surveyframe

Just here to verify a .sframe file's integrity hash or read its amendment log? Verify a file now. No R, no install. It runs entirely in your browser and nothing is uploaded.

surveyframe is a research-design-first survey package for R. Most survey tools collect answers and return counts. surveyframe begins at the research design and carries it through to a written results report.

The unit of work is the instrument, a typed sframe object that stores three things together:

1. The questions. Items, choice sets, scales, branching rules, and attention checks.
2. The analysis plan. A list of research questions, where each question is bound to a named statistical technique and to the variables that fill each role in that technique. The plan is written during design, before any data arrive.
3. The measurement or structural model. Constructs, indicators, and paths for EFA, CFA, CB-SEM, and PLS-SEM.

Because the plan and the model live inside the instrument, you never have to go back and match up questions, variables, and tests by hand. Say you're comparing satisfaction between first-time and repeat visitors: write that comparison into the plan once, alongside the questions, and it stays linked to the right variables from then on. When responses arrive, running the plan is a single step: each result comes back ready to write up, with a plot, a table, an APA statistic, an effect size where it applies, a writing prompt, and the reference that supports it. Plots switch between a colour palette for on-screen use and a black-and-white palette for print, both checked against WCAG contrast rules, so the same run produces a journal-ready figure with no separate step.

The package works offline during examples, tests, vignettes, and checks. Browser and Shiny entry points use open = FALSE or explicit launch functions, so automated checks do not open a browser.

Installation

Collecting data right now? Install 0.4.2 from GitHub. An external review of 0.4.1, the version on CRAN, found defects that silently alter or lose a participant's answer, and defects that give a wrong statistic. 0.4.2 corrects them and goes to CRAN on 26 September 2026. Until it is accepted there, this is the version to collect with:

remotes::install_github("MohammedAliSharafuddin/surveyframe")

NEWS.md lists every fix, and separately lists what changes for you: some scores and statistics move, and a Shiny-collected response gains a respondent_id column.

Install from CRAN:

install.packages("surveyframe")

To get changes not yet released to CRAN:

remotes::install_github("MohammedAliSharafuddin/surveyframe")

In an interactive session, library(surveyframe) prints a short start-up note with a demo to try, the commands to start with, and how to cite the package. suppressPackageStartupMessages(library(surveyframe)) loads it quietly, and scripts and rendered documents never show it.

Optional packages are only needed for selected features:

install.packages(c("shiny", "psych", "googlesheets4", "digest", "MASS", "nnet"))

Syntax generation works without installing lavaan or seminr. Install those packages when you want to fit the generated CFA, CB-SEM, or PLS-SEM models.

Already have data?

surveyframe is not a replacement for whatever collection tool your institution already has approved: Qualtrics, REDCap, Google Forms, or a paper form typed up afterward. It reads response data as a plain CSV or data.frame from any of them: export from your collection tool, rename columns to match your instrument's item IDs (or build the instrument to match the export), and load it. If you have collected responses in a CSV or Google Sheet and want to start from the analysis step, build a minimal instrument that matches your column names and load the data directly:

library(surveyframe)

# 1. Describe the items you already collected
cs  <- sf_choices("agree5", 1:5,
        c("Strongly disagree", "Disagree", "Neutral", "Agree", "Strongly agree"))
i1  <- sf_item("q1", "Item 1", type = "likert", choice_set = "agree5", scale_id = "S")
i2  <- sf_item("q2", "Item 2", type = "likert", choice_set = "agree5", scale_id = "S")
sc  <- sf_scale("S", "My scale", items = c("q1", "q2"))
instr <- sf_instrument("My study", components = list(cs, i1, i2, sc))

# 2. Load your CSV
responses <- read_responses("my_data.csv", instr, strict = FALSE)

# 3. Score and analyse
scored  <- score_scales(responses, instr)
results <- run_analysis_plan(scored, instr)

Documentation workflow

Choose the route that matches what you need:

• Try it in five minutes: Learn by example: 22 small surveys.
• Prefer a visual interface: SurveyBuilder, SurveyStudio, and the dashboard.
• Build and collect: Building a survey instrument, then Deploying and collecting.
• Already have responses: Analysing survey responses.
• Read the complete case study: Digital marketing and tourism services.

