Research data sourcing and analysis
NextDataLab is the research data arm of NextPUBGlobal. We design the study, source the data, run the analysis and write the methods — so that when a referee questions your sampling, your assumptions or your model, there is an answer already documented.
A vertical of NextPUBGlobal · A Careerantra Solutions company
Every project runs the same nine stages, and you can see which one you are in. Illustrative view of the client dashboard.
Reporting and integrity standards we work to
Where papers actually fail
A manuscript that reads badly can be rewritten in a fortnight. A study whose sample cannot carry its claim cannot be rewritten at all. By the time a referee says so, the funding is spent, the cohort has dispersed and the only honest fix is to collect again.
The three sentences that end a paper are almost always the same ones. The sample size is not justified. The analysis does not match the design. The results cannot be reproduced from what is described. None of them is about writing. All of them are decided before a single word of the manuscript exists.
NextDataLab exists to take those decisions early, on the record, with someone accountable for them. That is the whole proposition: not a faster analysis, but an analysis that was chosen for a reason you can state out loud.
What we do
Most researchers come to us in the middle — with data already collected, or an analysis a reviewer has questioned. We take work at any point. It costs less when it starts at the top.
Protocols, variable maps, sampling frames, a priori power calculations, instrument development and validation, ethics and consent documentation.
Fieldwork we run or supervise, and existing data located, licensed and merged into something you can actually model.
The analysis the design supports, run correctly, checked independently, and written up so a referee can follow every number back to a line in a script.
Our promise
We will tell you which test your design supports, run it correctly, and write the methods so a referee can follow them. What we will not do is move a threshold, drop a case that spoils a distribution, or report the one specification of twelve that worked.
Reported as computed, including the null ones.
All disclosed, not only the best.
Never altered, always returned.
Yours. We go in the acknowledgements.
Disciplines
Subject specialists across the fields our clients publish in, including interdisciplinary work that does not sit neatly inside one of them.
How we work
For the individual researcher with a study to finish, and for the institution that needs research capacity it does not have on staff. Both run the same nine stages, with every check placed before the point where fixing it becomes expensive.
For researchers and scholars
Choosing a test after seeing the data is how p-hacking happens by accident. In our sequence the analysis plan is Stage 3 and collection is Stage 5. Fixing the plan first is what makes the eventual result defensible — and it is free to do.
For institutions and departments
A statistics and data unit your faculty can book, without the cost of hiring one — plus the training that reduces demand on it over time, because a department that can size its own samples stops producing unpublishable studies.
Insights
Practical, source-checked writing on research data and statistics — the things nobody explains until a reviewer has already asked.
Why a post-hoc power analysis convinces nobody, what an a priori calculation needs as input, and how to report it in three sentences.
Read the noteWhen the assumption actually matters, when the central limit theorem covers you, and the three alternatives worth considering before you transform anything.
Read the noteHow to state a missingness mechanism, when multiple imputation is worth the complexity, and what to write in the methods either way.
Read the noteSend the proposal, the dataset description or the reviewer's comment. We come back with what the design supports, which analysis fits, what it would cost — or an honest note that you do not need us.
About NextDataLab
NextPUBGlobal spends its days on whether a manuscript will survive an editor. Working on that, one pattern kept repeating: the papers that failed were rarely badly written. They were papers where the sample could not support the claim, the analysis did not match the design, or nobody could reconstruct how a number was produced. NextDataLab exists to fix that earlier, where it is still cheap.
Why we exist
Language editing is a real service and it solves a real problem. But it operates at the end of a process, on a document, after every consequential decision has already been made. A copy editor can fix a sentence about a sample. Nobody can fix the sample.
Sitting inside a publication support business, that limit becomes obvious quickly. You watch a well-written manuscript come back from a journal with three statistical comments, and you watch the author discover that answering them properly means going back eighteen months. The paper was not rejected because of how it read. It was rejected because of a decision taken before the first participant was recruited.
NextDataLab was set up to move upstream of that moment — into design, sampling, measurement, collection and analysis — and to be accountable for the decisions taken there. The unit of work is not a document. It is a study.
A null finding you can defend is worth more than a significant one you cannot.
Who we are for
Whether we are the right people, whether this is the right moment, and what physically arrives at the end.
How we operate
These are not aspirations. Each one has cost us work, and each one is the reason a supervisor or an ethics committee can accept our involvement without a conversation.
