The HPMS sample is a stratified random sample of physical roadway sections. The data collected through the HPMS relates to physical characteristics and operating conditions on each of the sampled roadway sections.
Samples are stratified as follows:
State – Rural/small urban/urbanized – Roadway functional system – Traffic volume group
All publicly owned roadway sections meeting the stratum criteria are eligible for selection as a sample on a random basis. The determination of the number of samples needed in each traffic volume group (the sampling rate) is based on the traffic volume (AADT) data item because of its high variability compared to other data items in the HPMS.
Samples are expanded based upon the ratio of total roadway miles in a traffic volume group to the total of sampled miles of roadway in that volume group.
The sample is considered a semi-fixed sample panel. Samples are generally fixed over time; however, they may migrate from one traffic volume group to another, usually because of reclassification of one of the stratum criteria. Since traffic volume is a dynamic variable, sections most often move from one traffic volume group to another as a result of AADT growth. Samples also move when there are changes to their rural/urban characteristics as a result of census population changes, and when the functional class changes as land is developed and the traffic function of a roadway changes. Maintaining the HPMS sample requires the addition (and deletion) of samples over time to account for the shifts of existing sample sections between traffic volume groups.
As a result, active management of the HPMS sample is necessary to assure that the sample continues to represent the full extent of roads in the HPMS, to account for road system changes over time, and to account for the normal growth in traffic. Sample management also allows the State to maintain a cost effective sample plan with neither too many nor too few samples.
There are a number of practices that should guide all sample management activities:
Samples usually become biased when the State violates the random selection rule or imposes other criteria on the sample selection process. Some activities that result in sample bias include:
Existing HPMS samples may be biased because the State did not apply a random selection process to the original sample. However, as a result of poor sample management practices, even a sample initially selected on a random basis may become biased if samples are added or deleted on other than a random basis over time.
In general, biased samples are best corrected by randomly selecting an entirely new sample. An alternative the State may consider to remedy past bias is to make any further sample adjustments on a random basis; over time, this should lead to a more unbiased sample. Also, the State may consider making specific adjustments to a sample to eliminate bias directly by adding samples in such a way that the effects of the bias are negated. This last process is somewhat complex in that the bias identification, sample analysis, and sample adjustment must be carried out at the traffic volume group level.
For instance, to determine if a rural major collector traffic volume group is biased to State ownership, it would be necessary to look at the ownership of miles in that particular traffic volume group, both full extent and sample. If the sample reflects the inclusion of both State and non-state owned miles, and the representation in the sample and full extent are reasonably similar, it is likely that the sample was chosen in an unbiased manner.
While this may be obvious if there is a substantial representation of both types of ownership in the sample as well as in the full extent, it may be less than clear where the full extent contains very few miles of either ownership type. A full extent comprised of 90% State owned roads could easily yield a sample containing only State owned roads as a result of random selection. On the basis of probability, a full extent comprised of 80% State owned roads could be expected to have between 1 and 3 non-State owned samples out of every 10 as a result of random selection. The best way to determine bias, however, is to know that the State engaged in a biased, non-random, selection process.
If an adjustment to a sample with a clear State owned road bias is necessary, it only should be made at the volume group level. An appropriate adjustment could consist of randomly adding non-State owned road sections to the sample and randomly deleting the same number of State owned road sections until the number of State and non-State owned road sections had similar representation in both the full extent and sample for that particular traffic volume group. This would keep the total number of samples, and hence the sampling rate, the same. This kind of adjustment should not be done routinely and should be considered only where there is clear evidence that a sample has indeed been biased by the use of improper sampling practices.
A periodic sample management review should be done in cooperation with the State HPMS counterpart. As part of the review of the State's sample management activities, FHWA needs to be able to determine the following: