Effects of Covid-19 pandemic on mode choice behavior of working Filipinos in Metro Manila

Abstract

The COVID-19 pandemic and the resulting control measures by the government impacted the travel behavior of Filipinos. With work trips being a major part of trips generated, this study aimed to investigate the effects of the pandemic on working Filipinos' mode choice. Data were collected from employed residents of Metro Manila using an online survey. About 48% of the respondents preferred using public transportation for work pre-pandemic. This decreased to 22% during the transport lockdown and increased slightly to 25% after the lockdown was lifted. Active transport share increased by 3% during the lockdown but went back down after it was lifted. Using multinomial logistic regression, the significant factors that influence the pre-pandemic respondents’ mode choice were age, household income, travel cost, and vehicle ownership. During the lockdown, travel distance, travel time, and sex assigned at birth became significant also. After the lockdown, the significant factors reverted to that pre-pandemic.

1. Introduction

1.1. Background

The Coronavirus Disease 2019(COVID-19) is considered a global pandemic mainly due to the speed and scale of the transmission of this disease (Henri and Kluge, 2020). The Philippine government initially responded to the threat of COVID-19 by imposing control and preventive measures to prohibit unnecessary mobility circulation such as lockdowns. The first-ever lockdown imposed in the country, the Enhanced Community Quarantine (ECQ), started on March 16, 2020. This type of quarantine is described as the implementation of temporary measures which impose stringent limitations on the movement and transportation of people (Inter-Agency Task Force, 2020). In this study, the period before March 2020 was considered the period “before” transport lockdown. The second period, labeled as “during” transport lockdown, was from March 2020 to August 2021 since the latest ECQ implemented at the time of writing in the chosen cities happened until August 20, 2021 (Manuel, 2021). Lastly, the third period was after August 2021 or “after” the transport lockdown, which is the period after the latest ECQ at the time.

These travel restrictions and guidelines implemented by the government caused a reduction in the service capacity of transport systems. As a result, a shortage in supply was observed, specifically in public transportation in Metro Manila (Ramos, 2021). Despite this, people still had to go out for different purposes, with work trips being one of the most common (Abdullah et al., 2020). As not everyone can work from home and/or have private vehicles, an estimated 4.98 million public transport trips are needed daily if 50% of the pre-pandemic travel demand returns (Rey, 2020). The limited transportation availability and high passenger demand during the lockdowns caused crowding in the streets and delays in usual work trips.

A literature review identified how the pandemic affected work trips in different countries. In the US, work-related travel dropped from 2.2 average daily trips made pre-pandemic to 1.7 during the pandemic (Tefft et al., 2021). In Spain, mobility to workplaces decreased to 80% compared, with pre-pandemic trends with public transport being the most affected mode (Awad-Nunez et al., 2021). In India, a study revealed that about 45% reported no work-related travel, 23.6% reduced travel, and the rest answered the same as pre-pandemic (Pawar et al., 2021). Though there are numerous existing studies related to this topic in other countries, the research on the effects of the COVID-19 pandemic on mode choice behavior in the Philippines is very limited. To fill up this gap, the study focused on answering how the mode choice behavior of employed Filipinos in the country has changed because of the effects of the pandemic. Moreover, this can help inform policy researchers and makers in better preparing for future crisis-related lockdowns and planning labor-related transport policies in the country.

1.2. Objectives

The main purpose of the study was to investigate how employed Filipinos' mode choice behavior in Metro Manila, Philippines changed in relation to the COVID-19 pandemic-related lockdowns. Specifically, the study sought to

• 1.Determine the significant factors that affect the mode choice of working Filipinos in Metro Manila; and
• 2.Identify the effects of the COVID-19 pandemic-related lockdowns on these factors, and on their mode choice distribution.

There was a limited sampling due to the pandemic. Those who do not have access to the internet were not able to take part in the study since an online survey was utilized for data gathering. The survey was also conducted during the third period (after transport lockdown) only, thus, the respondents had to refresh their minds to answer for the first two periods (before and during transport lockdown). In addition, the number of participants who answered the survey was computed using the predicted working population in Metro Manila.

2. Review of related literature

2.1. Effects of the COVID-19 pandemic on mode choice

Transport and work-related policies induced by the COVID-19 pandemic, coupled with the perception of people of their safety from the COVID-19 virus, caused a shift in travel behavior (Bhaduri et al., 2020; De Vos, 2020). One of the main effects of the pandemic-related policies seen throughout different countries is the shift from public transport to private modes due to the desire to reduce transmission. During the initial month of the COVID-19 pandemic, public transport comprised 7% of the trips in Australia, down from 15% the months before (Beck and Hensher, 2020), while in Budapest, the modal share of public transportation declined from 43% to 18% and private vehicle use rose from 43% to 65% (Bucksy, 2020). There had also been an increase in the use of active modes such as walking and cycling due to similar reasons for avoiding closed-contact enclosed modes. Abdullah et al. (2020) showed a shift in the travel mode of 1203 people from different countries. From 32% before the pandemic, the modal share of the private vehicle turned 39%, while active transport comprised 20% from the initial 12%. Similarly, in the study by Beck and Hensher (2020), the use of active transportation rose to 20% from 14% of the trips before the pandemic.

At the time of writing, published research on similar effects of the pandemic-related lockdowns in mode choice behavior in the Philippines remained scarce. That said, there was a clearly observable shift, especially in major urban areas in the country with historically high public transport mode shares.

2.2. Pre-pandemic mode share in Metro Manila

The urban transport system in the Philippines is primarily road-based, composed of public utility jeepneys, buses, taxis, tricycles, and pedicabs. In Metro Manila, jeepneys and buses dominate road-based public transport wherein buses serve 805 routes, while jeepneys travel on 785 different routes. Taxis, tricycles, and pedicabs provide express services, but the latter two transport modes are limited to serve in local areas (Asian Development Bank, 2012).

Fig. 1 presents the 2014 distribution of trips by mode and the composition of trips made by public transportation in Metro Manila. Based on Fig. 1, most trips were generated by public transport, using public utility jeepneys.

Fig. 1.

Trip composition by mode (left) and trip composition of public transportation (right) Source: JICA & DOTC, 2015.

