Volkswagen Scandal

Among the many issues at stake for the company was one of public perception. Anecdotal evidence at the time of the incident suggested irreparable harm to the Volkswagen brand. So could Volkswagen recover in the short term in this regard? And, the broader question, how can you measure brand perception in times of scandal, particularly in an era where social media can cause negative news to proliferate and reverberate over time?

In the absence of direct empirical evidence, we wanted to find a way to tackle this important issue. We began our research with some key questions: How does social media sentiment change as a consequence of a public relations crisis? How does the public react to recovery efforts initiated by the company? How do topics of conversation shift as a consequence of a brand scandal and subsequent recovery efforts?

We examined more than 100,000 tweets to analyze how the public sentiment changed over time after the breakout of the scandal. Our approach to capturing themes in the evolving scandal involved sampling a few date windows; therefore, we did not examine data for every single day. The following periods were selected: September 29, 2015–October 7, 2015; October 18, 2015–October 27, 2015; January 1, 2016–January 7, 2016; and January 17, 2016–January 25, 2016. These periods align with some of the events relating to the scandal, and also represent periods during and following the scandal. We explored the daily tweets from these periods by considering all possible events that might have affected the public sentiment over Volkswagen. Entire sets of tweets including the word “Volkswagen” were in our initial data set. We made several observations about how the scandal unfolded in the public conversation, broken out into the following categories.

Frequency. The number of times the scandal was mentioned on Twitter varied dramatically day by day, and the mentions seemed to parallel specific actions taken by Volkswagen to issue apologies or by regulatory agencies to place responsibility or issue punishments.

For example, after an article in The Guardian on September 30 revealed that the scandal has affected 1.2 million Volkswagen diesel vehicles, the number of tweets increased for the next two days. Subsequently, we observed a decrease in the range of number of tweets, from 5,000–7,000 to 1,000–2,000, except around January 6, which coincided with the following headline: “U.S. Sues Volkswagen in Diesel Emissions Scandal.”

Another exceptional surge in the number of tweets was on October 19, which could be explained by articles regarding the governments of France and Spain pushing the scandal investigations. We conjecture that the amount of tweets reflect the level of public interest in the scandal.

Vocabulary. We also identified the most-frequent words in tweets for each day by mining Twitter for all mentions of the brand name “Volkswagen” during the aforementioned time periods, including retweets. We then conducted topic modeling on the tweets using the text-mining library within the statistical program, excluding words that were obvious, and thus less meaningful in our analysis (“vehicle,” “Volkswagen,” and “car,” among others). We narrowed the number of words down to the five most frequently mentioned on each day. In some cases, when there were multiple words with similar frequencies, we had more than five words per day.