Logging enumerator behavior¶
Collect can log the behavior of enumerators as they navigate through a form. This log has many uses including discovering:
- questions that take a long time to answer
- how enumerators typically navigate through a form
- enumerators who take a particularly long or short time to answer
- if enumerators were at the correct location when filling out a form
- when, why and who changed answers
This information can inform form design and training or feed into data validation processes.
If using Aggregate, Aggregate 1.5.0+ required
If a version of Aggregate lower than 1.5.0 is used, audit files will not be saved on the server.
To enable logging for a form, add a row of type audit and name audit in an XLSForm:
A form may contain at most one row of type audit.
You may add the location of events to the log. To do this, add the following parameters to the XLSForm. All three parameters are required.
high-accuracy: The most accurate location provided by the device, regardless of power use.
balanced: Block level accuracy (~100 meters). Uses less power than high-accuracy.
low-power: City level accuracy (~10 kilometers). Uses less power than balanced.
no-power: No locations will be returned unless another application on the device has requested location updates. Uses no additional power.
- The desired minimum time, in seconds, at which location updates will be fetched by the device.
- The maximum time, in seconds, locations will be considered valid by the device. Must be greater than or equal to location-min-interval.
|audit||audit||location-priority=balanced location-min-interval=60 location-max-age=120|
When location tracking is enabled, ODK Collect requests location updates from Android periodically, with an interval determined by location-min-interval. The requests are sent with location-priority to ensure Android does not use more power than is desired.
When Collect receives the location updates, it stores the locations in a timestamped cache. At the time of an event, Collect checks the cache for locations stored over the last location-max-age and returns the most accurate location in the cache.
For the most accurate locations, set location-priority to high-accuracy. For the most recent locations, use low numbers for location-min-interval and location-max-age.
Location tracking can be an invasion of privacy. Users of ODK Collect will be informed that their location is being tracked when they open a form with this feature enabled.
Users can control their privacy by disabling location providers in Android, refusing to grant Collect location permissions, or by disabling location tracking of specific forms in Collect.
Disabling location tracking will not prevent users from filling out forms, but these changes are logged as events in the log.
You can enable change tracking so that old answers and new answers will be added to the question events. To do this, add the following parameter to the XLSForm: track-changes=true.
New in version 1.25: ODK Collect v1.25.0
You can add to track-changes-reasons=on-form-edit to prompt enumerators to enter a reason before they save changes to a form:
This will prevent filled out forms being edited without a reason being given. If a reason is given the form will be saved normally and the audit log will include a change reason event with the reason recorded in the change-reason column.
If your form needs a record of the identity of the enumerator you can use identify-user=true.
This will cause Collect to prompt the enumerator for their identity before filling out or editing a form instance. In the audit log, a user column will be included that will be populated for each event. The enumerator will not be able to fill in or edit the form without entering a non-blank identity.
identify-user is useful for data collection workflows where devices might be passed between multiple enumerators for data verification or completion.
In cases where a device will only ever used by a single enumerator, it might make more sense to use username metadata. This will write the username to each submission instead of to the audit log.
Central will export a CSV with audits from all submissions if an export is requested for a form with an audit.
If using Aggregate, audit logs can be reviewed and downloaded for further analysis using Briefcase.
In Aggregate 1.5.0+, audit logs can be viewed by clicking on the media icon in the meta audit column on the Submissions page:
This displays a popup with the audit contents:
Aggregate currently only displays event, node, start, and end in the audit popup. To view locations, changed answers, or to perform more sophisticated analysis, logs can be downloaded along with their submissions using Briefcase.
If a form includes an audit, Collect will create an
audit.csv file as the form is filled out. The
audit.csv file has the following structure:
Values in the event column represent a particular user action such as opening a form, saving a form, or displaying a question. Possible event types are described in the Event types section.
Values in the node column represent the node in the form that the event refers to, if applicable.
