Descriptive assessment sampling coverage compares observed and valid sampling cells with a predeclared frame of relevant routines, times, settings, and partners. Report observed cells divided by planned cells, valid cells divided by planned cells, and the location of missingness. A high overall percentage can still omit the conditions most relevant to the assessment question.

Build the frame before observing

Descriptive assessment sampling coverage begins by translating the clinical question into a planned frame of routines, time bands, settings, communication partners, relevant supports, and ordinary contextual conditions. Define how many observation cells are needed in each combination and what makes a cell complete and valid. A frame is a prospective coverage map, not a quota for provoking events.

Use stable cell IDs and record planned date range, duration, responsible observer, observation method, and person-specific access requirements. Keep the frame tied to the current question. If the question changes from transitions across the day to mealtime communication, create a revised version instead of relabeling completed cells.

Track each coverage state

Every planned cell should end in one state: observed and valid, observed but invalid, not observed, rescheduled, declined, or unresolved. An observation may be invalid because visibility was lost, definitions changed, timestamps failed, the wrong routine occurred, or essential context was unavailable. Preserve the reason and source.

Report observation coverage as observed cells divided by planned cells. Report valid coverage as valid cells divided by planned cells. A separate yield measure can use valid divided by observed, but it cannot replace the planned denominator because it hides unobserved cells.

Verify Gideon’s frame

The plan has three routines × two time bands × four repetitions = 24 planned cells. Nineteen cells are observed: 19 / 24 = 0.7917, or 79.2% observation coverage. Eighteen are valid: 18 / 24 = 75% valid coverage.

Five cells were not observed because 24 − 19 = 5. One observed cell was invalid because 19 − 18 = 1. Together, six cells are missing or invalid. Four of those six concern the evening transition. The overall 75% value therefore cannot be described as balanced coverage across contexts.

Valid yield among observed cells is 18 / 19 = 94.7%, which shows that most completed observations passed validity rules. It says nothing about the five unobserved cells. Reporting yield alone would conceal the concentrated evening gap.

Audit context-level denominators

Create a matrix with planned, observed, valid, missing, invalid, and declined counts for each routine and time band. Add partner, setting, support, weekday, and other predeclared strata when relevant. Repeated morning observations cannot fill an evening-transition cell because the contextual exposure differs.

Review whether cell duration and visibility are comparable. A five-minute partial transition may not satisfy a twenty-minute cell rule. Avoid combining several short fragments after the fact unless that construction was predeclared and preserves the clinical question.

Handle missingness and participant choice

Missingness can be operational, such as staff absence or device failure, or informative, such as a routine that is frequently canceled, difficult to access, or declined. Preserve the distinction. Do not code a person’s dissent as invalid data or pressure participation to improve coverage.

Set a safe rescheduling rule, maximum observation burden, and stop conditions. If evening observation is infeasible or unwanted, record that limitation and consider other ethically and clinically appropriate evidence. No overall percentage authorizes creating a difficult routine.

Evaluate sequence-data readiness

Coverage is necessary for probability analysis and still separate from event yield. A valid cell may contain no target response, while another may contain several response anchors with truncated consequence windows. Report valid observation cells, valid time, response exposures, complete sequence windows, and event counts separately.

Check observer agreement and timestamp precision for a representative subset when event-sequence calculations depend on fine timing. High sampling coverage with weak event definitions or unreliable clocks does not produce trustworthy conditional probabilities.

Graph and report coverage

Use a heat map or matrix with routine on one axis and time band on the other. Display planned, valid, and missing counts in text so color is not the only signal. A flow diagram can show 24 planned to 19 observed to 18 valid, with the evening-transition deficit called out.

Suggested wording: “The sampling frame contained 24 planned cells across three routines and two time bands. Nineteen cells were observed (79.2%), and 18 were valid (75%). Five cells were unobserved and one observed cell was invalid. Four of the six missing or invalid cells involved evening transitions, limiting conclusions for that context.”

Keep inference inside the sampled frame

Representative coverage is judged against the declared question and frame, not a universal percentage. Seventy-five percent could be adequate for one descriptive update and seriously deficient when the missing quarter contains the central context. Observations also describe the people, supports, settings, and dates sampled.

Predictor research, contingency-space analysis, and lag-sequential work depend on clearly defined observations and sequences. Comparative descriptive-method research cautions against treating descriptive association as experimental function.

Include accessible review and safety limits

Review the frame with Gideon using his preferred communication. Ask which routines matter, when observation feels acceptable, what conditions change his experience, and what should remain private. Keep AAC available under ASHA guidance.

Observation cannot delay food, water, bathroom use, mobility, prescribed health care, rest, relationships, or emergency help. Record assent and dissent without treating them as data-quality defects. The BACB ethics hub and CASP summary offer professional context; medical, safety, privacy, payer, and legal matters remain with qualified owners.

Clinician checklist and practical limitations

Before interpreting the sample, confirm:

  • routines, time bands, settings, partners, supports, repetition targets, and validity rules were declared;
  • all 24 cells reconcile to observed, valid, missing, invalid, declined, or unresolved states;
  • 19/24 = 79.2%, 18/24 = 75%, and 18/19 = 94.7% reproduce;
  • missingness is reported by context, especially the four evening-transition gaps;
  • event counts and complete sequence windows remain separate from cell coverage;
  • the display is accessible and participant choice is respected; and
  • the report states supported contexts, excluded inference, next evidence step, owner, and date.

Coverage percentages cannot prove representative behavior, measurement validity, or function. Gideon’s planned frame reflects one assessment question and may omit relevant health or social contexts. Reopen it when the question, routine, setting, access, health, observer, definitions, or software changes, and preserve previous frames and coverage decisions.

Related resources

Sources