*The same results are obtained on repeated administration of an instrument. This is assessed in 2 ways:
A. Test–retest reliability
: the administration of the same instrument to the same subjects under similar conditions on 2 or more occasions (repeated measures). Tests from different times are compared & expressed as a Pearson r. (remember this test? a statistical test that is calculated to reflect the degree of relationship between 2 variables, -0.3 to 0 to +0.3=no relationship) Used in experimental & quasiexperimental situations.
B. Parallel or alternate form reliability
: can be tested only if 2 comparable forms of the same instrument exist. Subjects take one form of a test then a second form of a test or use one form of a tool then a second form of the tool at a later date.
*The researcher must ensure that each tool is measuring the same concept. The 2 forms of the tool must be highly correlated before being used. Rarely used in nursing as nursing concepts change over time. For this to be reliable the concept would need to be constant.
*Randomly divide the questions on a tool in 2 sets. The 2 forms can be used independent of each other & are considered equivalent measures.
*The reliability of an instrument to measure the same concept. If the items on a tool correlate or compliment each other it is one-dimensional.
*Homogeneity is assessed by 4 methods:
A. Item-to-total correlation
: measures the relationship between each of the items in a tool & the total scale. If a item is not highly correlated it is removed. If the value is > 0.3 it is acceptable.
B. Split-half reliability
: items are split in ½ & compared. This measures the consistency in content of the tool. If the scores of the 2 halves are similar the tool is reliable. Levels should be at least 0.75.
C. Kuder–Richardson coefficient:
this is used when the responses in a tool are dichotomous (yes/no or true/false). Minimum acceptable level is 0.7.
D. Cronbach's alpha
is the most commonly used measure of the internal consistency of a measurement tool.
Each item on the scale is compared with each other.
A total score is then used in the analysis of the data.
You are looking for a Cronbach's alpha of at least 0.7 to indicate a high degree of internal consistency or that the items in a tool are indeed correlated & measuring the concept you are trying to measure.
It is not a statistical test but a measure of sameness or consistency amongst items on a tool.
*Is the consistency or agreement amongst observers using the same measurement tool or the consistency or agreement between alternate forms of a tool.
*Two methods to test this are:
1. Parallel or alternate form
. (see previous slides for this)
2. Interrater reliability:
use when data is an observation. This requires training for sameness in data collectors. Observers can agree in the form of a % that a tool indeed measures what it should or a test can be used called Cohen’s Kappa (this is much more reliable). A Kappa score of 0.8 or > indicates good interrater reliability.