**4. Results**

In the first part of Section 4.1 below, we explore whether traffic congestion intensities are different across space and time in the study area based on data from the INRIX 2017 Global Traffic Scorecard [29]. In the second part of Section 4.1, we examine how traffic congestion levels are different over space and time based on the 1022 trips made by the 250 selected participants of the survey. In Section 4.2, we compare two assessments of individual exposures to traffic congestion for the 250 participants: one that considers only commuting trips and the other one that considers both commuting and non-commuting trips. In Section 4.3, we explore how three individuals from the same household are exposed to traffic congestion in unique ways over space and time through visualizations of their space-time trajectories.

### *4.1. Spatiotemporal Variations in Traffic Congestion Intensities*

In this section, we answer the first research question: Do spatiotemporal variations exist in traffic congestion intensities? We empirically examine whether traffic congestion intensities are different across space and time in the study area based on data from the INRIX 2017 Global Traffic Scorecard [29] and the trips made by the 250 selected participants. Using the first data source, traffic congestion intensity is assessed in terms of the percentage of congestion travel time that drivers experience out of gross travel time; using the second data source, traffic congestion intensity is derived as the ratio of travel time that considers traffic congestion to free-flow travel time. This is an important question because spatiotemporal variabilities in traffic congestion intensities may exacerbate the UGCoP. Although it is widely known that traffic congestion intensities are different over space and time [55], here we empirically investigate its precise spatiotemporal configurations in the study area.

First, we investigate general spatiotemporal variations of traffic congestion intensities in the study area using data from the INRIX 2017 Global Traffic Scorecard, which provides data on traffic congestion for over 1360 cities around the world (Table 2) [29]. Although the Los Angeles Metropolitan Area is globally notorious for its severe traffic congestion, and as these data indicate, we can observe the spatiotemporal heterogeneity of the traffic congestion in the area: (1) There are temporal variations in traffic congestion. For instance, inter-city drivers experience traffic congestion for 22% of their gross travel time during peak hours while 10% of their gross time occurs during non-peak hours (e.g., around noon) [29]. (2) There are spatial variations in traffic congestion. For example, during non-peak hours, intra-city drivers experience traffic congestion for 13% of their gross travel time, while inter-city drivers experience traffic congestion only for 10% of the gross travel time [29].

Second, we examine the spatiotemporal variations in traffic congestion intensities based on the 1022 trips made by the 250 selected participants (note that, here, traffic congestion intensity for a trip is derived as the ratio of travel time that considers traffic congestion to free-flow travel time; see Figure 2 and earlier description on how these two travel times for each trip are derived). Figure 4 illustrates traffic congestion intensity (the vertical axis) variations of these 1022 trips by trip departure time (the horizontal axis). The ratio of congestion and free-flow travel times is widely used in practice to represent the severity of traffic congestion at the road-segment level (e.g., [29,46,47]). For example, if traffic congestion intensity is 1.5, it means that the travel time that considers traffic congestion is

1.5 times longer than the free-flow travel time due to traffic congestion. Therefore, the minimum value of the traffic congestion intensity is 1.0 because free-flow travels give the minimum travel time.

**Figure 4.** Observed traffic congestion intensity over space and time.

We can observe the following two things in Figure 4. First, there are temporal variations in traffic congestion intensity. Not surprisingly, traffic congestion is generally severe during peak hours (e.g., 7–9 A.M., 3–6 P.M.). Moreover, although traffic congestion during non-peak hours is less severe than that of peak-hours, traffic congestion is still observed at any time during a day. This corroborates our earlier observation that there are temporal variations in traffic congestion in the Los Angeles Metropolitan Area. Second, there are spatial variations in traffic congestion intensity. Vertical distributions of the observations (i.e., indicated by the range of the boxes) represent various traffic congestion intensities at different locations at each hour. For example, the range of the box at 6 P.M. is wider than that at 12 P.M., which means more spatial variations in traffic congestion intensities exist at 6 P.M.

These two findings answer our first research question: Do spatiotemporal variations exist in traffic congestion intensities? Based on the observations made from Table 2 and Figure 4, it is clear that spatiotemporal variations of traffic congestion intensities exist in the study area. In what follows, we continue our analysis to compare individuals' exposures to traffic congestion obtained from two assessments based on the trips made by the 250 selected participants: one that only considers commuting trips and one that considers both commuting and non-commuting trips (i.e., taking into account individuals' activity-travel patterns).

**Table 2.** Spatiotemporal variations in traffic congestion in the study area (Source: [29]).


(a) Percentage of congestion travel time that drivers experience out of gross travel time. (b) Intra-trips refer trips that occur within the city. (c) Inter-trips refer trips that occur into/out of the city.

