• Step 2—Review monthly patterns and refine groups. Agencies should plot and visually review the monthly patterns of each group and make necessary modifications based on specific travel and roadway characteristics within each state. For instance, some functional classes may need to be merged if their travel patterns are similar, or new groups may need to be created for sites exhibiting significant recreational traffic or business activities.

  In situations where two CCSs happen to be nearby on the same road, one of them should be excluded from the factor group to prevent double weighting of the same traffic patterns. Failure to exclude one of the two CCSs can skew the group adjustment factors and can artificially inflate the group’s computed precision level. Agencies should consider using, if feasible, one CCS and reserving the other as a backup option in the event of service loss at the primary CCS. This applies to all grouping methods, not just the traditional approach.

• Step 3—Compute variability metrics. Analysts should compute the three within-group variability metrics presented in Chapter 2 (CV, ACV, WACV) to quantify and better understand the homogeneity of each group as well as that of each set of groups. These metrics can be used to compare factor groups in a data-driven manner:
  • CV: The CV (Equation 10) for each month and group. Typical monthly CV values at nonrecreational sites are usually less than 10 percent. Any monthly CV greater than 10 percent should be further investigated to determine the reasons behind the high traffic variability within that month. High CV values can potentially be attributed to dissimilar sites exhibiting different patterns that may need to be grouped differently, potential outliers that need to be removed, or other unusual events that need to be taken into consideration. Figure 7 illustrates an example of monthly patterns containing a potential outlier.
  • ACV: The ACV (Equation 11) across all 12 months within each group. Factor groups with high ACV values (> 15 percent) tend to be heterogeneous and may need to be modified. Analysts should determine if there are other attributes that can potentially divide a highly heterogeneous group into smaller, more homogeneous groups. If this is not feasible, a highly heterogeneous group may need to be dissolved/disassembled by assigning its CCSs to other groups, as long as these reassignments do not affect the ACV of the remaining groups. In general, adding new CCSs to a highly heterogeneous group containing more than six to eight CCSs should be avoided. Though the addition of new CCSs to a group tends to improve (i.e., decrease) the precision of the group adjustment factors (WAP), it tends to increase its variability and decrease its AADT accuracy.
  • WACV: The WACV (Equation 12) across all factor groups included in each set. It encapsulates the overall within-group variability of all factor groups developed using a specific attribute or combination of attributes. A lower WACV value indicates better homogeneity. This metric aids in identifying which attribute or combination of attributes yields the most homogeneous groups. In cases where two grouping methods generate a similar number of factor groups, preference should be given to the method with the most homogeneity, signified by the lowest WACV. WACV tends to increase as the number of CCSs within a group increases, and vice versa.
• Step 4—Select final set of factor groups. Agencies should confirm if the variability of the produced factor groups aligns with their assumptions and determine if a specific grouping attribute (or combination of attributes) produces more homogeneous groups than others. The selection of the final factor groups should rely on an assessment of the within-group variability metrics described above, coupled with analysts’ knowledge of traffic patterns within each state.
• Step 5—Determine required sample size per factor group. Once the factor groups are selected, analysts should calculate the required number of CCSs per factor group to meet the desired precision level (see Equations 1 and 13 in Chapter 2) for each group adjustment factor.

• Step 6—Evaluate effectiveness of group adjustment factors applied to SDC counts. After developing AADT estimates from factored SDCs, analysts should evaluate whether an individual SDC has been assigned to the most appropriate factor group. In general, AADTs tend to be fairly stable over time. If they fluctuate from one year to the next, the factor group may be a poor fit for the SDC. One approach to check factor group assignment is to reevaluate historical counts using temporal factors from different factor groups and assess their stability through the CV of the AADT estimates. A more effective yet resource-intensive approach involves conducting quarterly counts within the same year at an SDC location. Each quarterly SDC can be separately annualized multiple times by using the appropriate adjustment factor(s) from a different factor group (i.e., four AADT estimates can be generated using factors from each factor group). The factor group that results in the lowest CV of the four AADT estimates should be considered as the best fit for that SDC location. This evaluation should be performed regardless of the method used to develop the factor group. NCDOT uses this approach to assign seasonal counts to six factor groups. The approach is described in detail in Appendix A.

