I would think that WikiLeaks/Twitter followers would be young, but not necessarily likely voters. On the other hand with older voters having land lines, and younger voters mobiles, pollsters have a hard job finding the correct statistical mix of likely voters.
OK, Given that, maybe this is supported by the USC Dornfife Poll internals. Look at the 18-34, African Descent (A/A) and Hispanic.
Comparison of Trump and Clinton demographic margins from 1 week ago vs. today:
Demographic———Aug 14————Aug 21-——Net change
18-34——————HC+13.8-———DT+0.8-——DT+14.6
35-64——————HC+1.8————DT+1.0-——DT+2.8
65+———————DT+1.0————DT+5.3-——DT+4.3
High school———DT+2.6————DT+13.9——DT+11.3
Some college-——DT+4.5————DT+8.0-——DT+3.5
College degree-—HC+19.6-———HT+15.0——DT+4.6
White——————DT+15.9-———DT+20.1——DT+4.2
A/A———————HC+82.3-———HC+70.1——DT+12.2
Hispanic-————HC+39.1-———HC+15.5——DT+23.6
Female-—————HC+12.9-———HC+10.4——DT+2.5
Male-——————DT+4.2————DT+14.8——DT+10.6
<$35K——————HC+18.7-———HC+14.8——DT+3.9
$35K-$75K————DT+10.0-———DT+16.8——DT+6.8
>$75K——————HC+7.9————DT+0.6-——DT+8.5
http://www.freerepublic.com/focus/f-news/3461473/posts