## How do I create a scale in SPSS when I have items that are formulated in the opposite direction?

Often, you won’t be interested in the questions you ask participants as such, but you are interested in the latent construct that is indicated by a number of variables. In this case you will want to create a scale variable, constructed from your items. To do this you will take the average score of the questions. This means that the scores must all have the same valence: if you score higher on one item, you will score higher on the other. In other words, the correlations between items must be positive.

If you have items that are formulated negatively, e.g., that have negative factor loadings in a factor analysis or negative item-scale correlations in a reliability analysis, you should first recode them in the opposite direction.

Consider for instance the following two items van a need for cognition scale (Cacioppo & Petty, 1984):

1. I would prefer complex to simple problems    (ranging from 1 to 7)
2. Thinking is not my idea of fun   (ranging from 1 to 7)

Need for cognition is an internal trait that consists of people’s preference for knowing things. Statement 1 is positively formulated, and statement 2 negatively. So the lower one scores on statement 2, the higher the score is for need for cognition. You therefore want to recode the scores for statement 2 in such a way that 7 becomes 1 and 1 becomes 7. You can do this with TRANSFORM –> RECODE INTO DIFFERENT VARIABLES (7=1; 6=2; 5=3; 4=4; 3=5; 2=6; 1=7).

When you’re done recoding, you must ensure that your variables measure one construct and that they form a reliable scale. To see of how many components your scale consists, you perform a factor analysis. To measure the reliability of your scale you calculate Chronbach’s Alpha or if it’s just two variables a correlation.

When you have taken these steps and your scale seems to be one reliable scale, you can construct your scale variable.

In SPSS, go to ‘Transform’ –> ‘Compute variable’ –> [target variable=scale name] [numeric expression: MEAN(VAR1, VAR2, VAR3, VAR4) ]