Table 10 lists the strengths and weaknesses of the traditional approach.

The main advantage of the traditional approach is the ability to create well-defined groups based on one or more rules/attributes that are readily available. These attributes are also used to directly assign SDCs to one of the factor groups. In addition, the traditional approach is simple and easy to understand and communicate to others. Because of its simplicity and intuitiveness, the approach has been widely used by many agencies for several decades. Additionally, the traditional approach can be applied to all roadway functional classes, including the lower classes, FC6 and FC7, where a small number of CCSs may exist.

The main disadvantage of the traditional approach is that it may produce internally heterogeneous groups that may contain sites with highly variable patterns (Schneider and Tsapakis 2009). The group adjustment factors may not be representative of all different patterns within a group, potentially resulting in low accuracy of AADT estimates, particularly compared to clustering (Tsapakis 2009). For example, Table 4 shows that even two clusters generated by cluster analysis (M10) are significantly more accurate (i.e., lower MAPE), more homogeneous (i.e., lower WACV), and have more precise group factors (i.e., lower WAP) than any of the four traditional methods, which produce a much higher number of factor groups. The group factors of the traditional approach may not meet the target precision level (±10 percent) recommended by the TMG at 95 percent confidence for nonrecreational roads (FHWA 2022). The approach heavily relies on engineering judgment, which may be biased. Further, it may be challenging to maintain factor groups due to periodical changes in the functional classification of some roads.

Table 10. Strengths and weaknesses of traditional approach.

The research team validated nine traditional methods. Five of them (M1–M5) were applied to all functional classes, and four methods (M6–M9) were applied to lower functional classes, minor collectors (FC6), and local roads (FC7). The nine methods are listed below.

• Methods applied to all functional classes (FC1–FC7):
  • M1: No factoring.
  • M2: Functional classification.
  • M3: Functional classification combined with rural/urban area type.
  • M4: 5 volume groups.
  • M5: 10 volume groups.
• Methods applied to lower functional classes (FC6 and FC7):
  • M6: Using factors from FC5 to annualize counts on FC6 and FC7.
  • M7: Using factors from 5U to annualize counts on 7U, and factors from 5R to expand counts on 6R and 7R (no CCSs were available on 6U).
  • M8: Using factors from FC6 to annualize counts on FC7.
  • M9: Using factors from 6R to annualize counts on 7R (no CCSs were available on 6U).

Table 11 shows the most important aggregate performance metrics of the existing methods across all 45 states examined in this project. The validation results are separately provided for the analyses conducted using weekday sample counts (Monday–Friday) extracted from CCSs located on (1) all FCs, (2) FC6, and (3) FC7. The last column shows the average number of CCSs per group.

Table 11. Performance metrics of existing methods across all states and years for different FC groups and weekday counts (Monday–Friday).

Note that these nine methods evaluated in NCHRP Project 07-30 relied solely on one or more grouping variables without applying knowledge of local patterns to identify recreational sites and potentially reassign CCSs to other factor groups based on geography or unique regional factors. Therefore, the methods studied in this research represent “base” versions and the results could be improved if local knowledge of the network is used to further refine the factor groups created by each method. The main lessons learned from the validation of the methods (M1–M5) that were applied to all functional classes are:

