.- help for ^icc23^ .-

^Calculation of ICC Models 2 and 3^ ^---------------------------------^

.^icc23^ <dv> <classvar> <within_var>, MOdel(#) LEvel(#)

^Description^ ^-----------^

^icc23^ computes the intra-class correlation for random effects models based on > repeated measures ANOVA. These models are ICC[2,1], ICC[2,k], ICC[3,1], and ICC[3,k], a > s described by Shrout and Fleiss, 1979 (see reference below). (For the ICC[1,1] and ICC[1, > k] models based on a one-way ANOVA, see @loneway@). ^icc23^ runs a repeated measures ANO > VA to derive the appropriate estimates and degrees of freedom. In the event there is a sign > ificant F-test for the ^classvar^ (e.g., a significant differernce among raters), the program > will provide the p-value from the ANOVA table.

Data must be in the "long" format. If not, use the @reshape@ command to reconf > igure the data.

Four types of ICC models are considered:

ICC[2,1]: reflects the case where the same group of subjects is rated > by k raters, interest is in the reliability of individual scores. In thi > s model, raters are considered a representative sample of a population of si > milar raters. This model is a two-way random effects model.

ICC[2,k]: is the same approach as ICC[2,1] above, but interest is in t > he reliability of the MEAN score, rather than among single observations.

ICC[3,1]: reflects the case where the same group of subjects is rated > by k raters, interest is in the reliability of individual scores. In thi > s model, the only raters of interest are those participating in the study (e.g > ., there is no intention of generalizing the raters' scores to a larger pop > ulation of raters). This model is considered a "mixed" model (subjects are rando > m, raters are fixed). ICC[3,k]: is the same approach as ICC[3,1] above, but interest is in t > he reliability of the MEAN score, rather than among single observations.

Three inputs are required:

dv is the dependent variable

classvar the class variable refers to factor that is repeated within > subjects, e.g., raters, devices, time points, etc., e.g., the variable which would b > e entered as the repeated() variable in the ANOVA option.

within_var refers to the "within subject" variable, e.g., subjects bein > g assessed

^NOTE: The order of entry of the variables is critical!^

^Options^ ^-------^ ^MO^del(#) refers to the type of model to be estimated. ICC model 2 is > the default ( producing ICC[2,1] and ICC[2,k] estimates). ^LE^vel(#) the degree of precision of the confidence interval, entered > as a decimal. The default is 95%, i.e., level(.95)

Examples --------

For ICC[2,1] and ICC[2,k]: Two-way random effects (Subjects and raters are cons > idered to be sampled from larger populations); 95% CIs are assume > d. ^.icc23 score rater person_id^

For ICC[3,1] and ICC[3,k]: Two-way mixed model: Subjects are random, but raters > are fixed (i.e., the raters are not considered a sample -- the > y are the only raters of interest); 90% CIs are requested. ^.icc23 score rater person_id, model(3) level(.90)^

References ---------- Shrout PE, FLeiss JL. Intraclass correlation: uses in assessing rater reliabili > ty. Psychol Bull, 1979; 86: 420-428.

Portney LG, Watkins MP. Foundations of Clinical Research: Applications to Pract > ice (2nd ed.). Prentice-Hall, Inc: Upper Saddle River, NJ., 2000.

Author ------ Paul F. Visintainer, PhD Baystate Health System Springfield, MA 01089 visint46@gmail.com

Luis C.Orozco, MD, MSc Facultad de Salud Universidad Industrial de Santander Colombia lcorovar@gmail.com

Also see -------- Manual or on-line help for: @loneway@, @reshape@, @iclassr@, @iclassr2@