The percentage share of public transportation in Metro Manila is greater than the modal share of public transportation in Jakarta, which is 27% (Prayudyanto and Thohir, 2017), and in Kuala Lumpur, which is 20% (NKRA-UPT, as cited in Endut et al., 2015). But its trips generated using private vehicles are lower than Jakarta, with 73% of trips made by personal vehicles (Prayudyanto and Thohir, 2017).

2.3. Factors affecting mode choice

The mode choice behavior of people is affected by several factors. One of these factors is sex. Studies in Malaysia (Arasan and Vedagiri, 2009) and India (Nurdden et al., 2007) found that females are less likely to drive personal vehicles and prefer public transportation than males. This result is comparable with the findings of Chang and Wu (2005) that men in Taiwan prefer to drive for themselves and use personal vehicles. Age is another mode choice influencing variable. According to Racca and Ratledge (2003), walk trips are more common in the younger generation, while for ages 65 and above, it is more likely that people take trips as a passenger than a driver. The study by Mayo and Taboada (2020) on the mode choice of commuters in Metro Cebu, Philippines, revealed that varying factors influence different age groups. People ages 36 to 44 are more concerned with cost, those ages 45 to 53 are more concerned with comfort, while the people ages above 60 are keener on their safety. Income is also a significant factor in mode choice as people belonging to the lowest income group are more inclined to use public transport (Racca and Ratledge, 2003; Bajracharya and Shrestha, 2017).

As car ownership increases, the more likely people use a car; car owners choose private transport more than other travel modes (Abdullah et al., 2020; Racca and Ratledge, 2003). Vehicle trip travel time influenced by the travel distance also affects mode choice. In the Philippines, De Guzman and Diaz (2005) suggested that people tend to use private vehicles more as the total travel time increases, while Racca and Ratledge (2003) mentioned that public transport involves shorter travel distances than those taken by personal vehicles.

The trip purpose is another mode choice influencing factor to consider. For work trips, public transport and carpooling are more used, while active transport, such as bicycling and walking, is often related to travel to school (Racca and Ratledge, 2003).

Due to the COVID-19 pandemic and the travel guidelines imposed by the government, the travel behavior of people changed. It was observed that new mode choice influencing factors arise too. Infection-related factors have become a high priority for people. Commuters are concerned about social distance, vehicle cleanliness, and whether the other passengers wear face masks (Abdullah et al., 2020).

The factors discussed in the previous paragraphs were the ones investigated in this study to determine if there was a change in the mode choice influencing factors before, during, and after the pandemic-related lockdowns. Based on the presented literature, there is a high public transportation use in the Philippines, but there are not many studies on how the COVID-19-related lockdowns affected these factors, as well as the mode choice distribution. Therefore, this study seeks to fill in this gap and the results can be used as a basis for further research and data-driven policy shift.

3. Methodology

An online revealed-preference survey was used for data collection due to the limitations brought by the pandemic. The study focused on investigating two types of factors, namely socio-demographic factors (age, sex assigned at birth, household income, household size, and vehicle ownership) and travel characteristics (travel mode, travel distance, travel time, and travel cost). Respondents were asked to consider these factors for the periods before, during, and after the transport lockdown.

The study area was National Capital Region (NCR), commonly known as Metro Manila. The respondents are those who resided and worked in this region for all three periods being studied in the study. Due to the ongoing pandemic at the time of the study, the selected sampling method was snowball sampling. The initial respondents, who were private companies and barangay officials in Metro Manila, helped the researchers identify other potential samples. The survey was disseminated first to them on different online platforms, and they shared it with their co-workers or employees for the next set of respondents to do the same. Each response from the survey was checked to determine if the criteria for the study's respondents were all met before using it as part of the data. After obtaining the number of valid responses, the formula presented in equation (1), Slovin's formula (Ellen, 2020), was used to compute the sampling margin of error.

$$n=\frac{N}{1+{Ne}^2}$$ (1)

where, n is the sample size, N is the population size, and e is the margin of error. The population size considered was the estimated working population in Metro Manila obtained from the latest Philippine Statistics Agency data available at the time of study.

Five different categories of land transportation systems were studied: public transportation (Jeepneys, trains, and buses), private vehicles (personal cars and motorcycles), private hire (Share-a-ride, Grab, and Taxis), active transportation (walking and bicycling), and shuttle services (services offered by the government, companies, and other organizations).

For data analysis, multinomial logistic regression was used to determine the actual effects of each parameter being studied on the mode choice of respondents. This analysis was performed with the aid of the statistical software SPSS. The utility of a transportation mode was represented as a function of the attributes of the weighted coefficients that affect choosing that specific mode, represented mathematically by equation (2) (Khan, 2007).

$$U_{mi}={\theta }_1{X}_{mi1}+{\theta }_2{X}_{mi2}+\cdots +{\theta }_k{X}_{mik}$$ (2)

where,

• U_mi: net utility function for mode m for person i,

• x_mi1, …, x_mik: attributes of mode m for person i, and

• θ₁, …, θ_k: weights attached to each factor

Meanwhile, equation (3) presents the probability of an individual choosing a travel mode using the multinomial logit model (Khan, 2007).

$$P_{in}=\frac{e^{V_{in}}}{\sum _{i=1}^N{e}^{V_{im}}}$$ (3)

where,

• V_in: utility function of mode n for the person i,

• V _im: utility function of any mode m in the choice set for person i,

• P_in: probability of person i selecting mode n,

• N: total number of available travel modes for person i

Since the study included some categorical factors, such as travel mode, sex assigned at birth, and vehicle ownership, encoding was done to convert these categories into numerical codes for the software to analyze (Ray, 2015). Table 1 shows the corresponding numerical codes used for the variables.

Table 1.

Code for qualitative parameters.

