none. We need to control for these differences to get more accurate between-person relations for time spent travelling. Second, we want to allow for some form of heterogeneity with respect to travel modes in the model. This adds information that is highly relevant to policy makers, road operators, and transit planners. To do so, we add the most often used commute mode in the last pre-pandemic wave (2019). This allows us to attain information on the extent to which commute demand for the various modes has been affected by working from home because of the pandemic. The commute mode is operationalised using two dummy variables: people who commuted by public transport and people who commuted by car. The reference group is then the people who commuted by other travel modes.

3.3. Data description

Means and standard deviations for the four time-variant variables are given in Table 2.

Table 2 Means and standard deviations for the time-variant variables

		2017	2018	2019	2020	2021
Hours worked (hours/week)	Mean	35	35	35	35	35
	Median	36	36	36	36	36
	SD	8.9	8.6	8.5	8.4	8.0
Hours WFH (hours/week)	Mean	3	3	3	12	10
	Median	0	0	0	4	3
	SD	6.3	6.8	6.5	15	13
Commute travel time (hours / 3 days)	Mean	1.30	1.27	1.23	0.67	1.30
	Median	1.00	0.92	0.80	0.17	1.00
	SD	1.43	1.43	1.47	1.07	1.43
Leisure travel time (hours /3 days)	Mean	1.98	2.15	2.03	1.73	1.98
	Median	1.45	1.62	1.57	1.20	1.45
	SD	1.95	2.12	1.95	1.90	1.95

All variables’ mean and standard deviation values are stable across the three pre-pandemic years. This stability disappears when comparing both 2020 and 2021 to the years before the pandemic, except for the number of hours worked per week. The average number of hours worked from home substantially increased in the year 2019/2020, before levelling off at an average of 10 hours worked per week in 2021. Both commute and leisure travel time decreased: from 74 minutes per 3 days for commute travel in 2019 to 40 minutes in 2020 and from 122 minutes per 3 days for leisure travel to 104 minutes. The decrease in commute travel time thus was more substantial than the decrease in leisure travel time.

Correlations between the time-variant variables for both the years before and the years during the COVID-19 pandemic are given in Table 3.

Table 3 Correlations between time-variant variables

	Hours worked	Hours WFH	Commute travel time	Leisure travel time
Hours worked	1	0.31	0.13	-0.041
Hours WFH	0.26	1	-0.24	0.040
Commute travel time	0.20	-0.068	1	-0.11
Leisure travel time	-0.035	0.057	-0.18	1
Bivariate correlations in the lower-left (bold) are from the pooled 2017-2019 data; correlations in the upper-right (italic) are from the pooled 2020-2021 data

As expected, based on the conceptual model, the correlation between working from home and commute travel is negative both before and during the COVID-19 pandemic. The magnitude of the correlation is much larger during than it was before the pandemic. The correlation of hours worked from home and leisure travel time is positive – again in line with the expectation set out in the conceptual model – and did not change in magnitude nearly as much. The correlations between commute and leisure travel time are negative, and the correlations between working and working from home are positive.

3.4. Research method

To determine the longitudinal relations between the variables, we use a random-intercept cross-lagged panel model (RI-CLPM), which is an extension of the cross-lagged panel model (CLPM; (Finkel, 2011). The CLPM is an often-used model to study panel data, with the purpose of empirically testing the bidirectional effects between multiple concepts that are measured over time. To do so, the CLPM specifies auto-regressive relationships, which are supposed to control for the stability of a variable over time. The cross-lagged relationships between the constructs are then supposed to represent the causal processes between the variables. As pointed out in Hamaker et al. (2015), this approach assumes that the value of each variable for every person varies over time around the same sample mean. This assumption is problematic, as most variables do in fact contain stable differences between individuals. For example, some individuals will persistently work more hours per week than others across all measurements, which is ignored by the CLPM.

Hamaker et al. (2015) therefore argues that researchers should not only control for temporal stability across the sample, but also for the time-invariant stability of each variable on the level of the individual. Doing so effectively separates within-person effects over time from constant between-person differences. This is achieved by including random intercepts, which account for the trait-like, time-invariant stability of the variables. The random intercepts thus capture the between-persons differences, allowing the (auto)-regressive structure to specifically capture within-person effects. Figure 2 contains a graphical representation of both the base CLPM (in black) and the additions of the RI-CLPM (in dark red).

