You may not have a single dataset with all the subjects together, but instead have to load one dataset per subject and combine them afterwards. For this kind of operation the map_dfr function is useful (similar to apply, it is in the purrr package, from the magic world of the tidyverse).
Loading the dataset
list.files lets you search and see which files match a specific pattern. In my case I go into the folder and load all the data in .csv format (I use the head function to show you the first 6)
I apply the map_dfr function to all the files just obtained with the list.files function…
d <-list.files(path ="dataRaw", pattern ="*.csv", full.names =TRUE)%>%map_dfr(~read_csv(.x))
Error in `dplyr::bind_rows()`:
! Can't combine `..37$session` <double> and `..38$session` <character>.
Loading the dataset
Error in dplyr::bind_rows(): ! Can’t combine ..37$session and ..38$session.
For some reason, in some files the session information is read as double instead of character. So inside map_dfr I also add mutate
d <-list.files(path ="dataRaw", pattern ="*.csv", full.names =TRUE) %>%map_dfr(~read_csv(.x) %>%mutate(condition =as.character(condition),session =as.character(session)) %>%subset(RUN >=1)) # remove all the extra rows I don't need
Looking at the variables
For example, I check that the subjects used the assigned IDs. The unique() function is very useful because it returns a vector without duplicated elements (in my dataset I had 448 trials per subject, so each id is repeated 448 times).
I convert the variables I am interested in to factors, and keep only those I need for visualization and analysis. We can use mutate() to mutate/change the variable type, and select() to select the variables of interest:
Sometimes some subjects are excluded a priori because they did not follow the “rules” of the experiment; in this case you can exclude them right away
dat_clean <-subset(dat, # PSQI = 14 id !="GNG82"&# not respect sleep schedule id !="GNG16"& id !="GNG21"& id !="GNG40"& id !="GNG46"& id !="GNG51"& id !="GNG52"& id !="GNG58"& id !="GNG65"& id !="GNG72"& id !="GNG75"& id !="GNG83"& id !="GNG92"& id !="GNG95"& id !="GNG96"& id !="GNG105")# you can also create a vector of IDs to exclude and then use %in%
Excluding subjects
In my case I also wanted to exclude everyone with accuracy below 70%. To compute descriptive statistics, the group_by and summarise functions from the dplyr package (tidyverse again) are very useful
summ_acc <- dat_clean %>%group_by(id, target)%>%summarise(acc_mean =round(mean(acc),2),n =n())# subjects GNG102 GNG38 GNG59 GNG59 had very low accuracysumm_acc$id[which(summ_acc$acc_mean < .7)]
After cleaning the data it is good to check how much information is left, for example the number of subjects per group. n_distinct counts the number of unique combinations..
# A tibble: 2 × 2
C n_subj
<fct> <int>
1 FS 34
2 SD 32
Now let’s visualize them
One of the best packages for data visualization is ggplot2; you can do practically anything with it.
The first reason to visualize data is to see their distribution, the presence of extreme values… ggplot takes as input which variables to use for x and y (or only one of the two, depending on what we want to plot); I can also specify colors and various other things we will see…
theme_set(theme_pubr())dat_clean2 %>%subset(target =="go.png") %>%ggplot(aes(x= rt, fill = P))+geom_density(alpha = .5)
Accuracy
dat_clean2 %>%subset(target =="nogo.png") %>%group_by(id,P)%>%# group the dataset by id and P# compute the mean accuracy per condition/subjectsummarise(acc_m =mean(acc))%>%# ggplot(aes(y= acc_m, x = P, color = P))+geom_jitter(alpha = .8, size =2.5)+# dotsstat_summary(color ="black") # add mean and se
Customization
A simple and nice thing to do is to create your own color palette to use whenever you want in your plots. You can define it at the beginning or write it in a separate script and load it with the source() command, as we saw for functions. At this link you can find lots of (colorblind-friendly) colors
my_plot <- dat_clean2 %>%subset(target =="go.png"& rt > .15) %>%ggplot(aes(x= rt, color = P, fill = P))+geom_density(alpha = .5) +facet_wrap(~C)+scale_color_manual(values = my_pal$pal1)+scale_fill_manual(values = my_pal$pal1)+xlab("RTs in sec")# create the plot folderdir.create("plot")# save inside the plot folderggsave("plot/my_plot.jpg", my_plot,dpi =600, width =18, height =10, units ="cm")
Tables
Another thing you may find useful is creating tables with, for example, descriptive statistics or the output of a statistical model. For these operations the flextable or kableExtra packages are very useful:
doc <- officer::read_docx() # Create an empty Word documentdoc <-body_add_flextable(doc, value = tab) # Add the flextableprint(doc, target ="tabella.docx") # Save the file
