functions

Predefined functions

quickpsy incorporates 3 predefined functions to characterize the shape of the psychometric function: cum_normal_fun, logistic_fun and weibull_fun.

Let’s see, for example, the definition of the logistic_fun

library(quickpsy)
logistic_fun
#> function (x, p) 
#> (1 + exp(-p[2] * (x - p[1])))^(-1)
#> <bytecode: 0x5a54127c05f0>
#> <environment: namespace:quickpsy>

By default, quickspy fits cum_normal_fun, but you can change the function using the fun argument

fit <- quickpsy(qpdat, phase, resp, grouping = c("participant", "cond"), 
                fun = logistic_fun, bootstrap = "none") 

plot(fit, color = cond)

You can see that it is not necessary to include initial parameters to fit a predefined function.

Functions defined by the user

In the fun argument you can incorporate any function. For non-predefined functions you need to include initial parameters.

gompertz_fun <- function(x, p) exp(-p[1] * exp(-p[2] * x))

fit <- quickpsy(qpdat, phase, resp, grouping = c("participant", "cond"), 
                fun = gompertz_fun, parini = c(0.1, 0.02), bootstrap = "none") 

plot(fit)

The function does not need to be monotonic

n <- 100
x <- seq(-.3, .3, .1) 
k <- c(10, 45, 90, 100, 60, 10, 5)
dat <- data.frame(x, k, n)

gaussian <- function(x, p) exp(-0.5*((x - p[2]) / p[3])^2) / (1 + exp(-p[1]))

fit <- quickpsy(dat, x, k, n, fun = gaussian, parini = c(0, 0, 0), 
                thresholds = FALSE, # because the threshold is not well defined
                bootstrap = "none") 
plot(fit)

Data frame of functions (dev version of quickpsy)

In the fun argument you can also incorporate a data frame of functions specifying with function you want to fit to every group (the data frame should be a tibble to support lists of functions).

One advantage of this approach is that you can easily specify shared parameters across conditions, such as the slope parameter in the following example using the logistic function

library(dplyr)

fun_df <- tibble(cond = c("cond1", "cond2"), 
                 fun = c(function(x, p) (1 + exp(-p[2] * (x - p[1])))^(-1), 
                         function(x, p) (1 + exp(-p[2] * (x - p[3])))^(-1)))

fit <- quickpsy(qpdat, phase, resp, grouping = c("participant", "cond"), 
                fun = fun_df, 
                parini = c(-100, 0.02, -100), 
                bootstrap = "none") 

plot(fit, color = cond)

This is another example using a shared leftward and rightward asymptote across conditions using the cumulative normal function

library(dplyr)

fun_df <- tibble(cond = c("cond1", "cond2"), 
                 fun = c(function(x, p) p[5] + (1 - p[5] - p[6]) * pnorm(x, p[1], p[2]), 
                         function(x, p) p[5] + (1 - p[5] - p[6]) * pnorm(x, p[3], p[4])))

fit <- quickpsy(qpdat, phase, resp, grouping = c("participant", "cond"), 
                fun = fun_df, 
                parini = c(-100, 50, -100, 50, 0.01, 0.01), 
                bootstrap = "none") 
#> Warning: 1 group has a Hessian that is not positive-definite (a parameter at a
#> bound or a flat/ill-conditioned likelihood); asymptotic confidence intervals
#> for it are NA. Use bootstrap = "parametric" to obtain confidence intervals for
#> these.

plot(fit, color = cond)

For a given data set you can try different models and then compare them using likelihood ratio tests or the Akaike information criterion (see model_selection section).