7.8 Household Travel Survey and Replica Data Comparison Final Memo Memorandum TO: Mid America Regional Council FROM: Jason Lemp, Amit Mondal, Haiyun Lin, and Brent Selby, Cambridge Systematics DATE: November 26, 2019 RE: Task 7 - Analysis of Replica Data and Household Travel Survey This draft final technical memorandum describes our analysis of the recently completed How We Move KC household travel survey data for the Kansas City area as they compare to the Replica data for the region. Sideby-side comparisons of the datasets along a variety of dimensions are presented to identify key differences and similarities. In addition, the strengths and weaknesses of the Replica data are identified and recommendations are made on how these data can be used. 1.0 Introduction 1.1 MARC Household Travel Survey The How We Move KC household travel survey1 (HTS) for the region was conducted in the spring of 2019, and asked residents in the Kansas City region to provide demographic and travel pattern information about their households. In addition, the survey included a one-day travel diary where all travel and activities for each person in each surveyed household were recorded. Travel diary data were collected from survey respondents on weekdays during April and May 2019. The data were processed and cleaned to ensure that the final set of households in the sample was sufficient for modeling and other analyses for which the survey sample will be used. The final sample size of the survey was about 3,800 households and was weighted using the American Community Survey (ACS) 5-year data from 2013 to 2017. CS prepared all summaries of the household travel survey data presented in this report. Segmentation and categorization of the survey data were performed in order to provide consistent tables with those used in the Replica summaries. All survey summaries use weighted data. 1.2 Replica Replica is a planning tool that provides estimates of travel patterns and movements in a region. It uses a process that creates synthetic representations of all households and persons in the region, segmented across a variety of dimensions such as household income or person age. It also simulates the travel patterns of each synthesized household and person, providing travel diary like information that can be summarized in various ways. While the Replica model can be viewed as providing an estimate of all travel movements in a region, these estimates are simulations, they are not observed from data. Travel patterns in Replica are reported for an average day in a particular 3-month period. A variety of travel pattern information is generated and includes the following variables: 1 For more information about the survey, refer to: 2018 Kansas City Regional Household Travel Survey Final Report, Prepared by Westat, November 2019. 2018 Kansas City Regional Household Travel Survey Final Report 1 • Trip purpose; • Trip start/end locations; • Travel mode; and • Trip start/end times. A collection of data sources is used to support the models that underlie the synthetic travel pattern data, including Census data, land use and real estate data, mobile phone location data, transportation infrastructure data, and ground truth data like traffic counts. For the analyses conducted in this report, Sidewalk Labs prepared summaries of the Replica model data based upon a data summary request prepared by CS. The summaries used Replica data for an average weekday from the spring quarter (months of March, April, and May) in 2019. The Kansas City Replica model features travel beyond the 8-county region boundaries. In order to ensure a one-to-one comparison with the MARC household travel survey data (which collected data only for households in the 8-county region), Replica summaries were prepared for households inside the 8-county region, which includes Cass, Clay, Jackson, and Platte Counties in Missouri and Johnson, Leavenworth, Miami, and Wyandotte Counties in Kansas. 1.3 Purpose of this Technical Document This technical memorandum focuses on the evaluation of the Replica data, especially with respect to the household travel survey data. Side-by-side comparisons of household survey and Replica data summaries are presented and the differences between the datasets are documented. In some cases, other datasets were summarized as additional points of comparison, including ACS data and National Household Travel Survey (NHTS) data. ACS data were used to provide additional context to household formation analyses while NHTS data were used to provide additional travel patterns at a national level, especially in cases where the MARC household survey data were too thin to draw firm conclusions. The key areas of focus for the analyses in this document include the following: • Household Formation – These analyses include examining demographics and other key household level attributes of the Replica synthetic population. • Travel Patterns – This section focuses on the analysis of trip rates, trip purposes, trip lengths, travel modes, and time of day travel by market segment with special focus on key trip segments like transit and trips made using Transportation Network Companies (TNC). • Logical Checks – This includes an examination of the reasonableness of Replica data summaries based on our professional experience and general expectations about the region’s travel patterns. Analyses provided in this document focus on identifying strengths and weaknesses of the Replica data and how these data can be used in transportation planning and modeling applications. The analyses and documentation aim at providing answers to the following key questions: • Do the Replica travel patterns match patterns observed in the household travel survey and other data sources and do these make sense and meet our expectations? • In what ways do the Replica data provide value over and above the value that the household travel survey provides? • How should the Replica data be used and what pitfalls should be avoided in using these data? 2018 Kansas City Regional Household Travel Survey Final Report 2 Section 2 focuses on household formation while Section 3 focuses on travel pattern analyses of the data. The final Section 4 provides our conclusions and recommendations moving forward. 2018 Kansas City Regional Household Travel Survey Final Report 3 2.0 Household Formation Analyses 2.1 Regional Population and Trip-Making One of the basic elements of a tool like Replica is the total synthesized population and the total trips estimated by the tool. Table 1 shows the total number of households, total population, and total region-wide daily trips generated in Replica and compared to the MARC HTS, ACS, and NHTS2 samples. In total, Replica reports 5 percent more households in the region than the MARC HTS or ACS, while Replica’s total population figure is 2 percent lower than MARC HTS and 3 percent higher than ACS. These differences result in a range of estimates of average household size from a low of 2.39 reported for Replica and a high of 2.57 reported for MARC HTS. Replica also reports 9 percent fewer trips in total than MARC HTS. Table 1. Total Population and Trip-Making across Data Sources NHTS Metric Total Households Replica MARC HTS ACS (all of U.S., weekdays) 770,343 84,434,466 807,958 770,343 1,929,939 1,977,768 2.39 2.57 6,156,334 6,757,803 Household Trip Rate 7.62 8.77 n/a 8.86 Person Trip Rate 3.19 3.42 n/a 3.48 Total Population Persons per Household Total Trips 1,878,026 215,100,755 2.44 2.55 n/a 747,721,033 These differences are not large, but taken together, they result in daily trip rates that are lower than reported by the MARC HTS. Specifically, Replica reports a 13 percent lower household trip rate and an 7 percent lower person trip rate. It is worth noting here that non-school travel by children is not reported in Replica due to privacy concerns, whereas non-school travel is included in the HTS summaries. This would partially explain the differences in total trips and trip rates between the data sources. When compared to the national level estimates, the NHTS reports household and person trip rates that are roughly in line with the MARC survey. Traditionally a rule of thumb for travel modeling has been to expect 4 daily trips per person and 9-10 daily trips per household. However, it has been reported that trip rates are dropping3, largely as a result of reductions in shopping and personal business trips. In this respect, both the MARC and Replica estimates of lower household and person trip rates are probably reasonable estimates of current travel patterns with the MARC HTS survey providing estimates that are closer to past experience. 2 NHTS totals are for the entire U.S., but rate statistics like persons per household and trip rates are useful for comparison purposes. 