• Not factoring counts (M1) tends to be the least accurate approach and should be avoided. It results in a significant overestimation of AADT if counts are taken on weekdays and an underestimation of AADT when counts are conducted on Saturday or Sunday. In general, the average weekend traffic on nonrecreational roads tends to be lower than the AADT. Traditional methods should be preferred over not factoring counts.
• There is a high variability in the AADT accuracy (MAPE) of the remaining four traditional methods (M2–M5) across all 45 states. In many states (e.g., Figure 5b) FC_RU tends to be a more accurate grouping/assignment approach than using only FC. However, in some states (e.g., Figure 5a), both variables result in similar levels of AADT accuracy. Likewise, volume groups are a more effective grouping method than functional class in some, but not all, states.
• There is a certain point beyond which dividing a group into subgroups based on the same variable does not improve the AADT accuracy (MAPE), even though the within-group variability (WACV) continues to decrease. For example, in most states, splitting five volume groups (M4) into ten volume groups (M5) does not improve AADT accuracy, even though it tends to reduce the within-group variability, as expected. This suggests that temporal traffic patterns do not significantly change among the smaller ten volume ranges of M5. This happens because each volume group can contain highly variable monthly day-of-week patterns/factors, which depend on various spatiotemporal characteristics, not just the volume of each CCS. In general, a grouping variable (e.g., FC in M2 or volume in M4) can only capture some of the variability in the CCSs’ monthly day-of-week patterns.
• Adding CCSs to a group tends to make the group more heterogeneous (i.e., higher WACV) but improves the precision of the group adjustment factors (i.e., lower WAP). For example, FC (M2) produces up to seven factor groups per state and year, with each group containing on average 14.8 CCSs. FC_RU (M3), on the other hand, results in up to 14 smaller groups per state and year, with each group containing, on average, 8.9 CCSs. The smaller groups of FC_RU are, on average, more homogeneous (WACV = 10.4 percent) than the bigger groups of M2 (WACV = 11.8 percent), but the group adjustment factors are less precise (i.e., higher WAP). This happens because the group factor precision (WAP, Equation 15) is affected to a greater extent by the more pronounced changes in the number of CCSs within a group (i.e., the sample size, n, included in the denominator of the precision equation, Equation 1) than from the smaller changes observed in the within-group variability, which is captured by the CV, included in the numerator of Equation 1.

The main lessons learned from the validation of the methods that were applied to lower functional classes, FC6 and FC7, are:

• Not factoring counts (M1) taken on lower functional classes is one of the least accurate approaches and should be avoided.
• Having CCSs on all lower functional classes helps to improve AADT accuracy. The methods with the highest accuracy are FC_RU, FC, and volume groups, all of which included CCSs on FC6 and FC7. All three methods resulted in similar levels of AADT accuracy when they were applied to FC6. In the case of FC7, FC_RU and FC had the best performance.
• Splitting five volume groups (M4) into ten volume groups (M5) did not provide any significant improvement in AADT accuracy.
• Applying factors from FC5 to counts taken on FC6 and FC7 should be avoided, if possible. M6 and M7 resulted in less accurate AADTs than FC, FC_RU, and volume groups (M2–M5).

• Applying factors from 6R and 6U to counts on 7R and 7U (M9), respectively, produces AADT estimates that are slightly less accurate than when factors are applied from the same FC (M2) or FC_RU (M3). In addition, M9 produced significantly more accurate AADT estimates than M6–M8, which means that annualizing counts on local roads using factors from 6R and 6U is a more effective approach than using factors developed for FC5, 5R, 5U, or FC6.
• In many states, accounting for the rural/urban area type in the development of factor groupstends to improve AADT accuracy. For example, in many states, FC_RU is slightly more accurate than FC. Further, M9 tends to be more accurate than M8. In the latter, factors from 6R were applied to counts taken on both 7R and 7U, whereas in M9, factors from 6R were applied only to counts on 7R.
• The results from the validation of midweek counts (Tuesday and Thursday) are similar to those described above for weekday counts (Monday–Friday).

Volume Groups

The volume factor grouping approach involves creating a set of traffic volume groups, with each group having a specific AADT range. CCSs are assigned to the volume groups based on their AADT. Group adjustment factors are separately calculated for each volume group, and SDCs are assigned to the volume group in which their ADT falls. Volume groups can also be combined with other methods, such as functional classification and rural/urban area type. Table 12 shows the strengths and weaknesses of the volume factor grouping approach.

