# curve_compare

0th

Percentile

##### Compare Two Functions and Produces An AUC Score

Compares the p-value/s-value, and likelihood functions and computes an AUC number.

##### Usage
curve_compare(data1, data2, type = "c", plot = TRUE, ...)
##### Arguments
data1

The first dataframe produced by one of the interval functions in which the intervals are stored.

data2

The second dataframe produced by one of the interval functions in which the intervals are stored.

type

Choose whether to plot a "consonance" function, a "surprisal" function or "likelihood". The default option is set to "c". The type must be set in quotes, for example curve_compare (type = "s") or curve_compare(type = "c"). Other options include "pd" for the consonance distribution function, and "cd" for the consonance density function, "l1" for relative likelihood, "l2" for log-likelihood, "l3" for likelihood and "d" for deviance function.

plot

by default it is set to TRUE and will use the plot_compare() function to plot the two functions.

...

Can be used to pass further arguments to plot_compare().

##### Value

Computes an AUC score and returns a plot that graphs two functions.

plot_compare()

ggcurve()

curve_table()

##### Aliases
• curve_compare
##### Examples
# NOT RUN {
library(concurve)
GroupA <- rnorm(50)
GroupB <- rnorm(50)
RandomData <- data.frame(GroupA, GroupB)
intervalsdf <- curve_mean(GroupA, GroupB, data = RandomData)
GroupA2 <- rnorm(50)
GroupB2 <- rnorm(50)
RandomData2 <- data.frame(GroupA2, GroupB2)
model <- lm(GroupA2 ~ GroupB2, data = RandomData2)
randomframe <- curve_gen(model, "GroupB2")
curve_compare(intervalsdf[[1]], randomframe[[1]])
curve_compare(intervalsdf[[1]], randomframe[[1]], type = "s")
# }
# NOT RUN {
# }

Documentation reproduced from package concurve, version 2.5.0, License: GPL-3 | file LICENSE

### Community examples

Looks like there are no examples yet.