BioPsychoSocial Assessment Tools for the Elderly – Glossary of Terms
Cronbach’s alpha is a generalized formula used to express the internal consistency reliability of a test.
Validity indicated by comparing scores on a measurement with those obtained by applying alternative, equivalent measurements at the same time.
Used when there is no criterion against which to evaluate the validity of a measurement. Validity assessed by comparing the results of several contrasting tests of validity with predictions from a theoretical model.
The extent to which two or more instruments that purport to be measuring the same topic agree with each other. Correlations should be high (closer to 1 than zero) to provide evidence that the items all converge on the same construct.
The extent to which a measurement covers all aspects of the topic it purports to measure.
A measure of association that indicates the degree to which two or more sets of observations fit a linear relationship (range -1 to +1). A correlation close to 0 indicates no association between the observations; as correlations rise, it becomes more possible to predict the value of the second observation from a knowledge of the first. Formula most often used is Pearson’s r for interval or ratio scale level. Kendall’s tau and Spearman’s rho correlation may be used to indicate the association between variables measured at the ordinal level.
Validity indicated by comparing results obtained using a measurement scale with a “gold standard” or indicator of the true situation.
A multivariate statistical procedure that indicates how adequately a set of variables differentiate between two or more groups of people who are known to differ on some characteristic.
The extent to which scores on a measurement distinguish between individuals or populations that would be expected to differ. Need to show that measures that should not be related are in reality NOT related. Correlations should be low (closer to zero than 1).
A standardized indicator of the ability of scores on a measure to distinguish between two groups (typically, cases and controls).
A statistical technique that reduces a large number of interrelated observations to a smaller number of common dimensions or factors. A factor is a cluster of variables that are highly related to each other.
The extent to which results obtained by different raters or interviewers using the same measurement method will agree. Values above 0.85 may be considered acceptable.
In testing the reliability of a measurement, correlation coefficients such as Pearson’s r may be used to compare the ratings of a number of patients made by two raters. Intraclass correlation generalizes this procedure and expresses agreement among more than two raters.
The correlation between each item or question in a health measurement and the total score; used as an indication of the internal consistency or homogeneity of the scale, suggesting how far each question contributes to the overall theme being measured.
As a coefficient of agreement between two raters, kappa expresses the level of agreement that is observed beyond the level that would be expected by chance alone.
The accuracy with which a measurement predicts some future event, such as mortality.
An analysis of validity of a screening test that combines indicators of sensitivity and specificity. A statistical description of the overall performance of the test is given by calculating the area under the ROC curve (AUC). This statistic runs from 0.5 (indicating prediction no better than chance) to 1.0 (perfect accuracy).
The proportion of variance in a measurement that is not error variance. Can be assessed in many ways. Commonly, reliability refers to the stability of a measurement, influenced by the internal consistency of the method. Reliability coefficients range from 0 to 1.0. Good reliability requires scores of 0.80 or higher, when important decisions are to be made with test scores, a reliability coefficient of 0.90 is the minimum.
Spearman’s
rho correlation formula is used to indicate the association between variables measured at the ordinal level, and is termed a “rank order correlation.”
The ability of a measurement method or screening test to identify those who have a condition, in percent. The complement of sensitivity is the false negative rate - number of cases who are falsely classified as not having the condition.
The ability of a measurement to correctly identify those who do not have the condition in question. The complement of specificity is the false positive rate – the number of people the test falsely classifies as having the condition.
The stability or repeatability of a measurement is evaluated in terms of the agreement between a measurement applied to a sample of people and the same measurement being repeated later (one to two weeks later). Assuming that the state being measured has not changed, any change in scores can be regarded as error variance, and hence the level of agreement is used as an indicator or reliability.
The extent to which a measurement method measures what it is intended to. More generally, the range of interpretations that can be appropriately placed on a measure.