Exploring the uncertainty in capacity estimation at roundabouts
 Orazio Giuffrè^{1},
 Anna Granà^{1}Email author and
 Maria Luisa Tumminello^{1}
https://doi.org/10.1007/s1254401702388
© The Author(s) 2017
Received: 12 July 2016
Accepted: 21 March 2017
Published: 30 March 2017
Abstract
Purpose
In gapacceptance theory the critical and the followup headways have a significant role in determining roundabout entry capacities which in turn depend on circulating flow rates under a specified arrival headway distribution. Calculation considers single mean values of the gapacceptance parameters, neglecting the inherent variations in these random variables and providing a single value of entry capacity. The purpose of this paper is to derive the entry capacity distribution which accounts for the variations of the contributing (random) variables and suggest how to consider this issue in the operational analysis of the roundabouts.
Methods
We performed a Monte Carlo simulation to get the distribution of entry capacity and found Crystal Ball software effective for performing the random sampling from the probability density functions of each contributing parameter. A steadystate model of capacity was used for performing many runs; in each run, the values of each contributing parameter were randomly drawn from the corresponding distributions.
Results
The paper presents the first simulations and the entry capacity distributions at roundabouts, once the probability distributions of the headways were assumed. The results of the analysis were expressed probabilistically, meaning that the probability distributions of capacity rather than the simple point estimates were obtained.
Conclusions
Comparing the capacity values based on a metaanalytic estimation of critical and followup headways and the capacity functions based on the probability distributions of the model parameters, more insights in developing an appropriate approach to capacity estimation at roundabouts can be gained.
Keywords
1 Introduction
1.1 The background
The transportation decisionmaking process about a road facility or a transport system, as a consequence of planning and design activities or operational analysis, often exposes planners and designers to many sources of variability and uncertainty [1]. In transportation engineering, despite considerable information can be derived from new technologies and can be incorporated into the traditional performance measurements (see e.g. [2–8]), the effect of variability and uncertainty in input parameters on outputs is not often taken into account in the capacity analysis of roads and intersections. The assessment of the effects of a design choice on one or more parameters that are used when an operational analysis is being carried out, requests information on the sources of uncertainty that have affected them and the relation among them [9]. Since the variability is a chancecaused variation and depends on the facility or the system that is being considered, whereas uncertainty is the lack in the analyst’s knowledge of the parameters which define the physical system to be modeled, the combination of variability and uncertainty can erode the ability for making predictions about the future [10]; moreover, high levels of uncertainty can characterize longterm predictions [11]. The analysts typically produce a single number that explains the performance of the road facility, but usually do not give a statement of a likely range of variation in the result nor try to quantify the impact of this uncertainty on capacity estimation [12–14]. When the deterministic models are developed, and used to characterize a whatever process governing the road traffic phenomena, they should be applied for many iterations; for each iteration, rather than selecting the mean or the median value for each parameter, the values of model parameters should be randomly drawn from the corresponding probability distributions. By this way, the results can be expressed in probabilistic terms [15] [16]. Despite the required tasks may be complex, at least they include: i) to identify the possible sources of uncertainty for the problem under consideration; ii) to determine the main variables involved in the probabilistic analysis; iii) to assign the probability distributions to these variables.
The uncertainty analysis, indeed, aims to assess various aspects of a model as the statistical properties of the outputs when stochastic input parameters are considered [17]. Many methods exist for incorporating uncertainty into the quantitative estimates of the performance parameters. Monte Carlo simulation is commonly used by researchers and practicing engineers as a method for propagating uncertainties in model inputs into uncertainties in results; lots of “addins” can be now inserted in spreadsheets, or computer programming can help to develop custom solutions [1].
In the case of the capacity analysis at intersections and roundabouts, the impact of uncertainty depends on the kind of problem to be faced and/or solved. The analysts may need to identify how many lanes are required for a given approach of a roundabout, or know which control type (stop or traffic signal) is most appropriate for a given intersection, etc. According to Kyte et al. [13], the analysts should account for uncertainty when the capacity and levelofservice of a given intersection and/or roundabout should be estimated, and should explain how this component can affect the problem or decision under consideration. Moreover, the analyst should be aware of the large observed variation in driver behavior at intersections and roundabouts [18, 19].
1.2 Review of related literature on roundabouts
Over the last decades, considerable advances in highway operational analysis have been made to develop and apply statistical methods adequate for accommodating different traffic conditions that can be hard to handle through conventional statistical models and methods. Operational analyses often need a large volume of data that have to be generated via experimental measurements, each being appropriate for performing different investigations. However, the observations that we make never exactly match the surveyed processes, since several factors can interfere with the measurement; as a consequence, many sources of uncertainty in the measured variables can arise and increase the risk of misapplying the statistical techniques that are used in the subsequent analysis. In the context of typical experimental measurements at intersections and roundabouts, the analysts often need data which should be qualitative and/or quantitative: the qualitative data are usually considered descriptive and can result subjective in comparison to the quantitative data that are gathered in an experimentally repeatable manner. However, qualitative information is usually closer to phenomenon under examination, but can be subjected to interpretation by individual analyst. Sources of disagreement among the data can arise for different kinds of intersection (unsignalized intersection, signalized intersection and roundabout), since they are so different for geometric design, operations and driver behavior, as well as for the same kind of intersection; in the last case, indeed, the influence of the geometric design features on operations and driver behavior must be clearly taken into consideration especially when regression analysis is used to develop the relationship between the variables under examination.
When calculation of entry capacity  or whatever efficiency measures  at roundabouts, for a chosen observation period, is performed, steadiness and variability in traffic demand, as well as saturated or oversaturated conditions of one or more entries, have to be specified. Thus, the analysis with and without statistical equilibrium should be required and, based on traffic conditions at entries, probabilistic, deterministic, or timedependent models can be used. Although many studies have been done to analyze the operations of (existing or planned) roundabouts, even now the topical discussion between gap acceptance theory and empirical regression models characterizes the general situation about the estimation of entry capacity at roundabouts [20]. Based on their specific advantages and drawbacks, empirical and analytical models for capacity estimation at steadystate roundabouts are used in different countries. Empirical regression models are generated from saturated roundabout entries and, based on a wide data collection, establish relationships between capacity and geometric design features [21]. Based on the concept of gap acceptance, analytical models, in turn, can be developed starting from uncongested conditions; they require that the behavioral parameters are specified [8].
Roundabouts present interesting challenges in gap acceptance modelling, since roundabouts typically use gap acceptance rules. Indeed, capacity and service times at roundabout entries rely upon the possibilities of the minor street drivers to meet enough gaps between the circulating vehicles and safely merge (or cross) the conflict spaces. These possibilities are depending on the flow rate of circulating streams and the arrival headway distributions, as well as individual drivers, vehicle and environment characteristics that affect each individual gap acceptance behaviour. The reader interested in modern roundabouts design and technologies, as well as calculation of roundabout performances, is referred to [8, 22], respectively. Since vehicles enter a roundabout using the gaps between circulating vehicles, the entry capacity is correlated to the driver gap acceptance and its estimation is based on the critical headway and the followup headway when the analytical (gapacceptance) models are used to analyze roundabout operations and performances. Thus, the accurate estimation of roundabout capacity is widely dependent on the equally accurate estimation of the critical headway and the followup headway; however, the analyst should manage the confidence intervals around the estimates of critical headway and the followup headway, but there are no procedures for assessing and measuring the uncertainty in capacity estimation at roundabouts.
