The betting website PredictIt now puts the odds of Donald Trump winning the White House this November at 22%. I don’t think that estimate is too far off: Trump is ahead in virtually every recent Republican primary poll, and he trails Hillary Clinton by 3.4 percentage points in a general election matchup (according to the RealClearPolitics polling average). If there is, say, a 1-in-2 chance that he wins the Republican nomination and a slightly less than 1-in-2 chance (conditional on winning the nomination) that he defeats the Democratic candidate in the general election, then 22% seems just about right.

At the very least, the possibility of a Trump victory is not so remote as to make it too soon to ask ourselves: what would life under President Trump look like? To some extent, the answer depends on what the next Congress looks like — many of Trump’s proposals (like lowering the top individual income tax rate to 25%, ending birthright citizenship, and issuing concealed carry permits that are valid in 50 states) would require statutory changes. But not all of them. For example, Trump could accelerate the pace of deportations without any new legislation (though it would help if Congress allocated funds to triple the number of Immigration and Customs Enforcement officers, as Trump hassaid it should). Moreover, 8 U.S.C. § 1182(f) already gives the President broad authority to “suspend the entry of all aliens or any class of aliens” when he (or she) finds that their entry “would be detrimental to the interests of the United States.” (Others — including my colleague Eric Posner — have discussed whether the First or Fifth Amendment would prevent President Trump from banning Muslim immigrants from entering; I won’t weigh in on that question except to point out that President Trump presumably could ban some or all immigrants under § 1182(f) as long as he did not target immigrants of a particular religion, race, ethnicity, gender, or other “suspect class.”)

What checks and balances might prevent President Trump from taking executive actions along these lines (e.g., an executive order banning all immigrants from entering the United States for a period of time)? Conceivably, Congress could repeal § 1182(f) — though President Trump presumably would veto the repeal legislation, and Congress could override the veto only with the support of two-thirds of the House and two-thirds of the Senate. Alternatively, a federal court might block President Trump’s executive order — though I don’t know what theory the court would use. Could a court set aside the President’s finding that entry of these immigrants “would be detrimental to the interests of the United States”? The President’s action might be “arbitrary and capricious” under the standards of the Administrative Procedure Act, but the Supreme Court held in Franklin v. Massachusetts that the President is not an agency subject to the APA.

That leaves, as perhaps an unlikely suspect, the Department of Homeland Security. By this, I don’t mean the Secretary of Homeland Security — Trump would want to pick a loyalist for that post. I mean the Department of Homeland Security rank-and-file — the hundreds of thousands of DHS employees whose active participation would be necessary to implement Trump’s policy (even if they have no formal say in the matter under the Constitution).

In a dissenting opinion three years ago, Chief Justice John Roberts emphasized the extent to which administrative agencies can prevent a president from carrying out his or her agenda. Citing work by two of his colleagues on the court, Roberts wrote:

Although the Constitution empowers the President to keep federal officers accountable, administrative agencies enjoy in practice a significant degree of independence. As scholars have noted, “no President (or his executive office staff) could, and presumably none would wish to, supervise so broad a swath of regulatory activity.” Kagan, Presidential Administration, 114 Harv. L.Rev. 2245, 2250 (2001); see also S. Breyer, Making Our Democracy Work 110 (2010) (“the president may not have the time or willingness to review [agency] decisions”). President Truman colorfully described his power over the administrative state by complaining, “I thought I was the president, but when it comes to these bureaucrats, I can’t do a damn thing.” . . . President Kennedy once told a constituent, “I agree with you, but I don’t know if the government will.”

Roberts wrote those words to make a very different point: in the case three years ago, he was arguing that the Supreme Court should not defer to the Federal Communications Commission on a particular issue. Moreover, presidential power vis-à-vis “independent” executive branch agencies such as the FCC is even weaker than with respect to Cabinet departments. But the point applies across the executive branch — and it would almost certainly apply to a President Trump. As president, Trump might be able to do more than “a damn thing,” but he wouldn’t be able to do everything he wanted to — or even everything the Constitution allows him to. He may become Commander-in-Chief over a military of 1.5 million soldiers and sailors, but an even larger corps of 2.7 million federal civilian employees would still stand in his way.

