Pew Research Center concluded after a year-long study that reactions on Twitter do not necessarily match up with public opinion about politics as measured by surveys. According to DataSift and Topsy, two social analytics companies that work with Twitter to provide insights on public sentiment, the study may have missed the mark.
Released on March 4, 2013, the study compared Twitter reactions to survey results for eight major political speeches, events, and policy decisions from the presidential debates through President Barack Obama’s election.
“At times the Twitter conversation is more liberal than survey responses, while at other times it is more conservative. Often it is the overall negativity that stands out,” the researchers concluded. “Much of the difference may have to do with both the narrow sliver of the public represented on Twitter as well as who among that slice chose to take part in any one conversation.”
The 2012 presidential elections were a prime example of this. President Barack Obama’s victory over Republican candidate Mitt Romney on Nov. 6 drew an overwhelmingly positive response from the Twitter community. According to Pew, 77 percent of post-election Twitter comments were positive about the outcome of the election, while only 23 percent were negative.
In contrast, a survey of voters taken after the election revealed that 52 percent reported being happy about Obama’s reelection while 45 percent said they were unhappy. This survey’s numbers are closer to the actual election results, in which Obama won the election with 50.8 percent of the popular vote over Romney at 47.49 percent.
Rishab Ghosh, co-founder and chief scientist at Topsy, argued that “survey results are highly dependent on the wordings of questions asked,” he said. “It is unclear from the Pew study that the sentiment for positive and negative was any way comparable to the opinion poll questions – so it is not surprising that they found different results (e.g. we are not sure how broad Twitter sentiment for certain keywords could be compared to the answers in a poll to a specifically worded question on the Presidential candidates’ debate performance).”
DataSift added that tweets, like surveys, can be broken down by demographic information such as age, gender, and location. “You can’t make a broad statement about ‘what Twitter thinks’ about elections or any other topic without getting into the demographic details of users,” a company spokesperson said. “This is a common error that even the largest companies make.”
Pew has, in fact, conducted surveys that show the demographics of the user bases on different social networks. On Twitter, researchers found that the largest groups within the 16 percent of social media users who use the micro-blogging site are African-Americans, adults between the ages of 18 and 29, and people who live in urban areas.
Another survey about news consumption showed that Twitter users leaned to the left, with 57 percent of people who posted news on Twitter reporting that they were Democrats or leaned Democratic, compared with 46 percent of the general public.
The “Pew research on Twitter’s liberal bias is compelling, but misses the mark,” added DataSift. “…The misconception around Pew’s research points to the need for better insight and analysis to understand not just what is being said, but who is saying it down to the individual user level. This concept is relevant not just for politicians, but vital for brands as well. Those who don’t get it are taking the wrong approach.”
Pew researchers also noted that the volume of responses generated by each event varies — there were 14 million Tweets about Obama’s reelection versus more than five million Tweets about the first presidential debate — as do the segments of the population who choose to participate. “Those who tweeted about the California same-sex marriage ruling were likely not the same group as those who tweeted about Obama’s inaugural or Romney’s selection of Paul Ryan,” the Pew report noted.
Topsy actually worked with Democratic and Republican polling firms Mellman Group and North Star Opinion Research on the Twitter Political Index during the 2012 presidential elections. After analyzing sentiment on 60 billion Tweets in real time, Topsy analysts found a strong correlation between Twitter sentiment and 2 years worth of Gallup Poll data for President Obama. The company also used Twitter data to correctly predict election results for almost all U.S states. “So we are convinced Twitter data is very useful for understanding public opinion; of course, analysts of this data need to be clear about what questions they are asking,” said Ghosh.
He added, “For political opinions, we are quite confident that Twitter provides a broad spectrum of opinions enough to complement opinion polls – with some features, such as the large sample size, the lack of a self-selection response bias, and the ability to get instant opinions, compensating for others, such as the difficulty of identifying a controlled sample, or posing clearly worded questions to get clear responses to them.”