i am planning to check the effectiveness of a teaching programme. so i prepare a questionnaire for assessing the effectiveness . is it necessary to find the internal consistancy.
Well, you'll need to know whether your questionnaire is reliable and valid. Therefore it is necessary to investigate the reliability, but Cronbach's alpha is seldom a good idea.
Here is a link to an article which explains this:
www.unt.edu/rss/class/Jon/Benchmarks/Omega_JDS_Jun2012.pdf
Regards, Karin
With the information that is provided, suggesting the best approach to determining reliability is difficult. Generally, the Cronbach Alpha is what you do when a more desirable approach is not available. If you, indeed, are using a checklist, it is most likely that Alpha would NOT be the best. Assuming you will be using a checklist and your sample size is large enough, I suggest that the simplest procedure would be to randomly divide your sample into two samples and correlate the percentage of checks for the two samples. Be sure to square your correlation coefficient to obtain the percentage of variance accounted across samples. This would at least get you an estimate of the consistency.
What Dr. McLean has suggested (randomly dividing a sample followed by the correlation calculation) is what I have done at times for a quick estimate of the score reliability of a test. Note I said "score reliability of a test" and not "reliability of a test." Also, the Starkeweather UNT short paper, noted by Frau Schermelleh-Engel, is certainly a good resource to review.
It depends on the assumptions about the items, the main problem with alpha is that it is biased by the number of items and the tau-equivalence assumptions (every loading is equivalent).
There are several options to assess reliability under many when the assumptions are not met by the data, a recent paper in the SEM journal adress this problem:
http://www.tandfonline.com/doi/full/10.1080/10705511.2014.938597
The ultimate question is, of course, the degree of "exactness" needed for any measurement. If the question deals with only a quick estimate, then one would not check for violations of, for instance, homogeneity of variance, additivity, or normality. If a higher level of exactness is needed, e.g. the SAT, then one would spend approximately $1.5 million dollars a question to develop the test instrument, and taking the test from the test room would be a felony. All the comments given are excellent and again: the question deals with the level of exactness sought in the measurement.
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS