Listen, if Facebook was able to determine relationships blossoming and waning using data, then you should see that data can tell you just about anything. You just have to know what and how you want to look at the data.
Let’s take Mr. Stephen Colbert’s current situation as our example to get things started. If you haven’t heard, he is enjoying the glory of Twitter and the wrath of the 140-character activist. In short: #CancelColbert became a trending hashtag. You can see how “the Twit Hit the Fan” in the video below.
So, after reminding the new guy to NEVER delete a tweet EVER again (it’s a Cardinal Twin), The Colbert Report staff would have wanted to see what the conversation is looking like out on Twitter. First step is determining what they would want to see.
We know about the hashtag, but we also know not everyone will use it. So first, we create a basic boolean query to enter in to the data collection tool: colbert AND (cancel OR quit) – and then we run it for a week-long timeframe to make sure we have a good base query to start with.
As you can see, this gives us a pretty decent result, and we could dive in a bit deeper to add some additional keywords specific to the content in question like “redskins” or “asian,” but for the sake of this example, we will keep it simple. (We will revisit tweaking queries later.)
Next, they would need to determine what data is important to them for this report. Not all reports are created equal.
This is where I preach about being brave when it comes to reporting. There is a standard report style that you start with, but based on what you are looking at AND what you want to know, that report will evolve, and you should encourage that evolution. Another thing to consider is who will be viewing the report. How do they like to see data? Will they understand what they are looking at? How do we make sure they are getting what they need from the report. Again, customization is key to a useful and actionable analytics report.
So, back to what Colbert and team might want to see. The popularity graph above is a good place to start. How much traction is this conversation getting? We can see the the majority of the volume was on the 28th, with a significant drop off the next day, but there has still been steady volume happening around this conversation. We may then want to divide up the data into two timeframes: day of and residual days. But again, for the sake of this example, we will keep it as one timeframe.
The next area of interest would likely be Reach.
As you can see, just about half of the tweets came from those who only tweeted once, the other half of the volume was from RTs. You can also see engagement was limited to about one tweet per tweeter.
Other useful pieces to look at would be text analytics and sentiment.
Here you can see some other hashtags that popped up during the conversation, which may add some insight for response (or for them, material for the show).
In the word cloud, you see some obvious words (bigger and bolder = more volume) and then some interesting ones like racist and outrage, which would be words you might want to dive into deeper.
Next is sentiment. Here, we see that a large amount of negative sentiment is coming from people siding with Colbert and not the #CancelColbert movement. We also see a high level of neutral sentiment. More on this in the wrap up.
Lastly, we would include some demographics info, because we know that the Colbert staff would find this info potentially useful.
For the Colbert team, this would be a great start to a comprehensive report on the conversation. Next, we would want to give a recap analysis of the data so next steps can be decided.
The (Quick-and-Dirty) Analysis
The majority of the conversation happened on the 28th, and then dropped significantly. Volume has remained steady at ~500 mentions/ day. Approximately half of the volume were single tweets and the other half were RTs. A surprisingly small amount of tweets actually used the hashtag #CancelColbert that trended during the heavy volume time period. Sentiment analysis confirms this in that most tweets were in support of Colbert. Negative sentiment included the cancel conversation, but also supporters of the show showing distaste for the people who were getting offended. There was a heavy neutral sentiment set which we would lean towards positive, but designate as primarily info sharing, which is just people sharing the story vs adding commentary. We can see some other topics that we may want to dive into with an updated query like “keep colbert,” “don’t cancel,” racist and outrage to see what the specific conversation and conversation volume is around these key words/phrases. The demographic info available shows a heavy concentration in “liberal” states.
This report helps the team start off with some data guide next steps. Do they respond? Do they work this data into actionable items? If so, how? Do we want to dive in deeper? Is there something else we want to see? These are all valid questions that one should be asking when viewing an analytics report. Colbert and team may look at this and say, “We basically already knew this, but the data can now back up what we were thinking,” and that is perfectly OK. That might be all you are looking for, confirmation or an argument against current assumptions. Back and forth is important in getting to the right data. True data analysis takes time and careful consideration. Be sure to clarify what you need to know if speed is a factor. Do you only need a quick volume report? Can we do the deep dive as a second step?
If you need help analyzing your social conversation data, Raidious can help.