River
Y CombinatorBacked by Y Combinator
FREE TEMPLATE

Survey Questionnaire Design Template

The construct-to-item map that cuts questions nobody will analyse, the response formats you cannot fix after fielding, and the base size of every routed block.

Free download  ·  No account needed

Construct to Item Map

[Study name], [wave or round]

Expected completes [frame size times the response rate you expect]
Draft item count [before this sheet]
Kept item count [after this sheet]
Read the sheet right to left tables first, then analyses, then items

Construct              Items  Planned analysis          Table    Verdict
[primary outcome]      [n]    [the specific analysis]   [Tbl n]  keep
[primary exposure]     [n]    [the specific analysis]   [Tbl n]  keep
[covariate]            [n]    [the specific analysis]   [Tbl n]  keep
[construct]            [n]    none                      none     cut
[construct]            [n]    none                      none     cut
                       ----                                       
KEPT SUBTOTAL          [n]
CUT SUBTOTAL           [n]
DRAFT TOTAL            [n]    kept plus cut must close

The last two columns are the whole sheet. "Descriptive statistics" is not an analysis and "the results section" is not a table. An item that cannot fill both does not enter the instrument.

Cut list, grouped by reason

Duplicates a construct already measured / no research question behind the construct / construct named with no planned analysis / demographic with neither an analysis nor a benchmark to weight against.

The pack is seven files, one standing rule and eight prompts. Three sheets carry the arithmetic: the construct-to-item map, the response option register and the routing logic. Four documents carry the instrument itself, the pilot protocol, the ethics note, and a page on which four decisions stop being reversible the moment you field. The standing rule is that an item earns its place by naming the table it appears in.

Every sheet ships filled in for one illustrative workforce survey. A 68-item draft comes out at 41. Eleven items duplicated a construct already measured, eight named a construct with no planned analysis, five were demographics with nothing to weight against, and three had no research question behind them. Median completion time falls from 21 minutes to 13. Five of ten response formats could not support their planned analysis, and four of those five are unfixable once the instrument is in the field.

Federal survey standards require the written plan to set out the tabulations and analytic results that will inform decisions before the instrument exists. A template with a Question column and a Response Type column reverses that order. Open the pack in River and the map gets built from your tables. Or download the documents and sheets and fill them in yourself, then hand the fielded data to the cleaning record and the methods section. Whether the instrument measures the construct it claims to is a separate pilot-stage check; more research packs sit alongside both.

Every sheet in the pack

The Construct to Item Map, the Response Option Register, and the Routing Logic.

Construct to Item Map

Read right to left. The table decides the analysis, the analysis decides the item, and an item with no table has no purpose to be necessary for.

ConstructResearch question it answersItemsPlanned analysisTable
Wellbeing (primary outcome)Did wellbeing change between wave 3 and wave 4?8Mean scale score; paired change against wave 3Table 3, Figure 2
Unpaid overtime hours (exposure)Is unpaid overtime associated with lower wellbeing?4Continuous exposure in the main modelTable 4 cols 1–3
Job demandDoes demand account for the association?5Covariate, mean of five itemsTable 4
Supervisor supportDoes support moderate it?4Covariate and interaction termTable 4, Table 5
Contract and working hoursWhich contract groups carry the exposure?6Stratifier; drives the overtime routingTable 1, Table 2
Sickness absenceDoes absence move with wellbeing?3Secondary outcome, count modelTable 6
Carer responsibilitiesDo employees with caring duties differ?4Subgroup definitionTable 7
Tenure and roleStandard adjustment4CovariateTable 1, Table 4
Consent and administrationEligibility and record linkage3Defines exclusion rules E2 and E5Exclusion cascade
Kept subtotal41

The 27 items that could not fill the last two columns, grouped by the reason they failed:

Reason it was cutItems
Duplicates a construct already measured, with no reliability comparison planned11
Construct named, no planned analysis ("good to know", "a future paper")8
Demographic with neither a planned analysis nor a benchmark to weight against5
No research question behind the construct at all3
Cut subtotal27
Draft total: 41 kept plus 27 cut68

The column closes, so nothing was reclassified without being recorded. Median completion time goes from 21 minutes to 13, and the 8 minutes is what actually settles the argument. A principal investigator will defend an item and will not defend eight minutes.

Response Option Register

A response format is not a presentation choice. It sets the measurement level, and the measurement level decides which analyses exist.

ItemFormat as draftedPlanned analysisSupports itWhat it foreclosesFixable after fielding
AgeSix age bandsContinuous covariate; weighting against a benchmark published in single yearsNoA mean age, a continuous term, and any weight built on that benchmarkNo
Wellbeing items (8)Four-point forced choiceMean scale score compared against wave 3NoThe wave 3 comparison, which is why wave 4 is being fielded at allNo
Unpaid overtime hoursSelect one of five bandsContinuous exposure; hours compared against wave 3NoA mean, a per-hour coefficient, and the wave 3 comparisonNo
Benefits usedSelect all that applyPercentage using each benefitNoA stated denominator. The percentages sum past 100No
Contract typeFree textFour-level stratifier in Table 2NoA clean stratifier. Free text produces spellings, not levelsPartly
Overtime provisionYes / NoEligibility gate for the overtime blockYesNothingn/a
Dependants needing careYes / NoGate and subgroup definitionYesNothingn/a
TenureNumeric monthsCovariateYesNothingn/a
Sickness absence daysNumeric days in 12 monthsSecondary outcome, count modelYesNothingn/a
EmployerCoded list of 11 employersCluster identifier for the variance estimatorYesNothingn/a
5 of 10 formats cannot support their planned analysis4 of those 5 are unrepairable after fielding

Age is the clearest one. Bands cannot be unbanded, midpoint imputation is a different variable, and there is no re-contacting 3,400 respondents. Ask the year of birth and band it at analysis time; the decision stays open. In a survey platform the difference between these two is one dropdown selection, with no warning attached, made by whoever was building the form.

