ingridleiria Abstract and Title
Most people who ever encounter this research will encounter only the title and the abstract. That is not a claim about a ratio, it is a claim about where the objects live: databases index the title and the abstract, search returns them, alerts deliver them, reference lists carry the title alone, and paywalls stop at the abstract. The full text is behind all of that. So the two shortest pieces of the paper are the ones most people read, and they are what most readers decide on. That includes the handling editor deciding whether to desk reject, the referee deciding whether to accept the invitation, the researcher deciding whether to download it, and the search engine deciding whether to return it at all.
The failures are specific. An abstract written from intentions rather than results, because it was drafted early and never revised, so it promises a question the paper stopped asking. An abstract with no number in it, which forces the reader to open the paper to learn anything and, more often, to move on. An abstract whose verbs are stronger than the results section's, which is the single most common overclaim in academic writing and the one a referee sees first. A title naming the topic instead of the finding, which is invisible to anyone searching for what the paper actually shows. And a title with a clever pun, which ages badly, translates badly, and is skipped by every reader scanning a list of forty.
These are cheap to fix and expensive to leave. This is the last prose written in the manuscript, after results-writing and discussion-and-conclusion and after introduction-writer, and it is written from the finished paper rather than from what the author remembers about it.
When to use this, and when not to
Use it to write an abstract from a finished paper, to repair or shorten one, to convert an unstructured abstract into a structured one or the reverse, to generate and rank title candidates, to choose keywords and JEL codes, to write highlights, and to produce a second-language abstract where the venue or the university requires one.
Use it also when the paper is over a submission form's word limit, when a conference requires an abstract before the analysis is complete, and whenever a number in the paper has changed since the abstract was last touched.
Do not use it to write the introduction, which is introduction-writer and which the abstract compresses rather than duplicates. Do not use it to choose the venue whose limits and conventions this skill writes to, which is journal-targeting. Do not use it to write the cover letter or the highlights of a submission package beyond the fields listed here. Do not use it to write a talk title or a poster, which have different economics and are covered by conference-presentation-deck.
Do not write an abstract before the results are final. Every rule in this skill depends on there being a number to state, and an abstract written earlier will be wrong in exactly the place readers look first.
What you need before starting
The finished results with the headline number, its units and its exhibit. Missing: the abstract cannot be written. Where a deadline forces a draft anyway, write it with a visible placeholder in the finding sentence and mark the document so it cannot be submitted by accident.
The research question as a sentence. It becomes sentence one, compressed. Missing: take it from the introduction's third move; if it is not there either, the problem is upstream and belongs to research-design.
The design name, the data source, the units, the years and the sample size. Missing: read the data and strategy sections. These five facts occupy one sentence and they are what tells an editor whether to take the paper seriously.
The discussion's comparison sentence. The one that makes the magnitude legible. Missing: write it here and then check it matches the discussion, since the two must agree.
The target's limits and format. Word limit, structured or unstructured, whether citations are permitted, keyword count, whether highlights are required and at what character limit, whether JEL codes or other subject classification terms are needed. Missing: take them from the submission system and from three recent articles, since the guidelines and the practice often differ.
The keyword and classification lists in force. The journal's own keyword list where it has one, and whichever subject classification the field uses. Economics uses JEL. Health, medicine and much of public health use MeSH. Psychology uses the APA thesaurus of psychological index terms, which is what PsycINFO indexes on. Education uses ERIC descriptors. Many journals in sociology, management and the qualitative traditions have no controlled vocabulary at all and take free keywords. Missing: use the field's vocabulary from recent articles in the target and mark the codes as needing verification; do not guess a code or a descriptor from memory, since every one of these lists is revised.
The verbs used in the results section. Missing: read them. The abstract's calibration is copied from the results, not chosen fresh.
The method
Assemble the five facts and the finding before writing. Question, design, data, magnitude with units, comparison. Every abstract failure below traces back to writing before these are on the page.
Write sentence one: the question and why it is open. What is asked, in what setting, in one sentence, occasionally two. No throat-clearing about the importance of the topic in general, no "in recent years", no sentence that would be true of forty other papers. The rule: if the first sentence could open any paper in the field, it is doing no work and is deleted.