The resulting auto-regressive coefficients can then be interpreted as within-individual carry-over effects (Mulder & Hamaker, 2020), meaning that a positive effect indicates that a higher (or lower) than expected value is likely to also have a higher (or lower) than expected value during the next observation, where the expected value is based on the average, trait-like, value per respondent. Similarly, the cross-lagged effects represent that an individual with a higher than expected value on one variable also has a higher than expected value on the other variable in the next measurement. These effects are directional and since they represent within-individual changes, they can be more correctly assumed to represent causal processes on the within-individual level than parameter estimates from CLPMs. The RI-CLPM also allows us to estimate the correlations between the random intercepts, which can be interpreted as the general between-person relations between the concepts associated with the random intercepts (Mulder & Hamaker, 2020).

We use two extensions to the model discussed by Mulder & Hamaker (2020), both relating to the incorporation of time-invariant predictors of the time-variant variables. Time-invariant variables are stable over time, and in this model specification we expect them to have affected the time-variant variables. The first extension is the use of the number of weekend days as a predictor of the random intercepts. Effectively, this controls the trait-like stability of the variables for the differences in our measuring instrument that allots a different number of weekend days to the respondents. Second, we regress the time-variant variables for the waves during the pandemic on the commuting mode used in the last wave before the pandemic. This gives us additional insight into the effects of working from home on the use of the different travel modes.

Finally, we test whether the within-person effects are different during the COVID-19 pandemic than they were before the pandemic by imposing equality constraints on the effects before the pandemic (2017 -> 2018 -> 2019) and on the effects during the pandemic (2019->2020->2021). We effectively estimate two sets of effects, allowing a direct comparison between the effects before- and during the pandemic.

3.5. Model specification and estimation procedure

A schematic overview of the variables and relationships is given in the model in Figure 3. For the sake of visual clarity, the random-intercept structure and the correlations between variables are omitted from this figure. The regression structure is simplified as well. The arrow from work to work between 2017 (denoted with an a) for instance contains four different effects (cross- and autoregressive parameters between total hours worked and hours worked at home). These four effects are each kept stable between the two pre-pandemic wave-pairs. In total, there are sixteen regression parameters for each wave pair.

The models are estimated using the R-package Lavaan (Rosseel, 2012). First, a CLPM is estimated. The fit of this model is unsatisfactory, based on most goodness-of-fit statistics for structural equation models as given in Hooper et al. (2008): chi-square value is 1199 with 176 degrees of freedom (p-value is 0.000). RMSEA is 0.073 and CFI is 0.912. Following the estimation of the CLPM, a RI-CLPM is estimated. Model fit of the RI-CLPM is much better. Chi-square value of the model is 338.9 with 154 degrees of freedom, for a p-value of 0.000. This means that strictly speaking the chi-square test rejects the hypothesis that the model-based covariance matrix fits the sample covariance matrix. However, the chi-square test is sensitive to larger sample sizes, and the chi-square value divided by the degrees of freedom indicates good model fit (Hooper et al. 2008). The RMSEA is 0.043, which also indicates good model fit. Finally, the CFI is 0.975, again indicating good model fit.

This section describes and interprets the empirical results from the estimated model fitted to longitudinal data in the Netherlands. The section is structured around the expected relations from the conceptual model given in Section 2. First, the results relating to the effects of working from home on both commute and leisure travel time are interpreted. Then, the effects of travel time on working from home and the relations between the two travel time variables are discussed. Finally, we discuss the stability of the variables and take a brief look at the effect of commuting by different modes.

As described in Section 3, the RI-CLPM estimates two sets of effects: the within-person directional effects, which vary over time, and the between-person correlations, which do not vary over time. Both sets can be analysed together to attain a comprehensive overview of the relationships between working from home and travel behaviour.