3 See, e.g., http://onlinepubs.trb.org/onlinepubs/Conferences/2018/NHTS/McGuckinTravelTrends.pdf. 2018 Kansas City Regional Household Travel Survey Final Report 4 2.2 Household Composition In addition to the total number of households and population, it is important to understand the composition of households across key demographic attributes like household size, household income, number of workers, and number of vehicles. Since ACS data were used to weight the MARC HTS sample, we would expect that the distributions of households across these attributes will match closely between these two datasets. Since Replica also uses Census data products to calibrate their underlying model, we expect a close match as well. Figure 1 shows the distribution of households by household size for Replica, the MARC HTS, and ACS. The ACS and MARC HTS distributions match very closely, while Replica shows a slightly larger number of oneperson households and fewer larger households. Figure 1. Percent of Region-wide Households by Household Size 40.0 35.0 Household % 30.0 25.0 20.0 15.0 10.0 5.0 0.0 1 2 3 4 5/+ Household Size MARC HTS Replica ACS Figure 2 presents the distribution of households by the number of workers in each household. Results are very similar across all three datasets. Household % Figure 2. Percent of Region-wide Households by Number of Workers 90.0 80.0 70.0 60.0 50.0 40.0 30.0 20.0 10.0 0.0 0 1/+ Workers MARC HTS 2018 Kansas City Regional Household Travel Survey Final Report Replica ACS 5 Figure 3 shows the distribution of household income across the three datasets. While there is slightly more fluctuation in the results of both Replica and the MARC HTS compared to ACS, the three datasets are mostly well-aligned. One notable difference is that Replica indicates a larger number of low income ($0-10,000) households than the ACS. Figure 3. Percent of Region-wide Households by Household Income Category 30.0 Household % 25.0 20.0 15.0 10.0 5.0 0.0 Income MARC HTS Replica ACS Figure 4 shows the distribution of vehicle ownership across the datasets. Again, the three datasets match very well along this dimension. It is worth noting that 3.8 percent of households in Replica are categorized as “Unknown” for vehicle ownership. Figure 4. Percent of Region-wide Households by Vehicle Ownership 45.0 40.0 Household % 35.0 30.0 25.0 20.0 15.0 10.0 5.0 0.0 0 1 2 3/+ Vehicle Ownership MARC HTS 2018 Kansas City Regional Household Travel Survey Final Report Replica ACS 6 Unknown 2.3 Household Two-Way Cross-Classifications The analyses in the previous sub-section focus on the composition of households along a single household attribute. Since these attributes are often related to one another, we examine the household composition along two household attributes at a time. Because the ACS does not provide this information, comparisons were made between the MARC HTS and Replica. Table 2 shows the two-way distribution of households from the weighted MARC HTS compared to Replica. Values in the table are expressed as the percentage of all region-wide households. In general, the patterns of household size and income distribution match quite well between the two datasets. We should note that Replica data show a larger number of one-person, and low income ($0-10,000) households compared to the HTS. 2018 Kansas City Regional Household Travel Survey Final Report 7 Table 2. Household Size vs. Household Income of MARC HTS vs. Replica (Percent of Region-wide Households) 2018 Kansas City Regional Household Travel Survey Final Report 8 MARC HTS Household Size Household Income 1 2 3 4 0K-10K 3.04 1.32 0.55 0.66 0.37 5.93 10K-20K 4.18 1.60 0.82 0.21 0.74 7.55 20K-30K 4.37 2.02 1.32 0.53 0.48 8.73 30K-40K 5.21 3.13 1.26 0.56 0.62 10.77 40K-50K 2.84 2.35 1.17 0.30 0.60 7.26 50K-60K 2.66 2.75 1.06 0.70 0.42 7.59 60K-100K 4.97 9.25 4.12 3.72 1.90 23.95 100K-150K 1.19 6.15 2.57 3.57 2.02 15.50 150K-200K 0.21 2.53 1.34 2.05 0.77 6.90 200K+ 0.19 2.15 0.76 1.21 0.69 5.00 Unknown 0.35 0.27 0.17 0.03 0.00 0.82 29.21 33.53 15.14 13.53 8.59 100.00 Total 5/+ Total Replica Household Size Household Income 1 2 3 4 0K-10K 6.77 0.84 0.50 0.28 0.22 8.61 10K-20K 4.83 1.42 0.79 0.47 0.36 7.87 20K-30K 4.33 2.18 0.78 0.61 0.51 8.42 30K-40K 3.73 2.61 0.99 0.74 0.34 8.41 40K-50K 3.28 3.48 1.03 0.73 0.67 9.19 2018 Kansas City Regional Household Travel Survey Final Report 5/+ 9 Total 50K-60K 2.35 2.75 1.06 0.60 0.73 7.49 60K-100K 4.51 9.21 4.26 3.31 2.31 23.59 100K-150K 1.22 5.43 3.02 3.13 1.96 14.75 150K-200K 0.32 2.16 1.08 1.33 0.91 5.80 200K+ 0.40 2.26 1.02 1.30 0.88 5.87 Unknown 0.00 0.00 0.00 0.00 0.00 0.00 31.75 32.34 14.53 12.50 8.88 100.00 Total Table 3 shows the two-way distribution of household workers and household income for the MARC HTS and for Replica. Patterns are again very similar between the two data sources, but the zero-worker and low income category has a somewhat larger share of households in Replica. 2018 Kansas City Regional Household Travel Survey Final Report 10 Table 3. Household Workers vs. Household Income of MARC HTS vs. Replica (Percent of Region-wide Households) 2018 Kansas City Regional Household Travel Survey Final Report 11 MARC HTS Workers Household Income 0 1/+ Total 0K-10K 3.9 2.0 5.9 10K-20K 4.2 3.4 7.6 20K-30K 3.1 5.7 8.7 30K-40K 2.8 8.0 10.8 40K-50K 1.3 5.9 7.3 50K-60K 1.5 6.1 7.6 60K-100K 3.6 20.3 24.0 100K-150K 1.3 14.2 15.5 150K-200K 0.5 6.4 6.9 200K+ 0.2 4.8 5.0 Unknown 0.3 0.5 0.8 22.8 77.2 100.0 Total Replica Workers Household Income 0 1/+ Total 0K-10K 6.9 1.7 8.6 10K-20K 4.3 3.6 7.9 20K-30K 3.2 5.2 8.4 30K-40K 2.1 6.3 8.4 40K-50K 2.2 7.0 9.2 2018 Kansas City Regional Household Travel Survey Final Report 12 50K-60K 1.4 6.1 7.5 60K-100K 2.8 20.8 23.6 100K-150K 1.1 13.7 14.8 150K-200K 0.3 5.5 5.8 200K+ 0.6 5.3 5.9 Unknown 0.0 0.0 0.0 24.7 75.3 100.0 Total Table 4 shows the two-way distribution of household size and vehicle ownership for MARC HTS and Replica. Comparisons between the two datasets suggest that they follow similar patterns. 2018 Kansas City Regional Household Travel Survey Final Report 13 Table 4. Household Size vs. Vehicle Ownership of MARC HTS vs. Replica (Percent of Region-wide Households) MARC HTS Household Size Vehicle Ownership 1 2 3 4 0 4.20 1.07 0.34 0.19 0.00 5.79 1 20.76 6.65 2.97 1.40 1.37 33.15 2 3.51 19.81 6.28 7.50 3.14 40.24 3/+ 0.74 6.00 5.55 4.44 4.08 20.81 Unknown 0.00 0.00 0.00 0.00 0.00 0.00 29.21 33.53 15.14 13.53 8.59 100.00 Total 5/+ Total Replica Household Size Vehicle Ownership 1 2 3 4 0 4.10 0.97 0.43 0.28 0.18 5.95 1 19.52 6.81 2.85 1.59 1.10 31.87 2 3.51 18.75 6.16 6.17 3.88 38.46 3/+ 0.82 5.81 5.10 4.45 3.72 19.90 Unknown 3.81 0.00 0.00 0.00 0.00 3.81 31.75 32.34 14.53 12.50 8.88 100.00 Total 5/+ Total Table 5 shows the two-way distribution of vehicle ownership and household income for MARC HTS and Replica. Again, the patterns in the two-way distribution are broadly similar. One element that we noticed with the Replica data is that there is a segment of the synthetic households that falls into the category of “Unknown” vehicle ownership. This same household segment is also categorized as having household size of one (see Table 4), as having a low household income of $0-10,000 (see Table 5), and as having zero workers (not shown explicitly). This segment makes up 3.8 percent of region-wide households in the Replica model. Based upon the fact that no other household size, household income, or number of worker 2018 Kansas City Regional Household Travel Survey Final Report 14 designation has unknown vehicle ownership levels, it may be that the household size, household income, and number of worker designations are not real estimates of households in those categories, but simply reflect this other/unknown household category. Further clarification from Sidewalk Labs on this question may be warranted. 2018 Kansas City Regional Household Travel Survey Final Report 15 Table 5. Vehicle Ownership vs. Household Income of MARC HTS vs. Replica (Percent of Region-wide Households) 2018 Kansas City Regional Household Travel Survey Final Report 16 MARC HTS Vehicle Ownership Household Income 0 1 2 0K-10K 2.55 2.43 0.74 0.20 0.00 3.37 10K-20K 1.77 4.48 1.00 0.31 0.00 5.79 20K-30K 0.63 5.31 1.94 0.85 0.00 8.10 30K-40K 0.29 6.15 3.41 0.93 0.00 10.48 40K-50K 0.09 3.90 2.53 0.74 0.00 7.17 50K-60K 0.09 3.23 3.11 1.16 0.00 7.50 60K-100K 0.22 5.39 12.46 5.88 0.00 23.74 100K-150K 0.09 1.33 8.80 5.27 0.00 15.41 150K-200K 0.00 0.37 3.40 3.13 0.00 6.90 200K+ 0.02 0.23 2.58 2.16 0.00 4.98 Unknown 0.04 0.33 0.26 0.19 0.00 0.78 Total 5.79 34.15 42.24 20.81 0.00 94.21 3/+ Unknown Total Replica Vehicle Ownership Household Income 0 1 2 0K-10K 1.61 2.31 0.67 0.22 3.81 8.6 10K-20K 1.79 4.51 1.26 0.30 0.00 7.9 20K-30K 0.96 5.05 1.91 0.49 0.00 8.4 30K-40K 0.44 4.71 2.62 0.64 0.00 8.4 40K-50K 0.36 4.22 3.54 1.07 0.00 9.2 2018 Kansas City Regional Household Travel Survey Final Report 3/+ 17 Unknown Total 50K-60K 0.21 2.96 3.11 1.20 0.00 7.5 60K-100K 0.38 5.68 11.73 5.80 0.00 23.6 100K-150K 0.10 1.59 7.84 5.23 0.00 14.8 150K-200K 0.03 0.39 2.88 2.49 0.00 5.8 200K+ 0.06 0.43 2.90 2.48 0.00 5.9 Unknown 0.00 0.00 0.00 0.00 0.00 0.0 Total 5.95 32.86 40.46 19.92 3.81 100.00 2018 Kansas City Regional Household Travel Survey Final Report 18 3.0 Travel Pattern Analyses This Section focuses on the analysis of travel patterns contrasting the MARC survey and the Replica dataset as two sources of data. We first focus on Trip Rates in Section 3.1 examining how trip making varies by household characteristics. In Section 3.2 we analyze the distances traveled in total and by purpose. Section 3.3 focuses on origindestination markets examining the county-to-county flows and the district-to-district flows. In Section 3.4 we focus on the temporal distribution of trips examining the start times for trips across different purposes. Section 3.5 presents the implied market shares by mode in the two data sources examining differences by age and examining the airport access as a key travel generator. Finally, Sections 3.6 and 3.7 discuss the transit trips and the on-demand TNC trips observed in each of the two data sources. 