### *4.2. Differences in Individual Exposures to Traffic Congestion between the Two Assessments*

We conduct a paired sample *t*-test to see whether individual exposures to traffic congestion for each participant are significantly different between the commute-only assessment and the assessment that also considers participants' activity-travel patterns (i.e., considering both commuting and non-commuting trips) (please see Section 3.2.2 and Figure 2 for a detailed explanation of the method). Here, individual exposure to traffic congestion for each trip is estimated by subtracting its free-flow travel time from its estimated travel time (which considers traffic congestion and is estimated using the Google Maps API). This is the additional travel time for a trip due to traffic delay or congestion.

For each participant, we obtain a commute-only exposure measure by adding the additional travel times incurred by the commuting trips and another exposure measure by adding the additional travel times of both commuting and non-commuting trips. We then compare the difference between these two exposure measures for each participant (and thus a paired sample *t*-test is used).

Table 3 indicates that the mean difference in participants' exposure to traffic congestion is 6.66 min, which means that the duration for which a participant experiences traffic congestion increases on average by 6.66 min (47.78%) when participants' activity-travel patterns are considered, compared to the commute-only assessment. The result of the paired sample *t*-test confirms that the differences in exposures to traffic congestion between the two assessments are statistically significant (*p* < 0.001). Figure 5 visualizes the results presented in Table 3. The box plots also show that the average exposure to traffic congestion when individuals' activity-travel patterns are considered is higher than the average exposure obtained in the commute-only assessment.

**Table 3.** Mean exposure to traffic congestion in the commute-only assessment and the assessment that also considers participants' activity-travel patterns.

**Figure 5.** Box plots of individuals' traffic congestion exposure for (**a**) the commute-only assessment and (**b**) the assessment that also considers participants' activity-travel patterns. (Outliers are not presented.)

Moreover, we examine the probability distribution function of individual exposure to traffic congestion. Figure 6 presents the histograms of individuals' traffic congestion exposure levels for the two assessments. As the histograms show, when individuals' activity-travel patterns are considered (right histogram), the shape of the probability distribution function becomes less skewed and converges to its mean. Table 4 shows that skewness (from 2.313 to 1.724) and kurtosis (from 9.766 to 5.982) of the histogram decrease after activity-travel patterns are considered. The results indicate that the probability distribution function of individual exposures to traffic congestion shows a tendency to converge to its average when individuals' activity-travel patterns are considered.

This phenomenon can also be understood as a manifestation of the neighborhood effect averaging problem (NEAP) observed by Kwan [36]. However, there are two important differences between our observations here and the original interpretation of the NEAP put forward by Kwan [36]. First, both exposure assessments (i.e., one that considers only commuting trips and one that considers people's entire activity-travel patterns) in this research are mobility-based. In other words, the commute-only exposure assessment is not residence-based because it already included some portion of individuals' daily mobility (i.e., commuting trips). However, the original articulation of the NEAP compares residence-based exposures with mobility-based exposures. This indicates that the NEAP can also be encountered in environmental exposure assessments when only parts (instead of all) of people's daily mobility are ignored.

Second, the probability distribution functions of the two exposure assessments in this study are not bell-shaped but heavily skewed. The original notion concerning the NEAP only focuses on distributions of individual exposures that approximate a bell-shaped distribution (one such distributions is the normal distribution), but distribution functions in our research are heavily skewed. This indicates that the neighborhood effect averaging problem can also be encountered when the probability distributions of individual exposures are not bell-shaped. These two differences between our observations here and the original interpretation of the NEAP extends the original interpretation of the NEAP in important ways.

**Figure 6.** The histograms of traffic congestion exposure in (**a**) the commute-only assessment and (**b**) the assessment that considers participants' activity-travel patterns.

**Table 4.** Skewness and kurtosis of the histograms for both assessments.


Additionally, we examine in detail how considering individuals' activity-travel patterns impact the relative levels of individual exposure to traffic congestion. Figure 7 illustrates the standardized (*z*-score) individual exposures to traffic congestion. The horizontal axis displays individual exposures of the commute-only cases, while the vertical axis represents individual exposures obtained by the assessment that considers participants' activity-travel patterns. For example, points in the first quadrant (top-right) represent cases when individual exposures to traffic congestion are higher than the average in both assessments. On the contrary, points in the third quadrant (bottom-left) indicates that individual exposures to traffic congestion are lower than its average in both assessments. A closer examination of the graph yields a couple of important findings.

