The effects of methodological uncertainties on repair prioritization: An exploratory analysis of national road bridges in Lao PDR
2025 年 6 巻 1 号 p. 51-61
詳細
2025 年 6 巻 1 号 p. 51-61
A systematic approach to determining the maintenance priority of road bridges is important for ensuring that limited resources are directed to the bridges in greatest need. Multicriteria decision-making is often used for priority evaluation, but the multiplicity of non-equivalent methods may lead to undesirable results. The goal of this study is to evaluate the effect of methodological uncertainties on the prioritization of road bridges by applying uncertainty analysis to propagate the effect of different criteria selection, criteria weighting, and criteria aggregation methods to the bridge priority ranking. A prioritization framework and data on national road bridges from Lao PDR were used. It was found that the bridges with the highest and lowest average priority were the least sensitive to methodological choice, whereas the sensitivity of bridges with intermediate levels of priority varied widely. In-depth examination revealed that even if two bridges shared a similar level of priority, their sensitivity to methodological uncertainty varied depending on the bridge condition and characteristics and how these were scored. The results illustrate how uncertainty analysis can be utilized for improving the robustness of decision-making for bridge management, but it is important to develop the methodological alternatives by transparent and scientific means.
Road bridges are essential for economic growth, as they facilitate the safe and efficient movement of people and goods. During their service life, bridges are exposed to a diversity of environmental and mechanical loadings that cause damage and result in deterioration of structural performance. Maintenance activities are therefore necessary to ensure continued safety by identifying any damages present on the structure (inspection), assessing the performance level based on the condition (diagnosis), executing appropriate interventions, if necessary (intervention), and preserving the details of all activities and their outcomes (record)1).
While maintenance is currently a serious challenge for developed countries with aging and deteriorating bridge infrastructure, such as Japan, Germany, and the USA2-4), countries with less mature bridge infrastructure will face similar issues in the near future. However, effective maintenance is more difficult to implement in developing countries due to limited resources5), including insufficient budget for maintenance6). As a result, damaged bridges needing intervention may go neglected until resources become available, which increases the risk of accidents and disruption for road users7).
A systematic approach to evaluating the priority of bridges for maintenance activities becomes important when facing budget restraints. A literature review shows that multicriteria decision making (MCDA) is often utilized for tackling the bridge prioritization problem. In the USA, state-level Departments of Transportation (DOT) have generally adopted a weighted scoring approach, wherein the bridge priority is calculated based on a set of prioritization criteria8), which include those related to the health of the structure, such as structural condition rating or structural deficiency, as well as criteria related to bridge features, such as load capacity, functional class, vertical clearance, detour length, or daily traffic. Weights are then applied to these criteria to reflect their relative importance towards maintenance prioritization. Examples of other MCDA approaches to the prioritization of bridges include the Analytical Hierarchy Process (AHP) and multi-attribute utility theory9,10), but these are generally based on the same concept of aggregating a set of weighted criteria for determining bridge priority.
As with any decision-making problem, however, the ranks generated by a prioritization algorithm are dependent on the methods adopted for the analysis. Major methodological choices include the selection of criteria and the weights applied to those criteria, as well as the selection of the MCDA framework, which dictates how criteria and weights are incorporated to generate each bridge’s “priority.” This multiplicity of approaches introduces methodological uncertainty into the prioritization, which leads to uncertainty in the output scores and the ranks derived from those scores11). In the absence of a rigourously established framework, methodological choices may be made subjectively or without sufficient scientific evidence, leading to erroneous conclusions due to uncertainties within the decision-making process. In the case of the USA, for example, it was reported that many of the weighting schemes adopted by state DOTs were developed without scientific rigor, and there is a lack of consensus regarding the decision-making process for bridge prioritization12).
To ensure the robustness and reliability of bridge maintenance planning based on priority scores or ranks, it is necessary to understand the effects of methodological uncertainties on the prioritization process and its outputs. The goal of this study is therefore to evaluate how methodological choices for evaluating priority affect the prioritization of road bridges for repair activities. This will be achieved by first conducting uncertainty analysis, wherein an ensemble of methodological choices will be applied to produce a set of priority rankings for a set of road bridges, followed by probabilistic examination of these rankings to evaluate the degree to which they are affected by methodological uncertainties in the prioritization process.
