Pinker (The Sense of Style, Viking, 2014) names the main cause of bad academic prose: the writer cannot remember what it was like not to know what they know. Abbreviations that were defined once in chapter two, the district that was described in the proposal but not the paper, the 'obvious' step in the identification argument. The cure is a reader who does not know, and the discipline to believe them when they say they were lost.

Indian English is a standard variety with its own dictionary entries. Spelling (colour, programme), 'lakh' and 'crore' with the figure in millions beside them, Indian place names without gloss for an Indian journal: all fine. What to change is only what stops the reader, and an editor's job is to say which that is. Be consistent within a paper; the journal's style guide decides between British and American spelling, and either is correct.

Economics papers front-load the whole hourglass into the introduction: problem, question, method, main result with the number, why it is credible, what it adds, and a roadmap, in three to five pages. Cochrane and Thomson both insist on it. A reader who stops after the introduction should know everything except the details. Development studies and sociology journals let the introduction stay shorter and put the result later; read three recent articles in the target journal and copy the shape.

150 to 250 words. Setting and question, one sentence. Data and method, one or two. Main result with the number and its direction, one or two. Mechanism or heterogeneity if it is the point. Implication, one. No citations, no abbreviations, no 'this paper explores'. Write it last, rewrite it first.

A policy implication names an actor, an action and the evidence for expecting the action to work. 'Governments should invest in women's empowerment' is none of the three. 'A state SHG programme that adds a savings component at about Rs 40 per member per month could expect the decision-making effect we measure; whether it survives government delivery is the open question' is all three, and it is honest about the limit.

External validity. A reader deciding whether the result transfers needs to know what was particular about this place: the density of front-line workers, the presence of other programmes, the local wage. Every fact in the setting section should reappear in the discussion when you say where the result might and might not hold.

The same three parts in different words. What was done (how many interviews, with whom, selected how, over what period, in what language, transcribed and translated by whom); what makes it credible (the analytic approach, coding, who coded, how disagreements were handled, saturation); and how the reader can check (excerpts tied to the claims, a reflexivity statement, the codebook in an appendix). 'Thematic analysis was conducted' is one sentence where a page is owed.

Significant means statistically significant and should say at what level; for importance use a different word. Effect and impact claim causation; use association where that is what you have. Marginally significant means not significant. Trend toward means nothing. Robust means survived the specific checks you ran, which should be listed.

An introduction that states the shared question and how the papers answer it; three self-contained papers, each publishable, each with its own literature, data, methods and conclusion; a concluding chapter that reads across them. Standard in economics and increasingly accepted in Indian social-science departments, but check the university's ordinance, which may require a specific chapter list.

After a draft, write one sentence beside each paragraph saying what it does. Read only the sentences. Paragraphs that do the same job are merged; paragraphs that do two jobs are split; paragraphs whose sentence cannot be written are cut; paragraphs out of order are moved. It takes an hour on an article and finds the structural problems that reading the prose hides, because the prose is yours and reads well to you.

'The null result comes from the only study in which groups had no savings component, which is consistent with the mechanism we test.' Disagreement explained is a contribution of the review itself.

Census tables, NFHS and PLFS reports, budget documents, programme guidelines and CAG audits are primary sources and should be cited as such, with the issuing body as author, the exact title, the year and table number. 'According to government data' is not a citation. Where a figure has been revised, cite the version used and the date.

The theory section is often a simple model whose comparative statics are the hypotheses. It should be as simple as will produce the predictions the empirics test, and no simpler; a model with no empirical implication is a model for a different paper. If the predictions are obvious without algebra, state them in words and put the model in an appendix.

A reference manager stores every paper with its PDF and notes, inserts citations in Word, LibreOffice or Google Docs, and rebuilds the bibliography in any style in seconds. Zotero is free and open, works offline, and with the Better BibTeX plugin feeds LaTeX and Quarto. Starting a thesis without one means retyping four hundred references by hand in the last month, in the wrong style, with errors that examiners find.

Before: 'The utilisation of a difference-in-differences estimation strategy was undertaken in order to facilitate the identification of the causal impact of programme exposure on the educational outcomes of children in beneficiary households.' (33 words, no character, one buried action.)

After: 'We use difference-in-differences to estimate the programme's effect on children's schooling.' (13 words.) Nothing was lost.

Economists say 'treatment', 'identification' and 'exogenous'; sociologists say 'intervention', 'causal claim' and 'external'. Public health says 'exposure' and 'confounder'; development studies says 'programme' and 'context'. A paper submitted to a journal in one field using another field's dialect reads as foreign before the argument starts. Read the target journal and borrow its words for the same ideas.

Indian supervisors often correct grammar and leave the argument alone, either from courtesy or from pressure of time. The grammar corrections are welcome; the silence on the argument is not approval. Ask directly: 'Is the contribution enough for the thesis?' and 'Which chapter is weakest?' Both questions are answerable in one sentence and both are asked too late.

Columns are specifications, from the sparest on the left to the fullest on the right, so the reader sees what each addition does. The coefficient of interest is the first row. Controls appear as a row of Yes/No, not as thirty coefficients. n, the control mean and R² close the table. Significance stars are still common; several journals now prefer standard errors or intervals alone, and a table that reads well without stars is a better table.

Line for time; dot-and-interval for coefficients; bar for a few categories; histogram or density for a distribution; scatter with a fitted line for a relationship; small multiples for the same plot across groups. Pie charts almost never (Cleveland and McGill, JASA 1984, 79:531, showed readers judge angles badly); stacked bars only when the total matters more than the parts.