Write sentence two: the data and the design. Source, units, years, sample size, and the identification strategy named. "Using administrative records for 1,284 schools between 2016 and 2022 and a regression discontinuity around the eligibility threshold" is the whole sentence, and it tells an editor more than a paragraph of motivation would.
Write sentence three: the finding, with its magnitude. Direction, size in interpretable units, and precision where it matters. "Attainment rises by 0.09 standard deviations in mathematics, with no detectable effect in reading" is an abstract sentence. "We find significant effects" is not, and reviewers read it as an author who does not want to commit.
Write sentence four: the mechanism or heterogeneity, only if the paper establishes it. If the paper does not establish one, this sentence does not exist. An abstract that speculates is worse than an abstract that is short.
Write sentence five: the contribution or implication, sized to the evidence, ending on the finding. Not a call for future research, not a claim about what the field should now do. One sentence, matching the conclusion.
Check every number against its exhibit, at the same rounding used everywhere else in the paper. This check finds an error in a large share of drafts, because the abstract is the section most often revised last and least often rechecked.
Run the verb pass. Compare the abstract's verbs against the results section's, word for word. Any upgrade is corrected in the abstract, never by upgrading the results. Where the design supports only association, the abstract says so; there is no venue where an abstract may claim more than the paper.
Cut to the limit by removing sentences, not by compressing them. An abstract at 320 words with a 250 limit almost always contains a sentence with no job, usually a second motivation sentence or a methods detail. Compressing every sentence by twenty percent produces a dense, unreadable paragraph; removing one sentence produces a clean one.
Generate eight to twelve title candidates across formats, then rank them against the criteria below. Generating fewer than eight produces variations on the first idea, which is nearly always a descriptive title of the topic.
Choose keywords, classification codes and highlights, each against the current list for the field the target sits in, each checked rather than recalled.
Read the abstract cold, as a stranger. Does it say what was asked, on what data, with what design, what was found, and how big? If any of the five is missing, it is not finished, whatever the word count says.
Diagnosing a weak abstract
Mark each sentence with the job it performs from the list of five, or "none". This takes three minutes and locates the problem precisely.
| What the marking shows | What it means | The fix |
|---|---|---|
| Two sentences performing job one | The author is motivating rather than reporting | Delete the weaker one; it is usually the first |
| No sentence performing job three | The abstract predates the results | Insert the finding with its number from the exhibit |
| Job three present without a magnitude | Hedging, or the number is not settled | Add the size in interpretable units |
| No sentence performing job two | The design is weak or the author is uneasy about it | Name the data and the design; if that reads badly, the problem is the design |
| Sentences marked "none" | Padding, usually inherited from an earlier draft | Delete |
| Job five is a call for future research | The paper's contribution has not been decided | Replace with the contribution from the conclusion |
| Verbs stronger than the results section's | The abstract was written in one sitting to sound important | Copy the results section's verbs |
Show the marking before the rewrite. It is the part authors learn from, and it makes the rewrite arguable rather than imposed.
Titles and how to rank them
Generate across all five formats before ranking. The formats:
Declarative, stating the finding: "Fee caps raised maternal employment in adopting municipalities". Highest information per word and the most searchable. Some journals dislike it; check three recent issues.
Descriptive, stating the object: "The effect of childcare fee caps on maternal employment". Safe, common, and invisible in a list unless the topic is itself unusual.
Question: "Do childcare fee caps raise maternal employment?" Works when the question is genuinely contested and the field expects the form; reads as an undergraduate essay when the question is not contested.
Colon form, broad concept then specific mechanism and population: "Childcare costs and mothers' work: evidence from staggered municipal fee caps". The most common form in the social sciences and the safest default, because it carries the topic for browsers and the specifics for searchers.
Short phrase, for working papers and talks: "The fee cap effect". Not for a journal submission.
