Model Validation and Reasonableness Checking Manual

3.0 Trip Generation

The trip generation model estimates the number of motorized person trips to and from each TAZ in the study area. In this step of the travel forecasting process, socioeconomic data are used to estimate the number of daily motorized person trips within the study area, i.e. internal-internal, and with origins or destinations outside the study area, i.e. external-internal or internal-external.

The trip generation model estimates trip productions and trip attractions. For transportation planning purposes, a trip production is a trip end made at the home location for home-based trips and the origin location for non-home-based trips. For example, if a person travels from home to work and then from work to home on a certain day, that person would be considered to have two home-based work trip productions at his or her home and two home-based work trip attractions at his or her work location.

In most metropolitan area transportation models, trips are stratified by purpose. Typical trip purposes can include: home-based work; home-based non-work such as shopping, school, other; and non-home-based.

The trip generation model typically has a number of components including the following:

Socioeconomic Disaggregation Submodels -- These models provide data in sufficient detail to apply disaggregate trip production models. For example, one may need to estimate households by income group and household size given zonal households, populations, and median household income. Other models can be used to project auto ownership for households.
Trip Production Models -- These models estimate trip productions on a traffic analysis zone level. Productions are typically a function of population or number of households (or both) along with a measure of wealth such as income or autos. Other explanatory variables might be used (e.g. number of workers, life-cycle, etc.)
Trip Attraction Models -- These models estimate trip attractions on a traffic analysis zone level. Attractions are typically a function of socioeconomic activity - households, employment by type, school enrollment - but can also be land-use based (e.g. gross floor area for manufacturing, retail, government, open space, etc.).

Two other components of trip generation include:

Estimation of external trip ends
Procedure for balancing trip productions and attractions

3.1 Socioeconomic Disaggregation Submodels

Model Description
The socioeconomic submodels play an important role in forecasting the inputs to disaggregate trip generation models. While the detailed demographic data required for trip generation is available for the base year from the Census, land use forecasting procedures will typically only produce aggregate zone level estimates of households, population, median income, and vehicles. As a result, socioeconomic submodels are needed to develop disaggregate zonal estimates.

It has been ascertained in a number of other studies that the mix of disaggregated households is fairly similar for any spatial grouping given the average values. For example, if the average household size in a zone is 1.5 persons per household, it is logical to anticipate that there will be large numbers of one- and two-person households and fewer households with more than three persons. In order to develop a model, household data are summarized for small ranges of the zonal average, whether it be household size, income, or autos owned, to provide average aggregate estimates of the mix of households.

The primary data source used for calibration is typically the Census Transportation Planning Package (CTPP) at either the TAZ or Census tract level. For example, CTPP Table 1-17 lists the number of households by household size and vehicles available. A household travel survey may be used as a secondary source for verifying the distributions since it is not as robust as the Census data. The CTPP provides a breakdown of households by zone for the households size, auto ownership, and income group classifications.

An example of a household size disaggregation model is shown in Figure 3-1. A similar set of curves can be developed for other socioeconomic variables.

Figure 3-1
Household Size Disaggregation Model

Another type of procedure used by regions are disaggregate vehicle ownership (or availability) models which predict the number of vehicles available to households for each traffic analysis zone. These models typically incorporate a number of socioeconomic variables, the most important of which is income level. The model structure can vary from empirical curves to discrete choice models, but the type of aggregate validation checks used is roughly the same for all procedures.

Table 3-1 displays typical percentages of households by autos owned and income level.

Table 3-1
Percent of Households by Autos Owned and Income

Urbanized Area Size = 200,000 - 499,999
INCOME	AUTOS OWNED
INCOME	0	1	2	3+
Low	17	51	24	8
Medium	2	32	53	13
High	0	13	53	34
Weighted Avg.	7	32	42	19
Source: NCHRP 365 Note: In 1990 dollars, Low Income = less than $20,000; Medium Income = $20-39,999; and High Income = $40,000 and up.

Validation Tests
The models can be validated against the zonal level Census data used to develop them. The models would first be applied using the calibration data (e.g., for a household size submodel, using the observed average household size ). The result of this step would be observed and estimated households by household size for each zone.