Similar to the traditional approach, volume groups are well-defined based on traffic volume ranges or sometimes additional characteristics that may be readily available. It is easy to explain volume factor groups to stakeholders and assign counts to groups. Further, the volume groups are applicable to all functional classes, including FC6 and FC7.

The main disadvantage of the approach is that it may produce heterogeneous groups that contain sites with different temporal and time-of-day patterns. As a result, the group adjustment factors may not meet the precision level (±10 percent) recommended by TMG for nonrecreational roads (FHWA 2022). Likewise, the group adjustment factors may not be representative of all different roads within a group, potentially resulting in low accuracy of AADT estimates derived from annualized SDCs (FHWA 2022). The approach is subject to engineering judgment because analysts must select the total number of volume groups and the range of each group. Further, the counts may be assigned to the wrong group because the true AADT at each SDC location is not known. To reduce the uncertainty and human bias related to this selection, analysts should follow the five steps described below. These steps are similar to those described in the “traditional approach.”

• Step 1—Develop multiple sets of factor groups. Analysts should initially consider developing multiple sets of different volume groups, with each set being constructed using different bins. As previously explained, typically, a minimum of five factor groups are needed in each set.

Table 12. Strengths and weaknesses of volume factor groups.

For example, an initial set of five volume groups may be based on the following ranges, which were developed in FHWA pooled-fund study TPF-5(384) (FHWA 2021):

• 0–1,999 vehicles per day (vpd).
• 2,000–9,999 vpd.
• 10,000–34,999 vpd.
• 35,000–84,999 vpd.
• 85,000 vpd.

Separate recreational groups may need to be created, as explained in the previous section.

• Step 2—Review and modify groups. Analysts should review and potentially modify the volume groups by adjusting the volume ranges based on the distribution of traffic volumes within each state. For instance, a predominantly rural state lacking high-volume roads with AADT exceeding 85,000 vpd might consider eliminating the fifth volume group and modifying the range of the other four lower-volume groups accordingly.
• Step 3—Compute variability metrics. After developing two or more sets of volume groups, analysts should employ the variability metrics described in Chapter 2 (CV, ACV, WACV) to quantify and better understand the homogeneity of each group as well as that of each set of groups.
• Step 4—Select final set of factor groups. Analysts should identify the set that yields the most homogeneous groups. The sets to be compared should have the same or a similar number of groups. This is important because, as previously explained, subdividing five volume groups (M4) into ten smaller groups (M5) tends to increase the homogeneity (i.e., reduce the WACV) of the new smaller groups without improving the AADT accuracy.
• Step 5—Determine required sample size per factor group. Once the factor groups are selected, analysts should calculate the required number of CCSs per factor group to meet the desired precision level (see Equations 1 and 13 in Chapter 2) for each group adjustment factor.
• Step 6—Evaluate effectiveness of group adjustment factors applied to SDC counts. After developing AADT estimates from factored SDCs, analysts should evaluate whether an individual SDC has been assigned to the most appropriate factor group that ideally should have the most similar traffic patterns with those at the SDC location. Step 6 of the Traditional Approach describes two ways to perform this evaluation.

Recreational Patterns

Special attention should be given to recreational groups that cannot be automatically identified using volume groups nor the traditional approach. Agencies should review monthly profiles of CCS data to identify potential recreational patterns, but most importantly use their knowledge of specific generators of recreational activities and roads or areas that typically carry such traffic. Additionally, cluster analysis can potentially identify and group some or all recreational sites together in a more automated manner, but users should review and refine the produced clusters if needed. Currently, no method is available to determine all recreational patterns that may exist in a state. A potential, though not explicit, indicator of recreational traffic at a specific roadway location is when the following inequality is satisfied (FHWA 2022):

Taking into account the various recreational patterns identified, agencies should determine the number of recreational factor groups needed within each state. Specific roads and areas associated with each recreational group should be identified. SDCs carried out in these recreational areas should be annualized using adjustment factors corresponding to the respective recreational group. It is essential to document the road segments to which these recreational patterns have been assigned. Periodic reviews should be conducted to maintain the precision and relevance of the recreational factors applied.