In the Peak Search Technique routes are segmented based on a user defined length. Each segment is then subdivided based on a user defined window length. Performance measures are calculated at the subsegment levels along with the individual coefficient of variation.
Here, coefficient of variation, CV = σ/µ
Where, σ refers to the standard deviation of the pertinent performance measures for the sub-segments and µ refers to the mean value of crash counts over the analysis period. In order to identify locations experiencing statistically significant occurrences of crashes a certain minimum threshold is considered. If none of the CVs for sub-segments within a particular segment is less than or equal to the minimum threshold, the segment is not considered for ranking and window length is increased. This iteration can be repeated until the length of the sub-segment equals to the entire segment length.
An example of the peak searching technique is described below with the following assumptions:
- Network Element – 0.85 Segment of Route A.
- Window Length = 0.10 Mile.
- Resulting Sub-Segments – Nine (9) 0.10 Mile Sub-Segments.
- Standard Deviation: σ =1.014 and N.D. refers to Not Determinable.
Subsegments | Start | End | Crash Counts (x) | CV (based on crash counts) |
1 | 1 | 1.1 | 0 | N.D. |
2 | 1.1 | 1.2 | 0 | N.D. |
3 | 1.2 | 1.3 | 3 | 0.338 |
4 | 1.3 | 1.4 | 0 | N.D. |
5 | 1.4 | 1.5 | 0 | N.D. |
6 | 1.5 | 1.6 | 1 | 1.014 |
7 | 1.6 | 1.7 | 0 | N.D. |
8 | 1.7 | 1.8 | 0 | N.D. |
9 | 1.8 | 1.85 | 0 | N.D. |
Now, if the agency’s specific minimum threshold for coefficient of variation (CV) is 0.3, none of the subsegments will be considered as having over-representation of crashes. If the agency’s specified threshold is 0.35, then one of the sub-segments (#3) fulfills the minimum requirements. In such case, the segment X can be ranked as having a highest sub-segment crash count of 3 (out of total crash count 4). Therefore, the sub-segment #3 (having a length of 0.1 mile) holds 75% of the total crash counts. Therefore, the basis of peak-searching is to find out the highest concentration of crashes within a segment.
