When an EEG event duration is missing, what does the record actually establish?
· PhyPA methods editorial desk · Worked example and worksheet
Two rows in an EEG event table have the same duration entry, n/a. In one case the export log says the stop marker was omitted. In the other case, the available notes say nothing. Both cells can carry the same missing-value marker. They leave the next researcher with different leads.
1. Read the field definition before choosing a value
The Brain Imaging Data Structure (BIDS) defines event duration in seconds. An event modeled as an impulse can have duration 0. An unavailable duration can be n/a. These entries mean different things even if a spreadsheet invites you to format the whole column as numbers. See the BIDS events specification and its rules for missing and non-applicable values in tabular files.
Use a separate working ledger to record the event reference, the literal value, the evidence available for it and the next check. The downloadable CSV is a teaching ledger. An actual BIDS events file has its own required columns, units and tab-separated format.
2. Trace each row to a stated piece of evidence
All four records below are invented. They describe metadata for one example run and contain no signal measurements.
- ev-01: The two marker times in the example export are 3.00 and 3.60 seconds. Their difference gives a duration of
0.60seconds. - ev-02: The example protocol defines a cue marker as an impulse. Enter
0. The zero expresses the modeling convention. It says nothing about how long a physical device took to present a cue. - ev-03: The example export log explicitly reports omission of this event’s stop marker. Keep
n/aand record which log gives that explanation. Nothing in the export establishes when the event ended. - ev-04: The duration field is unavailable and the delivered notes contain no explanation. Keep
n/aand write “cause unresolved.” Borrowing ev-03’s explanation would create a second claim from a single log entry.
3. Check what zero-filling would change
Of four events, two have numeric durations. The mean of those two recorded values, including the deliberately instantaneous cue, is (0.60 + 0) / 2 = 0.30 seconds. Replacing both missing values with zero produces (0.60 + 0 + 0 + 0) / 4 = 0.15 seconds. The second calculation gives the undocumented events a duration and changes the denominator.
This arithmetic checks the handling of the example, not whether an average across these event types would answer a research question. For a report, write which types were included and give the observed and missing counts. The benchmark notes guide provides space to record the dataset version and processing decisions alongside a result.
4. Leave a question someone can resolve
For ev-03, ask whether an earlier export retains the omitted marker. For ev-04, identify the source log or export specification that could establish what happened. If neither can be found, retain the unresolved status. If a later file supplies an offset, keep the original reference and identify the new file and the revision in which the value changed.
Download the four-record missing-data ledger and try sorting by the evidence column. The two missing rows should stay recognizable as separate cases. If those records later feed a benchmark, the participant split audit gives a separate check on how records were assigned to evaluation folds.