3.1 Trip Rates Average daily trip rates vary across different types of households reflecting the impact of household size, income, trip travel purpose, and a person’s age on their overall trip making and trip rates. Figure 5 shows the average household trip rates by household size from the MARC HTS, Replica, and the NHTS nationwide sample. Trends across the datasets are consistent, with trip rates rising with larger household sizes. Replica trip rates are slightly lower across the board, which is consistent with the earlier finding that the overall trip rates are lower in the Replica model. Figure 5. Average Household Trip Rates by Household Size 20.0 Household Trip Rate 18.0 16.0 14.0 12.0 10.0 8.0 6.0 4.0 2.0 0.0 1 2 3 4 5 or more Household Size MARC HTS Replica NHTS (all of U.S.) Figure 6 shows average household trip rates by household income from the same three datasets4. Again, trip rate trends across income categories are consistent with the number of trips increasing with income as expected. 4 Note that the NHTS income categories do not align exactly with those used in the MARC HTS and those shown in the figure. The NHTS values represent our estimates of the trip rates for each income category based upon estimates of which income category households fell. 2018 Kansas City Regional Household Travel Survey Final Report 19 When we compare the patterns across the three datasets, we notice only slight differences in the magnitudes of trip rates with Replica data providing lower trip rates consistent with our earlier observations. Figure 6. Average Household Trip Rates by Household Income Household Trip Rate 14.0 12.0 10.0 8.0 6.0 4.0 2.0 0.0 Household Income MARC HTS Replica NHTS (all of U.S.) Note that the “Unknown” category includes all survey responses where the household income could not be ascertained (including refused and do not know). 3.1.1 Trips by Purpose Figure 7 shows trip rates by trip purpose for the three datasets. Eight trip purposes are defined in Figure 7 which shows the percentage of each trip category as a share of the total daily trips. Overall, the MARC HTS and NHTS samples are very consistent in terms of the assigned trip purposes for those datasets. Replica data show several differences compared to these two survey data sources that are discussed separately for four groups of trip purposes: • Work-related travel. Replica shows a higher incidence of work travel. The Home-based Work and Work-based Other trips have a share that is about 20 to 25 percent higher compared to the survey data. • School and College travel. Replica shows a much higher incidence of school and college related travel. Compared with the 6 to 7 percent share of school trips in the two surveys, Replica shows 11 percent of trips heading to school. Similarly, the two surveys suggest a 1 percent trip share for college trips compared to a 3 percent share of college trips in the Replica data. • Shopping, Social/Recreational, and Other home-based travel. The distribution of the mix of trips related to home-based shopping and other trip purposes is different in the Replica compared to the two surveys. Replica sees higher levels of shopping and lower levels of other trip making. Social/recreational trips are comparable between NHTS and Replica (7 versus 6 percent) and a bit higher in the MARC survey (11 percent). • Non-home based travel. The share of total non-home based trips (Work-Based Other plus Other-Based Other) in Replica is about 28 percent which is comparable with the survey shares of 29 to 32 percent for these trip purposes. In other words, the split of trips between home-based and non-home based across the three datasets is consistent. 2018 Kansas City Regional Household Travel Survey Final Report 20 Percent of Trips Figure 7. Percent of Trips by Trip Purpose 100% 90% 80% 70% 60% 50% 40% 30% 20% 10% 0% MARC HTS Replica NHTS (all of U.S.) Data Source HB Work HB School HB College HB Shopping HB Social/Recreation HB Other Work Based Other Other Based Other Note: HB is an abbreviation for “Home-Based”, meaning that one end of the trip is at home. We also checked trip rates by purpose for different age categories to confirm a reasonable correspondence. For instance, we found that young children made no HB Work or HB College trips, which is a result that we expected. We also found that no adults made HB School trips, which was also an expected result. The Replica model was reasonable along these logic checks. Trip purpose, as used in travel modeling, describes both the actual purpose of the activity as well as whether the trip is home-based or non-home based. The Replica data match the split between home-based and non-home based trips reasonably well. However, there are differences between the surveys and Replica when it comes to travel purposes. This likely reflects differences in how this information is collected or processed for each data source. In surveys, the purpose of an activity is collected directly from the survey respondent. The Replica model uses cell phone location data to infer trip ends and the activity purposes at those trip ends since activity purpose is not directly observed. As such, it makes sense that there are differences between the survey data and Replica. Understanding how trip purposes are defined by Replica would help users to understand how the trip purpose information should be used. This is especially the case for users that are accustomed to traditional survey trip purpose definitions. 3.1.2 Trips by Age Figure 8 shows the variations in person trip making by age category. Trip rates for most age categories (20 years to 89 years) are quite comparable between the two datasets. However, Replica shows much lower trip rates for children which are 26 percent lower for children less than 10 years old and 31 percent lower for children 10 to 19 years old. This is largely because Replica simulates only the school travel for children (leaving out other trip types) due to privacy concerns. We also see much higher trip rate in Replica for adults who are 90 years or older. 2018 Kansas City Regional Household Travel Survey Final Report 21 Trips Per Person Figure 8. Average Person Trip Rates by Age Category 4.50 4.00 3.50 3.00 2.50 2.00 1.50 1.00 0.50 0.00 00-09 yo 10-19 yo 20-29 yo 30-39 yo 40-49 yo 50-59 yo 60-69 yo 70-79 yo 80-89 yo 90 yo/+ Unknown Age Category MARC HTS Replica 3.2 Trip Lengths by Purpose and Time of Day Figure 9 depicts the distribution of trip lengths from the MARC HTS as they compare with the Replica patterns. Overall, Replica predicts fewer short trips that are less than seven miles and slightly more long trips. These distributions imply average trip lengths of 6.5 miles for the MARC HTS and 8.1 miles for Replica, which is 25 percent higher than the survey estimate.5 Figure 9. Trip Length Distribution Trips Percent 20.0 15.0 10.0 5.0 0.0 0 5 10 15 20 25 30 35 40 Trip Length (mi) MARC HTS Replica Our experience using cell phone data aligns with this finding. When inferring trips from cell phone data, certain criteria are typically used, including spatial clustering algorithms and minimum stop durations, both of which can 5 Note that we excluded trips over 75 miles in the calculation of average trip lengths. These trips make up a very small percentage of overall trips in both datasets. This was primarily done to avoid inclusion of long-distance trips (e.g., those made by air) made by residents of the region into or out of the region (i.e., trips where one trip end is external to the region). 2018 Kansas City Regional Household Travel Survey Final Report 22 create difficulties in detecting short trips. These difficulties can result in lower levels of short trip making resulting in higher average trip lengths. Average trip lengths can be further subdivided by trip purpose, as shown in Figure 10. Trips with one end at work or at school (HB Work, HB School, and WB Other) have very similar average trip lengths between the two datasets. On the other hand, average trip lengths for other trip purposes are higher, on average, in the Replica dataset reflecting the observation made earlier about the nature of the assumptions made when considering activities and stop durations. Figure 10. Average Trip Lengths by Trip Purpose 16.0 Average Trip Length (mi) 14.0 12.0 10.0 8.0 6.0 4.0 2.0 0.0 HB Work HB School HB College HB Shopping HB Soc/Rec MARC HTS HB Other WB Other OB Other Replica Figure 11 shows average trip lengths for different start times during the day. Most hours of the day show noticeably higher average trip lengths from Replica except for the period between 11 pm and 6 am, where MARC HTS shows longer trips on average. In particular, the high average trip lengths in the survey between 4 AM and 6 AM probably correspond to individuals that have very long commutes, and these commute trips represent a larger share of the trips beginning during this time period. Replica’s very low trip lengths of less than one mile from 1 AM to 3 AM are likely not reasonable. 2018 Kansas City Regional Household Travel Survey Final Report 23 Figure 11. Average Trip Lengths by Start Hour 30.0 Average Trip Length (mi) 25.0 20.0 15.0 10.0 5.0 0.0 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 Trip Start Hour MARC HTS Replica 3.3 Origin-Destination Trip Flows The geographic distribution of daily trips in the region provides a key input to planning and modeling activities. The comparisons of the HTS and the Replica data focus on origin-destination (O-D) travel patterns derived from each data source. 