Follow up to newspaper sentiment analysis

Justin Weinburg kindly shared the post from yesterday. I think he raises some good questions. There's one thing  I didn't mention in the post that I think is worthwhile. Sentiment analysis goes word-by-word, and the dataframe I was working with had more than 37,000 words. So the results wouldn't be skewed by a few hundred negative words...at least not the negative words that were in the data. It would have to be a fair number of "motherfucker"s to really make a difference. And that's, to my mind, the power of sentiment analysis. It offers a birds-eye view of sentiment made up by the contributions of tens of thousands of terms. There's no single article or term that causes the downward trend. It's a reflection of a wider cultural attitude.