Skip to content

From question to analysis

Choose a question. See what you can learn from it.

Explore question types, try answering and see the data you get. Then choose an analysis and view an R example.

This educational guide uses synthetic data. Example answers are neither stored nor sent. R code shows standalone examples, not calculations executed here.

How do you carry out analysis in R? ↓

Single choice

Try the question

Which service channel do you use most often?

Change the answer to see how it is recorded.

What data will you get?

One category per person. A dropdown changes how you select, not the measurement level.

What can you calculate or understand?

Counts, percentages and cross-tabulations. Compare response distributions across groups; the chi-squared test requires appropriate expected counts and independent observations.

How should you interpret the result?

A category code is not a number to average. For small counts, consider Fisher’s exact test.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

channel <- factor(c("online", "phone", "online", "office", "phone", "online"))
counts <- table(channel)
counts
100 * prop.table(counts)

Yes / No

Try the question

Were you able to resolve your issue?

Change the answer to see how it is recorded.

What data will you get?

One category per person. A dropdown changes how you select, not the measurement level.

What can you calculate or understand?

Counts, percentages and cross-tabulations. Compare response distributions across groups; the chi-squared test requires appropriate expected counts and independent observations.

How should you interpret the result?

A category code is not a number to average. For small counts, consider Fisher’s exact test.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

channel <- factor(c("online", "phone", "online", "office", "phone", "online"))
counts <- table(channel)
counts
100 * prop.table(counts)

Multiple choice

Try the question

Which service channels do you use?

Change the answer to see how it is recorded.

What data will you get?

A separate 0/1 indicator for each option. A person can select several answers.

What can you calculate or understand?

The number and percentage of people selecting each option, and co-occurrence of selections.

How should you interpret the result?

State the percentage base. Percentages of people may sum to more than 100%; selections by one person are not independent respondents.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

answers <- data.frame(online = c(1, 1, 0, 1), phone = c(0, 1, 1, 1))
colSums(answers)
100 * colMeans(answers)

Scale

Try the question

How would you rate the service from 1 to 5?

Change the answer to see how it is recorded.

What data will you get?

Ordered ratings. Define the endpoints and use a consistent scale direction.

What can you calculate or understand?

Distribution, median and quartiles. Spearman correlation for rank relationships; means and parametric tests require justification of measurement level and assumptions.

How should you interpret the result?

A single ordinal rating is not automatically a continuous measure. Report the distribution, not just the mean.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

score <- c(1, 2, 3, 3, 4, 4, 5, NA)
table(score, useNA = "ifany")
median(score, na.rm = TRUE)
quantile(score, na.rm = TRUE)

Rating

Try the question

How would you rate the service from 1 to 5?

Change the answer to see how it is recorded.

What data will you get?

Ordered ratings. Define the endpoints and use a consistent scale direction.

What can you calculate or understand?

Distribution, median and quartiles. Spearman correlation for rank relationships; means and parametric tests require justification of measurement level and assumptions.

How should you interpret the result?

A single ordinal rating is not automatically a continuous measure. Report the distribution, not just the mean.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

score <- c(1, 2, 3, 3, 4, 4, 5, NA)
table(score, useNA = "ifany")
median(score, na.rm = TRUE)
quantile(score, na.rm = TRUE)

NPS

Try the question

How likely are you to recommend us on a scale from 0 to 10?

Change the answer to see how it is recorded.

What data will you get?

A rating from 0–10: detractors 0–6, passives 7–8, promoters 9–10.

What can you calculate or understand?

NPS is the percentage of promoters minus the percentage of detractors. Also report the base size and full rating distribution.

How should you interpret the result?

NPS ranges from −100 to 100. It is neither the average rating nor the percentage satisfied. Small samples give unstable results.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

score <- c(0, 6, 7, 8, 9, 10, 10, NA)
valid <- score[!is.na(score)]
stopifnot(length(valid) > 0, all(valid %in% 0:10))
100 * (mean(valid >= 9) - mean(valid <= 6))

Ranking

Try the question

Rank speed, price and availability.
  1. Speed
  2. Pricing
  3. Availability

Change the answer to see how it is recorded.

What data will you get?

Each option’s position within the same person’s ranking. A lower rank means a higher priority.

What can you calculate or understand?

Share of first places, median rank and position comparisons. Friedman’s test can be used for complete, paired rankings.

How should you interpret the result?

Rankings show order, not the size of preference differences. Define how missing values and ties are handled.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

ranks <- matrix(c(1,2,3, 2,1,3, 1,3,2, 3,1,2), ncol = 3, byrow = TRUE)
colnames(ranks) <- c("speed", "price", "availability")
colMeans(ranks == 1)
friedman.test(ranks)

Matrix

Try the question

Rate speed, clarity of information and courtesy on the same scale.

Change the answer to see how it is recorded.

What data will you get?

A separate variable for each row. The shared format makes aspects easier to compare.

What can you calculate or understand?

Distributions and medians per row. For items measuring a common construct: scale reliability, Cronbach’s alpha and factor analysis after checking assumptions.