To carry out these analyses, this study will utilize a prioritization framework recently established in the Lao People’s Democratic Republic (Lao PDR). As part of a capacity development project supported by the Japan International Cooperation Agency (JICA), a new bridge management system, including an MCDA-based prioritization module, was developed for the Ministry of Public Works and Transport (MPWT). However, the project established only the initial settings for repair prioritization, and it is anticipated that Lao stakeholders will modify this initial framework according to their needs and as they develop experience with its application.
As the prioritization framework from Lao PDR is used for this study, the results will have direct significance for future efforts to improve the system in that country. However, it is also expected that the application of uncertainty analysis to the issue of bridge prioritization will serve as a useful reference and highlight issues that should be considered by bridge managers and maintenance engineers looking to improve the robustness of their decision-making processes.
In 2020, MPWT initiated a capacity development project for bridge maintenance and management supported by JICA. The goal of this project (henceforth referred to as “JICA-BMM”) was to implement effective management practices, apply maintence technologies that extend the lifespan of bridge structures, and establish an institutional maintenance framework for realizing a resilient road network. The project was comprised of four pillars: achieve and institutionalize academic, public, and private collaboration; establish maintenance and management objectives and a sustainable technology transfer program; enhance the quality of inspection, diagnosis, and repair activities; and ensure seamless project management considering the risks of COVID- 1913).
One output of the JICA-BMM project was an analytical module for the bridge management system, which included an MCDA framework that aimed to realize the rational prioritization of bridges for repair activities. The proposed framework utilizes a simple weighted-sum method for determining bridge priority, which is shown in Eqn.(1)14).
Where: Ps is the priority score, wi is the weight given to criteria i, ci is the score of criteria i, and n is the number of criteria.
The project also established a tentative set of five prioritization criteria and their corresponding weights based on the project experts’ judgment of their relative importance for bridge prioritization. Table 1 describes these five criteria in detail: element (or bridge span) health index, element importance, road importance, motorized traffic index (MTI), and impact on regional economy.
The methods for evaluating the health indices (c1) were developed during the JICA-BMM project in the bridge inspection and diagnosis manuals15). These manuals also specified the importance of the various structural elements (c2). These prioritization criteria are updated accordingly when new inspection data on the bridge are obtained. The road importance (c3) is derived from the road classification established by the national government, which spans six categories from national roads, which have the highest importance, to rural roads and specific roads, which have the lowest importance. The motorized traffic index (c4) criterion was adopted to reflect the loading on the bridge. These data are included in the bridge inventory and are calculated based on traffic monitoring. Finally, impact on regional economy (c5) is another feature included in the bridge inventory that qualitatively reflects the bridge’s role in regional economic development.
(1) Target bridges
This analysis aims to examine the methodological uncertainty in the prioritization of bridges on the national roads in Lao PDR. At the time of the JICA- BMM project, there were 829 national road bridges, but only 664 bridges possessed complete inspection and inventory data. Of these 664 bridges, 97.0% were classified as concrete bridges and 3.0% as steel.
(2) Scoring of bridges for prioritization
Along with the weights shown in Table 1, a rating scheme that converts the bridge data to a three-point ordinal score for each criterion was also established by the JICA-BMM project. The default scheme is given in Table 2, and the distribution of the 664 bridges following this scheme is shown in Fig.1 for the five prioritization criteria. It can be seen that 100% of the bridges scored “1,” or the lowest priority, for element (or bridge span) health index (c1), and 100% scored “3,” or the highest priority, for road importance (c3). Furthermore, nearly 100% scored “3” for element importance (c2). These results show that all bridges being considered for analysis are in generally good condition and require at most only small-scale repair, and that all have the same importance due to the selection of the national road network for the analysis.
