Driver Turn-Taking Behavior

in Congested Freeway Merges

Michael J. Cassidy and Soyoung Ahn

Data from four merge locations in northern California and Toronto, A merge like that shown in Figure 1a has a capacity, µ, that can
Ontario, Canada, unveil a notable feature of driver turn taking. It was be realized when queues reside on one or both of its approaches and
observed that queued vehicles from the on-ramp and freeway traffic there is no exogenous flow restriction from downstream. This max-
streams entered a congested merge in some (nearly) fixed ratio that was imum outflow is described with the capacity line in Figure 1b. The line
independent of the merge outflow. Drivers in competing traffic streams indicates that merge capacity falls below µ only when the demand
thus entered the merge by adopting some definite turn-taking behavior, from one approach (labeled “1” in Figure 1) is small and the other
and this behavior was not influenced by the severity of the exogenous flow approach (the on-ramp labeled “2”) lacks sufficient capacity to keep
restriction from downstream. The findings validate part of an existing the merge saturated.
theory of merging traffic and should be considered when any new such There is evidence that the capacity of a merge is influenced by the
theories are developed. traffic states on one or both of its approaches in such a way(s) that the
simple form for the capacity line in Figure 1b may not adequately
describe the merge discharge mechanism (3, 4). This issue lies beyond
This paper concerns the manner in which two competing traffic streams the scope of the present work, however, and is instead the subject of
enter a merge that is fully congested; i.e., a merge in which an exoge- other studies (5, 6 ).
nous queue from downstream has spilled over to both its approaches. The present study is concerned with the case in which an exogenous
Data were extracted from three such locations in California’s San queue from downstream engulfs the merge and spills over to both of
Francisco Bay Area by using video and from one site in Toronto, its approaches. Both approaches thus have a reservoir of vehicles,
Ontario, Canada, by using loop detectors. These data indicate that at such that drivers enter the merge at maximum rates (subject to the
each site drivers from the on-ramp and freeway merge together in constraints from the downstream restriction). In this case, the theory
what can reasonably be described as a definite ratio. Although the ratio assumes that the approaches supply inflows to the merge in a definite
varied across sites, its value at each site was approximately fixed and ratio; i.e., drivers enter the merge by taking turns in some regular
was independent of the congested outflow from the merge. This fashion. By specifying this ratio, the delays and queue lengths on both
means that the turn-taking behavior that determines a merge ratio is merge approaches can be predicted. (The solution comes by con-
unaffected by the severity of the exogenous flow restriction down- structing separate queuing diagrams for each approach. The depar-
stream of the merge (at least for data measured over time scales of ture rates for these diagrams are obtained by specifying the merge
2 min or more and for the range of congested flows measured here). ratio and the constrained outflow from the merge, as exemplified by
The findings are consistent with a theory of merging traffic proposed point P in Figure 1b.)
by Daganzo (1, 2). The theory itself and the general approach to test- The merge ratio is given by a line like the one shown in Figure 1b.
ing it are briefly described in the following section. The empirical evi- A straight merge ratio line was suggested by Daganzo (1) because it
dence supporting the theory is then provided. Finally, the practical is the simplest form, and, until now, data that might suggest an alter-
implications of the findings and future research needs are discussed. native form have not been available. Thus, when queues reside on
the approaches, the merge ratio is assumed to be independent of the
merge outflow; i.e., the ratio is fixed, irrespective of the severity of
BACKGROUND the exogenous restriction. The same ratio is presumed to occur when
the merge is an active bottleneck, i.e., when both of its approaches are
The need for theory to predict traffic conditions at merges is indis- queued and no exogenous restriction exists at all. The ratio is given
putable, since merges are such common freeway elements and the as µ2/µ1 in Figure 1b.
conditions that they induce can have far-reaching effects on freeway The data from the present study indicate that the merge ratio is as
systems. What may be the best-known and most rational merge theory described in the theory; i.e., driver turn-taking behavior produced a
is the one proposed by Daganzo (1, 2). A brief description of it is well-defined merge ratio that is evidently fixed. These data were taken
given below. from four fully congested merges. Although all four sites are junctions
formed by a freeway and an on-ramp, each has a distinct geometry
and thus exhibited a merge ratio that was different from the others.
M. J. Cassidy, 416C McLaughlin Hall, and S. Ahn, 416F McLaughlin Hall, Department None of the sites has a ramp meter or any other control devices to
of Civil and Environmental Engineering and Institute of Transportation Studies, influence merging behavior.
University of California, Berkeley, Berkeley, CA 94720. Congested inflows to three merge locations in the San Francisco
Transportation Research Record: Journal of the Transportation Research Board,
Bay Area were manually extracted from videos taken of the sites.
No. 1934, Transportation Research Board of the National Academies, Washington, Congested inflows from a merge location in Toronto were collected
D.C., 2005, pp. 140–147. from loop detectors installed there. In all cases, the analyses were

