pPolis2024-05-27

Summary for thumbnail: Politicians' opinion vectors visualized in Polis-like fashion and then grouped by party.

• LDP (right) and Reiwa, Social Democratic Party, Communist Party (left)

2024-05-27

• Joint research by Taniguchi Laboratory, University of Tokyo and Asahi Shimbun

I've tried two different ways of rounding a 5-point scale to 3 with no particularly interesting results...

Essentially, the clusters aren't separated, so the boundaries when divided into two clusters are in close proximity.

• Of course, the initial swing will replace the area around the boundary.

I've tried changing the colors by political party, but I can't find the meaning, something is buggy...

Explained variance: [0.29123858 0.18823937] Hmmm, this doesn't represent more than half of the variance in two dimensions, and the projection into a space stretched by a few people with extreme answers makes it hard to tell the difference between the parties.

Oh, is that possible?

• I'm told that the data is divided into clusters and mixed with data that is only for the LDP and the Constituent Assembly.
• It must be a bug.

:

----------------------------------
significant comments for group  0
----------------------------------
 
comment 6 : We should not hesitate to attack enemy bases when we anticipate attacks from other countries.
Count of +1 votes in group:  8.0
Count of 0 votes in group:  51.0
Count of -1 votes in group:  204.0
3.0% / 19.4% / 77.6%
 
Count of +1 votes outside group:  263.0
Count of 0 votes outside group:  47.0
Count of -1 votes outside group:  13.0
81.4% / 14.6% / 4.0
 
pvalue:  5.1690034351040005e-82
 
comment 10 : I want the Prime Minister to visit Yasukuni Shrine.
Count of +1 votes in group:  14.0
Count of 0 votes in group:  48.0
Count of -1 votes in group:  158.0
6.4% / 21.8% / 71.8%
 
Count of +1 votes outside group:  194.0
Count of 0 votes outside group:  81.0
Count of -1 votes outside group:  9.0
68.3% / 28.5% / 3.2
 
pvalue:  2.5450099098536335e-56
 
comment 5 : Japan's defense capability should be strengthened.
Count of +1 votes in group:  78.0
Count of 0 votes in group:  59.0
Count of -1 votes in group:  125.0
29.8% / 22.5% / 47.7%
 
Count of +1 votes outside group:  318.0
Count of 0 votes outside group:  8.0
Count of -1 votes outside group:  0.0
97.5% / 2.5% / 0.0
 
pvalue:  4.270713622876674e-49
 
comment 11 : Relocation of Futenma base in Okinawa Prefecture to Henoko is unavoidable
Count of +1 votes in group:  18.0
Count of 0 votes in group:  44.0
Count of -1 votes in group:  157.0
8.2% / 20.1% / 71.7%
 
Count of +1 votes outside group:  187.0
Count of 0 votes outside group:  80.0
Count of -1 votes outside group:  19.0
65.4% / 28.0% / 6.6
 
pvalue:  8.135497457098475e-47
 
comment 32 : The Japan-U.S. Security Treaty should be strengthened to ensure cooperation by the U.S. in times of crisis / We should be cautious about strengthening the Japan-U.S. Security Treaty to avoid getting involved in wars that do not involve Japan.
Count of +1 votes in group:  79.0
Count of 0 votes in group:  42.0
Count of -1 votes in group:  139.0
30.4% / 16.2% / 53.5%
 
Count of +1 votes outside group:  242.0
Count of 0 votes outside group:  57.0
Count of -1 votes outside group:  27.0
74.2% / 17.5% / 8.3
 
pvalue:  1.6118502287656484e-32
total votes in group:  218.0
----------------------------------
significant comments for group  1
----------------------------------
 
comment 22 : Permanent foreign residents should be allowed local suffrage.
Count of +1 votes in group:  7.0
Count of 0 votes in group:  33.0
Count of -1 votes in group:  247.0
2.4% / 11.5% / 86.1%
 
Count of +1 votes outside group:  156.0
Count of 0 votes outside group:  43.0
Count of -1 votes outside group:  20.0
71.2% / 19.6% / 9.1
 
pvalue:  1.0860551845560161e-65
 
comment 37 : Q5_6
Count of +1 votes in group:  16.0
Count of 0 votes in group:  111.0
Count of -1 votes in group:  199.0
4.9% / 34.0% / 61.0%
 