Continue as needed with these:

• Scale reliability and validity
• EFA, CFA, CB-SEM, and PLS-SEM syntax
• Multi-criteria decision analysis
• Small-sample inference
• Text and open-ended response analysis

Read all eleven vignettes inside R with:

browseVignettes("surveyframe")

An instrument is the research design

library(surveyframe)

agree5 <- sf_choices(
  "agree5",
  values = 1:5,
  labels = c("Strongly disagree", "Disagree", "Neutral", "Agree", "Strongly agree")
)

visitor_type_choices <- sf_choices(
  "visitor_type",
  values = c("first_time", "repeat"),
  labels = c("First-time visitor", "Repeat visitor")
)

sat_1 <- sf_item("sat_1", "The service was reliable.",
  type = "likert", choice_set = "agree5", scale_id = "sat")
sat_2 <- sf_item("sat_2", "The service was responsive.",
  type = "likert", choice_set = "agree5", scale_id = "sat")
sat_3 <- sf_item("sat_3", "I would recommend the service.",
  type = "likert", choice_set = "agree5", scale_id = "sat")

visitor_type <- sf_item("visitor_type", "Visitor type", type = "single_choice",
  choice_set = "visitor_type")
sat <- sf_scale("sat", "Satisfaction", items = c("sat_1", "sat_2", "sat_3"))

instr <- sf_instrument(
  "Service Survey",
  components = list(
    agree5, visitor_type_choices, sat_1, sat_2, sat_3, visitor_type, sat
  ),
  analysis_plan = list(
    list(
      id                = "RQ1",
      research_question = "Do first-time and repeat visitors differ in satisfaction?",
      family            = "group_comparison",
      method            = "mann_whitney",
      roles             = list(group = "visitor_type", outcome = "sat"),
      options           = list(alpha = 0.05)
    )
  )
)

write_sframe(instr, tempfile(fileext = ".sframe"))

# See the instrument as a survey a respondent would fill in:
export_static_survey(instr, open = FALSE)

write_sframe() validates the instrument and writes the validated object, including the validation flag, the analysis plan, and any saved model specifications. export_static_survey() renders it as a self-contained HTML survey, the same function covered in "Visual tools" below.

Import and score

responses <- data.frame(
  respondent_id = paste0("R", 1:5),
  sat_1 = c(4, 5, 3, 4, NA),
  sat_2 = c(5, 4, 3, 4, 5),
  sat_3 = c(4, 5, 2, 4, 4),
  visitor_type = c("first_time", "repeat", "first_time", "repeat", "first_time")
)

resp <- read_responses(responses, instr, respondent_id = "respondent_id", strict = FALSE)

score_scales(resp, instr)
missing_data_report(resp, instr)

Run the analysis plan

Each block binds a research question to a technique and to the variables that fill each role. run_analysis_plan() runs every block and returns one result per question. Earlier .sframe files using variables and test fields remain compatible.

results <- run_analysis_plan(resp, instr)
results

Supported method IDs include descriptives, missing data, quality checks, reliability, EFA readiness and solutions, CFA, CB-SEM, and PLS-SEM syntax, chi-square, Fisher's exact test, McNemar, Cochran's Q, t-tests, Mann-Whitney, Wilcoxon, one- and two-way ANOVA, ANCOVA, repeated-measures ANOVA, Kruskal-Wallis, Friedman, Pearson, Spearman, and Kendall correlations, partial correlations, linear and logistic regression, mediation, and moderation. Each technique reports an APA statistic, an effect size where it applies, a writing prompt, and the reference that supports it.

Render the results report

render_results(results, instr, output_file = tempfile(fileext = ".html"))

The report holds one section per research question, with the APA result, the writing prompt, a space for the interpretation, and a reference list compiled from the techniques used.

Reliability, EFA, and CFA

if (requireNamespace("psych", quietly = TRUE)) {
  reliability_report(resp, instr, omega = FALSE)
  efa_report(resp, instr)
}

cfa_syntax(instr)
cfa_lavaan_syntax(instr, ordered = TRUE)

Model layer

model <- sf_model(
  "model_1",
  "Satisfaction model",
  type = "cb_sem",
  constructs = list(
    sf_construct("SAT", "Satisfaction", c("sat_1", "sat_2", "sat_3"))
  )
)

instr <- add_model(instr, model)
model_json(model)
sem_lavaan_syntax(model, instr)

PLS-SEM syntax

pls_model <- sf_model(
  "pls_1",
  "Satisfaction and loyalty PLS model",
  type = "pls_sem",
  constructs = list(
    sf_construct("SAT", "Satisfaction", c("sat_1", "sat_2"), mode = "composite"),
    sf_construct("LOY", "Loyalty", "sat_3", mode = "single_item")
  ),
  paths = list(sf_path("SAT", "LOY")),
  options = list(bootstrap = 5000)
)

seminr_syntax(pls_model)

A full study report

render_report(
  instr,
  data = resp,
  output_file = tempfile(fileext = ".html"),
  include_codebook = TRUE,
  include_quality = TRUE,
  include_missing = TRUE,
  include_descriptives = TRUE,
  include_analysis = TRUE,
  include_models = TRUE
)

The built-in HTML fallback does not require Quarto. If the Quarto CLI is available locally, render_report() can use the bundled template.