Rank the candidates on these criteria, in this order:
| Criterion | Test |
|---|---|
| Does the reader learn the finding or the question | Read the title alone and state what the paper is about |
| Are the population and setting named | A reader in another country should know whether it applies to them |
| Would a search on the paper's key terms return it | Do the terms a searcher would type appear in it |
| Is it under about fifteen words | Count them |
| Does it avoid a pun, an allusion or a joke | Would it still read well in ten years and translate cleanly |
| Is it distinct from existing titles | Search the exact candidate; a title matching a published paper is rejected |
Check the shortlist for collisions before committing. A title identical or near-identical to an existing paper causes citation confusion and looks careless.
Keywords, classification codes and highlights
Keywords, five to seven, drawn from the field's vocabulary as it appears in recent articles in the target, and covering three dimensions: the topic, the method, and the setting or population. Repeating words already in the title wastes them, since search covers both; use the keyword slots for the synonyms a searcher might use instead of yours. Where the journal supplies a controlled list, use it exactly.
JEL codes, two to four for economics venues, verified against the current classification rather than remembered. The first should be the primary field, the others the method and the applied area. Where a connected lookup or the official classification is unavailable, take the codes from three recent articles on the same topic in the same journal and say that is how they were chosen.
Classification outside economics. JEL is an economics convention and most other fields use something else, so the slot is filled with the target's own vocabulary rather than left empty or filled with JEL out of habit. Health and public health journals index on MeSH, and choosing the terms deliberately matters because MeSH drives what a PubMed search retrieves; take them from the MeSH browser or from the indexed record of two recent articles on the same topic. Psychology journals use the APA thesaurus of psychological index terms, the vocabulary PsycINFO indexes on, usually three to six terms. Education journals use ERIC descriptors, which are also a controlled thesaurus and also worth checking rather than guessing. Sociology, management and most qualitative venues take free keywords, in which case the keyword rules above carry the whole load and the synonym discipline matters more, not less. Whichever list applies, record which vocabulary was used and how it was verified, exactly as with JEL.
Highlights, where required, three to five bullets under the character limit the journal sets, commonly 85 characters. Each is a finding with a number, not a description of the paper. "Fee caps raised maternal employment by 2.1 percentage points" is a highlight; "This paper studies childcare policy" is not. They are extracted from the abstract, not written fresh.
Second-language versions. Where a resumo, resumen or other second-language abstract is required, translate the finished abstract preserving the numbers and the sentence structure, then check the technical terms against how the field writes them in that language rather than translating them literally. Statistical and design vocabulary is the part that goes wrong: the local convention for naming a design is often not the literal translation of the English name.
Citations in the abstract. Most venues forbid them. Where one is required or permitted, use the same hyperlinked author-year form as the body, with the DOI behind it, and ensure the work appears in the reference list.
Worked example
Situation. Dr. Ines Kovac had a finished paper on a national tutoring programme, an accepted internal draft, and an abstract of 312 words against a journal limit of 200. The abstract had been written fourteen months earlier when the design was a matched comparison; the paper now used a regression discontinuity. Submission was in three days.
Task. A 200-word abstract, a title, five keywords, three JEL codes and four highlights, all consistent with the final paper.
Action. The marking took four minutes and found the problem. Sentences one and two both performed job one, motivation, and between them consumed 71 words. Sentence three named the old design. Sentence four said the programme "significantly improved outcomes" with no number. Two sentences at the end were marked "none": one described the data cleaning, and one said the findings had implications for policymakers in developing countries generally, which was neither true nor checkable.
The rebuild produced five sentences. Job one: the programme reached 340,000 pupils at an annual cost of about 96 million and its renewal was under review, with no credible estimate of its effect on measured attainment. Job two: administrative test records for 1,284 schools from 2016 to 2022, and a regression discontinuity around the eligibility score. Job three: mathematics attainment rose by 0.09 standard deviations, 95 percent confidence interval 0.03 to 0.15, with no detectable effect in reading and an interval excluding effects above 0.04. Job four: the effect was concentrated in schools that had been below the median attainment level before the programme. Job five: the estimated cost per standard deviation gained, which placed it against two alternatives the ministry was considering.
That came to 198 words. The number check found one problem: the abstract said 0.09 and the table said 0.086, which was correct at the paper's rounding convention, but the confidence interval had been copied from an earlier specification and read 0.04 to 0.16 rather than 0.03 to 0.15. That error had survived three internal readings.
