jwdink / eyetrackingr Goto Github PK

Lots of variability across our functions about when an arg takes a single AOI/condition, multiple AOIs/conditions, etc.

Then auto-convert into readme/API?

Rethink remove_trackloss and keep_trackloss

No need to take the mean for each subject*item, might as well maintain the original subject*trial structure. Only makes a difference if a subject sees the same item on multiple trials.

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3719041/

Ensure that package keeps behaving exactly the same when we update.

http://r-pkgs.had.co.nz/tests.html

This could be a theoretically and practically good way to compute divergences

Non-trivial because requires rethinking how data is shuffled. Probably will eventually implement by user supplying vector of predictor_columns, then each is shuffled consecutively on each iteration (according to whether it's within or between subjects). Doing this might make the shuffle_by arg more cumbersome than it already is, so might want to rethink that whole system.

E.g., if I try to plot a data.frame that came out of the time_analysis function, but I somehow accidentally removed its class, then what happens?

To avoid cluttering up issues forum, will put brainstorming for all future enhancements here.

• Twitter
• Colleagues
• This stack-overflow question: http://stackoverflow.com/questions/21992426/r-package-to-analyse-eye-tracking-data
• ...

from Yurovksy & Frank (2015, Developmental Science)

I see the following problems:

checking R code for possible problems ... NOTE
plot.time_sequence_data: warning in stat_summary(fun.dat = mean_se,
  geom = "ribbon", alpha = 0.25, colour = NA): partial argument match
  of 'fun.dat' to 'fun.data'

fairly rare that we use these. we should either fix this bug and add lots more links between documentation pages, or avoid bug by removing the few links we do have

It references running subset_by_window() with 15500 - 21000ms, but I don't actually see this call in the vignette?

Right now, it seems to just draw an NA for that entire bin, thus creating gaps in the clusters.

http://www.inside-r.org/packages/cran/zoo/docs/rollapply

Also add unit-tests for this to test_arbitrary_dv.R, for both within and between subjects.

• Create make_look_data() listing each look (i.e., consecutive samples in AOI)
• Create analyze_longest_looks() to compare longest looks to each AOI by condition, etc.

Hi,

which eye tracking systems have you used in your research? Can you recommend me some systems which can be used with eyetrackingR?

Thanks,
Gerald

Abstract:

Eye-tracking is an increasingly popular method within cognitive science. However, researchers lack a standardized framework and set of tools that are (a) easy to use and (b) based on free and open-source software. We offer a framework and accompanying package in R that addresses this gap. Our tools address tasks along the pipeline from raw data to analysis and visualization. We offer several popular types of analyses, including linear and growth curve time analyses, onset-contingent reaction time analyses, and cluster mass analyses. We also offer novel non-parametric techniques that address limitations in current analyses, with a particular emphasis on statistically estimating when looking behavior diverges across experimental conditions (rather than just whether it has diverged at some time-point). Finally, our R package is optimized for processing very large datasets extremely quickly. Our paper will outline the proposed workflow, discuss the logic and limitations of each analysis, and provide a walkthrough tutorial.

Hello,

I have a question regarding plotting the lines of multiple AoI's in one plot when visualizing the time course.
I tried so many options already but I still get the same outcome in most cases. Four separate plots, one per AOI (for me being: AOI1, AOI2, AOI3, AOI4 with binary data), rather than one plot with the four lines with different colors in one plot. But I also have one column AOI consisting of numbers 1 to 4 in which 1 equals AOI1, 2 equals AOI2, and so forth. In your example of the growth curve analysis, both the fixations to the Animate and Inanimate AOI are included in one plot.

See the uploaded image for the outcome I get when I tried to plot my data with codes as:

response_time2 <- make_time_sequence_data(response_window_clean, time_bin_size = 50, predictor_columns =c("AOI"), aois =c("AOI1", "AOI2", "AOI3", "AOI4"))

plot(response_time2, predictor_column = "AOI" ) + theme_light() + coord_cartesian(ylim = c(0,1))

Also when I tried to get rid of the predictor columns I got the same result.

I followed the instructions given by the eyetrackingR package, such as:

eyesdata <- make_eyetrackingr_data(data, participant_column = "SUBJECT", trial_column = "TRIAL", item_column = "ITEM", time_column = "TIMESTAMP", trackloss_column = "TRACKLOSS", aoi_columns = c('AOI1', 'AOI2','AOI3','AOI4'), treat_non_aoi_looks_as_missing = TRUE)

response_window_clean <- clean_by_trackloss(data = eyesdata, trial_prop_thresh = .25)

It would be great if you could help me with this issue.
I am already struggling with it for a while, and so far now one around me was able to help solving this problem with me.

Let me know if you need anything more