Time Sampling

Interval and Momentary Sampling

There are two main types of time sampling—interval sampling and momentary sampling. Interval sampling is a method, wherein the number of times a specific behavior was observed during a predetermined interval of time (e.g., 1 day). For example, a researcher interested in smartphone use may track the number of times an individual unlocks their phone in a 24-hour period. Similarly, a researcher interested in feeding behavior might track the number of times a child eats in a 24-hour period. An advantage of interval sampling is the rich information obtained about the frequency and duration of specifically defined behaviors. A disadvantage is that interval sampling requires researchers and/or collection systems’ undivided and continuous attention. In contrast, momentary sampling is a method wherein the presence or level of a specific behavior is observed only periodically at predetermined moments in time (e.g., each hour on the hour). In momentary sampling, the researcher might record whether or not the phone was being used or whether or not the child was being fed at 0:00, 1:00, 2:00, . . . 23:00. An advantage of the momentary sampling method is the low effort required. The researcher or participant is only engaged at specific points in time and thus does not need to attend to or continuously monitor behavior. The disadvantage is that behaviors occurring in the intervals between assessments are completely missed. In both types of sampling, the length of the assessment intervals or of the intervals between momentary assessments has severe implications for what data are collected and what types of inferences can be made.

Sampling Frequency

Generally, in interval sampling designs, the length of the interval should be much longer than the length of the behavior of interest, so that multiple (e.g., >3) occurrences could occur within any given interval. Importantly, the tallying requires that the behavior has a clearly defined beginning and end. In momentary sampling designs, the [Page 2247]interval between assessments should be short relative to the rate of change of the phenomena of interest. That is, the behavior should span multiple assessments. The Nyquist–Shannon sampling theorem provides an algorithmic framework for considering how often to sample. Formally, a behavioral process is considered as a set of change components that span a finite bandwidth B. For example, activity of the heart has change components that manifest at between 0.5 and 100 Hertz. The sampling theorem states that the continuous signal (i.e., pattern of change) can be fully reconstructed from discrete samples obtained at a sampling frequency (fs) that is at least twice the maximum relevant bandwidth, fs > 2B. That is, study of the heart requires sampling at least 200 times per second (200 Hertz). Indeed, electrocardiogram recordings typically engage precision by time sampling at 500 Hertz. Pushing out a few time scales for further illustration, the Nyquist–Shannon theorem suggests that measurement of a diurnal (24 hour) cycle that manifests as a single, pure sinusoid change component would require sampling at a minimum of twice per day. Capture of more complex (i.e., realistic) functions would require more intensive sampling, especially in the presence of measurement noise.

...