Section 7.1 Highway Assignment Validation

The ARC model employs a standard equilibrium methodology to assign vehicle trips to the transportation network using the bi-conjugate Frank-Wolfe algorithm. The assignment is assumed to reach convergence when the relative gap is less than $10^{-4}$ for three successive iterations. The highway assignments use a generalized cost function that includes travel time, toll, and distance. These values are converted to cost using value-of-time (VOT) and auto operating costs. The generalized cost function is as follows:

Cost = (time * VOT) + toll cost + (distance * operating cost)

The passenger car VOT used in the assignments is $21.50 and is based on average wage rates in the Atlanta region. The truck VOT is $36.00 and was based on a review of other truck models throughout the United States. The auto operating cost is $0.1729 per mile and includes the costs associated with fuel, maintenance, and tires using information from AAA. The AAA costs are broken down in Figure 7-1. The operating cost for trucks is $0.5360 per mile based on a fuel cost of $3.34 per gallon and a fuel efficiency of 6.2 miles per gallon. The truck costs were obtained from the American Transportation Research Institute (http://atri-online.org/wp-content/uploads/2017/10/ATRI-Operational-Costs-of-Trucking-2017-10-2017.pdf).

Figure 7-1. Auto Operating Cost

The ARC model includes highway assignments split into five time of day periods as follows:

Early AM = 3:00am to 6:00am
AM Peak = 6:00am to 10:00am
Midday = 10:00am to 3:00pm
PM Peak = 3:00pm to 7:00pm
Evening = 7:00pm to 3:00am

Each assignment includes the following vehicle classes:

SOV (non-toll eligible)
HOV 2 car (non-toll eligible)
HOV 3+ car (non-toll eligible)
SOV (toll eligible)
HOV 2 car (toll eligible)
HOV 3+ car (toll eligible)
Commercial vehicle
Medium Truck
Heavy Truck: I-285 by-pass
Heavy Truck: remaining

Section 7.1.1 Free-Flow Speeds and Capacities

Prior to the highway assignment, network link free-flow speeds and capacities are calculated based on the link facility type and area type. The area types used in the model in the subsequent tables are as follows:

ATYPE1 = CBD
ATYPE2 = Urban Commercial
ATYPE3 = Urban Residential
ATYPE4 = Suburban Commercial
ATYPE5 = Suburban Residential
ATYPE6 = Exurban
ATYPE7 = Rural

For free-flow speeds, data from the FHWA’s National Performance Management Research Data Set (NPMRDS) from 2013 were joined to ARC’s network using the Traffic Message Channels (TMCs). Average free-flow speeds were initially calculated by facility type and area type and were then modified such that the speeds for a given facility type increase as the area types transition from urban to rural. In cases where the facility type or area type did not include data, the free-flow speeds were asserted based on similar facilities. The resulting free-flow speed lookup table is provided in Table 7-1. In addition to the lookup table, several other characteristics determine the final free-flow speed including:

Ramps identified as “loop” ramps - free-flow speed set to 35 mph
Principal arterial speeds varied by number of lanes for CBD area types
Links with observed speed - free-flow speed is computed as the average of the observed early AM speed and the look table speed

Table 7-1 Free-Flow Speed Lookup Table

Facility Type	NAME	ATYPE1	ATYPE2	ATYPE3	ATYPE4	ATYPE5	ATYPE6	ATYPE7
0	centroid connector	7	11	11	11	11	14	14
1	interstate/freeway	62	63	63	63	64	65	66
2	expressway	43	46	49	52	55	58	61
3	parkway	43	46	49	52	55	58	61
4	freeway HOV (concurrent)	64	65	65	65	66	67	68
5	freeway HOV (barrier separated)	64	65	65	65	66	67	68
6	freeway truck only	62	63	63	63	64	65	66
7	system to system ramp	50	50	50	55	55	55	55
8	exit ramp	35	35	35	35	35	35	35
9	entrance ramp	35	35	35	35	35	35	35
10	principal arterial	23	26	31	35	41	48	53
11	minor arterial	21	26	29	33	38	43	48
12	arterial HOV	21	26	29	33	38	43	48
13	arterial truck only	21	26	29	33	38	43	48
14	collector	17	23	24	26	30	35	45