Parameter	Category	Code
Travel Mode	Public Transportation	1
	Private Vehicle	2
	Private Hire	3
	Active Transportation	4
	Shuttle Service	5
Sex Assigned at Birth	Male	1
	Female	2
Vehicle Ownership	Do not own a personal vehicle	0
	Do own a personal vehicle	1

The multinomial logistic model should have no multicollinearity, which means there should be no highly correlated independent variables (Shrestha, 2020; Bagozzi and Yi, 1988). In line with this, Pearson correlation analyses were performed. Once the condition of no multicollinearity was satisfied, the likelihood ratio test was done to show the relationship between the dependent variable (mode choice) and the independent variables (International Business Machines, 2020). For each model, the p-values from the likelihood ratio test were used to determine the independent variables that significantly influence people's mode choice (e.g., variables with a p-value less than 0.05). Meanwhile, Akaike's Information Criteria (AIC) and Bayesian Information Criteria (BIC) values were considered in assessing each multinomial logistic model's performance. The lower these values are, the better the model fits the data (Bevans, 2020). They can be computed theoretically as follows:

$$AIC=\left(\frac{-2}{N}*LL\right)+\left(2*\frac{k}{N}\right)$$ (4)

where,

N is the number of examples in the training dataset, LL is the log-likelihood of the model on the training dataset, and k is the number of parameters in the model.

$$BIC=\left(-2*LL\right)+\left[\log \left(N\right)*k\right]$$ (5)

where,

log() has the base-e called the natural logarithm, LL is the log-likelihood of the model, N is the number of examples in the training dataset, and k is the number of parameters in the model.

In addition, models were validated by comparing the predicted with actual mode choice decisions per respondent (Ding and Zhang, 2016). Lastly, the mode choice distributions in three different periods were also determined and compared to each other to investigate the evident changes that occurred using Sankey diagrams. These diagrams presented the mode composition per period and mode choice changes between time periods. It also represents the percentage shift per mode between the time periods considered.

4. Results and discussion

4.1. Demographics

Since the study focused on determining the changes in mode choice behavior by examining the distribution of mode choices in three different periods which are before (before March 2020), during (from March 2020 to August 2021), and after (after August 2021) transport lockdown of working Filipinos in Metro Manila, the respondents who answered the questionnaires were only those who reside and work in this region before the COVID-19 pandemic and until the time of the study. Based on the 2015 Census, Metro Manila has a population of 22 448 173 and an annual population growth rate of 1.58%. According to PSA (2020), the region has a 60.5% labor force participation rate (LFPR). Therefore, assuming that the population growth rate and LFPR in the region remained the same, its 2021 population was approximately 13 615 897, while its 2021 labor force population would be around 8 237 618.

The total number of responses collected in this research was 264. But the sample size was reduced to 204 respondents using the criteria mentioned earlier. Using Slovin's formula mentioned previously, this sample size corresponded to a 7% margin of error. According to Surresh and Chandrashekara (2012), the acceptable range for margin of error in survey types of studies is from 5% to 10%. Hence, the computed value was deemed acceptable.

Table 2 shows the distribution of the respondents in terms of sex assigned at birth, age, and household income. Majority of respondents are female, comprising 55.4. This is similar to the population data obtained from PSA (2018). In terms of age, most of the respondents come are in their 20s. This is mostly due to the data gathering being done online since younger people are more active on social media. Comparing this to the actual age distribution in Metro Manila (PSA, 2015), they are similar with the highest percentage being in the prime working age group (25–54) and the lowest percentage being in the elderly age group (65 and above). For household income, there are seven classifications based on PSA, but this study grouped them further into four only: poor to low income (less than ₱23,000), lower-middle to middle-middle income (₱23,000 to ₱81,999), upper-middle to upper income (₱82,000 to ₱233,999), and rich (more than ₱233,999). Most respondents belong to the lower middle to middle-middle income households. The difference between the survey and population percentages were about 7% only.

Table 2.

Profile of the survey respondents.

Parameter	Category	Percentage of Respondents		Percentage of Population	Source for Population Data
Sex assigned at birth	Female	55.4%		50.5%	(PSA, 2018)
	Male	44.6%		49.5%
Age	15–19	0.5%	17.2%	27.7%	PSA (2015)
	20–24	16.7%
	25–29	30.9%	76.1%	58.2%
	30–34	14.2%
	35–39	10.3%
	40–44	6.9%
	45–49	7.4%
	50–54	6.4%
	55–59	3.4%	5.9%	8.7%
	60–64	2.5%
	65 and above	1.0%		5.4%
Household Income	Less than ₱23 000	33.3%		26.6%	PSA (2018)
	₱23 000 - ₱81 999	44.6%	66.7%	73.4%
	₱82 000 - ₱233 999	14.7%
	More than ₱233 999	7.4%

It can be seen in Fig. 2, Fig. 3, Fig. 4 that most of the respondents have a household size ranging from four to six people. The actual average household size in Metro Manila is 4.2 (PSA, 2016). There was a sizeable decrease in household size during the COVID-19 related transport lockdown. The reduction in household size may be attributed to the loss of a family member, change of household or residency, and other similar reasons. On the other hand, there was not much difference in the household size observed between the second and third periods. Fig. 5, Fig. 6, Fig. 7 .

Fig. 2.

Household size distribution in three periods.

Fig. 3.

Household size distribution in three periods.

Fig. 4.

Household size distribution in three periods.

Fig. 5.

Vehicle ownership distribution in three periods.

Fig. 6.

Vehicle ownership distribution in three periods.

Fig. 7.

Vehicle ownership distribution in three periods.

As for vehicle ownership, 53.4% of the respondents owned at least one personal vehicle before the transport lockdown. During the transport lockdown, the percentage increased to 55.4%. People encountered difficulty in commuting due to reduced public transportation capacity and/or feared contracting the virus when using other modes. After the transport lockdown, the percentage remained at 55.4%.

As shown in Fig. 8, Fig. 9, Fig. 10 , only 12.3% of the respondents worked from home (WFH) before the transport lockdown. During the transport lockdown, this increased considerably to 47.1%. This is a clear indication effect of the transport lockdowns, as well as the shift of businesses and schools to remote work. Meanwhile, since the third period was after the latest ECQ, the percentage of respondents who WFH dropped to 36.8%. That said, this percentage is still relatively high. This is the result of the combination of some businesses allowing for a higher level of on-site work, while some still allowing (or even requiring) WFH arrangements.

Fig. 8.

