README.md

In thibault-gajdos/RTconflict: What the Package Does (One Line, Title Case)

Contents

1. Overview
2. Data
3. ELF
4. Interference Distribution Index

Overview

This package provides tools to analyse response time (RT) distributions for conflict tasks in R. Specifically, it allows to compute:

• the Error Location Function and the Error Location Index proposed in Servant, M., Gajdos, T., & Davranche, K. (2018). ELF: A new measure of response capture. Psychonomic Bulletin & Review, 1-9.

• Interference Distribution Index proposed in Gajdos, T., Servant, M., Hasbroucq, T., & Davranche, K. (2019, december 20). A measure of the Interference Effect Distribution. https://doi.org/10.31234/osf.io/7f5n8

To install, use the following command:

devtools::install_github("thibault-gajdos/RTconflict")

Data

We use simulated data for a conflict task.

There are 20 subjects (from 1 to 20), performing a conflict task with 200 congruent and 200 incongruent trials, in two distinct conditions (A and B). Therefore, there is 800 trials by subject.

data.conflict %>%
    group_by(compatible, condition) %>%
    summarise(acc = mean(acc), rt = mean(rt))

# A tibble: 4 x 4
# Groups:   compatible [2]
  compatible condition   acc    rt
  <fct>      <fct>     <dbl> <dbl>
1 c          A         0.876 0.494
2 c          B         0.880 0.493
3 i          A         0.878 0.516
4 i          B         0.880 0.515

Error Location Function

We start with a single subject (subject 1). Error Location Functions are particularly meaningfull for incongruent trials. With thus restrict attention to these trials.

First, we compute ELF in each condition.

d1A <- data.conflict %>% filter(sujet == 1, condition == 'A', compatible == 'i')
p.1A <- c(1:nrow(d1A))/nrow(d1A)
elf.1A <- elf(d1A$rt, d1A$acc)
df1A <- data.frame(elf = elf.1A, p = p.1A)  %>%
    mutate(condition = 'A')

d1B <- data.conflict %>% filter(sujet == 1, condition == 'B', compatible == 'i')
elf.1B <- elf(d1B$rt, d1B$acc)
p.1B <- c(1:nrow(d1B))/nrow(d1B)
df1B <- data.frame(elf = elf.1B, p = p.1B) %>%
    mutate(condition = 'B')

df1 <- rbind(df1A, df1B)

We now plot ELF.

plot.elf1 <- ggplot(data = df1, aes(x = p, y = elf, group = condition, colour = condition)) +
    geom_line() +
    xlab('p') +
    ylab('% of total errors') + 
    ggtitle('Error Location Function')
plot.elf1

The Error Location Index (ELI) is the area under ELF.

eli(elf.1A)

[1] 0.6492

eli(elf.1B)

[1] 0.6522222

So, as suggested by the ELF plots, response capture seems larger in condition B than in condition A. Now, we will turn to aggregate ELF accross subects.

dA <- data.conflict %>% filter(compatible == 'i', condition == 'A')
elf.A <- elf.g(dA$rt, dA$acc, dA$sujet)

dB <- data.conflict %>% filter(compatible == 'i', condition == 'B')
elf.B <- elf.g(dB$rt, dB$acc, dB$sujet)

We can now plot these aggregated ELF functions.

d <- data.frame(p = c(0:100,0:100), elf = c(elf.A, elf.B), condition = c(rep('A',101),rep('B',101)))

plot.elf <- ggplot(data = d, aes(x = p, y = elf, group = condition, colour = condition)) +
    geom_line() +
    xlab('p') +
    ylab('% of total errors') + 
    ggtitle('Aggregated Error Location Function')
plot.elf

Interference Distribution Index

To study inhibition, we focus on correct trials. We again start by focusing on one subject, and this time consider only one condition (A). We compute the Interference Distribution Index.

data1A <- data.conflict %>% filter(sujet == 1, condition == 'A', acc == 1)
LD.1A <- lorenz(data1A$rt, data1A$compatible)
glimpse(LD.1A)

List of 2
 $ inhib: num 0.755
 $ q    :'data.frame':  175 obs. of  6 variables:
  ..$ p    : num [1:175] 0.00571 0.01143 0.01714 0.02286 0.02857 ...
  ..$ qc   : num [1:175] 0.327 0.331 0.334 0.338 0.338 ...
  ..$ qi   : num [1:175] 0.365 0.372 0.372 0.382 0.384 ...
  ..$ delta: num [1:175] 0.0377 0.0406 0.0382 0.0443 0.046 ...
  ..$ m    : num [1:175] 0.346 0.352 0.353 0.36 0.361 ...
  ..$ c    : num [1:175] 0.00676 0.01405 0.02091 0.02887 0.03713 ...

