surveyframe Field Types: Multiple Choice

Declare a multiple_choice question in surveyframe and see the real codebook, exported survey, multi-response summary table, and Cochran’s Q test it produces. Copy-pasteable R throughout.

Author

Affiliation

Mohammed Ali Sharafuddin

FlairMI

Keywords

surveyframe, multiple_choice, R package, survey design, analysis plan, cochran’s Q, sf_item

TL;DR: multiple_choice lets a respondent select several options at once, a nominal set rather than a single answer. This post declares one with sf_item(), shows the real codebook and exported survey, then runs a real Cochran’s Q test to check whether selection rates actually differ across options, alongside the multi-response summary table every “select all that apply” question needs. Every table and chart on this page, except the one screenshot of the exported survey, was generated by the R code shown, at render time, from surveyframe 0.3.3 installed straight from CRAN.

Try this yourself. Install surveyframe, paste the code blocks below into RStudio in order, and you will have a working instrument, a real exported survey, and a real analysis report before you reach the bottom of this page.

install.packages("surveyframe")

The case

You run a small store and just finished a promotion across several channels: Instagram, Facebook, TikTok, an email newsletter, and in-store signage. You want to know which of those channels people actually noticed, and a visitor may have noticed more than one. That rules out single_choice, since forcing one answer would throw away real information. It is a multiple_choice question: select all that apply.

Declare the item

A multiple_choice item is declared exactly like single_choice: an id, a label, a choice_set, and whether it is required. The difference is not in how you declare it, it is in how a respondent answers it and how the responses come back.

choices <- sf_choices(
  "channels_used",
  values = c("instagram", "facebook", "tiktok", "email", "in_store"),
  labels = c("Instagram", "Facebook", "TikTok", "Email newsletter", "In-store signage")
)

item <- sf_item(
  id = "channels_used",
  label = "Which of these did you notice our store's promotion on? Select all that apply.",
  type = "multiple_choice",
  choice_set = "channels_used",
  required = TRUE
)

instrument <- sf_instrument(
  title = "Promotion Awareness Survey",
  version = "0.1.0",
  components = list(choices, item)
)

See it as a codebook

codebook_report() produces the same two tables as it does for any item type: the items and their choice sets. Real output from the instrument declared above.

cb <- codebook_report(instrument, format = "md")

Table 1
Survey Items

id	label	type	choice_set	scale_id	reverse	required
channels_used	Which of these did you notice our store’s promotion on? Select all that apply.	multiple_choice	channels_used		FALSE	TRUE

Table 2
Choice Sets

choice_set_id	value	label
channels_used	instagram	Instagram
channels_used	facebook	Facebook
channels_used	tiktok	TikTok
channels_used	email	Email newsletter
channels_used	in_store	In-store signage

Export it and see what a respondent sees

export_static_survey() writes the same kind of self-contained HTML file it writes for any item type. What changes is the input: five checkboxes sharing one field name instead of one set of radio buttons, so a respondent can tick more than one.

export_static_survey(
  instrument,
  output_path = "promotion_survey.html",
  open = TRUE
)

This is the one piece of this page that is a screenshot rather than live output. What follows is that file, unedited, at phone width, with Instagram and TikTok both ticked.

Real output of export_static_survey() for the channels_used item: five bordered checkbox cards lettered A to E, two of them ticked (Instagram and TikTok), and a Submit button.

Write the analysis plan before you collect anything

A single multiple_choice item does not come back as one column. It comes back as one column per option, each TRUE or FALSE, since a respondent can tick any combination of the five. The frequency method from the single_choice post only tabulates one variable at a time, so it is the wrong tool here: run it on five binary columns and it quietly tabulates only the first one, dropping the other four (worth knowing before you rely on it).

The method built for exactly this shape is cochran_q: it tests whether the selection rate differs across a set of related binary measures. Declared with roles = list(measures = ...), naming every column, before any response exists.

instrument <- sf_instrument(
  title = "Promotion Awareness Survey",
  version = "0.1.0",
  components = list(choices, item),
  analysis_plan = list(
    list(
      id = "RQ1",
      research_question = "Do selection rates differ across the five channels?",
      family = "association",
      method = "cochran_q",
      roles = list(measures = c(
        "channels_used_instagram", "channels_used_facebook", "channels_used_tiktok",
        "channels_used_email", "channels_used_in_store"
      ))
    )
  )
)

Run the plan against responses

In your own project this is read_responses() reading a CSV or Google Sheet, with one logical column per checkbox option. Here it is 140 rows of synthetic data, generated once with a fixed seed so this page renders the same result every time.

set.seed(42)
n <- 140
responses <- data.frame(
  channels_used_instagram = sample(c(TRUE, FALSE), n, replace = TRUE, prob = c(0.55, 0.45)),
  channels_used_facebook  = sample(c(TRUE, FALSE), n, replace = TRUE, prob = c(0.35, 0.65)),
  channels_used_tiktok    = sample(c(TRUE, FALSE), n, replace = TRUE, prob = c(0.40, 0.60)),
  channels_used_email     = sample(c(TRUE, FALSE), n, replace = TRUE, prob = c(0.20, 0.80)),
  channels_used_in_store  = sample(c(TRUE, FALSE), n, replace = TRUE, prob = c(0.15, 0.85))
)

results <- run_analysis_plan(responses, instrument)
rq1 <- results[[1]]

Cochran’s Q(4, N = 140) = 43.56, p < .001

That result answers one question: are the five rates different at all. It does not say which channel leads, or by how much, which is what a reader actually wants to see first.

The multi-response summary table

This table is not surveyframe output, it is five lines of base R against the same responses data frame: how many respondents ticked each option, and what share of the sample that is. Every multiple_choice write-up needs a table shaped like this, whichever tool produces it.

labels_map <- setNames(choices$labels, paste0("channels_used_", choices$values))
summary_tbl <- data.frame(
  Channel = labels_map[names(responses)],
  Selected = colSums(responses),
  Percent = round(100 * colMeans(responses), 1)
)
rownames(summary_tbl) <- NULL
summary_tbl <- summary_tbl[order(-summary_tbl$Selected), ]

Table 3
Multi-Response Summary for channels_used

Channel	Selected	Percent
Instagram	65	46.4
TikTok	47	33.6
Facebook	42	30.0
Email newsletter	30	21.4
In-store signage	17	12.1

Instagram leads at 46.4%, in-store signage trails at 12.1%. Cochran’s Q above confirms that spread is real rather than sampling noise. Neither number replaces the other: the summary table shows the shape, the test shows whether the shape means anything.

Adapt this to your own survey

To reuse this for a different multiple_choice question, change four things and leave the rest:

1. The id and label in sf_item()
2. The values and labels in sf_choices()
3. The column names in the plan’s roles$measures, and in responses, to match your new id (surveyframe names each binary column <id>_<value>)
4. The probabilities in the synthetic responses, or point read_responses() at your real data

The codebook, the export, and the multi-response summary table work the same way regardless of how many options the question offers.

Get surveyframe

install.packages("surveyframe")

• Documentation: mohammedalisharafuddin.github.io/surveyframe
• Source: github.com/MohammedAliSharafuddin/surveyframe
• Release notes: surveyframe 0.3.3 is on CRAN
• The single_choice walkthrough this post builds on: surveyframe Field Types: Single Choice