Flanker task

In a flanker task, participants identify a central stimulus (as quickly and accurately) as possible, while ignoring distracting stimuli presented on the left or right of the central stimulus (the flankers).

For example, the stimulus could be “HHH”, and the correct response would be H. This is called a compatible (or congruent) stimulus because the flanking Hs are the same as the central stimulus. Alternatively, the stimulus could be “HSH”, and the correct resposne would be S. This is called an incompatible (or incongruent) stimulus because the flanking Hs are different from the central stimulus.

The data for this assignment come from a flanker task where participants responded to many flanker stimuli over several trials.

Load the data and libraries you will use

I will help you with some sample code that compiles all of the text files in a single long format data frame.

The data is contained in this .zip file: FlankerData.zip

The code chunk below assumes that you have placed the folder FlankerData into your R project folder.

library(data.table)
library(dplyr)
library(ggplot2)
library(bit64)


# get the file names
file_names <- list.files(path="FlankerData")

# create headers for each column
the_headers <- c("stimulus","congruency","proportion",
                 "block","condition","dualtask","unknown",
                 "stimulus_onset","response_time","response","Subject")
# Load data
# create empty dataframe
all_data<-data.frame()

# loop to add each file to the dataframe
for(i in file_names){
  one_subject <- fread(paste("FlankerData/",i, sep=""))
  names(one_subject) <- the_headers
  one_subject$Subject <- rep(i,dim(one_subject)[1])
  one_subject <- cbind(one_subject, trial= 1:dim(one_subject)[1])
  all_data <- rbind(all_data,one_subject)
}

Pre-processing

get accuracy for each trial

A correct response occurs when the letter in the response column is the same as the letter in the middle position of item in the stimulus column. Create an accuracy column that codes whether the response was correct or incorrect on each trial (coding can be TRUE/FALSE, 0/1, or some other coding scheme that identifies correct vs incorrect)

i <- 1:length(all_data)
for (stim in i){
  stim<- all_data$stimulus 
  #stim <- strsplit(stim, split = '')
  stim <- substr(stim, 2, 2) 
  stim<-tolower(stim)
}
all_data <- cbind(all_data,stim)

new_all<- all_data %>%
  mutate(accuracy = response==stim)

Get Reaction time on each trial

The stimulus_onset column gives a computer timestamp in milliseconds indicating when the stimulus was presented. The response_time column is a timestamp in milliseconds for the response. The difference between the two (response_time - stimulus_onset) is the reaction time in milliseconds. Add a column that calculates the reaction time on each trial.

**tip:** notice that the numbers in response_time and stimulus_onset have the class integer64. Unfortunately, ggplot does not play nice with integers in this format. you will need to make sure your RT column is in the class integer or numeric.

new_all<- new_all %>%
  mutate(ReactionTime = (response_time - stimulus_onset))

Checks

Check how many trials each subject completed in the congruent and incongruent conditions, the mean accuracy for each subject in each congruency condition, and the mean RT for each subject in each congruency condition.

check<- new_all %>%
  mutate(Subject= as.factor(Subject),
         congruency= as.factor(congruency)) %>%
  group_by(Subject,congruency) %>%
  summarise(num_trials = length(congruency),
            mean_accuracy = mean(accuracy),
            mean_RT = mean(ReactionTime))
            
knitr::kable(check)
Subject congruency num_trials mean_accuracy mean_RT
1.txt C 96 0.9166667 550
1.txt I 96 0.9270833 548
10.txt C 96 0.9479167 1075
10.txt I 96 0.9166667 1140
11.txt C 96 0.9375000 708
11.txt I 96 0.9583333 852
12.txt C 96 0.9270833 622
12.txt I 96 0.0833333 682
13.txt C 96 0.8958333 545
13.txt I 96 0.8229167 598
14.txt C 96 0.9687500 719
14.txt I 96 0.9375000 742
15.txt C 96 0.9895833 631
15.txt I 96 0.9791667 689
16.txt C 96 0.9583333 572
16.txt I 96 0.9687500 584
17.txt C 96 0.9687500 633
17.txt I 96 0.9479167 620
18.txt C 96 1.0000000 802
18.txt I 96 0.9583333 817
19.txt C 96 0.9791667 1002
19.txt I 96 0.9895833 1105
2.txt C 96 1.0000000 1002
2.txt I 96 0.9583333 1008
20.txt C 96 0.9895833 669
20.txt I 96 1.0000000 690
21.txt C 96 1.0000000 840
21.txt I 96 1.0000000 904
22.txt C 96 0.9687500 795
22.txt I 96 0.9479167 713
3.txt C 96 0.9895833 812
3.txt I 96 0.9687500 803
4.txt C 96 0.9895833 815
4.txt I 96 0.9791667 901
5.txt C 96 0.9791667 819
5.txt I 96 0.9687500 941
6.txt C 96 0.9687500 667
6.txt I 96 0.9687500 688
7.txt C 96 0.9895833 1053
7.txt I 96 1.0000000 1146
8.txt C 96 0.8645833 611
8.txt I 96 0.9895833 632
9.txt C 96 0.9687500 695
9.txt I 96 0.9583333 776

