Table 1 Variables used in analysis.

 Commute mode switch behavior

Multi-class categorical

\({y}_{{ij}}\) records commute mode switch behavior of each commuter between the start and end months of each phase. \({y}_{{ij}}={{y}_{{ij}}}_{s} \sim {{y}_{{ij}}}_{e}\), where \({{y}_{{ij}}}_{s}\) is the start month commute mode type and \({{y}_{{ij}}}_{e}\) is the end month commute mode type. \({{y}_{{ij}}}_{s}\) and \({{y}_{{ij}}}_{e}\) can take the following three values: private car, public transit, and active travel, thus \({y}_{{ij}}\) has nine categories in total.

 Carbon emissions change

Continuous

\({y}_{{ij}}\) records carbon dioxide emission change per kilometer of each commuter between the start and end months of each phase due to commute mode switch.

 Phase

Dummy

\({t}_{j}=1\) if phase is j and \({t}_{j}=0\) if phase is not j.

 Gender

Dummy

\({X}_{{ij}}=1\) for male commuters, and \({X}_{{ij}}=0\) for female commuters.

In the pandemic context, men and young people were less sensitive to crowding, and they were more likely to travel by public transit than private cars and active modes.

Das et al. (2021)

Basnak et al. (2022)

Mussone and Changizi (2023)

Liu and Lee (2023)

 Age

Continuous

\({X}_{{ij}}=27\) for commuter aged 25–29, \({X}_{{ij}}=32\) for commuter aged 30–34, \({X}_{{ij}}=37\) for commuter aged 35–39, \({X}_{{ij}}=42\) for commuter aged 40–44, \({X}_{{ij}}=47\) for commuter aged 45–49, and \({X}_{{ij}}=52\) for commuter aged 50–54.

 Affluence index

Dummy

\({X}_{{ij}}\) corresponds to six categories ranging from 3 to 8, with 3 being the least affluent and 8 being the most affluent. The raw data has nine categories ranging from 1 to 9, and categories 1–3 and 8–9 are combined respectively because they contain too few observations.

In the pandemic context, high-income individuals were more sensitive to crowding, and they were more likely to travel by private cars than public transit; low-income individuals had less control over their travel mode choices, and they were more likely to be captive riders of affordable modes such as public transit.

Das et al. (2021)

Dingil and Esztergár-Kiss (2021)

Parker et al. (2021)

Basnak et al. (2022)

He et al. (2022)

 Commute time

Continuous

\({X}_{{ij}}\) equals the logarithm of the median commute duration of each commuter in each phase.

In the pandemic context, longer travel time increased the possibility of traveling by private cars versus public transit, and the possibility of traveling by public transit versus active modes; long travel time could prevent mode switch.

Das et al. (2021)

Dingil and Esztergár-Kiss (2021)

Mussone and Changizi (2023)

 Spatial position of workplace

Dummy

\({X}_{{ij}}\) corresponds to five categories ranging from 2 to 6, representing the number of the smallest ring in which each commuter’s work/home location is positioned.

In the pandemic context, urban areas with different development levels, land use characteristics and population composition responded differently in terms of travel behavior.

Hu and Chen (2021)

Liu and Lee (2023)

 Spatial position of residence

 Bus accessibility of workplace

Dummy

\({X}_{{ij}}=1\) if the \(1.5{km}\times 1.5{km}\) grid containing the commuter’s work/home location has at least one bus stop/subway station, and \({X}_{{ij}}=0\) otherwise.

In the pandemic context (and in general), people with higher accessibility to public transit were more likely to travel by public transit.

Cheng et al. (2020)

Das et al. (2021)

 Bus accessibility of residence

 Subway accessibility of workplace

 Subway accessibility of residence

 Season

Dummy

\({X}_{{ij}}=1\) for summer phases (Apr-Nov), and \({X}_{{ij}}=0\) for winter phases (Nov-Apr)

In general, travel mode choice is sensitive to weather conditions and seasonality; active modes are most influenced by weather factors.

Böcker et al. (2016)

Hyland et al. (2018)

 Initial commute mode

Dummy

\({X}_{{ij}}\) corresponds to five categories (private car, bus, subway, bicycle, and walking) recording each commuter’s commute mode in the start month of each phase.

In general, travel mode choice is habitual.

Verplanken et al. (1997)

Wood et al. (2002)

Zhao and Gao (2022)