(3) Adjustment of score rating scheme
As the variability of the target bridges is zero for two of the prioritization criteria (c1 and c3) and close to zero for another criterion (c2), it is evident that the prioritization of the 664 bridges will be affected only by the remaining two criteria (motorized traffic index and impact on regional economy). However, to apply uncertainty analysis effectively, it is necessary for the target bridges to exhibit variability along all the analytical dimensions. The rating scheme is therefore rexamined and adjusted.
First, as only bridges on the national road network were considered in this analysis, the distribution of c3 is understandable and cannot be modified. It was thus decided to remove this criterion from the priority calculation in this analysis, as it is not possible to distinguish the road importance among bridges that are only present on the national road network.
On the other hand, the target 664 bridges did exhibit variability in their raw data for c1 and c2, so adjusting the rating score scheme may produce a dataset with greater variability. Examination of the raw data found that it was difficult to distinguish between national road bridges when using the bridge span health index, so it was decided to focus criteria c1 on the element health - specifically, on the most damaged element in each structure. In addition, review of the rating schemes in the UK, China, and Sri Lanka revealed differing approaches to determing the criticality of condition for prioritization16-18), but it was decided to adopt a higher health index threshold that would enable more proactive risk management and provide decision makers with advanced notice for maintenance planning. This also aligns with the need for longer planning horizons in Lao PDR due to budget constraints. The revised rating score scheme for c1 and the associated repair activities are shown in Table 2.
The importance of the element factors (c2) was also updated to produce a more diverse distribution. First, the element factor of the most damaged element, which was used for determining the health index of the bridge for c1, was chosen. Next, the rating score scheme was adjusted by increasing the lower bound of the highest rating score from 0.50 to 0.80 so that bridges with damage to the most important elements would receive the highest priority. The ranges of the intermediate and lower scores were also increased accordingly. The adjusted rating score scheme for c2 is also given in Table 2.
The updated distributions of the 664 target bridges for the retained four prioritization criteria are shown in Fig.2. While the distributions of c4 and c5 remain unchanged, the distributions of c1 and c2 now exhibit greater variability when compared to the distributions produced by the default schemes. The scores based on the revised rating scheme will be utilized for the uncertainty analysis.
(1) Analysis framework
The prioritization Eqn. (1) involves three different sources of methodological uncertainty: the selection of prioritization criteria, the weighting of those criteria, and the method of aggregating those criteria. The extent to which these sources of uncertainty affect the priority of bridges for repair prioritization will be investigated through an ensemble approach, wherein different methodological choices will be exhaustively combined and the priority ranking of the bridges, which is derived from the priority score, will be treated probabilistically. Fig.3 shows the analysis framework illustrating the combinations investigated in this study. The details of these methodological choices will be discussed in the following sections.
(2) Criteria selection
Prioritization criteria are intended to serve as an operational representation of the complex attributes of road bridges and to translate these attributes into quantifiable and comparable measures19). The selection of prioritization criteria is dependent on many factors, including data availability and appropriateness for the analysis, and thus there is a notable degree of subjectivity in their selection.
Although the JICA-BMM originally proposed five criteria based on the expert judgment, after data preprocessing only four criteria were retained. To test the effect of these criteria on the bridge prioritization, a one-in one-out approach was adopted, whereby alternative sets of criteria were constructed by selectively removing each indicator in turn. As a result, five sets of criteria were generated: one set containing all four criteria, and four sets containing each indicator excluded in turn.
(3) Criteria weighting
The weights applied to the prioritization criteria represent value judgments that reflect the importance of each criterion in the context of the prioritization framework20). Choice of weighting in MCDA is often a subjective process, and stakeholders may possess different judgment values regarding the importance of criteria in a decision-making problem. The selection of weights is therefore a major source of uncertainty for bridge prioritization.
In this analysis, the effect of weighting on bridge prioritization was examined by generating alternative weighting schemes that placed a strong importance on each prioritization criterion in turn while minimizing the importance of the other criteria (ws3 through ws6). In addition, the default weighting scenario with c3 excluded (5:2:1:1 for c1:c2:c4:c5) as well as an equal weighting scenario were also included (ws1 and ws2, respectively). The weight ratios for the six weighting schemes are summarized in Table 3. These ratios were referred to during the uncertainty analysis to calculate the appropriate weights for each criteria set in each simulation.