Several possible validation tests are described below:

Compare observed and estimated households by socioeconomic subgroups. The differences can be examined in absolute terms and the coefficient of determination (R²) can be calculated over all strata (e.g. 0-1 Avg. Household Size, 1-2 AHHS, 2-3 AHHS,...). Look for systematic biases. An example of the socioeconomic subgroups is shown in Table 3-2.
Calculate correlation (or coefficient of determination R²) of shares of observed and estimated households by subgroups. R² can be inflated since it also measures "zone size effects", i.e. zones with a lot of households result in a lot of households while zones with few households result in few households by group. Using the shares instead of absolute values helps to factor out "zone size effects."
Calculate correlation (or coefficient of determination R²) and plot the relationship between the observed and estimated households for each household size group at the district or census tract level. Look for geographic biases. An example scatterplot is shown in Figure 3-2.

Other types of models might be used to estimate socioeconomic variables. For example, a few regions have developed disaggregate choice model to predict vehicle ownership (or availability). The methods outlined above can be used to validate these models. Other, more disaggregate, tests can also be performed (see discussion in Chapter 5.0 - Mode Choice).

Table 3-2
Observed and Estimated Households by Size Subgroups

Household Size Range	1 Person Households		2 Person Households		3 Person Households		4 Person Households		Total Households
Household Size Range	Obs.	Est.	Obs.	Est.	Obs.	Est.	Obs.	Est.	Total Households
1.00 - 1.04	181	181	0	0	0	0	0	0	181
1.05 - 1.14	0	0	0	0	0	0	0	0	0
1.15 - 1.24	0	0	0	0	0	0	0	0	0
1.25 - 1.34	308	293	28	33	5	15	8	8	349
1.35 - 1.44	29	35	17	7	0	3	0	2	46
1.45 - 1.54	382	391	115	119	38	34	33	23	568
1.55 - 1.64	1,531	1,452	515	622	183	190	144	109	2,373
1.65 - 1.74	987	912	425	512	114	150	136	88	1,662
1.75 - 1.84	1,656	1,749	1,211	1,232	472	357	231	232	3,570
1.85 - 1.94	1,208	1,187	938	991	313	320	250	211	2,709
1.95 - 2.04	1,126	1,259	1,263	1,207	555	452	283	310	3,227
2.05 - 2.14	1,178	1,185	1,211	1,280	600	542	396	379	3,385
2.15 - 2.24	2,714	2,569	2,981	3,074	1,279	1,450	1,312	1,193	8,286
2.25 - 2.34	1,373	1,346	1,562	1,710	1,028	886	775	796	4,738
2.35 - 2.44	1,962	1,908	2,570	2,648	1,534	1,481	1,415	1,444	7,481
2.45 - 2.54	2,948	2,886	4,465	4,405	2,367	2,595	2,877	2,772	12,657
2.55 - 2.64	2,431	2,469	4,172	4,124	2,486	2,567	3,076	3,005	12,165
2.65 - 2.74	2,362	2,316	4,251	4,221	2,664	2,805	3,591	3,526	12,868
2.75 - 2.84	1,506	1,425	2,871	2,832	2,030	2,040	2,500	2,610	8,907
2.85 - 2.94	659	708	1,722	1,518	1,102	1,189	1,577	1,645	5,060
2.95 - 3.04	605	590	1,554	1,332	1,013	1,146	1,585	1,689	4,757
3.05 - 3.14	242	201	622	478	289	460	708	722	1,861
3.15 - 3.24	188	172	492	420	439	449	693	770	1,812
3.25 - 3.34	58	19	22	52	45	62	121	113	246
3.35 - 3.44	0	0	0	0	0	0	0	0	0
3.45 - 3.54	17	14	83	42	64	60	93	141	257
3.55 - 3.64	0	9	56	25	42	39	79	104	177
3.65 - 3.74	6	2	12	6	8	10	27	34	53
3.75 - 3.84	0	0	0	0	0	0	0	0	0
3.85 - 3.94	28	6	47	18	49	34	94	160	218
3.95 or more	143	10	178	31	87	73	113	406	521
Total	25,828	25,296	33,383	32,940	18,806	19,408	22,117	22,490	100,134

Figure 3-2
Observed vs. Estimated Households by Census Tract

3.2 Trip Productions

Model Description
Trip production models have been based primarily on one of two basic structures: (1) regression equations, and (2) cross-classification trip rates. While earlier trip generation models were based on the regression method, most of the recently developed models are now based on the cross-classification method.