The wrong turn was the title. The first candidate everyone liked was "Tutoring at the margin: a lesson in targeting", which used a pun on "lesson" and won the room. It was dropped for three reasons applied from the ranking table: a reader learns neither the finding nor the setting from it; a search for tutoring effects or attainment would not return it; and the pun does not survive translation, which mattered because the university required a second-language abstract and the title travelled with it. The eventual title was the colon form: "Tutoring and attainment in primary schools: regression discontinuity evidence from a national programme". Twelve words, both terms a searcher would use, setting named.
Keywords avoided repeating title words and picked up the synonyms: remedial education, pupil attainment, programme evaluation, eligibility threshold, education policy. JEL codes were checked against the current classification rather than remembered, which changed one of the three the author had used on a previous paper. Highlights were extracted from the abstract, four of them, each under 85 characters and each carrying a number.
Result. Submitted on time. The interval error was the material catch, since it would have been in the published abstract and would have disagreed with Table 3 in the same paper. Total time about three hours, of which forty minutes went to the title discussion that ended with the pun being discarded.
A second scenario, where it goes differently
A conference abstract due six weeks before the analysis will be finished, and a structured abstract for a management journal. Both change the rules in ways worth naming.
The conference case is the harder one because the honest constraint is that there is no finding yet. What works: write jobs one and two in full, since the question and the design are settled, then state precisely what will be reported and by when, in the future tense, with the sample already in hand. What does not work: writing a finding sentence in the conditional and hoping the results cooperate. Every year, someone presents a poster whose abstract promises an effect the paper did not find, and the audience notices. Where the conference permits an abstract update before the programme is printed, note the date and diarise it.
The structured abstract changes the container and not the content. Purpose, Design and methodology, Findings, Originality and value, each with a heading and often each with its own word allowance. The five jobs map onto the four headings without loss: job one to Purpose, job two to Design, jobs three and four to Findings, job five to Originality. The trap specific to this form is that Originality invites a claim about novelty, and the safe version states what is new in terms a referee can check, such as the first estimate using administrative rather than self-reported outcomes, rather than asserting that the study is novel.
Output
ABSTRACT [n words against a limit of m]
[1. Question and setting]
[2. Data and design: source, units, years, N, strategy named]
[3. Finding with magnitude and precision]
[4. Mechanism or heterogeneity, only if established]
[5. Contribution or implication, ending on the finding]
TITLE
Chosen: [the title]
| # | Candidate | Format | Finding or question visible | Setting named | Searchable | Words | Collision |
KEYWORDS
[5 to 7, covering topic, method, setting, not repeating title words]
CLASSIFICATION CODES
[The vocabulary the target uses: JEL for economics, MeSH for health, APA thesaurus
or PsycINFO terms for psychology, ERIC descriptors for education, free keywords
where the venue has none. 2 to 6 terms, verified against the current list, with
how they were verified]
HIGHLIGHTS
[3 to 5, each under the character limit, each a finding with a number]
CHECKS
| Number in the abstract | Value | Exhibit and cell | Rounding matches |
Verbs match the results section: yes / no
Second-language version required: yes / no, and its status
Failure modes
Written from intentions. Recognise it when the design named in the abstract is not the design in the paper, or when there is no number. It happens because the abstract was drafted early. Rewrite from the finished results, never patch.
No number. Recognise it by "significant", "positive", "substantial" with nothing quantified. Readers cannot evaluate the paper and will not open it to try.
Verb inflation. Recognise it by comparing the abstract's verbs with the results section's. The abstract is where overclaiming is most visible and most damaging, because the referee reads it first and rereads the paper in its light.
Two motivation sentences. Recognise it by marking the sentences; jobs one and one. Delete the weaker.
Compressing instead of cutting. Recognise it by an abstract at exactly the limit with no sentence over eleven words and no air in it. Restore the sentence structure and delete a whole sentence instead.
A pun in the title. Recognise it by the fact that the room laughed. It costs search visibility permanently and translation immediately.
A descriptive title where the paper has a clear finding. Recognise it when the title would fit a paper that found the opposite. If the finding is clean, state it.
Keywords repeating the title. Recognise it by overlap. Search already covers the title; use the slots for the synonyms someone else would type.
