Below is the uncorrected machine-read text of this chapter, intended to provide our own search engines and external engines with highly rich, chapter-representative searchable text of each book. Because it is UNCORRECTED material, please consider the following text as a useful but insufficient proxy for the authoritative book pages.
30 The research team identified two types of confirmation as evidence of the approachâs ability to predict urban infill trip generation and to demonstrate its validity to the transporta- tion profession: 1. Verification â A process that focuses on ensuring that the proposed methodology was correctly developed (e.g., the process/equations were correctly translated, the expected variables cancel) and that there are no gross errors or over- sights in the theory, the translation of the theory into a procedure, or the implementation of the procedure. The research team is confident that this form of estimating infill trip generation is easily verified, although aspects of the proposed method may require further research. 2. Validation â The act of demonstrating, at a reasonable level of confidence, that the methodologiesâ predictions are able to repeatedly match empirical dataâin this case, vehicle traffic generation of infill development. Validation requires a substantial amount of empirical data, represent- ing a wide range of contexts, to show statistical significance. Ample empirical data help to smooth out the peaks and valleys typically found in small datasets, as well as help to isolate anomalies and outliers in the data. One of the challenges for this research study was a lack of resources to collect data in sufficient amounts to validate the method- ology. Validation of the proposed methodology will take time as members of the transportation profession con- tribute data from their own research or as a result of their work on development projects. Validation procedures take into consideration the following: ⢠Selecting and surveying sites for validation. This step involves the selection of one or more existing develop- ments for which the estimation method will be tested. For validation, data need to be collected at a minimum of five sites. Data for the selected validation sites will have either already been collected or will be collected once selected. While difficult to find in more urban contexts (urban cen- ter and urban core), the ideal site for validation has self- contained and exclusive parking for users of the site and is designed so that all of the traffic generated by the site can be counted automatically. If this is not possible, manual data collection is sufficient. It is essential to obtain accurate data regarding the siteâs occupied units of development (e.g., dwelling units, gross floor area, and gross leasable floor area) and other relevant information representing the time the count data are collected. The person planning and implementing the data collection should be familiar with ITEâs procedures for conducting local trip generation studies (4, 14). ⢠Assemble data needed to validate method. The mini- mum data to validate a site are (a) the number of vehicle trips generated by the site during a 2- or 3-hour period typically encompassing the peak hour (15), (b) the devel- opment units representing the independent variable used to estimate vehicle trips and the vacancy rate of develop- ment units, (c) the land use and transportation character- istics used to describe context in which the site is located, and (d) the data needed to apply the methods outlined in Chapter 4 (baseline and infill mode share and vehicle occupancy). ⢠Estimate infill vehicular trip generation. Use the desired method from Chapter 4 (proxy site method or household travel survey method) to predict the siteâs vehicular trip generation. ⢠Evaluate the methodâs performance. Compare predicted to actual trip generation data for each validation sample or compare to the average of the data collected from mul- tiple validation sites, apply applicable statistical tests to assess accuracy of validation results, assess the validity of the method for the subject under study, and identify any needs for adjustments or additional data in the proposed method. C H A P T E R 5 Confirming the Proposed Approach for Estimating Infill Trip Generation
