Focus Model Calibration 1.0August 2010
Focus Model Calibration 1.0
Denver Regional Council of Governments (DRCOG) Activity-Based Travel Model
8/5/2010
Suzanne Childress
Contents
Table List
Introduction
GISDK Modifications
Erik and Shahida---
Population Synthesizer
Regular Work Location
Regular School Location
Auto Availability Choice
Daily Activity Pattern Choice
Exact Number of Tours
Work Tour Destination Type
Work-Based Subtour Generation
Tour Primary Destination Choice
Tour Main Mode Choice
Tour Time of Day Choice
Intermediate Stop Generation Choice
Intermediate Stop Location Choice
Trip Mode Choice
Trip Departure Time Choice
Highway Assignment
Transit Assignment
Conclusions and Further Calibration Directions
Appendix 1: Basic Aggregate Model Results
Appendix 2: Assignment Model Summary File
Table List
Table 1. Focus Model Flow
Table 2. Observed and Modeled Vehicle Miles Traveled on Links with Counts
Table 3. Observed and Modeled Transit Boardings
Table 4. 2005 Regional Controls on Number of Households
Table 5. Regional Percentages- Households By Age of Householder
Table 6. Regional Percentages- Households by Presence of Children
Table 7. Regional Controls Input and Output
Table 8. 2005 PopSyn/ACS Number of Households By County
Table 9. Difference in Percents Households by Income Group: ACS and PopSyn
Table 10. Households Size By County Difference in Percents
Table 11. Income Group Controls by Percents
Table 12. Household Size Controls and Outputs by Percents
Table 13. 2005 Employed People By County- PopSyn vs ACS
Table 14. 2005 State Demographer Forecast Persons by Age Cohort By County
Table 15. 2005 PopSyn Outputs Persons By Age Cohort By County
Table 16. Percent Difference State Demographer - PopSyn Cohort By County
Table 17. Average Distance To Work by Person Characteristics
Table 18. Average Work Skimmed Distance and Logsums By Home District
Table 19. Modeled and Observed Percent of Workers Regularly Working at Home
Table 20. CTPP Target and Model Results- Workers by Home to Regular Workplace
Table 21. Average Distance to School by Student Grade Level
Table 22. Percent of Students attending a school located within school district boundaries
Table 23. Percent of Modeled 2005 Households by Auto Ownership By County
Table 24. Percent of Observed 2005 ACS Households by Auto Ownership By County
Table 25. Percent of Modeled Households by Income Group By Auto Ownership
Table 26. Percent of Observed TBI 1997 Households By Income Group By Auto Ownership
Table 27. Modeled versus Observed Percent of Persons Making Tours by Purpose
Table 28. Modeled versus Observed Percent of Persons Making Stops By Purpose
Table 29. 2005 Modeled Persons by Type by Average Number of Home-Based Tours by Purpose
Table 30. 1997 TBI Persons by Type by Average Number of Home-based Tours by Purpose
Table 31. Average Tours per Person by Home District Observed and Modeled
Table 32. Modeled and Observed % of Work Tours to the Regular Workplace
Table 33. Percent of Work Tours by Number of Subtours
Table 34. Average Modeled and Observed Tour Straight Line Distance
Table 35. Target SuperDistrict to SuperDistrict Non-Mandatory Tours
Table 36. Modeled SuperDistrict to SuperDistrict Non-Mandatory Tours
Table 37. Modeled Number of Tours Destined to District Compared with Number of People and Jobs
Table 38. Target and Modeled Work Mode Share All Destinations
Table 39. Target and Modeled Non-Work Mode Share All Destinations
Table 40. Target and Modeled Work Mode Share CBD Destinations
Table 41. Target and Modeled Non-Work Mode Share CBD Destinations
Table 42. Average Modeled Stops by Stop Purpose and Person Type
Table 43. Average Observed TBI Stops by Stop Type and Person Type
Table 44. Average Modeled Trip Distance (for trips with intermediate stops) by trip purpose
Table 45. Number of Modeled Intermediate Stops Compared to Number of People and Jobs By District
Table 46. Work Trip Mode Choice Shares Targets and Modeled
Table 47. Observed and Modeled 2005 VMT on links with counts
Table 48. Total Vehicle Miles Traveled By Facility Type on Links with Counts
Table 49. Total Vehicle Miles Traveled By Area Type on Links with Counts
Table 50. Modeled and Observed Volumes By Screenline
Table 51. Modeled Transit Trips and Boardings
Table 52. Average RTD Observed and Modeled Boardings By Sub-Mode
Introduction
The Focus activity-based travel model was recently developed and calibrated by Denver Regional Council of Governments and Cambridge Systematics. As activity-based travel model calibration is a new relatively new frontier, the calibration required several innovations for data comparison and calibration modification of variables, coefficients, and constants.