Where you engage us early enough, the analysis plan and the sample size are fixed and written down before collection begins. Where you do not, we say plainly what the existing data can and cannot support.
Nothing is done by hand in a spreadsheet. Cleaning, recoding, exclusion and analysis all live in a script with a changelog, so any result can be traced back to the raw file it came from.
The person who runs the model is not the person who reproduces it, and neither of them signs off the methods text. Three separate checks, deliberately not performed by the same pair of hands.
We write methods and results — the parts that describe what was done and what was found. Introduction, discussion and conclusion are yours, because they are the parts that constitute authorship.
The first scoping call is free and ends with a recommendation, not a quote. Where a supervisor or a departmental statistician can answer the question, that is what you will hear.
Raw files are never altered. They are returned, and deleted from our systems at the end of the agreed retention period, with the deletion documented.
The order of work
Compressed here to five moments; the full sequence runs to nine stages, each with its own deliverable.
A proposal, a dataset description, or three sentences from a referee. We read it and reply with what the design can carry.
Variables, hypotheses, sampling frame, power calculation, instrument, ethics pack. Everything that is expensive to change later is decided here.
Fieldwork we run or supervise, or existing datasets located and merged. Cleaning happens in a script, with the missingness mechanism stated.
By us, then independently from the raw file by an analyst who was not on the project. If a number does not reproduce, nothing ships.
Methods and results text, journal-ready tables, the reproducibility pack, the repository deposit and the data availability statement.
NextDataLab is the research data vertical of NextPUBGlobal, a Careerantra Solutions company. The publishing side works on manuscript development, journal selection and editorial compliance; this side works on everything that has to be true before a manuscript is worth writing.
Clients use one, the other, or both. Where both are involved, the data team hands over a methods section the editorial team does not have to invent, and the editorial team flags reviewer comments the data team answers with a reanalysis.
The two are deliberately staffed and checked separately. An editor does not sign off statistics, and an analyst does not sign off prose.
Unit 603–604, Floor 6, Tower B, Bhutani Alphathum, Sector 90, Noida 201305, India. We work with researchers and institutions internationally, in English, across time zones, and bill in USD, EUR, GBP or INR.
A scoping call costs nothing and ends with a written note on what the design supports — including, sometimes, that you do not need us.
What we do
Most researchers come to us in the middle — with data already collected, or an analysis a reviewer has questioned. We take work at any point on this list. It costs less, and the result is stronger, when it starts at the top.
The eleven entries below are sequential, not a menu of alternatives. Each one assumes the decisions taken in the one above it, which is why entering at Stage 8 with an undocumented dataset is more work than entering at Stage 2 with nothing but a question.
Each entry says what the work involves and what physically arrives at the end of it. Where you already have that deliverable — a validated instrument, an approved ethics pack, a clean dataset — we start after it and price accordingly.
Turning a question into a testable design: variables, hypotheses, unit of analysis, comparison groups, and the reporting guideline the study will be written against. The output is a protocol you can hand to a committee, not a conversation you have to remember.
Frame definition, sampling strategy, inclusion and exclusion rules, and an a priori power calculation that tells you the sample size the claim needs — before recruitment, not after a reviewer asks. Delivered with the justification paragraph, written for your methods section.
Questionnaires, scales and interview guides built from the construct up, then piloted and tested — content validity, reliability, and factor structure where the instrument is new or adapted. Adapting a published scale without revalidating it is one of the more common reasons a psychometric paper is returned.
Ethics committee and IRB submissions, participant information sheets, consent forms, data management plans and the anonymisation approach — prepared in the format your committee expects, and consistent with the protocol rather than written separately from it.
Fieldwork we run or supervise: online and offline surveys, clinical and institutional records, structured observation, in-depth interviews and focus groups, with enumerator training and live field monitoring. Response quality is watched while collection is open, when it can still be corrected.
Locating, acquiring and merging existing data: national statistical series, health and demographic surveys, World Bank and OECD indicators, Scopus and Web of Science exports, patent, market and institutional repositories — with licences and terms of use checked before anything is merged.
Coding, recoding, missing-data treatment with a stated mechanism, outlier handling, transformation and assumption testing — all recorded in a script rather than done by hand in a spreadsheet, so that every exclusion has a reason attached to it.
From descriptives to structural equation models, survival analysis, multilevel modelling, econometric panels and machine learning — chosen for the design and the data, not for what is fashionable in the field. Effect sizes and intervals are reported alongside p-values as a matter of course.