Values in the start and end columns are timestamps represented as the number of milliseconds since midnight, January 1, 1970 UTC. This is known as epoch time and provides a standard way of representing date/time even across timezones. The Timestamps section contains more information about timestamps.
If both location tracking and change tracking are enabled in the log, the CSV will look like this:
|location tracking enabled||1550615022671|
|location permissions granted||1550615068610|
|location providers enabled||1550615068665|
|location tracking disabled||1550615095914||37.4229983||-122.084||14.086999893188477|
|location tracking enabled||1550615099271||37.4229983||-122.084||14.086999893188477|
Values in the latitude and longitude columns represent the latitude and longitude in decimal degrees. Values in the accuracy column represents accuracy in seconds.
Locations will often be repeated in the log. This is because locations are not captured at the time of the event, but rather retrieved from a cache of the most accurate points captured over the last location-max-age.
Answers will be recorded only if they differ (if the new answer is different than the old one), otherwise, cells should be empty. Answers which contain commas will be surrounded by double quotes.
The event column of the audit log can have the following values:
|form start||Start filling in the form||No||start only||If enabled and available||No|
|question||View a question||Yes||Yes||If enabled and available||If enabled|
|group questions||View multiple questions on one screen (
||Yes||Yes||If enabled and available||No|
|jump||View the jump screen||No||start only||If enabled and available||No|
|add repeat||Add a repeat||Yes||Yes||If enabled and available||No|
|delete repeat||Delete a repeat||Yes||Yes||If enabled and available||No|
|end screen||View the end screen||No||Yes||If enabled and available||No|
|form save||Save the form||No||start only||If enabled and available||No|
|form exit||Exit the form||No||start only||If enabled and available||No|
|form resume||Resume the form||No||start only||If enabled and available||No|
|form finalize||Finalize the form||No||start only||If enabled and available||No|
|save error||Error trying to save||No||start only||If enabled and available||No|
|finalize error||Error trying to finalize the form (probably encryption related)||No||start only||If enabled and available||No|
|constraint error||Constraint or required error on finalize||No||start only||If enabled and available||No|
|location tracking enabled/disabled||Toggle location tracking in Collect||No||Yes||If enabled and available||No|
|location providers enabled/disabled||Toggle location providers in Android||No||Yes||If enabled and available||No|
|location permissions granted/not granted||Toggle location permission in Android||No||Yes||If enabled and available||No|
If we relied entirely on the time reported by the device for timestamps, users or the network could change the device time and manipulate the correctness of the audit log. For this reason, we only use device time for the form start timestamp. All subsequent event timestamps are the result of elapsed time, which users cannot change, added to the form start timestamp. This means that while the timestamps themselves may potentially be inaccurate, the time elapsed within and between the timestamps are always accurate within one form editing session.
Using epoch time makes it easy to compute elapsed time by subtracting start from end. For example, given the following log:
The enumerator spent
1488761809157 - 1488761807868 = 1289 milliseconds on the screen showing the
/data/name question. This corresponds to
1289 / 1000 = 1.289 seconds.
To convert from epoch time to time in UTC in most common spreadsheet programs, divide the epoch time by 86400000 ms per day and add 25569 days between January 1, 1900 (what spreadsheet programs use as "day zero") and January 1, 1970. For example, to convert the timestamp
(1488761807868 / 86400000) + 25569 = 42800.03944
When the cell is set to type date time in common spreadsheet programs, it will show
3/6/2017 0:56:48 UTC. A common workflow if device time is needed in a human-readable format will be to add a column for the calculation above and change that column's type to date time.
- If the device is turned off while a form is being filled, Collect will not record a log entry for the screen that was shown at the time of device shutdown. Events before and after the shutdown will be logged.
- Editing a saved form that was saved using different audit log options can result in a corrupt audit. It might take place when a user saves a form then updates a form definition (changing audit log options) and tries to edit the saved form.