Joins

Why the cheatsheet

• Examples for those of us who don’t speak SQL so good. There are lots of Venn diagrams re: SQL joins on the interwebs, but I wanted R examples.
• Full documentation for the dplyr package, which is developed by Hadley Wickham and Romain Francois on GitHub.
• The vignette on Two-table verbs covers the joins shown here.

inner_join(superheroes, publishers)

inner_join(x, y): Return all rows from x where there are matching values in y, and all columns from x and y. If there are multiple matches between x and y, all combination of the matches are returned. This is a mutating join.

(ijsp <- inner_join(superheroes, publishers))

## # A tibble: 6 × 5
##       name alignment gender publisher yr_founded
##      <chr>     <chr>  <chr>     <chr>      <int>
## 1  Magneto       bad   male    Marvel       1939
## 2    Storm      good female    Marvel       1939
## 3 Mystique       bad female    Marvel       1939
## 4   Batman      good   male        DC       1934
## 5    Joker       bad   male        DC       1934
## 6 Catwoman       bad female        DC       1934

We lose Hellboy in the join because, although he appears in x = superheroes, his publisher Dark Horse Comics does not appear in y = publishers. The join result has all variables from x = superheroes plus yr_founded, from y.

semi_join(superheroes, publishers)

semi_join(x, y): Return all rows from x where there are matching values in y, keeping just columns from x. A semi join differs from an inner join because an inner join will return one row of x for each matching row of y, where a semi join will never duplicate rows of x. This is a filtering join.

(sjsp <- semi_join(superheroes, publishers))

## # A tibble: 6 × 4
##       name alignment gender publisher
##      <chr>     <chr>  <chr>     <chr>
## 1   Batman      good   male        DC
## 2    Joker       bad   male        DC
## 3 Catwoman       bad female        DC
## 4  Magneto       bad   male    Marvel
## 5    Storm      good female    Marvel
## 6 Mystique       bad female    Marvel

We get a similar result as with inner_join() but the join result contains only the variables originally found in x = superheroes. But note the row order has changed.

left_join(superheroes, publishers)

left_join(x, y): Return all rows from x, and all columns from x and y. If there are multiple matches between x and y, all combination of the matches are returned. This is a mutating join.

(ljsp <- left_join(superheroes, publishers))

## # A tibble: 7 × 5
##       name alignment gender         publisher yr_founded
##      <chr>     <chr>  <chr>             <chr>      <int>
## 1  Magneto       bad   male            Marvel       1939
## 2    Storm      good female            Marvel       1939
## 3 Mystique       bad female            Marvel       1939
## 4   Batman      good   male                DC       1934
## 5    Joker       bad   male                DC       1934
## 6 Catwoman       bad female                DC       1934
## 7  Hellboy      good   male Dark Horse Comics         NA

We basically get x = superheroes back, but with the addition of variable yr_founded, which is unique to y = publishers. Hellboy, whose publisher does not appear in y = publishers, has an NA for yr_founded.

anti_join(superheroes, publishers)

anti_join(x, y): Return all rows from x where there are not matching values in y, keeping just columns from x. This is a filtering join.

(ajsp <- anti_join(superheroes, publishers))

## # A tibble: 1 × 4
##      name alignment gender         publisher
##     <chr>     <chr>  <chr>             <chr>
## 1 Hellboy      good   male Dark Horse Comics

We keep only Hellboy now (and do not get yr_founded).

inner_join(publishers, superheroes)

inner_join(x, y): Return all rows from x where there are matching values in y, and all columns from x and y. If there are multiple matches between x and y, all combination of the matches are returned. This is a mutating join.

(ijps <- inner_join(publishers, superheroes))

## # A tibble: 6 × 5
##   publisher yr_founded     name alignment gender
##       <chr>      <int>    <chr>     <chr>  <chr>
## 1        DC       1934   Batman      good   male
## 2        DC       1934    Joker       bad   male
## 3        DC       1934 Catwoman       bad female
## 4    Marvel       1939  Magneto       bad   male
## 5    Marvel       1939    Storm      good female
## 6    Marvel       1939 Mystique       bad female

In a way, this does illustrate multiple matches, if you think about it from the x = publishers direction. Every publisher that has a match in y = superheroes appears multiple times in the result, once for each match. In fact, we’re getting the same result as with inner_join(superheroes, publishers), up to variable order (which you should also never rely on in an analysis).

semi_join(publishers, superheroes)

semi_join(x, y): Return all rows from x where there are matching values in y, keeping just columns from x. A semi join differs from an inner join because an inner join will return one row of x for each matching row of y, where a semi join will never duplicate rows of x. This is a filtering join.

(sjps <- semi_join(x = publishers, y = superheroes))

## # A tibble: 2 × 2
##   publisher yr_founded
##       <chr>      <int>
## 1    Marvel       1939
## 2        DC       1934

Now the effects of switching the x and y roles is more clear. The result resembles x = publishers, but the publisher Image is lost, because there are no observations where publisher == "Image" in y = superheroes.

left_join(publishers, superheroes)

left_join(x, y): Return all rows from x, and all columns from x and y. If there are multiple matches between x and y, all combination of the matches are returned. This is a mutating join.

(ljps <- left_join(publishers, superheroes))

## # A tibble: 7 × 5
##   publisher yr_founded     name alignment gender
##       <chr>      <int>    <chr>     <chr>  <chr>
## 1        DC       1934   Batman      good   male
## 2        DC       1934    Joker       bad   male
## 3        DC       1934 Catwoman       bad female
## 4    Marvel       1939  Magneto       bad   male
## 5    Marvel       1939    Storm      good female
## 6    Marvel       1939 Mystique       bad female
## 7     Image       1992     <NA>      <NA>   <NA>

We get a similar result as with inner_join() but the publisher Image survives in the join, even though no superheroes from Image appear in y = superheroes. As a result, Image has NAs for name, alignment, and gender.

anti_join(publishers, superheroes)

anti_join(x, y): Return all rows from x where there are not matching values in y, keeping just columns from x. This is a filtering join.

(ajps <- anti_join(publishers, superheroes))

## # A tibble: 1 × 2
##   publisher yr_founded
##       <chr>      <int>
## 1     Image       1992

We keep only publisher Image now (and the variables found in x = publishers).

full_join(superheroes, publishers)

full_join(x, y): Return all rows and all columns from both x and y. Where there are not matching values, returns NA for the one missing. This is a mutating join.

(fjsp <- full_join(superheroes, publishers))

## # A tibble: 8 × 5
##       name alignment gender         publisher yr_founded
##      <chr>     <chr>  <chr>             <chr>      <int>
## 1  Magneto       bad   male            Marvel       1939
## 2    Storm      good female            Marvel       1939
## 3 Mystique       bad female            Marvel       1939
## 4   Batman      good   male                DC       1934
## 5    Joker       bad   male                DC       1934
## 6 Catwoman       bad female                DC       1934
## 7  Hellboy      good   male Dark Horse Comics         NA
## 8     <NA>      <NA>   <NA>             Image       1992

We get all rows of x = superheroes plus a new row from y = publishers, containing the publisher Image. We get all variables from x = superheroes AND all variables from y = publishers. Any row that derives solely from one table or the other carries NAs in the variables found only in the other table.