Hourly capacities were based in part from general Level-of-Service (LOS) E Highway Capacity Manual assumptions and were then asserted such that as area types transition from urban to rural, the hourly capacities increase. Similarly, the capacities decrease from limited access facilities (e.g., interstates) to facilities with less restricted access (e.g., arterials). The LOS E hourly capacities used in the model are provided in Table 7-2.

Table 7-2 LOS E Hourly Capacities

Facility Type	NAME	ATYPE1	ATYPE2	ATYPE3	ATYPE4	ATYPE5	ATYPE6	ATYPE7
0	centroid connector	10,000	10,000	10,000	10,000	10,000	10,000	10,000
1	interstate/freeway	1,900	1,900	2,000	2,000	2,050	2,100	2,100
2	expressway	1,200	1,200	1,300	1,350	1,400	1,450	1,450
3	parkway	1,150	1,150	1,250	1,300	1,350	1,400	1,400
4	freeway HOV (concurrent)	1,900	1,900	2,000	2,000	2,050	2,100	2,100
5	freeway HOV (barrier separated)	1,900	1,900	2,000	2,000	2,050	2,100	2,100
6	freeway truck only	1,900	1,900	2,000	2,000	2,050	2,100	2,100
7	system to system ramp	1,300	1,400	1,500	1,600	1,700	1,700	1,700
8	exit ramp	800	850	850	850	850	900	900
9	entrance ramp	900	900	950	950	1,000	1,050	1,100
10	principal arterial	1,000	1,050	1,100	1,150	1,200	1,250	1,300
11	minor arterial	900	900	950	1,000	1,000	1,050	1,100
12	arterial HOV	1,000	1,050	1,100	1,150	1,200	1,250	1,300
13	arterial truck only	900	900	950	1,000	1,000	1,050	1,100
14	collector	750	800	800	850	850	900	900

During the calibration of the highway assignment, the model had difficulty in replicating observed interstate speeds near major system-to-system interchanges. This is primarily due to the fact that the user equilibrium highway assignment does not account for operational issues occurring in these segments, weaving for example, or the presence of queues. As a means to improve the model’s ability to predict congested speeds at these locations, a network attribute (WEAVEFLAG) was introduced to identify the interstate links adjacent to the major interchanges. The link capacities at these locations are subjected to a modification in the originally calculated capacity when the number of lanes is greater than four. The equation is structured as follows in those cases:

Weave section capacity = Initial Capacity * $0.98^{(lanes-1)}$

Finally, to compute the period level capacity, rather than multiply the hourly capacity by the number of hours in a time period, adjustments were made to reflect the peaking that occurs within the modeled time periods. These period level adjustments were made based on available GDOT hourly traffic count data and are as follows:

Early AM = 1.66
AM Peak = 3.66
Midday = 4.70
PM Peak = 3.66
Evening = 3.91

Section 7.1.2 Volume Delay Functions

The calibration of the highway assignment included updating the volume delay functions (VDF curves). These curves are a modified version of the BPR function with coefficients that vary by facility type. The general formula for the VDF curves is as follows:

$T_c$ = T0 * A * V/C + D * ($V/C^{B}$)

where,

$T_c$ = congested time
T0 = free-flow time
V/C = volume to capacity ratio
A, B, D = calibrated coefficients, see Table 7-3

Graphical representation of the VDF curves is provided in Figure 7-2.