Work arrangement of the respondents in three periods.

Fig. 9.

Work arrangement of the respondents in three periods.

Fig. 10.

Work arrangement of the respondents in three periods.

4.2. Trip characteristics

The transportation mode most available (i.e., mode people can use or have the access to in traveling to their workplace) before the transport lockdown was public transportation. This transport mode was available to 110 respondents. Urban transport in the Philippines is dominated by jeepneys and buses (Asian Development Bank, 2012). However, when the ECQ was imposed, there was a significant drop in public transportation availability as seen in the results. The government restricted the mobility of people, and transport facilities were not allowed to operate. The availability of both private hire and active transportation also went down during this period. Some private hire drivers chose not to operate anymore due to low profits brought upon by curfews and lockdowns during the pandemic (Suntay, as cited in Balinbin, 2021). Meanwhile, one possible reason behind the reduction in active transportation was people's fear of being exposed to the virus. Moreover, during the transport lockdown, the availability of private vehicles almost doubled. There were also more shuttle services available to the respondents during this period compared to before the transport lockdown. The increase in shuttle service availability may be due to the DOLE-DTI advisory that required companies to provide shuttle services for their workers and the other non-government organizations that responded to the lack of transportation. After the transport lockdown, when there was no more ECQ, the transportation modes' availability slowly went back similar to before the transport lockdown situation. There were more public transportation, private hire, and active transportation available during this time and fewer private hire and shuttle services. Fig. 11 .

Fig. 11.

Availability of transportations mode in three periods.

In terms of the travel distance, as shown in Fig. 12, Fig. 13, Fig. 14 , most respondents had workplaces less than 5 km (short) from their homes in all three periods. Meanwhile, the respondents with workplaces that are 10–14 km (long) and 15 km and above (extra-long) from their residences almost had the same percentages in all three periods. The only reasons for the shift in travel distance during the pandemic were the changes in the workplace and home addresses. Fig. 15, Fig. 16, Fig. 17 .

Fig. 12.

Travel distance distribution in three periods.

Fig. 13.

Travel distance distribution in three periods.

Fig. 14.

Travel distance distribution in three periods.

Fig. 15.

Travel time distribution in three periods.

Fig. 16.

Travel time distribution in three periods.

Fig. 17.

Travel time distribution in three periods.

The distribution of the travel time was almost the same as the travel distance. More than 40% of the respondents spent 30 min or less traveling to work in all three periods based on the figures above. For travel time, there was an option “work from home” to account for those who WFH for a certain period and did not travel at all to work. During the transport lockdown, 10.3% of the respondents did not work on-site at all. After the transport lockdown, some reverted back to on-site work as suggested by the decreased number of respondents who answered “work from home” during this period. Fig. 18, Fig. 19, Fig. 20 .

Fig. 18.

Travel cost distribution in three periods.

Fig. 19.

Travel cost distribution in three periods.

Fig. 20.

Travel cost distribution in three periods.

There was not much change in the travel cost in the three periods. The range with the highest number of respondents in all three periods was ₱200 to ₱799 (USD4-16), while those with the least ranged from ₱0 to ₱199 (USD0-4).

4.3. Mode choice influencing factors

4.3.1. Before transport lockdown

Using Pearson correlation analysis, as shown in Table 3 , no pair of independent variables have a correlation coefficient greater than 0.80. This finding suggests that no variables are highly correlated with each other which means that there is no multicollinearity, and all the variables may be considered in multinomial logistic regression. Table 4 .

Table 3.

Collinearity among independent variables (before transport lockdown).

	Sex assigned at birth	Age	Household size	Household income	Vehicle ownership	Travel distance	Travel time	Travel cost
Sex assigned at birth	–
Age	−0.083	–
Household size	−0.034	0.058	–
Household income	−0.034	−0.100	−0.038	–
Vehicle ownership	−0.144	0.083	0.054	0.346	–
Travel distance	0.089	−0.154	−0.084	0.089	0.072	–
Travel time	0.038	−0.131	−0.042	0.186	−0.036	0.646	–
Travel cost	0.030	−0.096	0.012	0.206	0.209	0.404	0.455	–

Table 4.

Likelihood ratio tests (before transport lockdown).

Likelihood Ratio Tests
Effect	Chi-Square	df	Sig
Intercept	0.000	0
Age	15.952	4	0.003
Household size	5.726	4	0.221
Household income	15.454	4	0.004
Travel distance	7.399	4	0.116
Travel time	8.538	4	0.074
Travel cost	48.407	4	0.000
Sex assigned at birth	5.578	4	0.233
Vehicle ownership	46.759	4	0.000

For this study, public transportation was used as the reference category where the other transport modes were compared. As discussed in the methodology, variables with a significance value less than 0.05 are considered significant in affecting the mode choice of working Filipinos in Metro Manila. Hence, the significant variables before the transport lockdown were age (X _A ), household income (X _HI ), travel cost (X _TC ), and vehicle ownership (X _VO ). Their individual effects are discussed in the sections that follow.

The utility function of the modes for the period before the transport lockdown follows the format of equation (6) while the coefficients per parameter are shown in Table 5 .

$$U=\alpha X_A+\beta X_{HI}+\gamma X_{TC}+\delta X_{VO}+c$$ (6)

Table 5.

Coefficients and constants of the utility functions of the transportation modes (before transport lockdown).

Transportation Mode	α	β	γ	δ	c
PV	0.256	−0.279	−0.007	−3.598	0.036
PH	0.217	1.388	−0.160	1.142	−6.937
AT	0.388	0.270	−3.272	−0.501	2.364
SS	0.536	−1.994	0.221	0.609	−3.399

The results suggest that the preference of people for using private vehicles increases with age (α = 0.256). Older people have higher chances of choosing private vehicles over public transportation. This situation may be the case since older people have a more comfortable travel experience when using personal vehicles; transferring from one mode to another is not an issue. In terms of household income, the negative coefficient (β = −0.279) implies that people with lower total household incomes are more likely to choose private vehicles. This is an unexpected case and in contrast with the findings of Raca and Ratledge (2003) and Bajracharya and Shresth (2017) who said that lower-income people are more likely to use public transportation. The negative coefficient (γ = −0.007) for travel cost indicates that people are more inclined to choose public transportation when travel cost increases. People who do not own private vehicles are also more inclined to use public vehicles since there are no available cars for them to use.

