Home首页 · Paths路径 · Summarize, then compare先概括，再比较
Summarize, then compare先概括，再比较
Two headlines about a training program, and a sheet of workers who did or did not take the subsidy. The headlines are many word counts. Before anyone is matched, those counts have to be collapsed into one score. The stops are the counts, a hand-placed map, an adjustment, then a match.两则关于培训项目的标题，和一张工人有没有领补贴的表。标题先是很多个词数。给人配对之前，这些计数要收成一个分数。这几站是：词数、一张手写的词图、一次调整，然后一次匹配。
-
Text · many counts文本 · 很多个计数
Two short lines about a wage program share most of their words and disagree on the verb. A bag of words still scores them as close. That vector is not yet a covariate you can match on.两则关于工资项目的短句共享大部分词，动词却相反。词袋仍把它们打成接近。这个向量还不是你可以拿去匹配的协变量。
-
Text · a summary you choose文本 · 你选定的概括
The word map places “pay” near “earnings” by hand, so two different wordings sit close in two dimensions. A summary is a choice about what to keep. This page does not hand that summary to a regression.词图用手把 pay 放在 earnings 附近，于是两种不同措辞在二维里靠近。概括是你决定留下什么。这一页不把这个概括送进回归。
-
Causal · adjust for the summary因果 · 对概括做调整
Treat the one-number summary of the headlines as household savings: something that pushes both subsidy use and later pay. Leave the subsidy’s own effect at zero and turn that push up. The line can still rise. Putting savings into the line is one repair, on the terms that page states.把标题收成的那个数当成家庭积蓄：它同时推动领补贴和后来的月薪。把补贴本身的作用留在零，并把这个推力调高。线仍可能上升。把积蓄放进这条线，是那一页所说的一种修法。
-
Causal · match on one score因果 · 在一个分数上匹配
Treat that same one-number summary as prior math. Joining is pushed hard by it, and the allowed gap is 0.4. Pairing keeps students who are close on that number and drops the rest. On this page the number is prior math, not a score built from the headlines.把同一个收成的数当成先前的数学。它把报名推得很用力，允许的差距是 0.4。配对留下在这个数上接近的学生，其余丢掉。在这一页，这个数是先前的数学，不是从标题建出来的分数。