Saturday, November 26, 2016

Concentration of Hate Groups Predict Hate Crimes (if you consider DC) and Trump Vote (if you don't)





Earlier this year the Southern Poverty Law Center created a map showing the concentration of hate groups in the US.  I adjusted the number of groups in each state by the population in each state  showing a different picture.  Just after the election the FBI issued a new report that hate crimes in the US have increased in 2015.  I thought I would take a look at any possible association between hate crimes, hate groups, and Trump's percentage of the vote.


The graph above shows the linear association between hate groups and hate crimes (both adjusted for population in groups or crimes per million) .  There is a statistically significant relationship with 25.8% of the variability accounted for.  The regression line suggests that for every increase of one hate group per million there is an increase of hate crimes by 2.41 incidents per million.  The District of Columbia has an extremely high rate of both hate groups (26.78 groups/million) and hate crimes (96.69 incidents per million).  Massachusetts has the second highest rate of hate crimes (60.49 incidents/million) but a low rate of hate groups.  

I looked at the relationship first with DC removed (the relationship became negative and non significant) then with DC and MA removed (little changed).  The relationship without these two states is shown below.  




I also looked at the association between hate crimes, hate groups, and Trump’s % of the vote in each state.  The association between hate crimes and Trump’s % of the vote is confounded by DC and MA having high rates of reported hate crimes and low % of the vote for Trump.  Trump had 4% of the vote in DC.  With DC removed there is a positive relationship between hate group concentration and Trump’s % of the vote. There is a negative relationship between the concentration of hate groups and Trump's % of the vote if DC is included as can be seen in the graph below.


If DC is excluded the relationship between hate group concentration and Trump's % of the vote becomes positive with an increase in the concentration of hate groups/million by one leading to a predicted increase of 3.08% in Trump's % of the vote as can be seen in the graph below.  


Clearly DC is a powerful outlier compared to the other 50 states in income, life expectancy, hate groups, and now hate crimes.  Others dismissed DC having a high rate of hate groups by stating that they just had their national headquarters there.  The high rate of reported hate crimes supports that these groups are active in DC with a large minority population suggests that these groups are indeed active there. The alt-right conference below discussed below also supports the activity there.  
 
There are states with high rates of hate groups but low rates of reported hate crimes such as Arkansas and Mississippi (with zero reported hate crimes).  Are the groups more cautious  in these states or are hate crimes less likely to be reported there?  I don't know the answer to these questions but they merit further study.
**Related Posts**

More Hate Groups in States Where Trump and Clinton Win (and in DC Where He Lost)


SPLC Hate Group Update: Washington, DC has an Increase in Activity


 

A Wave of Hate Groups in California? No in Washington, DC

Friday, November 18, 2016

On Facebook, Fake News, and the Election

 
There has been a flurry of stories regarding fake news on Facebook and what role it may have had on the election outcome.  Some of these stories are summarized on my Facebook page.  The above clip features Craig Silverman discussing a study he did looking at the engagement (likes, comments, and shares with their friends) that the fake news sites were receiving relative to established news sources such as the Washington Post and the New York TimesIt showed elevated engagement right before the election.
 

Facebook and Google have said that they will crackdown on these sites to control the amount of misinformation on the web.  Facebook is now one of my largest traffic sources and Google is the owner of blogspot where this site is based.  I am concerned about how these sites will 'crack down'.  

Fact checking is time consuming.  The more obscure the fact the longer it takes to check.  This is not a fake news site though it may not be considered mainstream.  Will an algorithm be used to filter these sites.  I am concerned about how a computer algorithm will eliminate these sites.  Could someone who objects to my site report it and have it blackballed?  The internet is still like the Wild West and it takes work to filter out what is true and what is not. Who can process all of the information.

**Related Posts** 

Change in Facebook Followings for PA Candidates Before and After the Debate

Facebook Primary 2016, August Update, Does it Predict Support?


