Originally Posted by noimagination

Thank you. People ask this type of question all of the time without thinking about what they're actually asking.

OP: Think about what you're asking of the "study". You want a study comparing the "safety" of riding 2 abreast vs. single file. Ignoring, for the moment, that you have not defined what "safety" means, for the sake of argument I'm going to assume that you mean "bicycle accident rate (hit by car) per mile traveled" (even this is probably not precise enough).

In order to compare the two riding modes, you'd have to have the following data:
Miles traveled by bicycles riding single file.
Miles traveled by bicycles riding 2 abreast.
Accidents between autos and bicycles while bicycles were traveling single file.
Accidents between autos and bicycles while bicycles were traveling 2 abreast.
Then you'd have to control for a variety of additional factors where the likelihood of the accident is not affected by single file vs 2 abreast riding, for example:
- accidents between cyclists and oncoming car
- accidents between cyclists and car entering the roadway
- etc.

The data needed are not collected. I don't think that there is any database listing the number of cycling miles single file, 2 abreast, 3 abreast, etc. So, you'd have to come up with a methodology for estimating the numbers, and you'd have to justify the methodology.

Also, accident reports are not collected this way - there is no database listing car/bike accidents and whether the cyclists were riding single file or 2 abreast (or 3, or 4, or....), nor many other factors that might affect whether or not single file vs 2 abreast might have an effect on the likelihood of the accident. So, the researcher would have to read all auto-bike accident reports individually to determine whether they fit into the data set. Again, it would be difficult, if not impossible, to formulate unambiguous criteria, so the researcher would probably be forced to make judgments that would be open to challenge.

Once all of that was done, they would have to determine if there were a sufficient number of incidents so that the results would be statistically significant.

I'm sure I'm vastly oversimplifying, but only after all of the above were done would the study mean anything. This is also why it takes work to determine if you can believe someone when they say "Studies have shown....". There are a lot of BS "studies" that prove absolutely nothing.

Controlled scientific experiments are one thing. Statistical studies of data, especially of data collected for reasons unrelated (or only weakly related) to the purpose of the study, are a whole different kettle of fish.