Transcription, coding frameworks, thematic and content analysis, grounded theory, framework analysis and mixed-methods integration, with an audit trail and intercoder agreement where the design calls for it — and a reflexive account where the guideline expects one.
Systematic search strategies, screening and extraction, risk-of-bias assessment, meta-analysis and meta-regression; plus bibliometric and scientometric mapping for reviews and research-performance work. The search strategy is written so a stranger can re-run it and get your numbers.
The methods and results sections written to the target journal's requirements, the reproducibility pack assembled, the repository deposit prepared, and statistical reviewer comments answered with the reanalysis they ask for rather than a paragraph explaining why it is unnecessary.
Environments
So that the files you receive open on your machine, and your student or co-author can run them again next year without asking us for a licence.
Describe the study in a paragraph. We will tell you which stage you are actually at — which is often earlier than people expect — and what it would take from there.
Methods library
Filter by what you are trying to establish, the kind of variable you are working with and the software you have. Each entry states when the method applies and what it produces. Use it to check our thinking, or to arrive at the first call already knowing what to ask for.
The most expensive mistake in applied statistics is picking a test because it is familiar, then discovering that the design cannot support it. The second most expensive is picking one after looking at the data, which turns an honest study into an accidental fishing expedition.
So the library is organised the way the decision is actually made: start from what you are trying to establish, narrow by the kind of variable you have, and only then look at what your software can run. If two methods survive that filter, the choice between them is a conversation — and usually a short one.
Nothing here is a substitute for a methodologist reading your protocol. It is a way to arrive at that conversation already knowing the vocabulary, and to check that what we propose is what your design implies.
The families, briefly
Ten families cover almost every design we see. If yours does not fit one of them, that is worth a conversation rather than a filter.
Is the outcome different between these groups, or before and after? t-tests, ANOVA and their rank-based equivalents, chosen by how many groups there are and whether the same people appear twice.
Do these two things move together? Correlations, chi-square, agreement statistics — descriptive claims that do not, on their own, license a causal sentence.
What predicts the outcome, holding other things constant? Linear, logistic, ordinal, count and multilevel models, with the choice driven by the shape of the outcome variable.
Would the outcome have differed had the exposure differed? Matching, differences-in-differences, instrumental variables, fixed effects and discontinuity designs — for observational data with a defensible identification strategy.
How long until something happens, and what changes it? Kaplan-Meier, Cox models and time-series forecasting, with censoring handled properly rather than dropped.
How do we measure something we cannot observe directly? Factor analysis, structural equation modelling, PLS-SEM, mediation and moderation — the backbone of most management and psychology submissions.
Too many correlated variables to look at one at a time. Component analysis, clustering and discriminant analysis, usually as a step towards a model rather than a finding in itself.
What do people actually say, and what does it mean? Thematic, framework, content and grounded theory analysis, with coding frames, audit trails and intercoder agreement where the guideline expects it.
What does the literature, taken together, support? Meta-analysis, meta-regression and risk-of-bias appraisal; plus co-citation and performance mapping for reviews and research-office reporting.
Environments we work in
We work in your environment where you have one, so that the files you receive open on your machine and your student or co-author can run them again next year. Where you have no preference, we default to R for quantitative work and NVivo for qualitative, and supply the output in a form your co-authors can read without either.
Send the design and the variable list. You will get a short written answer naming the method, the assumption to check first, and what the reporting guideline expects you to state.
The method
The sequence below is the whole of what we do. Nothing in it is unusual on its own — what matters is where each step sits, and which of them you have to sign before the next one starts.
The analysis plan is Stage 3 and data collection is Stage 5. That single ordering is the difference between a study a statistical reviewer accepts and one they send back — because it means the tests were chosen while they were still a design decision, and not after somebody had seen which comparison looked promising.
It also removes the temptation that produces the problem, which is rarely dishonesty. It is deadline pressure meeting an ambiguous dataset at two in the morning.
Choosing a test after seeing the data is how p-hacking happens by accident.
Every stage ends with something you can hold: a document, a script, a dataset or a certificate. A stage that produces only a conversation has not finished.
Same activities, different order
Pre-specification is free at the start and impossible to retrofit at revision. This is the entire reversal.
The sample size is justified after the fact, which a reviewer can see.
The plan is dated and signed, so the method is a record rather than a claim.