In terms of private hire use, older people prefer traveling using this mode to public transportation (α = 0.217). Older people may have felt more at ease with point-to-point travel using private hire vehicles. But compared to private vehicles, the magnitude is lesser, as suggested by the coefficients. As household income increases, the utility of using private hire vehicles also increases (β = 1.388). As the household income of people increases, the utility of using private hire vehicles increases (β = 1.388) since people belonging to higher income groups have more capacity to pay for this mode. But when travel costs (γ = −0.160) begin to rise, people start to choose public transportation more. Since public transportation is cheaper, commuters prefer this transport mode over private hire. Travelers without personal vehicles are also more likely to use private hire (δ = 1.142). Since these people do not have readily available family vehicles, they are pushed to use other travel modes.

Focusing on active transportation, its utility increases as age (α = 0.388) and household income increase (β = 0.270). With increasing travel costs, the utility of active transportation decreases. As travel cost rises (γ = −3.272), people tend to use public transportation. People with no personal vehicles are also inclined to use public transportation than active transportation.

Meanwhile, as age and travel costs rise, people are more inclined to use shuttle services (α = 0.536 and γ = 0.221, respectively). But as household income increases, the utility of shuttle services goes down (β = −1.994). Commuters with no personal vehicles are also more likely to use shuttle services than public transportation (δ = 0.609). Table 6 .

Table 6.

Percent correct (before transport lockdown).

Classification
Observed	Public transportation (e.g., train, jeep, bus, or van)	Private vehicle (e.g., personal car or motorcycle)	Private hire (e.g., share-a-ride, Grab, or taxi)	Active transportation (e.g., walking or cycling)	Shuttle service (e.g., shuttle services from companies, LGUs, and other offices)	Percent Correct
Public transportation (e.g., train, jeep, bus, or van)	65	14	0	5	2	75.6%
Private vehicle (e.g., personal car or motorcycle)	6	42	0	5	0	79.2%
Private hire (e.g., share-a-ride, Grab, or taxi)	4	1	0	1	0	0.0%
Active transportation (e.g., walking or cycling	1	2	0	25	1	86.2%
Shuttle service (e.g., shuttle services from companies, LGUs, and other offices)	4	1	0	2	1	12.5%
Overall Percentage	44.0%	33.0%	0.0%	20.9%	2.2%	73.1%

To check the accuracy of the model, the AIC and BIC parameters were checked together with the percentage correctness of the model. From Table 7 , the model had AIC and BIC values of 412.770 and 266.093, respectively. These values were lower compared to the AIC and BIC of models from other attempts, thus indicating that this had a better fit than them. The resulting percentage correct of the model is 73.1%, which is relatively high for this kind of studies. Table 8 .

Table 7.

Model fitting information (before transport lockdown).

Model Fitting Criteria				Likelihood Ratio Tests
Model	AIC	BIC	−2 Log Likelihood	Chi-Square	df	Sig
Intercept Only	412.770	425.586	404.770
Final	266.093	330.174	226.093	178.677	16	0.000

Table 8.

Collinearity among independent variables (during transport lockdown).

	Sex assigned at birth	Age	Household size	Household income	Vehicle ownership	Travel distance	Travel time	Travel cost
Sex assigned at birth	–
Age	−0.083	–
Household size	−0.096	−0.027	–
Household income	−0.075	−0.150	0.043	–
Vehicle ownership	−0.171	0.012	0.080	0.303	–
Travel distance	0.043	−0.134	−0.050	0.159	0.099	–
Travel time	0.022	−0.051	−0.043	0.116	0.022	0.634	–
Travel cost	0.066	−0.070	−0.023	0.210	0.272	0.309	0.314	–

4.3.2. During transport lockdown

Similar to the first period, there are no pair of independent variables with Pearson correlation coefficients higher than 0.80 during the 2nd period, which means that the assumption for no multicollinearity in multinomial logistic regression is checked. Table 9 .

Table 9.

Likelihood ratio tests (during transport lockdown).

Likelihood Ratio Tests
Effect	Chi-Square	df	Sig
Intercept	0.000	0
Age	8.556	4	0.073
Household size	1.646	4	0.800
Household income	21.095	4	<.001
Travel distance	10.380	4	0.034
Travel time	18.310	4	0.001
Travel cost	67.136	4	<0.001
Sex assigned at birth	11.742	4	0.019
Vehicle ownership	50.695	4	<0.001

During this period, household income (X _HI ), travel distance (X _TD ), travel time (X _TT ), travel cost (X _TC ), sex assigned at birth (X _S ), and vehicle ownership (X _VO ) are significant in defining the mode choice of working Filipinos in Metro Manila.

The utility function of the modes for the period during the transport lockdown has the form shown in equation (7), while the coefficients obtained are shown in the table that follows. Table 10

$$U=\alpha X_{HI}+\beta X_{TD}+\gamma X_{TT}+\delta X_{TC}+{\epsilon X}_{VO}+{\zeta X}_S+c$$ (7)

Table 10.

Coefficients and constants of the utility functions of the transportation modes (during transport lockdown).

Transportation Mode	α	β	γ	δ	ε	ζ	c
PV	0.995	0.645	−1.222	0.248	−3.251	−0.325	0.590
PH	1.858	0.457	−0.588	0.161	−0.320	−1.981	−3.665
AT	1.166	−0.777	−0.901	−3.300	0.158	0.788	4.958
SS	0.453	0.430	−0.494	−0.474	−0.279	−0.873	0.050

The influence of household income in choosing private vehicles is unlike the first period. Respondents with higher household incomes prefer to use personal vehicles than public transportation (α = 0.995). As travel distance and travel cost increase, people are more inclined to use public transportation (β = 0.645 and δ = 0.248). But when travel time increases, people choose private vehicles over public transportation (γ = −1.222). Females were found to prefer traveling using private vehicles over males (ζ = −0.325). Still, commuters with no personal vehicles are more likely to travel via public transportation (ε = −3.251).