4.1. The effects of working from home on travel time

The relationship between working from home and travel time is interpreted using both the within-person and between-person coefficients estimated in the RI-CLPM. First, we will look at the within-person estimates. The unstandardized within-individual effects between the four time-variant variables, as well as the effect of the time-invariant predictor is given in Table 4. As explained in Section 3.4, the auto-regressive coefficients can be interpreted as within-individual carry-over effects and the cross-lagged coefficients represent that an individual with a higher than expected value on one variable also has a higher than expected value on the other variable in the next measurement.

Table 4 Unstandardized within-individual effects between time-variant variables

	Dependent variables
Independent variables	Effects before COVID-19 pandemic (2017 – 2019)				Effects during COVID-19 pandemic (2019 – 2021)
	Work hours	WFH hours	Commute time travelled	Leisure time travelled	Work hours	WFH hours	Commute time travelled	Leisure time travelled
Time-variant variables
Work hours	0.422***	0.034	0.003	-0.014	0.221***	0.035	-0.011	-0.018
WFH hours	0.035	0.405***	-0.016**	0.011	0.018***	0.602***	-0.018***	0.008**
Commute time travelled	0.061	-0.264***	0.365***	0.014	-0.238**	-0.378	0.006	0.085**
Leisure time travelled	0.002	-0.111	0.035**	0.056*	-0.204***	-0.266*	0.012	0.110***
Time-invariant predictors
PT Commuter^a					0.879*	9.467***	-0.165	0.216
Car Commuter^a					0.437*	-0.174	0.039	-0.111
In this and following tables, ^*p<0.01, ^ p<0.05, and ^*p<0.1 a. reference: all other travel modes as main commute mode

We find negative effects from working from home on the travelled commute time both before the pandemic (-0.016) and during the pandemic (-0.018). Hence, this effect did not change much during the pandemic compared to before the pandemic. The unstandardized effect of working from home on leisure travel time became smaller but did become statistically significant. These estimates provide evidence for both the substitution and complementary effects, however, the latter is only found to be statistically significant during the pandemic. In either case, the substitution effect appears to be stronger compared to the compensatory effect. Furthermore, COVID-19 has not had a substantial effect on the unstandardized effects of working from home on travel behaviour.

However, it is important to realize that the standardized estimates can still change drastically, even if the unstandardized estimates are relatively stable. Since we saw in Section 3.3 that the distribution of the variables changed because of the COVID-19 pandemic, the standardized estimates are likely to change. In particular, the smaller standard deviation of travel time, combined with the larger standard deviation of working from home, will result in a larger standardized effect of working from home on travel time during the pandemic than before the pandemic. We do indeed find that the standardized effect of working from home on commute time, which unstandardized was relatively stable, roughly triples during the pandemic (from -0.063 to -0.204). An increase of one standard deviation of working from home thus has a much greater effect on the commute time travelled. Even though the unstandardized effect itself was relatively stable, working from home determined to a much greater extent the commute travel time than before the pandemic.

The interpretation of these within-person effects can be complemented by an analysis of the between-person relations, estimated by the RI-CLPM as the correlations between the random intercepts. These correlations represent the time-invariant relations between the variables across the respondents. The random intercepts catch all time-invariant confounders and give an indication of the general relations between the variables. Table 5 contains the correlation coefficients between the four random intercepts.

Table 5 Correlations between the random intercepts

	Hours worked at home	Commute travel time	Leisure travel time
Hours worked	0.197***	0.478***	-0.068*
Hours worked at home		-0.075	0.115**
Commute travel time	-		-0.090*

The positive correlation (0.197) between the number of hours worked and the number of hours worked at home for example indicates that people who work more hours also work more hours from home. Similarly, people who work more also tend to commute more (correlation = 0.478). So far, these findings are not very surprising.

However, that changes when we turn our attention to the relations between working from home and travel time. We see an insignificant correlation (-0.075) between working from home and commute travel time, which is different compared to the significant negative within-person effects between working from home and commute travel time. This means that people who work from home do not, in general, travel less time for commuting: there is no significant between-persons relationship between commuting time and working from home. However, variations within the individual do have a significant effect: a higher than expected value for working from home results in a lower than expected value for commute travel time. The positive correlation between working from home and leisure travel time (0.115) aligns more closely with the within-persons effects, although there is a discrepancy here too. Before the pandemic, we found no statistically significant effect of working from home on leisure travel time. We only found such an effect during the pandemic.

