How worried should we be about Donald Trump becoming president?

This is not the same question as “what are the odds that he’s going to win?”

The reason is because it makes sense to be quite concerned about a potential catastrophe, even if the odds are low that it will happen.

My view is that a Trump presidency would be a genuine catastrophe, at a completely different level than anything the country has ever seen before in terms of presidential politics. Besides being almost unbelievably ignorant about pretty much every topic relevant to the office, Trump appears to be a mentally unstable person who shouldn’t be in command of a set of steak knives, let alone a set of nuclear launch codes.

So, in my view, a ten percent chance of Trump becoming president is in itself a sort of ongoing disaster, as in Apocalypse x .1 = Omigod This Can’t Be Happening. Which brings us to Thomas Edsall:

There is also strong evidence that most traditional public opinion surveys inadvertently hide a segment of Trump’s supporters. Many voters are reluctant to admit to a live interviewer that they back a candidate who has adopted such divisive positions.

In matchups between Trump and Hillary Clinton, Trump does much better in polls conducted online, in which respondents click their answers on a computer screen, rather than in person-to-person landline and cellphone surveys.

An aggregation by RealClearPolitics of 10 recent telephone polls gives Clinton a nine-point lead over Trump. In contrast, the combined results for the YouGov and Morning Consult polls, which rely on online surveys, place Clinton’s lead at four points.

Why is this important? Because an online survey, whatever other flaws it might have, resembles an anonymous voting booth far more than what you tell a pollster does.

In a May 2015 report, Pew Research analyzed the differences between results derived from telephone polling and those from online Internet polling. Pew determined that the biggest differences in answers elicited via these two survey modes were on questions on which social desirability bias — that is, “the desire of respondents to avoid embarrassment and project a favorable image to others” — played a role.

In a detailed analysis of phone versus online polling in Republican primaries, Kyle A. Dropp, the executive director of polling and data science at Morning Consult, writes:

Trump’s advantage in online polls compared with live telephone polling is eight or nine percentage points among likely voters.

This difference, Dropp notes, is driven largely by more educated voters — those who would be most concerned with “social desirability.”

These findings suggest that Trump will head into the general election with support from voters who are reluctant to admit their preferences to a live person in a phone survey, but who may well be inclined to cast a ballot for Trump on Election Day.

This is a kind of variation on the so-called Bradley effect, the continued existence of which is controversial. Will there be a Trump effect, and if so will it be strong enough to overcome the many excellent reasons to think that Trump will be an exceptionally weak candidate?

Other reasons to worry include:

The argument that Hillary Clinton is herself an exceptionally weak candidate.

This argument is usually based on some combination of the claims that she’s a bad campaigner, that she’s going to have to deal with serious legal complications because of the email thing or whatever other “scandal” the GOP cooks up, and that she’s a prototypically establishment figure at a time when the zeitgeist is going very much in an anti-establishment direction. I think the first two claims are seriously overstated, while the third is a genuine cause for concern.

The argument that Trump is a unique candidate, and that existing precedents for analyzing his strengths and weaknesses are of relatively little value.

Trump certainly is a unique candidate, but his uniqueness seems to be manifested in ways that are unambiguously bad for his election chances (there’s never been a major party candidate with anything like both his negatives, and his extra-political fame. In other words everybody knows him and most people hate him because they know him, which seems like an exceptionally unpromising combination). So the argument that you can’t really assess his chances seems like bare assertion at best.

The argument that something could happen that could change the present political landscape in a radical way: a big terrorist attack, an HRC health crisis, or who knows what else.

Again, this seems to add up to the claim that while things look bad for Trump now, they could look different five months from now, which is true by definition, and therefore trivial.

The argument that Trump will get massive media coverage, even beyond that given to a typical presidential nominee, because he’s great for ratings, and that this coverage will normalize him, because of the reflexive structure of political coverage in this country (“both sides do it, the truth lies somewhere in the middle,” etc.)

I think this is probably the single best reason to worry that Trump’s chances are better than conventional analysis would suggest.

In any case, there’s going to be a lot of gin in my freezer between now and November.