Transit station mobility recovery in Latin American cities during the COVID-19 pandemic

Design and setting

An ecological longitudinal study was conducted with subnational urban units in Latin America from March 2020 to September 2022. Daily observations were aggregated to monthly means to reduce day-of-week noise, to stabilize municipalities with intermittent reporting and to align the mobility series with the monthly policy index. The study period begins with the first month for which mobility is available for every country and ends with the last complete month before the mobility series was discontinued in October 2022. The design follows the Strengthening the Reporting of Observational Studies in Epidemiology statement for observational studies. Two complementary analytic levels were used. The municipality-level sample described the recovery of cities, and the country-month panel estimated the association with policy stringency, which is a national attribute that does not vary within countries. The ecological design was chosen because both the mobility outcome and the policy exposure are available only in aggregated form, and because it allows the recovery of hundreds of cities to be described with a single comparable measure.

Data sources

Mobility data were obtained from the Google COVID-19 Community Mobility Reports, which provide daily percentage changes in the number of visits to places classified as transit stations, including subway, bus and train stations.(8) The reference value is the median of the corresponding day of the week during the period from January 3 to February 6, 2020, so a value of zero corresponds to the pre-pandemic baseline and a value of minus 40 indicates a 40 percent reduction. The data are produced from aggregated, anonymized location history of users who opted into this service, and they are reported at the national level and, where available, at the first and second administrative levels. Country-level files for 2020, 2021 and 2022 were downloaded programmatically from the public repository at google.com/covid19/mobility; the series cover February 2020 to October 2022. Policy data were obtained from the Oxford COVID-19 Government Response Tracker, which records a StringencyIndex for every country and day on a scale from 0 to 100, with higher values indicating stricter containment and closure policies.(10) The index is a composite of nine policy indicators, including school and workplace closures, stay-at-home requirements, restrictions on gatherings and public transport closures, normalized to a common scale. The national file was downloaded from the public GitHub repository of the tracker on the same date. Both sources are public, aggregated, free of charge and do not require authentication.

The three annual mobility files of each country were concatenated and duplicate records were removed. Daily records were then collapsed into monthly means for each country and for each municipality. The daily stringency values were collapsed into monthly means per country in the same way, so that every mobility observation was aligned with the policy index of the same country and calendar month. No imputation was applied. A monthly value was computed only from the valid daily observations available in that month, and months with no valid observations were treated as missing. The mobility series ends on October 15, 2022, when Google discontinued the reports, which is why the analysis window closes in September 2022.

Study units

The municipality-level sample comprised the municipalities that met the eligibility criteria, distributed across the eight Latin American countries for which Google reported municipality-level transit mobility: Argentina, Bolivia, Brazil, Chile, Colombia, Ecuador, Guatemala and Peru. A municipality was eligible when it had at least 400 valid daily transit observations in total and at least 24 months with at least ten valid daily observations between February 2020 and September 2022. These thresholds excluded places whose series were too sparse to support stable monthly means while retaining municipalities with established transit activity. The country-month panel comprised all 14 Latin American countries with a national transit series and a matching StringencyIndex. Mexico, Uruguay, Paraguay and the Dominican Republic report subnational mobility only at the first administrative level, and Costa Rica and Panama report only national mobility; these six countries were included in the country panel but not in the municipality-level sample, because a state or a country is not equivalent to a city. Municipality-level transit mobility was reported for 3,469 municipalities in the eight countries; 2,545 were excluded for insufficient coverage, leaving 924 municipalities in the city sample. The six countries without municipality-level data differ from the eight with it in the granularity of their reporting rather than in the availability of a national series, and they were retained in the country panel to avoid restricting the analysis to countries with finer data. The national series are complete for all 14 countries in every month of the study window.

Variables

The outcome was the monthly mean of the daily transit station mobility, expressed as percentage change from the pre-pandemic baseline. Recovery was quantified with two measures. The trough was defined as the lowest monthly mean reached during 2020, and the recovery level was defined as the mean mobility of July to September 2022, the last complete quarter of the series. A city was classified as recovered when its third-quarter 2022 mean was at or above zero and as near-recovered when it was within ten percentage points of the baseline. The transit station category includes subway, bus and train stations, so it captures the segment of urban mobility most dependent on collective transport, which is the segment most directly affected by capacity limits and service suspensions. The exposure was the monthly mean of the national StringencyIndex, computed by averaging the daily index of each country within each calendar month. The country-level mean StringencyIndex over the study period ranged from 42.2 to 60.8, providing variation across countries in the average strictness of the responses.

