After introducing the Lott study, revealing the matched partner's donation decision, and revealing whether or not the matched partner learned about Lott's study prior to making the donation decision, we ask respondents the following open-ended question: "Why do you think your matched respondent chose to donate to Fund the Wall?" This document specifies how we intend to analyze participants' responses to this question.

We are interested in classifying perceived motives according to the two dimensions we analyze in our experiment: "bias" and "gullibility". We start with five words for each category. For bias, our words are "racist", "biased", "xenophobic", "intolerant", and "prejudiced". For gullibility, our words are "convinced", "persuaded", "gullible", "naive", and "sucker". For each of the ten seed words, we identify synonyms on www.thesaurus.com, include all synonyms that are classified by the website as "most relevant". These synonyms are reported in `synonyms.json`, which is also included in our AEA pre-registration under "Docs & Materials". Using the SnowballStemmer function from the Python library `nltk`, we will stem each word to allow us to detect respondents who are using different forms of the word (e.g. "bias" instead of "biased"). We will manually stem instances of "racism" or "racist" to "racis", as the SnowballStemmer function leaves both words unchanged. Combining seed stems and synonym stems, this procedure leaves us with 23 "extended bias" stems and 30 "extended gullibility" stems. 

Using these stems, we will generate "bias_word" and "gullibility_word" indicator variables that take value 1 if the participant's response includes a bias stem or gullibility stem, respectively. We will then estimate regressions identical to those in the primary "receiver" analysis in the paper, replacing the dependent variable with these indicator variables. 

The Python code file we will use for this analysis is included in our AEA pre-registration under "Docs & Materials".