The within-persons and between-persons relations between travel time and working from home thus seem to differ from one another. This implies that research methods that are unable to separate the within-persons from between-persons effect will, to varying extents, either under- or overestimate the effects of working from home on travel behaviour.

4.2. The relations between travel time and the effects on working

Based on the theory of constant travel time budgets, we expected a negative relationship between commute and leisure travel time. However, the model estimates either insignificant or positive within-person effects between these two variables. One explanation might be that as people move through life, they go through life-events that result in an increase in mobility in general. Furthermore, it is likely that the effect during the pandemic at least partly reflects a temporary (un)willingness to travel due to underlying concerns about an infection with the virus. Turning our attention to the between-person correlations between the two travel time variables however, we do find a negative relationship between commute and leisure travel time. This means that people who spend more time travelling for leisure purposes in general spend less time commuting. Again, the within-persons and between-persons effects seem to differ markedly.

We also find some effects of commute and leisure travel time on working from home. Before the pandemic, we find a negative within-persons effect from commute time on working from home. As explained in Section 2, we expected a positive effect due to people with longer commutes netting a larger gain by working from home more often. One explanation for this difference can be found in the operationalisation of commute travel time as the total amount of time travelled for commuting purposes. This does not necessarily encompass the amount of time travelled per commute.

During the pandemic, we find significant negative within-person effects of both commute and leisure travel time on working hours. This suggests that people that start to travel less, will then work less hours in the following year. There is no clear theoretical explanation for this finding. Perhaps the model is picking up certain life-events that more frequently occurred during the pandemic, where changes in travel and work are related.

4.3. Stability & Mode heterogeneity

The sizes of the auto-regressive parameters also differ between the estimates before and during the pandemic. Most striking is the complete disappearance, both in size and statistical significance, of the auto-regressive parameter of travelled commute time. This means that people who commuted more than expected in one measurement did not necessarily do so in the following measurement. This is most likely related to both the much stronger effect of working from home and the much higher prevalence of working from home in general.

Finally, we also included the most-used commuting mode during the last measurement before the pandemic (September 2019) as a time-invariant predictor of the variables in the following waves during the COVID-19 pandemic. The PT commuters differed statistically from all other commuters in terms of number of hours they work from home. People whose most often used commuting mode before the pandemic was PT are more likely to work from home during the pandemic (roughly 10 hours more per week compared to the reference). Car commuters were not more likely to work from home during the pandemic. Both PT and car commuters worked slightly more per week than people who used other modes to commute.

In this study, we set out to a) get more insights on the (bidirectional) relations between working from home and travel behaviour and b) whether these relations changed because of the COVID-19 pandemic. To do so, we estimated cross-lagged structural equation models on panel data pertaining both to the situation before and during the COVID-19 pandemic.

The results show that as expected, people who increase the amount of time they spent working from home reduce their commute travel. We also find that an increase in working from home results in an increase in leisure travel time, but this effect could only be established during the COVID-19 pandemic. Both before and during the pandemic, the negative effect of working from home on commuting travel time is greater than the positive effect on leisure travel time. The net effect of working from home on the amount of time people spent travelling thus seems to be negative both before and during the pandemic.

This finding contradicts most of the recent literature on the effects of working from home on travel behaviour, where people who work from home are generally found to travel more than people who do not (Caldarola & Sorrell, 2022; Wöhner, 2022). The results show that the between-person relations are more similar to these findings, as the positive association between leisure travel time and working from home is greater than the (not-significant) negative association between commute travel time and working from home. We postulate therefore that the difference between these findings at least partly result from our panel design, and that separating the within-person and between-person relations offers new insights into the relations between working from home and travel behaviour.

Turning our attention to the differences in the relations between working from home and travel behaviour before and during the pandemic, we see that the effect of an additional hour worked from home on commute travel time was relatively stable. However, the extent to which commute travel time is determined by working from home nearly tripled during the pandemic. This is the result of both a smaller variance in commute travel time and a larger variance in working from home. The extent to which working from home determines commute behaviour after the pandemic will in our expectations be somewhere in-between the pre-pandemic and post-pandemic relationships. We base this expectation on the indications that post-pandemic levels of working from home will be lower than during the pandemic, but higher than were before the pandemic.

