Functionally, you make a great point. That reminds me, BEM does this too: Blocks and elements are separated by a double underscore, and multiple-word elements are joined with dashes (e.g. .block__item-name). As you mention, this makes it impossible to quickly select by double-clicking.

For URLs, apparently dashes are preferred because Google will interpret words separated by underscores as a single word (i.e. "hall_of_fame" = "halloffame"). So we opt for worse user experience in order to rank better... Turns out Matt Cutts tried to get Google to interpret underscores as separators, just like dashes, but it was never implemented.

There's quite a few reasons, but here's a few that are still on my mind:

• The vast majority of internet users do not really know that underscore is a character. If you're making a TV/Radio/Podcast commercial, and say "underscore" most interpret that to be "underline" and do not know how to type that into the URL

• Everyone knows what a dash is, it doesn't require a modifier key to type in. It's also in default keyboards on mobile (aka not behind the weird "symbols" third keyboard in iOS)

• It's very common to use an underline to style URLs. Type has gotten better on the web since then, but it is still difficult to tell the difference between a space and an underscore in an underline style.

• While some text editing modes do select the whole segment with an underscore and not a dash, some are opposite, some are neither. Very few computer users use this segment-but-not-whole-entity selection mode, or even know of its existence.

• Over the past twenty years or so, web syntax (HTML, CSS, Javascript) has settled away from snake case and toward camel case and dash casing. Dash casing in particular works well in an acronym heavy environment like the web.