This summer, Professor Ho and I worked on a behavioral economics project examining the effectiveness of various apology types on the ability of celebrities and public figures to return to society after being “canceled.”

For our work, we looked first to well publicized celebrity apologies. We collected a sample of celebrity apologies based on two online databases: a New York Times dataset of celebrities who were publicly reprimanded during the #metoo movement and a dataset compiled by the online apology blog SorryWatch who noted and analyzed a number of celebrity apologies in the recent decades, culminating in over 200 apologies. We then used Amazon Mturk to categorize and assess the effectiveness of these apologies through a survey we constructed.

We additionally worked with academic access from Twitter to download databases of tweets referencing the celebrities within our dataset. These databases of tweets consist of a sample of tweets from six distinct periods of interest: 

1.  Random collection of tweets from the year prior to the cancellation
2. A database of tweets directly before the news of the cancellable event was published
3. A database of tweets directly after the news of the cancellable event was published
4. A database of tweets directly before the celebrity apologized
5. A database of tweets directly after the celebrity apologized
6. Random collection of tweets from the year following the issuing of the apology

A screenshot of some of the code used to create our tweet databases.

We then used the open source program VADERsentiment to analyze the overall sentiment of the tweets contained in our databases to quantify how public sentiment towards the celebrity changed over time. Our initial results are promising and in line with some of our initial hypotheses, and I hope that our continued collaboration on this project will yield interesting results. 

This image depicts an example output of our sentiment analysis of the tweet dataset for Kevin Spacey.