Each Focus model component was calibrated individually and then the entire model was calibrated aggregately against roadway counts and RTD transit boardings.Table 1 shows the Focus model component flow. The network skimming, assignment and airport models were adapted from the earlier trip-based TransCAD model. All location, mode choice, and time of day model travel components were re-estimated for Focus using the 1997 Travel Behavior Inventory (TBI) survey of regional households.
Table 1. Focus Model Flow
- Population Synthesizer
- TransCAD Initialization
- TransCAD Trip Generation
- TransCAD Skimming
- Population Synthesizer
- Size Sum Variable Calculator
- Regular Workplace Location
- Regular School Location
- Auto Availability
- Aggregate Destination Choice Logsum Generation
- Daily Activity Pattern
- Exact Number of Tours
- Work Tour Destination Type
In this calibration, the model was calibrated up to 2005 data from the 1997 estimation. Data from 2005 was used wherever possible to ensure that the model was correctly capturing observed 2005 Denver travel behavior. The following 2005 datasets were used to calibrate against: 2005 American Community Survey (ACS), 2005 Colorado state demographer data, 2005 Colorado Department of Transportation (CDOT) highway counts, 2005 HPMS estimated regional VMT, 2005 Regional Transportation District (RTD) transit boardings and 2005 Compass trip-based model results. Unfortunately, because activity-based travel models are relatively new, large datasets were not available for some of the detailed travel behavior output by the model and many individual components had to be calibrated against the 1997 weighted expanded TBI survey data. The TBI and other older datasets like the 2000 Census were used in combination with growth factors to account for regional growth to 2005. One other dataset that was used for calibration in combination with growth factors was the 2000 Census Transportation Planning (CTPP) journey-to-work data. A new 2010 regional travel survey, Front Range Travel Counts and the 2010 Census should allow for a refreshed Focus calibration during 2011.
Once comparisons were made of model results against the observed datasets, each model component was calibrated. The calibration involved changing utility function constants, coefficients, and adding variables. Then the model was re-run, results compared again, and modifications made again. This process was iterated as time allowed until satisfactory results were achieved.
The major regional level model results of the calibration are shown in Table 2 and Table 3. These tables demonstrate that the aggregate model results match the observed counts and transit boardings well.
Table 2. Observed and Modeled Vehicle Miles Traveled on Links with Counts
Observed VMT / Modeled VMT20,506,768 / 20,906,583
Table 3. Observed and Modeled Transit Boardings
Observed Transit Boardings / ModeledTransit Boardings
269,741 / 263,508
The remainder of document first details how each individual model was calibrated. It finishes with the aggregate highway and transit assignment results, as well appendices showing overall model summaries.
GISDK Modifications
The first model component that was changed for calibration was the GISDK code based used for network skimming and assignment.
Erik and Shahida---
base code set came from Compass 2.0 / matching code used from estimation.
changes made over Compass:
- use of trips passed from SQL Server focus, combined with old Compass DIA, I-E, E-E and commercial trips;
- auto operating cost based on VOC
- 10 ten time period assignment changes
- 4 transit periods
- many additional variables skimmed for focus model- ex generalized time; piecewise linear skim variables
- Value of Time by area type changed to be more consistent with focus mode choice models
Population Synthesizer
After network skimming is completed, the Population Synthesizer (PopSyn) creates a recordset of individual regional households and persons. This section presents the results of validation tests performed on PopSyn developed for Focus.The Population Synthesizer creates a forecast of individual households and persons for chosen year. It operates by drawing household and person records from the 2000 Public Use Microsample (PUMS) with the goal of matching forecasted demographic controls. The 2005 synthesized population is validated by observing how well the synthesized data matches both the inputs and independent data sources.
The inputs are also examined to verify their accuracy, and uncontrolled variables are compared to the American Community Survey(ACS) estimates and state demographer data. The controls are being adequately maintained in the model run. Overall the validation shows that PopSyn’s results are acceptable enough for use in the FOCUS.
However, a few discrepancies between PopSyn’s data and external data sources were identified in this validation, as shown in the bullets below.
- The 2005 ACS estimated 3% fewer households in the 6 major counties[1] than was given in the 2005 land use forecast given as input to PopSyn. The 2005 ACS estimated 5% fewer households in Denver County than PopSyn.
- Some of the variables that were uncontrolled like workers and person by age came out of PopSyn different from ACS and the state demographer data:
- PopSyn produced 6% more workers in the six counties than the ACS estimated.
- PopSyn produced 10% more persons age 25-44 and 9% fewer persons age 15-24 in the six counties than the state demographer’s forecast.