(4) Criteria aggregation
The default method of aggregating the priority criteria is linear aggregation. To evaluate the effect of aggregation method on the prioritization, geometric aggregation (Eqn. 2) was also utilized.
Where: Ps is the priority score, wi is the weight given to criteria i, ci is the score of criteria i, and n is the number of criteria. Geometric, or multiplicative, aggregation is less compensatory than linear aggregation21), and it was expected that geometric aggregation may affect the bridge prioritization through trade-offs between the criteria.
(5) Probabilistic ranking and its interpretation
Through exhaustive combinations of the five sets of prioritization criteria, six weighting schemes, and two aggregation methods, each bridge received 52 different priority scores. It is noted that, while the full exhaustive combination should produce 60 scores for each bridge, the results when applying the high importance weighting schemes together with the criteria set excluding the indicator with high importance (for example, criteria set cs2, which drops c1, and weighting scheme ws3, which places high importance on c1) would produce the same result as applying the equal weighting scheme to that indicator set. The eight combinations leading to duplicate results were thus removed from the analysis.
After generating the priority scores of all bridges for a single ensemble of methodological choices, the bridges were assigned their priority rank based on the ranking of their priority score relative to all bridges for that ensemble. The bridge with the highest score was ranked highest priority, or priority rank “1,” followed by the bridge with the second highest score, which was ranked second, and so forth until all bridges were assigned a priority rank. When carried out for all methodological combinations, this produced a set of 52 priority ranks for each bridge.
The effects of the three sources of methodological uncertainties on the bridge prioritization were then evaluated by statistical analysis of the distribution of the priority ranks22). The average priority rank of each bridge was used to evaluate its overall priority when accounting for methodological uncertainties, and the standard deviation of the priority ranks was used as an indicator of the degree of sensivity of each bridge to the methodological uncertainties. The effects of methodological uncertainty on each bridge’s priority could therefore be evaluated based on the standard deviation, with higher values indicating larger effects and lower values indicating smaller effects.
(1) Relationship between average and standard deviation of the priority ranks
Fig.4(a) visualizes the results of the uncertainty analysis through a scatter plot of the relationship between the average and standard deviation of the 52 simulated priority ranks for each of the 664 national road bridges. The quintile ranges are also shown, together with the actual percentage of bridges within each quintile. The quintile percentages are not exactly 20% due to many bridges possessing the same rank. The number of bridges with tied ranks can be better seen in Fig.4(b).
The averages and standard deviations of the ranks follow a second-order polynomial trend, and bridges with the highest or lowest priority, on average, tend to have smaller standard deviations. It can therefore be concluded that these bridges are, in fact, high or low priority bridges because their ranks are less affected by methodological uncertainties in the prioritization process. In other words, their priority ranks remains relatively unchanged regardless of the methodological combination used to calculate the priority score. Conversely, bridges concentrated in the middle of the curve appear more sensitive to the choice of methodology, and there may be more uncertainty regarding their actual priority.
Although the second-order polynominal equation provides a reasonable fit to the dataset, there remains a great deal of variability in the standard deviations that cannot be explained by the average rank alone. To better understand the effect of methodological uncertainties on the bridge priority rankings, the relationships between the averages and standard deviations of the priority ranks will be examined in greater detail within select quintiles.
(2) Intra-quintile analysis for select quintiles
a) Top quintile (highest priority)
The averages and standard deviations for the bridges in the top quintile are shown in Fig.5. The effect of uncertainty on bridges in this quintile decreases linearly as the average rank increases, with the highest-ranked bridge, No. 7022, having an average priority rank of 1 and a standard deviation of 0. This bridge is the only one to possess a rating score of “3” for all prioritization criteria, and therefore it always received the highest priority score and, consequently, the highest rank, regardless of the methodological combination.