How should you interpret the result?

Similar-looking questions do not necessarily form one scale. Reverse-code relevant items and check construct validity; a high alpha does not establish validity.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

items <- data.frame(a = c(1,2,3,4,5,3), b = c(2,2,3,4,5,4), c = c(1,3,3,4,4,3))
sapply(items, median)
# Optional package for reliability:
# psych::alpha(items)

Advanced matrix

Try the question

For each service, give a rating, visit count and comment.

Change the answer to see how it is recorded.

What data will you get?

Cells can use different types. Keep variables and their measurement levels separate.

What can you calculate or understand?

Analyse numbers as numbers, choices as categories and text through thematic coding. Join results by response identifier.

How should you interpret the result?

Do not average an entire mixed-type matrix. Check the export before reshaping it into an analysis table.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

answers <- data.frame(rating = c(3,4,5), visits = c(1,4,2), comment = c("slow", "clear", "helpful"))
summary(answers[c("rating", "visits")])

Text block

Try the question

In a moment we will ask about your last visit. Your answers are confidential.

In a moment we will ask about your last visit. Your answers are confidential. This example block does not collect an answer.

Change the answer to see how it is recorded.

What data will you get?

No data. The block shows an instruction or introduction and has no answer field.

What can you calculate or understand?

It is not analysed like a question. When interpreting results, take into account which instruction respondents saw before the following questions.

How should you interpret the result?

The wording of an instruction can influence answers. Changing the block between study waves makes comparisons harder; do not put questions you expect answers to in it.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

# A text block stores no answer column.
# Document its wording next to the dataset, e.g. in a codebook.

Number

Try the question

How many minutes did you wait?

Change the answer to see how it is recorded.

What data will you get?

A numeric value with a unit and range. Zero, a missing answer and “not applicable” have different meanings.

What can you calculate or understand?

Mean, median, standard deviation, quartiles and confidence intervals. For independent groups: Welch’s t-test or Welch’s ANOVA. For relationships: correlation or regression.

How should you interpret the result?

Check the distribution, outliers and independence. Correlation does not prove causation. Use paired analysis for before-and-after measurements.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

d <- data.frame(wait = c(3,5,4,7,8,12,9,10), group = rep(c("A", "B"), each = 4))
summary(d$wait)
sd(d$wait)
t.test(wait ~ group, data = d, var.equal = FALSE)
oneway.test(wait ~ group, data = d, var.equal = FALSE)

Short text

Try the question

What made the service hardest to use?

Change the answer to see how it is recorded.

What data will you get?

Responses in people’s own words, with context and the option of thematic coding.

What can you calculate or understand?

Define codes, assign them to responses and check coding consistency. Compare themes with quantitative ratings and approved quotations.

How should you interpret the result?

Word counts do not measure a theme’s importance. Code frequency does not establish population prevalence; quotations require checks for possible identification.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

coded <- data.frame(id = 1:4, topic = c("time", "clarity", "time", "access"))
table(coded$topic)
# Codes were assigned after reading the answers.

Longer response

Try the question

What made the service hardest to use?

Change the answer to see how it is recorded.

What data will you get?

Responses in people’s own words, with context and the option of thematic coding.

What can you calculate or understand?

Define codes, assign them to responses and check coding consistency. Compare themes with quantitative ratings and approved quotations.

How should you interpret the result?

Word counts do not measure a theme’s importance. Code frequency does not establish population prevalence; quotations require checks for possible identification.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

coded <- data.frame(id = 1:4, topic = c("time", "clarity", "time", "access"))
table(coded$topic)
# Codes were assigned after reading the answers.

Date

Try the question

When was the service used?

Change the answer to see how it is recorded.

What data will you get?

Dates, times or timestamps. Preserve the time zone and format.

What can you calculate or understand?

Counts by period, trends and time between events. Account for seasonality and different period lengths when comparing.

How should you interpret the result?

Time of day is cyclical: the mean of 23:00 and 01:00 should not indicate noon. A missing date does not mean zero time.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

dates <- as.Date(c("2026-01-03", "2026-01-10", "2026-02-02"))
table(format(dates, "%Y-%m"))
as.numeric(diff(sort(dates)))

Time

Try the question

When was the service used?

Change the answer to see how it is recorded.

What data will you get?

Dates, times or timestamps. Preserve the time zone and format.

What can you calculate or understand?

Counts by period, trends and time between events. Account for seasonality and different period lengths when comparing.

How should you interpret the result?

Time of day is cyclical: the mean of 23:00 and 01:00 should not indicate noon. A missing date does not mean zero time.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

dates <- as.Date(c("2026-01-03", "2026-01-10", "2026-02-02"))
table(format(dates, "%Y-%m"))
as.numeric(diff(sort(dates)))

Date and time

Try the question

When was the service used?

Change the answer to see how it is recorded.

What data will you get?

Dates, times or timestamps. Preserve the time zone and format.

What can you calculate or understand?

Counts by period, trends and time between events. Account for seasonality and different period lengths when comparing.

How should you interpret the result?