Table 7-3 VDF Curve Parameters

Facility Type	A	B	D
Freeway Basic	0.1	6.0	0.60
Freeway Weave	0.2	5.5	1.25
Expressway	0.0	4.0	1.00
Parkway	0.0	4.0	1.25
Ramp	0.1	4.0	1.00
Principal Arterial	0.1	4.0	0.45
Minor Arterial	0.1	4.0	0.45
Collector	0.1	4.0	0.45

Figure 7-2. VDF Curves

Section 7.1.3 Vehicle Miles Traveled

Validation of the highway assignment results included comparisons of regional vehicle miles traveled (VMT). To compare regional VMT, GDOT HPMS summaries of average annual daily traffic (AADT) were summarized by functional classification for each county. As the model is designed to estimate an average weekday, the AADT-based VMT was converted to represent average weekday VMT by reviewing data from GDOT permanent count stations which continuously record traffic data. This resulted in the following conversion factors:

13-county interstates = 1.03
13-county non-interstate = 1.07
8-county interstates = 1.003
8-county non-interstate = 1.065

After converting the GDOT HPMS VMT at the county-level, the observed and estimated data were compared at the regional level as shown in Table 7-4. While the data is provided for both collectors and local roads, the network does not include all of these facilities in the region which explains the large differences in VMT. However, when viewing interstates, principal arterials, and minor arterials, the model is within 5% of the observed VMT.

Table 7-4 Observed vs. Estimated Regional VMT

FUNCTIONAL CLASSIFICATION	GDOT 2015 (AWDT)	MODEL 2015	PERCENT DIFFERENCE
Interstate	57,399,000	54,422,000	-5%
Principal Arterial	27,417,000	26,377,000	-4%
Minor Arterial	34,652,000	34,277,000	-1%
Collector	12,960,000	11,242,000	-13%
Local	48,488,000	11,341,000	-77%
Total	180,916,000	137,659,000	-24%
Arterial and Above	119,468,000	115,076,000	-4%
Collector and Above	132,428,000	126,318,000	-5%

Section 7.1.4 Traffic Counts

GDOT maintains an extensive traffic counting program which includes daily volume counts, vehicle classification counts, and hourly counts. These count locations were joined to the model network in more than 5,000 locations. For the daily volume counts, the same conversion factors used for the AADT to AWDT VMT calculations were applied to the observed counts for comparing against the model estimates. The daily counts were compared against the model in several ways including:

Volume groups
Facility type
Area type

The statistical summaries for these comparisons included the RMSE, % RMSE, and volume-to-count ratios. The analysis is provided in Tables 7-5, 7-6, and 7-7 below. As shown in these tables, the model matches GDOT counts very well at the regional level, particularly for higher volume roadways such as interstates and principal arterials. As expected, the model is less accurate for lower volume roadways. The region-wide % RMSE across all facilities is 38%; however, when calculating the % RMSE for facilities with > 5,000 vehicles per day, it drops to 18%.

Table 7-5 Highway Validation Statistics by Volume Group

Volume Group	Observations	Observed Counts	Estimated Volumes	RMSE	%RMSE	Volume/Count Ratio
< 2,500	3,283	3,981,000	4,979,000	1,421	117%	1.25
2,500 - 4,999	2,109	7,736,000	7,967,000	2,105	57%	1.03
5,000 - 9,999	2,681	19,120,000	18,710,000	3,161	44%	0.98
10,000 - 24,999	2,300	35,163,000	31,083,000	4,891	32%	0.88
25,000 - 49,999	365	12,389,000	11,715,000	7,918	23%	0.95
50,000 - 74,999	172	10,913,000	10,421,000	10,202	16%	0.95
75,000 - 99,999	103	8,793,000	8,506,000	10,373	12%	0.97
>= 100,000	104	12,806,000	12,705,000	10,697	9%	0.99
Total	11,117	110,901,000	106,086,000	3,812	38%	0.96