3.3.1 O-D Trip Flows by County Table 6 shows the county-to-county trip flows as a percentage of the total trips. Trends in the county-to-county flows are generally similar between the HTS and Replica. One notable difference is observed in Johnson County, where the HTS shows about 35 percent of trips originating from the county compared to Replica which shows only about 30 percent of trips originating from the county (see row totals in Table 6). This could be a result of the relatively higher share of overall households in Johnson County from the survey (about 33 percent of households compared to the ACS total of about 30 percent).6 6 Note that while the HTS summaries are weighted using the expansion factors by household, the weights did not apply at the district or county level, which is the reason for the disparity between ACS and HTS. 2018 Kansas City Regional Household Travel Survey Final Report 24 Table 6. County-to-County Trip Flow Percentages 2018 Kansas City Regional Household Travel Survey Final Report 25 MARC HTS Clay County, MO Jackson County, MO Johnson County, KS Leavenworth County, KS Miami County, KS Platte County, MO Wyandotte County, KS Buffer Region 0.24 0.02 0.08 0.15 0.26 0.05 0.01 0.15 0.02 0.98 Cass County, MO 0.04 2.61 0.01 0.45 0.17 0.00 0.01 0.01 0.01 3.31 Clay County, MO 0.07 0.01 7.70 1.14 0.27 0.04 0.00 0.75 0.11 10.08 Jackson County, MO 0.23 0.47 1.07 29.68 2.93 0.04 0.01 0.48 0.58 35.50 Johnson County, KS 0.30 0.17 0.30 2.90 29.32 0.11 0.22 0.21 1.09 34.62 Leavenworth County, KS 0.04 0.00 0.03 0.04 0.10 1.97 0.00 0.03 0.18 2.40 Miami County, KS 0.01 0.01 0.00 0.01 0.21 0.00 0.59 0.00 0.01 0.84 Platte County, MO 0.09 0.01 0.80 0.44 0.25 0.03 0.00 3.15 0.14 4.91 Wyandotte County, KS 0.04 0.01 0.12 0.57 1.10 0.17 0.01 0.14 5.22 7.37 Total 1.07 3.30 10.10 35.38 34.60 2.40 0.84 4.93 7.37 100.0 Trip Start Total Cass County, MO Buffer Region Trip End Replica Jackson County, MO Johnson County, KS Leavenworth County, KS Miami County, KS Platte County, MO Wyandotte County, KS 0.41 0.08 0.13 0.27 0.28 0.12 0.05 0.07 0.07 1.48 Cass County, MO 0.08 2.80 0.04 0.87 0.31 0.00 0.03 0.01 0.04 4.18 Clay County, MO 0.13 0.04 7.61 1.52 0.38 0.03 0.00 0.82 0.26 10.78 Jackson County, MO 0.26 0.87 1.52 28.85 3.16 0.08 0.03 0.51 0.89 36.17 Johnson County, KS 0.28 0.30 0.37 3.17 23.48 0.19 0.30 0.23 1.43 29.76 Leavenworth County, KS 0.12 0.00 0.03 0.08 0.19 2.64 0.00 0.11 0.26 3.44 Miami County, KS 0.05 0.03 0.00 0.03 0.30 0.00 0.86 0.00 0.02 1.30 Platte County, MO 0.08 0.01 0.82 0.51 0.22 0.11 0.00 2.73 0.24 4.73 2018 Kansas City Regional Household Travel Survey Final Report 26 Total Clay County, MO Buffer Region Trip Start Buffer Region Cass County, MO Trip End Wyandotte County, KS 0.07 0.04 0.26 0.88 1.43 0.26 0.02 0.24 4.95 8.15 Total 1.48 4.18 10.79 36.17 29.76 3.44 1.30 4.73 8.15 100.0 Table 7 shows county-to-county trip flow percentages for HBW trips between the two datasets. There is generally agreement between the two datasets, and similar patterns are present in the HBW trip patterns as were seen in Table 6 of overall trips, not surprisingly, but some differences are present in comparing HBW to overall trips. Specifically, relative HBW trip rates are lower for intra-county trips and higher for inter-county trips. Table 7. County-to-County HBW Trip Flow Percentages MARC HTS Cass County, MO Clay County, MO Jackson County, MO Johnson County, KS Leavenworth County, KS Miami County, KS Platte County, MO Wyandotte County, KS Buffer Region 0.00 0.04 0.06 0.20 0.43 0.11 0.00 0.15 0.04 1.03 Cass County, MO 0.03 1.20 0.01 0.75 0.46 0.00 0.01 0.02 0.01 2.50 Clay County, MO 0.06 0.01 3.95 2.86 0.95 0.14 0.00 0.68 0.25 8.90 Jackson County, MO 0.08 0.94 2.43 25.19 6.84 0.05 0.00 0.84 1.26 37.63 Johnson County, KS 0.46 0.46 0.88 6.74 23.68 0.22 0.50 0.53 2.17 35.63 Leavenworth Co, KS 0.17 0.00 0.12 0.06 0.19 1.48 0.00 0.03 0.27 2.33 Miami County, KS 0.03 0.01 0.00 0.06 0.43 0.00 0.18 0.00 0.04 0.75 Platte County, MO 0.12 0.02 0.79 0.85 0.66 0.05 0.00 1.60 0.38 4.48 Wyandotte County, KS 0.00 0.01 0.36 1.01 1.86 0.22 0.04 0.30 2.94 6.76 Total 0.95 2.69 8.60 37.73 35.51 2.27 0.73 4.15 7.36 100.0 Total Trip Start Buffer Region Trip End Replica Cass County, MO Clay County, MO Jackson County, MO Johnson County, KS Leavenworth County, KS Miami County, KS Platte County, MO Wyandotte County, KS Buffer Region 0.00 0.03 0.07 0.12 0.19 0.11 0.03 0.08 0.05 0.67 Cass County, MO 0.03 1.68 0.05 1.22 0.53 0.00 0.01 0.02 0.07 3.62 2018 Kansas City Regional Household Travel Survey Final Report 27 Total Trip Start Buffer Region Trip End Clay County, MO 0.08 0.05 5.34 2.69 0.68 0.05 0.00 1.16 0.45 10.50 Jackson County, MO 0.14 1.13 2.55 23.94 5.27 0.10 0.04 0.89 1.13 35.19 Johnson County, KS 0.22 0.44 0.66 5.37 23.32 0.26 0.41 0.35 2.28 33.30 Leavenworth Co, KS 0.12 0.00 0.05 0.11 0.28 1.96 0.00 0.12 0.30 2.93 Miami County, KS 0.03 0.01 0.00 0.05 0.49 0.00 0.55 0.00 0.04 1.17 Platte County, MO 0.08 0.02 1.12 0.94 0.37 0.13 0.00 1.85 0.29 4.80 Wyandotte County, KS 0.06 0.06 0.42 1.13 2.28 0.27 0.03 0.28 3.28 7.82 Total 0.76 3.42 10.26 35.58 33.40 2.88 1.08 4.75 7.89 100.0 3.3.2 O-D Trip Flows by District We also examined trip-making at the trip district level by breaking the region into 16 districts. Figure 12 shows the district definition boundaries. Leavenworth (District 8), Miami (District 11), and Cass (District 13) Counties 2018 Kansas City Regional Household Travel Survey Final Report 28 are each treated as districts on their own, while each of the other counties are divided into two or three districts each. District 5 represents the Kansas City airport. Figure 12. MARC Region District Definitions The overall breakdown of trip origins and destinations by district is shown in Figure 13. Similar to the countyto-county flow table above, Districts 7, 10, and 15, which make up Johnson County, are shown to have higher numbers of trips in the HTS than in Replica. Overall, however, trends across the districts generally match between the two datasets. 2018 Kansas City Regional Household Travel Survey Final Report 29 Percent of Trips Figure 13. Percent of Trip Origins and Destinations by District 18.0 16.0 14.0 12.0 10.0 8.0 6.0 4.0 2.0 0.0 District MARC HTS Trip Origins Replica HTS Trip Origins MARC HTS Trip Destinations Replica Trip Destinations Table 8 shows the percent of trips to each destination district given a fixed origin district. For instance, in the first row, the numbers in that row indicate the percent of trips that originate in origin District 1 that have a destination in each destination district. The following observations can be made from the comparison of HTS data to Replica data: • The diagonal elements of the table reflect trip making within each district and are all higher in the HTS compared to the Replica data. This is consistent with the shorter trip lengths in the HTS since intradistrict trips are shorter than inter-district trips. • While there are differences in intra-district trips, the trends in the distribution of trips between different districts tend to be similar between the two datasets in most cases. • o In both sources of data most trips are within the same district followed by trips that are to/from neighboring districts. o As an example, both datasets indicate that trips originating in District 9 (Kansas City Downtown) are most likely intra-district (48 percent in HTS and 36 percent in Replica), followed by trips destined for District 14 (20 percent in each dataset), and then followed by the third most likely destination in District 7 (9 percent each). o Some differences still exist for selected origin-destination pairs. For instance, the fourth and fifth most likely destinations for District 9 trip origins in the HTS is District 4 (6 percent) and District 12 (5 percent) while for Replica, these rankings are reversed with District 12 (9 percent) and District 4 (6 percent). With that being said, it is important to remember that as the trip shares become smaller, the HTS sample sizes become smaller and less reliable, so some of these variations could be attributed to noise in the underlying data. Noise in the HTS data is likely the reason for differences in the trip destinations for trips from airport District 5, where HTS data are quite thin with a sample size of 60-70 trips. Table 8. Trip Destination District Percentages for each Origin Trip District MARC HTS 2018 Kansas City Regional Household Travel Survey Final Report 30 Destination District (%) Origin District 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Total 1 72 5 9 6 0 0 2 2 2 0 0 0 0 0 0 3 100 2 1 61 2 24 0 0 2 1 2 0 0 2 0 3 1 1 100 3 1 1 64 16 1 1 2 0 5 1 0 1 0 3 1 2 100 4 0 9 8 65 0 1 2 0 5 0 0 2 0 5 0 1 100 5 1 3 6 11 6 0 22 1 9 10 0 15 3 4 7 1 100 6 0 0 2 1 0 57 8 6 2 2 0 1 0 2 3 17 100 7 0 0 1 1 0 2 68 0 4 6 0 1 0 5 8 3 100 8 1 1 0 1 0 8 4 81 1 1 0 1 0 0 1 1 100 9 0 1 3 6 0 1 9 0 48 1 0 5 1 20 3 2 100 10 0 0 0 1 0 1 11 0 1 70 1 0 0 2 12 1 100 11 0 0 0 0 0 1 3 0 0 15 71 0 1 1 8 0 100 12 0 0 0 1 0 0 1 0 2 0 0 83 1 7 1 0 100 13 0 0 0 0 0 0 1 0 1 1 0 6 79 7 3 0 100 14 0 1 1 2 0 1 6 0 10 1 0 7 2 64 4 1 100 15 0 0 0 0 0 1 14 1 2 13 1 2 1 6 57 1 100 16 1 0 2 2 0 19 13 1 5 2 0 1 0 5 3 47 100 Replica Destination District (%) Origin District 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 Total 1 56 4 13 6 1 1 1 8 3 1 0 1 0 2 1 1 100 2 1 61 4 19 0 1 1 0 3 0 0 4 0 3 1 1 100 3 3 3 53 16 1 3 3 1 5 1 0 2 0 5 1 3 100 4 1 9 8 58 0 1 2 0 6 1 0 4 0 6 1 2 100 5 11 6 29 17 7 3 4 1 2 2 0 5 0 8 4 1 100 6 0 1 3 2 0 51 12 6 3 3 0 1 0 3 3 12 100 2018 Kansas City Regional Household Travel Survey Final Report 31 7 0 0 1 1 0 4 58 1 5 7 0 2 1 6 11 4 100 8 2 0 1 1 0 6 4 78 1 2 0 0 0 1 1 1 100 9 0 2 3 6 0 2 9 1 36 2 0 9 2 20 4 4 100 10 0 0 0 1 0 1 11 1 2 62 2 1 1 3 14 1 100 11 0 0 0 0 0 1 5 0 1 12 69 0 2 1 7 1 100 12 0 1 1 2 0 0 2 0 4 1 0 74 2 11 2 1 100 13 0 0 0 1 0 0 2 0 3 1 1 9 68 10 4 0 100 14 0 1 1 3 0 1 5 0 9 