140
Cassidy and Ahn 141

Approach 1 (freeway) flows even fell as low as a few hundred vehicles per hour (vph) per
lane. At another site, some of the measurements were taken while
congested outflows were very near capacity. Presentation of these data
Queue 1
follows below.
2

e EMPIRICAL FINDINGS
eu
Qu
Approach 2
(on-ramp) The findings are presented in this section. To this end, bivariate plots
(a) of the inflow streams from each of the merge sites studied are pro-
vided. It is shown that these plots exhibit relatively little scatter and
note here that, with the exception of one site (discussed later), there
was no evidence of heteroscedasticity; i.e., no systematic pattern in
Flow from Approach 2 (on-ramp)

Capacity line
the residuals arose between each site’s data and the best-fit lines used
A constrained outflow line to describe these data (7 ).
Ramp It is further shown that these best-fit lines indicate that the merge
capacity ratio at each site can be described as fixed; i.e., the data can be de-
Merge ratio line
scribed with a straight line. The possibility of fitting the bivariate data
with power functions of the form y = αxβ was explored for each site’s
2
P data. These functions, however, virtually always produced estimates
of β that were not significantly different from 1. The one exception to
-1
-1 this was linked to factors exogenous to the merge, as will be explained
later in this report. For the site that did exhibit heteroscedastic data,
1
a suitably adjusted (heteroscedastically robust) standard error was
Flow from Approach 1 (freeway) used to verify the linearity of its merge ratio.
(b) Lines were fit to the data by the method of weighted least squares,
in which each datum point was assigned a weight according to the
FIGURE 1 (a) Hypothetical merge and (b) merge theory. duration of its sampling interval; data taken over longer intervals
were assigned proportionally larger weights. (As discussed below,
each datum point was extracted while the merge outflow was nearly
performed only for those data collected from periods when the con- stationary, and such periods varied in duration.)
gested merge outflow was nearly stationary. (A description of the The two merges highlighted with shading in Figure 2 formed
simple method used to distinguish stationary outflows and the ratio- along eastbound Interstate 80 by the on-ramps from Powell Street
nale for sampling data in this manner are provided in the next sec- and Ashby Avenue near San Francisco and are discussed first. The
tion.) With perhaps one exception, the data taken from each site reflect upstream merge (at Powell) has five regular-use freeway travel lanes.
a wide range of congested conditions. At one site, congested out- The downstream merge (at Ashby Avenue) has four regular-use

City of Emeryville
City of Berkeley
Powell St. Merge
Ashby Ave. Merge
X1 X2 X3 X4
Location for
measuring inflows
215 m 75 m 150 m

HOV Lane HOV Lane HOV Lane

Shoulder
lane

Powell St. University Ave.

Shoulder Ashby Ave.
on-ramp 150 m lane
Ashby Ave. off-ramp
off-ramp on-ramp

305 m 500 m 1500 m

Location for
measuring
inflows

FIGURE 2 Eastbound Interstate 80.