It can further be seen that two data points, (101,79) and (105,126) possess fairly similar average ranks but different standard deviations, indicating they are affected by methodological uncertainties differently despite having similar overall priority. Fig.4(b) shows that there are seven bridges at each of these two points, so two representative bridges, No. 7105 and No. 7123, were selected for further examination. Both are concrete bridges, but the most damaged element for No. 7015 was the bridge deck, whereas the most damaged element for No. 7123 was the bridge railing. The score ratings of these two bridges and their priority rank statistics are summarized in Table 4. Both bridges possess the same scores for criteria c4 and c5, but they differ for c1 and c2. No. 7105 is rated as “2” for both these criteria, but No. 7123 is rated “3” for c1 and “1” for c2. These scores indicate that although the bridge railing of No. 7123 is in a more damaged state than the bridge deck of No. 7105, the bridge deck is a more important element than the railing. When aggregated linearly, the average priority ranks are the same for both sets of rating scores but, when using geometric aggregation, the average of the combination of “3” and “1” will be lower than that of “2” and “2.” This comparison highlights the difference between linear and geometric aggregation vis a vis how they handle trade-offs between prioritization criteria.
Finally, the distributions of the simulated priority ranks used for calculating the representative averages and standard deviations of the two select bridges are shown in Fig.6. The difference between the standard deviations of No. 7105 and No. 7123 can be clearly understood from these distributions. While No. 7123 is often evaluated as having very high priority, there are some scenarios in which it is ranked much lower, which pulls its average rank down. No. 7105, on the other, is not evaluated as high priority as often as No. 7123, but its distribution is less dispersed and, thus, its average value is less affected by outliers. The distribution of No. 7123 also highlights the degree to which the priority of a single bridge may vary depending on the choice of methodologies for prioritization, as the highest rank it obtained was 1st, while the lowest was 350th.
b) Middle quintile (moderate priority)
Fig.7 shows the averages and standard deviations for the bridges with moderate priority. Unlike the top quintile, there is no clear relationship between these values in the middle quintile. This quintile also has the largest disparity in standard deviation between bridges with similar average priority ranks. Data points (249,72) and (262,219) fall beside each other with regards to average rank, but the former is the least affected by uncertainty in the middle quintile, whereas the latter is the most affected in the entire dataset. There are three bridges at (249,72), all with nearly identical characteristics, so No. 7339 was chosen for further analysis. On the other hand, only No. 8219, is present at (262,219), making it one of the most unique observations in the dataset. Both of the selected bridges are concrete, with No. 7339 possessing the most severe damage to its bridge deck and No. 8219 to its foundation. The scores of the two bridges, however, are completely divergent, as No. 7339 is rated “2” for all four criteria, whereas No. 8219 scored “3” for c1 and c2 and “1” for c4 and c5 (Table 5). No. 8219 thus represents a perfect trade- off between condition-related criteria and inventory- related criteria relative to No. 7339, as the sum of the rating scores is equal for both bridges yet no criteria are scored the same between them.
The average priority ranks of these two bridges and their sensitivity to methodological uncertainties is further examined considering the distributions of their simulated priority ranks (Fig.8). All four of the rating scores of No. 7339 fall in the middle of the rating scale, and the lack of trade-offs between the four prioritization criteria mean the average priority rank of No. 7339 is insensitive to the aggregation method. The lack of variability among the criteria also diminishes the effect of criteria set selection, as the one-in one-out approach does not produce meaningfully different sets of values for weighting and aggregation. The sensitivity of No. 7339 is therefore due almost entirely to uncertainty arising from the choice of weighting, which reduces its variability overall and leads to a distribution with a single peak concentrated in the middle.
The distribution of No. 8219 is the polar opposite of No. 7339. The largest concentrations of ranks are at the extreme ends of the distribution, indicating that No. 8219 is likely to be evaluated as either a very high priority bridge or a very low priority bridge, depending on the choice of method. This may be attributed to the variability of the rating scores. The “3-3-1-1” combination represents the most extreme trade-off possible for the four-criteria, three-level evaluation adopted in this prioritization framework. Consequently, the effects of criteria set selection and aggregation method on the priority rank of No. 8219 are the largest among all bridges, leading to a large percentage of ranks that fall at either end of the distribution depending on whether a “3” criterion or “1” criterion was dropped from the criteria set.
c) Bottom quintile (lowest priority)
Finally, the averages and standard deviations of the priority ranks for the bridges in the bottom quintile are shown in Fig.9. Although the observations in this quintile exhibit a downward trend, with lower priority leading to less sensitivity to methodological uncertainties, the linear relationship between average rank and its standard deviation is not as strong as it was for the top quintile.
