-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathresults.rtf
132 lines (108 loc) · 8.04 KB
/
results.rtf
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
{\rtf1\ansi\ansicpg1252\cocoartf1404\cocoasubrtf470
{\fonttbl\f0\fswiss\fcharset0 Helvetica;\f1\fnil\fcharset0 Cousine;}
{\colortbl;\red255\green255\blue255;\red189\green189\blue189;\red180\green36\blue25;\red0\green0\blue233;
\red152\green152\blue152;}
\margl1440\margr1440\vieww12240\viewh14500\viewkind0
\pard\tx560\tx1120\tx1680\tx2240\tx2800\tx3360\tx3920\tx4480\tx5040\tx5600\tx6160\tx6720\pardirnatural\partightenfactor0
\f0\fs24 \cf0 \CocoaLigature0 \
2020-03-04 by Brendan O'Connor\
Biden-Sanders margins, averaged over all Super Tuesday states, conditional on voter attributes, from WaPo exit polls. Attributes are shown in order of how polarized-by/sorted-by they are -- how much variability of the margin is explained by the attribute (with a std dev metric, last column; max=100).
\b Age the most polarizing attribute
\b0 ! More so than political ideology, race, party, or especially education status.
\f1 \
\
attrib value mar biden sanders totshare sd\
\cf2 1\cf0 AGE 18-29 -\cf3 43.5\cf0 12.4 56.0 0.142 25.9 \
\cf2 2\cf0 AGE 30-44 -\cf3 27.0\cf0 17.5 44.5 0.228 25.9 \
\cf2 3\cf0 AGE 45-64 12.8 34.8 21.9 0.353 25.9 \
\cf2 4\cf0 AGE 65 or over 25.4 40.5 15.0 0.279 25.9 \
\fs4 \
\fs24 \cf2 5\cf0 SINGLEPAYER20 Support -\cf3 16.4\cf0 24.6 41.0 0.564 24.2 \
\cf2 6\cf0 SINGLEPAYER20 Oppose 34.4 46.2 11.8 0.381 24.2 \
\fs4 \cf2 \
\fs24 7\cf0 FAVSOCIALISM Favorable -\cf3 20.8\cf0 21.6 42.4 0.527 21.2 \
\cf2 8\cf0 FAVSOCIALISM Unfavorable 24.9 40.4 15.5 0.367 21.2 \
\fs4 \cf2 \
\fs24 9\cf0 PHIL4 Very liberal -\cf3 24.3\cf0 21.3 45.6 0.250 18.5 \
\cf2 10\cf0 PHIL4 Somewhat liberal 4.63 33.1 28.5 0.354 18.5 \
\cf2 11\cf0 PHIL4 Moderate or cons. 23.0 42.5 19.5 0.396 18.5 \
\fs4 \cf2 \
\fs24 12\cf0 WINORISS20 Agr. w. you on issues -\cf3 15.7\cf0 24.7 40.3 0.349 15.0 \
\cf2 13\cf0 WINORISS20 Can beat Trump 16.1 39.0 22.9 0.618 15.0 \
\fs4 \cf2 \
\fs24 14\cf0 TIMEPRIFEWDAYS Earlier than that -\cf3 4.42\cf0 29.8 34.3 0.649 14.9 \
\cf2 15\cf0 TIMEPRIFEWDAYS Last few days 27.2 45.2 18.0 0.338 14.9 \
\fs4 \cf2 \
\fs24 16\cf0 RACE Asian -\cf3 24.9\cf0 13.6 38.5 0.049\ul 4\ulnone 14.5 \
\cf2 17\cf0 RACE Hispanic/Latino -\cf3 11.7\cf0 28.4 40.1 0.194 14.5 \
\cf2 18\cf0 RACE White 4.13 31.6 27.5 0.623 14.5 \
\cf2 19\cf0 RACE Black 30.6 50.0 19.4 0.172 14.5 \
\fs4 \cf2 \
\fs24 20\cf0 PARTY Indep. or sth. else -\cf3 12.0\cf0 25.0 37.1 0.275 11.0 \
\cf2 21\cf0 PARTY Democrat 10.8 37.3 26.5 0.690 11.0 \
\cf2 22\cf0 PARTY Republican 21 43 22 0.07 11.0 \
\fs4 \cf2 \
\fs24 23\cf0 ISSDEM20B Income inequality -\cf3 9.26\cf0 26.3 35.6 0.208 8.20 \
\cf2 24\cf0 ISSDEM20B Climate change 5.95 33.4 27.4 0.216 8.20 \
\cf2 25\cf0 ISSDEM20B Health care 6.27 35.6 29.3 0.411 8.20 \
\cf2 26\cf0 ISSDEM20B Race relations 20.1 43.4 23.3 0.106 8.20 \
\fs4 \cf2 \
\fs24 27\cf0 SEX Male -\cf3 0.780\cf0 32.0 32.8 0.438 5.22 \
\cf2 28\cf0 SEX Female 9.74 35.5 25.8 0.562 5.22 \
\fs4 \cf2 \
\fs24 29\cf0 EDUCCOLL College graduate 4.74 31.0 26.3 0.489 0.119\
\cf2 30\cf0 EDUCCOLL No college degree 4.98 36.9 31.9 0.511 0.119\
\
\
\f0 Exit poll data for margins conditional on subgroups by state are from here:\
\pard\pardeftab720\sl280\partightenfactor0
{\field{\*\fldinst{HYPERLINK "https://www.washingtonpost.com/graphics/politics/exit-polls-2020-super-tuesday-primary/"}}{\fldrslt \cf4 \expnd0\expndtw0\kerning0
\ul \ulc4 \CocoaLigature1 https://www.washingtonpost.com/graphics/politics/exit-polls-2020-super-tuesday-primary/}}\cf4 \expnd0\expndtw0\kerning0
\ul \ulc4 \CocoaLigature1 \
