gam in geom_smooth doesn't use s() when manually specified:

[Ax3man]

Consider:

ggplot(diamonds, aes(price, carat)) + geom_smooth()

##`geom_smooth()` using method = 'gam'

But with manual specification of the method:

ggplot(diamonds, aes(price, carat)) + geom_smooth(method = "gam")

Something seems to off with the default formula, since

ggplot(diamonds, aes(price, carat)) + geom_smooth(method = "gam", formula = y ~ s(x, bs = "cs"))

produces same as the first plot. Is this intentional? For what it is worth, this behavior was already there in v1.0.0.

[hadley]

Will be fixed by #1951

[tjmahr]

This particular issue still happens.

library(ggplot2)

iris2 <- iris[sample(seq_len(nrow(iris)), 1000, replace = TRUE), ]

ggplot(iris2) + 
  aes(x = Sepal.Length, y = Sepal.Width) + 
  stat_smooth() + 
  ggtitle("stat_smooth()")

# gam
ggplot(iris2) + 
  aes(x = Sepal.Length, y = Sepal.Width) + 
  stat_smooth(method = "gam") + 
  ggtitle("stat_smooth(method = \"gam\")")

# gam with formula
ggplot(iris2) + 
  aes(x = Sepal.Length, y = Sepal.Width) + 
  stat_smooth(method = "gam", formula = y ~ s(x, bs = "cs"))+ 
  ggtitle("stat_smooth(method = \"gam\", formula = y ~ s(x, bs = \"cs\")")

 package     * version    date       source                            
 ggplot2     * 2.2.1.9000 2018-05-29 Github (tidyverse/ggplot2@4299917)

[batpigandme]

Isn't it applying the same formula?

library(ggplot2)

iris2 <- iris[sample(seq_len(nrow(iris)), 1000, replace = TRUE), ]

ggplot(iris2) + 
  aes(x = Sepal.Length, y = Sepal.Width) + 
  stat_smooth() + 
  ggtitle("stat_smooth()")
#> `geom_smooth()` using method = 'gam' and formula 'y ~ s(x, bs = "cs")'

# gam
ggplot(iris2) + 
  aes(x = Sepal.Length, y = Sepal.Width) + 
  stat_smooth(method = "gam") + 
  ggtitle("stat_smooth(method = \"gam\")")

# gam with formula
ggplot(iris2) + 
  aes(x = Sepal.Length, y = Sepal.Width) + 
  stat_smooth(method = "gam", formula = y ~ s(x, bs = "cs"))+ 
  ggtitle("stat_smooth(method = \"gam\", formula = y ~ s(x, bs = \"cs\")")

Created on 2018-06-01 by the reprex package (v0.2.0).

[tjmahr]

The idea is that the default formula used should change to a smooth when “gam” is given.

[batpigandme]

Oh, I see what you're saying. I thought the problem was that the first and last were the same – what you're saying is that the behaviour that occurs when no method is specified should occur when "gam" is given, and it's currently happening when either no method is given or gam and formula are given.

[hadley]

This behaviour is correct; we report the formula but do not automatically change it. That's up to you.

[clauswilke]

Maybe it would help to be a bit more explicit in the documentation about how to manually switch to gam for arbitrary sample sizes. E.g.:

If you want to use the default gam model for a dataset with fewer than 1,000 observations, set method = "gam", formula = y ~ s(x, bs = "cs").

I also noticed that the documentation still says formula = y ~ mgcv::s(x, bs = "cs") and that needs to be modified to formula = y ~ s(x, bs = "cs").

[lock]

This old issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with reprex) and link to this issue. https://reprex.tidyverse.org/