Based on the literature, we expected that the compensation effect could partly be explained by the theory of constant travel time budgets, which states that people on average spend a similar, constant amount of time on total travel. If this is the case, then we would also find negative direct relations between commute and leisure travel time. We do indeed find a negative relation between commute and leisure travel time between-persons, indicating that people who commute more in general spend less time traveling for leisure purposes. However, within-persons we find positive effects. A person that spends more time travelling for commute purposes will then also spend more time travelling for leisure purposes. This would indicate that the mechanism behind the theory of constant travel time budgets works on the between-person level, for example due to negative feedback loops between-persons as a result of congestion or crowding. However, our within-person findings could be biased, either due to the fact that people in a panel only get older and thus go through certain life-events that might make them more or less mobile in general, or due to COVID-19 resulting in some people travelling less in general due to the threat of the virus.

Finally, we find that PT commuters were much more likely to work from home more hours per week than non-PT commuters. Working from home has thus affected public transport demand much more than the demand for the other modes. This finding is in-line with more descriptive analyses of the adoption and use of working from home because of the pandemic.

These conclusions have implications for policy makers, road authorities, and public transport operators. First, we do find that working from home reduces commute travel time. Working from home thus seems to be an effective strategy for policy makers to reduce peak-hour travel (which is largely based on commute travel) and thus congestion on roads and crowding in public transport. The increase in leisure travel (found only during the COVID-19 pandemic) can partly compensate the reduced commute travel, but this shift will probably lead to a more spread-out activity pattern across the day and thus smaller peaks in travel demand. This finding is in line with expectations of the effects of the pandemic on travel behaviour in the literature (van Wee & Witlox, 2021). It is important to note that working from home is only one of many factors affecting total travel demand, and that the relevant actors should not overestimate its effects compared to all other factors that influence travel demand compared to for example broader economic or socio-demographic factors. Second, since the extent to which working from home determined commute travel time was much greater during the pandemic than it was before the pandemic, policy makers and transport operators should be careful when extrapolating the apparent influence of working from home on travel demand that we’ve seen during the pandemic.

Future research is needed to further understand the relations between working from home and travel behaviour. A few specific points to address are mentioned below. First, this study uses data from before and during the pandemic to set expectations for future relations between working from home. The pandemic is not fully over, and it will be especially important to keep doing research into the relations as the world transitions out of the pandemic. Second, our study only partly addressed the heterogeneity of the impact of working from home with respect to travel modes. More in-depth studies into the varying effects of working from home on especially public transport and active mode use would be a valuable addition. Studies could also investigate other sources of heterogeneity, such as differences between urban and rural areas or specific groups of workers. The last limitation of this study that we want to highlight here is that the effects are calculated using fixed timespans of one year between measurement waves. The effects of working from home are expected to vary over time, with people for example making different long-term decisions due to working from home. Hence, estimating effects of working from home on travel behaviour using different spans of time could be a promising avenue for future research as well.

Author Contributions Statement (CReDiT):

R.M. Faber – Conceptualization, Data curation, Methodology, Formal Analysis, Visualization, Writing – original draft

M. Hamersma – Conceptualization, Project administration, Supervision, Writing – review & editing

J. Brimaire – Formal Analysis, Investigation, Writing – original draft

M. Kroesen – Conceptualization, Methodology, Supervision, Writing – review & editing

E.J.E. Molin – Conceptualization, Methodology, Supervision, Writing – review & editing

Competing Interests

The authors declare that they have no known competing financial or non-financial interests that could have appeared to influence the work reported in this paper.

