Showing papers on "Intraclass correlation published in 1982"

Statistical methods in ophthalmology: an adjustment for the intraclass correlation between eyes.

Abstract: For the cases of normally- and binomially-distributed outcome variables, methods are presented for analyzing ophthalmologic data to which a person may have contributed two eyes worth of information, the values from the two eyes being highly correlated. A frequently-used method of analysis, where each eye is treated as an independent random variable, is shown to be invalid in the presence of intraclass correlation: it yields true p-values two to six times as large as nominal p-values when realistic assumptions are made about the degree of correlation between eyes. These results may be applicable to other medical specialties, such as otolaryngology, where highly-correlated replicate observations are obtained from individuals.

The reliability of axis II of DSM-III.

Abstract: The authors studied the reliability of axis V of DSM-III by analyzing ratings of 97 psychiatric inpatients made by a multidisciplinary team of clinicians. The intraclass correlation coefficient for ratings of the overall sample was .49, lower than the figure found during the DSM-III field trials. The psychiatric diagnosis, age, ethnicity, marital status, and sex of the patient did not significantly influence reliability. The authors recommend more explicit instructions in the administration of the scale, the use of a standardized interview, and more training for raters as ways of increasing reliability.

Design considerations in the estimation of intraclass correlation

Abstract: SUMMARY The design of family studies to estimate the value of an intraclass correlation coefficient p is considered when ni individuals are to be selected from each of k families, i = 1, 2, …, k. In particular, the accuracy of a balanced design (ni=n, i = 1,2,…,k) for estimating p is compared with the accuracy of an unbalanced ‘natural’ design, in which the ni are sampled at random from family size distributions that tend to occur in practice. It is found for two different estimators of p that the balanced design is usually preferable, but only to a small degree if the number of families sampled is greater than 50.

Measuring Pairwise Agreement Among Many Observers. II. Some Improvements and Additions

Abstract: Weighted kappa was defined as a measure of pairwise interobserver agreement for the case where the observers judging one subject are not necessarily the same as those judging another subject. In this paper improved formulas for the large sample variance of the weighted kappa statistic are derived, a new definition of interclass kappa coefficients is suggested, and the intraclass correlation coefficient is shown to be a special case of weighted kappa.

On the estimation and testing of interclass correlations: the general case of multiple replicates for each variable

Abstract: Methods have recently been developed for the estimation and testing of mother-child correlations. In this report, these methods are extended to the general case of assessing interclass correlations where multiple replicates are allowed for each of the two classes of individuals under consideration. An algorithm is presented for obtaining the maximal likelihood estimator and an asymptotic test of significance is provided. In addition, a computationally convenient significance test is derived based on the pairwise estimator whereby one estimates the effective number of degrees of freedom in a family as a function of the number of replicates and the estimated intraclass correlation for each of the two types of individuals and sums up the effective degrees of freedom over all families in the sample. These methods are shown to be applicable to more general situations than the analysis of familial data, including the assessment of correlations between two variables measured at one point in time or the same variable measured at two points in time.

Some problems concerning the reliability and structure of the scales in the inpatient multidimensional psychiatric scale (IMPS).

Abstract: Two problems concerning the IMPS are revealed. First, the importance of an appropriate reliability measure is demonstrated in a sample of 124 inpatients. Widely divergent results are found using three different intraclass correlation coefficients. The one producing the highest results was used by Klett and McNair (1966) and by Behrends et al. (1971). However, the results from this measure only apply to certain types of investigation. A measure more applicable to most investigations in psychiatric research (the intraclass correlation coefficient of individual ratings with between-rater variance included) produced results that were clearly lower, although still acceptable, with an average correlation of 0.72 between the 12 scales. Secondly, the multiplication of item scores by 2 or 8 when forming the scale scores is shown to lead to unusual and undesirable scale distributions in a sample of 1932 inpatients. The resulting implications for assessing change scores on some of the IMPS scales are discussed.

Nonparametric measures of intraclass correlation

Abstract: Three nonparametric measures of intraclass correlation based on the notion of concordance are considered. Their unbiased estimators and nonparametric tests based on the estimators are studied and it is shown that an analogue of the Kendall's tau provides small variance estimator and relatively powerful test. Furthermore, the approximate variance of the estimator is given when the correlation is small in the normal model.

Testing multivariate means when the covariance matrix has intraclass correlation structure

Abstract: Fisher's method of combining independent tests is used to construct tests of means of multivariate normal populations when the covariance matrix has intraclass correlation structure. Monte Carlo studies are reported which show that the tests are more powerful than Hotelling's T 2-test in both one and two sample situations.

Intraclass correlation: Estimation of the reliability of ratings

Abstract: The interactive FORTRAN program INTRACORR calculates intraclass correlations; both maximum likelihood and unbiased estimates of the population correlation are calculated. These estimates are available for individual measurements and for the mean of a set of measurements. An option identifies the number of measurements needed to obtain a correlation coefficient of some specified magnitude. The program was written in FORTRAN IV-plus for a Digital Equipment Corporation VAX-I 1/780. Intraclass correlation is a general approach for determining the reliability or agreement of a set of observations. The approach in its various forms uses the meansquare terms generated by a repeated-measures analysis of variance to estimate true score and observed score variability and, provided the proper assumptions are met, gives a measure directly interpretable as a reliability coefficient. Since Fisher (1958) first introduced the notion, it has undergone considerable development by a number of different authors (Bartko, 1966, 1976; Ebel, 1951; Gulliksen, 1950; Horst, 1949; Shrout & Fleiss, 1979; Winer, 1971). Several versions exist, each of which assumes a different linear model under which the variance components are estimated. Three of these versions, explicated by Shrout and Fleiss, are calculated by INTRACORR. A brief description of each of these models is presented in this paper to help the reader evaluate the program. In Modell, each case is rated by a different set of k judges, assumed to be sampled from a larger population of judges. A rating of the jth individual by the ith judge can be represented as follows: x(ij):= m + bG) + w(ij), in this case, m = the overall population mean of ratings, blj) =the effect associated with the jth case, and w(ij) =the combined effect of the ith judge, the interaction of the ith judge with the jth case, and an error component associated with the ijth observation. Model 2, described by Bartko (I 966) and Shrout and Fleiss (1979), is appropriate to the situation in which each of a set of k raters views all n cases. Like Modell, raters are assumed to constitute a random sample from some population of raters. The underlying model for the ith judge's rating of the jth case is x(ij) =m + btj) + r(i) +br(ij) +e(ij). Here, m and bG) are defined as before, rei) =the effect of the ith rater, br(i j) := the effect of the ith rater with the jth case, and e(ij) =an

Measures of Multivariate Relationships for the Uniform Covariance Case

Abstract: Simple procedures for obtaining the m.l. estimates of multiple correlation, canonical correlation, partial canonical correlation and bi-partial canonical correlation have been derived for situations where the underlying covariance structure is uniform (equal variances and equal positive covariances). It has been shown that for any even-sized p × p symmetric R matrix with intra-class correlation structure, the largest canonical correlation is identically equal to the maximum eccentricity of the p-dimensional correlation ellipsoid. Using an empirical variance con variance matrix based on attribute ratings of coffee, the computational short-cuts in the m.l. estimation of various measures of multivariate relationships have been demonstrated.