The most commonly used methods (although not the only ones) for estimating the critical headway  or the minimum acceptable gap during which a minorstreet vehicle can enter an unsignalized intersection or a roundabout  are the methods of Raff [23] and Troutbeck [24, 25]. The method of Raff [23] is based on macroscopic model and it is used in many countries because of its simplicity. In Troutbeck’s microscopic model the probability of the critical headway is calculated through the maximum likelihood method which requires an iteration process. Based on the outcome of a comprehensive analysis of technical literature (see e.g. [13, 19, 20, 26]), the maximum likelihood method has been suggested and used to estimate the critical headway; see e.g. the bestknown manuals for traffic and transportation engineering [18, 27]. Differently from the critical headway, the followup headway  or the time between two successive departures of minorstreet vehicles which use the same majorstreet gap when a continuous queue on the minor street is observed  can be directly measured on the field [28]. According to the gap acceptance theory, drivers are considered consistent and uniform. Gap acceptance models provide estimates of entry capacity based on constant values of the critical headway and followup headway which, in turn, represent average values for all observed drivers. However, since variability and heterogeneity characterize drivers’ population, the assumptions above introduced can produce erroneous or unreasonably high estimates of roundabout capacity. Tian et al. [26] highlighted that the accurate estimation of the critical headway and the followup headway can be reached when one considers the specific conditions of the site, as the geometric design of the intersection and the approach grade, the types of vehicle and traffic movements. Respecting this, Kyte et al. [29] found that data relating to a specific site usually produce higher forecasts than the forecasts based on general values; thus, a highlevel uncertainty may be associated with the gap acceptance parameters and great variability in estimation of sitespecific critical headway and followup headway can be observed.
Empirical evidence shows that the measurements of the critical headway and the followup headway at roundabouts vary depending on the geometric elements of the intersection layout and the circulating stream bunching characteristics; under changes in traffic demand, the measurements of the gap acceptance parameters at a multilane roundabout can be also influenced by dominant and/or subdominant arrival flows at entry lanes [24, 25]. Considering constant values of the critical headway and the followup headway, the capacity calculation always represents average conditions. However, the critical headway and the followup headway being stochastically distributed cannot be considered as constant values but each of them should be represented by a distribution of a set of values.
1.3 Research objectives and organization of the paper
Based on the considerations above, the purpose of this paper is to consider which variables significantly affect the entry capacity estimation and suggest how to investigate this problem in the operational analysis at roundabouts. The paper presents the results of a research study aimed at finding the probability distributions of entry capacity for singlelane, doublelane and turbo roundabouts, once the probability distributions of the critical headway and the followup headway were assumed. On this regards, we proposed a methodological path based on an application of Crystal Ball software in order to perform a Monte Carlo simulation (see [30] for further application of CrystalBall for performing a MonteCarlo simulation also in other fields of interest). The use of CrystalBall is illustrated with three working examples of roundabout (i.e. the singlelane roundabout, the doublelane roundabout and the turbo roundabout), dealing with a capacity model of nonlinear features and the correlated variables. Thus, a capacity model at steadystate conditions was then used for performing many iterations; in each iteration, the values of the model parameters were randomly drawn from the corresponding probability distributions. The results of the analysis were expressed probabilistically, meaning that the probability distributions of the capacity at each entry lane rather than the simple point estimates of the performance measure were obtained. At last, the comparison was also done between the capacity estimations based on the mean values of behavioral parameters  as derived from a systematic review performed in a previous study by the Authors [31]  and capacity functions based on the probability distributions of the model parameters.
Starting from the results of the metaanalysis of the systematic review briefly summarized in section 2, only for what we deem congruent with the goal of the research presented in this paper, section 3 will describe the procedure and the analysis developed to determine the probability distributions of the entry capacity at singlelane, doublelane and turbo roundabouts, will present the computational approach that incorporates uncertainty in roundabout capacity analysis, and will discuss the main results. At last, conclusions will be summarized in section 4.
2 Preliminary research activities
2.1 Searching for studies for a systematic review
Without wanting make a judgment on the validity of the methods of capacity estimation (see section 1), and recognizing that the efforts attributable to field observations are always praiseworthy, the Authors have set themselves the goal of exploring the advantages of using statistical methods to synthesize data rather than taking the results collected for one or more case studies of a same kind of roundabout (singlelane, doublelane or turbo roundabout) where field observations had been made.
Bearing in mind that the synthesis of empirical estimates of the behavioral parameters at roundabouts should be based on published researches, in a previous work the Authors undertook an extensive review of the benchmark studies currently available in the worldwide literature to collect the mean values of the critical headway and the followup headway at roundabouts and address the variation of each parameter across studies [31]. Thus, we focused on a metaanalysis of effect sizes, that is the analysis where each (primary) study yields an estimate of statistical mean values of the critical headway and the followup headway (hereinafter the effect sizes), assessed the dispersion in these effects and then computed a summary effect [32]. The choice was inspired by the results of applications of metaanalysis carried out in other research areas and for other aspects of transportation data analysis; see e.g. [33].
The reader is referred to Giuffrè et al. [31] both for further details on the set of rules used for doing the literature search and determining which studies should be included in or excluded from the analysis, the total data set extracted from the available relevant literature on the topic, the methods applied to obtain the values of the critical and follow up headways considered in the calculations, and for a more effective presentation of the results of the metaanalysis of effect sizes that was performed as part of the literature review through the randomeffects model. It is noteworthy that some important studies were excluded from our investigation, since experiences in capacity estimation at roundabouts carried out in countries as Great Britain and France were based on a capacity formula which is not a gap acceptancebased model [31]. It should be also noted that the research did not attempt to answer the question of how drivers change their acceptance characteristics and adapt themselves to accept shorter or large headways based on traffic levels of the circulating volume. Indeed, the choice of the behavioural parameters that we selected for the metaanalysis was also based on how arrivals in the circulating stream were estimated in primary studies.
2.2 The randomeffect metaanalysis
Based on the effect size collected from each selected study, the metaanalysis performed the statistical synthesis of the data and produced a single summary effect of which the statistical significance was also assessed [34]. According to [33] the mean values of the two parameters (or the effect sizes), the standard deviations and sample size were the data input of metaanalysis. Since each effect size varied from a study to another study, the metaanalysis of effect sizes was carried out to combine the data used in all selected studies through the randomeffects model. In order to produce a more precise estimation of the mean value of the distribution of effect sizes (namely the summary effect) both the original variance withinstudy and the variance betweenstudies were considered [31, 32]. Thus, we computed a weighted mean value, assuming that the weight for each study was the inverse of the study’s variance; the last one is the sum of the two components of the variance as above introduced. Based on the dispersion in the effects across studies, we computed the summary effect which represented the weighted mean of the single effects [32, 33].
The metaanalytic estimates for critical headway at roundabouts [31]
Roundabout  Entry lane  Circulating lane  Random estimate (se)  95% LL and UL  Zvalue  pvalue  Q  I ^{ 2 } 