142 Transportation Research Record 1934

freeway travel lanes. Data from the high-occupancy vehicle (HOV) tion of an afternoon rush (on August 19, 2002). They display virtual
lane were excluded from the analyses because its traffic conditions departures at Location X2, such that the persistent vertical displace-
were generally much different from those in the adjacent lanes. (Vehi- ments between the curves verify that the freeway segment upstream
cle maneuvers into and out of the HOV lane were rare near the merge of the Powell Street merge was queued. [The reader can refer to a
locations.) previous report (8, pp. 25–46), for example, for a further discussion
The freeway stretch is part of a “laboratory” in which traffic is of virtual departure curves.]
monitored by using several video cameras mounted on the rooftop As a notable aside, the vertical displacements were made more
of a nearby 30-story building. The data used for the present study of visible to the naked eye by plotting the curves on an oblique coordi-
these two merges were manually extracted from videos shot from nate system. Because they are actually presented here in orthogonal
this rooftop. coordinates, the curves display the quantity O(t) = V(t) − q0 × (t − t0),
During each afternoon peak period, the queue from a downstream i.e., the oblique virtual vehicle count to time t, O(t), is the virtual
bottleneck engulfs these merges and spills over to their approaches. vehicle count to time t, V(t), minus a background reduction; the latter
Data were taken while merge approaches were queued. This ensured is some specified rate, q0, multiplied by the interval extending from
that inflows to each merge occurred at maximum rates (subject to the curve’s start time, t0, to t. (Further discussion of the construction
the downstream constraints), a condition necessary for the observation of these O-curves can be found in a number of reports, including
of merge ratios. The data-processing method used to verify the pres- those of Cassidy and Windover (9) and Munoz and Daganzo (10).
ence of these queues is described next. The method was applied to O-curves for the Powell Street on-ramp are presented in Figure 3b.
a portion of the data collected at the upstreammost merge at Powell Their persistent vertical displacements illustrate that the ramp was
Street. queued during the period shown in Figure 3b. Finally, the O-curves in
Figure 3a displays cumulative curves of freeway vehicle count Figure 3c verify that a short freeway segment immediately down-
versus time, t, at Locations X1 and X2 (Figure 2). These cumulative stream of the merge was also queued; i.e., a queue from a downstream
curves were constructed from the counts collected over a 10-min por- bottleneck restricted the outflows from the merge.

450 120
V(t) – qo × (t - to), qo=2000 vph

V(t) – qo × (t - to), qo=100 vph

400
100
350
X1
300 80
X1
250
60
200
X2
150 40
X2
100
20
50

0 0
17:40

17:41

17:42

17:43

17:44

17:45

17:46

17:47

17:48

17:49

17:50

17:40

17:41

17:42

17:43

17:44

17:45

17:46

17:47

17:48

17:49

17:50
t (measured at X2) t (measured at X2)
(a) (b)

450
V(t) – qo × (t - to), qo=2500 vph

400

350
X3
300

250

200

150 X4
100

50

0
17:40

17:41

17:42

17:43

17:44

17:45

17:46

17:47

17:48

17:49

17:50

t (measured at X4)
(c)

FIGURE 3 O-curves for (a) freeway approach at X 1 and X 2 , (b) on-ramp at X 1 and X 2 , and (c) at X 3 and X 4 .
Cassidy and Ahn 143

As noted above, merge ratios were sampled only when outflows 1200
from the merge were nearly stationary. This sampling method was
used to exclude data when disturbances interrupted turn taking. In
1000
general, only the data from transition periods of a minute or so that
often arose whenever merge outflows changed were excluded.

y, inflow from on-ramp (vph)