Facebook and Twitter Primary: The Final Five

Fact Checking the Factcheckers

The Need for CSI Without Dead Bodies (& Similar Websites)

 

 

 

Saturday, November 12, 2016

Facebook Followings Predict The Outcomes of Elections in PA



Candidate
Election Result 
in PA (%)
FB followers 
Nov 8
FB engaged 
Nov 8
% engaged 
Nov 8
Net gain 
Nov 8
% Gain Nov 8
Clinton (D)
47.61
16,771
52,344
312.1
      5,847
53.5
Trump (R)
48.83
44,222
6,394
14.5
    31,160
238.6
Johnson (L)
2.39
764
47
6.2
              33
4.5
Stein (G)
0.82
542
55
10.1
              49
9.9
gap between D and R

1.22
27,451
-45,950
-297.65
25,313
185.03
correlation with Result

0.85
0.66
0.58
0.73
0.74

The election is now over. Now the process of what the result all means now begins.  I have been posting on the Facebook followings of candidates for nine races in Pennsylvania.  I collected data from the candidates pages on the followings (those who click like on the page), the number and % engaged (those who click on, like or share posts from the page), and the number and % gain in followers since the day after the second presidential debate.  Above is a summary of the Presidential election result for Pennsylvania and the data from the state campaign Facebook pages for the four major presidential candidates collected on the morning of November 8.  The gaps between Clinton and Trump for each measure are presented in the first row in bold.  The bottom row in bold shows the correlation of each Facebook (FB) measure with the election result.  The variable with the strongest correlation with the result was the total number of followers at 0.85 (the correlation coefficient is on a scale of -1 to +1 with the number being further away from zero suggesting a stronger relationship).  The tables below summarize the US Senate Race and several other down ballot races for the US House, State Senate, and State House.  

US Senate
Election Result
in PA (%)
FB followers
Nov 8
FB engaged
Nov 8
% engaged
Nov 8
Net gain
Nov 8
% Gain
Nov 8
Pat Toomey (R)
48.94
156,446
16,276
10.4
        3,849
2.5
Katie McGinty (D)
47.21
25,291
13,155
52.0
        3,750
17.4
gap between D & R
1.73
131,155
3,121
-41.61
99
-14.89
 

The senate race between Pat Toomey and Katie McGinty was narrowly won by Toomey by 1.73%.  He had a big advantage in FB followers (by 131,155), net gain in followers and number engaged.  McGinty did have an advantage in % engaged and % gain in FB followers.



US House 12th
Election Result
in PA (%)
FB followers
Nov 8
FB engaged
Nov 8
% engaged
Nov 8
Net gain
Nov 8
% Gain
Nov 8
Keith Rothfus (R)
62.1
12,748
143
1.1
   22
0.2
Erin
McClelland (D)
37.9
3,720
1,799
48.4
217
6.2
gap between D and R
              24.2
        9,028
     -1,656
           -47
 -195
     -6

The US House race in the 12th district in PA was won by incumbent Keith Rothfus (R) by 24.2%.  He had an advantage in FB Followers but his opponent Erin McClelland had the advantage in the engagement and the gain in followers.  

US House 9th
Election Result
in PA (%)
FB followers
Nov 8
FB engaged
Nov 8
% engaged
Nov 8
Net gain
Nov 8
% Gain
Nov 8
Bill Shuster (R)
63.42
6,589
3,488
52.9
      401
6.5
Art Halvorson (D)
36.58
2,909
2,573
88.4
      705
32.0
gap between D & R
              26.84
3,680
           915
           -36
     -304
       -26
 

The US House 9th District race was won by Bill Shuster by 26.84%.  Shuster had the advantage in followers and in the number engaged.  his opponent Halvorson had the advantage in % and net gain as well as the % engaged.  As an aside my cousin Casey Contres was managing Shuster's campaign.




State Senate 35th
Election Result
in PA (%)
FB followers
Nov 8
FB engaged
Nov 8
% engaged
Nov 8
Net gain
Nov 8
% Gain
Nov 8
Wayne
Langerholc (R)
62.54
1540
320
20.8
        64
4.3
Ed Cernic (D)
37.46
498
347
69.7
       146
41.5
gap between D & R
        25.08
        1,042
           -27
         -48.9
        -82
        -37.1

The State Senate race in the 35th district, like the Presidential, race had no incumbent.  Wayne Langerholc won with 62.54% of the vote.  He had a 3 to one advantage in Facebook followers.  Cernic (who, in the interest of full disclosure, I had worked for) had an advantage in the number and % engaged and in the net and % gain in followers.

