Politico: Shy Clinton Vote?

[Mr. Morden]

If the polls end up being "off", then I think it's far more likely to happen because the pollsters did a bad job of constructing their samples than because the poll respondents were untruthful about who they supported.

[Mr. Morden]

The "shy Clinton vote" extends beyond middle-aged or elderly women: there are loads of shy Clinton supporters among millennials. There's a lot of vitriol that's immediately directed towards any millennial, particularly if they're among white men, for saying anything positive about Clinton so male millennials are inclined to keep their support for Clinton close to the vest. I learned about this inadvertently when I got into a public argument with a Trump supporter and a bunch of white engineering majors, out of the blue, started saying that they prefer Clinton to Trump.

Hillary Clinton is about as maligned as Osama Bin Laden in parts of the country so I'm not surprised that this phenomenon is real...

But you're talking about these people's unwillingness to acknowledge Clinton support in social settings.  We don't know if that translates to unwillingness to acknowledge it to pollsters.  I'm skeptical of the idea that we can infer a "shy X" effect for either candidate in the polls before the election.

For that matter, even after the election, if the polls are wrong, how do we determine if it's because people lied to pollsters or because the pollsters did a poor job of creating their samples?  Even if the pollsters get the topline demographic mix "right", there could still be hidden selection effects that prevented them from reaching the "right" kinds of Hispanics, or non-college whites or whatever.

[Mr. Morden]

If such an effect exists, it's likely only in live-interview polls (where you are talking out loud to a human) rather than in robopolls or internet polling.  Obviously, the latter two have their fair share of even worse systematic issues, so it would be hard to disentangle regardless.

Right.  People often suggest the "Shy X" theory whenever there's a discrepancy between live phone polls and robo-polls/internet polls.  The implication then is that we should believe the latter over the former, because people will lie to human beings but not to robots.  Except....there's no evidence that robo-polls and internet polls are more accurate than live phone polls.  If anything, the evidence runs in the other direction.

[Mr. Morden]

Should also note that the article linked in the OP doesn't say anything to indicate that there's a "Shy Clinton vote".  It's the opposite: The article says that there are educated women who won't tell their friends and families that they support Clinton.  And the evidence they use is.....polls.  That is, they're assuming that the polls are right, meaning that people are being truthful when they talk to pollsters.  But that they're not always being truthful when they talk to people in real life.