Exclusion

It is common to exclude Reaction times that are very slow. There are many methods and procedures for excluding outlying reaction times. To keep it simple, exclude all RTs that are longer than 2000 ms

excluded <- new_all %>%
  filter(ReactionTime<2000)

Analysis

Reaction Time analysis

  1. Get the individual subject mean reaction times for correct congruent and incongruent trials.
RT_analysis <- excluded %>%
  mutate(Subject= as.factor(Subject),
         congruency= as.factor(congruency)) %>%
  filter(accuracy == TRUE) %>%
  group_by(Subject,congruency) %>%
  summarise(sub_mean = mean(ReactionTime))

knitr::kable(RT_analysis)
Subject congruency sub_mean
1.txt C 556
1.txt I 551
10.txt C 898
10.txt I 986
11.txt C 714
11.txt I 826
12.txt C 612
12.txt I 567
13.txt C 531
13.txt I 635
14.txt C 661
14.txt I 721
15.txt C 631
15.txt I 690
16.txt C 571
16.txt I 582
17.txt C 619
17.txt I 622
18.txt C 802
18.txt I 810
19.txt C 984
19.txt I 1043
2.txt C 919
2.txt I 952
20.txt C 671
20.txt I 690
21.txt C 840
21.txt I 884
22.txt C 747
22.txt I 746
3.txt C 811
3.txt I 809
4.txt C 815
4.txt I 844
5.txt C 784
5.txt I 882
6.txt C 667
6.txt I 691
7.txt C 1024
7.txt I 1076
8.txt C 601
8.txt I 633
9.txt C 695
9.txt I 779
  1. Get the overall mean RTs and SEMs (standard errors of the mean) for the congruent and incongruent condition. Make a table and graph.
overall <- RT_analysis %>%
  group_by(congruency)%>%
  summarise(mean_RT = mean(sub_mean),
            SEM = sd(sub_mean)/sqrt(length(sub_mean))) %>%
  mutate(mean_RT = as.numeric(mean_RT))
  
library(xtable)
knitr::kable(xtable(overall))
congruency mean_RT SEM
C 734 29.80150
I 773 32.58174 
ggplot(overall, aes(x=congruency, y=mean_RT, fill=congruency))+
  geom_bar(stat="identity")+
  theme_classic()+
  geom_errorbar(aes(ymin=mean_RT - SEM,
                    ymax = mean_RT + SEM),
                position = position_dodge(width = .8), width = .4)

  1. Compute the flanker effect for each subject, taking the difference between their mean incongruent and congruent RT. Then plot the mean flanker effect, along with the SEM of the mean flanker effect

**tip:** Not all problems have an easy solution in dplyr, this is one them. You may have an easier time using logical indexing of the dataframe to solve this part.

Sub_flanker <- as.numeric(check[check$congruency=="I",]$mean_RT - check[check$congruency=="C",]$mean_RT)

Sub_means <- mean(Sub_flanker)
Sub_SEM <- sd(Sub_flanker)/sqrt(length(Sub_flanker))

Sub_flank<- data.frame(DV = "flanker effect", Sub_means, Sub_SEM)

ggplot(Sub_flank, aes(x = DV, y = Sub_means))+
  geom_bar(stat = "identity")+
  theme_classic(base_size = 14)+
  geom_errorbar(aes(ymin = Sub_means - Sub_SEM,
                    ymax = Sub_means + Sub_SEM),
                position = position_dodge(width = 1.2),
                width = .4, color = "green")+
  ylab("Mean Flanker Effect")

Exploratory analysis

Multiple questions may often be asked of data, especially questions that may not have been of original interest to the researchers.

In flanker experiments, like this one, it is well known that the flanker effect is modulated by the nature of the previous trial. Specifically, the flanker effect on trial n (the current trial), is larger when the previous trial (trial n-1) involved a congruent item, compared to an incongruent item.

Transform the data to conduct a sequence analysis. The dataframe should already include a factor (column) for the congruency level of trial n. Make another column that codes for the congruency level of trial n-1 (the previous trial). This creates a 2x2 design with trial n congruency x trial n-1 congruency.

First get teh subject means for each condition, then create a table and plot for teh overall means and SEMs in each condition. This should include:

  1. trial n congruent : trial n-1 congruent
  2. trial n incongruent : trial n-1 congruent
  3. trial n congruent : trial n-1 incongruent
  4. trial n incongruent : trial n-1 incongruent

**tip:** be careful, note that the first trial in each experiment can not be included, because it had no preceding trial