Table 7-6 Highway Validation Statistics by Facility Type

Facility Type	Observations	Observed Counts	Estimated Volumes	RMSE	%RMSE	Volume/Count Ratio
Interstate / Freeway	786	42,472,000	41,428,000	8,360	15%	0.98
Principal Arterial	1,491	21,190,000	20,364,000	4,902	34%	0.96
Minor Arterial	3,828	26,485,000	25,644,000	2,999	43%	0.97
Collector	3,929	9,170,000	7,933,000	1,822	78%	0.87
Ramps	1,083	11,585,637	10,717,384	4,641	43%	0.93

Table 7-7 Highway Validation Statistics by Area Type

Area Type	Observations	Observed Counts	Estimated Volumes	RMSE	%RMSE	Volume/Count Ratio
Area Type 1	482	11,613,000	11,028,000	5,909	25%	0.95
Area Type 2	797	12,680,000	11,636,000	4,869	31%	0.92
Area Type 3	1,470	18,159,000	16,747,000	4,633	38%	0.92
Area Type 4	1,845	22,556,000	21,400,000	4,285	35%	0.95
Area Type 5	3,955	36,617,000	34,943,000	3,556	38%	0.95
Area Type 6	1,236	5,434,000	5,893,000	2,213	50%	1.08
Area Type 7	1,332	3,844,000	4,440,000	1,805	63%	1.16

The observed and estimated daily volumes were also graphed using a scatterplot which is provided in Figure 7-3. As illustrated, the correlation coefficient of 0.95 and the trendline indicate the model generally reasonably estimates daily volumes when compared to observed counts.

Figure 7-3. Daily Estimated Volumes vs. Observed Counts

Hourly data from GDOT’s permanent count stations were aggregated to match the model’s five time period assignments. This comparison is provided in Table 7-8 and shows the model estimated volumes match period-level counts reasonably well. The model does appear to be underestimating travel in the evening time period as evidenced by the volume-to-count ratio of 0.80 for that time period. However, it should be noted that there were only 78 permanent count stations which makes it difficult to draw concrete conclusions that warrant modifications to the model. Scatterplots for each time period were also prepared and are provided in Figures 7-4 through 7-8.

Table 7-8 Highway Validation Statistics by Time of Day

Time Period	Observed Counts	Count % Share	Estimated Volumes	Volume % Share	Volume/Count Ratio
Early AM	399,580	5%	379,100	5%	0.95
AM Peak	1,758,715	23%	1,820,900	25%	1.04
Midday	2,089,199	27%	1,900,600	26%	0.91
PM Peak	1,997,101	26%	1,982,400	27%	0.99
Evening	1,417,771	19%	1,130,200	16%	0.80
Total	7,662,366	100%	7,213,200	100%	0.94

Figure 7-4. Early AM Estimated Volumes vs. Observed Counts

Figure 7-5. AM Peak Estimated Volumes vs. Observed Counts

Figure 7-6. Midday Estimated Volumes vs. Observed Counts

Figure 7-7. PM Peak Estimated Volumes vs. Observed Counts

Figure 7-8. Evening Estimated Volumes vs. Observed Counts

In addition to comparing the model against all vehicle counts, GDOT’s vehicle classification percentages were added to the highway network. These percentages were then multiplied by the total vehicle counts to compute the respective truck volumes and subsequently compared against the model estimates by facility type, area type, and inside/outside I-285. The validation summaries are provided below in Tables 7-9 through 7-11.

GDOT’s available data include more locations with an overall truck percentage than locations that split the truck percentage into medium and heavy duty trucks. Generally, the model’s estimation of total truck traffic, medium trucks, and heavy trucks matches the observed data as evidenced by the overall volume-to-count ratios. When viewing by area type, the model appears to overestimate medium and heavy trucks in the CBD; however, there are relatively few count locations which included the two truck types in this area type. As shown in Table 7-11, the model appears to be estimating the heavy truck traffic inside I-285 reasonably well which is important given that heavy trucks without a destination inside I-285 are prohibited from using I-75/I-85 to travel through the city of Atlanta.