31 5.1 Selecting a Method for Verification Any of the context classifications presented in this report, if also meeting transit proximity criteria, can have develop- ment that qualifies as infill. Therefore, the approach and the methods of applying the approach presented in this report needed to be applicable across a spectrum of contexts. Testing and verification of the approach focused on (a) the house- hold travel survey method of deriving adjustment factors, and (b) contexts that span the GU/UC classifications, for the following reasons: ⢠Extracting the adjustment factors from HTS data in the household travel survey method is the most complex of the methods, and the verification process was an opportunity for the research team to derive adjustment factors from a second source of HTS linked-trip data. ⢠GU/UC contexts (commonly called midtown or down- town fringe) make up a large portion of urbanized met- ropolitan areasâareas significantly larger than urban core contextsâand, therefore, GU/UC contexts are applicable to a greater number of potential validation sites. ⢠GU/UC areas represent the middle of the range of infill area types (from urban core to suburban center), but the crite- ria for identifying GU/UC areas spans a broad spectrum, eliminating the potential hindrance of a limited source of validation sites using a single narrowly defined context. ⢠Given the limited resources for the verification step, the research team selected to maximize the number of sites in a combined context zone rather than spread the limited resources over all of the context zones. Although the testing of the household travel survey method was limited to general urban and urban-center contexts, the method is applicable to the extremes ranging from suburban center to urban core. The research team anticipates that use by the transportation profession of the methods presented in this report will help identify ways of improving the methods and produce data for future verification and validation of the methods for the full range of contexts. 5.2 Verifying the Travel Survey Method: A Case Study from Metropolitan Washington, D.C. The research team evaluated five metropolitan areas that had current HTSs for use in verifying the travel survey method. The team selected the Washington, D.C., metropolitan area, whose metropolitan planning organization (MPO)âthe Metropoli- tan Washington Council of Governments (MWCOG)âis the source for the HTS data. The Washington, D.C., HTS surveyed 11,000 households and contains a database of approximately 88,000 linked trips. A principal reason for selecting the Washington, D.C., HTS was the recency of its survey, which was completed in 2008. This is an important consideration given that the veri- fication process relies on 2011 traffic count data (counted in conjunction with this research study). Survey data substan- tially older than the counts would make it difficult to reconcile differences between the predicted and actual trip generation of the sites. 5.3 Application of the Method and Results of the Verification Data were collected at a limited number of sites for test- ing the reasonableness of the methodâs results, for verifying the methodâs procedures and computations, and to serve as a catalyst for continued data collection for future validation. Verifying the method included a reasonableness review of the procedures and results of the method based on sound engineering practice and the experience of the research team. Although this form of verification does not yield a definitive answer, the results support the research teamâs confidence that the procedures and data used in the method will pro- duce consistent, logical, and reasonably accurate results that professional peers and users of the method will find credible. 5.3.1 Expected Results of the Case Study Analysis The research team reviewed much of the literature on the travel characteristics of urban infill development and has itself conducted focused research on the theorem that development in urban contexts generate less traffic than the same develop- ment in suburban contexts. While the principal investigator and the members of the research team, in their professional judgment and their personal opinions, have confidence that the theorem is correct, they seek empirical data that can be linked statistically to their model and require validation through consistent and reproducible predictions. Following are the research teamâs two most prominent theoretical expec- tations from the case study: ⢠Land uses in urban contexts qualifying as infill, in proxim- ity to rail or high-frequency bus transit, demonstrate mea- surably lower vehicular trip generation than an equal type and size of development in suburban contexts or urban contexts that do not qualify as infill. ⢠Trip estimates derived from land uses in suburban contexts will be consistent with the land usesâ trip generation using baseline rates or regression equations published in the ITE Trip Generation Manual.
