---

name: r-dplyr-wrangling description: > Use when writing or fixing dplyr code, when a filter returns zero rows or a recoded column has unexpected NA, when a grouped summary has missing groups or stays grouped, or when explaining these verbs to a learner. Governs reshaping a data frame in R with dplyr before it feeds a dashboard. Do not use for reading files or APIs (r-data-prep), statistical tests (r-stats-foundations) or chart code (r-dashboard-design).

R dplyr wrangling

Stance: check the data before and after every verb (frozen); base R, dplyr or data.table is flexible and the project's style wins. Teach one verb at a time, run it, look at the output, then combine.

Hard rules

• Never filter with == NA or == "NA". Mechanism: symptom is zero rows; cause is that a missing value is not equal to anything; fix is is.na(). Reason: the learner would conclude the data is empty.
• Never leave a case_when() without a default when unmatched rows matter. Mechanism: unmatched rows become NA and drop out of later counts. Fix: .default = "Other" or an explicit check.
• Never drop rows silently. Mechanism: na.omit() and drop_na() change the totals a dashboard shows; fix is to count the rows before and after and say so.
• Do not leave data grouped. Mechanism: a grouped table makes later summarise(), mutate() and slice() run per group and gives odd totals.
• Keep the original column until the new one is checked. Mechanism: overwriting loses the evidence needed to find a wrong rule.
• Use made-up or the learner's own data for examples, and label made-up data as such. Reason: real health records are personal.

Agent dispatch

Before handling a request, follow the agent selection in r-datadash, including when this skill is invoked directly. Give the selected agent this skill's instructions. If no agent is available, continue yourself.

Gate

| Situation | Action | |---|---| | Choose, drop, reorder or rename columns | select() or rename() | | Keep or drop rows by a rule | filter(); decide what happens to NA first | | New or recoded column | mutate(); case_when() for several rules, if_else() for two | | Summary table or counts | group_by() then summarise(), or count(); then ungroup() | | A value per row relative to its group | group_by() then mutate() or arrange(), then ungroup() | | Reading files, APIs or dates | Hand to r-data-prep | | Data holds personal information | Stop and ask what may be kept; summarise before showing |

Steps

1. Detect. Look at the columns and types (glimpse()), the first rows and the missing values. Name the question the table must answer.
2. Build one verb at a time. Run each step, print it, and check the row and column counts. Use verb-patterns.md for the pattern and the trap of each verb.
3. Handle NA on purpose. Missing values never match == or !=. Use is.na() and decide in words whether missing rows are kept, dropped or labelled.
4. Order the rules. In case_when() the first matching rule wins, so write the most specific rule first and finish with an explicit .default. Check that no real value fell through to NA.
5. Finish grouped work. After summarise() on two or more grouping variables, or after any grouped mutate(), call ungroup() or set .groups = "drop".
6. Verify. Compare one total with the source (the group counts add up to the row count), print the final shape, and say which checks were not made.

Checklist

- [ ] Types and missing values looked at before any verb
- [ ] Each verb run and its output looked at
- [ ] NA handled with is.na(), not == NA
- [ ] case_when rules ordered, with an explicit default
- [ ] Group counts add up to the row count
- [ ] Data ungrouped before further use
- [ ] Column names are snake_case with no spaces

Example

Input: "My filter for adults returns nothing, and age_group is NA for some rows." Output: the age column is text, or has NA values that < cannot compare. Convert it to numeric on purpose, add a rule or label for NA, put the rules in order from the most specific, and report how many rows ended up in each group so a missing group is visible.

References

• verb-patterns.md: each verb's pattern, trap and a small made-up example.

Package docs: dplyr and tidyr on CRAN. Check the installed version for any argument you are unsure about, for example .default in case_when() needs dplyr 1.1.0 or later.

Next

• Load and clean first: r-data-prep
• Test or model it: r-stats-foundations
• Show it: r-dashboard-design, r-shiny-dashboards, r-quarto-dashboards