Generalizing from segments to sessions: Should it be done?

Abstract

This article addresses two methodological questions that are crucial for research on the natural language in the counseling process: (a) What fraction, or segment, of an interview best represents the entire interview? (b) Do different starting points yield different results? Using a variety of data sets and process measures, we tested segment generalizability in different formats of counseling (individual and family) and with different coding systems (of client and counselor behavior). Results from three studies suggested that, whereas starting points made little difference, generalizing from any size segment to a whole session should not be done when interviews are examined individually. On the other hand, when interviews are aggregated, even fairly small segments can be sampled reliably. In making their own sampling decisions, investigators should consider theoretical and clinical significance and conduct subsample generalizability tests.This article addresses two methodological questions that are crucial for research on the counseling process: (a) What fraction of an interview, if any, best represents the entire interview? (b) At what starting point in the interview should this "best" sample be drawn?Increased awareness of the richness in case study data has rekindled interest in conducting detailed content analyses of the natural language of counseling. The smaller-is-better perspective (Greenberg & Pinsof, 1986) has resulted in a number of studies in which thousands of client or counselor statements are categorized along one or more dimensions (e.g.

In Vivo Feedback Predicts Parent Behavior Change in the Attachment and Biobehavioral Catch-up Intervention

“…In the current study, in which the clips were assigned to both independent coders and clinicians for use in commenting-focused supervision, this length of measurement was also considered feasible by both parties. Though process variables coded from short segments may not accurately represent individual sessions, aggregating short segments results in a reasonably accurate estimate of overall process (Friedlander et al, 1988). As such, session 1–5 data were aggregated at the case level, as described below.…”

“…These composite variables were intended to increase reliability and accuracy of in the moment feedback measurement by averaging across multiple observations (cf. Friedlander et al, 1988). The decision to average variables across sessions 1–5 was guided by Kazdin and Nock’s (2003) discussion of establishing a timeline in mediation/mechanism research.…”

Reflective listening and open‐ended questions in counselling: Preferences moderated by social skills and cognitive ability

“…The audio-taped transcripts were an average of 4 min 33 sec long. It might be argued that these sections were too short, but Friedlander, Ellis, Siegel, Raymond, and Haase (1988) have shown that for aggregated sessions even short segments can be representative.…”

The Relationship Between Counselor Verbal Response Modes and the Working Alliance in Career Counseling

“…Preparation ofthe audiotapes. Because the unit of investigation in this study is aggregated data and not individual sessions, it is recommended that researchers use segments of sessions rather than entire tapes (Friedlander, Ellis, Siegel, Raymond, & Haase, 1988). Friedlander and her colleagues used a variety of data sets and process measures to test segment generalizability in different formats ofcounseling with different coding systems (including the system used in this study).…”