1 0 10 3 60 4 1 100 15 0 0 1 1 0 1 17 1 3 13 1 3 2 7 49 1 100 16 0 1 3 3 0 12 13 1 6 3 0 3 1 6 3 46 100 3.4 Trip Start Times Time of day patterns are a critical input to planning and modeling activities and are analyzed by examining the trip start times in the MARC HTS and in the Replica results. Trip start times vary a great deal across trip purposes. Table 9 presents the percent of trips by trip purpose that start in each hour during the day and are captured in the MARC HTS and Replica datasets. We should note that both datasets capture the general pattern of a greater concentration of daily trips during the AM and PM peaks although there are other differences in the peaking characteristics by trip purpose: • Home-Based Work Trips – Both datasets show similar peaking patterns but the peak periods in the Replica data are less well-defined than in the HTS. About 28 percent and 24 percent of trip making occurs in the AM and PM peak periods (6-9 AM and 3-6 PM) in the Replica dataset compared to about 35 percent and 30 percent of peak period trip making in the HTS dataset. • Home-Based School and College Trips – Both datasets show morning and afternoon peaks in travel, but the peak hours are offset by one hour. The Replica morning and afternoon peaks occur at 8-9 AM and 4-5 PM while the HTS suggests that peaks occur at 7-8 AM and 3-4 PM. The HTS peaks are consistent with patterns that we typically see from household travel survey data. Another feature of the Replica data that needs to be examined is the apparent lack of any trips starting at any other hour of the day outside the morning and afternoon peak periods (7-10 AM and 3-6 PM). These patterns suggest that the school and college trip purposes in Replica may only include traditional school and college patterns that start in the morning, stay for the whole day, and leave in the late afternoon. • Home-Based Other Trips – Here we grouped home-based shopping, social/recreation, and other trips into a single category. A key difference between the trip start time patterns between these two datasets is the morning peak patterns where the HTS captures a higher percentage compared to Replica. This likely reflects the short stops that travelers make on their commute to work which are not captured well by Replica. Short stops can be difficult to capture with cell phone data because stops need to be inferred and cannot be observed directly. On the other hand, the Replica data show higher levels of trip making in the late evening and early morning hours consistent with trends in other cell phone datasets. In addition, respondents to a traditional diary survey tend to underreport trips of this nature which are considered less important and may be omitted more often. 2018 Kansas City Regional Household Travel Survey Final Report 32 • Non-Home Based Trips – The trends in trip start times are generally similar between the two datasets. Like home-based other trips, trip making in the late evening and early morning hours is slightly higher in the Replica data. 2018 Kansas City Regional Household Travel Survey Final Report 33 Table 9. Distribution of Trip Start Times by Trip Purpose 2018 Kansas City Regional Household Travel Survey Final Report 34 MARC HTS Trip Start Hour HB Work Replica HB School & College HB Other NonHome Based HB Work HB School & College HB Other NonHome Based 12:00 AM 0.7 0.0 0.1 0.2 0.7 0.0 0.5 0.1 1:00 AM 0.2 0.0 0.0 0.1 0.0 0.0 0.6 0.3 2:00 AM 0.1 0.0 0.0 0.0 0.0 0.0 0.6 0.4 3:00 AM 0.3 0.0 0.2 0.1 0.6 0.0 0.3 0.3 4:00 AM 0.9 0.0 0.2 0.1 1.3 0.0 0.8 0.6 5:00 AM 4.1 0.1 0.8 0.5 2.7 0.0 1.5 0.9 6:00 AM 10.1 6.4 3.0 1.9 6.2 0.0 2.4 1.4 7:00 AM 16.6 26.9 10.6 6.2 11.1 5.9 4.1 3.2 8:00 AM 8.5 15.7 9.6 5.7 10.6 26.4 5.1 4.5 9:00 AM 3.1 1.2 4.9 4.8 6.3 17.7 5.1 6.0 10:00 AM 2.0 1.0 4.9 6.3 3.8 0.0 5.3 6.5 11:00 AM 2.4 0.7 4.7 8.8 3.3 0.0 5.5 8.0 12:00 PM 3.4 1.0 4.4 9.9 3.5 0.0 5.9 9.4 1:00 PM 2.6 0.8 5.0 7.3 3.7 0.0 6.0 8.7 2:00 PM 3.5 10.2 5.9 8.3 4.3 0.0 6.0 8.2 3:00 PM 7.0 21.5 9.6 9.0 6.4 14.7 6.8 8.5 4:00 PM 11.8 7.8 8.7 8.5 8.1 29.4 7.6 8.2 5:00 PM 11.1 3.9 8.5 8.4 9.4 5.9 8.5 7.8 6:00 PM 4.0 1.4 7.1 5.4 6.2 0.0 8.5 6.3 7:00 PM 2.2 0.4 4.3 4.0 3.5 0.0 7.1 4.7 8:00 PM 1.5 0.4 4.1 2.4 2.5 0.0 5.4 2.9 9:00 PM 1.2 0.2 2.2 1.2 2.0 0.0 3.3 1.7 2018 Kansas City Regional Household Travel Survey Final Report 35 10:00 PM 1.7 0.2 0.8 0.6 1.8 0.0 1.9 0.8 11:00 PM 1.0 0.0 0.4 0.3 2.1 0.0 1.1 0.8 100.0 100.0 100.0 100.0 100.0 100.0 100.0 100.0 Total Note: “HB Other” includes HB Shopping, HB Social/Recreation, and HB Other. 3.5 Trip Mode Analysis Table 10 shows the trip mode shares by trip purpose for the two datasets. There are several notable elements of the results, including the following: • Auto-Driver vs. Auto-Passenger – The auto-driver mode is defined as a trip that uses a private auto where the individual is also the driver while the auto-passenger mode is defined as a trip that uses a private auto where the individual is a passenger. While differences exist between MARC HTS and Replica in the split between these two modes, taken together, the overall auto mode shares are similar across all trip purposes. • Auto-Passenger Mode – The auto-passenger mode shares from Replica seem quite low for several trip purposes, including home-based shopping, social/recreation, and other and other-based trips. In each case, the share of travelers riding as a passenger is 2 percent or less. These mode shares suggest average vehicle occupancies for these trip purposes of less than 1.02 each, which is even lower than we typically observe for home-based work trips which have a typical occupancy of 1.1. This pattern is likely related to the fact that Replica does not model travel made by children except for school travel and children tend to make up a large percentage of auto passengers. However, even if children (less than 18 years old) are removed from MARC HTS summaries for these trip purposes, the HTS shows higher levels of adult auto-passenger trips than Replica, with average vehicle occupancies on the order of 1.15 to 1.20. • School Bus Mode – The school bus mode was coded as “Other” in the MARC HTS summary and as “Auto-Passenger” in the Replica summary. The different convention used accounts for the disparity in mode shares for HB school trips in the two datasets. • Public Transit Mode – The trends in public transit mode shares differ between the HTS and Replica. Transit mode shares computed from the MARC survey are all higher than those obtained from Replica, and for many trip purposes, these shares are higher by factors of two or more. Moreover, the MARC survey suggests that the trip purposes with the highest transit mode shares are HB School and HB College (10.0 and 2.6 percent, respectively), while Replica suggests that WB Other and OB Other trip purposes (0.6 and 0.8 percent) have the highest transit mode shares. The finding from Replica is at odds with our typical understanding of non-home based trips, which tend to be shorter and are part of more complex tours with more than one stop where transit is less competitive. The survey is more reasonable in this respect as non-home based transit mode shares are lower than most other trip purposes. Note that the HTS transit trip samples is about 450 trip records. • Walking Mode – The HTS shows relatively low walking mode shares for HB Work trips and the highest walking mode shares for HB School, HB College, and HB Social/Recreation trips. While Replica also depicts low walking mode shares for HB Work trips, walking mode shares for HB School and HB College trips are lower than other trips. This is unexpected as school and college trips tend to include many short trips in walkable environments which make walking more attractive, even for children going to school. Replica indicates the highest walking mode shares for WB Other and OB Other trips. 2018 Kansas City Regional Household Travel Survey Final Report 36 • Biking Mode – The results for biking mode are reasonable and comparable although we should note that the sample size of the bike mode in the HTS is small. • On-Demand Mode – The HTS on-demand trip sample size is only 51 trips. On-demand mode shares from Replica are relatively low for work, school, and college trips reflecting a reasonable result. Additional data would be needed to draw firmer conclusions. 