So reporting the 25th is a way of showing what happened to the whole bottom quartile of a class. They're the group most likely to fail the bar exam.
Here are the stats for Brooklyn, the school mentioned in the article: http://www.lstscorereports.com/schools/brooklyn/
The median score dropped 7 points in four years. The bottom score dropped 9 points in four years The top 75th score only dropped 6 points in four years.
Those are massive drops for such a short time. I'm an LSAT instructor, and there is a vast difference between a 162 and a 153.
So a school having the class drop from 170 to 160 shouldn't have much of an effect on the passage rate. But if they start letting in a large number low LSAT scores (below the 150 threshold) you'll really decrease the passage rate. Even if the median is still the same.
I work in the field, it would be good to have a hard number to present to students.
Anecdotally, I notice a big difference in those who can't break 150 vs. other students.
http://www.lstscorereports.com/schools/brooklyn/
She picked 25 because that's roughly the population of the students who failed (i.e., the weak students at the school are much weaker than they used to be, so they are now failing.)
The school in question had a significant drop (9 points) for the average score of their lowest quarter. So a corresponding drop on the bar exam, in Moeser's analysis, is a natural consequence.
To me, that screams data mining. The average score of the 25th percentile is a very strange statistic. Why not use something more conventional like the median of the entire population? Then you could honestly say: the majority of people at this school are getting worse. The use of an obscure statistic makes it seem they went hunting for a statistic to spin a story upon.