For private hires, the higher the household income, the higher the probability that a person uses this mode (α = 1.858). The utility of private hire also increases with increasing travel distance (β = 0.457) and travel costs (δ = 0.161) but decreases with the increase in travel time (γ = −0.588). During the first period, this is not the case for travel distance since people are more inclined to avoid private hires as travel distance becomes longer. The change may be due to the COVID-19 pandemic. People choose to travel via private hire since they feel like it is less likely that they will contact the virus while using this transportation mode. Males prefer public transportation to females (ζ = −1.981), and it is more likely that persons with no personal vehicles travel via private hire during this period (ε = −0.320).

For active transportation, those part of higher-income groups prefers this mode to public transportation (α = 1.166). But as the travel characteristics, travel time, distance, and costs rise, people are more inclined to choose public transportation over active ones. This situation may be the case since bicycling or walking is tiring at longer times and distances. Unlike private vehicles and private hire uses, there is a higher probability that males travel via active transportation than females (ζ = 0.788). Commuters with personal cars are less probable to resort to bicycling or walking (ε = 0.158).

An increase in household income and travel distance results in rising in the utility of shuttle services (α = 0.453 and β = 0.430, respectively). The predicted influence of household income is different from the first period. Like active transportation, it is less likely that people travel via shuttle services as the travel characteristics (travel distance, time, and cost) rise. Compared to females, males are more inclined to choose public transportation than shuttle service (ζ = −0.883). Unlike the first period, there is a higher probability that people with no available personal vehicles choose public transportation over shuttle services (ε = −0.279). Table 11 .

Table 11.

Percent correct (during transport lockdown).

Classification
Observed	Public transportation (e.g., train, jeep, bus, or van)	Private vehicle (e.g., personal car or motorcycle)	Private hire (e.g., share-a-ride, Grab, or taxi)	Active transportation (e.g., walking or cycling)	Shuttle service (e.g., shuttle services from companies, LGUs, and other offices)	Percent Correct
Public transportation (e.g., train, jeep, bus, or van)	30	7	2	5	0	68.2%
Private vehicle (e.g., personal car or motorcycle)	4	56	4	6	0	80.0%
Private hire (e.g., share-a-ride, Grab, or taxi)	3	7	7	1	0	38.9%
Active transportation (e.g., walking or cycling	1	3	0	30	0	88.2%
Shuttle service (e.g., shuttle services from companies, LGUs, and other offices)	7	3	2	3	1	6.3%
Overall Percentage	24.7%	41.8%	8.2%	24.7%	0.5%	68.1%

The mode choice model during the transport lockdown had an AIC value of 459.022 and BIC value of 331.497 based on Table 12 . These values were lower compared to the obtained AIC and BIC values of models from other trials, thus having a better fit. The percent correct of the model during the transport lockdown shows that it is capable to predict 68.1% of the choices of the trip makers correctlyTable 13 .

Table 12.

Model Fitting information (During the Transport Lockdown).

Model Fitting Criteria				Likelihood Ratio Tests
Model	AIC	BIC	−2 Log Likelihood	Chi-Square	df	Sig
Intercept Only	495.022	507.838	487.022
Final	331.497	421.210	275.497	211.525	24	<0.001

Table 13.

Collinearity among independent variables (after transport lockdown).

	Sex assigned at birth	Age	Household size	Household income	Vehicle ownership	Travel distance	Travel time	Travel cost
Sex assigned at birth	–
Age	−0.083	–
Household size	−0.106	−0.007	–
Household income	−0.038	−0.146	0.005	–
Vehicle ownership	−.171	0.007	0.032	0.271	–
Travel distance	0.024	−0.151	−0.084	0.153	0.101	–
Travel time	0.056	−0.061	−0.057	0.144	0.020	0.653	–
Travel cost	0.044	−.077	−0.052	0.248	0.201	0.321	0.418	–

4.3.3. After the transport lockdown

Just like in the previous periods, there is no multicollinearity in the data set. Table 14 .

Table 14.

Likelihood ratio tests (after transport lockdown).

Likelihood Ratio Tests
Effect	Chi-Square	df	Sig
Intercept	0.000	0
Age	10.555	4	0.032
Household size	7.862	4	0.097
Household income	15.714	4	0.003
Travel distance	6.718	4	0.152
Travel time	8.465	4	0.076
Travel cost	57.190	4	0.000
Sex assigned at birth	3.873	4	0.423
Vehicle ownership	53.564	4	0.000

Age (X_A), household income (X_HI), travel cost (X_TC), and vehicle ownership (X_VO) were the independent variables that significantly affect mode choice after the transport lockdown. The significant factors identified for this period were similar with the period before the transport lockdown. Based on this, the utility function of the modes after the transport lockdown has the following form. Table 15

$$U=\alpha X_A+\beta X_{HI}+\gamma X_{TC}+\delta X_{VO}+c$$ (8)

Table 15.

Coefficients and constants of the utility functions of the transportation modes (after transport lockdown).

Transportation Mode	α	β	γ	δ	c
PV	0.071	0.838	−0.240	−3.174	0.610
PH	−0.119	1.404	−0.116	−0.403	−2.688
AT	0.322	0.878	−3.264	−0.940	2.583
SS	0.263	0.444	−0.074	−0.063	−2.872

In terms of private vehicle use, age influences the utility of a mode choice the same way as in the first period. The utility of private vehicles goes up as age increases (α = 0.071). The effect of household income is like the second period. As household income increases, it is more likely that private vehicle is chosen over public transportation (β = 0.838). Travel cost has a different effect during the third period. After the transport lockdown, an increase in travel costs results in more public transportation use (γ = −0.240). Still, people with no personal vehicles choose to travel via public transportation (δ = −3.174).

Regarding the utility of private hire, age now has a different influence. Before the transport lockdown, as the age of people becomes higher, it is more likely that they will choose private hire. After the transport lockdown, public transportation is more preferred with increasing age (α = −0.119). Household income has the same influence on the utility of private hire for all three periods. People with higher household incomes are more inclined to ride private hire (β = 1.404). The effect of travel cost goes back similar to the first period. With higher travel costs, commuters are more probable to use public transportation than private hire (γ = −0.116). The influence of vehicle ownership remained to be like the second period. People with no access to personal vehicles choose public transportation over private hire (δ = −0.403). Table 16 .