The sequence
Three phases, nine stages. Scroll and the spine draws to whichever stage you are reading; the gates clear as their conditions are met.
We read the proposal, the dataset description or the referee's comments verbatim, and tell you what the design can and cannot support. This stage is free, and it ends in writing whether or not you engage us — including, sometimes, with a recommendation to proceed without us.
Variables, hypotheses, unit of analysis and comparison structure are fixed, and the reporting guideline the study will be written against is chosen now rather than at submission. The unit of analysis is written down even when it seems obvious — especially then, since most multilevel errors are a unit-of-analysis error nobody recorded.
The primary analysis is named and justified against the design, the secondary and subgroup analyses are declared in advance, and the handling of missing data and outliers is decided by rule rather than by inspection. The sample size is calculated for the smallest effect worth finding.
Where the data already exists, we report the precision your sample supports instead of a post-hoc power calculation — which is mathematically redundant with the p-value and signals to a reviewer that the plan came after the result.
Questionnaires, scales and interview guides built from the construct up or adapted, then piloted and tested for reliability and validity. A published scale translated or moved to a new population is a new instrument, and we revalidate it rather than cite the original.
Ethics and consent documents are prepared in your committee's format, and we check that the consent obtained actually covers the analysis and the deposit you intend — the clause nobody reads until a journal asks for the data.
Fieldwork we run or supervise, with enumerator training and response quality monitored while collection is open rather than audited after it closes — when nothing can be corrected. Or existing datasets located, licensed and merged, with provenance and access date recorded for every source.
Your raw file is copied once, set read-only and never opened again. Everything downstream reads from that copy.
Coding, recoding, missing-data treatment with a stated mechanism, outlier decisions taken by the rule set in the plan, and assumption testing — all of it in a script rather than by hand in a spreadsheet. Every exclusion carries a reason, so when a reviewer asks why n fell from 412 to 388 there is a line to point at.
The pre-specified analysis is run, with the assumptions your model actually requires tested and reported — not a standard battery that buries the check that mattered. Where an assumption fails we choose openly between transforming, switching to a robust alternative and respecifying, and record which and why.
Effect sizes and confidence intervals accompany every p-value, including for null results, and every specification fitted goes in the analysis log — not only the one that worked.
A second analyst who was not on the project receives your raw file and the scripts — never the output tables, and never a summary of what to expect — and reproduces every reported number in a clean session. Anything that does not match is logged and resolved by fixing the script, never by adjusting the report to agree with it.
Reproduction runs from the raw file, not from the cleaned dataset. Starting from the cleaned data would test the analysis script alone and leave every cleaning decision unchecked.
Methods and results written to the target journal's requirements in journal-ready past tense, with software and package versions named. Every number quoted in the text is cross-checked against the table it came from. The reproducibility pack is assembled, the repository deposit prepared with access conditions matching the consent, and the data availability statement drafted.
Delivered on a call rather than as a zip file, because most of the value transfers in that half hour — and because you cannot defend what nobody explained.
Stage 8, in detail
Ask anyone quoting for your analysis whether a second person re-runs it from your raw file before release. The answer separates a service from a supplier.
Illustrative. In practice 98% of analyses reproduce exactly on the first re-run; the remainder are returned to the client as a data query before anything is released. It is not an acceptance rate, and we do not present it as one.
The deliverable
Not a slide summary of your results. A pack your co-author can open in five years and run again.
Plus the acknowledgement wording, drafted for you. We are acknowledged; we do not take authorship, because statistical assistance does not meet the ICMJE criteria on its own.
Where people actually arrive
We take work at any point in the sequence. It costs less, and the result is stronger, the earlier it starts — but the honest position is that we are usually met in the middle.
A plan written after the data arrived is dated and labelled as such. It is the single easiest thing for a reviewer to catch, and the hardest to recover from.
Getting started
Within one business day, in writing, from the person who would actually run the work — not a sales team summarising it back to you.
We send the NDA before you send anything. Only after it is signed do you send the protocol, the dataset description or the referee's comments. Never identifiable participant data by email.
Thirty to forty-five minutes on what the design can and cannot support, ending in a written note whether or not you engage us.
If a supervisor or a departmental statistician can answer the question, that is what the note will say. It happens often enough that we mention it here.
The scoping call is free, ends in writing, and occasionally ends with us telling you to save your money.
