ELSEVIER
DISCOURSE ANALYSIS
PROCEDURES:
RELIABILITY ISSUES
KAREN HUX, DIXIE SANGER, ROBERT REID.
and AMY MASCHKA
Department of Special Education and Communication Disorders, UniversiO' of Nebraska-Lincoln
Performing discourse analyses to supplement assessment procedures and facilitate interven-
tion planning is only valuable if the observations are reliable. The purpose of the present
study was to evaluate and compare four methods of assessing reliability on one discourse
analysis procedure--a modified version of Damico's Clinical Discourse Analysis (1985a,
1985b, 1992). The selected methods were: (a) Pearson product-moment correlations, (b) in-
terobserver agreement, (c) Cohen's kappa, and (d) generalizability coefficients. Results
showed high correlation coefficients and high percentages of interobserver agreement when
error type was not taken into account. However, interobserver agreement percentages ob-
tained solely for target behavior occurrences and Cohen's kappa revealed that much of the
agreement between raters was due to chance and the high frequency of target behavior non-
occurrence. Generalizability coefficients revealed that the procedure was fair to good for
discriminating among persons with differing levels of language competency for some as-
pects of communication performance but was less than desirable for others; the aggregate
score was below recommended standards for differentiating among people for diagnostic
purposes. © 1997 by Elsevier Science Inc.
Educational Objectives: The readers will understand the strengths and weakness of lbur
methods of assessing reliability and will become aware of the importance of reliability when
using discourse analysis procedures for diagnostic purposes.
INTRODUCTION
The popularity of obtaining and analyzing samples of children's discourse as
part of language assessment procedures has steadily increased over the past
several years (Gallagher, 1983; Owens, 1995; Stickler, 1987). This popularity
is attributable, in part, to the inherent disadvantages of many standardized lan-
guage measures (Owens, 1995; Damico, 1985a; Naremore, Densmore, & Har-
man, 1995). Put succinctly, discourse analyses are important because stan-
dardized tests alone do not provide sufficient information about how children
use language. However, despite their potential contribution, discourse analy-
Address correspondenceto Karen Hux, Ph.D., 318E Barkley Memorial Center, University of Ne-
braska-Lincoln, Lincoln, NE 68503-0738.
J. COMMUN. DISORD. 30 (1997), 133-150
© 1997 by Elsevier Science Inc.
655 Avenue of the Americas, New York, NY 10010
0021-9924/97/$17.00
Pll S0021-9924(96)00060-3