2018 Kansas City Regional Household Travel Survey Final Report 37 Table 10. Trip Mode Percent by Trip Purpose MARC HTS Auto-Driver OB Other WB Other HB Other HB Soc/Rec HB College HB School HB Work Trip Mode HB Shopping Trip Purpose 90.0 10.9 77.2 74.0 56.0 70.5 84.8 68.4 Auto-Passenger 5.0 40.7 6.5 18.2 26.7 21.0 6.7 23.9 Transit 1.5 10.0 2.6 1.6 0.7 1.7 0.8 1.3 Walking 2.2 9.2 13.6 5.8 14.8 5.2 6.7 4.0 Biking 0.6 0.9 0.0 0.1 0.9 0.2 0.1 0.2 On-Demand Auto 0.4 0.3 0.0 0.1 0.2 0.4 0.0 0.1 Other 0.3 28.0 0.0 0.1 0.8 1.0 1.0 2.1 Replica Auto-Driver OB Other WB Other HB Other HB Soc/Rec HB College HB School HB Work Trip Mode HB Shopping Trip Purpose 89.2 17.5 87.9 88.6 90.7 88.7 86.5 83.1 Auto-Passenger 7.6 76.2 7.1 1.5 1.3 1.4 2.0 0.2 Transit 0.5 0.2 0.4 0.5 0.4 0.5 0.6 0.8 Walking 0.8 3.0 1.4 7.9 5.9 7.9 9.2 13.0 Biking 0.2 2.0 0.3 0.4 0.3 0.4 0.3 0.4 On-Demand Auto 0.0 0.1 0.1 0.4 0.4 0.4 0.4 0.5 Other 1.7 1.2 2.7 0.7 1.0 0.6 1.0 2.1 2018 Kansas City Regional Household Travel Survey Final Report 38 3.5.1 Average Trip Lengths by Mode Table 11 shows average trip lengths by trip mode. Several key observations can be made: • Replica’s estimate of average trip length for the auto-driver mode is slightly higher than that of the HTS (by 14 percent), but the difference is lower than average difference in trip lengths discussed earlier (of 25 percent). • Replica’s estimate of average auto-passenger trip lengths is significantly lower than that of the HTS (by 43 percent). Because auto-passenger trips are made up by a large share of children, and Replica does not capture travel by children outside of school travel, it stands to reason that this mode may be largely dominated by school trips in Replica, which tend to be shorter on average. Note that in both datasets auto-passenger trip lengths are shorter than auto-driver trip lengths, which is reasonable. • Public transit trips are significantly shorter as estimated by Replica than the HTS (by 35 percent). This trip mode makes up a smaller share of trips in Replica than the HTS. • Replica’s estimate of average walking trip lengths is significantly longer than the HTS (over 100 percent greater). The NHTS for the entire nation suggests average walking trip lengths of about 0.9 miles, which is between the two estimates. • Replica also estimates a higher average trip length for trips made by bike mode (by 52 percent), though the HTS sample of such trips is quite small, which makes the HTS estimate less reliable. The national NHTS sample indicates average biking trip lengths of about 2.2 miles, which is between Replica and MARC HTS. • The difference in On-demand Auto average trip lengths between HTS and Replica (of 21 percent) is approximately the same as the average difference in trip lengths overall between the two datasets (of 25 percent). Table 11. Average Trip Lengths by Mode Trip Mode MARC HTS Replica Percentage Difference Auto-Driver 8.4 9.6 14% Auto-Passenger 7.0 4.0 -43% Transit 7.3 4.7 -35% Walking 0.4 1.0 129% Biking 1.9 2.9 52% On-Demand Auto 7.4 8.9 21% 39.1 6.6 -83% Other 3.5.2 Mode Shares by Age Group 2018 Kansas City Regional Household Travel Survey Final Report 39 Table 12 shows the HTS and Replica mode shares by age category. As noted above, the auto-passenger mode share is quite low for the majority of age groups in Replica with the exception of children, reflecting the school travel that is modeled for these travelers in Replica. Replica shows a trend of decreasing auto-passenger mode shares as age category increases, while the HTS suggest auto-passenger mode share decreases to the 40-49 year age group and then increases with increasing age thereafter. There are also differences in estimates of transit and walking mode shares across age category. In particular, while the HTS suggests the highest levels of transit usage by individuals under 30 years (and 90 years or more), Replica indicates transit usage is lowest for younger individuals. Replica also indicates very low levels of walking mode usage for children under 10 years, whereas the HTS shows much higher levels of walking for these children. 2018 Kansas City Regional Household Travel Survey Final Report 40 Table 12. Replica Mode Percent by Person Age Category 2018 Kansas City Regional Household Travel Survey Final Report 41 MARC HTS Trip Mode Age Category AutoDriver AutoPassenger Transit Walking Biking OnDemand Auto Other 00-09 yo 0.31 74.75 4.36 7.79 1.01 0.23 11.54 10-19 yo 22.69 52.52 5.24 5.83 0.50 0.13 13.09 20-29 yo 75.79 11.60 1.79 9.62 0.56 0.41 0.22 30-39 yo 81.69 8.87 0.88 7.11 0.42 0.35 0.68 40-49 yo 85.46 7.99 0.78 4.80 0.38 0.08 0.53 50-59 yo 82.02 9.37 1.75 5.37 0.32 0.30 0.86 60-69 yo 79.58 12.40 1.02 6.06 0.44 0.06 0.44 70-79 yo 74.59 18.20 0.90 5.86 0.09 0.03 0.32 80-89 yo 72.69 23.09 0.00 3.43 0.00 0.00 0.79 90 yo/+ 43.75 45.83 4.17 6.25 0.00 0.00 0.00 Total 72.02 17.58 1.59 6.22 0.40 0.18 2.00 Replica Trip Mode Age Category AutoDriver AutoPassenger Transit Walking Biking OnDemand Auto Other 00-09 yo 0.35 98.87 0.01 0.27 0.15 0.00 0.34 10-19 yo 45.07 44.63 0.28 5.51 2.52 0.19 1.79 20-29 yo 85.15 4.60 0.49 7.48 0.43 0.26 1.59 30-39 yo 86.16 3.64 0.48 7.80 0.38 0.30 1.23 40-49 yo 87.40 2.95 0.51 7.31 0.27 0.32 1.25 2018 Kansas City Regional Household Travel Survey Final Report 42 50-59 yo 87.82 2.50 0.69 7.13 0.23 0.34 1.30 60-69 yo 89.44 1.49 0.66 6.61 0.23 0.38 1.20 70-79 yo 91.59 0.43 0.54 5.87 0.22 0.47 0.87 80-89 yo 92.69 0.73 0.62 4.72 0.20 0.41 0.64 90 yo/+ 94.84 0.31 0.69 3.27 0.21 0.36 0.33 Total 80.18 10.63 0.52 6.64 0.48 0.31 1.25 3.5.3 Mode Shares for Airport Access The airport trips were analyzed separately given the different mode use patterns to this special generator in the region. Figure 14 shows the mode shares for trips originating at the airport. Several observations can be made: • Replica data indicates a much lower share of auto-passenger mode share (6 percent) compared to the HTS share of 16 percent. These mode shares indicate an average vehicle occupancy for trips from the airport of 1.3 in the HTS sample compared to an occupancy of 1.1 in the Replica data. Given that leisure trips make up a significant portion of airport trips, the auto-passenger share from Replica seems very low. However, it is worth noting that Replica models the “typical day” of local residents and does not include visitors. As such, the majority of travel to/from the airport modeled by Replica probably includes only airport workers. The private auto mode shares seem more reasonable in this context. • Replica also estimates that 5 percent of trips made from the airport are by walking or biking, which is almost certainly high given the difficulty of accessing the airport by nonmotorized modes. • On the other hand, Replica’s estimate of 2 percent mode share for transit is probably reasonable and is consistent with the levels of transit observed in the HTS. • Last, the on-demand mode share of 2 percent in the HTS is nearly 10 times larger than the 0.2 percent mode share estimated by Replica. Replica’s estimate seems low, especially given that Replica’s regionwide estimate of on-demand mode share is higher than that. However, this could also be related to Replica modeling the “typical day”, which means that most airport travel in the Replica data is for work purposes. 2018 Kansas City Regional Household Travel Survey Final Report 43 Figure 14. Trip Mode Shares for Trips Originating at Airport Other On-Demand Auto Mode Biking Walking Transit Auto-Passenger Auto-Driver 0 10 20 30 40 50 60 70 80 90 Mode Share MARC HTS Replica 3.6 Transit Trips In total, the weighted HTS shows a total of 130,000 transit trips regionwide compared to Replica which indicates that there are about 32,000 transit trips. We previously developed transit trip targets for calibration and validation of the base year travel demand model for 2015, which were scaled to reproduce total transit ridership in line with observed transit boardings in the region. Those targets indicated a regionwide transit trip total of 46,000 trips without the streetcar which was not yet operational in 2015 and was not included in this estimate. The HTS and Replica both miss the mark in this respect, though it is worth pointing at that transit ridership may have changed in the four years since 2015, especially with the opening of the Streetcar service in the region. Tables 13, 14, and 15 show the breakdown of the total share of trips in each dataset made by transit. Table 13 shows the breakdown of transit trips by household income category. The two datasets show similar shares of transit trips by income breakdown. Table 14 shows transit trips by the number of household vehicles. Replica estimates a larger number of transit trips being made by households with zero vehicle ownership, while the HTS shows more transit trips being made by households with 1 or more vehicles, especially those with exactly 2 vehicles. Table 15 shows transit trips by race/ethnicity and suggests similar patterns between the two sources of travel data. 2018 Kansas City Regional Household Travel Survey Final Report 44 Table 13. Percent of Total Transit Trips by Household Income Category Household Income ($) MARC HTS Replica 0K-10K 20.2 19.7 10K-20K 16.3 20.6 20K-30K 13.6 12.1 30K-40K 7.7 9.7 40K-50K 4.3 7.1 50K-60K 2.8 4.7 60K-100K 18.0 14.0 100K-150K 12.9 7.6 150K-200K 6.1 2.5 200K+ 3.7 2.0 Unknown 0.0 0.0 Table 14. Percent of Total Transit Trips by Household Number of Vehicles Vehicles MARC HTS Replica 0 34.8 53.9 1 21.2 17.8 2 30.1 16.0 3/+ 13.9 10.3 Unknown 0.0 2.0 Table 15. Percent of Total Transit Trips by Person Race/Ethnicity Person Race/Ethnicity MARC HTS Replica White Non-Hispanic 55.0 39.8 Black Non-Hispanic 29.7 47.8 2018 Kansas City Regional Household Travel Survey Final Report 45 Hispanic Any Race 8.1 7.7 Asian Non-Hispanic 1.9 1.9 Multiracial Non-Hispanic 4.4 1.8 American Indian and Alaska Native NonHispanic 0.0 0.7 Native Hawaiian and Pacific Islander NonHispanic 0.0 0.3 Other Non-Hispanic 0.0 0.1 Unknown 0.9 0.0 Figure 15 shows the distribution of transit trips by the trip origin district. The two datasets show several differences along this comparison. Replica shows a much larger share of transit trips beginning in Districts 9 and 14, which represent Kansas City’s downtown area and the area to the southeast of downtown (36 percent and 48 percent in Replica vs. 23 percent and 29 percent in HTS). However, Replica indicates a much smaller share of transit trips beginning in several districts, including District 12 representing eastern Jackson County (3 percent in Replica vs. 15 percent in HTS) and District 16 in Kansas City, KS (5 percent vs. 9 percent in HTS). Transit boarding data could be useful in identifying the dataset with more reasonable estimates. Figure 15. Percent of Total Transit Trips by Trip Origin District 60.0 Transit Trips % 50.0 40.0 30.0 20.0 10.0 0.0 Origin District MARC HTS Replica In contrast to Figure 15, Figure 16 shows the percent of total transit trips by the home location district of the trip maker. We expect Figures 15 and 16 to show similar trends, since most trips (especially transit trips) are homebased trips. This pattern is generally what we find in both datasets, with the exception of District 9 (Kansas City downtown) where many of the trip origins are trips from a workplace back home. However, MARC HTS indicates a smaller difference between trip origins and home locations (23 percent vs. 16 percent) for this district than does Replica (36 percent vs. 16 percent). Replica also shows several districts with zero trip origins but non-zero trips by residents of the district (including Districts 1, 2, 8, 11, and 13). 