Regression models for trip generation were generally developed when origin-destination surveys were conducted for relatively large sample sizes. The large sample sizes provided enough samples of trips to cover most of the geographic area surveyed. This type of model is aggregate since the model is developed using data at the zonal level rather than the household level.

Regression equations explain the variation in a dependent variable, in this case, trips, based on one or more independent, or explanatory, variables. For example, a work trip production model may have the form:

Home-Based Work Trips = a + b * (households) + c * (workers) + d * (autos)

A distinct disadvantage with multivariate regression equations is that explanatory variables are often interrelated and correlated with each other. Interaction effects occur when one independent variable depends on the value of another independent variable. For example, zones with more households would also be expected to have more workers and more autos. Another weakness of regression models is that a large value for the constant a can distort the number of trips estimated for a zone.

Zonal models can only explain the variation in trip making behavior between zones, yet the main variations in person trips data occurs at the household level. In order to overcome this weakness, current state-of-the-practice models typically uses a set of trip production rates stratified by relevant characteristics of households for a given purpose. Trip rates can then be used to estimate trip productions by multiplying the rate by the total number of households in a category or cell.

While the use of a single category, such as auto ownership, will explain some of the variation in the number of trips, the use of multiple variables tends to improve the predictive ability of the model. Stratification of trip rates is often done with at least two independent variables such as income level, auto ownership, number of persons, household density range, and/or number of workers. These variables have been shown to be directly related to trip generation characteristics. Most models will use household size and a wealth variable, such as income or auto ownership, as the independent variables. Base data used for the calibration of trip production models is usually a regional household travel survey.

Cross-classification models are better than regression models in their ability to handle non-linear functions of variables. For example, a four-person household may not produce twice as many trips as a two-person household. Another advantage is that they are calibrated using disaggregate household data, which requires a smaller sample size than is required for more aggregate zone level calibration. The use of disaggregate data (i.e., households) reduces errors due to averaging. The main disadvantage of this approach is the need to forecast the number of households in each category.

There are a number of sources of error in the development of trip generation models. Sampling error and bias in the travel survey affect the trip generation rates. In some cases, the model may not be specified correctly with the relevant explanatory variables.

Validation Tests
The first validation checks which should be made for the trip production models involve examination of total and purpose-specific household trip rates. The most important of these regionwide checks are described below (from aggregate to disaggregate):

Calculate total person trip productions per household or per capita. Examples of typical trip rates are available from the forthcoming publication NCHRP 365 Travel Estimation Techniques for Urban Areas. Table 3-3 shows trip estimation variables by urban size. Table 3-4 shows the average trips per household for a number of regions which was obtained from recent household travel surveys. Note that trip rates range from 8 to 14 trips per household on a typical day. The NCHRP 365 report concludes that urban size may not have a significant impact on variation in trip rates; geographical characteristics and level of service by mode may play a more important role. Variations in trip rates per household might be caused by variation in household sizes; trip rates per capita avoids this problem. A rule of thumb for models calibrated in the past decade is that the total person trips in motorized vehicles per capita should be over 3.0 and, very likely, in the range of 3.5 to 4.0. Note that comparisons of total trips should be consistent in terms of modes (motorized trips vs. all modes) and amount of trip linking.
Calculate total person trips by purpose. Since trip generation models are stratified by purpose, the number of trips by purpose generated by the model is very important. Table 3-5 compares trips rates for a number of regions by purpose. Tables 3-3 and 3-6 compare the percentage of trips by purpose.