The bottom quintile also contains the bridges with the lowest average priority rank (638) among all 664 bridges. The rating score for these seven bridges is “1” for all four criteria; as a result, their priority scores are always “1” regardless of the ensemble of methods. However, unlike the bridge with the highest average priority rank (with scores of “3” for all criteria), the standard deviation of the lowest priority bridges is not zero. There exist some methodological combinations wherein these bridges with “1” for all criteria receive the same priority score as bridges with three “1” criteria and one “2” or “3” criterion due to the one-in one-out criteria sets. In those cases, these lowest priority bridges are tied in rank with other bridges, which results in their priority rank deviating from the other combinations in which these seven alone are the lowest-ranked bridges.
In the bottom quintile, there are again data points with similar average priority ranks but differing standard deviations. Two of these - (538,105) and (550,31) - are chosen for further analysis. Data point (538,105) contains just two bridges, from which No. 7575 was selected. On the other hand, (550,31) is the largest group of bridges in the entire dataset, with 72 concrete bridges sharing the same average priority rank and standard deviation. Among these, No. 7245 was selected for comparison with No. 7575.
Table 6 summarizes the bridge rating scores and Fig.10 shows the distribution of the simulated ranks for Nos. 7245 and 7575. The difference between the two bridges is the rating scores of c2 and c4, with No. 7245 scoring higher on c4 and No. 7575 higher on c2. Despite this minor difference, however, the distributions of the two bridges are completely unalike. This may be attributed to the choice of weighting scheme. Were the default weighting scheme ws1 excluded, the remaining five schemes would provide equal treatment of the four prioritization criteria - one scheme in which they are weighted equally, and four scenarios in which they are weighted strongly in turn. The average priority ranks of No. 7245 and No. 7575 with ws1 excluded would therefore be equal, as the “1-1-2-1” and “1-2- 1-1” rating scores are functionally equivalent for those combinations.
The default weighting scheme, however, increases the weight given to the element factor c2, which is scored higher for bridge No. 7575. As a result, when including this weighting scheme in the uncertainty analysis, there are some ensembles for which the priority score of No. 7575 becames much higher due to the increased weight placed on c2. The inclusion of the default weighting scheme is therefore the primary cause of the increased sensitivity of No. 7575 to methodological uncertainties relative to No. 7245, as it introduces asymmetry to uncertainty arising from the selection of weighting scheme.
The preceding analytical results revealed various issues facing the rational prioritization of road bridges for repair, both in the case of Lao PDR and for bridge management in general. Although this study focused on the prioritization process, both the data pre-processing and analysis results suggest that stakeholders in Lao PDR may need to revisit the rating scores used for normalizing the bridge condition and characteristics. The default thresholds used for categorization resulted in nearly uniform data distributions for three prioritization criteria. These criteria account for 80% of the weight in the default weighting scheme, and thus differences in the remaining two prioritization criteria may be insufficient to clearly distinguish the bridges for repair activities. The rating schemes were adjusted for the purposes of this study, but more in-depth discussions should be carried out for improving these rating schemes.
The overall relationship between the average priority ranks and their standard deviations revealed that bridges ranked as the highest and lowest priority, on average, tended to be less affected by the uncertainty arising from methodological choices in the prioritization process. For road bridge managers in Lao PDR, this result suggests that there is a high likelihood that the bridges identified using the current method are actually high and low priority bridges. However, bridges with intermediate levels of priority are more suspect, as the current prioritization method may be identifying some bridges as moderate or low priority when they are, in fact, high priority.
