A few years back several members of the Central Vermont Railway Historical Society tabulated the total number of boxcars (and only boxcars) going through White River Jct, VT over a several day period in 1954. (The period and locale were chosen because there were relatively complete train lists for that period of time).
I thought this might help guide the creation of an accurate freight car fleet for the layout (by percentage of road name) so I was very interested in the results. After looking the resulting data I'm not convinced it's helpful for modeling purposes.
The sample total was 3,605 cars. There were about 60 or different reporting marks represented (basically, name a North American railroad of the time and it appeared at least once ...)
By far the most common roadname, with more than 50% of the total, was Canadian National. Since the CV was a subsidiary of the CN, that was not really all that much of a surprise.
I prepared the somewhat useless pie chart above to have an image with the blog posting - I'm afraid there's little useful data to be gleaned from it - except that if you go with a statisical approach to a modeled fleet 3/4 or so of the fleet should be made up of boxcars from the 10 railroads listed.
The table below shows the breakdown by roadname of most of the remaining 55+ reporting marks.
Each accounted for far fewer cars - or for a total so small it was insignificant.
I noted the percentage of the total by roadname didn't come even remotely close to reflecting the national fleet, although the totals seem to reflect some regional "bias" (greater percentages of New England/Northeastern region road names, but not by much). I was especially shocked at how few NYC and PRR cars (based on the % of these roads rosters compared with the national fleet at the time) appeared in our sample data. Not sure what I learned from this exercise, except that out of a fleet of 100 boxcars fully half should be CN, with almost any other road name represented as long as you don't include too many of any one road name. The thing is, if one were to model a roster to these percentages and then compare the resulting trains to prototype photos, the result may be defendable as somewhat "authentic," but I don't think they'd look right!