y = 1.155x
The nearly stationary periods were distinguished by visually inspect- 800 R2 = 0.976
ing curves of the cumulative vehicle count, N, measured in all (regular-
use) freeway lanes just downstream of the merge. Nearly linear
600
portions of these N-curves marked periods of nearly stationary out-
flows because the flows are equal to the slopes of piecewise linear
approximations to the N-curve. An N-curve was considered to have 400
a nearly linear trend when its deviations from a best-fit (straight) line
never exceeded a specified threshold. The threshold used in each
instance ranged from 5 to 20 vehicles, depending on the merge out- 200
non-incident data
flow at the time that the data were extracted; larger thresholds were incident data
used for higher outflows. [A similar, albeit more involved method for 0
identifying near-stationary traffic states is described by Cassidy (11)] 0 200 400 600 800 1000
As already noted, this strategy for data collection meant that the sam- x, inflow from freeway shoulder lane (vph)
pling intervals were different for each datum point. The intervals (a)
used in this work ranged from 2 to 20 min.
The findings from the present data analyses are now presented sep- 1200
arately for each site. The presentation begins with the data collected
at the Powell Street merge.
1000
y, inflow from on-ramp (vph) y = 0.185x
Freeway 80 at Powell Street R2 = 0.773
800
Attention is first focused on the turn-taking behaviors exhibited by
drivers from the Powell Street on-ramp and those from the freeway’s 600
shoulder lane (only). Figure 4a is used to this end. It is a scatter plot
of the queued inflow from the shoulder lane versus the queued on-ramp
inflow. The ratios of these inflow streams are not what have previously 400
been defined as the merge ratio, since they do not reflect drivers in
all (regular-use) travel lanes. The data in Figure 4a are notable none-
200
theless, in that they were taken from the traffic streams that compete non-incident data
most directly for entry to the merge. (Because the abscissa in Figure 4a incident data
displays measurements taken only in the freeway shoulder lane, each 0
of its datum points was extracted from a period marked by nearly 0 1000 2000 3000 4000 5000 6000
stationary outflows in that lane only.) These data are used to verify x, inflow from all regular-use freeway lanes (vph)
that driver turn-taking behavior in these streams is not influenced by (b)
merge outflow and to distinguish this finding from the noisy data that
FIGURE 4 (a) Inflows from freeway shoulder lane versus inflows
can be created by factors exogenous to the merge. from Powell Street on-ramp and (b) inflows from all regular-use
The unshaded datum points in Figure 4a were measured over a freeway lanes and inflows from Powell Street on-ramp.
period spanning 50 min during a typical (nonincident) afternoon peak
period on August 19, 2002. The blackened points were collected
over a 30-min period on March 20, 2001, while a downstream incident lane traffic streams than with the merge ratios that arise when all
created denser queuing at the merge. The incident caused outflows regular-use freeway lanes are considered.
to drop to as low as about 400 vph in the shoulder lane. (The reader Unfortunately, the reproducibility of turn-taking behavior at this
can verify this by using both axes in Figure 4a.) Powell Street merge becomes less evident by plotting the inflows from
The data in Figure 4a exhibit a linear trend. A best-fit line (a line that all five regular-use freeway lanes against the on-ramp flows. Such a
passes through the origin) generated an R2 value of .976, as annotated plot is provided in Figure 4b. The data were sampled when merge
in Figure 4a. As also annotated in Figure 4a, the slope of this line is outflows (in all regular-use lanes) were nearly stationary. The scatter
very nearly 1 (slope = 1.155). The finding indicates that on-ramp and in these data is greater than in the previous figure, and the R2 value
freeway shoulder lane drivers tend, on average, to take turns merging for the best-fit line (.773) is lower as well.
in what is very nearly a one-to-one alternating fashion. This kind of Notably, the four encircled datum points in Figure 4b lie especially
turn taking has been called the “zipper effect” (12). far from the best-fit line. One can visualize a straight line passing
Scatter plots measured at the other study sites also showed that through the origin that fits the circled datum points in Figure 4b quite
on-ramp and shoulder lane drivers exhibit reproducible turn taking; well. This indicates that the estimated merge ratio (the slope of the
the data in each plot exhibit a strictly linear trend with relatively little line) for these circled data would be markedly different from the ratio
scatter. These plots (which exclude inflows from the freeway’s inner estimated for the other data in Figure 4b. Given this difference, the
lanes) are not presented here, however, because the present study is data in Figure 4b are not best described with a straight line. This dif-
less concerned with the interaction between on-ramp and shoulder ference, however, did not arise through an endogenous change in
144 Transportation Research Record 1934