Singlelane  4.27 (0.11)  (4.05; 4.49)  37.46  0.000  23.28  1.19  
Double lane  Right  3.82 (0.13)  (3.56; 4.08)  28.79  0.00  15.12  0.00  
Left  4.17 (0.13)  (3.90; 4.42)  31.24  0.00  16.78  0.00  
Turbo (major road)  Left  3.60 (0.06)  (3.49; 3.72)  61.22  0.00  4.84  37.98  
Turbo (major road)  Right  3.91 (0.25)  (3.42; 4.40)  15.66  0.00  1.00  0.00  
Turbo (minor road)  Left  Outer  3.07 (0.15)  (2.78; 3.36)  20.85  0.00  2.91  31.22 
Turbo (minor road)  Left  Inner  3.20 (0.03)  (3.15; 3.26)  106.36  0.00  1.87  0.00 
Turbo (minor road)  Right  3.83 (0.20)  (3.43; 4.23)  18.70  0.00  10.66  81.23 
The metaanalytic estimates for followup headway at roundabouts [31]
Roundabout  Entry lane  Random estimate (se)  95% LL and UL  Zvalue  pvalue  Q  I ^{ 2 } 

Singlelane  3.10 (0.07)  (2.96; 3.25)  41.82  0.00  33.90  20.37  
Double lane  Left  2.85 (0.10)  (2.66; 3.04)  29.58  0.00  22.32  19.36 
Right  2.72 (0.08)  (2.57; 2.87)  35.74  0.00  17.77  9.98 
Comparing the summary effect (random) and the effect size in each single study, one can observe that the summary effect may also differ significantly from the estimation of each study; indeed, it is independent and not based on similar real world data. Thus, the metaanalytic estimate for the critical headway was found nearly consistent across all studies and, compared to single studies, provided a more reliable result for the parameters of interest.
3 Incorporating uncertainty in capacity analysis for roundabouts: The case study
3.1 The starting point hypothesis
In the following we refer to the Hagring’s model [39] that has been particularized for the three roundabouts here studied. Within the values of the inscribed circle diameter and other singlelane roundabout’s dimensions, we can mainly identify the mini roundabouts and the compact roundabouts, whereas within the values of the inscribed circle diameter and other doublelane roundabout’s dimensions we can also recognize the compact roundabouts and the large roundabouts [28, 31]. For the selected case study of turbo roundabout, the basic turbo roundabout geometry was considered with an inner radius of 12 m, an outer radius of 22.45 m, an inside roadway width of 5.30 m and an outsider roadway width of 5.00 m [40].