Improvements to PopSyn require refinement to its inputs, which are controls on regional and zonal data. A better economic forecast will result in better regional demographics output by PopSyn. The land use model is used to create zonal controls; greater detail and accuracy with the land use model will improve PopSyn’s ability to project the types of households in each zone in future years. Better ability to forecast shifts in income groups by zone (gentrification, change in real income) would improve PopSyn ability to forecast households in the correct income groups in future years.
The recommendations for PopSyn’s improvement are the following:
- Recommendation 1. Obtain a new economic forecast with greater detail and accuracy, including a regional forecast for households by number of adults, number of workers, number of children, and age of householder.
- Recommendation 2. Upgrade the land use model to make changes in income group distributions over time by zone. Develop the land use model’s ability to forecast more demographic characteristics (i.e. number of households by number of workers/number of children). Improvements to the land use model will allow PopSyn to output the types of households in each zone more accurately.
2005 PopSynControl Variables
The first validation tests described in this document concern the controlled household characteristics. The Population Synthesizer uses two sets of controlled variables for household characteristics: regional-level controls and zonal-level controls. For the 2005 PopSyn run, the regional controls come from the 2035 economic forecast [2]. The zonal controls are based on the land use model and 2000 Census data.
Two separate questions are posed about both the regional and zonal controls:
First, are the controls valid? Do they match other data sources? Secondly, assuming the controls are valid, do the output control totals match the input control totals?
PopSyn Regional Controls
The regional controls provide targets for the Population Synthesizer to attempt to match demographics for the travel model region. These controls were created based on the 2035 economic forecast.
Because of some discrepancies between the shares of households by age of householder and households by type between the economic forecast for 2005 and the 2005 ACS, we chose to scale the economic forecast to the 2005 ACS shares for these categories. Also, the 2005 Economic Forecast was scaled to match the 2005 land use model for total regional households. The scaled regional controls are for total households by household type and age of householder as follows in Table 4.
Table 4. 2005 Regional Controls on Number of Households
Household Type / Ageof Householder / NUMBER OF HOUSEHOLDS FORECASTED AND SCALED TO LAND USE MODEL AND ACS
YEAR / 2005 / 2010 / 2015 / 2020 / 2025 / 2030 / 2035
One Adult
No Kids / 18-44 / 127,464 / 126,304 / 129,259 / 137,321 / 150,649 / 164,723 / 175,435
One Adult
No Kids / 45-64 / 91,024 / 103,164 / 108,649 / 115,192 / 114,561 / 116,427 / 121,503
One Adult
No Kids / 65 & over / 68,418 / 80,059 / 102,289 / 132,151 / 164,172 / 194,074 / 214,240
One Adult
With Kids / 18-44 / 48,015 / 48,338 / 50,312 / 54,382 / 60,598 / 67,129 / 72,354
One Adult
With Kids / 45-64 / 11,950 / 14,544 / 16,158 / 18,029 / 18,815 / 19,874 / 21,648
One Adult
With Kids / 65 & over / 2,148 / 3,134 / 4,617 / 6,749 / 9,318 / 11,947 / 14,340
Two or More Adults
No Kids / 18-44 / 136,051 / 136,853 / 140,596 / 150,177 / 165,863 / 181,653 / 194,438
Two or More
Adults
No Kids / 45-64 / 194,971 / 221,177 / 233,632 / 248,607 / 247,845 / 252,207 / 262,493
Two or More Adults
No Kids / 65 & over / 80,399 / 97,097 / 127,451 / 167,025 / 208,702 / 246,262 / 271,536
Two or More Adults
With Kids / 18-44 / 224,469 / 223,099 / 230,728 / 246,932 / 272,744 / 302,626 / 326,056
Two or More Adults
With Kids / 45-64 / 83,292 / 95,479 / 101,751 / 109,268 / 109,869 / 112,571 / 117,921
Two or More Adults
withKids / 65 & over / 4,177 / 5,177 / 6,918 / 9,255 / 11,827 / 14,254 / 16,079
Table 5compares the percentages of the households with householders in three age categories: 18-44, 45-64, and 65+ for the 2005 base year and for future years. Note that the 2005 controls and the 2005 ACS are nearly identical because of the controls were scaled to match the ACS.