Table 16.

Percent correct (after transport lockdown).

Classification
Observed	Public transportation (e.g., train, jeep, bus, or van)	Private vehicle (e.g., personal car or motorcycle)	Private hire (e.g., share-a-ride, Grab, or taxi)	Active transportation (e.g., walking or cycling)	Shuttle service (e.g., shuttle services from companies, LGUs, and other offices)	Percent Correct
Public transportation (e.g., train, jeep, bus, or van)	32	11	2	4	0	65.3%
Private vehicle (e.g., personal car or motorcycle)	6	62	0	5	0	84.9%
Private hire (e.g., share-a-ride, Grab, or taxi)	7	6	2	1	0	12.5%
Active transportation (e.g., walking or cycling	1	7	0	20	1	69.0%
Shuttle service (e.g., shuttle services from companies, LGUs, and other offices)	8	4	0	2	1	6.7%
Overall Percentage	29.7%	49.5%	2.2%	17.6%	1.1%	64.3%

The influence of all the variables in the model on the utility of active transportation is similar to their effects during the first period. As age and household income increase, the utility of active transportation rises (α = 0.322 and β = 0.878, respectively). With increasing travel costs, it is more likely that people will choose public transportation over active modes (γ = −3.264). Commuters with no personal vehicles choose to ride public transportation than resort to active transportation (δ = −0.940). Meanwhile, the probability of using shuttle services rise as age also increases (α = 0.263). This behavior is also observed in the first period. An increase in household income causes a rise in the utility of shuttle services (β = 0.444). The rise in travel costs means that people are more likely to use public transportation, which is also the case during the transport lockdown (γ = −0.074). People with no personal vehicles remained to use public transportation over shuttle services, like in the second period (δ = −0.063). Table 17 .

Table 17.

Model fitting information (after transport lockdown).

Model Fitting Criteria				Likelihood Ratio Tests
Model	AIC	BIC	−2 Log Likelihood	Chi-Square	df	Sig
Intercept Only	466.150	478.966	458.150
Final	328.624	392.704	288.624	169.526	16	<0.001

The mode choice model after the transport lockdown has AIC and BIC values of 466.150 and 328.624, respectively. These values are lower when compared to the AIC and BIC value of other attempts in modeling. The model is also seen to be capable to predict 64.3% of the choices of the trip makers correctly.

The identified significant mode choice influencing factors in the three periods are summarized in Table 18 . As presented, the first and third periods had the same mode choice influencing factors. This shows that the mode choice behavior has relatively returned to pre-lockdown state once the lockdowns were lifted, at least in terms of the factors significantly affecting this mode choice. Moreover, three mode choice influencing factors: household income, travel cost, and vehicle ownership, were common during the three periods. This shows that these factors are significant regardless of the state of lockdown.

Table 18.

Significant mode choice influencing factors in three periods.

	Before the Transport Lockdown	During the Transport Lockdown	After the Transport Lockdown
Age	✓	✗	✓
Sex assigned at birth	✗	✓	✗
Household income	✓	✓	✓
Household size	✗	✗	✗
Vehicle ownership	✓	✓	✓
Travel distance	✗	✓	✗
Travel time	✗	✓	✗
Travel cost	✓	✓	✓

Table 19 summarizes the coefficients of the significant variables in all three periods. Based on this, the sign of household income changed for private vehicle and shuttle service. In period 1, it was negative which means people with lower household incomes are more likely to choose private vehicles and shuttle services over public transportation. But in periods 2 and 3, the sign became positive indicating that those with lower household incomes prefer using public transportation than these two modes. For all the modes, the magnitude increased from period 1 to period 2 and decreased to period 3. These indicate that the significant factors have the greatest effect on the utility of the modes in period 2.

Table 19.

Coefficients of the common significant factors in three periods.

	Household Income			Travel Cost			Vehicle Ownership
	Period 1	Period 2	Period 3	Period 1	Period 2	Period 3	Period 1	Period 2	Period 3
Private vehicle	−0.279	0.995	0.838	0.007	0.248	−0.240	−3.598	−3.251	−3.174
Private hire	1.388	1.858	1.404	−0.160	0.161	−0.116	1.142	−0.320	−0.403
Active transportation	0.270	1.166	0.878	−3.272	−3.300	−3.264	−0.501	0.158	−0.940
Shuttle service	−1.994	0.453	0.444	0.221	−0.474	−0.074	0.609	−0.279	−0.063

For period 1, the sign of travel cost coefficient for private vehicle and shuttle service was positive and negative for private hire and active transportation. However, during the transport lockdown, the sign in private hire and shuttle service changed. In period 3, the effect of travel cost on all the modes changed to negative. This indicates that public transportation is more preferred over all the other modes as travel cost goes up.

The sign of vehicle ownership coefficient for private vehicle in all three periods remained to be negative which indicates that those who own private vehicles are more inclined to use private vehicles. Meanwhile, for private hire and shuttle service, the sign was positive in period 1, but changed to negative in periods 2 and 3. This means that pre-pandemic, those with private vehicles tend to choose public transportation more than these two modes. But during the pandemic, those with private vehicles prefer public transportation less which may be due to people's fear of getting exposed to different people. Moreover, the magnitude was greatest in period 1 for these modes, showing that this parameter influenced the utility of these modes the most for this period.

4.4. Mode choice distribution

Fig. 21, Fig. 22, Fig. 23 present three different diagrams which show the changes and shifts in the mode choice distribution of the respondents in the two periods being compared. The left side shows the mode share for the previous period, while the values on the right side are for those in the latter period. The thickness of the lines represents the percentage. Each line connecting the two sides represents those that went from a mode in the previous period (left side) to a mode in the latter period (right side). Note that WFH was not prevalent in Metro Manila before the pandemic-related lockdown, hence there were no respondents from this category in the pre-pandemic period. This was seen to be more significant during and after lockdown periods.

Fig. 21.

Change of mode choice distribution before and during the transport lockdown.

Fig. 22.