Table 3-3
Typical Trip Estimation Variables from NCHRP 365
Urban Area = 200,000 - 499,999

Income	Avg. Autos Per HH	Avg. Daily Pers Trips Per HH	Avg. Daily Veh. Trips Per HH	% Average Daily Person Trips by Purpose
				HBW	HBO	NHB
Low	1.3	6.8	5.4	17	60	23
Medium	1.8	9.5	8.3	20	56	24
High	2.4	12.4	11.2	23	52	25
Wtd. Avg.	1.8	9.0	7.8	21	56	23
HH Size	Avg. Autos Per HH	Avg. Daily Pers Trips Per HH	Avg. Daily Veh. Trips Per HH	% Average Daily Person Trips by Purpose
				HBW	HBO	NHB
1 Person	1.0	3.6	3.2	20	56	24
2 Person	1.9	7.0	6.3	23	53	24
3 Person	2.1	11.3	10.3	22	54	24
4 Person	2.2	13.4	11.2	18	61	21
5 Person+	2.4	16.8	13.5	19	59	22
Wtd. Avg.	1.8	9.0	7.8	21	56	23

Table 3-4
Average Motorized Person Trips per Household by Region

Region	Survey Year	Population	Person Trips/HH
Dallas-Ft.Worth	1984	1,000,000	8.68
Charlotte, NC	1985	511,433	9.29
Vancouver, WA	1985	259,000	5.83
San Diego, CA	1986	2,498,000	14.30
Northern NJ	1986	1,278,000	7.75
Austin, TX	1986	536,693	7.99
Reno,NV	1987	254,000	8.58
Phoenix, AZ	1989	840,000	8.98
Puget Sound	1989	2,559,000	12.20
St.Louis, MO	1990	2,444,000	9.05
Nashua, NH	1990	154,000	10.08
Pittsburg, PA	1990	2,323,000	10.72
Twin Cities, MN	1990	2,464,000	10.11
Atlanta, GA	1991	2,834,000	9.81
Source: FHWA Analysis of Survey Trip Rates (Unpublished)

Table 3-5
Comparison of Person Trips per Household

Purpose	Houston¹	Dallas/Ft. Worth²	Denver²	San Francisco²	Atlanta²	Delaware Valley³
Purpose	1985 Models	1984 Trvl. Sur.	1985 Trvl Sur	1985 Trvl Sur.	1980 Trvl Sur.	1986 Trvl Sur.
HBW	1.71	2.29	1.96	1.89	1.95	2.27
HBNW	4.80	4.32	3.40	4.49	4.45	4.19
NHB	2.96	2.07	1.97	2.35	1.87	1.64
Total	9.47	8.68	7.33	8.71	8.27	8.10

Table 3-6
Comparison of Percentage of Person Trips by Purpose

Purpose	Houston¹	Dallas/Ft. Worth²	Denver²	San Francisco²	Minn/St. Paul⁴	Atlanta²
Purpose	1985 Models	1984 Trvl. Sur.	1985 Trvl Sur	1985 Trvl Sur.	1982 Trvl Sur.	1980 Trvl Sur.
HBW	18.1%	27.0%	26.0%	23.6%	17.9%	23.6%
HBNW	50.6%	47.7%	47.0%	49.7%	53.7%	53.8%
NHB	31.3%	25.3%	27.0%	26.7%	28.4%	22.6%
Total	100%	100%	100%	100.0%	100.0%	100%

Sources:
1 - "Development, Update and Calibration of 1985 Travel Models for the Houston-Galveston Region", Prepared by the Houston-Galveston Area Council and Texas Transportation Institute, June 1991.
2 - "The 1984 Home Interview Survey in the Dallas-Ft. Worth Area: Changes in Travel Patterns, 1964-1984, Transportation Research Record 1134, Transportation Research Board, Washington, D.C., 1987.
3 - "Interregional Stability of Household Trip Generation Rates from the 1986 New Jersey Home Interview Survey", Transportation Research Record 1220, Transportation Research Board, Washington, D.C., 1989.
4 - "Calibration and Adjustment of System Planning Models", FHWA, December 1990.

