Tracking the Mood of Americans: Use Twitter if you want to prove they’re happy
July 25, 2010 | 3 Comments
An article in the New York Times this week discussed a research project that is attempting to track the mood of Americans using Twitter as the data source. The project involves researchers from Northeastern University College of Computer and Information Sciences and Harvard Medical School. It is certainly reasonable that a group of scientists can develop algorithms that accurately predict the mood of Americans. However, Twitter data is not simply and instantly predictive of the general population of Americans. Given that only 7% of people who are online even use Twitter, it is risky, and can easily lead to wrong conclusions.
Want to see a real example? No problem.
Let’s look at consumer opinions related to one specific product, the iPad.
- First, we gathered thousands of opinions from across the internet, from blogs, microblogs, forums, question and answer sites, personal sites, all of which mentioned the iPad. Sites like YouTube, Blogger, Twitter, and thousands more were included.
- Then, we categorized all of the conversations into two groups, 1) everything from Twitter and 2) the entire internet space.
- Next, we determined the level of emotion for every online conversation. Specifically, we determined whether the emotion of the conversations was extremely happy, somewhat happy, neutral, somewhat unhappy, or extremely unhappy.
- Finally, we created the pretty little charts that you see on the right of this page.
What’s the first thing you notice from these charts?
Not one single chart has two bars that look the same. What is the percentage of tweets that reflect an extremely happy opinion? 15%. What is the comparable number for the entire internet? 5.6%. I hope it’s not just me, but 15% doesn’t look like 5.6%, not even if the 5.6% is averaged up to 6%. There is a big difference in the percentage of people who have extremely happy opinions on Twitter vs the entire Internet.
The same trend is apparent when we look at the percentage of people who are extremely unhappy with the iPad. 11.3% of tweeple are extremely unhappy compared to just 1.9% of the entire internet space. All five of the charts lead to the same conclusions. Twitter results do not equal Internet results.
It’s not 1 to 1
Clearly, the relationship between Twitter data and total internet data is not 1 to 1. It’s impossible to gather Twitter data, analyze the sentiment, and be confident that it represents a wide, more general audience.
Perhaps people on Twitter have more extreme opinions than everyone else; perhaps they are less likely to guard their remarks so that the more extreme opinions are shared; perhaps Twitter opinions are in fact the closest to the average American opinion. Whatever the reason, it is undeniable that the mood on Twitter is unlike anywhere else.
Prepare to be wrong. Prepare to explain contradictions. Generalize Twitter mood at your own risk.
Category conversition | Tags: business research,harvard,internet research,invalid,mood of americans,new york times,qualitative research,research examples,sampling,twitter,twitter mood,validity,weighting
Social Networks : Share
3 Comments so far