Time of day is cyclical: the mean of 23:00 and 01:00 should not indicate noon. A missing date does not mean zero time.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

dates <- as.Date(c("2026-01-03", "2026-01-10", "2026-02-02"))
table(format(dates, "%Y-%m"))
as.numeric(diff(sort(dates)))

Email address

Try the question

How can we contact you?

Change the answer to see how it is recorded.

What data will you get?

Contact details or a link, with format validation. Collect them only when required for the research purpose.

What can you calculate or understand?

Check completeness and possible duplicates within an authorised process. Use a pseudonymous identifier for opinion analysis.

How should you interpret the result?

A valid format does not verify that an address exists or that contact is authorised. Do not include raw contact details in a public report.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

complete <- c(TRUE, FALSE, TRUE, TRUE)
table(complete)
# No contact details are needed in this example.

Phone

Try the question

How can we contact you?

Change the answer to see how it is recorded.

What data will you get?

Contact details or a link, with format validation. Collect them only when required for the research purpose.

What can you calculate or understand?

Check completeness and possible duplicates within an authorised process. Use a pseudonymous identifier for opinion analysis.

How should you interpret the result?

A valid format does not verify that an address exists or that contact is authorised. Do not include raw contact details in a public report.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

complete <- c(TRUE, FALSE, TRUE, TRUE)
table(complete)
# No contact details are needed in this example.

Website address

Try the question

Enter the website address your feedback refers to.

Change the answer to see how it is recorded.

What data will you get?

Contact details or a link, with format validation. Collect them only when required for the research purpose.

What can you calculate or understand?

Check completeness and possible duplicates within an authorised process. Use a pseudonymous identifier for opinion analysis.

How should you interpret the result?

A valid format does not verify that an address exists or that contact is authorised. Do not include raw contact details in a public report.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

complete <- c(TRUE, FALSE, TRUE, TRUE)
table(complete)
# No contact details are needed in this example.

Point on a map

Try the question

Mark a location where you need a new service.

Demonstration diagram, without GPS or real coordinates.

Change the answer to see how it is recorded.

What data will you get?

A point selected by the respondent, without automatic GPS collection. This differs from the interview location.

What can you calculate or understand?

In the platform: aggregate points into a 100–10,000 m grid, apply a publication threshold of at least 5 and export aggregated GeoJSON. After export, R can visualise the grid using sf.

How should you interpret the result?

The module requires activation, a purpose, privacy settings and retention. Results below the threshold are suppressed; do not publish exact points. Grid size affects the result’s appearance.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

# After downloading an aggregated GeoJSON from the platform:
# install.packages("sf") # once, in your own R environment
# cells <- sf::st_read("spatial.geojson", quiet = TRUE)
# plot(sf::st_geometry(cells))
# Check attribute names with names(cells) before mapping counts.

Area selection

Try the question

Which of these areas needs a change?

Change the answer to see how it is recorded.

What data will you get?

The selected area’s identifier and a version of its Polygon or MultiPolygon geometry.

What can you calculate or understand?

In the platform: counts by area with a publication threshold and GeoJSON export. After export, R can combine aggregates with geometry and draw a map.

How should you interpret the result?

More selections do not imply higher risk per resident. Comparisons need the correct denominator; check boundary versions before combining them.

View an R example

A standalone example using synthetic data. Replace the data and column names with your own prepared dataset.

area <- factor(c("A", "B", "A", "C", "B", "A"))
table(area)
100 * prop.table(table(area))
# Use only privacy-safe aggregates for a public map.

How do you carry out analysis in R?

  1. 1. Prepare the dataset

    In Analysis Studio, select study data or import CSV or SAV. Check variable codes, measurement levels, missing values, filters and the analysis base. Your own R export may require reshaping multiple responses and matrices.

  2. 2. Choose the method

    Start with distributions and descriptive statistics. Then choose a cross-tabulation, correlation, difference test, regression or reliability analysis. Define the hypothesis, measurement level and whether observations are independent or paired.

  3. 3. Run and check the calculations

    Choose the R engine and an available analysis type, then select variables, groups and parameters. The R engine and packages must be configured. Spatial analyses in the platform use a separate aggregation engine; R supports further work after export.

  4. 4. Report the result and limitations

    Report N, missing values, filters, weights and method. Interpret effect size and confidence intervals where available; p alone does not indicate practical importance. Retain the data version and analysis report.

Which statistics does the platform support?

Core R analyses: descriptive statistics, cross-tabulations, Pearson, Spearman and Kendall correlations, linear regression, Welch’s t-test, Welch’s ANOVA, Cronbach’s alpha and weighted tables.

Extended methods include paired measurements, Wilcoxon tests, adjusted pairwise comparisons, correlation inference, survey design analysis, GLM, psychometrics, multiple imputation and k-means clustering. Appropriate packages and configuration are required. Not every method suits every dataset.

The Friedman, Fisher and sf mapping examples illustrate standalone R workflows; they do not imply dedicated buttons for those analyses in the platform.

The application interface, tutorials, recordings, training files and legal documents are currently in Polish. Contact us to discuss deployment in another language.