Statistical analysis

Distributions were summarized with medians and interquartile ranges because several of them were skewed. The association between policy stringency and transit mobility was estimated with ordinary least squares on the country-month panel of 434 observations. Three specifications were fitted. The crude model regressed monthly transit mobility on the StringencyIndex alone. The second model added country fixed effects, which absorb time-invariant differences between countries, and calendar-month fixed effects, which absorb seasonal patterns. The third model replaced the calendar-month effects with a linear time trend to remove the common secular trajectory of the pandemic while retaining within-country variation in stringency; this specification was prespecified as a sensitivity analysis because the crude association may otherwise confound the effect of restrictions with the shared course of the pandemic. Standard errors were clustered by country in all models to account for serial dependence within countries, and coefficients were expressed per ten-point increase in the index, with 95% confidence intervals. Ten points were chosen as the scaling unit because it represents approximately one tenth of the range of the index and corresponds to a substantively meaningful change in the intensity of restrictions. A cross-country Spearman correlation assessed the association between the country-level mean StringencyIndex over the study period and the third-quarter 2022 mobility, using one observation per country. This correlation is descriptive and was prespecified to complement the panel estimates, which identify the association from within-country variation over time. Statistical significance was set at 0.05, two-sided. Analyses were run in Python with pandas, statsmodels and SciPy, and the entire pipeline was executed twice to confirm that the derived estimates were identical.

Ethical considerations

The study used public, aggregated and anonymized secondary data that contain no individual identifiers. No ethical approval was required.

Study units

The country-month panel included 14 countries and 434 observations over 31 months, with complete national mobility and stringency series for every country and month. Of the 3,469 municipalities with reported transit mobility in the eight reporting countries, 2,545 were excluded for insufficient coverage and 924 cities were analyzed. Brazil contributed 519 municipalities, Argentina 180 and Peru 64; Chile, Colombia, Ecuador, Bolivia and Guatemala contributed 46, 48, 28, 23 and 16 municipalities, respectively (Table 1).

Because the transit category is reported only where Google detects transit station activity, the retained municipalities are those with functioning transit systems rather than a random sample of all municipalities.

Descriptive data

Among the 14 countries, transit station mobility reached a median trough of −69.8% below the baseline (IQR −79.1 to −62.6), with the regional minimum recorded in Apr 2020. The deepest troughs were observed in Peru (−84.2%) and Bolivia (−83.7%), and the shallowest in Brazil (−53.6%) and Uruguay (−55.5%) (Table 1). The trough occurred in April 2020 in twelve of the 14 countries, in May 2020 in Mexico and in June 2020 in Chile (Table 1). The country-level mean StringencyIndex over the study period ranged from 42.2 in Uruguay to 60.8 in Peru, and within every country the monthly index rose during the strict phases of 2020 and early 2021 and declined thereafter (Figure 1).

The regional mean fell from −32.3% below the baseline in March 2020 to −69.9% in Apr 2020, recovered to −24.7% below the baseline by December 2020, and dipped again to −36.6% in January 2021 during the second wave. The series crossed the baseline in November 2021, fell to −6.1% in January 2022, and then remained above the baseline from February 2022 onward, reaching 13.0% in August 2022 (Figure 1). By the third quarter of 2022 the median country stood 9.1% above the baseline (IQR 3.4 to 13.5).

The recovery level ranged from −11.9% below the baseline in Costa Rica to 46.1% above it in Colombia, with Ecuador and Bolivia also exceeding the baseline by more than twenty percentage points (Table 1).

Among the 924 municipalities, 52.5% were at or above the baseline and 64.8% were within ten percentage points of it in the third quarter of 2022. The median municipality reached a third-quarter 2022 level of 2.1% (IQR −21.4 to 25.2) after a median 2020 trough of −65.8% below the baseline.

The interquartile range of third-quarter 2022 mobility across municipalities therefore spanned more than forty percentage points, from −21.4% to 25.2%. Cities with the deepest 2020 troughs tended to recover less completely, although the dispersion around this pattern was wide (Figure 2).

Main results

In the crude analysis, each ten-point increase in the StringencyIndex was associated with a change of −8.6 percentage points in monthly transit mobility (95% CI −9.9 to −7.4; p <0.001). After adjustment for country and calendar month, the estimate was −9.0 percentage points per ten points of stringency (95% CI −10.2 to −7.8; p <0.001) (Table 2).

In every specification the confidence interval excluded zero, and the point estimates were negative. The crude model explained 0.60 of the variance in monthly transit mobility, and the adjusted model explained 0.77.