32 Case study sites were identified using the guidelines pre- sented in Chapter 4, and data were collected consistent with the procedures for deriving the adjustment factors using the minimum data collection variant. Data collected at the case study sites included: ⢠Vehicle counts at driveways of parking facilities exclusive to the site, ⢠Vehicle occupancy, ⢠Person trips entering and exiting the siteâs building, ⢠Observation of mode of access, and ⢠General observation of site conditions and surrounding context. With empirical data available, the research team was able to compare predicted and surveyed results of the household travel survey method. A secondary objective of the data col- lection was to refine the data collection protocol for the proxy site method. 5.3.2 Summary of Findings The following sections contain brief overviews of the results of applying the household travel survey method to the four land use categories used to develop the example adjustment factors from the HTS data presented in Chapter 4. 5.4 Derived Adjustment Factors Table 5.1 presents the methodology-derived adjustment factors (mode share and vehicle occupancy) for the GU/UC context zones by land use category and proximity to transit. The research team reviewed these findings for reasonable- ness. The MWCOG has not published a report summariz- ing the findings of their HTS, so the research team could not compare its findings on mode share and vehicle occupancy with mode-share cross-references to land use, trip purpose, or context prepared by MWCOG. 5.4.1 Residential Land Use Category The results in Table 5.2 show that the method results in sub- stantially higher peak hour trip generation at the three resi- dential infill case study sites when compared to the actual trips. The results range from a factor of two to as high as nearly three and a half times the actual trips. The research team expected that the method would overpredict or underpredict, but did not expect the large differences shown in Table 5.2. The three residential test sites generate low volumes of traf- fic, so the percentage difference between the predicted and actual trips can be misleadingly large. For example, the Colum- bia Uptown residential test site was determined to generate 13 vehicle trips in the a.m. peak hour, while the method pre- dicts the a.m. peak hour to be 25 trips. The absolute differ- ence of 12 trips remains a small number, but the percentage difference of 92% appears large. The research team considered that magnitude of the dif- ference between predicted and actual vehicle trips might be an anomaly or magnification of error related to the small number of actual trips. But because all of the residential sites had low actual vehicle trips, the research team was unable to confirm a magnification of error. When compared to trips estimated using ITE trip gen- eration rates, the method predicts about one-third to one- half fewer trips at all three study sites, as the research team expected. The difference between the predicted and ITE trip Table 5.1. Mode share and vehicle occupancy adjustment factors for Washington, D.C. Infill Adjustment Factors for GU/UC Contexts Within Walking Distance Of: High- Frequency Bus Stop Rail Station a.m. p.m. a.m. p.m. Residential Case Study Sites (ITE LUC 220) Transit 27.3% 24.0% 32.5% 27.7% Walk/bicycle 11.3% 13.4% 12.9% 15.8% Vehicle occupancy 1.27 1.32 1.30 1.34 General Office Case Study Sites (ITE LUC 710) Transit 33.4% 31.0% 38.8% 35.6% Walk/bicycle 9.8% 10.4% 11.9% 12.5% Vehicle occupancy 1.13 1.16 1.15 1.17 Retail/Shopping Center Case Study Sites (ITE LUC 820) Transit 15.4% 13.5% 19.7% 16.5% Walk/bicycle 29.6% 19.0% 35.4% 22.8% Vehicle occupancy 1.20 1.36 1.16 1.36 Restaurant Case Study Sites (ITE LUC 932) Transit 10.4% 13.8% 12.2% 16.1% Walk/bicycle 29.9% 17.6% 38.8% 22.4% Vehicle occupancy 1.36 1.71 1.35 1.69 Source: Mode share and vehicle occupancy adjustment factors were extracted from linked trip data records developed from the 2004 MWCOG HTS.
33 generation estimates appeared reasonable to the research team and, in fact, is similar to the findings from other research (2). The research team concludes that the difference between pre- dicted and actual vehicle-trip generation is great enough for the investigators to find the results inconclusive without data from additional sites. 5.4.2 Office Land Use Category The results of the comparison of the office case study sites are shown in Table 5.3. Similar to the residential sites, apply- ing the methodology to the office sites results in a relatively consistent infill automobile mode share in the siteâs respective TAZs (see Table 5.1). The method consistently overpredicts both a.m. and p.m. peak hour vehicle-trip generation when compared to the actual trips. Although the overprediction of the office sites is not as great as shown for the residential sites, application of the method results in predictions greater than 60% over actual trips. Also similar to the residential case study sites, the method- ology predicts peak hour trip generation at about one-half of the trips estimated using ITE rates. As with the residential sites, the research team concludes that the difference between actual and predicted trip generation varies enough to find the results inconclusive without data from additional office sites. Despite the inconclusiveness of the methodâs predictions when compared to surveys, it is clear that the selected sites (all close to rail stations), if analyzed using ITE trip generation rates, would result in