Table 7-9 Truck Validation Summaries by Facility Type

	All Truck Count Locations		Medium and Heavy Truck Count Locations
Facility Type	Observations	Volume / Count Ratio	Observations	Medium Truck Volume / Count Ratio	Heavy Truck Volume / Count Ratio
Interstate	84	1.19	72	1.51	0.88
Expressway / Parkway	106	1.10	64	0.88	0.83
Ramps	20	0.77	3	1.24	0.96
Arterials	3,953	0.91	1,042	0.92	1.04
Collectors	1,903	0.65	214	0.84	1.68
Total	6,066	0.96	1,395	1.10	0.94

Table 7-10 Truck Validation Summaries by Area Type

	All Truck Count Locations		Medium and Heavy Truck Count Locations
Area Type	Observations	Volume / Count Ratio	Observations	Medium Truck Volume / Count Ratio	Heavy Truck Volume / Count Ratio
1	173	1.14	23	1.57	1.66
2	393	0.99	92	1.13	0.98
3	758	1.00	200	1.16	0.93
4	903	1.00	203	1.20	1.05
5	2,181	0.92	471	1.03	0.89
6	832	0.89	210	0.98	0.93
7	826	0.94	196	0.96	0.81

Table 7-11 Truck Validation Summaries Inside/Outside I-285

	All Truck Count Locations		Medium and Heavy Truck Count Locations
Location	Observations	Volume / Count Ratio	Observations	Medium Truck Volume / Count Ratio	Heavy Truck Volume / Count Ratio
Outside I-285	5,209	0.96	1,247	1.10	0.93
Inside I-285	857	0.93	148	1.07	1.05

In addition to the above summaries, scatterplots were generated for all trucks, medium trucks, and heavy trucks to illustrate graphically the comparison between the estimated versus observed truck volumes. These plots are presented in Figures 7-9 through 7-11 below. The correlation coefficients are all higher than 0.85, with the heavy truck resulting in 0.95. When viewing Figure 7-10, the model does appear to be overestimating medium truck as evidenced by the trendline above the 45-degree line; however, when comparing the total truck volumes in Figure 7-9, this observation is not apparent. As noted previously, the number of observed locations which included medium truck percentages was much lower than the number of locations which included total trucks which could be skewing the medium truck results. That being said, in future updates to the truck model, another review of the medium truck component is likely warranted.

Figure 7-9. Estimated vs. Observed All Trucks

Figure 7-10. Estimated vs. Observed Medium Trucks

Figure 7-11. Estimated vs. Observed Heavy Trucks

Given the planned expansion of the express toll lane system, the model estimated volumes in the I-85 North Express Lanes were compared against observed data from the State Road and Tollway Authority (SRTA), which represent an average weekday from October 2015. The volumes were aggregated by segments, AM peak period, PM peak period, off-peak periods, and by direction of travel. During the assignment validation, an iterative process was used which involves running CT-RAMP, running assignments, and optimizing tolls. A graphical representation of this process is provided in Figure 7-12. Note that CT-RAMP must be run initially to develop toll/non-toll eligible trip tables (outer loop). Those trip tables are then assigned to the network and toll modifications are made within the inner loop where the assignment is run multiple times with the same trip tables. Once the toll volumes appear reasonable in the inner loop, the highway skims are rebuilt and CT-RAMP is run again to generate another set of trip tables. This process is continued until a balance between the toll rates between mode choice and assignment is reached.

The resulting model estimated toll volumes compared to the observed data are provided in Table 7-11 and via scatterplot in Figure 7-13. Generally, the model matches the AM and PM peak periods well, while the model tends to overestimate the off-peak periods. This is a primary function of two things:

In reality, the tolls are dynamically priced in small time intervals while the model toll rates are based on a single period spanning multiple hours.
Very little congestion occurs during the off-peak periods. As a result, the travel time savings between the express lanes and general purpose lanes is small and during the equilibrium assignment, even incremental changes to toll rates can result in large swings in express lane volumes under these conditions. Toll rates that are set too high yield zero toll trips in the lanes, which was avoided.

