Summary

Check that you can reorder observations, combine tables using an appropriate rule, work with text, and summarize values by group.

Before you start: table selection; function calls and returned values.
Setup: Tables are supplied below. Only the final reshape check needs the installed reshape2 package.
Your check: Check text keys, join/group observations and explain their table layout. Practise in the unit summary.

What you learned

Task	Tool	Main distinction
Sort values	`sort(x)`	Returns sorted values. Factor order depends on the levels.
Reorder a table	`df[order(df$column), ]`	`order()` supplies positions; reorder whole rows to preserve relationships between columns.
Append observations	`rbind()`	Stacks compatible tables; check names, types, and units.
Put columns side by side	`cbind()`	Aligns by position, not by a shared identifier; check row correspondence first.
Match tables by identifiers	`merge(..., by = "site")`	Matches key values even when row orders differ.
Work with text	`nchar()`, `paste()`, `sprintf()`, `grep()`, `sub()`, `strsplit()`	Inspect, construct, find, replace, or split strings.
Summarize by group	`aggregate()`	Reduces observations to group summaries.
Change table layout	Wide/long reshaping	Changes how measurements are arranged in rows and columns.

For merge(), specify the key explicitly. By default, all common column names are used as keys and only matching combinations are retained. all.x = TRUE keeps every row from the first table, adding missing values where there is no match; all = TRUE keeps unmatched rows from both. Duplicate keys can create multiple matches and increase the number of rows, so check keys and the result.

A short recap in R

These illustrative observations arrive in a different site order from the lookup table. The site identifier, not row position, determines the match.

sites <- data.frame(
  site = c("A", "B", "C"),
  habitat = c("forest", "grassland", "forest")
)
measurements <- data.frame(
  site = c("C", "A", "B"),
  height_cm = c(18, 12, 9)
)
combined <- merge(measurements, sites, by = "site")
combined <- combined[order(combined$height_cm), ]
combined
aggregate(height_cm ~ habitat, data = combined, FUN = mean)

The sorted table contains sites B, A, and C with heights 9, 12, and 18 cm. The mean height is 15 cm for forest and 9 cm for grassland. Aggregation produces one row per habitat; it does not retain each original observation.

Text and table layout

Check spelling, capitalization, and whitespace before matching text keys, and document any changes. Regular expressions describe patterns: characters such as . or * have special meanings. Use literal matching, such as fixed = TRUE where supported, when you mean the exact characters. grep() returns matching positions by default; grepl() returns a logical vector. sub() replaces the first match in each string; gsub() replaces all matches.

In a wide table, repeated measurements can occupy separate columns; in a long table, a variable identifies the measurement and another column holds its value. The lesson demonstrates melt() and dcast() from the additional package reshape2. Reshaping and aggregation serve different purposes, although casting may also aggregate when several observations belong to one output cell. Check the identifiers and intended unit of observation before either operation.

Try it yourself

Work from these small tables. Each measurement row is one observation; lookup should have one row per site.

lookup <- data.frame(site = c("A", "B"),
                     habitat = c("forest", "grassland"))
samples <- data.frame(site = c(" b ", "A", "C", "A"),
                      height_cm = c(9, 12, 5, 18))

Standardise the measurement keys: trimws() removes surrounding whitespace and toupper() makes letters uppercase. Check the resulting keys.
Sort complete rows by site, then height; the two A rows should show the secondary sort. Append a new B measurement of 10 cm to a copy using rbind().
Join the original four cleaned measurements to lookup, retaining unmatched sites. Check the keys and row count; do not use cbind() to attach these differently sized tables.
Compute mean height by known habitat. Keep the unmatched observation in the joined table; explicitly select known habitats for this summary and explain what is excluded.

Compare a possible solution

samples$site <- toupper(trimws(samples$site))
ordered <- samples[order(samples$site, samples$height_cm), ]
with_extra <- rbind(samples, data.frame(site = "B", height_cm = 10))
anyDuplicated(lookup$site)
samples$site[!samples$site %in% lookup$site]
joined <- merge(samples, lookup, by = "site", all.x = TRUE)
nrow(joined)
known <- joined[!is.na(joined$habitat), ]
aggregate(height_cm ~ habitat, data = known, FUN = mean)

ordered has A/12, A/18, B/9 and C/5; with_extra has five rows. The lookup has no duplicate key (0), and C is unmatched. The join keeps four observations; C’s habitat is NA. The known-habitat means are forest 15 cm and grassland 9 cm, using three observations. This grouping excludes C because its habitat is unavailable, not because its height is invalid.

For the reshaping outcome, use the reshape2 package introduced in the lesson. Turn the two seasonal columns below into a four-row long table, then restore the wide table. State what one row means in each layout. This extension requires the installed package; it does not require attaching it.

seasonal <- data.frame(site = c("A", "B"), spring = c(10, 7),
                       autumn = c(12, 9))

Compare the reshape check

seasonal_long <- reshape2::melt(seasonal, id.vars = "site",
                                measure.vars = c("spring", "autumn"),
                                variable.name = "season", value.name = "height_cm")
seasonal_back <- reshape2::dcast(seasonal_long, site ~ season,
                                 value.var = "height_cm")
seasonal_back <- seasonal_back[names(seasonal)]
dim(seasonal_long)
identical(seasonal, seasonal_back)

The long table has four rows and three columns: one row per site–season measurement. The wide table has one row per site. Casting can reorder columns, so the comparison restores their original order explicitly; it returns TRUE. No aggregation is needed because each site–season key is unique.

Ready to continue: you can choose binding versus matching, inspect text keys, summarize a declared group and explain a change of table layout.

Check your understanding

Try these questions before opening the suggested answers.

Why would cbind(measurements, sites) give incorrect site/habitat combinations in this example?
What happens to an unmatched measurement with the default merge(), and how could you retain it?
How do sorting, aggregation, and wide-to-long reshaping each change a table?