8
problem that can be quickly and easily handled.
1
The researcher need not even understand the
language in which the documents she is analyzing are written!
The intuition underlying the procedure can be illustrated by considering an extreme
example. Suppose someone who does not speak English is presented with two written opinions of
the U.S. Supreme Court. One is authored by a justice who is ideologically conservative, while the
other is the written by a more liberal justice. Upon closer inspection, such a person might discover
that the two texts --- which the reader cannot understand --- make use of systematically different
language. The words “rational,” “legitimate,” and “intent,” for example, appear frequently in the
opinion of the conservative justice, but only once in a while in the other opinion. The words
“strict,” “discrimination,” and “equality,” on the other hand, are more frequent in the liberal
justice’s opinion but hardly appear in the other opinion.
Our observer is then handed a new set of documents and is told that they are the written
opinions of the other seven members of the Supreme Court but about whose ideological positions
nothing is known. What would the observer conclude about the likely location of the justices on the
right-left continuum? The thrust of the Wordscore procedure is the claim that one can make use of
the frequencies with which different words appear in the different texts to draw some conclusions.
For example, if one were to observe that one of the opinions made heavy use of words primarily
found in the conservative justice’s opinion and infrequent use of words that are typical for the
opinion of the liberal justice, one would draw the conclusion that that new opinion probably
expresses a position that is closer to the position of the right-leaning justice than the left-leaning
justices. In other words, even in the absence of understanding the meaning of any of the words in
the documents, one could make use of the similarity in word use in documents whose positions one
1
The procedure is currently available as an add-on to STATA, which the authors have made available.