Change of mode choice distribution during and after the transport lockdown.

Fig. 23.

Change of mode choice distribution before and after the transport lockdown.

Before the transport lockdown, 47.5% of the respondents chose public transportation as their main mode of transport as they go to work, making up the greatest percentage among the other modes of transportation. People who chose private vehicles were 30.9% of the respondents, 14.2% chose active transportation, 3.9% chose shuttle services, and 3.5% chose vehicles for hire before the transport lockdown. It can be seen that there was a decrease in the percentage of respondents who used public transportation during transport lockdown. The value dropped by more than half due to the pandemic-related lockdowns. Public transportation users before the transport lockdown shifted to private vehicles, private hire, shuttle services, and active transportation. There were also public transportation users who started purely WFH instead. This is due to mostly three things: (1) the reduced capacity of public modes due to social distancing requirements, (2) the reduced demand due to WFH arrangements, and (3) the increased desire of trip makers to reduce contact with others in enclosed spaces.

Approximately three-fourths of private vehicle users before transport lockdown continued using the same mode during the transport lockdown. Few vehicle for hire and active transportation users who shifted to private vehicles during the transport lockdown. Meanwhile, all those who use shuttle services continued using this travel mode during the transport lockdown. The active transport mode share actually increased by almost 3% due to lockdowns, mostly from previous public mode users since active modes like walking and cycling can also be seen as private mode that reduces the risk of contact with others in an enclosed space.

Generally, there was minimal modal change after the transport lockdown. Public transportation users slightly increased from 21.6% to 25.1%. Most people who shifted to public transportation in the third period were active transportation and vehicles for hire users during the transport lockdown. The percentage of people who use private vehicles continued to increase. An additional 3.5% shifted to this mode during the third period. Most people who started to work on-site after WFH during the transport lockdown chose to travel via private vehicles. It should be noted that even after the time of this study, subsequent lockdowns were imposed again due to the rise in COVID-19 cases. Hence the comparison should be taken as a snapshot of the time considered, instead of as a lasting condition.

The last comparison was done to check if the distribution has returned to pre-pandemic levels after the initial lockdowns were relaxed. As mentioned earlier, the after-lockdown period should only be taken as a snapshot in time since subsequent lockdowns still ensued after. As shown in Fig. 23, the mode share of each transport mode after the transport lockdown did not yet fully return to pre-pandemic levels. Many public transportation users before the transport lockdown were using other modes of transport after the transport lockdown. Most of which are private vehicle users now. Due to the decrease in public transportation availability during the transport lockdown, people were forced to use other available travel modes. From Fig. 23, the percentage of active transportation users was equal before and after the transport lockdown. But it is evident that the users were not entirely the same; some shifted to active transport, while some shifted from active transportation to other modes. For shuttle service and vehicle for hire, it can be observed that the percentage of those who use these modes increased significantly. The modal share almost doubled. Most of the additional users of shuttle service and vehicles for hire were pre-pandemic public transportation users.

5. Conclusions and recommendations

The main objective of this study was to determine how the COVID-19 pandemic affect the mode choice behavior of working Filipinos in Metro Manila before (before March 2020), during (from March 2020 to August 2021), and after (after August 2021) the transport lockdown. The results of the study revealed the distribution of the mode choice of working Filipinos in Metro Manila and the corresponding factors that affect people's mode choice behaviors.

Different factors significantly affect the mode choice of people in the three periods studied. Before the transport lockdown, age, household income, travel cost, and vehicle ownership were significant. During the transport lockdown, the number of significant factors increased and turned to household income, travel distance, travel time, travel cost, sex assigned at birth, and vehicle ownership. After the transport lockdown, the significant factors reverted back to those before the transport lockdown. Based on these findings, there were three parameters that were significant in all the periods which are household income, travel cost, and vehicle ownership; however, the magnitude and sign of the parameters changed in each period. For active transportation, the coefficient of the parameter vehicle ownership varied both in magnitude and sign in every period. It was negative in period 1 then positive in period 2 and became negative again in period 3. The magnitude decreased from period 1 to period 2 and increased in period 3.

Results revealed the changes and shifts in the mode choice distribution of the respondents before, during, and after the transport lockdown. Before the transport lockdown, 47.5% of the respondents chose public transportation as their main mode of transport. But its modal share fell to 21.6% during the transport lockdown, then rose again to 25.1% after the transport lockdown. The shift from public transportation resulted in more people using private vehicles, private hire, active transportation, and shuttle services. There was a continuous increase in private vehicle use. The percentage share of private vehicles before the transport lockdown was 30.9%. It increased to 34.3% during the transport lockdown and continued to increase after the transport lockdown to 37.8%. Meanwhile, active transportation users increased from 14.2% before the transport lockdown to 16.6% during the transport lockdown, then reverted to 14.2% after the transport lockdown. Based on the results, the mode distribution in Metro Manila has not yet fully returned to before the transport lockdown situation.

With the gathered data, the study was able to provide additional information regarding this topic since there were no existing studies of this yet in Metro Manila, and transport policies made during the pandemic still lacked basis. As discussed previously, there was a decrease in public transportation users during the transport lockdown, and people started using other alternatives not preferred before the pandemic, like active transportation. This finding suggests a need to make roads bicycle-friendly and provide more walkable space for pedestrians. There was also an increase in private vehicle users during this period, indicating that planning to push people to switch to other travel modes aside from private vehicles and prevent roads from being congested is necessary. The plans may consider the variables that significantly affect the mode choice of working Filipinos in Metro Manila after the transport lockdown, namely age, household income, travel cost, and vehicle ownership.

The main limitation of the study was the limited number of samples because of the difficulty in obtaining responses during the pandemic. With that, conducting a pen-and-paper survey is recommended to encourage more participants and allow people with no internet access to participate in the study. Future researchers may explore other correlations using the existing data, such as the effects of the pandemic on different social classes, and determine which parameters have direct causality to the changes in mode choice. They are also encouraged to investigate when and how the travel behavior of the working population in Metro Manila will go back the same way as before the pandemic.

Declaration of competing interest

The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Katherine Dimaculangan reports financial support was provided by WebFocus Solutions.