Visual tools

launch_builder(open = FALSE)
export_static_survey(instr, open = FALSE)

Use launch_builder() to author the questionnaire, the plan, and the model and to export the .sframe file and model syntax. It runs no statistics. launch_studio() uploads responses, runs the plan on its Analysis Plan screen, and renders the report on its Export screen. launch_dashboard() is a read-only response explorer. Demo launchers are available for training:

launch_builder_demo(open = FALSE)
# launch_studio_demo()
# launch_dashboard_demo()

Interactive functions such as launch_builder(open = TRUE), launch_studio(), render_survey(), and launch_dashboard() are available for manual use. Tests and examples avoid opening browsers.

Roadmap

0.4.0 (CRAN, 2026-08-20) added three capability themes:

• Small-sample survey helpers, validated by a simulation study: Hodges-Lehmann and paired-Wilcoxon pseudomedian estimators, exact Fisher odds-ratio intervals, and Firth logistic regression.
• 10 multi-criteria decision methods (TOPSIS, AHP, ANP, DEMATEL, VIKOR, MOORA, SMART, WASPAS, PROMETHEE, ELECTRE), with 2 new question types for collecting judgements, weight-sensitivity analysis, and declared conjoint designs.
• Text and open-ended response analysis, 9 methods: term and n-gram frequency, keyword in context, co-occurrence networks, sentiment, document-feature matrices, and topic modelling via LDA or a structural topic model.

A disclosed-amendment and Git-linked provenance mechanism shipped alongside, on top of the existing .sframe integrity hash. Full detail in NEWS.md.

0.4.1 focuses on stability: fixes found by using surveyframe on real instruments, no new capability theme.

0.4.2 is a defect-fix release, correcting what an external review of 0.4.1 found across 8 batches: 171 findings, 154 fixed here and 17 deferred with a stated reason. The ones that matter most to a live study are in collection, where a participant's answer could be altered or lost with no signal, and in scoring, where the report path recomputed scale scores by a second method that ignored reverse coding and the declared minimum. Some numbers therefore move between 0.4.1 and 0.4.2, and NEWS.md says which, under "What you need to change". Reports also gained a folded "Show R code" block, and analysis_syntax() returns the statistical call behind a result with its variables and options resolved, so a reader can copy the code and reproduce the number.

When to use surveyframe, and when not to

surveyframe is a fit when a study's analysis has to be decided before data collection: a scale to validate, a pre-registered hypothesis test, a measurement model to fit, or an audit trail showing the plan wasn't changed after seeing results. Its core strength is a pre-declared, integrity-checked analysis plan bound to the instrument itself. It also interoperates with survey and srvyr, the standard tools for weighting and variance estimation on data from a complex probability sample.

Haven't decided the analysis yet? Collect first with any web-form tool, Google Forms, Qualtrics, REDCap, or the R package surveydown, and add the plan when you're ready: surveyframe reads exported CSV data from any of them (see "Already have data?" above).

If this package is ever archived by CRAN, the GitHub repository remains the canonical source: remotes::install_github("MohammedAliSharafuddin/surveyframe"). Each CRAN release is also deposited to Zenodo with its own DOI, so a specific version stays citable and retrievable independently of both CRAN's and GitHub's continued availability.

FAQ

Is surveyframe a replacement for Qualtrics, Google Forms, or REDCap? No. Those collect responses with no plan required up front. surveyframe declares the analysis plan before collection, and reads exported CSV data from any of them once the plan is added (see "Already have data?" above).

Does surveyframe do survey weighting or complex-sample variance estimation? No. survey and srvyr are the standard R tools for stratified, clustered, or weighted samples. surveyframe's instruments interoperate with that workflow instead of duplicating it.

Can I use surveyframe with data I've already collected? Yes. Build a minimal instrument that matches your column names, then load your CSV or data.frame directly with read_responses(). See "Already have data?" above for a worked example.

What is a pre-declared analysis plan, and why does it matter? A list of research questions written into the instrument at design time, each bound to a technique and to the variables that fill its roles, before any response arrives. It removes matching questions, variables, and tests up by hand, and it's the basis for the audit trail showing the plan wasn't changed after seeing results.

What happens if surveyframe is ever archived by CRAN? The GitHub repository stays the canonical source (remotes::install_github("MohammedAliSharafuddin/surveyframe")), and every CRAN release is separately deposited to Zenodo with its own DOI.

Citation

citation("surveyframe")

Related resources

• Sharafuddin, M. A., Jaleel, A. A., and Madhavan, M. (2026). Quantitative Analysis with Small Samples: A Practical Guide for Students and Early-Career Researchers (Version 0.1.0) [Book]. Zenodo. https://doi.org/10.5281/zenodo.20221929. A companion textbook on statistical inference when sample sizes are small, also available at https://flairmi.com/textbooks/smallsamplelab.html. It describes the small-sample methods that surveyframe added in 0.4.0 and when to prefer each one.
• Sharafuddin, M. A. (2026). surveyframe: A Pre-Declared, Reproducible Framework for Multi-Criteria Decision Analysis in Survey Research. Manuscript in preparation for Computo. Describes the 10 MCDM methods surveyframe added in 0.4.0.

License

MIT. See LICENSE.