\pard\tx560\tx1120\tx1680\tx2240\tx2800\tx3360\tx3920\tx4480\tx5040\tx5600\tx6160\tx6720\pardirnatural\partightenfactor0
\cf0 \kerning1\expnd0\expndtw0 \ulnone \CocoaLigature0 \
These were calculated via weighting by state size -- specifically, total number of votes (so far, as of today at ~1pm according to NYT website). You get slightly different results, looking worse for Sanders, if we use flat averaging (presumably since Calif counts for less).\
'totshare' is the fraction of voters with this attribute-value pair; they should sum to 1 across values of a single attribute.\
'sd' is the standard deviation of the margin across types of voters; it's weighted by totshare\
for each attr:\
avg margin = weighted mean of margin, weighted by num voters\
variance = weighted mean of sq. diff of per-attrvalue margin vs. avg margin, weighted by num voters\
sd = sqrt of variance\
\pard\tx560\tx1120\tx1680\tx2240\tx2800\tx3360\tx3920\tx4480\tx5040\tx5600\tx6160\tx6720\pardirnatural\partightenfactor0
\cf0 Thus this "stdev polarization metric" has a minimum value of 0, and a maximum value of 100 at perfect polarization (if there were two equally-sized populations, one of them 100-0 biden/sanders and the other 0-100 biden-sanders thus margins +100 and -100. Mean=0, avg squared diff=10,000, sqrt of that = 100).\
\pard\tx560\tx1120\tx1680\tx2240\tx2800\tx3360\tx3920\tx4480\tx5040\tx5600\tx6160\tx6720\pardirnatural\partightenfactor0
\cf0 \
This stdev metric is one possible way to measure how big the vote preference gaps are based on an attribute. You could also use shannon entropy-based MI or ANOVA-style variance decomposition (i think between-group varaince is the square of this metric?), etc. Earlier on I tried using stats on all 4 candidates as a multiclass problem with quadratic or shannon entropies, but I found it harder to interpret. The Biden-Sanders margin is quite interpretable.\
\
Note something is weird here -- if you average the margins per attribute, you get different overall margins. Ideally these would all be the same. It's possible I have a bug in my analysis, or I'm misinterpreting the WP exit poll data or even its formatting, or maybe not all states get the same exit poll questions, or there are accuracy issues? I'm not sure.\
\pard\tx560\tx1120\tx1680\tx2240\tx2800\tx3360\tx3920\tx4480\tx5040\tx5600\tx6160\tx6720\pardirnatural\partightenfactor0
\f1 \cf0 > e %>% group_by(attrib) %>% summarise(bs=allstatebs %*% totshare)\
\cf5 # A tibble: 12 x 2\cf0 \
attrib bs\
\cf5 <chr>\cf0 \cf5 <dbl>\cf0 \
\cf2 1\cf0 AGE -\cf3 0.713\cf0 \
\cf2 2\cf0 EDUCCOLL 4.86 \
\cf2 3\cf0 FAVSOCIALISM -\cf3 1.82\cf0 \
\cf2 4\cf0 ISSDEM20B 4.07 \
\cf2 5\cf0 PARTY 5.59 \
\cf2 6\cf0 PHIL4 4.67 \
\cf2 7\cf0 RACE 4.32 \
\cf2 8\cf0 SEX 5.14 \
\cf2 9\cf0 SINGLEPAYER20 3.88 \
\cf2 10\cf0 TIMEPRIFEWDAYS 6.31 \
\cf2 11\cf0 WINORISS20 4.50 \
\pard\tx560\tx1120\tx1680\tx2240\tx2800\tx3360\tx3920\tx4480\tx5040\tx5600\tx6160\tx6720\pardirnatural\partightenfactor0
\f0 \cf0 \
\
Vote totals from the elections are from NYT results pages:\
\pard\pardeftab720\sl280\partightenfactor0
{\field{\*\fldinst{HYPERLINK "https://www.nytimes.com/interactive/2020/03/03/us/elections/results-super-tuesday-primary-election.html"}}{\fldrslt \cf4 \expnd0\expndtw0\kerning0
\ul \CocoaLigature1 https://www.nytimes.com/interactive/2020/03/03/us/elections/results-super-tuesday-primary-election.html}}\
\pard\tx560\tx1120\tx1680\tx2240\tx2800\tx3360\tx3920\tx4480\tx5040\tx5600\tx6160\tx6720\pardirnatural\partightenfactor0
\cf0 \
CSV from saving React data from WP site into data1.json then\
% python parse2.py > d2.csv\
\
Analysis above from 'analysis3.r'\
\
}