A long time ago, in a land not so far away, before every would-be writer owned a personal computer and word processing software, manuscript preparation followed a fairly rigid format. Because manuscript format was rigid, it was easy to calculate manuscript word counts: One page contained 250 words. The opening page of a short story, because it began halfway down the page to accomodate the title, the byline, and the author's contact information, contained 125 words. The last page varied and need to be guesstimated based on how much of the page was actually filled.
Back then I could safely estimate that my 11-page short story manuscript contained 2,500 words. Editors, who usually used the same method to estimate word count, paid based on that estimated word count. At 5-cents/word, I could expect to receive a check for $125.
Not long after I began submitting electronic manuscripts (on disk initially; via e-mail these days), I noticed that publications paying on a per-word basis were paying less. Their per-word rates had not been reduced. Instead, they were using a word processing program's wordcount function to determine pay. That 2,500-word manuscript (using the "traditional" method of counting) may only contain 2,100 words. At 5-cents/word, the pay comes to $105.
In effect, publications that did not raise their per-word pay rates after the advent of electronic manuscripts actually reduced the amount of money they paid writers.