C _{ e } = entrylane capacity [pcu/h];

φ _{ j } = Cowan’s M3 parameter, i.e. the proportion of the free traffic on the circulating stream;

Q _{ c } = circulating traffic flow [pcu/h];

T _{ c } = critical headway [s];

T _{ f } = follow –up headway [s];

∆ = minimum headway of major flow [s];

j, k, l, m = indices for the lanes on the circulatory roadway.
Note that a Cowan’s M3 headway distribution  that explicitly takes into account the number of bunched vehicles through the φ parameter (equal to 1Δ⋅q _{ c, } where q _{ c } is the circulating traffic flow in pcu/s) representing the proportion of free vehicles  was assumed for each circulating stream; according to literature, the parameter Δ was assumed equal to 2.10 s.
The Hagring’s model [40] in eq. (1) was then specified for singlelane roundabout as follows:
where the circulating traffic flow (Q _{ c }) is split in two streams Q _{ c,i } and Q _{ c,e } representing the inner circulating flow and the outer circulating flow, respectively.
It is noteworthy that drivers entering a roundabout are not required to preselect the entry lane; moreover, vehicles using the outer circulating lane usually leave the roundabout at the next exit. The Hagring’s model [39] was also applied to the turbo roundabouts: for major entries, eq. (2) is applied to estimate the capacity of each entry lane, considering Q _{ c,e } instead of Q _{ c }; for minor entries, eq. (2) is applied to estimate the capacity of the rightentry lane, whereas eq. (3) is applied to estimate the capacity of the leftentry lane.
In order to reach a broadbased assessment of the variability of the behavioral parameters and incorporate uncertainty into the entry capacity estimation, we assumed that the critical headway and the followup headway could be captured over an observation period short enough to ensure a persistent steadystate condition and long enough to overstep the transient state. Under this hypothesis, the headways experienced by users during the observation period can be considered as sampled from the entire population; in this sense, they assume mean values that are within the distribution of the mean.
Based on the probability theory, if the initial (normal distributed) population (X) has mean μ and variance σ ^{ 2 }, the sampling distribution of the sample mean \( \overline{X} \) from samples of size n is assumed normally distributed \( \overline{X} \) ~ N(μ, σ ^{ 2 } /n).
This is also true for a population that is not normally distributed  namely the sampling distributions may also be assumed approximately normally distributed, regardless of the population distribution that one samples from  if the sample size is not too small (n≥30), and the population size, N, is at least twice the sample size. When the distribution of \( \overline{X} \) is unknown or differs from the normal distribution, according to the central limit theorem, \( \overset{}{\mathrm{X}} \) assumes a normal asymptotic distribution. In fact, as n increases, the density function of \( \overset{}{\mathrm{X}} \) approaches a normal distribution very rapidly, although the population distribution is strongly asymmetric; see e.g. [41].

