We are pleased to provide full access to all your data on EyesDecide. We provide a number of data export tools.
The Individual Recordings page allows you to download or export two types of data: Raw recordings, and Structured JSON.
This is the most convenient format for analysis of individual gaze and mouse tracks in other software. The data has been preprocessed to transform each session into a continuous recording with meta-data about media shown, the participant, unique IDs to cross-reference with AOI analyses exports and more.
All coordinates in the structured JSON download are participant screen coordinates (not your screen coordinates). Information about how the media viewed was positioned on screen is also included, so the viewing can be reconstructed.
This format is ideal for further analysis in your own custom software.
Since the data is complex and often quite large, we recommend using a service such as JSONLint to format it nicely for browsing.
To download the structured JSON data, open the Individual Recordings tab in your study and click ‘Download All‘.
You can also download the same format on a per-view basis by clicking the icons next to any playable view.
Raw Data Download
To access the Raw Data Download feature:
- Open any existing study with valid eye tracking results.
- Select the ‘Recordings’ tab.
- Click the ‘Download raw data’ link (see image below).
The data will appear in a new browser window as a HTML file. You can save the file to disk.
Note that all the data used by EyesDecide to reconstruct and replay gaze is provided without any filtering. This ensures you can reproduce any graphical output you see in EyesDecide, using other software. It also satisfies our objective to be transparent with your data.
Raw Data Format
The raw data is in HTML format but it contains some JSON preformatted text within the HTML and links to CSV files. To obtain all the data, you should download the CSV files also. These are much bigger – potentially 10MB or more in a large study with many views.
The data is structured as “div” elements. A single outermost div is the root of the data and has the id property “data”. Inside this element are one or more div elements with the class “content” or “content-view”. Each content refers to a media item. There is also one div per viewing, with the class “content-view”.
CSV File URLs
The CSV links are relative to our data store. Therefore you will need to create a complete link as follows.
Assume your file is ‘12345/gaze_SKH4398RKJLSSDF0D.csv’. The number indicates the view ID. The prefix ‘gaze’ is the type of data. The alphanumeric suffix is a pseudo-random key that prevents random access unless you already have the key.
The data store location is:
Therefore the complete file URL is:
The “content-view” div has the following attributes:
- id: A unique number identifying the view.
- class: The class attribute will always have the value “content-view”
- status: The value will be a string describing the status of the viewing. If the value is “after” or “view” the viewing was completed.
- note: Text entered by the viewer after viewing, describing what they saw.
created: Date and time at which the viewing occurred.
- name: The EyesDecide user who completed the viewing. If an anonymous URL was used, the username will be “email@example.com”
The “content-view” div has the following child elements containing data. They are all <pre> elements; the “class” attribute tells you what data is inside each element:
- result: A string describing the outcome of the viewing. If “complete” the viewing was successful and recorded gaze data.
- states: A sequence of numbers describing the user’s flow through the viewing process. It is not necessarily linear as there are optional and cyclic paths.
score: A JSON object containing the positions on screen the user clicked and the gaze values that were computed at the same time. This raw data allows the accuracy percentage score to be constructed.
- tags: The study organizer can annotate viewings with tags, to group them.
- settings: A JSON object describing the settings of the viewer’s computer. In particular, the dimensions of the screen in pixels and the Dots Per Inch (DPI). This allows the physical size of pixels to be computed. The start and end times of the actual media viewing step are also included. These are necessary because the CSV log data (below) records the entire session for which the camera was in use. Therefore, we need to truncate that data to the interval while media were being viewed.
CSV (Comma Separated Values) Data Format
- gaze.csv: One row per measurement. Each measurement includes screen X,Y gaze coordinates in pixels and a timestamp in milliseconds since Epoch (00:00 1st Jan 1970). The 4th measurement is a confidence indicator. Lower values indicate higher confidence (less error expected). As a rule of thumb, you can disregard measurements with a confidence higher than 8 or 9.
- mouse.csv: One row per measurement. Each measurement includes screen X,Y mouse coordinates in pixels and a timestamp in milliseconds since Epoch.
- error.csv: One row per measurement. Each measurement represents the occurrence of an error and includes a timestamp for the error.
- scroll.csv: One row per measurement. This is a log of all scroll actions performed by the user. This allows reconstruction of media positioning on screen.
Using all this raw data, it is possible to exactly reconstruct the user’s experience of the media viewed and their interactions with it using mouse and gaze.