I have a large convenience sample of vehicular flow counts, which I want to correlate with road network centrality metrics. Each object in my sample is a count location (point in geographical space) with an observed vehicular flow value, which I can relate to the specific centrality value of that point in the road network. My objective is to access the degree of statistical association between road network centrality values and observed vehicular flows.
The problem is that the count locations in my sample have a huge bias towards a specific class of road. My population being the entire road network, I know exactly the actual frequency of road classes, which is quite different from the road class frequency of my sample (i.e. the sample is not representative of the population). Could a possible method for correcting this sampling bias be to force a new distribution of road classes, by randomly sub-sampling the original sample according to the known frequency of road classes in the entire network?
In other words: if the true (population) frequency of road class A is 20%, and in my sample 50% of the count points are on A roads, can I randomly select a sub-set of the counts on A-roads so that their final frequency becomes 20%? And can I proceed like that for all other road classes, until my new (reduced) sample shows the same road class frequency as the whole population?
Sorry for the long question and many thanks for any help on this!