State Senate
41st
Election Result
in PA (%)
FB followers
Nov 8
FB engaged
Nov 8
% engaged
Nov 8
Net gain
Nov 8
% Gain
Nov 8
Don White (R)
68.32
966
4
0.4
                9
0.9
Tony DeLoreto (D)
27.42
874
770
88.1
           328
60.1
gap between D & R
40.9
92
-766
-87.7
-319
-59.1
 

The state Senate race in the 41st district was won by incumbent Don White with 68.32% of the vote.  He had a slight advantage in Facebook followers on his State Senate page (I couldn't find a campaign FB page for him).  Challenger DeLoreto had an advantage in the remaining categories.



State House 71st
Election Result
in PA (%)
FB followers
Nov 8
FB engaged
Nov 8
% engaged
Nov 8
Net gain
Nov 8
% Gain
Nov 8
Mark Amsdell (R)
41.23
268
50
18.7
         117
77.5
Bryan Barbin (D)
58.77
404
23
5.7
              -  
0.0
gap between D & R
-17.54
-136
27
13.0
117
77.5


The state house 71st district (my district) was won by incumbent Bryan Barbin with 58.77% of the vote over Mark Amsdell.  Barbin had the advantage in followers on his district page. He had no campaign page.  Amsdell had the advantage in the remaining categories.  

State House 72nd
Election Result
in PA (%)
FB followers
Nov 8
FB engaged
Nov 8
% engaged
Nov 8
Net gain
Nov 8
% Gain
Nov 8
Cecelia Houser (R)
42.29
684
566
82.7
         236
52.7
Frank Burns (D)
57.71
1266
150
11.8
           15
1.2
gap between D & R
     -15.42
         -582
           416
          70.9
        221
      51.5
 

In the 72nd State House district, incumbent Frank Burns won with 57.71% of the vote over Cecelia Houser.  Burns had the advantage in FB followers on his district page while Houser had the advantage on the other categories.  Like Barbin and DeLoreto, Burns just had a FB page for his district, not a campaign page.

State House 73rd
Election Result
in PA (%)
FB followers
Nov 8
FB engaged
Nov 8
% engaged
Nov 8
Net gain
Nov 8
% Gain
Nov 8
Tommy Sankey (R)
71.2
639
262
41.0
             52
8.9
Fred Weaver (D)
28.8
812
867
106.8
           185
29.5
gap between D & R
          42.4
         -173
         -605
        -65.8
        -133
   -20.6
 

The final race I followed was the one in the 73rd State House district.  Incumbent Tommy Sankey defeated challenger Fred Weaver by 42.4% of the vote.  Weaver had the advantage over Sankey's FB campaign page.   Sankey had more followers on his district page with more than 1,730 total followers on election day.

Of the nine races I looked at in Pennsylvania, eight of them were won by the candidate with more followers on their Facebook page for the candidate with more followers on their campaign page.  This gives a percentage of 89% with a margin of error of +/- 6.12%.  If the incumbent had a district page and a campaign page, I used the campaign page in the analysis.  In each case the incumbent may not have needed a campaign page as the incumbent won in each case.   

The advantage in Facebook followers in confounded with incumbency.  The two races with no incumbent (President and State Senate 35th district) followed were also won by the candidate with the larger FB Following.  Below I looked at the gap variables between the D and R candidates to see how they were correlated with the gap in the % of the vote.  The strongest relationship I found was between the % gain in FB followers and the gap in the % of the vote.  This was a negative correlation that was borderline statistically significant
 (p=0.051) accounting for 44% of the variability.  This suggests that as the gap in % gain in followers increases, the gap in the popular vote decreases.  This is summarized in the  graph below.  Trump had a very large increase in his FB following in PA which may have been a factor in his winning PA.



The size of a candidates FB following can be an indicator of the strength of his or her campaign.  Being races from PA for President and Senate and local races for West Central PA, this is admittedly a biased sample but it is relatively easy to select your own sample to see how the selected candidates' FB page followings correspond to their support on election day.

**Update**

Facebook founder Mark Zuckerberg posted on his FB page that the false stories on the site did not influence the outcome of the election but they will use truth filters in the future (the Full post can be read here).  He claimed that 99% of material is authentic.  This post did not look at the veracity of the posts by the different politicians pages.  It only looked at the engagement with the posts.  Since many of the top pages pay Facebook to improve their circulation would it pay for Zuckerberg to rigorously block content from these pages?

  **Related Posts***

Change in Facebook Followings for PA Candidates Before and After the Debate

The Facebook and Twitter General Election

What Effect Do 3rd Parties Have on Trump and Clinton?

Making Sense of the Pat Toomey-Joe Sestak Senate Race