overestimation. However, in all cases the methodâs estimates are much closer to what was observed in the field than to what was estimated using ITE rates. The comparison of the office siteâs actual trip generation with ITE trip generation estimates triggered further investigation. The research team expected that trip estimates for office buildings in GU/UC contexts would be lower than those in suburban contexts (the context presumed to be represented by ITE trip generation rates), but not to the extent observed in Table 5.3. Table 5.2. Comparison of actual versus predicted peak hour vehicle-trip generation (residential sites). Note: DUs = dwelling units. Residential Sites Columbia Uptown The Lencshire House The Beauregard Average of Residential Sites 90 DUs 125 DUs 45 DUs 87 DUs a.m. p.m. a.m. p.m. a.m. p.m. a.m. p.m. Predicted vehicle trips 25 30 39 47 10 12 25 30 Actual vehicle trips 13 12 19 14 7 6 13 11 Vehicle trips based on ITE avg. rates 40 47 55 65 20 23 38 45 Percent diff. (predicted vs. actual) 92% 150% 105% 236% 43% 100% 90% 178% Percent diff. (Predicted vs. ITE) -38% -36% -29% -28% -50% -48% -36% -34% Table 5.3. Comparison of actual versus predicted peak hour vehicle-trip generation (office sites). Office Sites 1920 N Street NW 1616 N Fort Myer Drive 1200 Wilson Boulevard Average of Office Sites 114 KSF 303 KSF 146 KSF 188 KSF a.m. p.m. a.m. p.m. a.m. p.m. a.m. p.m. Predicted vehicle trips 85 84 222 222 107 107 138 138 Actual vehicle trips 43 51 134 134 65 67 81 84 Vehicle trips based on ITE avg. rates 177 170 470 452 226 218 291 280 Percent diff. (predicted vs. actual) 98% 65% 66% 66% 65% 60% 71% 64% Percent diff. (predicted vs. ITE) -52% -51% -53% -51% -53% -51% -53% -51% Note: KSF = thousand square feet.
34 5.4.3 Retail and Restaurant Land Use Categories Table 5.4 and Table 5.5 compare predicted with actual peak hour vehicle trips for the retail and restaurant land use cate- gories, respectively. The application of the method to the two retail sites resulted in an outcome that was different from the outcome observed for office sites. The investigators expected the method to overpredict compared to actual vehicle trips, and expected the predicted trips to be substantially lower than trips estimated using ITE rates. The expected pattern for predicted versus actual did not occur, and instead the results shown in Table 5.4 are quite variable. The predicted trips for one of the case study sites are relatively close in the p.m. peak hour (a difference of 11%), while the a.m. peak hour is predicted about 51% lower than actual. At the second site, the predicted trips are 58% to 87% lower than the actual trips. The comparison of predicted to estimated trips using ITE data shows that the method produces consistently lower trip estimates, ranging from two-thirds to three-quarters of the ITE estimatesâa finding the researchers did not expect. The unexpected comparison between predicted and actual trips spurred further investigation of the data. The research team found a potential reason for the unusually large under- estimation at the 819 H Street site: the mix of retail estab- lishments in the actual shopping centerâin particular, a fast food restaurant, a pharmacy, dry cleaners, and a convenience store, which are all high traffic-generating uses by themselves (and high morning traffic generators). These uses, typically, do not experience much internalization of trips between them, and the combination of high-generating uses and lit- tle internal capture of trips will result in a high vehicle-trip generation. In summary, the research team finds the comparison of the retail sites inconclusive because of the small sample size, the high variability between predicted and actual trips, and the vari- ability when compared to ITE methods of estimating trip gen- eration. In retrospect, the research team believes the criteria for selecting retail case study sites should be more restrictive on the type and compatibility of uses within multi-use shop- ping centers. Future verification of the method could include trip generation estimates of individual land uses, as well as estimating trips for shopping as a single use. Only one restau- Table 5.4. Comparison of actual versus predicted peak hour vehicle-trip generation (retail sites). Retail Sites 1315 N. Rhode Island Ave. NE 819 H Street NE Average of Retail Sites 36 KSF 37 KSF 37 KSF a.m. p.m. a.m. p.m. a.m. p.m. Predicted vehicle trips 21 81 21 84 21 83 Actual vehicle trips 43 73 164 198 104 136 Vehicle trips based on ITE equation 84 212 86 217 85 215 Percent difference (predicted vs. actual) 51% 11% 87% 58% 80% 39% Percent diff. (predicted vs. ITE) 75% 62% 76% 61% 75% 62% Table 5.5. Comparison of actual versus predicted peak hour vehicle-trip generation (restaurant site). Restaurant Site 1333 Rhode Island Avenue NE 2.4 KSF 2.4 KSF a.m. Peak Hour p.m. Peak Hour Predicted vehicle trips 24 19 Actual vehicle trips 9 8 Vehicle trips based on ITE avg. rates 28 27 Percent diff. (predicted vs. actual) 167% 138% Percent diff. (predicted vs. ITE) -14% -30%