Funding

The authors received no specific funding for this work

Ahmed, A., & Stopher, P. (2014). Seventy Minutes Plus or Minus 10 — A Review of Travel Time Budget Studies. Transport Reviews, 34(5), 607–625. https://doi.org/10.1080/01441647.2014.946460
Andreev, P., Salomon, I., & Pliskin, N. (2010). Review: State of teleactivities. Transportation Research Part C: Emerging Technologies, 18(1), 3–20. https://doi.org/10.1016/j.trc.2009.04.017
Beck, M. J., & Hensher, D. A. (2020a). Insights into the impact of COVID-19 on household travel and activities in Australia – The early days under restrictions. Transport Policy, 96, 76–93. https://doi.org/10.1016/j.tranpol.2020.07.001
Beck, M. J., & Hensher, D. A. (2020b). Insights into the impact of COVID-19 on household travel and activities in Australia – The early days of easing restrictions. Transport Policy, 99, 95–119. https://doi.org/10.1016/j.tranpol.2020.08.004
Beck, M. J., Hensher, D. A., & Wei, E. (2020). Slowly coming out of COVID-19 restrictions in Australia: Implications for working from home and commuting trips by car and public transport. Journal of Transport Geography, 88. https://doi.org/10.1016/j.jtrangeo.2020.102846
Caldarola, B., & Sorrell, S. (2022). Do teleworkers travel less? Evidence from the English National Travel Survey. Transportation Research Part A: Policy and Practice, 159, 282–303. https://doi.org/10.1016/J.TRA.2022.03.026
de Abreu e Silva, J., & Melo, P. C. (2018). Does home-based telework reduce household total travel? A path analysis using single and two worker British households. Journal of Transport Geography, 73, 148–162. https://doi.org/10.1016/j.jtrangeo.2018.10.009
de Haas, M., Faber, R., & Hamersma, M. (2020). How COVID-19 and the Dutch ‘intelligent lockdown’ change activities, work and travel behaviour: Evidence from longitudinal data in the Netherlands. Transportation Research Interdisciplinary Perspectives, 6, 100150. https://doi.org/10.1016/j.trip.2020.100150
de Vos, D., Meijers, E., & van Ham, M. (2018). Working from home and the willingness to accept a longer commute. Annals of Regional Science, 61(2), 375–398. https://doi.org/10.1007/s00168-018-0873-6
Downey, L., Fonzone, A., Fountas, G., & Semple, T. (2021). Impact of COVID-19 on travel behaviour, transport, lifestyles and location choices in Scotland. https://doi.org/10.48550/arxiv.2104.10440
Elldér, E. (2020). Telework and daily travel: New evidence from Sweden. Journal of Transport Geography, 86. https://doi.org/10.1016/j.jtrangeo.2020.102777
Finkel, S. (2011). Causal Analysis with Panel Data. SAGE Publications. https://doi.org/10.4135/9781412983594
Hamaker, E. L., Kuiper, R. M., & Grasman, R. P. (2015). A critique of the cross-lagged panel model. Psychological Methods, 20(1), 102–116. https://doi.org/10.1037/a0038889
Harpaz, I. (2002). Advantages and disadvantages of telecommuting for the individual, organization and society. Work Study, 51(2), 74–80. https://doi.org/10.1108/00438020210418791
He, S. Y., & Hu, L. (2015). Telecommuting, income, and out-of-home activities. Travel Behaviour and Society, 2(3), 131–147. https://doi.org/10.1016/j.tbs.2014.12.003
Hoogendoorn-Lanser, S., Schaap, N. T. W., & Olde Kalter, M. J. (2015). The netherlands mobility panel: An innovative design approach for web-based longitudinal travel data collection. Transportation Research Procedia, 11, 311–329. https://doi.org/10.1016/j.trpro.2015.12.027
Hooper, D., Coughlan, J., & Mullen, M. R. (2008). Structural equation modelling: Guidelines for determining model fit. Electronic Journal of Business Research Methods, 6(1), 53–60. https://doi.org/10.21427/D79B73
Jain, T., Currie, G., & Aston, L. (2022). COVID and working from home: Long-term impacts and psycho-social determinants. Transportation Research Part A: Policy and Practice, 156, 52–68. https://doi.org/10.1016/J.TRA.2021.12.007
Kasraian, D., Maat, K., & van Wee, B. (2018). Urban developments and daily travel distances: Fixed, random and hybrid effects models using a Dutch pseudo-panel over three decades. Journal of Transport Geography, 72, 228–236. https://doi.org/10.1016/J.JTRANGEO.2018.09.006
Kolarova, V., Eisenmann, C., Nobis, C., Winkler, C., & Lenz, B. (2021). Analysing the impact of the COVID-19 outbreak on everyday travel behaviour in Germany and potential implications for future travel patterns. European Transport Research Review, 13(1), 1–11. https://doi.org/10.1186/S12544-021-00486-2/FIGURES/7
Kraut, R. E. (1989). Telecommuting: The Trade‐offs of Home Work. Journal of Communication, 39(3), 19–47. https://doi.org/10.1111/j.1460-2466.1989.tb01038.x
Milakis, D., & van Wee, B. (2018). “For me it is always like half an hour”: Exploring the acceptable travel time concept in the US and European contexts. Transport Policy, 64, 113–122. https://doi.org/10.1016/j.tranpol.2018.02.001
Mokhtarian, P. L. (1990). A typology of relationships between telecommunications and transportation. Transportation Research Part A: General, 24(3), 231–242. https://doi.org/10.1016/0191-2607(90)90060-J
