Switching citation styles late: what breaks

Two weeks before submission, someone tells you the department wants Vancouver and you have written in APA. Or a journal's guidelines say author–date and your manuscript is numbered.

Whether that is an afternoon or a fortnight depends entirely on which direction you are going, and the difference is not about how much punctuation changes.

Two of the four conversions are cosmetic

Conversions divide by whether they stay inside a family or cross between them.

APA → Harvard                    SAFE
  author–date to author–date
  → punctuation, ampersands, order of elements
  → identity unchanged: still author + year
  → a reference manager does this properly

Vancouver → IEEE                 SAFE
  numeric to numeric
  → bracket shape, list formatting
  → identity unchanged: still position in list

APA → Vancouver                  CAREFUL
  author–date to numeric
  → identity moves from (author, year) to a NUMBER

Vancouver → APA                  HARDEST
  numeric to author–date
  → identity must be RECOVERED from the list,
    because the text does not contain it

The first two are genuinely formatting. Your citations still identify their references the same way afterwards; only the characters differ. Do these whenever you like, including in your final week.

The other two are not formatting operations, and calling them that is how people lose a weekend.

What "identity" means here, and why it moves

Every citation has to say which reference it means. Different families do that differently, and a cross-family conversion swaps the mechanism.

AUTHOR–DATE
  (Okafor, 2021)
  identity = the author and the year
  → the reference list can be reordered freely
  → sorting is a convenience

NUMERIC
  [12]
  identity = the position in the list
  → the reference list CANNOT be reordered
  → the order IS the meaning

Converting between them does not restyle the
identity. It replaces it.

That is the whole risk in one line. In a numeric style, if the list order changes, every citation silently means something different — the text still reads [12] and [12] is now a different paper. Nothing about the document looks wrong.

A numeric citation is a pointer into a list. Move the list and you have not broken a link — you have quietly redirected it.

Author–date to numeric: the mechanical direction

This one is laborious but safe if done in the right order, because all the information you need is already in your text.

Every citation names its author and year, so each can be resolved to an entry and given a number. The number is assigned by first appearance in the text, which means the reference list must be rebuilt in that order rather than sorted alphabetically.

Three things to watch. Multiple citations in one bracket collapse: (Okafor, 2021; Chen, 2019; Duarte, 2020) becomes [12,13,14] and the internal order changes from whatever you wrote to numeric sequence. Repeat citations reuse the first number rather than getting new ones. And secondary citations need care, because only one of the two works named has an entry — so only one of them gets a number.

Numeric to author–date: the direction that hurts

Here is why this one is different in kind. [12] contains no author and no year. The information needed to write (Okafor, 2021) is not in the sentence — it is in the list, at position twelve.

So every citation in your text has to be resolved by looking it up, and the conversion inherits every existing numbering error rather than revealing it.

Your text        …improved adherence [12]…
List position 12  Okafor, T. (2021)

But an entry was deleted at position 7 in year
three, and the list renumbered while the text
did not.

Position 12 now holds   Duarte, P. (2020)

Before conversion   [12] — silently wrong, and
                    invisible to reading
After conversion    (Duarte, 2020) — silently
                    wrong, and now ATTRIBUTED
                    to a named person in your text

The conversion did not create the error; it was already there. What it does is promote it from an invisible pointer to a visible claim about who said what. A reader can no longer discover the mistake by following a number — they will simply believe your text.

Which is the practical warning worth taking from this post: check the numbering before converting away from it. Confirm that a sample of citations across the document still resolve to the entries you think they do. If several do not, you have found a numbering drift, and fixing it before conversion is an hour; discovering it afterwards means re-checking every citation in a document that no longer contains the numbers.

And if a drift shipped unnoticed, the consequence is not merely untidy: your text attributes findings to researchers who did not produce them. Nobody would call an unnoticed renumbering misconduct. Knowing about it and submitting anyway is a different matter, because at that point you are asserting an attribution you know to be false.

What your reference manager will and will not do

A manager restyles the list excellently and does considerably less to your text than people assume.

It will re-render every entry in the new style, reorder the list, handle the punctuation, and — if your citations are live fields rather than typed text — rewrite the in-text markers too.

It will not know that two of your entries are the same work, that one entry is cited under a co-author's name, that a citation points at nothing, or that your numbering drifted three years ago. It restyles what it holds. Nothing in a style change inspects whether the underlying pairing is correct.

Which produces the ordering rule: check first, convert second. A conversion applied to a list with defects produces a beautifully formatted list with the same defects, and you will have spent the effort twice.

Three things that surface during a conversion

Year suffixes appear from nowhere. Numeric styles hide the problem of two works by the same author in the same year, because numbers distinguish them. Convert to author–date and both become (Okafor, 2021). You now need 2021a and 2021b, and assigning them requires knowing which citation meant which work — which the numbers told you and the text does not.

Et al. rules change under you. The number of authors that triggers the abbreviation, and whether the reference list may use it, differ by style — materially, not cosmetically. Converting a list that used "et al." at six authors into a style that lists twenty means finding the full author lists again.

Orphans become visible. Numeric lists contain only cited works by construction, so converting to author–date and sorting alphabetically often reveals entries that were never cited at all — they had a number nobody used, and nothing made that obvious.

Typed citations versus live fields

One practical distinction decides whether a conversion takes an afternoon or a week, and most people do not know which situation they are in until they try.

LIVE FIELDS
  citations inserted by a reference manager
  → the document stores a link to a record
  → changing style rewrites text AND list
  → conversion is largely automatic

TYPED TEXT
  citations typed, or pasted, or converted to
  plain text at some point
  → the document stores characters
  → changing style rewrites the LIST ONLY
  → every in-text citation is manual

MIXED
  the common case: fields early, typed later,
  or fields lost when a file was converted
  → the manager silently updates half of them
  → and the halves no longer agree

The mixed case is worth checking for specifically, because it produces the most confusing failure: you run the conversion, the list restyles perfectly, and roughly half your in-text citations change while the rest stay as they were. It reads as a bug in the software and is simply a document with two kinds of citation in it.

To find out which you have, click on a citation in your text. If it highlights as a field or shows a grey background, it is live. If the cursor sits inside it like ordinary words, it is typed. Check several, in different chapters — a document that was passed between machines, converted between formats, or edited by a co-author without the same manager will frequently have both.

If your citations are typed text, a style change is a rewrite of every citation in the document, and no tool is going to do it for you.

The order that works

One: check the current document, in its current style. Every citation resolving to an entry, every entry cited, numbering verified if numeric. Fix what you find.

Two: resolve identity problems while the old identity still exists. Year suffixes, duplicate entries, works cited under two names. These are far cheaper to fix before the mechanism changes.

Three: convert. Manager if your citations are fields; carefully and by hand if they are typed text.

Four: check again, in the new style. Not because the conversion is untrustworthy, but because this is the pass that catches what the conversion could not know — and it is fast, because you fixed everything else first.

The whole thing is an afternoon if the list was sound and a fortnight if it was not, which is really an argument for checking at a moment when nobody has just told you to change styles.