While there is a growing body of bicycle and pedestrian count data, there is little documentation of how or if the quality of the data has been assessed and controlled. Here we offer a discussion of what a Quality Assurance/Quality Control (QA/QC) program might involve for a permanent bicycle and pedestrian count program. There are different levels of data validation and cleaning. We separate out these levels into the following three steps.
1. Validation at equipment set up. This important step may take multiple iterations to make sure the continuous counting technology is operating properly and correctly located.
A. Manually observe at least 10 bicyclists and/or pedestrians and, if possible, verify in the field that each is counted. If some are not counting, contact the manufacturer, adjust the equipment settings, or otherwise problem solve. Once most are being counted, continue on to Step B.
B. Create adjustment factors. Observe at least 100 bicyclists and/or pedestrians. For sites with high to medium volumes, this can usually be done by counting bicyclists and/or pedestrians for 1 or 2 peak hours. The observation can be by video or manually in the field. Video observation has the advantage that it can be checked by others later. Compare the manual counts to the automated detector counts. If possible, trouble shoot what might be causing any under or overcounting observed. If it can be fixed, make changes and redo this Step. If there are not obvious problems or the problems can’t be fixed moved on to Step C.
C. Compute a correction factor (actual /automated count) to account for under or overcounting.
2. Maintenance validation. At least once per year and any time changes that might impact the counting technology are made to the location, do a quick verification that the equipment is working. To catch any big problems with the equipment, observe at least 10 bicyclists and/or pedestrians and compare to the automated count.
If adequate bicycle traffic is not available at the location during counting, g
enerate counts by walking or bicycling across the detector. Generated counts are not as good as natural counts for validation purposes because they may not behave the same as the general public. For example, they could be slower or faster than normal traffic, use a slightly different bicycle type, or take a slightly different travel path. However, using generated counts is better than not doing any maintenance validation.
If the maintenance validation finds that equipment is not working correctly, contact the manufacturer, adjust the equipment settings, or otherwise problem solve. After attempting to fix the problem, return to Step B, above. The adjustment factor (Step C) will need to be recomputed. Make sure to note when any changes were made.
3. Data validation. Check for unusually high counts and suspect zero counts. It’s best to create an automated process for doing this, but either way, flag suspect counts. Check to see if high counts may have been caused by an organized event, and if so, leave them in the record. Delete days with zero counts if these are not explainable by weather events or other causes such as holidays. If equipment malfunction is the cause, delete these data from the record.
Check for unusually high counts and suspect zero counts. It’s best to create an automated process for doing this, but either way, flag suspect counts. Check to see if high counts may have been caused by an organized event, and if so, leave them in the record. Delete days with zero counts if these are not explainable by weather events or other natural causes such as cold weather or holidays. If low or high counts are caused by construction, consider removing them from the record. If equipment malfunction is the cause, delete these data from the record.
Turner and Lasley discuss QA for non-motorized counts and give an example of how they cleaned data from an infrared bicycle and pedestrian counter in their 2013 paper.