Who we work with
For the individual researcher with a study to finish, and for the institution that needs research capacity it does not have on staff. Both run the same nine stages, with every check placed before the point where fixing it becomes expensive.
The problem
Language and structure are fixable in a week. A sample that cannot carry the claim, a test that assumes something your data violates, or an analysis nobody can reproduce are not fixable at revision — they send you back to collection, a year later, with the funding spent.
What we do about it
You get a named analyst who reads the design before the data exists, an analysis plan you sign, and a second analyst who reproduces every number before it reaches you. When a referee raises a statistical comment, answering it is a week's work rather than a year's.
Open a stage to see what happens in it and what you receive at the end.
| Model | Suits | Indicative |
|---|---|---|
| Methodology consultation One session, written note after | Deciding design, sample size or test | from USD 120 |
| Single analysis Fixed scope, fixed fee | One dataset, one paper | from USD 450 |
| End-to-end study Design through deposit | Thesis chapters and funded projects | quoted |
| Reviewer response Reanalysis and letter | Statistical comments from a journal | from USD 300 |
Indicative only, in US dollars; billed in USD, EUR, GBP or INR. Priced on design complexity, dataset size and turnaround. The first scoping call is free and we will say if you do not need us.
The situation
Accreditation frameworks, rankings and funding decisions all read publication and citation performance. Most departments respond by asking individual faculty to publish more, without giving them the methodological support that makes publishable work possible.
What we provide
A statistics and data unit your faculty can book, without the cost of hiring one. Plus the training that reduces demand on it over time — because a department that can size its own samples stops producing unpublishable studies.
Configured to what the department already has, rather than sold as a fixed package.
Turnaround
On complete inputs, not on the enquiry. Where a client is late supplying data or a decision, the delivery date moves by the same number of days — and we confirm that in writing on the day it happens, not at the end.
| Engagement | Target | Clock starts |
|---|---|---|
| First reply to an enquiry | 1 business day | On receipt |
| Scoping note after the call | 2 business days | End of the call |
| Methodology consultation | Within 1 week | Signed scope |
| Single analysis | 2–3 weeks | Complete data and signed analysis plan |
| Reviewer response | 5–10 business days | Receipt of the referee report |
| End-to-end study | Quoted with the scope | Signed scope |
Often the answer is the smallest one — a single consultation that changes the design before anything is collected.
Our promise
That line governs every project we take. We will tell you which test your design supports, run it correctly, and write the methods so a referee can follow them. What we will not do is move a threshold, drop a case that spoils a distribution, run the model until something reaches significance, or report the one specification of twelve that worked.
The position, in full
If the analysis does not support the hypothesis, you get that result, with an honest reading of what it means and what a reviewer will make of it. A null finding you can defend is worth more than a significant one you cannot — and considerably more than one that gets retracted three years later.
This matters commercially, not only ethically. The market we sit in contains people who will produce a significant result on request. What they are selling is a liability with a p-value attached: the paper is accepted, then a reader cannot reproduce it, and the correction notice carries your name rather than theirs.
It is also why our name belongs in the acknowledgements rather than the author list. Statistical assistance and data curation do not meet the ICMJE authorship criteria on their own, and we will draft that disclosure for you — along with the data availability statement your journal now asks for.
Where an integrity question arises mid-project — an inconsistency in the raw file, a consent record that does not cover the analysis, a co-author asking for a case to be removed — we follow COPE guidance, tell you in writing, and stop work until it is resolved.
A significant result we cannot defend is not a deliverable. It is a future correction notice with your name on it.
Every delivery is re-run from the raw file by a second analyst who was not on the project. If a number does not reproduce, it does not ship. The remainder are cases returned to the client for a data query before release.
Separation of duties
An analysis is only as trustworthy as the weakest check it survived. Ours are separated on purpose: the analyst who runs the model is not the analyst who reproduces it, and neither of them signs off the methods text.
Runs the pre-specified analysis, tests and reports the assumptions, records every specification tried in the analysis log, and produces the script that generates each table and figure.
A second analyst who was not on the project takes the raw file and the script and re-runs everything independently. Any number that does not match is an exception, logged and resolved before release.
A third person, neither of the above, reads the methods and results text against the reporting guideline for the design — CONSORT, PRISMA, STROBE, COREQ or whichever applies — and signs it off.
Quality, ethics and confidentiality
Written out because a supervisor, an ethics committee or a research office will sometimes ask to see them, and because a promise nobody can check is not worth making.
