Measurement and Data Collection Flashcards

A behavior analyst is measuring vocal stereotypy using a 10-second partial interval recording system across a 30-minute session. The behavior occurs continuously for the first 5 minutes, then stops entirely. What will the data MOST likely show, and what is the key limitation?

Answer: 100% interval occurrence for the first 5 minutes and 0% thereafter; partial interval overestimates behavior duration when intervals are long relative to behavior length

Partial interval recording scores an interval as positive if the behavior occurs at any point during it. Since the behavior was continuous for 5 minutes (30 intervals), those all score positive, yielding 100% for that block and 0% after — accurately capturing presence/absence per interval. However, partial interval is known to overestimate duration because even a brief instance in a long interval scores the whole interval. The 16.7% figure is incorrect because PIR doesn't average across the session that way; it reports interval-by-interval. And while PIR does only yield percentage data (not rate), this is a secondary point, not the primary limitation identified in the scenario.

A technician is using momentary time sampling (MTS) every 60 seconds to measure on-task behavior during a 20-minute work period. Unknown to the technician, the client exhibits a cyclical pattern of on-task behavior that peaks and troughs every 60 seconds in near-perfect synchrony with the observation intervals. What is MOST likely to result?

Answer: The data will be systematically biased, either drastically overestimating or underestimating the true prevalence of on-task behavior

This is the aliasing problem — when the sampling interval synchronizes with the natural cycle of a behavior, MTS can capture only the peaks or only the troughs, producing a severely distorted picture. If the technician always samples during the on-task peak, prevalence will be vastly overestimated; if always during the trough, it will be vastly underestimated. This is a genuine and documented threat to MTS validity. The law of large numbers (choice B) applies to random error, not systematic bias. The error is not random (choice C) — it is systematic and directional. While rare, cyclical behaviors do occur in applied settings (choice D), making this a relevant consideration.

Two observers independently score the same session of self-injurious behavior using frequency recording. Observer A records 24 occurrences; Observer B records 16 occurrences. What is the interobserver agreement (IOA) calculated using the smaller-over-larger method, and under what circumstance would this method be MOST misleading?

Answer: IOA = 66.7%; it is most misleading when both observers record high total counts but disagree about which specific intervals or occurrences they observed

Smaller-over-larger IOA = (16/24) × 100 = 66.7%. The critical limitation of this method is that it only compares totals, not which specific occurrences were agreed upon. Two observers could both record 20 instances but disagree on every single one — recording different instances — yet achieve 100% IOA by this method. This makes it most misleading when total counts happen to be similar but actual moment-to-moment agreement is low, as can occur with high-frequency behaviors. Choice B describes a real limitation of frequency IOA generally (low-N sensitivity), but it is not the specific weakness of smaller-over-larger. Choices C and D contain computational errors.

A researcher wants to measure response latency for a child's compliance with instructions. After piloting the measurement system, she finds that latency values cluster bimodally: most responses occur within 2 seconds or after 30+ seconds, with very few responses in between. Which measure of central tendency should she AVOID reporting as the primary summary statistic, and why?

Answer: The mean, because in a bimodal distribution the mean falls in the sparse middle region and does not represent either typical response pattern

In a bimodal distribution, the mean is pulled toward the center of the distribution — the region with the fewest actual data points. Reporting the mean would suggest a 'typical' latency of, say, 15 seconds, when in reality nearly no responses occur at that latency. This fundamentally misrepresents the data. The median (choice B) is actually more robust than the mean for skewed or bimodal data, though it also has limitations. The mode (choice C) can be reported as two separate modes, which is informative. Range (choice D) has no distributional assumptions. The mean is uniquely deceptive here because it describes a central tendency where no data exists.

During baseline, a client's aggressive behavior shows a highly variable, ascending trend. A behavior analyst initiates an intervention in the next phase. After 5 sessions, aggression is lower on average than baseline but still shows an ascending trend. Which conclusion is MOST defensible from a behavior-analytic data interpretation standpoint?

Answer: The intervention's effect is ambiguous; the ascending trend indicates that if the current trajectory continues, aggression may eventually exceed baseline levels

Visual analysis in ABA examines level, trend, and variability both within and between phases. While the intervention produced a level change (lower mean), the persistence of an ascending trend within the intervention phase is a critical warning sign: if this trajectory continues, behavior will worsen past baseline levels. A defensible conclusion acknowledges both the apparent benefit (level) and the threat (trend). Choice B is incorrect because it ignores the trend entirely. Choice C overstates the requirement — a level change alone can be meaningful, but here the trend undermines it. Choice D is incorrect; the BACB does not mandate a specific minimum number of data points, and guidelines emphasize interpretive rigor, not arbitrary thresholds.

A BCBA is training a new RBT to take ABC (antecedent-behavior-consequence) data using a narrative recording format. After two training sessions, the RBT's ABC narratives consistently omit antecedents that occurred more than 30 seconds before the behavior and fail to record behaviors that last less than 3 seconds. These errors BEST illustrate which two distinct measurement threats?

Answer: Limited observer sensitivity (threshold effects) and response restriction artifacts from the recording format

The two errors represent distinct technical threats: (1) Omitting antecedents beyond 30 seconds reflects a temporal threshold effect — the observer's sensitivity window is limited, causing systematic exclusion of distal antecedents that may be functionally relevant. (2) Missing behaviors under 3 seconds reflects a duration threshold — the recording format or observer processing speed creates a floor below which events are not captured. Together, these are 'observer sensitivity limits' and 'response restriction artifacts.' Choice A (observer drift) refers to gradual degradation over time, and reactivity refers to subjects changing behavior when observed — neither fits. Choice C describes definitional problems, not observer-level measurement threats. Choice D describes reliability and fidelity issues, which are separate constructs from the perceptual/temporal gaps described.