driver turn taking. To the contrary, Figure 4a previously verified that The other sites used in this work did not have nearby downstream
the ratio between the on-ramp’s inflows and those from the freeway off-ramps. The authors find it no coincidence that these other sites
shoulder lane is approximately fixed over the entire range of the yielded data without the kind of scatter seen in Figure 4b. To verify
measured conditions. It turns out that the data scatter in Figure 4b is this statement regarding scatter, attention now turns to the second of
due to factors exogenous to the merge. the four study sites.
To advance this argument, note first that the longer delays generated
by the incident (on March 20, 2001) evidently motivated a number
Freeway 80 at Ashby Avenue
of drivers to divert from the freeway via the Ashby Avenue off-ramp
downstream; this off-ramp is located about 305 m beyond the Powell The merge at Ashby Avenue is more isolated than its upstream
Street merge, as shown in Figure 2. While the incident was present, the counterpart, in that the downstream off-ramp nearest the Ashby
exit flows at this downstream ramp were nearly double those during Avenue merge is approximately 1,500 m away (Figure 2). As described
the incident-free period; the average exit rates were about 820 and below, this off-ramp had no noticeable effect on merge operations
460 vph, respectively. More revealing still is the fact that the per- upstream.
centage of merge outflow to have exited via that off-ramp jumped Traffic data were collected from this merge for a 1-h period of an
from 10% during the incident-free period to greater than 40% during afternoon peak (on March 20, 2001). Roughly half of this period was
the incident. marked by typical, nonincident conditions. The other half was marked
These higher exiting rates eased the flow constraints in the shoulder by a downstream incident. This incident created queues at the Ashby
lane upstream of the off-ramp. Consequently, the freeway queue was Avenue merge that were denser than usual. However, the incident
not first-in, first-out at the merge. Instead, the average outflow from evidently had little effect on the proportion of vehicles that discharged
the shoulder lane (measured at X3) was about 750 vph during the from the merge via the shoulder lane. During the nonincident period,
incident, as evident from both axes in Figure 4a, and this rate was the shoulder lane outflow was approximately 10% greater than that
generally about three times greater than the coinciding average of the average outflow in the three adjacent freeway lanes. During
flows measured in the freeway’s four adjacent lanes. the incident, this difference grew to 15%, a modest increase.
These marked distinctions in the shoulder lane were observed only In the absence of the substantial exogenous effects of an off-ramp,
during the incident. Given these incident effects, on-ramp flows the site’s merge ratio was reproducible and fixed (approximately)
entering the merge became disproportionately high when they were for the entire range of conditions observed. Figure 5 is a plot of the
measured relative to the inflows from all regular-use freeway lanes. queued inflow to the merge from the four regular-use freeway lanes
Consequently, all blackened datum points in Figure 4b lie above the versus queued inflow from the Ashby Avenue on-ramp. The blackened
best-fit line. Furthermore, the circled blackened points (which lie the datum points are from the incident period. The unshaded points are
farthest from the line) were measured when the proportions of vehi- from the nonincident time.
cles exiting the Ashby Avenue off-ramp were at their highest. The These data exhibit a linear trend. The best-fit line yields an R2
circled points were measured during a period when 45% of value of .955. The slope of this line (the estimated merge ratio) is
the outflow from the Powell Street merge exited via the Ashby 0.371. This is an appreciably higher ratio than that estimated for the
Avenue off-ramp. In contrast, the other blackened points in Figure 4b upstream merge at Powell Street (0.185, as annotated in Figure 4b).
were measured when 37% of the outflow used this off-ramp. The higher ratio is explained by the merge geometries.

1800

1600

1400
y, inflow from on-ramp (vph)

1200

1000

800

600

400

200

0
0 1000 2000 3000 4000 5000
x, inflow from all regular-use freeway lanes (vph)

FIGURE 5 Inflows from all regular-use freeway lanes versus inflows from Ashby
Avenue on-ramp.
Cassidy and Ahn 145

As already noted in Figure 2, the Ashby Avenue merge has one shown in Figure 6a sampled vehicle counts, along with speeds and
less freeway travel lane than its upstream counterpart. This in itself occupancies, over 20-s intervals. These detectors furnished all of the
is expected to produce a higher merge ratio at Ashby Avenue. More- data needed for the site’s analysis. In this case, the presence of per-
over, the Ashby Avenue on-ramp consists of two lanes upstream of sistent on-ramp queues was not verified by means of O-curves but,
the merge (as also shown in Figure 2). It was observed that numerous rather, by use of the occupancies measured by the detectors. All data
drivers in the ramp’s left lane entered the freeway in advance of the from this site came when the occupancies from the ramp detectors
merge; these drivers maneuvered into the freeway shoulder lane exceeded 50%, and joint measurements of occupancy and flow
upstream of the merge gore. In effect, then, the ramp very nearly served fell on the congested branch of the occupancy–flow relation.
two lanes of inflowing traffic. This generated higher merge outflows A scatter plot of nearly stationary, queued inflows is presented in
in the freeway shoulder lane and increased the merge ratio. Figure 6b. The data were taken over 6 h spanning three afternoon peak
The finding described above is notable only in that it underscores periods (on March 19, 1997, and June 6 and 22, 1998). A wide range
the fact that the merge ratio is site specific. The ratio is influenced by of inflows was observed because of multiple bottlenecks downstream
a number of factors, not the least of which is merge geometry. Data of the merge. Some of these data were collected when congested
from a third merge, with its own distinct geometry, are presented next. outflows from the merge rose well above 2,000 vph per lane. This can
be verified by summing the inflows corresponding to some of the
Gardiner Expressway at Spadina Avenue (rightmost) datum points in Figure 6b and dividing each sum by 3, the
number of freeway lanes, excluding the short acceleration lane. Some
The next site is the segment of Toronto’s westbound Gardiner Express- of these other data were taken after the merge was overrun by denser
way, shown in Figure 6a. Unlike the other study sites, the merge has queues from a more restrictive bottleneck farther downstream.
an acceleration lane. Unlike the data for other sites, the data from this The data in Figure 6b exhibit a linear trend. The best-fit line yields
merge were not obtained by using video. Rather, the loop detectors an R2 value of .942. The estimated ratio for this site (0.365) is