As an option, one can specify the quantile method by setting "type" to a value between 1 and 9, 7 be the default. See the documentation of the quantile function in R for further details.

LD.1A contains two parts:

• inhib which is the Lorenz-inhibition index

• q which contains:

  • p: quantiles

  • qc: quantile function for compatible trials

  • qi: quantile function for incompatible trials

  • delta = (qi-qc)/2

  • m = (qc+qi)/2

  • c = cumulative function of delta

Let's plot the Lorenz-interference plot.

plot.1A <- ggplot(data = LD.1A$q, aes(x = p, y = c)) +
    geom_line(size = 1) +
    xlab('p') +
    ylab('% D(p)') +
    ggtitle('Lorenz-inhibition plot') 
plot.1A

Now, let compute an aggregate Lorenz-inhibition plot accross subjects for each condition. First we compute Lorenz-inhibition plots for each subject and each condition. Note that it might be the case that different subjects or conditions have different numbers of quantiles. To solve this, we constrain the number of quantiles by grouping quantiles by 1%.

x  <- NULL 
for (s in unique(data.conflict$sujet)){
    for (cond in unique(data.conflict$condition)){ 
    d  <- data.conflict %>% filter(sujet == s, condition == cond, acc == 1) 
    xx  <- lorenz(d$rt, d$compatible)$q %>%
                                   mutate(p  = round(p, digits = 2)) %>%
                                   group_by(p) %>% 
                                   summarise(c = mean(c)) %>%
                                   mutate(sujet = s, condition = cond)
    x  <- rbind(x, xx)
    }
}

Now we aggregate these Lorenz-Delta plots together

x.ag  <- x %>%
    group_by(p, condition) %>%
    summarise(c = mean(c))

and plot them.

plot  <- ggplot(data = x.ag, aes(x = p, y = c, group = condition, colour = condition)) +
    geom_line(size = 1) +
    xlab('p') +
    ylab('% D(p)') +
    ggtitle('aggregated  Lorenz-inhibition plot')
plot

data.inhib computes delta plots and the Inhibition Distribution Index for each subject and each condition.

data.inhib <- data.conflict %>% filter(acc == 1)
inhib  <- inhib.delta(rt = data.inhib$rt, comp = data.inhib$compatible, sujet = data.inhib$sujet, cond = data.inhib$condition)

The function generates three data frame: - i: Inhibition Distribution Index index by subject and condition - delta.slope: slope of the last segment of the delta plot by subject and condition. This is just for information, as we strongly advice against the use of this measure.

glimpse(inhib)

List of 2
 $ i          :'data.frame':    40 obs. of  3 variables:
  ..$ sujet: Factor w/ 20 levels "1","2","3","5",..: 1 1 2 2 3 3 4 4 5 5 ...
  ..$ cond : Factor w/ 2 levels "A","B": 1 2 1 2 1 2 1 2 1 2 ...
  ..$ index: num [1:40] 0.755 0.603 1.483 0.645 0.853 ...
 $ delta.slope:'data.frame':    40 obs. of  3 variables:
  ..$ sujet: Factor w/ 20 levels "1","2","3","4",..: 1 10 11 12 13 14 15 16 17 18 ...
  ..$ slope: num [1:40] -0.5493 0.0814 0.0715 -0.4077 -0.7678 ...
  ..$ cond : chr [1:40] "A" "A" "A" "A" ...

The mean of the Inhibition Distribution Index index is

inhib$i %>%  group_by(cond) %>%
    summarise(index = mean(index))

# A tibble: 2 x 2
  cond  index
  <fct> <dbl>
1 A     0.693
2 B     0.659

thibault-gajdos/RTconflict documentation built on March 2, 2020, 6:12 p.m.