These come up often enough to be worth stating in advance. None of them is a judgement about the person asking; most arrive from someone under a deadline who has been told this is normal.
| The request | What we do instead |
|---|---|
| “Can you make the result significant?” | Report what the data shows, with the effect size and interval, and help you write a discussion that survives the null. |
| “Can you generate the data? The deadline moved.” | Decline, without exception. Where a deadline is genuinely immovable, we help you scope a study that fits it honestly. |
| “Just drop the cases that break normality.” | Test whether normality matters for your design at all, then choose a method that tolerates the data you actually have. |
| “Write the whole paper and put my name on it.” | Write methods and results. Introduction, discussion and conclusion stay with you, because they are what authorship means. |
| “Add me as an author on your analysis.” | Nothing to add — we take no authorship. We draft the acknowledgement instead. |
| “Can you guarantee acceptance?” | No one can. We can make the methods section the part of your paper the reviewer does not argue with. |
The NDA is signed first. Then the study, the dataset description, or the three sentences from a referee that started all this.
Our team
Statisticians, methodologists, qualitative researchers and data engineers, supported by an advisory board drawn from medicine, management, economics and education. For this audience, knowing there is a named biostatistician is most of the buying decision — so the names are here rather than behind a form.
Every project is assigned a named analyst at the scoping call, and that person stays with it through delivery and through whatever the reviewers send back. They are the one who answers your email, and the one whose reading of your design you can argue with.
Behind them sit two people you will hear from less: the analyst who independently reproduces the work from your raw file, and the methods reviewer who checks the write-up against the reporting guideline. Both are deliberately not the person who ran the analysis.
Where a design is contested — an unusual identification strategy, a psychometric instrument being adapted across languages, a trial with a protocol amendment — it goes to the advisory board member for that field before we commit to it.
Nobody outside that list sees your data. Access is scoped to the named project team and logged.
Operating team
Each lead owns a stretch of the nine stages end to end, which is why a handover between them is a scheduled event rather than an email.
Owns study design, sampling and analysis plans across all disciplines. Runs the scoping call on most projects, and is the person who will tell you the design cannot carry the claim.
Owns clinical and health-sciences analysis and reviewer response: trials, epidemiology, survival models, and the statistical comments that come back from medical journals.
Owns coding frameworks, mixed-methods integration and audit trails, and the intercoder agreement reporting that COREQ and SRQR reviewers look for.
Owns sourcing, cleaning pipelines, version control and reproducibility — including the packs that go to repositories and the scripts your co-authors will run in five years.
Consulted on contested designs, and on the disciplinary conventions that decide whether a defensible analysis is also a publishable one.
Advises on trial design, ethics review and health data governance, including protocol amendments and registry requirements.
Advises on causal identification, panel data and impact evaluation, and on when an identification strategy will not survive a referee.
Advises on scale development, cross-cultural validation and structural models, including measurement invariance across groups.
Advises on data management policy, FAIR compliance and integrity, and on what accreditation reviews expect to see documented.
Replace the placeholder names and add photographs before the site goes live. Keep the role lines and the paragraph under each — for this audience, knowing there is a named biostatistician is most of the buying decision.
Discipline desks
The methods are general; the conventions are not. A defensible mediation model in management is written differently from a defensible one in clinical psychology, and reviewers notice.
What researchers say
Quote to be inserted.
Quote to be inserted.
Quote to be inserted.
Carry across the names already used on NextPUBGlobal and paste each person's actual words, with permission to name them and their institution. For this audience an unattributed quote reads as invented.
The scoping call is with the person who would run your study. If they are not the right fit for the design, they will say so and bring in whoever is.
Insights
Practical, source-checked writing on research data and statistics — the things nobody explains until a reviewer has already asked. Each note answers one question, states what the reporting guideline expects, and shows the two or three sentences you would actually put in the paper.
Almost every note here exists because the same question arrived three times in a month, usually from someone who had already collected their data. Power calculations after the fact. Normality tests treated as a gate rather than a diagnostic. Listwise deletion applied silently because the software default did it.
None of these are exotic errors. They are what a competent researcher does when nobody told them the convention has moved, and the cost of finding out from a referee instead of from a colleague is measured in months.
All notes
Why a post-hoc power analysis convinces nobody, what an a priori calculation needs as input, and how to report it in three sentences.
