After last week's try at predicting the bottom 2 got Amelia correct in 5th, but missed Misha's rise to safety, it was time for my model to deliver this week. The twitter stats came through on the Saturday night - earlier than before, but I wasn't too worried about the inconsistency as I'm told Saturday night generates most of the traffic. No reason to make any changes to the model this week, so I plugged the figures in and risked making an idiot of myself with the following figures:

Contestant	Vote Share
Little Mix	28.3%
Marcus	25.5%
Amelia	18.8%
Janet	16.6%
Misha	10.9%

And this week....it was spot on. The live results show announced a bottom 2 of Misha/Janet, with Janet being eliminated by a majority verdict (which strongly supports her not being bottom.) My model involves giving each contestant a weighting to reflect the link between their online support and voting base; Janet's is much larger than any other contestants, as she was continually safe (and allegedly topping the voting) despite having no greater twitter volumes than others.

Her drop in support was obvious this week, and for her to end up in the relatively narrow margin between Amelia and Misha is a real indication that it's an accurate model (or just plain lucky, take your pick...)

Here is the updated results graph:

And by popular* request here are the same figures compared to the average vote each week. This shows a bit more than the previous graph, as an act scoring the average vote each week will be a flat line at 100%, and an act heading upwards really is gaining support. On the previous graph, Janet is shown as gaining support from week 4 to week 5, but that's because there were 2 fewer acts in week 5; this one shows that she was actually being dragged closer to the average vote in week 5. I've taken Sami and Sophie off as it was too cluttered otherwise, and I've completely ignored Nu Vibe so all the percentages in week 1 will be a little lower in reality.

Anything to learn this week? A couple of thoughts:

• The author at my data source questions whether data is still useful this late in the contest. I'd suggest that it is - with the accurate prediction this week, and the acts and their fanbases are better established
• It looks like a 2 horse race at the top, however for Little Mix to top the vote after performing first is an great achievement if true - they should probably be slight favourites for the whole contest
• I'm aware I have a rival at the X Factor Tracker. Their site certainly looks better than mine, and we've come up with some similar predictions, although theirs seems to be point in time tracking so has shown different bottom 2 options within certain weekends. I don't know how their methodology works, in particular how they treat negative tweets for a contestant, and how (or if) they try to map the relationship between twitter response and voting patterns, which is the bridge my model is trying to cross. If anyone from that site is reading this one, do drop me a line on tpfkar at hotmail dot com - very interested to hear further details of your work.

(*) As with everything else on this site, popular is defined as a proportion rather than a number ;)