Classification codes from memory. Recognise it by JEL codes, MeSH headings or ERIC descriptors carried over from a previous paper, or by JEL codes appearing on a submission to a journal that does not use them. Verify against the current list for the target's field, and if that is not accessible, take them from recent articles in the target and say so.
Stale after a revision. Recognise it by any revision that changed a number. The abstract is the last thing rechecked and the first thing read; recheck it immediately before submission every time.
Edge cases
A null headline result. State it as a finding with the excluded range: "the estimates rule out effects larger than 0.04 standard deviations, below the 0.12 reported in earlier work". Never describe the paper as failing to find an effect, which sounds like a failed study rather than an informative one.
No word limit given. Write 150 to 200 words. Long abstracts are skimmed, and skimming loses the finding sentence more often than any other.
A thesis abstract. Longer, often one page, and it covers the whole thesis rather than one chapter: the thesis question, the chapters and what each establishes, the overall contribution. The five jobs become five paragraphs. University regulations usually set the exact length and often require a second-language version, so check the regulations rather than the field convention.
A paper with several outcomes and no single headline. Choose the outcome the question is about and lead with it; the others get one clause. An abstract that reports six outcomes evenly communicates none.
A descriptive paper with no model or question. The abstract will expose it, since job one has nothing to say and job three has no estimate. Do not paper over it. If the contribution is genuinely a measurement or a documented pattern, the abstract states what could not previously be measured, what the pattern is with numbers, and what changes now it is known; if there is no such claim, the fix is research-design, not phrasing.
A journal requiring both a structured and an unstructured abstract, which some do for different parts of their system. Write the unstructured one first and derive the structured version, since deriving in that direction preserves the flow.
Anonymous review. Check the abstract for anything identifying: a named dataset only one group holds, a self-citation phrased as "our earlier work", the name of a partner institution. This is easy to miss because the abstract is written last.
Quality bar
- Every sentence has one of the five jobs, and no sentence has none.
- The finding appears with a magnitude in interpretable units and its precision.
- The data source, units, years, sample size and design are all named in one sentence.
- Every number in the abstract appears in an exhibit at the same rounding, checked against the cell.
- The verbs are identical in strength to the results section's, checked in a dedicated pass.
- The title states the finding or the question, names the setting, is under about fifteen words, and does not collide with an existing title.
- Keywords and classification codes were checked against the current list for the target's field, JEL or MeSH or the APA thesaurus or ERIC descriptors as applicable, and the method of checking is recorded.
- The abstract was rewritten, not patched, after the most recent change to any number in the paper.
Adapting this to your context
The sentence jobs, word counts and title formats here come from empirical social science articles, mostly economics and adjacent applied fields, at journals taking unstructured abstracts of 150 to 250 words. The shape travels further than the specifics do.
- The five sentence jobs. They assume one headline estimate. A structured abstract in health or psychology usually wants Background, Methods, Results, Conclusions, sometimes Objective and Limitations too; map the five jobs onto those headings rather than fighting the form.
- "No number, no finding". That assumes an estimate exists. For qualitative work, replace the magnitude with the claim and its base: how many participants, over what period, and what the central pattern was.
- The classification slot. JEL is economics only. Use MeSH for health, APA thesaurus or PsycINFO terms for psychology, ERIC descriptors for education, free keywords elsewhere. A registered report or systematic review usually also needs its registration number in the abstract.
- The 85 character highlight limit and 200 word default. Both are publisher habits. Take the real limits from the submission form; medical journals often allow 300 words with mandatory subheadings.
- What not to change. The abstract is written last, from the finished results, every number checked against the cell it came from, and the verbs never claim more than the results section claims.
Related skills
full-manuscript-build places this last in the writing order and rechecks it in the cross-section pass. results-writing supplies the numbers and the verbs; discussion-and-conclusion supplies the comparison and the contribution sentence; introduction-writer is the section this compresses, and the two must agree. journal-targeting supplies the limits and conventions written to here. literature-verification and references-and-bibliography handle the rare abstract citation. conference-presentation-deck covers talk titles and the spoken version of the finding. peer-review-simulator reads the abstract first, exactly as an editor does, and is the last check before submission.