You surveyed 214 nurses across three hospitals about hand-hygiene practice. The responses are in a spreadsheet and the paper looks like it writes itself: 68 per cent good knowledge, 41 per cent good practice, done.
The four sentences that will sink it are already in the draft, and none is about statistics. There is no sampling frame, so nobody can say who the 214 are 214 out of. The response rate has a numerator and no honest denominator. The questionnaire is called validated because a previous paper used something similar. And the conclusion says nurses do something, when the study measured what they said.
All four are fixable before submission and close to unfixable afterwards; the figures are invented.
Reported practice is not observed practice, and the whole paper has to obey that
A questionnaire measures what a respondent was willing to write down, at one moment, in the setting where they were asked. That is a real measurement, but not the same as watching them.
People report the behaviour they believe is expected of them, and a question about their own compliance invites exactly that. So 41 per cent does not mean 41 per cent comply. It means 41 per cent reported practice consistent with the protocol — which the title, the aim and the abstract must say too.
Name your sampling frame before you name your sample
The sampling frame is the set of people who could have been approached, and how it was assembled. Not the people who answered.
Write it with a number in it: "all 620 nurses on the payroll of three tertiary hospitals in [city] as at 1 March, enumerated from the nursing establishment list". Where no list exists, describe the sweep — "all nurses on the medical and surgical wards during 12 day shifts in March" — a weaker frame, and still a frame. Having neither is fatal: the percentages then describe nobody.
The response rate needs a denominator you can defend
A response rate is a fraction and needs two honest numbers. Name both: "214 of 620 nurses invited returned a questionnaire (34.5 per cent)".
Three designs where an honest denominator does not exist
- An open link — a URL can be forwarded, so the number who saw it is unknown.
- Snowball recruitment — the frame grew during the study and was never enumerated.
- Uncounted refusals — a handout has a denominator only if somebody counted how many people were approached, including those who declined to take one.
Where none exists, say so — "distributed through an open link, so the number who saw the invitation is unknown and a conventional response rate cannot be computed" — then report what CHERRIES asks for: how many reached the first page, how many agreed, and how many finished. No universal minimum rate makes a survey publishable.
"Validated" is a claim about a named study in a named population
An instrument is validated in a population, for a purpose, by a study you can cite, and your claim is no wider than that study. Name all three: "we used the [name] scale, psychometrically evaluated in [population] by [citation], unchanged". If you dropped items, adapted wording or collapsed seven response options into five, say what you changed. Using a scale because a previous paper did is precedent, not a psychometric property.
If you wrote the instrument yourself, say so, and report what you did do
Self-developed instruments are ordinary; no evaluated scale exists for many specific practices. The failure is not writing one, but calling it something it is not.
Report the development in order: how the items were generated; expert content review; the pilot, and whether its responses stayed in; and internal consistency in this sample.
That is worth reporting. It is not validation: content review plus a pilot plus a coefficient shows that experts judged the items relevant, that they read clearly, and that they correlated with one another in one sample — not that the scores measure the thing you name.
Translation and back-translation: what to report
If the instrument was administered in another language, report who translated it, whether there was an independent back-translation, how discrepancies were reconciled, and whether the translated version was piloted.
Report the instrument in full, as a supplementary file
"A 32-item questionnaire covering knowledge, attitude and practice" is a label, not a description. Supply the instrument as administered: every item, its response options, the scoring rules, and the cut-offs used to label a score "good", in both language versions if there were two. The dataset behind the tables follows the same reasoning — see citing data, software and code.
STROBE for the cross-sectional frame, CHERRIES for anything distributed online
Our guide to CONSORT, PRISMA and STROBE covers which checklist fits which design. From STROBE, cross-sectional version: item 6 (eligibility and selection), item 9 (bias), item 13 (numbers at each stage, including non-response) and item 16 (estimates with their precision — see p-values and confidence intervals).
From CHERRIES: open versus closed, adaptive questioning, whether items were mandatory, duplicate prevention (cookies, IP check, log file analysis, registration), and separate view, participation and completion rates. Where sex or gender is analysed, the SAGER guidelines apply.
Write the conclusion the design supports
Three sentence patterns will carry an honest survey conclusion.
In this sample of 214 nurses from three tertiary hospitals, 41 per cent reported practice consistent with the hand-hygiene protocol.
Higher knowledge scores were associated with higher reported practice scores in this sample; the design does not establish the direction of that association.
Reported practice was lowest for [item], which identifies a target for observational audit rather than a measured gap.
Each names the sample and stops where the design stops. The verbs to delete: comply, adhere, underuse, improves, increases, causes, demonstrates, proves.
The pre-submission grid: frame, denominator, provenance, verb
Work down the twelve rows before you submit. Column three is what a reader may reasonably conclude when a row is missing.
| What to state | Where it goes | What a reader concludes if it is missing |
|---|---|---|
| The sampling frame: who could have been approached, and how that list was enumerated | Methods | The 68 per cent may describe only the nurses who felt like answering |
| Recruitment route and setting, with data-collection dates | Methods | An audit or training week may have fallen inside the study period |
| Inclusion and exclusion criteria for respondents | Methods | Students or agency staff may or may not be in the denominator |
| Distribution route (in person, printed, email, messaging app, QR code) and whether the link was open or closed | Methods | A denominator is assumed where an open link has none |
| The response rate as a fraction with both numbers named, or why it cannot be computed | Methods and Results | The rate was low and left out for that reason |
| Instrument provenance: self-developed, or adapted or used unchanged with a citation naming the validation population | Methods | The instrument may never have been used in a population like this |
| Pre-use work: expert content review (how many, which disciplines) and pilot (how many, whether data were included) | Methods | Nobody checked the items, or pilot responses were counted twice |
| Internal consistency for the scale as used, calculated in this sample, not quoted from the source | Results | A coefficient from another population says nothing about these items here |
| Translation and back-translation procedure, and how discrepancies were reconciled | Methods | Respondents may not have answered the question the authors think they asked |
| Handling of incomplete responses, and what counted as complete | Methods | Denominators shift between tables and cannot be reconstructed |
| How anonymity was maintained, and whether a response could be linked to an individual | Methods | The pressure on a respondent reporting their own compliance is unknown |
| The ethics document: committee, reference number and date, or the basis on which review was not required | Ethics statement | Nobody independent saw the study before it ran |
Then run the verbs.
| Claim as drafted | What the design supports |
|---|---|
| Nurses comply with the protocol in 41 per cent of cases | 41 per cent of respondents reported practice consistent with the protocol |
| Doctors underuse the guideline | Respondents indicated that they consult the guideline less often than [comparator] |
| Knowledge improves practice | Higher knowledge scores were associated with higher reported practice scores in this sample |
| The training programme increased awareness | Respondents reporting training reported higher awareness; the design cannot attribute that to the training |
| Hand hygiene is poor in this region | Reported practice was low in this sample, drawn from three hospitals in one city |
Finally, complete the STROBE cross-sectional checklist, and CHERRIES for an internet-distributed survey, with a page number against every item. Keep the completed copy with the manuscript, for yourself and for whichever journal asks for it: a citation is a claim, a completed checklist is evidence.
Our Author Guidelines cover manuscript preparation, and Publication Ethics covers approval and consent for research with human respondents, against the Declaration of Helsinki. See also statistical reporting and writing clear scientific English.