Precision, Reliability, Accuracy and Validity
Charles Sturt University
Measurement Error
•All measures within health, sport and physical assessment consists of the actual measurement plus an error.
•When comparing between groups, trials or over time in order to classify and quantify the ‘true’ measure need to minimise or maintain error of measurement
•This is important for accurate data and for research and experimental enquiry.
Charles Sturt University
•Precision-variability in repeated measures on same subject
•Reliability-variability across subjects or measurements
•Accuracy-how close it is to ‘true’ measurement value
•Validity-measures what is intended to be measured
Validity –“am I measuring what I think I’m measuring?”.
Eg.How does a skinfold prediction of % body fat compare with a DEXA result?
-Assessed by ICC (intraclass correlation co-efficient).
Measuring various components requires
Charles Sturt University
•Reliability -“how repeatable is the test from one trial to the next”.
•Calculation of inter and intra-tester reliability enables greater confidence in the accuracy of the results and the ability to detect meaningful changes
•ISAK recommend the following using data collected on 20 different people on different days:
-Skinfold: inter-tester reliability is 10%
-Intra-tester reliability is 7.5%
-Girths & Breadth: inter-tester reliability is 2%
-Intra-tester reliability is 1.5%
•The higher the reliability measurements (%) the lower the confidence in accuracyof results and ability to detect meaningful changes
refer: page 70 of student manual
Reliability
Charles Sturt University
Visual representation of accuracyversus precision
Note these two elements are independent
•Accuracydescribes the nearness of a measurement to the standard or true value
•Precision is the degree to which several measurements provide answers very close to each other
Accuracy and Precision
Charles Sturt University
Errors can occur due to:
•Intra-subject variability -biological, psychological
•Inter-tester variability -error between testers
•Intra-tester variability -error within a single tester
•The test apparatus -error due to set-up or apparatus
•Testers should establish their typical error of measure on 20 test trials. This way, if the change you see when testing is smaller than the TEM, you can figure the change is not a real change.
TEM = Technical error of measurement
Charles Sturt University
Major Descriptive Measures
•Mean -sum of all scores divided by total number of measurements.
•Standard Deviation -square root of the variance
•Descriptors of measurement error
-Technical Error of Measurement
(TEM)
-Coefficient of Variation (CV)
-Intra-class correlation (ICC)
Charles Sturt University
•Determined from repeated measures of the same test by the same tester on the same subject or can be same test, same subject, different tester.
•Is the standard deviation of the repeated measurements taken independently of one another on the same subject
TEM = √[(Σd2) /(2n)]
Where d2= the square of the difference between the two means
Relative TEM
%TEM = TEM/Mean x 100
Technical Error of Measurement
Charles Sturt University
•As a rough guide, normally measures with less than 5% relative TEM are acceptable, over 10% relative TEM becomes problematic.
•Helps describe how measurement error impacts upon the statistical power of the comparison.
•So if a measurement has a high TEM it willincrease thestandard deviation and reduce the likelihood of significant difference.
Technical Error of Measurement
Charles Sturt University
•The standard deviation is sometimes expressed as a percent of the mean, in which case it’s known as a coefficient of variation (variability of a dimension)
•The Coefficient of Variation (%) of a set of values is calculated as:
CV= SDX 100
Mean
•A statistical measure of the deviation of a variable from its mean, is useful to compare the SD to an expected value (given that the value is greater than zero)
•As with TEM, can use CV as a guide to determine whether small changes are likely to be as a result of some intervention, or just normal measurement error/variance
Coefficient of Variation
Charles Sturt University
•Correlations (r) are used to assess the reliability of any repeated measurements by the same measurer
–range from 0 (no relationship ) to 1 (perfect relationship)
•Most common measure of reliability from the same population is an ICC
•Normally calculated from an ANOVA but can be done independently –determine difference at same time.
•An ICC will
-indicate correlation between successive measurements on the same subject
-indicates the ability of the procedure to discriminate between subjects.
Intra Class Correlations (ICC)
The post Precision, Reliability, Accuracy and Validity appeared first on My Assignment Online.