--- title: "Ggplot Statistical transformation objects" output: html_notebook --- ```{r message=FALSE} library(gapminder) library(tidyverse) ``` # Geoms to display summarizing statistics They always take an x and compute and plot the corresponding summarizing y-values for it: central value and dispersion of some sort, or just one of these. You have two options to say what values the geom should render: - supply individual functions to compute the lower and higher dispersion ranges, and a function to compute the central value. Argument slots to fill: `fun`, `fun.min`, `fun.max`. - supply a summary function that returns both/all of them. Its output must be a named vector. Argument slot to fill: `fun.data`. You either have to write these functions yourself, or you can use a few well-matching ones from the `Hmisc` package or wrappers around them created for ggplot2. _Examples of summary functions_ ### `geom_errorbar` and `geom_linerange` Just ranges, without the central value __Compute standard error of the mean lifeExpectancy for each continent in each year. Render them as errorbars (i.e. without the mean).__ ```{r} gapminder %>% ggplot(aes(x = factor(year), y = lifeExp, color = continent)) + geom_point(alpha = 0.2, position = position_jitterdodge(jitter.width = 0.7, dodge.width = 0.7)) + stat_summary(fun.data = "mean_se", geom = "errorbar", size = 1) ``` __Compute standard deviation of the life expectancy for each continent in each year using linerange (i.e. without the mean)__ Use the `ggplot2::mean_sd` function. NB: it is a wrapper around the `Hmisc::smean_sdl` function. It is documented below. By default, the range represents standard error times two (double length). To alter this, one has to use the `mult` parameter just like in the original`Hmisc::smean_sdl` function. Look at the way `stat_summary` inputs these arguments: `fun.args = list()`. In fun.data, preferably use these functions: ``` Usage mean_cl_boot(x, ...) mean_cl_normal(x, ...) mean_sdl(x, ...) median_hilow(x, ...) Arguments x a numeric vector ... other arguments passed on to the respective Hmisc function. Value A data frame with columns y, ymin, and ymax. ``` These are wrappers around some summary function from the `Hmisc package`, and they use Hmisc's functions' parameters. Documentation of these functions follows below. ``` Usage smean.cl.normal(x, mult=qt((1+conf.int)/2,n-1), conf.int=.95, na.rm=TRUE) smean.sd(x, na.rm=TRUE) smean.sdl(x, mult=2, na.rm=TRUE) smean.cl.boot(x, conf.int=.95, B=1000, na.rm=TRUE, reps=FALSE) smedian.hilow(x, conf.int=.95, na.rm=TRUE) ``` These cannot be used directly as `fun.data`, since their output is different from the dataframe with column names `y`, `y.min`, and `y.max`. Example: ```{r message=FALSE} library(Hmisc) library(magrittr) c(1,1,1,10,10,10,10) %>% Hmisc::smean.cl.normal() %T>% str() #mean and confidence intervals ``` As seen above, the resulting output is a named vector, with names different from `y`, `y.min`, `y.max`. In all summary functions, we can supply either `fun.data`, or functions for each statistics separately. These arguments are called `fun` (the central value), `fun.min` (the lower dispersion value), and `fun.max` (the upper dispersion value). ```{r} gapminder %>% ggplot(aes(x = factor(year), y = lifeExp, color = continent)) + geom_point(alpha = 0.2, position = position_jitterdodge(jitter.width = 0.7, dodge.width = 0.7)) + stat_summary(fun.data = "mean_sdl", fun.args = list(mult = 1) , geom = "linerange", size = 0.7, position = position_dodge(width = 0.8)) ``` ###`geom_crossbar` and `geom_pointrange` These include the central value ```{r} gapminder %>% ggplot(aes(x = factor(year), y = lifeExp, color = continent)) + geom_point(alpha = 0.2, position = position_jitterdodge(jitter.width = 0.7, dodge.width = 0.7)) + stat_summary(fun.data = "mean_se", geom = "crossbar", size = 0.7) ``` The ```{r} gapminder %>% ggplot(aes(x = factor(year), y = lifeExp, color = continent)) + geom_point(alpha = 0.2, position = position_jitterdodge(jitter.width = 0.7, dodge.width = 0.7)) + stat_summary(fun.data = "mean_sdl", fun.args = list(mult = 1), geom = "pointrange", position = position_dodge(width = 0.8), size = 0.5) ``` With `fun`, `fun.min`, and `fun.max`: you have to write your own functions first :-/ . ```{r} low_f <- function(x) {quantile(x, probs = 0.25)} hi_f <- function(x) {quantile(x, probs = 0.75)} gapminder %>% ggplot(aes(x = factor(year), y = lifeExp, color = continent)) + geom_point(alpha = 0.2, position = position_jitterdodge(jitter.width = 0.7, dodge.width = 0.7)) + stat_summary(fun = "median", fun.min = "low_f", fun.max = "hi_f", geom = "pointrange", position = position_dodge(width = 0.8), size = 0.5) ``` ### `geom_smooth`, `stat_smooth` ```{r} gapminder %>% ggplot(aes(x = gdpPercap, y = lifeExp, color = continent)) + geom_point() + geom_smooth(method = "lm") ``` ### `geom_quantile`, `stat_quantile` ```{r} gapminder %>% filter(year > 1995) %>% ggplot(aes(x = gdpPercap, y = lifeExp)) + geom_point(alpha = 0.3) + geom_quantile(formula = y ~ x ,quantiles = c(0.01, 0.25, 0.5, 0.75), aes(color = factor(..quantile..)), size = 2) + geom_smooth( formula = y ~ x , method = "lm", color = "black", linetype = 2, se = FALSE)# + #facet_wrap(~continent) ``` # `stat_function` ```{r} gapminder %>% filter(continent == "Europe", year > 2000) %>% ggplot(aes(x = gdpPercap)) + geom_density() + stat_function(fun = dnorm, color = "red", args = list(mean = mean(filter(gapminder, continent == "Europe", year > 2000)$gdpPercap), sd = sd(filter(gapminder, continent == "Europe", year > 2000)$gdpPercap))) ```