Count of +1 votes outside group:  177.0
Count of 0 votes outside group:  59.0
Count of -1 votes outside group:  26.0
67.6% / 22.5% / 9.9
 
pvalue:  4.6226851092461706e-38
 
comment 31 : A quota system should be introduced to allocate a certain percentage of parliamentary seats and candidates to women.
Count of +1 votes in group:  58.0
Count of 0 votes in group:  112.0
Count of -1 votes in group:  152.0
18.0% / 34.8% / 47.2%
 
Count of +1 votes outside group:  216.0
Count of 0 votes outside group:  39.0
Count of -1 votes outside group:  7.0
82.4% / 14.9% / 2.7
 
pvalue:  6.57885322570923e-36
 
comment 9 : The three non-nuclear principles should be upheld.
Count of +1 votes in group:  92.0
Count of 0 votes in group:  100.0
Count of -1 votes in group:  133.0
28.3% / 30.8% / 40.9%
 
Count of +1 votes outside group:  249.0
Count of 0 votes outside group:  13.0
Count of -1 votes outside group:  2.0
94.3% / 4.9% / 0.8
 
pvalue:  1.6502232104242013e-34
 
comment 28 : The law should allow a married couple to continue to use their maiden names after marriage if they so desire.
Count of +1 votes in group:  123.0
Count of 0 votes in group:  68.0
Count of -1 votes in group:  133.0
38.0% / 21.0% / 41.0%
 
Count of +1 votes outside group:  246.0
Count of 0 votes outside group:  14.0
Count of -1 votes outside group:  3.0
93.5% / 5.3% / 1.1
 
pvalue:  2.1099888519589346e-33
total votes in group:  256.0

2024-05-28 I tried to cut them down to five each from the Liberal Democratic Party and the Communist Party. Ehhh, so the party still won't split? Really? Not a bug? :

comment 5 : Q4_1 Japan's defense capability should be strengthened.
Count of +1 votes in group:  5.0
Count of 0 votes in group:  0.0
Count of -1 votes in group:  0.0
100.0% / 0.0% / 0.0%
 
Count of +1 votes outside group:  0.0
Count of 0 votes outside group:  0.0
Count of -1 votes outside group:  5.0
0.0% / 0.0% / 100.0

Eh, I knew this was buggy. The mapping of data points to names and political parties seems crazy.

:

labels_=array([0, 1, 1, 1, 1, 0, 0, 0, 1, 0], dtype=int32)

I see the order of the PCA results has changed.

Ahhh, okay.

• To increase stability when the data size is large, the first 100 kMeans are multiplied, where the order is randomly shuffled.
• plot_embedding_with_clusters(centers, clusters_K_star)
• and
• def plot_embedding_with_clusters(embedding_, labels_):
• This embedding_ is not the same as embedding as a result of PCAing the matrix, because it is received in I'm so happy for you!

If you take 30 from the Liberal Democrats and 30 from the Communists, they split up nicely.

All data again, colored by party, giving up on naming names. A little meaningful bias in color, but hard to tell. Polis-like cluster boundaries do not make much sense.

• The proper interpretation is "there is only one cluster" or "the algorithm is 2-5, so I'm just forcing it into two clusters." It would be more interesting to draw boundaries for each party.

Reiwa and the Communist Party are in small clusters, while the NHK Party seems to be scattered. There seem to be two or three separate clusters inside. Japan Restoration has a tight core, but there are people scattered around the periphery.

• LDP (right) and Reiwa, Social Democratic Party, Communist Party (left)

• Democratic Party of Japan (left), New Komeito, People's Democratic Party (lower right), Japan Restoration Association (upper right)

I'm limited to seeing three or four at a time. I didn't enjoy it in simple Polis, but this is much more interesting.

2024-05-28 Q: Do the left and right sides of the PCA coincide with the political ideological left and right? A: I didn't manipulate it to do that, I think it just happened. Well, the X axis is the first principal component axis, so it is the direction with the largest variance, and it is perhaps not strange that it is strongly correlated with the political left and right. I think there's a 1/2 chance the left and right will be reversed.

pPolis pJCPolis

This page is auto-translated from /nishio/pPolis2024-05-27 using DeepL. If you looks something interesting but the auto-translated English is not good enough to understand it, feel free to let me know at @nishio_en. I'm very happy to spread my thought to non-Japanese readers.