Compare observed and estimated trips produced at the regional (aggregate) level. Apply the model to base year zonal data to estimate trips produced by zone, and sum over all zones. The estimated number of trips are compared with the observed number of trips, which comes from weighted (expanded to the regional universe of households) trip records from the household travel survey. Comparisons of observed and estimated trips can be made for a number of different classifications including the following:
- Trips by purpose (Home-Based Work, Home-Based Non-Work, etc.)
- Trips by geographical area (region, county, district, zone)
- Trips by income level or autos owned
Differences between observed and estimated trip totals may be due to both error in the trip generation model, as well as sampling error in the household travel survey. In the example shown in Table 3-7, the effect of the sampling error has been explicitly shown by presenting a range for the observed trips by purpose. In most cases, the modeled trips were within the range of the sampling error. However, for some of the purposes, the final estimates of trips were substantially greater than would be expected due to sampling error. In the validation process, this was attributed to under-reporting of trips on the survey and justified as a method to match regional vehicle miles of travel (VMT).
Calculate the coefficient of determination (R²) and plot the relationship between the observed and estimated trips (or trip rates) by districts. An example scatterplot is shown in Figure 3-4. The geographic level at which this test is performed depends on the number of observations per district; comparisons at the TAZ level would not be possible unless the sample size was very large. The use of total trips by district is a "biased" validation measure in the sense that large zones produce a lot of trips, small zones produce fewer trips. Thus, the resulting R² is measuring zone size. A better measure would be to calculate the observed and estimated average household trip rates at the zonal or district level and compare these values. This comparison is a better indicator of model performance, even though it results in lower values for R².
Compare observed and estimated trips produced at the household (disaggregate) level. Apply the model for each household in the survey to estimate trips produced (e.g. each 1-person, low-income household will produce 0.57 HBW trips, etc.) Compare the estimated trips with the observed number of trips by household (e.g. HH#1 has 0 HBW trips, HH#2 has 2 trips, etc.),

Table 3-7
Aggregate Trip Generation Checks
Albuquerque Travel Model Summaries for 1992

Item	Surveyed Range of Values¹	Modeled Value²
Home-Based Work	357,538-388,110	385,001
Home-Based School	230,658-250,382	232,441
Home-Based Shop	210,703-228,719	274,639
Home-Based Other	658,406-714,704	858,194
Non-home-Based Work-Related	197,819-220,845	272,132
Non-home-Based Other	368,079-410,925	506,352
Total Internal-Internal Trips	2,052,772-2,184,116	2,528,759
Total Trips per Person	3.64-3.87	4.49
HB Work Trips per Employee	1.29-1.40	1.39
HB Shop Trips per Retail Employee	4.15-4.50	5.40
¹ Surveyed range of person trips based on measured sample error from 1992 household survey ² Modeled person trips include increases to trips to match regional VMT Source: Barton-Aschman Associates, Inc.

Sensitivity of trips per capita can be checked relative to changes in average household size, workers per household, income level, and auto ownership (not all of these variables would be included in a single model). Such an analysis involves relating variations in individual independent variables (inputs) to the resulting changes in the dependent variable (output).

Figure 3-3 shows the number of trips per capita that have been surveyed in several cities over a number of years. As can be seen, there is a general trend that the number of trips per capita is increasing over time, albeit at a decreasing rate. One factor to consider is that surveys may also have improved (or changed) over the same time period in terms of capturing more of trips made by each household. The use of activity-based surveys may further increase the portion of trips recorded by the survey respondent.

While examination of trip rate trends might be considered an aggregate data check, it can also form the basis for a sensitivity check. For example, if the trip generation model for a future year results in static or decreasing trips per capita compared to the base year, concerns may be raised about the sensitivity of the trip generation model to the factors driving the increase in per capita trip-making.

Since observed data is not available for future years, validation of forecast models can rely as heavily on qualitative measures as quantitative ones. For example, are trip generation rates increasing, and is this trend consistent with household composition and income?

