Our Buzz Survey show who Twitter is talking about, not necessarily who Twitter supports. However, the conversation gravitates to the topic or person perceived as most important and drifts away from the least important. Subsequently, the Buzz Tracker can predict a rise or fall in polls.
Methods are discussed below. Trend lines have been included.
Methods: Tweets mentioning the candidates’ @screen_names are collected using the Twitter API and designated either a mention or a retweet. Only tweets containing the @screen_name are considered and only actual mentions are used – retweets are excluded. Next we correct for spammers and overly-enthusiastic supporters who might otherwise affect the outcome. Then, we calculate the percentage of total mentions each candidate has received for the day and the results are put in a graph spanning 14 days.
Why mentions only? Mentions are an indication of importance – a portion of the conversation is inevitably drawn to each candidate in order of their perceived importance.
State polls by nature include less data than the national polls. Less data leads to significantly wider day-to-day variance so we display linear trend lines for ease of interpretation.