Introduction to Data

• N = 131 Articles

• (Some duplicates, removed)

• Good variation in news outlets (n = 17)

• One weird issue:

Frequency of Media Outlets

Frequency of Media Outlets

Text Analysis: Most Common Words (n > 75)

Text Analysis: Most Common Words (n > 50)

Text Analysis: All Words

Text Analysis: Sentiment Analysis

• Note about the DV:

• The dependent variable is a “Sentiment score”, which is similar to the “tone” variable that we have used previously.

• Three major “Sentiment” Lexicons: Bing, NRC, AFINN.

• See section 2.1 here

Sentiment Scores By Article (BING)

Sentiment Scores By Year (BING)

Sentiment Scores By News Outlet (BING)

Sentiment Scores By News Outlet and Year (BING)

Sentiment Scores By Article (NRC)

Sentiment Scores By Year (NRC)

Sentiment Scores By News Outlet (NRC)

Sentiment Scores By News Outlet and Year (NRC)

Sentiment Scores By Article (AFINN)

Sentiment Scores By Year (AFINN)

Sentiment Scores By News Outlet (AFINN)

Sentiment Scores By News Outlet and Year (AFINN)

Correlation Between Sentiment Scores

Comparison of Three Sentiment Scores

Comparison Across Years

Comparison Across Years

Comparison Across Outlets

Conclusion / Things to Do

• Conclusion: Much better data than GDELT. Can construct our own variables.

• But data extraction is time consuming (whereas data cleaning is not)

• Need to extract More Data ASAP (IMPORTANT!!! )

• Move further with more text-as-data analysis

• Example of similar research before (Specifically analysis of text of news articles)?

• Other things?