Figure 3-3
Trips per Capita - Selected U.S. Cities (from course materials)

3.3 Trip Attractions

Model Description
The trip attraction model is used to predict the trip ends which are associated with the non-home end of the trip. The same trip purposes that are used for the trip production models are used for the trip attraction models. Two different approaches can be used to calibrate trip attraction models similar to those used for trip production models. The first method is to develop regression equations which relate the trip attractions to a number of explanatory variables such as population, households, employment, density, and school enrollment. A second method is to estimate regional trip-attraction rates, stratified by land use or employment category.

Attraction models are typically developed from the same household travel survey used to calibrate the trip production model. Data limitations are often a problem with trip attraction models. While household travel surveys provide excellent data for production models on the location, nature, and trip-making characteristics of households, much less information is available on activity locations. Nearly all household surveys are too small to provide stable zone level attraction data. As a result, attraction models are typically developed with regression equations using data aggregated to large districts. Zone-level calibration is more realistic if an establishment survey of major trip attractors has been conducted.

Validation Tests
Validation of trip attraction models should use the same basic procedures as for the trip production models. Trip attraction rates should be reviewed for reasonable relationships and compared with other areas. The following rates should be reviewed:

Home-based work person trip attractions per total employment
Home-based school trips per school enrollment
Home-based shop trips per retail employment

The trip attraction models can be applied to zonal input data to estimate trip attractions in the base year. A comparison of observed and estimated trips should be made at either the district or county level.

3.4 Special Generators

Special attention should be paid to identifying the location and magnitude of activity associated with major trip generators, including CBDs, shopping malls, suburban activity centers, hospitals, government installations such as military bases, airports, and colleges and universities. It is likely that some of these should be represented in the modeling system as special generators, particularly military bases, airports, and colleges and universities. These are major land uses for which the standard trip generation and distribution models are not expected to provide reliable estimates of their travel patterns.

Two sources of data against which to check trips produced at activity centers include local trip generation surveys and ITE trip generation rates⁽¹⁾. Local surveys, such as traffic impact studies, often provide detailed driveway traffic counts and may include occupancy information. ITE trip generation rates are classified by land use type. Since both of these sources give estimates of vehicle trips, these should be converted to person trips using an average auto occupancy.

3.5 Modeling Trips for Other Purposes

The previous sections have described trip generation for trips made internal to the study area by residents of the study area. In addition to those trips, trips for several other purposes need to be accounted for in the modeling of travel for the region. These trips include:

truck trips or those trips made by commercial vehicles in the region,
non-resident trips or trips made by non-residents of the modeling area while they are visiting the study area,
internal-external trips or trips made by residents and non-residents of the study area with one end inside of the study area and one end outside of the study area, and
external-external trips or those trips passing through the study area without stopping.

Truck Trips
Information on commercial vehicle travel within most regions is limited. In most regions where truck traffic is a minor component of the vehicular traffic, truck trips are estimated by simply factoring the auto trips. Classification counts which separate traffic volumes by type of vehicle are collected for a wide range of locations in the region. The percent of truck trips can be estimated and then applied to the auto vehicle trips before assignment. An alternative to simply factoring auto vehicles to obtain truck demand is to estimate truck trips separately using truck generation, distribution, and assignment models.

Internal Trips by Non-Residents
Non-residents of a region travel into the region for many purposes. Their trips into and out of the region are accounted for as internal-external trips (see below). However, while they are in the region, they make trips that are totally internal to the region before returning to their residences outside the region. In effect, these trips are non-home-based trips by non-residents of the region. In other areas with a great deal of travel made by tourists, a separate model is calibrated. However, for the most regions, a simple accounting of non-resident travel should be sufficient.

If overall trip generation appears low based on internal and external trip purposes, the non-home-based trips can be factored to reflect those trips made by non-residents. A simple factoring procedure is based on the assumption that the ratio of internal NHB trips to external-internal trips made by non-residents is equal to the ratio of NHB trips to home-based trips made by residents. This is expressed in the following equation where numerator should include the sum of all of the non-home based trips, both work-related and other purposes. The denominator should include the sum of all home-based trips for the residents of the region.

