Reliability Analyses: Assessing Tests and Surveys Using SPSS
A common method for determining the quality of a tests or survey is through reliability analyses. Reliability analyses measure how consistent a test is across time, across items or across raters. The easiest type of reliability to obtain is internal consistency reliability, because this type requires administering and scoring a test only once. The basic principle underlying internal consistency reliability is that all of the questions that are supposed to measure something should be related. A test or survey should not just be a set of unrelated questions. For example, if you have twenty-five questions on a statistics test, students who answered one question correctly should be more likely to answer other questions correctly.
The most common type reliability analysis is coefficient alpha, also sometimes called Cronbach's alpha. The alpha coefficient tests how well items measure a single construct, such as verbal ability, mathematics anxiety or brand loyalty.
Obtaining Coefficient Alpha using SPSS
To obtain coefficient alpha using SPSS:
- From the Analyze menu, select Scale.
- From the drop-down menu, select Reliability Analysis.
- Select Alpha from the Model drop-down menu.
- From the list on the left, select each variable you wish to use, and then click the blue arrow to move that item into the Items on the right.
- Once you have selected all the items for your test, click the OK button. You will get the output shown below.
Usually, it is advisable to do more detailed analyses, including inter-item correlations and descriptive statistics for both items and the total score.
If, from the Reliability Analysis menu shown above, you click the Statistics button, the menu below will appear. Simply check the statistics you want and then click the Continue button on the bottom right.
Caution: If you have a lot of variables, the inter-item correlations or covariance can produce a lot of output. For example, if you have 50 items on your test, this will give you 50*50 or 2,500 correlations.
Last updated:
February 22, 2010