Routing Logic

Skip logic is arithmetic, and the arithmetic is fully available before anyone is invited. Every condition multiplies.

IDBlockShown whenRateApplies toExpected baseCellsExpected cellVerdict
R1OvertimeContract carries an overtime provision59.0%3,4002,0064501Field
R2Hours detailR1 and any overtime reported93.9%2,0061,8844471Field
R3ReasonsR2 and more than 10 hours33.4%1,8846294157Field
R4Formal complaintR3 and a complaint was raised4.1%6292646Cut
R5Carer blockReports dependants needing care28.0%3,4009522476Field
R6Night-shift carerR5 and works night shifts11.2%9521071110Restructure
R7Absence detailReported any absence63.1%3,4002,1454536Field
R8Return to workR7 and absence over 7 days21.4%2,1454594115Field
R9OH referralR8 and referred12.6%45958414Restructure
9 routed blocks3 cannot support the table they were written for: 1 cut, 2 restructured

R4 is the one worth staring at. Four conditions deep at 59.0, 93.9, 33.4 and 4.1 per cent, so 0.76 per cent of expected completes reach it: 26 people, before crossing by anything. Its planned table has four cells, giving an expected six. "The expected cell is six respondents" ends an argument that "we might not get many" does not, and both sentences were available on the same day.

Illustrative study, the same fictional workforce survey the cleaning record uses. Rates come from that wave's observed incidence, so 59.0 per cent is 2,043 of 3,465. Each nested base is the parent base times the rate, not the completes times the rate, which is the error that makes deep branches look survivable.

What's in the pack

01

Construct to Item Map sheet

One row per construct with the research question behind it, the item count, the planned analysis, and the table or figure the items appear in. Kept and cut subtotals that have to close.

02

Response Option Register sheet

Every item's format written out in full, the measurement level it produces, the analysis it has to support, a plain yes or no, and whether the mistake survives fielding.

03

Routing Logic sheet

Each routed block with its condition, the expected rate and where the rate came from, the expected base, the smallest cell of its planned table, and a verdict of field, restructure or cut.

04

Instrument

The questionnaire block by block with each item's rationale beside it rather than in a separate file, because a rationale kept elsewhere is one nobody reads during a revision.

05

Pilot Protocol

What is being tested, on whom, and what result changes the instrument, all written before the pilot runs. Scoped to what is new, because the standard accepts components successfully fielded on a previous occasion.

06

Ethics Note

What respondents are told, by construct rather than by item, what justifies each item collected, and what was cut because nothing justified collecting it.

07

How This Instrument Was Built

The four decisions that stop being reversible the moment you field, which artifact catches each one, and the order the artifacts have to run in.

08

An item earns its place by naming its output

The standing space rule every prompt reads first, including the three questions that cut items and the list of things to reject on sight.

How to use it

  1. 1

    Open in River, or download it

    Open the pack in River and the agent builds the map from your research questions, or download the blank Word documents and CSV sheets instantly.

  2. 2

    Send the questions and the frame

    Your research questions, the tables your report has to contain, any prior wave you need to stay comparable to, and how many people you expect to complete.

  3. 3

    Build the map before anything else

    Every construct gets its research question, its item count, its planned analysis and its table. Anything that cannot fill the last two columns collects in the cut list.

  4. 4

    Check the formats and the routing

    Then the response formats, checked against the analysis each one has to support, and the expected base of every routed block. Both are knowable before fielding.

Frequently asked questions

Is this template free?

Yes. Download the whole pack as Word documents and CSV sheets with no signup and no credit card. Edit with AI is a separate, optional path where the agent builds the map from your research questions and audits a draft instrument you already have. The template library holds the rest.

What format are the downloaded files?

Word documents for the four documents and the rule, and CSV for the three sheets, in one zip. They open natively in Word, Pages, Google Docs, Excel, Numbers and Sheets with nothing to convert. Inside River the same content opens as live Docs and Sheets.

I already have a draft questionnaire. Can it work from that?

That is the best input it can get. It places every existing item against a construct, a planned analysis and a named table, then hands you the leftovers grouped by reason with the completion time before and after. Most drafts lose between a quarter and a half of their items.

Why is it so insistent about asking age as a number?

Because bands cannot be unbanded and there is no re-contact. A banded age has no mean, cannot enter a model as a continuous term, and cannot be weighted against a benchmark published in single years. Ask the number, band it at analysis time, and the choice stays open.

Does it shorten published scales to save time?

No, and it will tell you when altering one is the problem. Dropping items from a published scale to save time produces a new instrument with no published reliability and no comparability. The pack records whether each block is verbatim, a validated short form, or altered with the alterations named.

How does it know a routed block will be too small?

It computes them. Expected completes times the rate at each gate, using your prior wave's observed incidence where you have one, then the smallest cell of the subgroup table that block was written for. In the worked example three of nine blocks cannot support their table.

Does the pack handle the data once responses come in?

No. It stops at the moment the instrument is fielded, which is where the irreversible decisions are. Once responses arrive, the export preparation tool handles the platform quirks and the exclusions, and the researcher workspace holds the whole project.

Decide it now, or never decide it

Send your research questions, the tables your report has to contain, and how many people you expect to complete. The map and the base sizes come back with the arithmetic done.

Edit with AI