City of Toronto Location for

measuring inflows

570m 280m

Shoulder lane

Spadina Ave.
on-ramp Loop detectors

(a)

2000

1800

1600 y = 0.365x
y, inflow from on-ramp (vph)

1400 R2 = 0.942

1200

1000

800

600

400

200

0
0 1000 2000 3000 4000 5000 6000
x, inflow from all freeway lanes (vph)
(b)

FIGURE 6 Toronto: (a) westbound Gardiner Expressway and (b) inflows from all freeway
lanes versus inflows from Spadina Avenue on-ramp.
146 Transportation Research Record 1934

disproportionately high. (The ratio would have been 1⁄6 if the site ent study to exhibit heteroscedasticity. After this heteroscedasticity
exhibited a first-in, first-out queue along with the zipper effect.) The was addressed (by adjusting the standard error), however, a straight
high ratio arose because congested merge outflow was greater in line still provided a reasonable description of these data; the expo-
the freeway shoulder lane than in the adjacent lanes. nent in the best-fit power function was not significantly different
from 1. Although the resulting R2 value (.690) is the lowest coefficient
among the four sites studied here, it is nonetheless a respectable
State Route 24 at State Route 13
value for traffic data. The disproportionately high estimate for the ratio
The final site is the segment of eastbound State Route (SR) 24 and the (0.413) is the result of shoulder lane outflows that exceed those in
single-lane connector from SR-13 shown in Figure 7a. Like the first the adjacent lane.
two of the study sites, this merge resides in the San Francisco Bay
Area. It lies a short distance downstream of where SR-24 splits into CONCLUSIONS
two two-lane freeway segments. (The two split sections eventually
pass through separate bores of a downstream tunnel.) A video camera The data from multiple study sites indicate that when a merge is com-
was set up on a nearby hillside vantage point during two afternoon pletely queued, its inflows can be described with a ratio line of constant
peak periods (on June 2 and 3, 2004), and the data were manually slope. Drivers evidently merge by taking turns in a definite fashion,
extracted from the videos. and absent certain exogenous factors (e.g., because of a downstream
The queue from a downstream bottleneck engulfs the merge. The off-ramp), the ratio is unaffected by merge outflow.
observed range of congested inflows is not large at this site (because The findings mean that a congested merge can be modeled in a
the data were not obtained in the presence of any downstream incident, simple way. Namely, delays and queue lengths can be predicted for
for example). both approaches by specifying the merge ratio. One must also specify
This limited range of data renders the form of the merge ratio line traffic demands for the merge and the merge’s constrained outflow,
open to question. On top of this, this is the only data set in the pres- as described by Daganzo (1). The merge ratio can be estimated at a

City of Berkeley

Caldeco
tt
tunnel
Location for
measuring inflows
Shoulder lane

From SR-13
(a)

1400

1200
y, inflow from on-ramp (vph)

1000 y = 0.413x
R2 = 0.690
800

600

400

200

0
0 500 1000 1500 2000 2500 3000
x, inflow from all freeway lanes (vph)
(b)

