cp_wrapper: Find change points in a sequence of choices based on given...
In ontogenerator/cpdetectoR: Change point estimation from sequences of choices

Description Usage Arguments Value References Examples

Find change points in a sequence of choices based on given test and decision criterion

1	cp_wrapper(input, isDiscrete, test, Crit)

`input`	A vector of correct or incorrect decisions, coded as 1s and 0s
`isDiscrete`	A boolean parameter, true if the data to be analyzed are discrete, false if continuous, e.g. succesive intervals
`test`	Name of the test to be performed. The following tests are implemented: "binomial" (random rate), "chisquare", "ttest", "KS" (Komolgorov-Smirnov). The binomial test is used for finding changes in the rate parameter of a random rate process, either intermittently (isDiscrete = TRUE) or continuously (isDiscrete = FALSE) sampled. The chi square test is used with data where there has been a frequency change, as in, e.g. correct choices. The t test should be used for normally distributed data. The Komolgorov-Smirnof test is used for data with nonnormal or unknown distribution.
`Crit`	A real-valued decision criterion on logit. Values recommended by Gallistel et al. (2004) are between 1.3 and 6, which correspond to p values of 0.05 and 0.000001, respectively. The values are the logarithms of the odds against the null (no-change) hypothesis. logit = log[(1-p)/p], where p is the desired significance level

A data frame with the number of Trials (Trial) for each change point detected, the cumulative number of correct responses, or successes, (CumSs), and the ratio of correct/total choices (Slopes) for the trials since the last change point and the current change point

Gallistel CR, Fairhurst S, Balsam P (2004) The learning curve: Implications of a quantitative analysis. PNAS 101:13124-13131. doi: 10.1073/pnas.0404965101

# generate dummy data with a change point at trial 120
input <- c(rbinom(120, 1, 0.5), rbinom(200, 1, 0.8))
cp_wrapper(input, TRUE, "chisquare", 3)
cp_wrapper(input, TRUE, "binomial", 3)

# use included eyeblink data set:
eyeblink[,] # inspect data set
cp_wrapper(eyeblink, TRUE, "chisquare", 3)
# using small criterion, e.g. 2, fails due to computational problems
# better use "binomial" instead
cp_wrapper(eyeblink, TRUE, "binomial", 2)

# use included plusmaze data set:
plusmaze[,] # inspect data set
(cp.1 <- cp_wrapper(plusmaze, TRUE, "binomial", 1.3))
# decrease sensitivity and detect different change points
(cp.2 <- cp_wrapper(plusmaze, TRUE, "binomial", 2))
# the chisquare test detects more change points even with higher criterion
(cp.3 <- cp_wrapper(plusmaze, TRUE, "chisquare", 2))

# plotting data with ggplot2
library(ggplot2)
# first generate data.frame with cumulative responses:
my.data <- data.frame(Trial = 1:length(plusmaze[,]),
CumRespMeasure = cumsum(plusmaze)[,])
# plot cumulative learning curve with change points
ggplot(my.data) + geom_line(aes(Trial,CumRespMeasure)) +
 geom_point(data = cp.1, aes(Trial, CumSs), shape = 1, size = 6, color = "blue") +
 geom_point(data = cp.2, aes(Trial, CumSs), shape = 2, size = 3, color = "green") +
 geom_point(data = cp.3, aes(Trial, CumSs), shape = 3, size = 3, color = "red")

# plot average response rate per trial
ggplot() + geom_step(data = cp.1, aes(Trial, Slopes)) +
 ylab("Average Response Rate per Trial")