Figure 7-12. Toll Optimization Routine

Table 7-11 I-85 Express Lane Observed vs. Estimated Volumes

Segment	Period	Direction	Observed	Modeled
I-285 to Pleasantdale Rd	AM	NB	460	570
Pleasantdale Rd to Jimmy Carter Blvd	AM	NB	820	710
Jimmy Carter Blvd to Indian Trail Rd	AM	NB	870	1,090
Indian Trail Rd to Pleasant Hill Rd	AM	NB	740	810
Pleasant Hill Rd to Old Peachtree Rd	AM	NB	310	410
I-285 to Pleasantdale Rd	AM	SB	3,910	4,320
Pleasantdale Rd to Jimmy Carter Blvd	AM	SB	5,400	4,920
Jimmy Carter Blvd to Indian Trail Rd	AM	SB	5,060	4,690
Indian Trail Rd to Pleasant Hill Rd	AM	SB	2,710	3,030
Pleasant Hill Rd to Old Peachtree Rd	AM	SB	1,260	1,120
I-285 to Pleasantdale Rd	PM	NB	3,900	4,560
Pleasantdale Rd to Jimmy Carter Blvd	PM	NB	4,950	4,870
Jimmy Carter Blvd to Indian Trail Rd	PM	NB	4,660	4,320
Indian Trail Rd to Pleasant Hill Rd	PM	NB	3,890	4,460
Pleasant Hill Rd to Old Peachtree Rd	PM	NB	2,020	1,450
I-285 to Pleasantdale Rd	PM	SB	1,270	1,310
Pleasantdale Rd to Jimmy Carter Blvd	PM	SB	1,820	1,760
Jimmy Carter Blvd to Indian Trail Rd	PM	SB	1,750	1,830
Indian Trail Rd to Pleasant Hill Rd	PM	SB	930	1,110
Pleasant Hill Rd to Old Peachtree Rd	PM	SB	480	250
I-285 to Pleasantdale Rd	OP	NB	1,860	2,480
Pleasantdale Rd to Jimmy Carter Blvd	OP	NB	3,110	4,220
Jimmy Carter Blvd to Indian Trail Rd	OP	NB	3,180	4,860
Indian Trail Rd to Pleasant Hill Rd	OP	NB	2,770	4,100
Pleasant Hill Rd to Old Peachtree Rd	OP	NB	1,340	1,330
I-285 to Pleasantdale Rd	OP	SB	2,100	3,390
Pleasantdale Rd to Jimmy Carter Blvd	OP	SB	3,080	4,230
Jimmy Carter Blvd to Indian Trail Rd	OP	SB	2,990	4,500
Indian Trail Rd to Pleasant Hill Rd	OP	SB	1,660	4,040
Pleasant Hill Rd to Old Peachtree Rd	OP	SB	1,050	1,160

Figure 7-13. I-85 Express Lanes Estimated vs. Observed Volumes

Section 7.1.5 Speeds

The highway assignment validation also included comparisons to observed NPMRDS speeds where available. The raw speed data was averaged to reflect the ARC time-periods and joined to the highway network. The results are provided in the scatterplots below in Figure 7-14 through Figure 7-18. As expected, the model matches the speeds in the off-peak periods well, with the correlation coefficients ranging from 0.87 to 0.91 in the early AM, midday, and evening periods. The AM and PM peak period speeds do not match as closely with the observed speeds which is primarily a function of the static user equilibrium assignment procedures which cannot account for operational characteristics such as signal timing, merge/diverge/weaving, and queue formations that exist in real world conditions. However, in the context of a regional planning model, ARC’s assignment matches the AM and PM peak speeds reasonably well as evidenced by the plots.