One previously-identified example of this issue is bridge No. 8219, which belongs to the middle quintile. From a bridge manager standpoint, the prioritization results of this bridge may be cause for concern because both c1 and c2 - the two criteria representing the structural condition of the bridge - of No. 8219 are scored as most severe, indicating that the safety of the bridge may be at risk. However, its average priority rank falls within the middle quintile, and some ensembles evaluate No. 8219 as one of the lowest priority bridges, despite its critical condition. It is therefore recommended that the characteristics of the national road bridges in Lao PDR with intermediate priority be examined in detail to ensure that the necessary resources are allocated to maintain the bridge safety. The tradeoff observed for No. 8219 also highlights a more general point of caution regarding the use of criteria unrelated to structural performance in the prioritization process, as their inclusion may shift resources away from structurally- deficient bridges to bridges that score higher on supplementary prioritization criteria.
While this study clarified how methodological uncertainties may affect bridge prioritization, the results presented here should be interpreted in the context of the methodological alternatives explored in the uncertainty analysis. The criteria sets and weighting schemes were selected considering theoretical extremes relative to the default approach and, in some cases, do not represent reasonable scenarios for practical bridge maintenance. In particular, the alternative scenarios that placed little weight on the condition of the bridge, or excluded the structural condition entirely, would be difficult to justify from the perspective of bridge safety. It is therefore recommended that, when using uncertainty analysis to evaluate a prioritization framework, the methodological choices considered should represent practical alternatives, and that they be generated through transparent and scientific methods.
Finally, it is reiterated that this study evaluated the effects of methodological uncertainties by examining the variability of the priority of individual bridges using the standard deviation, a simple yet useful metric for quantifying variability. However, the focus on individual bridges limits the conclusions that could be made about the effects of uncertainty on the prioritization process itself. Variance decomposition is one technique that can quantify the degree to which each source of uncertainty contributes to output variability over the entire dataset22). This information may be used to improve a prioritization framework by fixing those methods that have little effect on output uncertainty and by identifying and guiding discussion on those methods that produce widely differing results.
The objective of this study was to evaluate the effect of methodological uncertainties on the prioritization of road bridges for repair activities. To realize this goal, uncertainty analysis was applied to a prioritization framework recently established in Lao PDR to propagate uncertainties arising from methodological choices to the priority ranks of national road bridges. The priority ranks generated by uncertainty analysis were probabilistically examined, with the average of the priority ranks representing the generalized priority of a bridge in the absence of consensus on a single prioritization method and the standard deviation of the priority ranks evaluating the effect of uncertainty on each bridge’s priority. The relationships between the average and standard deviation of the priority ranks of the bridges were examined both overall and for subsamples with the highest (top quintile), moderate (middle quintile), and lowest (bottom quintile) average priorities. Notable outcomes are summarized as follows.
• The highest and lowest priority bridges were the least sensitive to methodological uncertainties. In the top quintile, the sensitivity increased linearly with a decrease in the average rank; conversely, in the bottom quintile, the sensitivity tended to increase with an increase in the average rank.
• There was no clear relationship between average priority rank and its standard deviation for the middle three quintiles, and the sensitivity to uncertainty varied widely for bridges with intermediate levels of priority.
• In-depth analysis of three pairs of bridges revealed that even if two bridges shared similar priority levels, the sensitivity of their priority ranks to the choice of methodology varied widely. Minor differences in bridge rating scores were found to lead to distinctly different priority rank distributions.
• The effects of the different sources of uncertainty also varied, as some bridges were found to be insensitive to criteria set selection or aggregation method, while other bridges were highly sensitive to these methodological choices.
The analytical approach presented in this study can serve as a useful guide on how to evaluate the extent to which different perspectives affect the outcomes of the prioritization process. Such analyses can identify strengths and weaknesses in decision-making systems, and will contribute to evidence-based policy making for infrastructure asset management.
The authors would like to express their gratitude to the Japan International Cooperation Agency (JICA) for supporting this research through the Road Asset Management Platform (RAMP). Special thanks also go to the Ministry of Public Works and Transport (MPWT) in Lao PDR, particularly the JICA Bridge Maintenance and Management Project (JICA-BMM), for providing the necessary data and information.
