Within-client versus across-client averages use different analysis units. A within-client summary combines compatible observations for one person. An across-client summary gives each eligible person the declared weight. Pooling all sessions gives clients with more observations greater influence. Report client-level values, observation counts, weighting, spread, missingness, and individual patterns before interpreting a group result.

Define the decision and analysis unit

Begin with the question the summary must answer. “What was each person’s average across eligible observations?” is a within-client question. “What was the average client-level result for the eligible client cohort?” gives each client one analysis value. “What percentage of all opportunities across all clients met the definition?” pools opportunity-level numerators and denominators. These are different estimands and can produce different numbers from the same records.

Write the eligible clients, dates, phase, service, setting, outcome definition, observation rule, and maturity cutoff before calculation. Give each client a stable identifier and keep the source series linked. Do not add a person because their data are easy to retrieve or drop a person because their record remains incomplete.

Build compatible client-level values

For each client, verify the operational definition, unit, opportunity or exposure rule, phase, and observation schedule. A table of within-client versus across-client averages should begin with the declared within-client statistic. If using a mean of session percentages, show the session percentages and number of eligible sessions. If using pooled opportunities within a client, sum that client’s numerators and denominators first. Those methods are identical only when session denominators are equal or the values happen to align.

Preserve variability and time order. Two clients can each average 60% while one is stable at 60% and the other moves from 20% to 100%. Report count, range or another justified spread measure, phase, and graph alongside the average. A client-level clinical decision still requires the individual record.

Declare the weighting rule

An equal-client average gives each eligible client one vote regardless of session count. Add the client-level values and divide by the number of eligible clients. A session-weighted mean gives greater influence to clients with more sessions. An opportunity-weighted result pools numerators and denominators, giving more influence to records with more opportunities. Label the method in the table title or column header.

Choose weighting from the question, not from the preferred result. Equal-client weighting may suit a descriptive question about the average eligible client. Session weighting may suit a question about the average recorded session, provided sessions are comparable. Opportunity weighting may suit a question about all recorded opportunities. None of these summaries automatically supports a clinical decision for an individual.

Work the fictional calculations

Dev reviews Client A, whose client-level value is 80% across ten sessions, and Client B, whose value is 40% across two sessions. The equal-client average is (80% + 40%) / 2 = 60%. Its denominator is two clients.

For a session-weighted average, multiply each value by its session count: (80% × 10 + 40% × 2) / (10 + 2) = 880 / 12 = 73.3% after rounding. Its denominator is twelve sessions. Client A contributes ten-twelfths of the weight, so the result sits much closer to 80%.

The session-weighted result is not necessarily the same as pooling raw opportunities. Dev would need each session or client numerator and denominator for that calculation. The table therefore reports:

ClientClient-level valueEligible sessionsA80%10B40%2Equal-client summary60%2 clientsSession-weighted summary73.3%12 sessions

Neither summary is “the true average” without the analysis question and weighting label. Reporting only 73.3% would hide the lower value and limited observation count for Client B.

Handle missingness and unequal follow-up

Track expected, completed, canceled, missing, unusable, and late observations for each client. A client with two sessions may be newly enrolled, may have had interrupted service, or may have missing records. Those conditions influence both precision and weight. Keep the person in the eligible cohort under the predeclared rule and show the observation count.

Do not treat a missing session as 0% or remove an incomplete client after seeing the group value. Set a maturity rule, such as all clients eligible by a cutoff with data through a stated date. If a minimum number of observations is required for a particular summary, justify it in advance and report how many clients remain outside that mature subset.

Differences in phase length create another limit. Ten sessions can span two weeks for one person and three months for another. State the calendar window and service exposure. Avoid combining baseline and intervention values into one client average unless that mixed-phase question is deliberate and clearly labeled.

Graph individual and group results together

Use small client-level panels or raw time-series graphs when the audience needs to see change over time. A separate dot plot can show one declared client-level value per person, with session counts or confidence in the descriptive estimate visible nearby. Avoid a group bar that erases individual points.

A concise report could say: “Client A averaged 80% across ten eligible sessions and Client B averaged 40% across two. The equal-client mean was 60%. A session-weighted mean was 73.3%, with Client A contributing ten of twelve session weights. Individual graphs and missingness are reported separately.” This wording lets the reader understand why the summaries differ.

Interpret within descriptive and causal limits

An across-client average describes the declared cohort under the declared weighting. It does not show that the group is homogeneous, validate the measurement, establish treatment integrity, prove clinical benefit, or isolate cause. Changes in client mix can move the group average even if no individual changes. Unequal observation counts can make a weighted result reflect data availability as much as client performance.

NIST descriptive guidance distinguishes location, spread, and shape. Single-case design guidance and visual-analysis protocols support attention to individual ordered patterns. The BACB ethics hub and CASP public summary supply professional context, while the BCBA Test Content Outline is examination content.

Protect access, privacy, and client meaning

Ask each person, through accessible communication, whether the outcome matters and whether the summary leaves out burdens or unwanted effects. Maintain AAC access under ASHA’s guidance. Descriptive comparison should never restrict food, water, bathroom access, mobility, prescribed care, rest, relationships, or emergency help.

Across-client reports can create reidentification risk in small groups. Limit identifiers, protect extracts, and suppress or combine public cells only under an approved privacy rule. The clinical team should retain access to the individual source record needed for care even when an external report is aggregated.

Clinician review checklist and limitations

Before releasing the summary, confirm:

  • the question names the observation, session, client, or opportunity analysis unit;
  • eligibility, cutoff, definitions, phases, exposure, and source records are compatible;
  • each client’s value, session count, missingness, variability, and time pattern remain visible;
  • equal-client, session-weighted, or opportunity-weighted calculations reproduce;
  • reported denominators match the weights and are not limited to completed cases silently;
  • small groups and client details follow the approved privacy and access controls;
  • client feedback and meaningful-outcome review occurred through usable communication; and
  • conclusions remain descriptive and retain a qualified owner and next review date.

With only two clients, Dev’s group summaries are highly sensitive to either value and cannot characterize a broader population. Session weighting gives influence to service volume, which may itself reflect access or missingness. Equal weighting can give a sparse two-session value the same influence as a ten-session value. Report both when both answer useful questions, and reopen the table when eligibility, source data, weighting, or the cohort changes.

Related resources

Sources