35 rant site was included in the data collection site, and this fact alone makes any findings inconclusive. However, the research team wanted to see if the method resulted in the same pattern of overpredicting surveys and estimating substantially lower trips that ITE data produces, such as seen with residential and office uses, or if the method would produce inconsistent and widely variable findings like those observed in the retail sites. The data in Table 5.5 show that the single-restaurant data fol- low the same pattern as the residential and office dataâthe method overpredicting trips compared to surveys and having substantially lower trip estimates compared to average ITE trip generation rates. 5.5 Application of the Approach Using Data from the San Francisco Bay Area Household Travel Survey The research team conducted a second verification analy- sis using the adjustment factors extracted from the 2000 Bay Area travel survey during the development of the household travel survey method and surveyed trip generation data for several sites in the San Francisco Bay Area collected as part of a previous infill trip generation study (16). Table 5.6 presents the mode share and vehicle occupancy adjustment factors subdivided by proximity to rail and high-frequency bus tran- sit and representing the context of GU/UC in the San Fran- cisco Bay Area. The factors in Table 5.6 are a slightly modified version of the factors presented in Chapter 4 (Table 4.3). An additional LUC has been added to provide data for an LUC of coffee shop or bagel/donut shop. In contrast to the minimum data collection variant used for collecting data at the Washington, D.C., study sites, the San Francisco Bay Area study sites were selected as part of the California urban infill trip generation study (1) that used the approach of direct estimation of trip generation based on empirical data. The objective of the California study was to develop trip generation rates for each of the land use categories being studied. The data collection methodology used the techniques listed under the comprehensive data col- lection variant, which include cordon counts of person trips, automobile counts at driveways, and randomly sampling intercept surveys to obtain mode share and other informa- tion. Vehicle occupancy data were not collected as part of the California study. Seven of the California study sites (which are all located in the San Francisco Bay Area) were selected for this verifica- tion analysis. The seven sites were one multifamily residential apartment building, one general office building, retail (in the form of one copy shop and one florist on the ground floors of mixed-use buildings), one quality (sit-down) restaurant, one local coffee shop, and one bagel shop. 5.5.1 Findings and Overall Conclusions of the Analysis Table 5.7 presents the results of the trip generation analysis. The table is organized with the three right-most pairs of col- umns showing the predicted infill vehicle trips, the actual infill vehicle trips, and trips estimated using baseline ITE trip gen- eration rates. The rows following the trip comparison present the percentage difference between the predicted and actual trips and between the predicted and ITE estimated trips. Unlike the Washington, D.C., verification analysis, there is no discernible pattern of predicted trips overestimated when compared to actual trips and underestimated when compared Table 5.6. Mode share and vehicle occupancy adjustment factors for the San Francisco Bay Area.1 Infill Adjustment Factors for GU/ UC Contexts Within Walking Distance Of: High- Frequency Bus Stop Rail Station a.m. p.m. a.m. p.m. Multifamily Residential (ITE LUC 223) Transit 20.2% 17.5% 19.3% 16.2% Walk/bicycle 13.4% 13.3% 13.2% 13.7% Vehicle occupancy 1.61 1.60 1.58 1.61 General Office (ITE LUC 710) Transit 23.6% 22.4% 20.6% 20.6% Walk/bicycle 8.4% 8.7% 9.1% 9.4% Vehicle occupancy 1.36 1.27 1.35 1.27 Retail/Shopping Center (ITE LUC 820) Transit 12.7% 10.7% 13.1% 11.7% Walk/bicycle 11.4% 15.3% 12.3% 16.3% Vehicle occupancy 1.50 1.49 1.55 1.53 Quality (Sit-Down) Restaurant (ITE LUC 932) Transit 26.7% 14.3% 25.3% 15.5% Walk/bicycle 20.8% 16.6% 20.6% 19.8% Vehicle occupancy 1.37 2.07 1.44 2.13 Coffee Shop and Bagel/Donut Shop (ITE LUC's 936 and 939)2 Transit 23.6% 22.4% 20.6% 20.6% Walk/bicycle 8.4% 8.7% 9.1% 9.4% Vehicle occupancy 1.36 1.27 1.35 1.27 1 Mode share and vehicle occupancy adjustment factors were extracted from linked trip data records developed from the 2000 Bay Area travel survey, Metropolitan Transportation Commission. 2 Coffee shop and bagel/donut shop land use categories are too specific to extract mode share and vehicle occupancy factors from the travel survey data. Therefore, general office factors were used as representative of the primary trip purpose of people who use these categories.