C _{ i } = capacity at entry lane i, pcu/s;

Q _{ ei } = entry flow at the lane i, pcu/s.
It is noteworthy that this formula can be applied only when the ratio (Q _{ ei } /C _{ i }) <1. Besides, it must be said that the steadystate models of entry capacity are only a useful approximation if the duration of the analysis period is considerably greater than the duration calculated using the Morse’s expression [42]. Thus, in our application we assumed a period of observation equal to twice the time of the transient phenomenon and we calculated the number of entering vehicles over this period that we considered as sample size.
In order to apply the Morse’s formula (thereby determining the sample size n), an entering flow rate should be set, or the ratio of the entry flow to the entry capacity (Q _{ e } /C) should be specified upon the condition (Q _{ ei } /C _{ i }) <1; this means that only undersaturated conditions shall be considered. Thus, we considered three different values of the ratio of the entry flow to the entry capacity, i.e. 0.25, 0.50 and 0.70. Under these hypotheses, the corresponding values of n were found equal to 4, 12, 37, respectively, whereas half of them can be considered in steadystate conditions.
3.2 Uncertainty analysis
To understand uncertainty in roundabout capacity estimation, the probability distributions of the random variables of the capacity model had to be identified. Crystal Ball software was used to find the probability distributions of each parameter contributing to entry capacity of the roundabouts under examination. For this purpose, literature data sources were used to hypothesize the probability distributions of each contributing parameter, as described in the previous section 3.1. Since sample sizes generally affect precision, the weighted average value of the standard deviation σ of each single (primary) study used in the metaanalysis was calculated, multiplying each standard deviation by the corresponding sample size (adding those values and then dividing the total value by the total number of sample sizes).
For each roundabout and each contributing parameter, the normal distribution best seemed to fit the data; the (random) summary effect, or the metaanalytic estimation for each headway, is the mean of the distribution, whereas the standard deviation σ is weighted with regard to the sample size, as reported in the different primary studies [31].
The parameters of the sample distribution for the critical and the followup headways (Q _{ e } = 0.5·C)
Roundabout  Entry  Entry lane  Circulating lane  Mean [s]  \( \frac{\sigma}{\sqrt{n}} \)  [s] 