Table 5. Regional Percentages- Households By Age of Householder
% of Households by Age of HouseholderForecast Year or Estimate Sources / 18-44 / 45-64 / 65+
2005 ACS / 49% / 36% / 15%
2005 Economic Forecast Control / 50% / 36% / 14%
2010 Economic Forecast Control / 46% / 38% / 16%
2015 Economic Forecast Control / 44% / 37% / 19%
2020 Economic Forecast Control / 42% / 35% / 23%
2025 Economic Forecast Control / 42% / 32% / 26%
2030 Economic Forecast Control / 42% / 30% / 28%
2035 Economic Forecast Control / 42% / 29% / 29%
Table 6shows the percentage of households with and without children in the PopSyn controls and from the ACS for the 2005 base year and future years. The economic forecast controls in 2005 were scaled to the shares of households by presence of children from the ACS. Note that share of households with and without children are nearly identical in 2005 for the ACS and the economic forecast because the controls were scaled to match the ACS.
Table 6. Regional Percentages- Households by Presence of Children
Percent of Households withChildren / Percent of Households
Without Children
2005 ACS / 34% / 66%
2005 Economic Forecast / 35% / 65%
2010 Economic Forecast / 34% / 66%
2015 Economic Forecast / 33% / 67%
2020 Economic Forecast / 32% / 68%
2025 Economic Forecast / 32% / 68%
2030 Economic Forecast / 31% / 69%
2035 Economic Forecast / 31% / 69%
The second question that needs to be answered is whether the regional controls are being adequately maintained during PopSyn’s operation.Table 7below compares the regional control totals input to the PopSyn’s outputs. All the input controls match the output households with less than a 0.4% difference with several outputs substantially below this difference. Therefore, PopSyn is effectively maintaining the regional controls.
Table 7. Regional Controls Input and Output
Household Type / Age of Householder / % Difference Input and OutputOne Adult NoKids / 18-44 / -0.002%
One Adult No Kids / 45-64 / -0.001%
One Adult No Kids / 65 & over / -0.012%
One Adult with Kids / 18-44 / 0.042%
One Adult with Kids / 45-64 / 0.203%
One Adult with Kids / 65 & over / 0.383%
Two or More Adults No Kids / 18-44 / -0.007%
Two or More Adults No Kids / 45-64 / -0.006%
Two or More Adults No Kids / 65 & over / -0.010%
Two or More Adults with Kids / 18-44 / -0.012%
Two or More Adults with Kids / 45-64 / -0.011%
Two or More Adults with Kids / 65 & over / 0.312%
Zonal Controls
The zonal controls provide targets for the Population Synthesizer to match on characteristics of synthesized households on a zonal level. PopSyn uses the 2812 zone system developed for the FOCUS project. The zonal controls are results of the land use model and 2000 Census data. An improvement to the controls would have to result from an improvement in the land use model. The controls for each zone are targets for the following nine statistics:
ZONAL CONTROLS ( for each of the 2812 zones)
(1) Total Households in the Zone
(2) Percentage of Households in Zone with Income 0-30K
(3) Percentage of Households in Zone with Income 30-60K
(4) Percentage of Households in Zone with Income 60-100K
(5) Percentage of Households in Zone 100K+
(6) Percentage of Households of Size 1
(7) Percentage of Households of Size 2
(8) Percentage of Households of Size 3
(9) Percentage of Households of Size 4+
As with the regional controls, the first question we ask is: Are the controls themselves reasonable? Table 8compares the number of households from the 2005 land use model to the 2005 ACS. Table 8 shows that the 2005 land use model estimated 3% more households in the six counties overall than the 2005 ACS estimated. The land use modeling team asserted that differences in estimation methods between the land use model and ACS, as well as sampling error accounted for the differences seen inTable 8.
Table 8. 2005 PopSyn/ACS Number of Households By County
County / 2005 Land Use / 2005 ACS / Difference / %Difference / ACS Margin of ErrorAdams / 145256 / 141383 / 3873.5 / 3% / +/- 1.8%
Arapahoe / 211224 / 206250 / 4974.5 / 2% / +/- 1.4%
Boulder / 118178 / 113405 / 4773.3 / 4% / +/- 1.7%
Denver / 253764 / 241579 / 12185.2 / 5% / +/- 1.4%
Douglas / 87807 / 87654 / 153.3 / 0% / +/- 1.1%
Jefferson / 213322 / 211394 / 1928.0 / 1% / +/- 1.0%
Total / 1029553 / 1001665 / 27887.8 / 3%
Table 9shows the difference between the ACS and PopSyn income group distributions. The 2005 PopSyn estimates of percentages of households by income group came directly from the 2000 Census. The 2005 ACS estimates, in comparison, include inflationary effects from 2000 to 2005. As a result, the percent of households in the highest group of $100,000+ had more households in it in the ACS than in PopSyn. Furthermore, the $30-60K group had 4% fewer households in the ACS than in PopSyn. Improvements in the ability to forecast the number of households by income group changes over time would be required to cause PopSyn to more accurately predict income distributions.