2018 Kansas City Regional Household Travel Survey Final Report 46 Figure 16. Percent of Total Transit Trips by Home Location District 60.0 Transit Trips % 50.0 40.0 30.0 20.0 10.0 0.0 Home Location District MARC HTS Replica Table 16 shows access and egress distance distributions of transit trips in both datasets. The HTS shows much lower average access and egress distances, with 83 and 89 percent of transit trips have access and egress distances less than a half mile, compared to Replica which shows that 61 and 72 percent of transit trips have access and egress distances of less than half a mile. The HTS results are more consistent with our professional experience. Table 16. Percent of Transit Trips by Access and Egress Distances Access Distance Egress Distance Distance (mi) MARC HTS Replica MARC HTS Replica 0 - 0.25 57.8 28.8 71.9 42.3 0.25 - 0.50 25.3 32.6 17.0 30.1 0.50 - 0.75 7.3 18.5 3.6 13.4 0.75 - 1.00 4.0 9.5 2.2 6.9 1.00 - 1.25 3.3 6.2 1.5 4.7 1.25 - 1.50 0.6 4.3 1.3 2.6 1.50 - 1.75 0.3 0.0 0.1 0.0 1.75 - 2.00 0.7 0.0 0.0 0.0 2 or more 0.7 0.0 2.4 0.0 2018 Kansas City Regional Household Travel Survey Final Report 47 3.7 On-Demand Trips The on-demand mode represents a relatively new mode that has rapidly increased in usage over the past several years and very little data have been collected to understand it. Even with the increase in usage, the regionwide mode share in Kansas City is less than 1 percent, according to the HTS7. Figure 17 and Figure 18 compare ondemand trip data from the MARC HTS, Replica, and NHTS (for all of the U.S.). Figure 17 shows the percent of total on-demand trips by household income. The three datasets show some important differences, but the trends found in Replica are similar to the NHTS for the nation, with the exception that Replica estimates very few on-demand trips for the lowest income category. Overall, these results look reasonable. Figure 17. Percent of Total On-Demand Trips by Household Income 70 On-Demand Trips % 60 50 40 30 20 10 0 0K-10K 10K-50K 50K-100K 100K-150K 150K-200K 200K+ Unknown Household Income MARC HTS Replica NHTS (all of U.S.) Figure 18 shows the percent of total on-demand trips by vehicle ownership at the household level. In this case, the trends found in Replica contradict those found in both the MARC HTS and the NHTS. Replica estimates almost no on-demand trips for zero-vehicle households and suggests that about 80 percent of on-demand trips are made by households with 2 or more vehicles. In contrast, both the MARC HTS and NHTS indicate many more on-demand trips are made by zero-vehicle households (between 25 and 33 percent) with fewer on-demand trips made by households with 2 or more vehicles (25 to 37 percent). Given that zero-vehicle households have fewer mode options, the survey data seem reasonable. 7 NHTS data suggests that nationwide, the TNC mode share is also less than 1 percent. 2018 Kansas City Regional Household Travel Survey Final Report 48 Figure 18. Percent of Total On-Demand Trips by Household Number of Vehicles 60.0 On-Demand Trips % 50.0 40.0 30.0 20.0 10.0 0.0 0 1 2 3/+ Unknown Vehicle Ownership MARC HTS Replica NHTS (all of U.S.) Table 17 shows the percent of on-demand trips by trip origin and destination district for the MARC HTS and Replica. Two areas where we expect a fairly robust number of on-demand trips are those to/from the airport and to/from downtown. While the sample size of on-demand trips for the HTS is small (only about 50 trips), the HTS indicates that about 2 percent of on-demand trips have an origin or destination at the airport (District 5) and about 20 percent have an origin or destination near downtown (District 9). Replica, on the other hand, suggests a much lower number of on-demand trips at the airport (about 0.1 percent) and near downtown (about 5 to 6 percent). Although the airport estimate may be justified given the nature of the Replica sample, the downtown estimate of TNC usage seems very low given the concentration of traffic. 2018 Kansas City Regional Household Travel Survey Final Report 49 Table 17. Percent of Total On-Demand Trips by Trip Origin and Destination District Trip Origin District District MARC HTS 1 0.0 0.9 0.0 0.9 2 0.0 3.5 0.0 3.3 3 1.8 4.0 1.0 4.1 4 6.7 7.8 7.6 7.8 5 2.4 0.1 1.7 0.1 6 0.6 3.8 0.0 3.8 7 7.1 14.1 7.0 14.1 8 0.6 3.6 0.6 3.5 9 19.7 5.4 21.6 5.8 10 1.2 8.1 0.0 8.1 11 0.0 1.3 0.0 1.3 12 13.6 15.0 10.0 14.3 13 0.0 4.1 3.7 3.9 14 16.9 14.4 16.4 14.4 15 28.9 9.0 29.9 9.6 16 0.6 3.1 0.6 3.2 0.0 1.8 0.0 1.8 Buffer Region Replica Trip Destination District 2018 Kansas City Regional Household Travel Survey Final Report MARC HTS Replica 50 4.0 Conclusions and Recommendations The MARC HTS provides a snapshot of the travel made by residents of the eight-county Kansas City region. The dataset includes a vast array of information about each traveler and trip, including demographics, contextual travel information (like purpose and party size), and key trip attributes (like origins and destinations, travel mode, and time of day). This information is observed directly based upon the stated responses of survey respondents. While expansion factors were developed for each household in the survey to represent the full set of households in the region, the survey dataset that was collected remains a sample. Because it is a sample, the survey data will not necessarily accurately represent the incidence of travel patterns that are localized or occur infrequently, even when applying the expansion factors. Therefore, the survey is best utilized to analyze key travel trends and differences among large groups of travelers. Cutting the data too thin, as can be the case in analyzing origin-to-destination travel flows, can result in misleading inferences. On the other hand, Replica provides a synthetic version of all residents in the Kansas City region and simulates the travel by these residents on an average day.8 From this perspective, Replica can be viewed as a synthetic representation of all travel by residents in the Kansas City region. The synthetic population represents all households in the region and travel patterns are generated for each household. Like the HTS, Replica provides a snapshot of travel in the region at a point in time. Unlike the HTS, the underlying data used to generate trips by the synthetic population come from a mobile phone dataset, which while still a sample of trips in the region, is a very large sample. These data can be segmented in a variety of multi-dimensional ways while still providing robust estimates of travel patterns although it should be noted that it is possible to also cut these data too thin. Unlike the HTS, these data do not provide direct observations of trips made nor do they provide the context of each trip. Because the mobile phone data, from which the travel patterns in Replica are drawn, are a passively collected dataset, all trip information must be inferred from the mobile phone data including trip origins and destinations, travel mode, and trip purpose. It is important to remember that there could be systematic differences between the inference process and the ways in which travelers respond to surveys. This section summarizes our findings from the comparative analyses we conducted of Replica to the MARC HTS and other datasets. It also expands on how Replica data can be used in other analyses and what this source offers beyond the MARC HTS. 4.1 Summary Results can be summarized along two key areas: household formation and travel pattern analyses. In terms of household formation, the total number of regionwide households and total population are slightly different in Replica’s synthetic population, which has 5 percent more households and 2 percent fewer population compared to the MARC HTS or the ACS. Trip rates are slightly lower than the HTS resulting in a total regionwide trip estimate that is also slightly lower than the HTS. In our experience working with cell phone datasets, we have found them to consistently generate slightly lower trip rates than are seen in surveys, which is probably because trips must be inferred from the cell phone data rather than observed directly by a survey. Moreover, the distribution of households along key household market segments in the Replica model are quite similar to the distribution of households along the same market segments in the MARC HTS and ACS datasets. 8 Note that Replica also simulates visitor travel and truck travel, but these elements of the tool were not analyzed as part of this study. 