Belle Amie voted out

So in the evening voting show Belle Amie were shown to have received the lowest number of votes and left the competition. Does this tally with our predictions of yesterday?

The bottom 6 according to our analysis on twitter chatter contained both Katie and Belle Amie, and so this aspect of the analysis was correct. However we had Paije as the worst performer along with Belle Amie and Trayc. Xfactor does not release the exact ordering of the contestants, so we don't know who finished 3rd from bottom. But why does our analysis show a different bottom two than were voted for by the public?

We can see the reasons why when we examine the rankings that we posted yesterday in a little more detail.

The chart below shows normalised positive sentiment analysis for the contestants every day based on twitter comments. We haven't looked at negative sentiment since the Xfactor requires you to vote FOR your favourite contestant - therefore the more positive that people are being about a particular contestant, the more likely that contestant will do well in the public vote. The analysis is normalised so that each day can be more easily compared against each other - there is evidently much more chatter on the Saturday and Sundays about the Xfactor when the program actually airs.

We can see in the figure that Aiden tends to have the most positive comments, but we also see Katie doing well. Looking at the positive sentiment, it is clear that Paije for example consistently barely scores at all - this is due to the fact that there are very few comments being made about Paije at all on twitter. The same is true for Trayc and Belle Amie. On a number of days there is no chatter about Trayc at all, and although there is consistent chatter about Belle Amie it is small in comparison to Aiden, and even Katie.

This highlights a couple of obvious conclusions that all digital marketeers should be very aware of! Firstly, the demographic of twitter does not match that of the voting public. It may be possible to identify trends in the populace - for example the high scoring of Wagner throughout the week indicating that there is a large body of support for him. However particularly at this early stage where the vote is split it can be difficult to predict with any accuracy how the general public will vote from the twitter data alone.

This highlights the second conclusion which means that in order to improve our predictive capability, alternative data sources should be identified that more closely match the voting intentions of the general populace. It is unlikely that a single data source will cover this wide demographic and therefore a combination of data sources should be used.