FIGURE 7 San Francisco Bay Area: (a) eastbound SR-24 and (b) inflows from all
freeway lanes versus inflows from on-ramp.
Cassidy and Ahn 147

site by jointly measuring inflows from its approaches, as in the present ACKNOWLEDGMENT
work, i.e., when both approaches are queued. Because the ratio is
evidently fixed, one need not collect samples over a wide range of Funding for this work came from the University of California Trans-
flows to obtain an estimate. One should, however, use a sampling portation Center.
interval that is long relative to any periods marked by nonstationary
outflows brought by disturbances from downstream. REFERENCES
Whether or not a theory for prediction of the merge ratio can be
devised remains to be seen. A reviewer of the manuscript of this report 1. Daganzo, C. F. The Cell Transmission Model. Part II. Network Traffic.
offered one such theory, whereby the merge ratio for the junction of Transportation Research, Part B, Vol. 29, 1995, pp. 79–93.
a one-lane on-ramp and an n-lane freeway is taken as 1/n, i.e., the ratio 2. Daganzo, C. F. The Nature of Freeway Gridlock and How to Prevent It.
Proc., International Symposium on Transportation and Traffic Theory
of the maximum flows that may arise on the approaches when these (J. B. Lesort, ed.), Pergamon, Tarrytown, Pa., 1996, pp. 629–646.
flows are unconstrained by the merging maneuvers. This recipe does 3. Banks, J. H. Flow Processes at a Freeway Bottleneck. In Transportation
not seem to account for the turn taking that occurs between traffic from Research Record 1287, TRB, National Research Council, Washington,
the on-ramp and the freeway shoulder lane(s). Although the model D.C., 1990, pp. 20–28.
4. Cassidy, M. J., and R. L. Bertini. Some Traffic Features at Freeway Bot-
provides reasonable agreement with some of the present observations,
tlenecks. Transportation Research, Part B, Vol. 33, 1999, pp. 25–42.
this is evidently because the freeway queues at these merges were not 5. Cassidy, M. J., and J. Rudjanakanoknad. Increasing Capacity of an Iso-
first in, first out; i.e., the queued flows were different across lanes. lated Merge by Metering Its On-Ramp. Research Report UCB-ITS-RR-
Moreover, more complicated geometries (like the Ashby Avenue 2004-3. Institute of Transportation Studies, University of California,
on-ramp in Figure 2) present additional problems for this theory. Berkeley, 2004.
6. Papageorgiou, M., and A. Kotsialos. Freeway Ramp Metering: An
The merge ratio is specified as 1/(2n − 1) in another theory (2). Overview. IEEE Transactions on Intelligent Transportation Systems,
This assumes that the freeway queue discipline is first in, first out and Vol. 3, No. 4, 2002, pp. 271–281.
that drivers from the (single-lane) on-ramp take turns with the free- 7. Wooldridge, J. M. Introduction to Econometrics. Southwestern College/
way shoulder lane drivers in a one-to-one alternating fashion. These Thomson Learning, Cincinnati, Ohio, 2002.
8. Daganzo, C. F. Fundamentals of Transportation and Traffic Operations.
conditions do not always arise.
Elsevier, New York, 1997.
It seems that more work will be required to obtain a reliable theory 9. Cassidy, M. J., and J. R. Windover. Methodology for Assessing Dynam-
of the merge ratio. The present findings can be viewed as a neces- ics of Freeway Traffic Flow. In Transportation Research Record 1484,
sary first step to this end. It appears that this is the first study to show TRB, National Research Council, Washington, D.C., 1995, pp. 73–79.
that merge ratios are actually well defined and fixed for a given site. 10. Munoz, J. C., and C. F. Daganzo. Fingerprinting Traffic from Static
Freeway Sensors. Cooperative Transportation Dynamics, Vol. 1, 2002,
These are notable attributes of merging traffic that not only validate part pp. 1–11.
of an existing theory but that can also help guide in the development 11. Cassidy, M. J. Bivariate Relations in Nearly Stationary Highway Traffic.
of new theories. Transportation Research, Part B, Vol. 33, 1998, pp. 49–59.
As noted above, work is still needed to formulate a realistic theory 12. Newman, L. Freeway Operations Analysis Course Notes. Institute of
Transportation Studies, University Extension, University of California,
of merge capacity (in the absence of an exogenous restriction). Empir-
Berkeley, 1986.
ical evidence (already cited here) reveals that merge capacity is affected
by the traffic states on its approaches. Further field experiments are The Traffic Flow Theory and Characteristics Committee sponsored publication of
needed to understand this cause-and-effect relation more completely. this paper.