Mokhtarian, P. L. (1991). Telecommuting and travel: state of the practice, state of the art. Transportation, 18(4), 319–342. https://doi.org/10.1007/BF00186563
Mokhtarian, P. L. (2020). The Adoption and Travel Impacts of Teleworking: Will it be Different This Time? https://hbr.org/resources/images/
Mokhtarian, P. L., Collantes, G. O., & Gertz, C. (2004). Telecommuting, residential location, and commute-distance traveled: Evidence from State of California employees. Environment and Planning A, 36(10), 1877–1897. https://doi.org/10.1068/a36218
Mokhtarian, P. L., Handy, S. L., & Salomon, I. (1995). Methodological issues in the estimation of the travel, energy, and air quality impacts of telecommuting. Transportation Research Part A, 29(4), 283–302. https://doi.org/10.1016/0965-8564(94)00029-A
Molloy, J., Schatzmann, T., Schoeman, B., Tchervenkov, C., Hintermann, B., & Axhausen, K. W. (2021). Observed impacts of the Covid-19 first wave on travel behaviour in Switzerland based on a large GPS panel. Transport Policy, 104, 43–51. https://doi.org/10.1016/J.TRANPOL.2021.01.009
Mulder, J. D., & Hamaker, E. L. (2020). Three Extensions of the Random Intercept Cross-Lagged Panel Model. Structural Equation Modeling: A Multidisciplinary Journal, 28(4), 638–648. https://doi.org/10.1080/10705511.2020.1784738
Nilles, J. (1976). Telecommunications-transportation tradeoff: Options for tomorrow. https://dl.acm.org/doi/abs/10.5555/540203
Olson, M. H. (1983). Remote office work: Changing work patterns in space and time. Communications of the ACM, 26(3), 182–187. https://doi.org/10.1145/358061.358068
Rafiq, R., McNally, M. G., Sarwar Uddin, Y., & Ahmed, T. (2022). Impact of working from home on activity-travel behavior during the COVID-19 Pandemic: An aggregate structural analysis. Transportation Research Part A: Policy and Practice, 159, 35–54. https://doi.org/10.1016/J.TRA.2022.03.003
Rosseel, Y. (2012). lavaan: An R Package for Structural Equation Modeling. Journal of Statistical Software, 48, 1–36. https://doi.org/10.18637/JSS.V048.I02
Salomon, I. (1986). Telecommunications and travel relationships: a review. Transportation Research Part A: General, 20(3), 223–238. https://doi.org/10.1016/0191-2607(86)90096-8
Shamshiripour, A., Rahimi, E., Shabanpour, R., & Mohammadian, A. (Kouros). (2020). How is COVID-19 reshaping activity-travel behavior? Evidence from a comprehensive survey in Chicago. Transportation Research Interdisciplinary Perspectives, 7. https://doi.org/10.1016/j.trip.2020.100216
Shibayama, T., Sandholzer, F., … B. L.-E. J. of, & 2021, undefined. (2021). Impact of COVID-19 lockdown on commuting: a multi-country perspective. Journals.Open.Tudelft.Nl, 21(1), 70–93. https://doi.org/10.18757/ejtir.2021.21.1.5135
Shortall, R., Mouter, N., & van Wee, B. (2021). COVID-19 passenger transport measures and their impacts. Https://Doi.Org/10.1080/01441647.2021.1976307. https://doi.org/10.1080/01441647.2021.1976307
Siha, S. M., & Monroe, R. W. (2006). Telecommuting’s past and future: A literature review and research agenda. Business Process Management Journal, 12(4), 455–482. https://doi.org/10.1108/14637150610678078/FULL/PDF
Stiles, J., & Smart, M. J. (2021). Working at home and elsewhere: daily work location, telework, and travel among United States knowledge workers. Transportation, 48(5), 2461–2491. https://doi.org/10.1007/S11116-020-10136-6/TABLES/6
Susilo, Y. O., & Maat, K. (2007). The influence of built environment to the trends in commuting journeys in the Netherlands. Transportation, 34(5), 589–609. https://doi.org/10.1007/S11116-007-9129-5/TABLES/5
van Wee, B., & Witlox, F. (2021). COVID-19 and its long-term effects on activity participation and travel behaviour: A multiperspective view. Journal of Transport Geography, 95, 103144. https://doi.org/10.1016/J.JTRANGEO.2021.103144
Wöhner, F. (2022). Work flexibly, travel less? The impact of telework and flextime on mobility behavior in Switzerland. Journal of Transport Geography, 102, 103390. https://doi.org/10.1016/J.JTRANGEO.2022.103390
Yilmazkuday, H. (2020). COVID-19 spread and inter-county travel: Daily evidence from the U.S. Transportation Research Interdisciplinary Perspectives, 8, 100244. https://doi.org/10.1016/j.trip.2020.100244
Zhang, J., Hayashi, Y., & Frank, L. D. (2021). COVID-19 and Transport: Findings from a World-wide Expert Survey. Transport Policy. https://doi.org/10.1016/j.tranpol.2021.01.011
Zhang, N., Jia, W., Wang, P., Dung, C. H., Zhao, P., Leung, K., Su, B., Cheng, R., & Li, Y. (2021). Changes in local travel behaviour before and during the COVID-19 pandemic in Hong Kong. Cities, 112, 103139. https://doi.org/10.1016/J.CITIES.2021.103139
Zhu, P. (2012). Are telecommuting and personal travel complements or substitutes? Annals of Regional Science, 48(2), 619–639. https://doi.org/10.1007/s00168-011-0460-6
Zhu, P., Wang, L., Jiang, Y., & Zhou, J. (2018). Metropolitan size and the impacts of telecommuting on personal travel. Transportation, 45(2), 385–414. https://doi.org/10.1007/S11116-017-9846-3

