purrr::pmap and function with multiple conditions

Question

Whenever I use the purrr library function pmap() with a function that contains an if statement with multiple conditions, the if statement does not seem to work properly. To show you what I mean, here is a reproducible example using the gapminder dataset.

library(tidyverse)
library(gapminder)
library(broom)

# Nest the tibble into separate dataframes for each country-continent combination

by_country <- gapminder %>%
  group_by(country, continent) %>%
  nest()

Now I want to build a linear regression model for each grouped dataframe. The catch is that I want to use a different x-variable in my model depending on the country and continent. Here is my function where I suspect there is a problem with the if statement:

# My function
country_model <- function(df, cont, count) {

  if(cont == "Asia" & count == "Afghanistan") { # 2 conditions

    lm(lifeExp ~ year, data = df)

  } else {

    lm(lifeExp ~ pop, data = df)

  }
}

Now i'm going to take that function and apply it to all the grouped dataframes. What I expect is that the model summary output will show that the model for the Afghanistan dataset will have a coefficient for year rather than pop.

by_country2 <- by_country %>%
  mutate(model = pmap(list(data, continent, country), country_model),
         modelsum = map(model, tidy)) %>%
  unnest(modelsum, .drop = TRUE)

by_country2

My output showing that the coefficient for Afghanistan is pop, not year.

A tibble: 284 × 7
       country continent        term     estimate    std.error  statistic      p.value
        <fctr>    <fctr>       <chr>        <dbl>        <dbl>      <dbl>        <dbl>
1  Afghanistan      Asia (Intercept) 2.834615e+01 2.314395e+00  12.247758 2.410050e-07
2  Afghanistan      Asia         pop 5.771517e-07 1.343425e-07   4.296121 1.570999e-03
3      Albania    Europe (Intercept) 4.963274e+01 1.935933e+00  25.637630 1.871817e-10
4      Albania    Europe         pop 7.286188e-06 7.171585e-07  10.159802 1.374311e-06
5      Algeria    Africa (Intercept) 3.565187e+01 1.632853e+00  21.834099 9.087006e-10
6      Algeria    Africa         pop 1.176242e-06 7.588190e-08  15.500960 2.548769e-08
7       Angola    Africa (Intercept) 2.855043e+01 1.922225e+00  14.852803 3.843692e-08
8       Angola    Africa         pop 1.276860e-06 2.482137e-07   5.144195 4.351004e-04
9    Argentina  Americas (Intercept) 5.323586e+01 3.784907e-01 140.653008 8.102227e-18
10   Argentina  Americas         pop 5.532629e-07 1.282987e-08  43.123018 1.079775e-12
# ... with 274 more rows

What is bizarre to me is that when I use only 1 condition in my function if statement, then it seems to work perfectly:

country_model <- function(df, cont) {

  if(cont == "Asia") { # Only 1 condition

    lm(lifeExp ~ year, data = df)

  } else {

    lm(lifeExp ~ pop, data = df)

  }
}

by_country2 <- by_country %>%
  mutate(model = map2(data, continent, country_model),
         modelsum = map(model, tidy)) %>%
  unnest(modelsum, .drop = TRUE)

by_country2

# A tibble: 284 × 7
       country continent        term      estimate    std.error  statistic      p.value
        <fctr>    <fctr>       <chr>         <dbl>        <dbl>      <dbl>        <dbl>
1  Afghanistan      Asia (Intercept) -5.075343e+02 4.048416e+01 -12.536613 1.934055e-07
2  Afghanistan      Asia        year  2.753287e-01 2.045093e-02  13.462890 9.835213e-08
3      Albania    Europe (Intercept)  4.963274e+01 1.935933e+00  25.637630 1.871817e-10
4      Albania    Europe         pop  7.286188e-06 7.171585e-07  10.159802 1.374311e-06
5      Algeria    Africa (Intercept)  3.565187e+01 1.632853e+00  21.834099 9.087006e-10
6      Algeria    Africa         pop  1.176242e-06 7.588190e-08  15.500960 2.548769e-08
7       Angola    Africa (Intercept)  2.855043e+01 1.922225e+00  14.852803 3.843692e-08
8       Angola    Africa         pop  1.276860e-06 2.482137e-07   5.144195 4.351004e-04
9    Argentina  Americas (Intercept)  5.323586e+01 3.784907e-01 140.653008 8.102227e-18
10   Argentina  Americas         pop  5.532629e-07 1.282987e-08  43.123018 1.079775e-12
# ... with 274 more rows

I'm not sure if my problem is with pmap() or my if statement.

Answer 1

This is related to this GitHub issue.

It appears pmap is sending through continent and country as numerics, as can be confirmed by putting a print statement in your function.

test_fun <- function(df, cont, xx) {
  print(paste(cont, xx))
}

temp <-by_country %>%
  mutate(model = pmap(list(data, continent, country), test_fun))

Prints:

[1] "3 1"
[1] "4 2"
[1] "1 3"
[1] "1 4"
[1] "2 5"
[1] "5 6"
[1] "4 7"
[1] "3 8"
[1] "3 9"
etc

This doesn't happen in map2, and therefore your second try does work.

Coercing to character solves the issue:

by_country %>%
  mutate(model = pmap(list(data, as.character(continent), as.character(country)), country_model),
         modelsum = map(model, broom::tidy)) %>%
  unnest(modelsum, .drop = TRUE)

# A tibble: 284 x 7
       country continent        term      estimate    std.error  statistic      p.value
        <fctr>    <fctr>       <chr>         <dbl>        <dbl>      <dbl>        <dbl>
 1 Afghanistan      Asia (Intercept) -5.075343e+02 4.048416e+01 -12.536613 1.934055e-07
 2 Afghanistan      Asia        year  2.753287e-01 2.045093e-02  13.462890 9.835213e-08
 3     Albania    Europe (Intercept)  4.963274e+01 1.935933e+00  25.637630 1.871817e-10
 4     Albania    Europe         pop  7.286188e-06 7.171585e-07  10.159802 1.374311e-06
 5     Algeria    Africa (Intercept)  3.565187e+01 1.632853e+00  21.834099 9.087006e-10
 6     Algeria    Africa         pop  1.176242e-06 7.588190e-08  15.500960 2.548769e-08
 7      Angola    Africa (Intercept)  2.855043e+01 1.922225e+00  14.852803 3.843692e-08
 8      Angola    Africa         pop  1.276860e-06 2.482137e-07   5.144195 4.351004e-04
 9   Argentina  Americas (Intercept)  5.323586e+01 3.784907e-01 140.653008 8.102227e-18
10   Argentina  Americas         pop  5.532629e-07 1.282987e-08  43.123018 1.079775e-12
# ... with 274 more rows