Constructing Scales and Checking Their Reliability
There may be times when you wish to combine several variables that focus upon a related topic into a scale.
For example, the 2008 General Social asks specific “yes/no” variables regarding opinions on abortion under the following circumstances: defect of the baby (Abdefect), health of the mother (Abhealth), rape (Abrape), when a married woman wants no more children (abnomore), and because a woman is single (Absingle).
You can easily construct a scale about general opinions on abortion using SPSS. First, examine the questions that you intend to use to create your scale. Do all of the variables identically code the responses (e.g. “yes” = 1, “no” = 2)? Are some questions positively worded and other negatively phrased? Sometimes question wording is reversed to ensure that people are paying attention to the question. If you find that the questions are not the same in either coding or phrasing, go back and recode those variables so that the codings are identical. Also make sure that you mark all missing answers with the same code as well. (e.g. not applicable = 0, “don’t know” = 8, missing = 9).
Once you have addressed these issues, you can create a new scale variable that combines responses to the five specific questions. In Data View, go to: Transform, Compute Variable. In the field “Target Variable”, type the name of your new variable (AbTotal). Click the button “Type and Label”. Enter the label that you are giving to your new variable and make sure that the “numeric” box is checked. Highlight the first variable that you want included in your scale variable, click on the arrow and then use the “+” sign on the calculator. Go back and add each variable that you want included in your scale by using these steps. Your screen should look like this:
Click OK. Your new variable will be listed at the end of your variables. Remember to go to the Variable View and define your values of your new, recoded variable!
Reliability
Once you have created a scale, you should test to see if it is reliable; that is, to see if the scale items are internally consistent. The most commonly used test is Cronbach’s alpha coefficient. You can assume reliability if the coefficient is greater than .7. Be aware that the Cronbach test is highly dependent upon the number of items in the scale (especially less than 10).
To check on the reliability of your scale: click Analyze, Scale, and Reliability Analysis. Then move all of the items of your scale into the box “Items”. Make sure that “model” is Alpha. Type the name of your new variable into the “Scale Label Box”. Click on the “Statistics” box. It should look like this:
Then click Continue and then OK. The first part of your output will look like this:
Cronbach’s Alpha is greater than .7, so this scale has relatively good internal reliability. You will also want to check the Inter-Item Correlation box to ensure that all of the values are positive. This ensures that all of the variables in the scale are measuring the same characteristic. For the variable Abtot, here is that output:
Please note that these are basic tests to see if your scale is internally reliable. For additional information, I recommend that you refer to a good statistics book.
Please note: If you need to request accommodations with content linked to on this guide or with your SPSS Software for your coursework, on the basis of a disability, please contact Accessibility Resources and Services by emailing them. Requests for accommodations should be submitted as early as possible to allow for sufficient planning. If you have questions, please visit the Accessibility Resources and Services website.
SPSS eTutor by Dee Britton is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.
