Before we get going there's some statistical housekeeping to do. I gathered votes for the 27 regular-season episodes (not counting celebrity specials), but as I didn't have the poll idea until episode 3 the first two weeks received very few responses and so are excluded from everything that follows. In addition, all polls were on a per-episode basis, so keep this in mind when interpreting the results: a question that received a lot of votes may have done so because it was very good, or because it was the best of a bad bunch. If I run these polls next season there will definitely be a 'none of the above' option! Finally, all data correct at time of writing, but the polls are still running for anyone lagging behind.
My first point of interest was connections versus sequences. I usually prefer the latter, but what about you lot? There are a few ways to look at this, but the simplest is the total votes received by sequence and connection questions across the entire series. On this front sequences do indeed triumph, picking up 1,246 votes to 1,107 for connections for a 53%-47% 'victory'. On an episode-by-episode basis, meanwhile, the sequence round questions scored more votes than the connections 16 times out of 25 (remember we've excluded the first two episodes), while a sequence question was twice as likely to be voted Question of the Week. This doesn't represent an overwhelming preference, however, with formal statistical tests being disinclined to identify any real preference for sequences over connections.
There's also a potential source of bias working against the connections round: music. Every episode features a music connection question, while (by my count, at least) there were only six of the new-fangled music sequences. If viewers dislike the music rounds as much as the contestants seem to (read more on that here), this could be why connections seem to suffer slightly overall in popularity. The plot on the right summarizes the vote share by question type (if you're unfamiliar with boxplots, these give a rough idea of the spread of votes, with the thick black lines giving the average share of the votes a question received, and a taller box indicating a bigger range of votes). The evidence is pretty damning for the music connections, which averaged just 3% of votes per episode while music sequences, perhaps benefiting from the novelty factor, were the single most popular question type. I should be a good statistician though and acknolwedge the relatively small sample size: we can't say with any particular confidence that music sequences were significantly better than the rest, but we can say that music connections are not a hit with viewers at home (or at least the viewers at home who can be bothered to vote on my blog!). Oh, and in case you were wondering about picture qeustions, they weren't significantly different from 'normal' ones, so I haven't bothered to separate them out in this plot.Another important aspect to these data is that looking at average vote share isn't necessarily what we're interested in; after all, the average question will necessarily receive one-twelfth of the vote (which is why in the previous plot the connections and sequences are centred around the 8% mark). What's potentially more informative is how the best questions in each episode are distributed. Allowing tied questions to share the honours (it doesn't make much difference), there were 30 'questions of the week' across the series. If there were no substantial differences between music, picture and regular questions we'd expect to see 1 in 6 of these be pictures, about 1 in 10 be music (including music sequences) and the rest be regular style questions for a 5-3-22 split. In fact, just 2 picture questions earned the accolade, while 3 music questions and 25 'others' round out the set. This doesn't seem to represent a 'real' difference between the question types however (at least from a statistical perspective), which is unsurprising given the relatively small sample size.
Let's get the Hall of 'Shame' out of the way. Just four questions had picked up no votes at time of writing. Presented along with the total votes on their respective polls, they are:
Episode 9, Question 2: A music connection where song titles contained types of weather ending in 'y' (Rainy Night in Georgia, etc.). (Total votes: 113)
Episode 14, Question 8: A picture sequence of the four most common dog breeds in the UK. (Total votes: 84)
Episode 23, Question 6: Allegedy Only Connect's 'most expensive question' saw both teams stumped by Robert Doisneau's photographs. (Total votes: 83)
Episode 24, Question 8: A 'straight up' sequence testing your knowledge of Afghanistan provinces. (Total votes: 66)
Not too much to say there, so let's get straight onto the highlights. The top three questions of the series are those that received the highest shares of their respective polls. Obviously this could just mean they were lacking competition, but it's good enough for me. Here, then, are the Questions of the Series:
3rd place (24.8%): Web country codes and their elements (Episode 13)
Clues of Australian gold, Antiguan and Barbudan silver, Portugese platinum and Cuban copper were a lovely construction, rewarding patience or a very brave early buzz. There's lots to like here, but what's really remarkable is that this made the top three despite not even being that episode's question of the week. (Spoilers!)
2nd place (27.7%): Letters in the element name (Episode 23)
Another chemical element question makes the top three, as O6, B5 and Fe4 had the hallmarks of luring you into an early buzz if you thought it was atomic number (it isn't, although C6 as an opening clue would have been evil). What's more, while realizing 'tin' was the natural end to the sequence, Sn3 isn't the easiest symbol to remember for the answer. Simultaneously tough and a real 'kick yourself' classic.
1st place (29.9%): First four answers in this programme (Episode 13)
"Forget question of the week, this felt like question of the series." - this was my first reaction when the sequence 'Broken hearts, Sibling feuds, £2 coins' had us all struggling to remember something we'd seem mere minutes previously ('Bens', in case you're wondering). I stand by that opinion now, and while I'm sure there will be claims of a ballot biased by my words at the time, I'm nevertheless delighted to see this reflected in the data. My understanding is this idea had been in the pipeline for years, but logistics had prevented its use until now - and it's perhaps just as well this was saved for BBC Two's much larger audience.
So that's your lot - not the most scientifically rigorous process but it would nevertheless be hard to begrudge any of those top three. A cracking series with many more fantastic questions than I have time to list here. Here's to the next one!
