AKB48 Sousenkyo Senbatsu Prediction – Why Satou Amina is the real Cinderella Girl

Branching from my love of anime, I have been very into various idol groups over the last few years. I started off with Morning Musume, fell in love with Idoling!!! (still my favourite group) but has recently warmed up tremendously to AKB48. It’s very easy to find subtitled materials online so that has helped my consumption level and my ability to get to know many of the girls. My oshimen (favourite member) is Sato Amina because I am a sucker for an underdog and she personifies that more than anyone in the AKB48 family. It is well-known amongst fans the level to which she got shafted after the first Senbatsu Sousenkyo when she placed 8^th (for a more complete article of what makes Amina special, please read this blog by aakun. I started really paying attention to her because of it). I want to go one step further and try to see whether she there is any other girl who ranked better in the election despite getting no, or minimal, push whatsoever.

With the exception of Election singles (when fans elect who enter the senbatsu) and Janken singles (when senbatsu places are determined by a tournament of rock-paper-scissors), AKB48 management chooses who are representing the group for each singles. This allows them some power to give exposure to those who are popular or those who they want to groom to be the next popular members of AKB48. I am planning to use this as my proxy to measure how much each members get pushed and compare that against their election results, to see who well being pushed translate to increased or continued popularity among fans.

I decide to use the regression analysis, with senbatsu appearances as the X variables (independent) and sousenkyo ranking as the Y variable (dependent). In effect, I want to try to predict the sousenkyo ranking based on the senbatsu appearances. The difference between these predicted numbers and the actual results should give an indication of which members should not be as popular as their senbatsu fate suggests.

Here are the results. The Diff column represents [ACTUAL SOUSENKYO RANKING – PREDICTED RANKING]. Thus, the more negative the number, the more popular the associated girl is than she has the right to be given her senbatsu exposure. Conversely, the more positive the number, the more the push does not work as well as it should to make the girl popular.

Note that:

• Predicted ranking = Average ranking a member should have gotten based on their cumulative senbatsu selections
• Actual ranking = Average ranking over the past 5 sousenkyo results, as voted by fans
• Names highlighted in red are those who have not appeared in any senbatsu and the regression predicts them a ranking of 43.5 (see methodology for more detail)
• Names in white (10-43) are within one standard deviation of the mean, representing around 68% of all possible value. Name in lighter orange and blue are within 2 standard deviations, representing around 95% of the values. Names in brighter colour (1-2 & 51-54) are outside 2 standard deviations (See Wikipedia for more explanation). Thus, those are the ones who are not likely to get the results by chance.

Comments on the results:

• Amina is undoubtedly the Cinderella girl as far as not getting pushed and still getting amazing ranking. Her score is way ahead of Glad to see my analysis proving this quite emphatically. Her score is way below anyone and is actually outside the 3 x standard deviation range, meaning that she is statistically outside the 99.9% of all possible results – a true outlier indeed. I would attribute the discrepency to how awesome she is as an AKB48 member despite all the ignorance by the management. It makes me feel how incredible it is that she manages to cultivate such level of support despite getting zero senbatsu push whatsoever
• I’m also not too surprised to see Umeda, Kuramochi and Masuda ranking 2-3-4 respectively. From my impression of what I have read online, they are the members who have not exactly been pushed so hard yet still maintain a sizable fanbase that propelled them up the ranking once the sousenkyos came.
• On the same topic of getting no push, out of the 4^th generation that are still around, three (Amina, Oya and Nakata) have had no single senbatsu appearance whatsoever. The remaining two (Kuramochi and Fujie) had three each. Talk about a neglected generation…
• On the topic of generational comparison, I summed up the predicted ranking and actual ranking for each generation and this is what I got. Given their senbatsu appearance, 4^th gen prediction is among the lowest, barely on par with 7^th gen. On the other hand, 10^th and 11^th gens are pushed pretty hard, yet without the results we can yet see.
• 
• I read a post recently that said that 7-8-9 gen are forgotten-extinct-manager’s favourite in that order. The predicted ranking does show that as 9^th gen should outrank 7^th by 5.4 position (36.6 vs. 42.0) and 8^th gen is nowhere to be seen
• On the other end of the scale, Kawaei and Anna are the push that has not yet panned out. Suzuran and Lemon also rank quite high in that regards.
• 6th Gen is only one person (Akicha) so it is difficult to read too much into the result.
• When it comes to early generation pushes that have not been as popular as expected, Mii-chan and Myou come to mind, ranking 43^rd and 45^th respectively.
• The majority of the 1st gens, such as Kojima, Acchan and Takamina had received similar ranking to what their push should give them.

—

Last Words

Amina’s time for more exposure has definitely passed. I wonder if she got the push she should have, how popular she might actually be. She will still remain my favourite member regardless. It is interesting to do this analysis to try to confirm the anecdotes that have floated around the fandom. Push works, but there are always those, such as Amina, who refreshingly forges her own popularity through their own means, if only for a short while.

—

Methodologies

1. I collected the senbatsu data and members’ data, courtesy of stage.net website
2. For each member, I figured out how many singles she is active, namely when she first became a member or a kenkyuusei (i.e. 4^th gen is not announced as part of AKB48 until after BINGO! So those earlier singles do not count)
3. Find out the percentage of appearances each member made it in the senbatsu over total possible appearances. I.e. Watanabe Mayu could have appeared in 24 singles out of a maximum of 26 (because she’s 3^rd gen, she was not in AKB48 during the earlier singles). However, she appeared in senbatsu for 21 singles (missing out on Seifuku and Keibetsu, the two early singles). Therefore, her senbatsu score would have been 21/23 or 91%
4. For the X-variables, I eventually used center selection, frontline selection (i.e. media senbatsu) and general senbatsu position.
5. I ran a 3-variable regression in excel.
6. Using the regression, I then worked out the predicted ranking.

The regression cofficients that I ended up with are:

Predict ranking = 43.5 – 38.4 (Senbatsu Count) – 37.8 (Front Position Count) – 44.3 (Center Position Count)

• From this, if you appear in no senbatsu at all, like in the case of Amina, you will be predicted a ranking of 43.5.
• Any appearances in the senbatsu, or better, will lower the predicted ranking accordingly.
• The R-squared that I got from this particular regression is 76.9%, which is pretty good

–

Assumptions

• I am only doing this for the AKB48 members. I excluded all members that never made it into the sousenkyo senbatsu ranking.
• For the senbatsu count, I excluded sousenkyo and janken singles since those are not determined by the management (at least that’s what I will believe)
• For those who went unranked, I assigned the following scores for each sousenkyo to represent such status numerically: 1st = 40, 2nd = 50, 3rd = 50, 4th = 70, 5th = 70

---