Other analyses

With respect to the common secular trajectory of the pandemic, the association persisted but attenuated. When the calendar-month effects were replaced by a linear time trend, each ten-point increase in stringency was associated with a change of −3.2 percentage points (95% CI −4.1 to −2.2; p <0.001) (Table 2), an estimate smaller in absolute value than the month-adjusted estimate of −9.0 percentage points. The cross-country Spearman correlation between the mean StringencyIndex over the study period and third-quarter 2022 mobility was 0.09 (p = 0.76) (Figure 3; Table 3).

Main results

In this ecological panel of 14 Latin American countries and 924 cities, mobility to transit stations collapsed by roughly −70 percent below the pre-pandemic baseline in the first wave of 2020 and, on average, returned to or slightly above the baseline by the third quarter of 2022. The contemporaneous association between policy stringency and transit mobility was strong and negative in the crude and month-adjusted models, and it persisted, attenuated, once the common secular trend was removed. Across countries, the average strictness of government responses was not associated with the level of recovery reached by late 2022. Taken together, the two levels of analysis indicate that restrictions depressed transit mobility while they were in force but did not determine where mobility settled once they were removed.

Interpretation and comparison with other studies

The depth of the initial decline is consistent with earlier reports. Studies of traffic and public transport in eight large Latin American cities found reductions of a similar magnitude during the first months of the pandemic,(3) and mobile phone records documented large drops in population mobility across 314 Latin American cities.(4) The April 2020 trough also matches the timing of the most restrictive national measures in the region, when several governments suspended public transport or limited it to essential workers.(20,21) The recovery above the baseline by the third quarter of 2022 is more favorable than the incomplete recovery described in European cities, where rail and bus demand remained below pre-pandemic levels two years after the onset.(17,18,22) Part of this difference may reflect the short reference window used by Google, which compares late-pandemic mobility with January and February of 2020 rather than with an annual average, and part may reflect genuine ridership growth in Latin American systems that regained users faster than operators in high-income countries.(1,11) The contrast may also reflect the lower feasibility of remote work in much of the region and the scarcity of modal alternatives available to low-income workers, who returned to buses and metro lines sooner because they had few substitutes.(6,23) The city-level distribution was likewise uneven, with a substantial minority of municipalities still below the baseline by late 2022, concentrated among the systems that had declined most, a dispersion that mirrors the spatially uneven recovery documented elsewhere.(24,25) Recovery trajectories also differed by mode and by user group, with bus ridership recovering at different speeds across cities,(25,26) which cautions against treating a single mobility index as the whole story.

The negative association between stringency and contemporaneous transit mobility agrees with a large body of ecological work. Policy stringency has been repeatedly associated with reduced population movement, both globally(12,13) and in specific settings of the Global South,(15,27) and mode-specific analyses have shown that public transit is among the activities most sensitive to these restrictions.(2) The attenuation of the estimate after modeling the secular trend is expected. Part of the crude association reflects the shared pandemic trajectory in which restrictions were tightened while mobility fell and were relaxed while mobility recovered, so the trend-adjusted estimate isolates the component of the association that is specific to stringency rather than to the passage of time.(28) The plausibility of a direct effect rests on the composition of the index, which combines school and workplace closures, stay-at-home requirements and public transport closures, measures that restrict the legal ability to travel and signal risk to users.(12,14) The null cross-country correlation between average stringency and the recovery level suggests that the strictness of restrictions did not determine how far transit mobility had returned by late 2022. Recovery appeared instead to be shaped by the resumption of economic activity, by vaccination coverage and by the capacity of users to return to collective transport, factors that varied more within than between policy regimes and that rebounded once the acute restrictions were lifted.(19,23,29) The distinction between suppressing mobility and delaying recovery deserves emphasis. A policy that depresses transit use during a wave can be effective for infection control and yet leave no lasting trace on the level of transit use if demand returns once the measure is removed. The present results are compatible with that pattern: the within-period association was strong, whereas the between-country association with recovery was null. This does not imply that restrictions were without cost, because the revenue losses and service disruptions of the acute phase were real and may have damaged operators even when riders eventually returned.(16,27) This interpretation is consistent with evidence from Colombia and Santiago de Chile, where mobility and transport demand responded to local conditions beyond the stringency of national decrees,(6,30,31) and with findings from London that the decline and recovery of activity were spatially and socially uneven.(24) The heterogeneity across countries is as informative as the regional mean. Colombia and Ecuador ended the period more than twenty percentage points above the baseline, whereas Costa Rica and Paraguay remained below it, a spread that mirrors differences in the timing and intensity of epidemic waves, in vaccination and in the resumption of employment rather than differences in the average strictness of the response. In Bogotá, for example, the disruption of the bus rapid transit network was followed by a marked return of demand once economic activity resumed.(6,31) These local trajectories support the interpretation that transit recovery in the region was driven primarily by the resumption of the activities that generate trips, and only secondarily by the removal of restrictions.