2018 Kansas City Regional Household Travel Survey Final Report 51 The travel patterns we summarized were generally less consistent with one another compared to household formation characteristics and some of the travel patterns we identified in Replica were unexpected. We examined several key areas of travel patterns: • Trip Rates. Replica’s trip rates varied in expected ways across key demographic variables like household size and household income. Trip rates for children, however, were found low because Replica simulates only school travel for children. This could explain the slightly lower overall levels of trip-making in Replica compared to the HTS. Trip rates by trip purpose were also found to be different between Replica and the HTS. This might be explained by the difference in data collection: the HTS observes directly the respondent’s stated trip purpose while Replica must infer trip purpose from mobile phone data. Systematic differences may exist and understanding that these differences exist is important when working with Replica data. Nonetheless, the split of home-based vs. non-home based trip making in Replica was consistent with the HTS. This makes sense as the home location is the easiest location to infer from mobile phone data. • Trip Lengths. Replica’s trip lengths were found to be significantly higher, on average, than trip lengths in the HTS (8.1 miles vs. 6.5 miles). There are a couple of potential reasons that mobile phone data may display higher trip lengths: o This can occur because it is difficult to properly identify short stops using mobile phone data. When processing mobile phone data, false detection of stops that did not actually occur (e.g., stopped in traffic) must be balanced against not detecting short duration stops. Short duration stops often occur on the way to another location (e.g., dropping kids off at school or stopping for coffee on the way to work). As such, missing such stops often means creating longer trips with no stops although certain other metrics like VMT may be unaffected. o Very short distance trips can also be hard to detect based on mobile phone data. This is often due to spatial clustering of pings in the mobile phone data which groups pings that are spatially (and temporally) close to one another. If two actual activity locations are separated by a small enough distance, spatial clustering can treat them as the same activity location. • Trip Start Times. The overall trip start time patterns in Replica were reasonable but a few observations were made about trip start times. First, Replica seems to display less peaking of trip start times compared to the HTS, especially for home-based work trips which may reflect differences in the definition of trip purposes. Second, school and college trip start time distributions seemed to mimic the HTS except that they were offset by about one hour. Last, Replica’s home-based other trips were found to have only a single peak later in the day, whereas HTS also showed peaking in the morning. This disparity may be due to differences in short duration stop making, which is prevalent in the morning peak. • Trip Mode. Overall, the mode splits between auto, transit, and non-motorized travel were fairly reasonable. However, Replica diverged from our expectations along a couple of key areas. Because we do not know how mode is inferred in the Replica model, it is difficult to state definitively why these differences exist. o First, the split between auto-driver and auto-passenger did not seem reasonable for non-work trip purposes. These splits suggested average vehicle occupancies on the order of 1.1 or less for non-work purposes, whereas we expect vehicle occupancies in the range of 1.3 to 1.5 for such trip purposes. While the fact that non-school travel by children is not modeled in Replica and children make up a large percentage of auto-passenger trips explains some of this difference, we found that it did not explain all of the difference. o Second, the walking mode share for school and college trips was much lower than expected. While we usually expect school and college trips (compared to other trip purposes) to have among 2018 Kansas City Regional Household Travel Survey Final Report 52 the highest non-motorized mode shares, Replica’s walking mode share was much lower for school and college trips than for non-work and non-school and non-college trips. o Last, the walking and biking mode shares for trips originating at the airport were higher than seemed reasonable given the rarity of walking and biking to the airport. • Trip Flows. Trip flows are a bit more difficult to study because the HTS sample sizes start becoming rather small when cutting the data along multiple spatial dimensions. Nonetheless, one key difference between the HTS and Replica was the incidence of intra-district trip-making, where Replica estimated a lower percentage of such trips. As intra-district trips have shorter trip lengths generally, this result is consistent with the overall higher trip lengths observed in Replica. • Transit and On-demand Trips. Replica’s distribution of transit and on-demand trips across market segments was generally not consistent with our expectations that were reflected in the survey data. It is worth noting again the sample survey sample sizes for these modes (about 450 trips for transit and 50 for on-demand). o Total transit trip-making estimated by Replica was significantly lower than the regional model (by about 30 percent) and the HTS (by about 75 percent). o Replica’s average access and egress distances were higher than expected. o Replica’s geographic distribution of transit trips by district was inconsistent with the HTS. Due to the low HTS sample sizes of transit trips, this presents a good candidate to check against transit ridership data. o Replica estimated a large share of on-demand trips being made by households with two or more vehicles, which is in contrast to the MARC HTS and nationwide data found in the NHTS. o Replica estimated a smaller than expected share of on-demand trips that begin in the downtown area of Kansas City, where higher land use densities probably make these trips more popular. 4.2 Using Replica Data for Analyses There are a number of ways to effectively utilize Replica data that can provide value over and above the analyses that can be done with the HTS. The following uses of the data are not intended to be a comprehensive list of potential uses of the data, but • Model Estimation. While the Replica model links travel patterns from mobile phone data to synthetic households/persons and includes most of the required contextual travel information that is necessary for model estimation, we do not recommend that travel demand model components be estimated using these data. Based upon our analysis of the dataset, it is not clear whether the heuristic rules used to infer contextual travel information are robust enough to generate the inter-relationships we expect between key variables in the data (e.g., socioeconomics of transit travelers and trip mode designations). The HTS is better suited for estimating nuanced behavioral models. • Model Calibration/Validation. The Replica dataset is more appropriate for calibration and validation purposes. This may be especially true for trip distribution and time of day elements of the model, where the travel pattern data on which Replica is based are more robust than the HTS. Before using Replica in this way, it would be important to verify that the trip distribution patterns in Replica are not distorted by Replica’s modeling process, at least at the geographic resolution that will be used in calibration and validation. Some allowances and/or adjustments may be needed to account for the 2018 Kansas City Regional Household Travel Survey Final Report 53 discrepancies in average trip lengths between Replica and the HTS in calibrating trip distribution models. These adjustments can be done possibly by reweighting Replica trips. • Model Updates. If Replica data can be updated in the future (e.g., 3 to 5 years), they can be used to provide a lower-cost option for improving our understanding of changes in the region’s travel and for updating the regional model. While this is somewhat hypothetical at this point, measuring change between the existing Replica dataset and a hypothetical Replica dataset in the future would offer the opportunity to update the travel patterns in the model more frequently. Types of measurements that could be made include: o Trip rates by geography; o Temporal distribution of trips; o Trip lengths; and o Mix and shares of trip purposes. • External Travel. Because Replica estimates the travel generated to/from Buffer counties, including travel by residents of those counties, it can provide a good estimate of travel at external stations in the model. This includes estimates of the geographic distribution of trips inside the region to/from each external station and temporal distributions of travel. • Corridor and Subarea Analyses. Replica data can support corridor-level studies in cases of welldefined corridors or parts of an urban area. The data can be used to develop origin-to-destination trip tables that might be adjusted using matrix adjustment methods to better match traffic counts. This approach may be preferred to generating subarea models. These types of analyses cannot be done with the MARC HTS due to the sample size limitations of the survey. • Seasonality and Day of Week. Analysis of the travel patterns estimated by Replica by season and by day of week could provide useful information for certain types of applications. Measuring differences in these travel patterns for different seasons or across days of week would provide a benchmark from which to estimate differences in key metrics like VMT and VHT. 2018 Kansas City Regional Household Travel Survey Final Report 54