No competing interests reported.

Download PDF

Journal Publication

published 23 Jun, 2023

Read the published version in Transportation →

Editorial decision: Major revision
21 Jan, 2023
Reviews received at journal
23 Oct, 2022
Reviewers agreed at journal
17 Oct, 2022
Reviewers invited by journal
17 Oct, 2022
Editor assigned by journal
10 Oct, 2022
Submission checks completed at journal
23 Aug, 2022
First submitted to journal
22 Aug, 2022

You are reading this latest preprint version

The relations between working from home and travel behaviour: a panel analysis

Status:

Journal Publication

Version 1

Abstract

Figures

1. Introduction

2. Literature Overview & Conceptual Model

2.1. Conceptualising the relationship between working from home and travel behaviour

2.2. The effects of working from home on travel behaviour

2.3. The effects of the COVID-19 pandemic on working from home & travel behaviour

2.4. Conceptual Model

3. Research Methods & Data

3.1. Data & Sampling

3.2. Operationalisation

3.3. Data description

3.4. Research method

3.5. Model specification and estimation procedure

4. Results

4.1. The effects of working from home on travel time

4.2. The relations between travel time and the effects on working

4.3. Stability & Mode heterogeneity

5. Conclusion & Discussion

Declarations

References

Additional Declarations

Status:

Journal Publication

Version 1