When to use it
Use event recording when the target behavior is discrete: it has a clear start and a clear end, and you can count each instance without guessing where one stops and the next begins. Good examples are hitting, requesting with a mand, raising a hand, or throwing an object.
It works less well for:
- Very high-rate behavior, where you cannot count accurately (consider interval recording).
- Behavior that lasts a variable amount of time, such as crying or on-task behavior. Here duration usually tells you more than a count (consider duration recording).
- Behavior without clear boundaries, where two observers would disagree on what counts as “one.”
How to fill it in
- Write the operational definition before you observe. Describe what the behavior looks like (topography), what marks its onset and offset, and what does not count (exclusions). Example: “Hitting: any forceful contact between the client’s open or closed hand and another person’s body, from above the elbow. Excludes high-fives, accidental contact during play, and contact while the client is being prompted physically.”
- Set a rule for when a new instance starts. For behaviors that come in bursts, decide ahead of time how long a pause separates one instance from the next (for example, “3 seconds without hitting ends an episode”). Write it on the sheet. Without this rule, one observer’s “one episode” is another observer’s “six hits.”
- Record the start time when the observation period begins.
- Make one tally mark for each instance that meets the definition. Use a clicker or counter app if tallying would take your attention away from the client.
- Record the stop time when the observation ends, then total the tallies in the Count column.
- Calculate rate (see below) whenever observation periods differ in length.
- Note anything unusual in the Notes column, such as illness, a new staff member, a medication change, or a schedule change. These notes become event markers on the graph.
Common mistakes:
- Counting instances that don’t meet the definition because they “felt like” the behavior.
- Leaving out the observation time, which makes the count impossible to compare across days.
- Graphing raw counts from sessions of different lengths. Twelve hits in 2 hours and twelve hits in 20 minutes are very different.
- Changing the definition partway through a phase without marking the change.
The sheet
Client:Observer:Setting:
Target behavior:Operational definition:
New-instance rule (pause length between instances):
| Date | Start time | Stop time | Total minutes | Tally | Count | Rate (per min or per hour) | Initials | Notes |
|---|---|---|---|---|---|---|---|---|
Summarizing the data
Count is the total number of instances in the observation period. Report count alone only if every observation period is the same length.
Rate = count divided by observation time. Choose one unit (per minute or per hour) and use it for every session in the program so the graph stays comparable.
Worked example: a client hits 14 times during a 35-minute observation.
- Rate per minute = 14 / 35 = 0.4 per minute
- Rate per hour = 0.4 x 60 = 24 per hour
If the next session is 60 minutes with 18 hits, the rate is 18 per hour. The raw count went up (14 to 18), but the rate went down (24 to 18 per hour).
Graph rate on a line graph with sessions or dates on the x-axis. Mark phase changes with a vertical line and note events such as medication changes directly on the graph.
Tips
- If a behavior is too fast to count and too important to skip, a short timed sample (for example, count for 10 minutes at the start of each hour) is better than an inaccurate all-day count. Report it as rate for the sampled time only.
- Check interobserver agreement (IOA) regularly. For count data, total count IOA (smaller count divided by larger count, times 100) is the simplest method. Mean count-per-interval IOA is stricter. Many practices aim for 80% or higher on at least 20% of sessions, but requirements vary by organization and payer.
- Keep the observation window consistent across days when you can (same activity, same time of day). This makes changes in the data easier to interpret.
- If the count is zero, write 0. A blank cell looks the same as a missed observation.