The mechanisms that link stringency to transit mobility are several. Restrictions reduce mobility through legal prohibitions, through the suspension or reduction of service supply, and through the perception of risk that keeps potential riders at home even when travel is permitted.(12) The attenuation after the trend adjustment is consistent with a large component of the crude association operating through the shared timing of the pandemic rather than through any single policy instrument, and it is consistent with the observation that similar restrictions produced different mobility responses depending on the structure of the labor market and on the capacity of the state to enforce them.(16,21) In the United States, the decline in bus ridership was mediated in part by employment and by the shift to remote work,(32) a pattern consistent with the view that the return of trips, rather than the removal of restrictions, dominated the recovery. The regional mean ending above zero also reflects the short January-February 2020 reference window, which may understate a typical pre-pandemic level, a caveat already noted for indicators derived from these reports.(11) The experience of the pandemic also has implications for how transit systems should be monitored during shocks. The Google reports, discontinued in October 2022, provided a rare common yardstick for hundreds of cities, and their end left a gap that operator data alone do not fill.(11) This analysis shows that a discontinued but archived public series can still support retrospective evaluation, and that combining it with a policy index allows the response to restrictions to be separated from the recovery.

The main strengths of the study are its coverage, its reliance on public data and its prespecified sensitivity analysis. The sample spans 14 countries and 924 municipalities, which is broader than the single-city or single-country samples used in most recovery studies, and it links two independent, publicly archived sources rather than the records of a single operator. The three model specifications make explicit how much of the stringency association is attributable to the common course of the pandemic rather than to restrictions themselves, and the country-level correlation offers a direct test of the recovery question that the monthly panel cannot answer on its own. The use of the same indicator across all cities also avoids the comparability problems that arise when combining operator records with different definitions of ridership. The study thereby complements single-system analyses, which are more precise about local mechanisms but cannot characterize the regional pattern of recovery, and it provides a transparent, fully reproducible benchmark against which future post-pandemic monitoring efforts can then be compared.

Limitations

First, the study is ecological, and the exposure varies only at the national level, so the panel estimates describe country-level covariation and cannot be read as individual-level causal effects. Subnational policy data would be required to separate the effect of restrictions from other national conditions. Second, Google mobility is a proxy derived from a changing sample of users with location history enabled, and its coverage is thinner in lower-income municipalities, which may bias the city sample toward places with better digital penetration.(11) Validation against operator ridership records would strengthen the measurement of transit recovery.(32) Third, the cross-country correlation uses only 14 observations and has limited power to detect modest associations, so the null result should be interpreted as the absence of a large effect rather than as proof of no effect. Fourth, residual confounding by vaccination, economic activity, fare adjustments and mode substitution was not controlled, because consistent subnational measures of these variables were not available for the whole region.(33,34) Fifth, the eligibility filter required substantial longitudinal coverage and therefore excluded municipalities with sparse reporting, including the capitals of countries that report no municipality-level data at all. Sixth, the series end in October 2022, so the analysis cannot describe later phases of the recovery or the post-pandemic stabilization of demand.(7) Seventh, national stringency conceals subnational differences in policy, which were substantial in federal countries such as Brazil and Argentina, so the exposure may misclassify the restrictions actually experienced by specific cities; subnational policy coding would reduce this misclassification. Eighth, aggregating daily values into monthly means smooths within-month variation, so the estimates reflect the average association over each month and may understate the sharpness of the response in the days immediately after a policy change.

Generalization

These results generalize to Latin American cities with established transit systems and adequate Google mobility coverage, and they are most informative for the eight countries that report municipality-level data. The conclusions are therefore best read as a description of urban systems with formal collective transport, not of the entire region or of rural and informal mobility. The finding that average policy stringency did not predict the level of recovery reached by late 2022 cautions against interpreting short-term mobility suppression as long-term transit damage. Transport authorities should therefore monitor recovery with operational data after the acute phase of a shock, because aggregated mobility proxies and national policy indices do not by themselves reveal which cities and which user groups have returned to transit.(1,29) For operators and governments the implication is twofold. Financial support is most needed during the acute phase, when fare revenue collapses together with mobility, whereas the recovery phase calls for measures that bring riders back, such as service frequency, vehicle hygiene and fare integration, rather than for the lifting of restrictions alone.(7,34) The wide dispersion of recovery across cities also indicates that monitoring should be disaggregated to the municipality level, so that systems that lag behind can be identified and supported with specific measures. Future studies could extend this approach to other regions with archived mobility reports or combine it with operator records to validate the indicator at the city level.(32)