36 to ITE trip estimates. The surveys of many of the sites produced low traffic volumes, potentially introducing an exaggerated percentage difference based on a small number of trips. The method consistently results in a lower number of trips, by one-third to one-half of trips estimated using ITE rates, similar to the findings of the Washington, D.C., analysis for residential and office land uses. In general, the method consistently predicts closer to the actual number of trips in the San Francisco Bay Area analysis than in the Washington, D.C., analysis. Table 5.7. Comparison of actual versus predicted peak hour infill vehicle-trip generation. Site/Location ITE (LUC) Size Units Context Predicted Actual ITE Rate/Equation Estimate1 Infill Vehicle Trips Infill Vehicle Trips Infill Vehicle Trips a.m. p.m. a.m. p.m. a.m. p.m. Multifamily residential LUC 223 99 DUs Urban center 16 20 4 28 30 39 2116 Allston Way, Berkeley, CA Percentage diff. (predicted vs. actual) 304% -28% Percentage diff. (predicted vs. ITE rate based) -47% -49% General office building LUC 710 120.000 KSF Urban center 106 110 145 110 186 179 388 Sutter Street, San Francisco, CA Percentage diff. (predicted vs. actual) -27% 0% Percentage diff. (predicted vs. ITE rate based) -43% -39% Retail (copy center) LUC 820 3.000 KSF Urban center 2 5 n/a 12 3 7 2111 University Avenue, Berkeley, CA Percentage diff. (predicted vs. actual) n/a -58% Percentage Diff. (Predicted vs. ITE Rate Based) -33% -29% Retail (florist) LUC 820 2.400 KSF Urban center 2 4 2 7 2 6 2004 University Avenue, Berkeley, CA Percentage diff. (predicted vs. actual) -2% -43% Percentage diff. (predicted vs. ITE rate based) 0% -33% Quality (sit-down) restaurant2 LUC 932 3.000 KSF Urban center 11 7 14 13 17 12 337 3rd Street, San Francisco, CA Percentage diff. (predicted vs. actual) -20% -44% Percentage diff. (predicted vs. ITE rate based) -34% -42% Coffee shop3 LUC 936 4.500 KSF Urban center 309 115 81 35 528 183 1910 Oxford Street, Berkeley, CA Percentage diff. (predicted vs. actual) 284% 226% Percentage diff. (predicted vs. ITE rate based) -41% -37% Bagel/donut shop3 LUC 939 5.000 KSF Urban center 206 88 18 42 351 140 1370 University Avenue, Berkeley, CA Percentage diff. (predicted vs. actual) 1044% 108% Percentage diff. (predicted vs. ITE rate based) -41% -37% Notes: 1 Retail, quality restaurant, coffee shop, and bagel/donut shop trips estimated using ITE rates or equations are new trips and exclude pass-by trips. 2 Quality restaurant: a.m. peak represents the morning peak hour of the generator, and not the peak of the adjacent street traffic. 3 Analysis uses general office adjustment factors based on an assumption that coffee shop and bagel/donut customer trips are composed predominantly of work trip purposes. As a reasonableness check, the research team again com- pared the percentage difference between predicted and ITE estimates of the residential site with other published sources (17). This study and other research support a similar conclu- sion that, at least for residential uses, the use of ITE rates in estimating infill trip generation results in an overestimation of one-third to one-half. The research team anticipates that with more site data, the method presented in this report will produce results similar to those in other studies of infill development or TOD that are based on empirical data.