To compare ABC data across settings, calculate sequence and background probabilities separately for each setting using the same observable definitions and compatible windows. Report raw response, event, exposure, valid-observation, and missing counts. Compare like units and examine setting-specific access, people, tasks, health, and timing. A pooled probability can hide different patterns and unequal exposure.

Stratify before pooling

Create one table per setting, routine, or partner category named in the question. Keep an all-settings total only as an additional summary.

Match definitions and windows

A comparison requires the same response code, event code, timestamp rule, and sequence duration. Mark version changes instead of blending them.

Report exposure and coverage

Show valid observations, response events, conditioning exposures, and missed units for every stratum. Small or selectively observed settings need cautious interpretation.

Review background opportunity

An event can be more available in one setting. Compare conditional probability with the setting's own background probability before interpreting the sequence.

Build matched setting tables

To compare ABC data across settings, create the same four-cell or sequence table for each setting. Use the same response definition, event definition, window, timestamp precision, missing-state rule, and rounding. Record observer, activity mix, access conditions, health context, and coverage so the reviewer can identify differences in how the samples were collected.

Keep the denominators visible. Ten response events in one setting and six in another do not provide equal information. A setting with a high conditional percentage based on two events needs a different level of caution from one based on a large, representative sample.

Compare conditional and background values together

For each setting, place the event-after-response probability beside the event's background probability. The difference between those two descriptive values can help frame the next assessment question. Compare that pattern across settings only after confirming that background units and sequence units have compatible definitions.

Vega's setting A has 0.80 after responses and 0.30 in background intervals. Setting B has 0.50 in both places. The pooled 0.688 result loses that contrast. Retain counts, because 8 of 10 and 3 of 6 communicate more than the percentages alone.

Decide whether pooling answers anything useful

An all-settings total can describe the combined sample, but it should not replace setting rows when context is part of the question. Weighting by observations or response events changes the meaning of the summary. State the unit and purpose before producing a pooled figure.

When setting procedures differ, mark them as separate measurement versions and repair the comparison prospectively. Avoid adjusting one row after the fact to make it resemble another. Report the limitation and collect a matched sample when the comparison remains important.

Frame one clear comparison question

The setting-stratified ABC table for Vega names the observable events, sequence direction, time or interval window, planned exposure, valid denominator, missing-state rule, raw cells, calculation owner, and review date before percentages are interpreted.

Worked example: two settings, different backgrounds

Vega's setting A has the consequence after 8 of 10 responses, or 0.80, with a background probability of 15 of 50 intervals, or 0.30. Setting B has 3 of 6, or 0.50, with a background probability of 12 of 24, also 0.50. Pooling the consequence-after-response counts gives 11 of 16, or 0.688, and obscures the different background relations.

Audit setting-specific denominators

Vega's comparison retains 10 versus 6 response denominators and 50 versus 24 background units. The observation schedules, sequence window, event codes, and missed coverage are checked before the setting rows are interpreted together.

Interpret context without labeling the setting

A contextual difference may reflect distinct routines, communication partners, demands, supports, event availability, observation schedules, or measurement quality. It provides a question for qualified assessment and does not justify a label about the setting.

Protect access during observation

During Vega's context-specific probability patterns review, keep AAC, interpreters, mobility, food, water, bathroom use, prescribed care, health support, rest, relationships, and emergency help available. Record access or health changes and use accessible communication to ask about relevant events, comfort, assent, dissent, and priorities.

Use evidence within its stated scope

The BACB ethics hub and CASP public summary provide professional context. The BCBA Test Content Outline includes direct, indirect, and product measures; validity and reliability; representative measurement; and indirect, descriptive, and experimental functional assessment as examination content.

For Vega's context-specific probability patterns question, primary descriptive-assessment research compares conditional with background probabilities and shows why event direction, sequence windows, cell counts, and context matter. Antecedent versus consequent predictors, contingency space analysis, and comparisons of descriptive methods document limited correspondence between descriptive and experimental findings. Research on reducing functional-assessment ambiguity distinguishes observed association from experimental demonstration. ASHA says AAC users should always have access to their tools or devices.

Close the descriptive-data review

Review the setting-stratified ABC table with Vega and the responsible clinician. Record raw observations, definitions, windows, coverage, calculations, limitations, the selected next assessment step, its owner, and a review date. Reopen the method when the response, context, access, health, observer, software, or decision changes.

Clinician review checklist

  • Do all setting rows use compatible definitions, windows, and code versions?
  • Are response, event, background, valid-observation, and missing counts visible?
  • Is each conditional value paired with its setting-specific background value?
  • Are small or selectively observed samples identified?
  • Does any pooled result state its unit, weighting, and purpose?
  • Are differences in people, activity, access, and health context documented?
  • Does the report keep descriptive patterns separate from conclusions about function?

Related resources

Sources