Critical headway  
Singlelane  4.27  0.43  
Doublelane  Right  3.82  0.49  
Left  Inner  4.17  0.49  
Outer  3.81  0.49  
Turbo  Major  Left  3.60  0.31  
Turbo  Major  Right  3.91  0.47  
Turbo  Minor  Left  Inner  3.20  0.18  
Outer  3.07  0.27  
Turbo  Minor  Right  3.83  0.41  
Followup headway  
Singlelane  3.10  0.53  
Doublelane  Left  2.85  0.45  
Doublelane  Right  2.72  0.44 
A major task in our applications was to perform preliminary simulations in order to know how many iterations were needed. Thus, a quite high number of iterations was tried until very slight differences in the outputs led to the searched distributions; lastly, we opted for 10,000 trials. Once the probability distributions for the contributing parameters were set, simulation started. Indeed, the simulation methods generate sequences of random numbers to conduct simulation runs; thus, the probability density functions can be used to describe the physical system. As introduced in section 1, we used the Monte Carlo method and ran simulations with Crystal Ball software. The Monte Carlo method, indeed, selects a random set of input data values drawn from their individual probability distributions; these values are then used in the simulation model to obtain some output values.
In any overlay graph, that is for whatever roundabout examined, one can see bellshaped and symmetrical histograms, in which the central column represents the (mean) capacity corresponding to a specified value of the circulating flow; more numerous measures around the mean value can be observed. It should be noted that, when the circulating flow is low, the capacity distribution turns out to be “squashed” with respect to the abscissa axis. Such distribution is characterized by a high variance, or values highly dispersed, whereby the degree of uncertainty of the output in this case is of some importance. It should be noted again that, if one considers gradually more and more high values in the circulating flow, the distribution of capacity takes a higher and narrow shape, with values quite concentrated around the mean; so the result is found to be more stable.
The results of the simulations indicated that the uncertainty in capacity estimation could be high, especially when the opposing flow is low; in these cases, estimation through the mean values of the individual parameters can be far from the real value, providing a rough underestimation/overestimation of the latter value. The results indicated, indeed, that the actual capacity of the roundabout may be, with a probability of about 50%, higher than the capacity which can be estimated deterministically; based on this result, the traffic conditions could be better than the expected conditions.
At the same time, however, with the same probability, the capacity estimation based on the deterministic model may be an overestimation of the actual capacity, with the result that, for a given traffic demand, the oversaturated conditions at entries are not highlighted. Based on the results obtained, the deterministic estimation of capacity is not cautionary, but rather the risk of poor performances at roundabouts, especially when the circulating flow is low, is quite significant. However, the conclusions that are drawn from this research could be affected by the choice of one or another capacity model. The reader is advised that the use of other models that incorporate different processes could further improve understanding of uncertainty in capacity estimation at roundabouts.
4 Conclusions
The concept of gap acceptance is inherent in the traffic interaction which takes place when the minorstreet vehicles enter the intersection merging into or crossing a traffic major stream. Gap acceptance models are aimed at representing to what extent the minorstreet vehicles entering a roundabout will be able to use an acceptable gap between two consecutive vehicles in the major traffic stream. When a gap acceptance model is going to be developed, assumptions need to be made both for the psychotechnical headways (or the critical headway and the followup headway), and for the arrival headway distribution (or the distribution of the gaps between the vehicles in the different circulating streams), as well as for the distribution of traffic flows among the circulating lanes. The accuracy of the capacity estimation is primarily determined by the accuracy of the estimation of the critical headway and the followup headway. In calculation process, single mean values usually replace these random variables, disregarding their inherent variations, and thus providing a singlevalue of entry capacity. However, the model performance in predicting capacities can be limited, although when one is assessing the operational performance of an existing roundabout or deciding whether (or not) a roundabout should be planned, designed or built, an important role is played by the capacity estimations and levelofservice determinations. Hence, analysts would like to perform estimations as complete and correct as possible.
The paper analyzes how variations in behavioral factors affect the capacity estimates for different types of roundabouts. Uncertainty analysis in estimation of roundabout capacity was performed by using the Monte Carlo sampling and simulation procedure for finding the distributions of output variable values based on the distributions of input data values. A MonteCarlo simulation was then developed as statistical simulation method to obtain the probability distributions of the entry capacity at roundabouts. For this purpose, Crystal Ball software was found very suitable for performing MonteCarlo simulation and estimating uncertainty; thus it was used for the random sampling from the probability density functions that were chosen for the contributing parameters according to the adopted entry capacity model of nonlinear features.
The paper presents the results of first simulations which provided estimates of entry capacity distributions for different roundabouts, once the probability distributions of the critical headways and the followup headways were assumed. A deterministic capacity model was then used for performing many runs; in each run, the values of the contributing parameters were randomly drawn from the corresponding distributions. For the examined roundabouts, simulations with the Crystal Ball software provided the probability distributions of entry capacity when varying the circulating flow. Thus, the results of the analysis were expressed probabilistically, meaning that the probability distributions of capacity at each entry lane rather than the simple point estimates of the performance measure were obtained.
Based on these outputs, one can analyze variations in capacity and then make a comparison with the results derived from the application of the deterministic model. The end results were, indeed, the probability distributions of entry capacity for each roundabout, the capacity functions including the 5th percentile, the median (50th percentile) and the 95th percentile, the capacity functions which were built based on the adopted capacity formulation; the last overlapped the capacity functions corresponding to the median value below which the 50% of the outputs may be found. Although variation in capacity values are expected due to variations in parameters, and the deterministic estimate represent the average value, the results provide some insights about the situations in which the uncertainty of the capacity estimates might be particularly relevant. Indeed, the results of the simulations indicated that uncertainty in capacity estimates could be high, especially when the opposing flow was low; in these cases, estimation through the mean values of the individual parameters can be far from the real value, providing a rough underestimation/overestimation of the latter. The results, indeed, indicated that the actual capacity of the roundabouts may be, with a probability of about 50%, higher than the capacity which can be estimated deterministically. Based on this result, the traffic conditions could be better than the expected ones. However, at the same time, with the same probability, the capacity estimated by using the deterministic model may be an overestimation of the actual capacity, with the result that, for a given traffic demand, the oversaturated conditions at entries are not highlighted. Based on the results obtained, the deterministic capacity appears to be not cautionary, but rather the risk of poor performance at roundabouts, especially when the circulating flow is low, is quite significant. The reader should be advised that the conclusions drawn from this research were based on calculations performed by a particular capacity model and the use of other models that incorporate different processes could further improve the understanding of uncertainty in roundabout capacity estimation.

