Recode Variables,
Check Reliability, &
Compute New Variables
In Psychological Research, you will often have several items that represent an overall construct. Taken together, these items make up a measurement scale (often referred to as a “measure,” “scale,” or “instrument”).
If you are going to analyze the entire measure (not just individual items), you will need to do several things to prepare the measure for analysis:
1)Make sure that all items are coded in the same direction.
2)Check the Reliability of the measure.
3)Compute a scale score.
(Note: If the measure has separate subscales, follow these directions for each subscale)
First, you want to make sure that all the items are coded in the same direction. If there are items that are coded in the opposite direction, you’ll need to RECODE them.
For example:
The four items below are part of the “Attitudes Toward Research Methods” measurement scale.
1) Research Methods is a good learning experience Strongly Disagree Disagree Agree Strongly Agree
2) I am afraid of Research Methods Strongly Disagree Disagree Agree Strongly Agree
3) I look forward to Research Methods Strongly Disagree Disagree Agree Strongly Agree
4) I love Research Methods Strongly Disagree Disagree Agree Strongly Agree
Notice that a higher score on Q1, Q3, & Q4 = more positive attitudes, whereas a higher score on Q2 = more negative attitudes.
Before we analyze this measurement scale, we’ll need to Recode Q2 so that it is consistent with the other items.
Recoding Variables
On the SPSS Menu Bar, Click
Transform Recode into different variables
The Recode box will open:
A new box will open:
After you click “continue,” you will return to the 1st Recode box.
Click “OK” to update your variables.
Your new variable (Q2r in this example)
will show up at the end of your dataset.
(If you like,yYou can click and drag variables
to a new position in your dataset)
Other uses for the Recode Command
You can use the Recode command for many other transformations.
For example, you might want to recode a variable on a 4-point scale to a dichotomous scale (low vs. high). See Chapter 5 “creating subgroups” for more information.
The next step to prepare your Measure is to check itsreliability.
If you are using a published measure, you may not need to do any SPSS analyses. Information about reliability is usually provided in published journal articles in the Method or Results section. Look for information on test-retest reliability or internal consistency (alpha or split-half).
There are many published measures available, and it is highly recommended that you find an appropriate published measure to use. If you decide to create your own measure (with approval from your professor) or if you make major changes to a published measure, you will need to run reliability analyses.
Coefficient Alpha (or Cronbach’s Alpha or just plain Alpha) will be discussed here. It is a measure of internal consistency – in other words it assesses if all the items measure the same thing.
Reliability Analysis (Coefficient Alpha)
To check the reliability of the Attitudes toward Research Methods scale:
On SPSS Menu bar, Click Analyze Scale Reliability Analysis
The reliability analysis box will appear:
After you click “continue,” you will return to the 1st box.
Click “OK” to run & an output window will appear.
SPSS OUTPUT:
Reliability
Scale: ALL VARIABLES
Making Decisions based on Reliability Analysis
1)If you get an alpha that is a negative number or an extremely lownumber (less then .40), this suggests that you made an error coding your data. You’ll need to check your data and make appropriate corrections. Some this error happens if you needed to recode a variable and didn’t, or if you sent over both the original and recoded variable.
2) If you get an alpha below .70 and your data was correctly coded, check to see if omitting an item will raise the alpha. Repeat reliability analysis without that item, and continue until the alpha reaches an acceptable level. You should also look over the items and see if an item does not seem to fit with the others and consider omitting it.
3) If you get an alpha above .70, your measure has good internal consistency. You should still check individual items to see if omitting an item will raise the alpha.
4) The decision to omit an item should be based on how much it will help to improve the alpha and a review of the item itself. You may decide to keep an item if you believe it is important to your overall construct.
EXAMPLE: In the example on the previous page, we have a pretty decent alpha with all 4 items (.76). However, if we omit Q4 the alpha will increase to .88. This is a big increase! Additionally, Q4 doesn’t seem to fit as well with the other items because it assesses a general feeling of love. Consequently, we would omit it from further analyses.
To omit an item, you don’t actually delete it from your data set. You simply leave it out of future computations. For example, when we compute a scale score (next step), we would leave out Q4.
Compute New Variables
Items in a measure can be combined to represent an overall construct.
(For example, Q1, Q2r, & Q3 of our Research Methods Survey can be combined to represent attitudes toward Research Methods. Based on the reliability analyses, we would omit Q4.)
On SPSS Menu bar, Click Transform Compute
The compute variable box will open:
*Alternatively, you can type q1 + q2r + q3 OR point and click to get the equation.
If you have Missing Data because one or more of your participants left an item blank, SPSS won’t add up (sum) the items for those participants.
Instead of using the “sum” command, you can simply compute a mean score instead.
To compute a Mean score:
In the Numeric expression box, type: mean (q1, q2r, q3)