Inspecting Data Quality

After loading and initial inspection of raw gaze data, the next step is to assess data quality. High-quality data is essential for reliable analysis and interpretation of eye-tracking results. Common data quality issues include missing data, noise, drift, and artifacts caused by blinks or head movements.

Inspecting Data Loss

Data loss can occur during tracking due to blinks, tracking errors, or participants looking away from the screen. The length and frequency of consecutive data-loss segments are useful indicators of overall dataset usability. The function data_loss_histogram() plots the distribution of consecutive missing-data chunk lengths (expressed in samples or time), which makes it easy to see whether loss is mostly brief (e.g. short blinks) or prolonged (e.g. long tracking failures).

Show code cell source

Hide code cell source

import matplotlib.pyplot as plt

import pymovements as pm
from pymovements.gaze.experiment import Experiment

# Create an Experiment with the appropriate display geometry
experiment = Experiment(
    screen_width_px=1280,
    screen_height_px=1024,
    screen_width_cm=38,
    screen_height_cm=30.2,
    distance_cm=68,
    origin='upper left',
    sampling_rate=250.0,
)

# Load a dataset that contains simulated blinks / tracking loss
gaze_with_loss = pm.gaze.from_csv(
    '../examples/gaze-with-loss.csv',
    experiment=experiment,
    time_column='time',
    pixel_columns=['x', 'y']
)

# Plot the data-loss histogram in units of time
pm.plotting.data_loss_histogram(
    gaze_with_loss,
    column='pixel',
    unit='time',
    sampling_rate=experiment.sampling_rate,
)
plt.show()

Interpretation

Short, infrequent gaps often reflect normal blinks or brief signal dropouts and are commonly tolerated or interpolated in preprocessing. A large number of long gaps suggests severe tracking problems and may warrant re-collection or exclusion of affected trials or participants.