Non-residents of the region are assumed to behave like residents of the region for their non-home-based trip making. The non-home-based trip ratio for non-residents is set equal to the rate derived for residents shown above. Thus, the number of non-home-based trips made by non-residents of the region can be estimated using the following equation:

where:

NHB_nr is the non-home-based trips made by non-residents

NHB_rate is the non-home-based ratio made by residents

IX_nr is the proportion of total internal-external trips attributed to non-residents

IX is the total number of internal-external trips

This procedure is best illustrated by an example. In typical urban area models, NHB trips made by residents equal about 25 percent of total trips. Percentages can be substituted for trips in the above equation for NHB_ratio without changing the results.

Therefore the ratio of NHB to HBNW would be:

The equation for NHB trips made by non-residents is factored to convert vehicle trips to person trips. Discounting the auto occupancy rates and assuming that the proportion of total internal-external, external-internal trips (IX_nr) made by non-residents is about 90%, then the following equation would be used to compute internal non-home based trips made by non-residents:

While this calculation estimates a rate of 0.30, the final factor will be based on the calibration of the entire model set and will be based on the match of assigned volumes to observed count data. If the overall assigned volumes are consistently below the count data then this factor can be adjusted upward. Conversely, it can be reduced if the model over-assigns travel.

External Trip Generation
Internal-external trips have one trip end outside of the cordon and are modeled as vehicle trips. For the base year, the control total for a given external station is the daily traffic volume after through traffic has been subtracted out.

External-external, or through, trips have both trip ends outside of the study area. These trips are also modeled as vehicle trips. There is no trip generation model for this purpose since both ends of the trip occur outside of the area being modeled. An origin-destination matrix for the through vehicle trips is developed using the external cordon survey and is added to the other internal-based vehicle trips before traffic assignment.

The total number of external-based trips comes directly from daily counts at the external cordon. However, in many metropolitan areas, limited data are available on the percentage of cordon traffic that are through trips and the origin-destination movements of external trips. As a result, through trip percentages may be adjusted during validation of the assignment results in order to match observed traffic count volumes.

3.6 Balancing Productions and Attractions

The last step in trip generation modeling is the balancing of regional trip productions and attractions. The regional total of trip productions must be equal to the total of trip attractions for each trip purpose in order to apply the gravity model in the trip distribution step.

The estimated total trips produced at the household level should be equal to the total trips attracted at the activity centers. Each trip must have two ends, a production and an attraction. In reality, the estimation of trip productions and attractions will not be exactly equal. While trip production and attraction rates may contribute to the imbalance, the majority of the difference can be explained by the estimation of the number of households, the socioeconomic characteristics of the households, and the estimation of the number of employees by type.

The ratio of regionwide productions to attractions by purpose should fall in the range of 0.90 to 1.10 prior to balancing. If this is not the case, then socioeconomic data and trip rates should be reviewed again.

To bring the regional totals in balance, either the zonal productions or attractions are scaled to equal regional control totals. In the majority of cases, the control totals of trips are the regional totals of trip productions by purpose. This is due to the fact that we generally have a greater degree of confidence in household data than we do in employment data. This is particularly true when a home interview survey serves as the base for developing the trip production rates. The 100% inventory of households is used to develop the number of households by zone. The employment data from which the attractions are computed are less certain, not only on a regional basis, but more critically, at the traffic analysis zone level of geography. Although some regions have collected a complete inventory of employment, the trip attraction rates are usually calibrated from household travel survey data when no workplace survey is collected.

The exception is for non-home-based trips where trip attractions are used as the control totals and productions are scaled to match attraction totals. Special generators are another example where attractions would be the control total for the balancing process. If external trips are not treated as a separate purpose, then these may be held constant since the cordon line vehicle crossings serve as a control total. External-internal trips may need to be converted from vehicle to person trips if these are included with the productions and attractions.

Endnotes

1. Institute of Transportation Engineers, 1997. Trip Generation Sixth Edition. Publ. No. 1R-016D. Institute of Transportation Engineers, Washington, D.C.