Figure 7-14. Early AM Observed vs. Estimated Speeds

Figure 7-15. AM Peak Observed vs. Estimated Speeds

Figure 7-16. Midday Observed vs. Estimated Speeds

Figure 7-17. PM Peak Observed vs. Estimated Speeds

Figure 7-18. Evening Observed vs. Estimated Speeds

Section 7.2 Transit Assignment Validation

The ARC model assigns transit trips to the network using the Public Transport (PT) module in Cube. The trips are assigned to the single best path using PT’s algorithms for each time period, mode of access, and premium only vs. premium/non-premium transit modes. The assignment results are then aggregated to daily totals for comparison against data provided by the regional transit operators. The overall summary of total boardings by operator is provided in Table 7-12. As shown, the model matches total regional boardings well. The model tends to overestimate suburban transit providers but is within 7% overall for MARTA rail and 9% for MARTA buses, which account for approximately 85% of total regional transit ridership.

Table 7-12 Regional Transit Boardings

Operator / Mode	Observed	Modeled	Difference	% Difference
MARTA Rail	230,940	247,040	16,100	7%
MARTA Bus	201,370	184,100	-17,270	-9%
GRTA	6,370	5,440	-930	-15%
CCT	11,660	15,960	4,300	37%
GCT	6,430	7,900	1,470	23%
HAT	570	1,530	960	168%
CATS	120	130	10	8%
Shuttles	46,300	41,090	-5,210	-11%
Total	503,760	503,190	-570	0%

Section 7.2.1 MARTA Rail

The MARTA rail entries and boardings were further summarized at the station level. For the purposes of this documentation, entries represent patrons that enter the MARTA rail system at a given station. This is an important designation, particularly at Five Points, where a person can enter the system or transfer between MARTA lines. Boardings account for both entries and transfers between lines. The resulting entries and boardings by station are provided in Table 7-13. In the table, the difference between entries and total boardings is clearly shown at Five Points, where the observed entries are approximately 19,000 while the observed boardings are nearly 59,000. At most stations, transferring between lines is not possible, in which case the entries and boardings are identical.

While in some cases the percentage differences might appear large, these differences should be viewed in the context of the overall station activity. For example, Garnett Station shows a difference of 58%; however, the observed entries/boardings at this location are less than 2,000. Another way to view the results are by means of a scatterplot similar to the highway assignment validation results. The station entries are provided graphically in Figure 7-19. As shown, when viewing in this manner, the model matches the observed data well with a correlation coefficient of 0.93. Finally, a bar graph of the station boardings is provided in Figure 7-20. Since the total at Five Points is much higher than the other stations, it was removed from the graph to provide a closer look at the remaining stations.

