Reliability Analyses Using SAS
What do we mean by reliability? Reliability of a measurement is the degree to which it is consistent, across items, across raters, across time. There are several types of reliability analyses that can be conducted using SAS. Calculating Inter-rater or Test-retest Reliability Inter-rater reliability and test-retest reliability are quite simple, statistically speaking. The researcher obtains two measures, either the same test scored by two different raters (inter-rater reliability) or tests on the same person at two different times (test-retest reliability). For either type, it is a simple matter of correlation to determine the reliability coefficient, as follows:Proc corr data = datasetname ;Calculating Internal Consistency Reliability Surveys, tests and questionnaires are used across diverse fields, from health care to engineering to education. How reliable are those forty questions? How well do these items measure proper nutrition, product durability or knowledge of mathematics? One common way to assess the usefulness of a measure is “internal consistency reliability”. In brief, the assumption of internal consistency analysis is that if 40 items all measure the same underlying concept, e.g., product durability, that there should be correlation among these items. To obtain a Cronbach alpha internal reliability coefficient, the following program is used:
Var variable1 variable2 ;
Proc corr alpha data = datasetname ;Although it is not required, it is generally desirable to include the nocorr option in using alpha; otherwise, with, for example, 40 items, a 40 x 40 matrix of 1,600 correlations will be produced along with the reliability analyses. Let’s assume we are trying to evaluate family environment. We rate each family on 40 items:
Var list of variable names ;
- Number of meals the family eats together on an average day.
- Number of times during the one-hour interview the respondent makes a positive comment about a family member
Proc corr data = lab.family alpha nocorr ;Click here to view the results of the SAS alpha procedure. As you can see by the attached output, even without the correlation matrix, a lot of numbers are generated. The first page is descriptive statistics for all of the variables. Coefficient alpha is found on the second page, for both raw and standardized variables. On these last two pages, you will also find, for each variable, the correlation with the total of the raw and standardized variables and the alpha if the variable is deleted.
Var q1- q40 ;
Last updated:
September 16, 2008