First, the standardized (*z*-score) individual exposures to traffic congestion of most participants (203 participants, 81% of the selected subsample) range between −1 and 0 (see the focused area in the inset). Second, a majority of participants (48 out of 69) in the first quadrant are located in the blue triangular area. This indicates that individual exposures shift much closer to its mean value when activity-travel patterns are considered. These findings confirm an earlier observation that many individuals have exposure levels around the average value while fewer individuals have very high or low exposure levels, and considering individuals' activity-travel patterns leads the exposure level to converge to its mean [36].

Based on these results, we also answer the second question: Will exposures to traffic congestion based on commute-only assessments be significantly lower than those obtained from assessments that also consider individuals' activity-travel patterns in addition to commuting trips? We found that this is indeed the case. There are statistically significant differences between exposures evaluated with and without considering individuals' activity-travel patterns, indicating that the UGCoP is a serious issue. We also found that ignoring individuals' activity-travel patterns may exacerbate the NEAP. Therefore, we can conclude that overlooking people's activity-travel patterns may lead to serious methodological issues in the form of the UGCoP and the NEAP when assessing their exposures to traffic congestion.

**Figure 7.** Standardized (*z*-score) individual exposures to traffic congestion.

### *4.3. Space-Time Visualizations of Individual Exposures to Traffic Congestion*

Figure 8 illustrates the cumulative traffic congestion exposures of 3 individuals from the same residence (i.e., the same household) over the 24 hours of the survey day. Line A (blue) indicates the mother's exposure to traffic congestion, while Lines B (yellow) and C (red) represent the older son and

### *Int. J. Environ. Res. Public Health* **2019**, *16*, 89

the younger daughter respectively. Also, space-time visualizations of these individuals' activity-travel patterns are presented in Figure 9. The vertical axis (*t*) represents time, and the horizontal plane displays space (*<sup>x</sup>*, *y*). Each dot represents a 1-min interval in the trip trajectories obtained from the Google Maps API. The size of the dots indicates traffic congestion intensity. For instance, larger dots represent more intense traffic congestion. The vertical solid lines indicate durations when individuals are performing activities at fixed locations, as their location (*<sup>x</sup>*, *y*) does not change over time.

**Figure 8.** Cumulative exposures to traffic congestion of three selected individuals from the same household (A: Mother, B: Older Son, C: Younger Daughter).

**Figure 9.** Space-time visualizations of traffic congestion exposures of three individuals from the same household (A: Mother, B: Older Son, C: Younger Daughter).

The space-time illustration of individuals' trajectories clearly shows that individuals are exposed to traffic congestion in unique ways over space and time. First, although all 3 individuals have a similar travel-demand environment (e.g., actively employed and driving their own cars), each family member's traffic congestion exposure varies because of their idiosyncratic activity-travel patterns.

As Figure 9 shows, for example, the traffic congestion exposure of Person A (mother) ranges between 10 and 70, while that of Person C (younger daughter) varies between 10 and 20. This difference can be explained by the different activity-travel patterns between these two persons. Person A takes longer commuting trips, and she is heavily exposed to traffic congestion especially during her way back home. Person A is exposed to severe traffic congestion when she takes non-commuting trips, but the non-commuting trips do not significantly contribute to the total exposure because the length of the trips is relatively short. By contrast, Person C is less exposed to traffic congestion than Person A. Most trips that Person C takes are near her residence and relatively short, which enables her to avoid heavy exposure to traffic congestion.

Specifically, we can observe that the younger son (Person B) is exposed to heavier traffic congestion than the others in the household. His traffic congestion exposure becomes more severe when his activity-travel patterns are considered. This drastic increase is mainly because of his work-related trips happening near areas in South Los Angeles during non-peak hours (e.g., 9 A.M.–2 P.M.), when traffic congestion there is still severe.

Further, Figure 10 depicts the exposures to traffic congestion of 32 individuals from 15 households in the subsample for both the commute-only assessments and assessments that also consider their activity-travel patterns. The bar graph clearly shows that individuals from the same household are differently exposed to traffic congestion because individuals' activity-travel patterns are idiosyncratic.

Based on these results, we answer the third question: How are individuals uniquely exposed to traffic congestion as a result of their activity-travel patterns? The results confirm that individuals are idiosyncratically exposed to traffic congestion due to their distinctive activity-travel patterns. The results also corroborate previous studies [37,56], which argue that individuals from the same household are differently exposed to environmental influences or contexts. Therefore, these results may cast doubt on the validity of previous studies' exclusive focus on commuting trips during peak hours and using data aggregated over areas (e.g., census tracts) because they did not fully reflect individuals' unique activity-travel patterns.

**Figure 10.** Individual exposures to traffic congestion of 32 individuals from 15 households in the subsample (Same alphabet means the same household.)