Table 7-13 MARTA Rail Station Entries and Boardings

Station Name	Observed Entries	Estimated Entries	Difference (Entries)	% Difference (Entries)	Observed Boardings	Estimated Boardings	Difference (Boardings)	% Difference (Boardings)
NORTH SPRINGS	7,040	7,250	210	3%	7,040	7,250	210	3%
SANDY SPRINGS	2,760	2,760	0	0%	2,760	2,790	30	1%
DUNWOODY	4,230	5,500	1,270	30%	4,230	5,530	1,300	31%
MEDICAL CENTER	1,790	2,990	1,200	67%	1,790	2,990	1,200	67%
BUCKHEAD	3,730	3,050	-680	-18%	3,730	3,050	-680	-18%
DORAVILLE	5,980	6,300	320	5%	5,980	6,320	340	6%
CHAMBLEE	3,860	4,100	240	6%	3,860	4,110	250	6%
BROOKHAVEN	2,480	1,820	-660	-27%	2,480	1,820	-660	-27%
LENOX	3,990	2,700	-1,290	-32%	3,990	2,700	-1,290	-32%
LINDBERGH CENTER	8,650	10,140	1,490	17%	10,400	13,730	3,330	32%
ARTS CENTER	7,100	8,990	1,890	27%	7,100	9,030	1,930	27%
MIDTOWN	6,490	8,390	1,900	29%	6,490	8,440	1,950	30%
NORTH AVENUE	5,930	5,770	-160	-3%	5,930	5,800	-130	-2%
CIVIC CENTER	2,880	4,650	1,770	61%	2,880	4,670	1,790	62%
PEACHTREE CENTER	9,660	7,790	-1,870	-19%	9,660	7,810	-1,850	-19%
FIVE POINTS	19,120	20,320	1,200	6%	58,820	66,650	7,830	13%
GARNETT	1,720	2,720	1,000	58%	1,720	2,720	1,000	58%
WEST END	7,840	8,000	160	2%	7,840	8,000	160	2%
OAKLAND CITY	5,150	4,450	-700	-14%	5,150	4,500	-650	-13%
LAKEWOOD	2,760	2,630	-130	-5%	2,760	2,650	-110	-4%
EAST POINT	5,650	6,980	1,330	24%	5,650	6,980	1,330	24%
COLLEGE PARK	10,700	9,520	-1,180	-11%	10,700	9,540	-1,160	-11%
AIRPORT	11,470	11,300	-170	-1%	11,470	11,300	-170	-1%
HAMILTON E HOLMES	7,030	7,510	480	7%	7,030	7,520	490	7%
WEST LAKE	1,600	1,610	10	1%	1,600	1,670	70	4%
BANKHEAD	1,910	2,110	200	10%	1,910	2,110	200	10%
ASHBY	1,980	2,150	170	9%	2,640	2,470	-170	-6%
VINE CITY	1,010	1,340	330	33%	1,010	1,520	510	50%
DOME/GWCC/PHILLIPS	3,120	1,430	-1,690	-54%	3,120	1,540	-1,580	-51%
GEORGIA STATE	4,060	5,760	1,700	42%	4,060	5,790	1,730	43%
KING MEMORIAL	1,180	1,330	150	13%	1,180	1,330	150	13%
INMAN PARK	2,890	3,320	430	15%	2,890	3,370	480	17%
EDGEWOOD	1,190	1,830	640	54%	1,190	1,880	690	58%
EAST LAKE	1,350	1,410	60	4%	1,350	1,410	60	4%
DECATUR	4,000	3,510	-490	-12%	4,000	3,510	-490	-12%
AVONDALE	4,310	3,880	-430	-10%	4,310	3,900	-410	-10%
KENSINGTON	6,130	5,240	-890	-15%	6,130	5,240	-890	-15%
INDIAN CREEK	6,090	5,400	-690	-11%	6,090	5,400	-690	-11%
Total	188,830	195,950	7,120	4%	230,940	247,040	16,100	7%

Figure 7-19. Observed vs. Estimated MARTA Rail Entries

Figure 7-20. Observed vs. Estimated MARTA Rail Boardings

Section 7.2.2 Buses

As previously mentioned, the model matches overall regional bus totals well, but a review of the individual bus routes was also performed by preparing scatterplot comparisons. Two scatterplots were developed which include all buses except shuttles (Figure 7-21) and one of just MARTA buses (Figure 7-22) given it is the largest transit operator in the region. While shuttle buses are included in the transit network, in many cases, these are university shuttles transporting students in and around university campuses. In these instances, the regional model does include the same level of detail as exists in reality. For example, in the model, a university is likely to be located in one TAZ and cannot reflect the intra-campus connectivity that occurs. For these reasons, the shuttles were not included in the route-level analysis.

As shown in the figures, the model matches the regional and MARTA route level bus boardings reasonably well as evidenced by the trendline generally along the 45-degree line and correlation coefficients of 0.80 and 0.78, respectively.

Figure 7-21. Observed vs. Estimated Regional Buses

Figure 7-22. Observed vs. Estimated MARTA Buses

Section 7 - Trip Assignment