more traffic demand scenarios to reflect the different constraints (namely environmental and of context); especially at multilane roundabouts, dominant (or subdominant) arrival flows could influence the estimation of the gap acceptance parameters;

different assumptions about the arrival headway distributions and the distribution of vehicles among the circulating lanes (where possible and consistently with the chosen layout of the roundabout);

uncertainty analysis on the central island size (and/or the entry width) of the roundabout, which we have assumed as being not influent; this analysis can be included as well to reflect the different needs of the built environment.
Declarations
Open Access This article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
Authors’ Affiliations
References
 Grayman WM (2005) Incorporating uncertainty and variability in engineering analysis. J Water Resour Plan Manag 131(3):158–160. doi:10.1061/(ASCE)07339496(2005)131:3(158) View ArticleGoogle Scholar
 Uusitalo L, Lehikoinenb A, Hellec I, Myrberga K (2015) An overview of methods to evaluate uncertainty of deterministic models in decision support. Environ Model Softw 63:24–31. doi:10.1016/j.envsoft.2014.09.017 View ArticleGoogle Scholar
 Ameri M, Moayedfar R, Jafari F (2013) Determination the capacity of twolane suburban roads with neural networks and effect of speed on level of service. Eur Transp Res Rev 5(4):179–184. doi:10.1007/s125440130096y View ArticleGoogle Scholar
 Mukherjee A, Majhi S (2016) Characterisation of road bumps using smartphones. Eur Transp Res Rev 8:13. doi:10.1007/s1254401602001 View ArticleGoogle Scholar
 Wu J, Liu P, Tian ZZ, Xu C (2016) Operational analysis of the contraflow leftturn lane design at signalized intersections in China. Transp Res Part C: Emerg Technol 69:228–241. doi:10.1016/j.trc.2016.06.011 View ArticleGoogle Scholar
 Schnebele E, Tanyu BF, Cervone G, Waters N (2015) Review of remote sensing methodologies for pavement management and assessment. Eur Transp Res Rev 7:7. doi:10.1007/s1254401501566 View ArticleGoogle Scholar
 Arroju R, Gaddam HK, Vanumu LD, Rao KR (2015) Comparative evaluation of roundabout capacities under heterogeneous traffic conditions. J Mod Transp 23(4):310–324. doi:10.1007/s4053401500898 View ArticleGoogle Scholar
 Mauro R (2010) Calculation of roundabouts. SpringerVerlag, Berlin HeidelbergView ArticleGoogle Scholar
 Gao Y, Rong G, Yang L (2015) Analysis of order statistics of uncertain variables. J Uncertain Anal Appl 3:1. doi:10.1186/s4046701400251 View ArticleGoogle Scholar
 Vose D (2008) Risk analysis: a quantitative guide. John Wiley & Sons, Inc., New York, USAMATHGoogle Scholar
 Matas A, Raymond JL, Ruiz A (2012) Traffic forecasts under uncertainty and capacity constraints. Transportation 39(1):1–17. doi:10.1007/s1111601193251
 Bie J, Lo HK, Wong SC (2010) Capacity evaluation of multilane traffic roundabout. J Adv Transp 44(4):245–255. doi:10.1002/atr.124 View ArticleGoogle Scholar
 Kyte M, Dixon M, Basavaraju P (2003) Why field measurements differ from model estimates analysis  framework for capacity and levelofservice analysis of unsignalized intersections. Transp Res Rec 1852:32–39. doi:10.3141/185205 View ArticleGoogle Scholar
 Tarko A, Tian Z (2003) Example analysis and handling of uncertainty in the highway capacity manual with consideration of traffic diversion. Transp Res Rec 1852:40–46. doi:10.3141/185206 View ArticleGoogle Scholar
 Iman RL, Helton JC (1988) An investigation of uncertainty and sensitivity analysis techniques for computer models. Risk Anal 8(1):71–90. doi:10.1111/j.15396924.1988.tb01155.x View ArticleGoogle Scholar
 Cullen AC, Frey HC (1999) Probabilistic exposure assessment: a handbook for dealing with variability and uncertainty in models and inputs. Plenum press; New York, USAGoogle Scholar
 Bulleit WM (2008) Uncertainty in structural engineering. Pract Period Struct Des Constr 13(1):24–30. doi:10.1061/(ASCE)10840680(2008)13:1(24) View ArticleGoogle Scholar
 Highway Capacity Manual. (2010) Transportation research board, Special report 209, 5th editionGoogle Scholar
 Bunker JM (2014) Novel methods and the maximum likelihood estimation technique for estimating traffic critical gap. J Adv Transp 48(6):542–555. doi:10.1002/atr.1204 View ArticleGoogle Scholar
 Brilon W, Troutbeck R, König R (1999) Useful estimation procedures for critical gaps. Transp Res Part A Policy Pract 33(3–4):161–186. doi:10.1016/S09658564(98)000482 View ArticleGoogle Scholar