Would Homeland Security employees engage in outright civil disobedience under President Trump? Perhaps not. But between outright disobedience and vigorous enforcement of a president’s policies, there is a wide spectrum of potential bureaucratic responses. These responses will vary depending on the president’s policy — for example, if Trump orders Homeland Security to stop everyone except for U.S. citizens and lawful permanent residents at the border, then disobedience by an individual agent may be relatively easy to detect. (But if President Trump banned all foreigners, how would he fill the rooms at his hotels?) On the other hand, if Trump orders Homeland Security to stop every foreigner who arrives at a U.S. airport without a nonrefundable return ticket, then the individual agent might have more room to decide just how closely she will examine the airline refund policy in fine print.

These are, of course, hypothetical responses to hypothetical policies: we don’t yet know what a President Trump will do (and hopefully we never will know). We do, however, have many historical examples of “bureaucratic drift” — that is, of an agency pursuing goals that diverge from those of its presidential principal. The Environmental Protection Agency under President Reagan is one example: B. Dan Wood finds that “following the Reagan inauguration, the EPA bureaucracy bucked the administration and used its slack resources to substantially increase surveillance of pollution sources.” EPA Administrator Anne Gorsuch Burford managed to reverse this trend for a time, but her stormy tenure ended in resignation after less than two years, after which “[t]he committed bureaucrats who remained at EPA . . . again found slack to restore and intensify their surveillance activities.” Wood and coauthor Richard Waterman document a somewhat similar phenomenon at the National Highway Traffic Safety Administration under President Carter: while Carter and his NHTSA administrator both favored a more robust regulatory regime, the lower-level employee in charge of engineering evaluations did not share that regulatory philosophy, and the number of auto safety defect evaluations dropped sharply. And there are countless more cases of drifts less dramatic — many of which might not be observable to academics and other outsiders.

To be sure, the drift dynamic depends on the preferences of bureaucrats being different from the preferences of the president. And the former are difficult to measure. We do, however, have data on political contributions by employees at various federal agencies. And so we know that at Homeland Security, 61% of political contributions (by dollar amount) went to Democrats in the 2012 election cycle. We know that of the small number of Homeland Security employees whose 2016 presidential campaign contributions are reported along with their place of work in the OpenSecrets database, 58% donated to Democrats (primarily to Clinton) and none gave to Trump. And according to one poll of workers across federal agencies this past August, more than two-thirds said they had an unfavorable opinion of Donald Trump (the unfavorable figure was 65% for Ted Cruz, 54% for Clinton, 46% for Marco Rubio, and 40% for Bernie Sanders).

All of which is to say that in thinking about a Trump presidency (or, for that matter, an administration led by anyone whose views are far afield from the federal rank-and-file), we should think not only in formal separation-of-powers terms but also in practical terms of bureaucratic drift. Even where a President Trump would have constitutional and statutory authority to pursue his policies via executive action, he might not have the bureaucratic buy-in necessary to carry those policies through. Nor could he easily replace the federal rank-and-file with a bureaucracy of Trump followers: aside from the tenure protections of the Civil Service Reform Act, the existing rank-and-file is further insulated by the fact that outsiders couldn’t just step in and do their jobs. Say what you want about “good enough for government work”; the fact of the matter is that longtime federal employees have accumulated loads of agency-specific expertise that wouldn’t be easy for new hires to rebuild.

These practical observations raise difficult normative questions — questions that could consume anentire book. Assuming it is true that rank-and-file federal officials could make it difficult for a President Trump to implement his most extreme policies, that doesn’t necessarily mean that they should. After all, Trump will only become President Trump if he wins the November election, which would imbue even his most disturbing proposals with a degree of democratic legitimacy. Under those circumstances, what would be the ethical obligations of a rank-and-file federal worker unwilling to carry out Trump’s directives? As a lawyer rather than a philosopher, I’m not sure I’m competent to answer that question. But with Trump just days away from winning his second primary in South Carolina, I don’t think it’s too early to ask it.

@DanielJHemel