Thanks Hamid. That's a great idea and I tried to use this before. The problem is to determine the number of time frames for which the correlation is computed. Also, there is a lag due to driver's reaction time. This lag is different for individual drivers.
This is a good question. While I don't have a ready answer to you, I refer to the work on the traffic conflict technique by Laureshyn et al. 2010 (attached), and references therein. You may get some ideas through these sources.
The "correlation" idea comes from this paper - https://www.researchgate.net/publication/247246511_What_characterizes_a_free_vehicle_in_an_urban_area
However it was used only in urban area, and on one intersection only. I am not aware of many other applications of this methods - maybe someone can add information. Thank you.
Article What characterizes a “free vehicle” in an urban area?
I have another idea. Maybe you can try both, and I would be pleased to check results with you...
If you have a significative statistical distribution of adopted headway in your population, you can retain (at a given level of significance, e.g. 0.01) that a leader does not influence the follower when the interaction range is greater than xxxx.
I have some kind of data like this to share, but it would be useful to analyse data specific to your context. Which kind of data do you have?
Thanks everyone. I really like the idea of correlation and will definitely go through the reference that @Ambros shared. I have actually tried that before on my own with promising results, but as mentioned before, due to the differences in drivers' reaction times it is hard to use a fixed width of rolling correlation.
@Pariota I am using US-101 vehicle trajectories data from NGSIM website (http://ngsim-community.org/).
Remember, that the results might differ. I suppose you will get different results for non-urban motorways, different for clogged streets in agglomerations, different for rural areas