Level

Level describes vertical position

On a conventional line graph, higher points have a higher level on the y-axis and lower points have a lower level. The meaning depends on the measure. A higher count of independent communication may align with a selected goal, while a higher duration of pain behavior may signal concern. Direction alone says nothing about value.

State the measurement unit before interpreting level. Five requests per session, five requests per hour, and responses during five opportunities use different denominators. Changes in session length, opportunity count, observer definition, device access, or setting can move the plotted values without a comparable change in the underlying outcome.

The CASP ABA Practice Guidelines Version 3.0 public summary concerns ABA behavioral health treatment for people diagnosed with autism and places treatment within assessment, planning, implementation, evaluation, and care coordination. The detailed guidelines require a license. This page uses the public scope while presenting level analysis as an editorial teaching framework.

A summary can help without replacing the graph

Analysts often describe a condition's typical level with a mean or median. The mean is the sum of values divided by their count. The median is the middle ordered value and can be less affected by one unusually high or low observation.

Suppose fictional condition A contains 3, 4, 5, 6, and 7. Its mean and median are both 5, yet the data rise steadily. Condition B contains 5, 5, 5, 5, and 5. Its mean and median are also 5, while its trend is flat and its variability is zero.

Calling both conditions “level 5” hides the difference. Report the raw sequence, graph, unit, dates, trend, variability, missing observations, and any unusual event alongside the summary. A horizontal level line can orient the reader, provided its calculation and eligible observations are stated.

Within-condition level and level change answer different questions

Within a condition, level asks where its points generally sit. Between conditions, change in level asks how vertical position differs after a condition changes. Reviewers may compare whole-condition summaries, adjacent boundary windows, or model-based estimates. The chosen method should match the design and be named before interpretation.

The What Works Clearinghouse single-case technical documentation treats level, trend, and variability as within-phase features. Its between-phase review also considers immediacy, overlap, and consistency. One common immediacy comparison uses the final three observations of one phase and the first three of the next.

For example, the final three observations in a fictional baseline are 2, 3, and 3, with a mean of 2.67. The first three observations after a condition change are 6, 7, and 6, with a mean of 6.33. The boundary-window level difference is +3.66 after rounding each mean to two decimals.

That arithmetic describes the six observations. It does not establish why they differ. A rising baseline, a coincident schedule change, altered measurement, low procedural integrity, or an outlier could change the interpretation. Replication at design-controlled boundaries supplies stronger evidence than one comparison.

Visual analysis combines several dimensions

The authors of Systematic Protocols for Visual Analysis organize single-case review around level, trend, variability, immediacy, overlap, and consistency. Their review found substantial variation in how published protocols define and sequence these judgments, which supports stating the selected procedure rather than assuming every reviewer uses “level” identically.

A single-case design analysis review similarly separates inspection within phases from comparison across phases. It describes mean or median level lines as possible aids and emphasizes the broader design evidence needed for experimental control.

Level also needs human meaning. A numerical shift may be statistically or visually noticeable while remaining irrelevant to the person's priorities. Ask whether the outcome reflects a chosen goal, access to communication, comfort, participation, safety, burden, or another valued change. Preserve assent and dissent and include the person’s report when possible.

Common level errors

Common errors include comparing averages from unequal exposure, overlooking a strong trend, changing y-axis scales between graphs, treating a percentage like a raw count, and removing inconvenient observations after seeing the result. Another error is inferring improvement from an upward shift without considering what the measure represents.

Floor and ceiling effects can compress visible change. Sparse observations may miss important variation. Aggregating across people, settings, response forms, or weeks can conceal distinct patterns. Keep subgroup and case-level data available when pooled summaries guide decisions.

A practical level review

1. Define the measure, unit, opportunity, and desired interpretation.
2. Confirm axes, scale, dates, condition labels, and comparable measurement.
3. Inspect all points for vertical position, trend, variability, gaps, and outliers.
4. Calculate the named mean, median, or boundary comparison when useful.
5. Compare overlap, immediacy, similar phases, and replicated effects.
6. Check procedural integrity, concurrent changes, access, and client report.
7. Record uncertainty and the qualified decision that follows.