 Kimber, R. and Coombe, R. (1980). The traffic capacity of some major priority junctions. Crowthorne, Berkshire, UK: Transport and Road Research Laboratory, Supplementary Report 582Google Scholar
 Tollazzi T (2015) Alternative types of roundabouts: an informational guide. Springer tracts on transportation and traffic 6. SpringerVerlag. Springer international publishing. SwitzerlandGoogle Scholar
 Raff MS, Hart JW (1950) A volume warrant for urban stop signs. The Eno foundation for highway traffic control. Saugatuck, Conn, USA http://ntl.bts.gov/lib/26000/26700/26777/index.html Google Scholar
 Troutbeck RJ (1986) Average delay at an unsignalized intersection with two major streams each having a dichotomized headway distribution. Transp Sci 20(4):272–286. doi:10.1287/trsc.20.4.272m View ArticleGoogle Scholar
 Troutbeck RJ (1991) Recent Australian unsignalized intersection research and practices. Intersections without traffic signals II 1991; SpringerVerlag (Werner Brilon. Ed.): 238–257Google Scholar
 Tian Z, Vandehey M, Robinson BW, Kittelson W, Kyte M, Troutbeck R, Brilon W, Wu N (1999) Implementing the maximum likelihood methodology to measure a driver's critical gap. Transp Res A 33(3–4):187–197. doi:10.1016/S09658564(98)000445 Google Scholar
 Handbuch die Bemessung von Straben (HBS: German highway capacity manual). (2009) Forschungsgesellschaft für Straßen und Verkehrswesen (FGSV): CologneGoogle Scholar
 Rodegerdts L, Blogg M, Wemple E, Myers E, Kyte M, Dixon M, List G, Flannery A, Troutbeck R, Brilon W, Wu N, Persaud B, Lyon C, Harkey D, Carter D (2007) Roundabouts in the United States. NCHRP report 572. Transportation Research Board of the National Academic. Washington. D.C., USAGoogle Scholar
 Kyte M, Tian Z, Mir Z, Hameedmansoor Z, Kittelson W, Vandehey M, Robinson B, Brilon W, Bondzio L, Wu N, Troutbeck RJ (1996) NCHRP web document 5: capacity and level of Service at Unsignalized Intersections: final report volume 1: twoway stop controlled intersections. Transportation Research Board, Washington, D.C., USAGoogle Scholar
 Gonzalez AG, Herrador A, Asuero AG (2005) Uncertainty evaluation from MonteCarlo simulations by using CrystalBall software. Accred Qual Assur 10(4):149–154. doi:10.1007/s0076900408969 View ArticleGoogle Scholar
 Giuffrè O, Granà A, Tumminello ML (2016) Gapaccepteance parameters for roundabouts: a systematic review. Eur Transp Res Rev 8:2. doi:10.1007/s1254401501904 View ArticleGoogle Scholar
 Borenstein M, Hedges LV, Higgins JPT, Rothstein H R. (2009) Introduction to Metaanalysis. John Wiley & Sons, Ltd, Chichester, UK, doi:10.1002/9780470743386.ch13
 Elvik R (2011) Publication bias and timetrend bias in metaanalysis of bicycle helmet efficacy: a reanalysis of Attewell, Glase and McFadden, 2001. Accid Anal Prev 43(3):1245–1251. doi:10.1016/j.aap.2011.01.007 View ArticleGoogle Scholar
 Cochran WG (1950) The comparison of percentages in matched samples. Biometrika 37(3/4):256–266. doi:10.1093/biomet/37.34.256 MathSciNetView ArticleMATHGoogle Scholar
 Higgins J, Thompson SG, Deeks JJ, Altman DG (2003) Measuring inconsistency in metaanalyses. BMJ 327:557–560View ArticleGoogle Scholar
 Gazzarri A, Martello MT, Pratelli A, Souleyrette R (2013) Gap acceptance parameters for HCM 2010 roundabout capacity model applications in Italy. Wit Ser Transp Syst Traffic Eng 1:1–6Google Scholar
 Zheng D, Chitturi M, Bill A, Noyce D A (2011) Comprehensive evaluation of Wisconsin roundabouts volume 1: traffic operations. Wisconsin Traffic Operations and Safety Laboratory, Wisconsin, US, Available at http://wwwtopslabwiscedu/projects/410html Accessed 28 March 2013
 Fortuijn LGH (2009) Turbo roundabout. Estimation of capacity. Transp Res Rec 2130(2009):83–92. doi:10.3141/213011 View ArticleGoogle Scholar
 Hagring O, Rouphail NM, Sørensen HA (2003) Comparison of capacity models for two lane roundabouts. Transp Res Rec 1852:114–123. doi:10.3141/185215 View ArticleGoogle Scholar
 CROW (2008) Turborotondes [turbo roundabouts]. Publicatie 257. CROW Kenniscentrum voor verkeer, vervoer en infrastructuur [Dutch information and technology platform], The Netherlands, EdeGoogle Scholar
 Gravetter FJ, Wallnau LB (2013) Statistics for the behavioral sciences, 9th edn. Cengage Learning, Boston, USGoogle Scholar
 Morse PM (1982) Application of queuing theory, 2nd edn. Chapman Hall, LondonGoogle Scholar