Hi scholars,
I am trying to study the influence of the Big-5 and narcissism on content engagements on social media. I am using the 10-point / shortened version of the Big-5 (Gosling & Swann, 2003) and the 16-point NPI (Ames, Rose, & Anderson, 2006).
The Big-5 is on a 7 point Likert scale and the NPI is a set of 16 paired statements, where one is the narcissistic response. So, theoretically you get a score between 0 and 16. I am looking for ways to treat both Big-5 and the NPI on a common measure and I need some pov on how personality researchers use them together as predictors since they are on different scales, score differently, and assume different distributions.
Thank you!
Since your predictors are ordinal (you cannot assume that each statement on the NPI is equal to one unit) there is a question of interpretability. One thing you can do is to convert the scores to deciles or quintiles. While categorising variables entails some loss of statistical power, it's not very marked for deciles. In return, you get a predictor that is interpretable : the coefficient is for a 1-decile increment in the score.
How is your predicted variable measured?
Do you want to combine the two measures? If so, and if you have the sample size to do this, you could run a factor analysis on each, then do a subsequent clustering procedure on the resulting factors. This should allow you to do what you want to do.
It is treated statistically by means of the SPSS statistical bag
If you are thinking of running some sort of multiple regression analysis, it really doesn't matter that the tests use different scales. The B weights will be appropriate to each, and the beta weights will always be standardized (similar to correlation coefficients). You have 6 potential predictors, so you should have a good-sized sample - don't try this with N = 50, say. But with a couple of caveats, just go ahead!
One caveat is the issue raised by another respondent: do the scores on these tests qualify as interval measures (suitable for parametric statistics) - or are they ordinal? Frankly, most researchers just "assume" they have interval-level data - if you look at the literature using either of your tests, I'd bet that at least 95% of the papers use parametric statistics.
Another is that although you have a pretty good measure of narcissism, your measure of the Big Five is quite poor. Two items per scale is the minimum needed for any calculation of reliability - or for use in structural equation modeling, where each will be treated as a separate measure. This is the only reason such a scale exists: for use in statistically sophisticated analyses of huge (but crude) data sets. Again, the size of your sample is critical here. So make sure to report the internal consistencies of all the scales. And do not expect strong predictions.