dplyr bind_rows() and bind_cols() in R: Combine Tables

The bind_rows() and bind_cols() functions in dplyr stack data frames vertically (more rows) or horizontally (more columns). Unlike base R rbind() and cbind(), they handle column-name mismatches gracefully and return a tibble.

bind_rows(df1, df2)                       # stack vertically
bind_rows(df1, df2, df3)                  # multiple data frames
bind_rows(list_of_dfs)                    # from a list
bind_rows(df1, df2, .id = "source")       # add origin column
bind_cols(df1, df2)                       # stack horizontally (must match nrows)
bind_rows(df1, df2)                       # missing cols filled with NA
do.call(bind_rows, list_of_dfs)           # equivalent to bind_rows(list)

What bind_rows() and bind_cols() do in one sentence

bind_rows() stacks data frames VERTICALLY (more rows); bind_cols() stacks them HORIZONTALLY (more columns). They are pipe-friendly replacements for base R rbind() and cbind(), with smarter handling of mismatched columns and consistent tibble output.

bind_rows() is forgiving: missing columns in one input become NA in the output. bind_cols() is strict: the inputs must have the same number of rows.

Syntax

Both functions accept any number of data frames as separate arguments OR a single list of data frames. .id adds a column tracking which input each row came from.

The full signatures:

bind_rows(..., .id = NULL)
bind_cols(..., .name_repair = c("unique","universal","check_unique","minimal"))

... accepts any number of data frames (or a single list of them). .id (bind_rows only) creates an origin tracking column.

Six common patterns

1. Stack vertically with bind_rows

The two data frames share columns; the result has the union of all rows.

2. Track origin with .id

When inputs are NAMED arguments, .id = "quarter" creates a column with the name of each input.

3. Stack horizontally with bind_cols

bind_cols() requires the same number of rows. It does NOT match by key; it concatenates positionally (row 1 with row 1, etc.).

4. Mismatched columns get NA

b exists in df1 but not df2; c exists in df2 but not df1. The bound result has both columns; missing values are NA.

5. Stack a list of data frames

A common pattern: read N files via lapply()/map(), combine via bind_rows().

6. bind_cols with name conflict handling

When columns conflict, .name_repair = "universal" appends position numbers to make names unique. Default is "unique" which prepends a position prefix.

bind_rows() / bind_cols() vs base R

Base R rbind() and cbind() behave similarly but are stricter about column-name matches and convert to data.frame (not tibble).

When to use which:

• Use bind_rows() when columns might differ between inputs.
• Use rbind() when columns are guaranteed identical and you want zero dependencies.
• Use bind_cols() for tibble output; cbind() for base data.frame.

Common pitfalls

Pitfall 1: bind_cols silently aligns by row position, not key. If you have customer_data and order_data both with 1000 rows but different orders, bind_cols() returns garbage. Use left_join() to match by key.

Pitfall 2: type coercion with bind_rows. If column x is integer in df1 and character in df2, the bound result coerces to character (the more permissive type). This silent coercion can change downstream behavior.

Pitfall 3: factor columns may bind weirdly. Two factor columns with different levels get coerced to character or to the union of levels (depending on dplyr version). For predictable results, convert factors to character before binding, or use forcats::fct_unify() first.

Try it yourself

Related dplyr functions

After mastering bind_rows() and bind_cols(), look at:

• left_join(), inner_join(), full_join(): combine by KEY (not position)
• rows_insert(), rows_update(), rows_upsert(): surgical row updates by key
• Base R rbind(), cbind(): zero-dependency alternatives
• purrr::list_rbind(), list_cbind(): variants for purrr workflows
• do.call(rbind, list_of_dfs): pre-dplyr idiom, still works

For most data combination tasks, bind_rows() is the right choice. Reach for bind_cols() only when you really mean positional column-stacking (rare).

FAQ

What is the difference between bind_rows and rbind in R?

bind_rows() is from dplyr; rbind() is base R. Both stack data frames vertically. Differences: bind_rows() handles mismatched columns by filling NA (rbind errors), accepts lists directly, and returns a tibble. rbind() is faster on identically-shaped inputs and zero-dependency.

How do I combine multiple data frames in dplyr?

For VERTICAL stacking: bind_rows(df1, df2, df3) or bind_rows(list_of_dfs). For HORIZONTAL stacking: bind_cols(df1, df2) (rows must match). For COMBINING BY KEY: left_join() or other join verbs.

What is the difference between bind_cols and cbind?

bind_cols() is from dplyr; cbind() is base R. Both stack horizontally. bind_cols() returns a tibble and uses .name_repair for column-name conflict handling. cbind() returns a data.frame and may error or recycle on mismatched row counts.

How do I track the source of each row when binding?

Use .id with named inputs: bind_rows(a = df1, b = df2, .id = "source"). The result has a source column with values "a" and "b" indicating where each row came from.

Can I bind data frames with different columns in dplyr?

Yes, with bind_rows(). Missing columns are filled with NA. bind_rows(df1, df2) where df1 has columns (a, b) and df2 has columns (a, c) returns a data frame with columns (a, b, c) and NAs in the missing cells.