A matrix of themes by participant group, with counts or with representative phrases in the cells, shows the pattern a page of quotations hides. A table of participants (role, sex, district, number interviewed) replaces a paragraph. A code tree in the appendix shows the reader how the themes were built. None of this makes the work quantitative; it makes it checkable.

Building the package before submission is when the errors are found: the table made from an older data version, the figure whose axis was mislabelled, the coefficient copied from the wrong column. Every one of those is a correction after publication if a reader finds it first. Quarto and Stata's dyndoc remove the copying step altogether.

What a reader of the target journal would know without looking it up. That randomisation balances covariates in expectation: common knowledge in economics, cite it in a nursing journal. That Bihar is in eastern India: common knowledge. That Bihar's per capita income is the lowest among Indian states: a fact that changes, so cite it with the year. When in doubt, cite.

Splitting one study into the smallest publishable units to raise the count. Three papers on three outcomes from one survey, each with the same introduction, is salami; three papers asking different questions of the same data is a research programme. The test is whether each paper stands alone and cites the others.

A registered plan (AEA RCT Registry, OSF, the Registry for International Development Impact Evaluations) states the outcomes, the specification and the subgroups before the data are seen. It makes a null result publishable, because the reader knows it was not one of thirty tries. It is required by many funders of trials now and is good practice for any confirmatory quantitative study.

Typically 20–40 pages: the argument, the findings chapter by chapter, the contribution, the main tables. Examiners read it before the thesis and sometimes instead of parts of it, so it is written as the case for the degree, not as a summary. Many universities require a presentation to the department at this stage; treat it as the viva's rehearsal and invite the hardest questioners.

Supervisors leave, retire, fall ill or stop responding. The ordinance provides for a change of supervisor and the Research Advisory Committee is the route. Document the silence (dates of drafts sent, meetings requested) before raising it; a scholar with a record is taken seriously, and one without is a complaint. The UGC's 2022 regulations also set grievance procedures every university must have.

The research question is not stated clearly. The literature review describes rather than argues. The methods do not justify the choices made. The findings are presented without interpretation. The conclusions go beyond the evidence. The contribution is not stated. Sections 02 to 05 of this course are the answer to each; the list has not changed in decades because the faults are the natural ones.

Listen to the whole question. Pause. Answer the question asked, briefly, and stop. 'I don't know, but here is how I would find out' is a good answer. 'That is beyond the scope of this thesis' is a good answer once and a bad one three times. Disagree with an examiner when you are right, with evidence and without heat; examiners respect it and some ask a wrong question to see whether you will.

Shodhganga is also the largest open collection of Indian social-science research and almost nobody searches it. A thesis on your district from a neighbouring university, with its instrument and its descriptive tables, is worth more than most journal articles for context, and it is citable: author, year, title, degree, university, and the Shodhganga handle.

Is it indexed where your institution and funders look (Scopus, Web of Science, DOAJ, the ABDC or ABS lists for management schools)? Since February 2025 India has no national journal list: the UGC discontinued UGC-CARE and issued 36 suggestive parameters under eight heads for institutions to apply themselves, so check what your own university's committee accepts. Who reads it: academics only, or the policymakers and practitioners you want? Will it be found: is it open access, or behind a wall most Indian readers cannot pass?

An Indian institution's own working-paper series (IGIDR, ISI, CDS, NCAER, IIMs, the Centre for Policy Research) gives a paper a number, a date and a place on RePEc, and it is where seminar audiences look. If your institution has none, IZA and the World Bank's series take affiliated authors, and SSRN takes anyone. A working paper is also what a job market or a grant application can cite while the journal decides.

A monograph from a thesis is the humanities' and some social sciences' standard, and an academic press (Oxford University Press India, Cambridge, Sage, Routledge, Orient BlackSwan, Permanent Black) will send the proposal and sample chapters for review. The thesis is not the book: examiners wanted proof of competence; readers want an argument, and the literature review and the methods chapter mostly go. Budget a year of rewriting.

'The contribution is unclear' means the introduction failed. 'The identification is not convincing' means a specific assumption needs a specific test. 'The paper is too long' means the appendix is in the main text. 'The authors should cite ...' followed by five papers by one person is a referee asking to be cited; cite the relevant ones. 'The English needs work' means a language edit, and also that the argument was hard to follow.

Referee 1 wants the model; Referee 2 wants it gone. The editor's letter usually says which way to go; if not, do what strengthens the argument, say in the response that the referees differed and what you chose and why, and offer the other version in the appendix. Editors respect a reasoned choice more than a paper that tries to please both and satisfies neither.

An appeal succeeds when a referee made a demonstrable factual error that the editor relied on: misread the design, cited a result the paper does not contain. It does not succeed on a difference of judgement, and it burns the editor's goodwill. Most journals allow one appeal, in writing, and decide within weeks. Use it once in a career, if that.

An error found after publication is corrected with an erratum (the journal's) or a corrigendum (yours), published and linked. An error that changes the conclusions is a retraction, which authors can and should initiate themselves. Both are ordinary parts of science, and a self-reported correction costs a reputation nothing. An error discovered by someone else after you knew of it costs everything.

A model produces fluent, generic academic prose on any topic, and fluent generic prose is exactly what sections 02 to 05 of this course are teaching you not to write. The tell is not grammar; it is the absence of anything a stranger could check: no specific number, no named study, no sentence that could be wrong. Referees have learned the register and read it as a signal that the thinking was outsourced too.

Perfectionism (the first sentence must be right), fear (the supervisor will hate it), and the missing argument (there is nothing to say yet) look identical from outside and have different cures. Write the bad first draft; send the draft anyway; go back to the card with the one-sentence claim. If the claim cannot be written, the block is not in the writing.