Read the noteWhen the assumption actually matters, when the central limit theorem covers you, and the three alternatives worth considering before you transform anything.
Read the noteHow to state a missingness mechanism, when multiple imputation is worth the complexity, and what to write in the methods either way.
Read the noteWhat replaced the causal steps approach, why bootstrapped indirect effects are now expected, and how to report a mediation model in 2026.
Read the noteWhat PRISMA 2020 actually requires in the supplementary file, and the five omissions that get a review desk-rejected.
Read the noteEthical and legal restrictions are an acceptable answer. Silence is not. How to write the statement that satisfies the journal and your ethics committee.
Read the noteTitles, reading times and summaries are placeholders written to show the section working. Replace with your published articles and point each card at its own page — the card markup already carries a link, a kicker and a metadata line.
What we write about
Roughly in the order they cause trouble.
Sample size justification, sampling frames, inclusion rules, and the difference between a study that is underpowered and one that is simply small.
Which assumptions matter for which method, how to test them without turning the paper into a diagnostics report, and what to do when one fails.
Adapting published instruments, reporting reliability properly, factor structure, and measurement invariance when you compare across groups or languages.
What CONSORT, PRISMA, STROBE and COREQ actually require, where the checklists disagree with journal templates, and what belongs in the supplement.
Codebooks, repository deposits, FAIR principles, data availability statements, and what to do when the data genuinely cannot be shared.
No digest, no round-up, no newsletter about our newsletter. If a month passes without a question worth answering at length, nothing arrives.
Nothing has been sent — this form is not yet connected to a list. Wire it to your provider and this becomes the subscriber's confirmation.
Send it. If it comes up more than once, it becomes the next note — and you get the answer either way.
Start with a free methodology review
Share your proposal, your dataset description or the reviewer's comment. We come back with what the design supports, which analysis fits, what it would cost and how long it takes — or an honest note that you do not need us. We sign an NDA before you send anything.
What happens after you send this
Within one business day, in writing, from the person who would actually run the work — not a sales team summarising it back to you.
We send the NDA first. Only after it is signed do you send the protocol, the dataset description or the referee's comments. Never identifiable participant data by email.
Thirty to forty-five minutes on what the design can and cannot support. It ends with a written scoping note, whether or not you engage us.
If a supervisor or a departmental statistician can answer the question, that is what the note will say. It happens often enough that we mention it here.
For scholars, faculty and clinicians.
Nothing has been sent — this page is not yet connected to a mailbox. Wire the form to your backend or form service and this becomes the researcher's confirmation.
For universities, departments and research centres.
Nothing has been sent — connect this form to your backend and this becomes the institution's confirmation, with its reference number.
research@nextdatalab.com
Design, analysis and reviewer response.
institutions@nextdatalab.com
Capacity, training and governance.
+91 98186 45221
Weekdays, 10:00–19:00 IST.
Before you write
No — and the earlier you come, the less it costs. A proposal, a research question and a rough idea of the population is enough for a scoping call. If anything, arriving with data already collected narrows what we can do for you.
Statistical and data support is a normal, disclosable part of research, and most institutions expect it to appear in the acknowledgements. We draft that disclosure for you. What is not allowed — and what we do not do — is ghost-writing, fabricated data or authorship in exchange for analysis. Tell your supervisor we exist; the response is usually relief.
The research question, the design, the approximate sample size, the software you use if you have a preference, and the deadline. If a referee has already commented, send the comments verbatim — they tell us more in three sentences than a summary does in a page.
NDA first, always. Identifiers are stripped or pseudonymised on receipt, transfer is encrypted, and we do not accept participant data as email attachments. Access is limited to the named project team, and everything is deleted at the end of the agreed retention period with the deletion documented. We confirm ethics approval before touching human-subject data.
No, and anyone who does is selling something else. What we can do is make the methods section the part of your paper a reviewer does not argue with, and make sure that if a statistical comment does come back, answering it is a week's work rather than a year's.
Usually the Dean of Research, the research cell, or whoever owns the accreditation file — but a single department can engage us on its own. Use the institutional form; we reply with an outline scope and a call invitation rather than a brochure.
Sign in to track your projects, datasets and deliveries.
New to NextDataLab? Create an account
Front-end demo. This form validates input and shows the signed-in state only. Connect it to your authentication backend — session handling, password hashing, rate limiting and role checks — before deploying.