README.md

In abeith/eyeCleanR: Clean and Transform Eye-Tracking Data

eyeCleanR

This package can be used to clean raw eye-tracking data based on expected patterns for exploratory analysis. The current implementation focuses on sequential reading, where y coordinates are correlated with time and x coordinates follow a quasi-regular oscillatory pattern.

Installation

You can install this version of eyeCleanR using the following command:

devtools::install_github("abeith/eyeCleanR")

You can then use library(eyeCleanR) to load the package.

Vignettes

Use the following command to browse the documentation for this package.

browseVignettes("eyeCleanR")

Alternatively, to view vignettes in R, use vignette(package = "eyeCleanR") to see a list of available vignettes and then open the relevant vignette (e.g. vignette("new_lines", package = "eyeCleanR")).

Aim

The aim of this package is to facilitate the non-destructive detection of reading behaviours for exploratory analysis. The saccades package (von der Malsburg 2015) adopts a method set out by Engbert and Kliegl (2003) to identify saccades based on the velocity of eye movements. This approach draws the distinction between saccadic and microsaccadic movement, allowing for small movements within fixations and large movements between fixations. Where areas of interest (AOIs) are categorical, the timing and position of a fixation can be used to infer visual attention as a categorical variable. However, this requires the signal to noise ratio (SNR) of the observed point of gaze (POG) data to be large enough to distinguish between neighbouring AOIs. In naturalistic reading tasks this may not be the case as the spatial distance between lines is small. For example, a saccade from the first word in the first line of a text, to the first word in the second line of a text would may be smaller than the saccades from the first word to the second word within a line. This is problematic as two saccades of similar magnitude would be observed while the ordinal progression through the text would be very different.

eyeCleanR attempts to facilitate the identification of reading behaviours by cleaning the signal, rather than aggregating to fixations. The advantage of this is greater flexibility, allowing the researcher to detect patterns while avoiding destructive transformations. The detect_newline() function in particular addresses the problem of identifying saccades to a new line. While the amplitude of these saccades in the y coordinate are small, they are associated with large negative saccades in the x coordinate. Using a similar approach to Engbert and Kliegl (2003), velocity is used to identify when the magnitude of a saccade indicates a shift in attention. High velocity left saccades are assumed to be indicative of progression to a new line. This assumption is likely to hold for data with a high SNR and the reader following a predictable progressive reading pattern. For noisy data, or data with frequent regressions, additional data cleaning steps are necessary to reject large regressions and artefacts. The filter_points() function offers a simple method of rejecting points outside of the target area. Where this is not sufficient, the new_lines vignette provides an example of how modelling the function of y-axis position over time can be used to reject outliers.

The advantages of identifying saccades to a new line are demonstrated in the x_progress vignette. For languages read from left to right, reading progress within lines can be expressed variations in a continuous variable over time. This time series representation of reading progress allows observations to be modelled and compared to predicted patterns or other readers. The examples provided in the vignette demonstrate how data can be modelled and visualised. However, as there is no aggregation of data, the researcher is free to apply whichever method of analysis is appropriate for the research question. Non parametric approaches can also be used in the analysis of x coordinate time series, using packages such as infotheo (Meyer 2014) for Mutual Information calculations or crqa (Coco, James D. Dixon, and Nash 2018) for Cross-Recurrence Quantification Analysis.

References

Coco, Moreno I., Rick Dale with contributions of James D. Dixon, and John Nash. 2018. *Crqa: Cross-Recurrence Quantification Analysis for Categorical and Continuous Time-Series*. .

Engbert, Ralf, and Reinhold Kliegl. 2003. “Microsaccades Uncover the Orientation of Covert Attention.” *Vision Research* 43 (9): 1035–45. .

Meyer, Patrick E. 2014. *Infotheo: Information-Theoretic Measures*. .

von der Malsburg, Titus. 2015. *Saccades: Detection of Fixations in Eye-Tracking Data*. .

abeith/eyeCleanR documentation built on June 1, 2019, 12:43 a.m.