Removing unlikely pupil values involves up to three steps:
Deblink pupil data.
pupil_deblink()
Artifiact rejection.
pupil_artifact()
Evaluate and/or Remove trials
with large amounts of missing data.
pupil_missing()
To remove unlikely pupil values that can be caused due to blinking,
the standard technique is to deblink the pupil data,
which just means to remove pupil values during a blink (Geller et al.,
2020). You can do this using pupil_deblink().
Eye trackers have their own algorithm for detecting various eye events (fixations, saccades, and blinks). It is important to have columns in your raw data file that contain these eye event information.
Some eye tracker algorithms do not actually capture the entire duration of a blink, such that it can cause extreme dips in pupil values around the edges of a blink. This can result in unlikely pupil values.
The most simple method of dealing with this is to extend the deblink by some amount (Geller et al., 2020). This is easily implemented:
pupil_deblink(extend = 100)This will extend the deblink by 100ms to get something like:
You can use the plot and plot_trial
arguments to determine how much you need to extend the deblink by and
produce a plot like the one above.
pupil_deblink(extend = 100, plot = TRUE, plot_trial = c(1,2,3,4,5))To plot all trials use plot_trial = "all"
Unlikely pupil values can be caused by a host of other reasons. One characteristic that unlikely pupil values will display is a large change in pupil values over a short amount of time.
A recommended way of dealing with these types of pupil values is
using the Median Absolute Deviation (MAD), pupil speed, and a constant
(n) to determine a threshold on pupil speed (Geller et al., 2020; Kret
& Sjak-Shie, 2019). Any values that exceed the threshold get removed
from the data (set to missing, NA).
This method is implemented in pupil_artifact():
pupil_artifact(n = 16)You can use the plot and plot_trial
arguments to determine what value to set the constant (n) to
pupil_artifact(n = 16, plot = TRUE, plot_trial = c(1,2,3,4,5))To plot all trials use plot_trial = "all"
It can be a good idea to evaluate how much missing data each trial
has after data cleaning and before other preprocessing steps such as
interpolation. The pupil_missing() function will create a
column in the data with the amount of missing data on that trial. This
column can be used later on during analysis to remove trials with too
much missing data.
By specifying a missing allowed criterion
pupil_missing(missing_allowed = .90) you can also remove
trials with too much missing data. At this stage, it can be a good idea
to remove trials with large amounts of missing data.
One reason is that you will likely want to discard these trials anyways
(it is not worth interpolating the missing values because there is too
much missing). Another reason is more practical, in that trials with an
excessive amounts of missing data (particularly early in the trial) can
cause errors downstream when using the other preprocessing
functions.
The amount of missing data that you are comfortable allowing can be
set with the missing_allowed = argument in
pupil_missing()
pupil_missing(missing_allowed = .90)Geller, J., Winn, M. B., Mahr, T., & Mirman, D. (2020). GazeR: A Package for Processing Gaze Position and Pupil Size Data. Behavior Research Methods, 52(5), 2232–2255. https://doi.org/10.3758/s13428-020-01374-8
Kret, M. E., & Sjak-Shie, E. E. (2019). Preprocessing pupil size data: Guidelines and code. Behavior Research Methods, 51(3), 1336–1342. https://doi.org/10.3758/s13428-018-1075-y