Data standardization, moving the mean

#1
Hi, hope you're all well! :)

I'd be really thankful if someone could help me with a standardisation-code:

I have psychometric test data and had to standardise the values. It worked really well with the R-code "scale". Unfortunately, I can't log the variables any more (in order to improve linearity) because many values have become negative as a result.

I would therefore like to move the scale from 0=mean to 5=mean.

I would really, really appreciate your help.

Best wishes,

Amanda
 

TheEcologist

Global Moderator
#2
How about transforming it first and then standardizing? It's fine to do this.

Although, generally, I would not advise any transformations or standardisations, mostly due to the difficulties in interpreting the results in terms of the original measurements. Log-transformations mean you are now working on a relative scale, and standardization means you are looking at standardizes distances from the mean.. so now you are looking at the relative standardized distances from the mean. Hence, it really doesn't help explaining natural patterns in all cases. This while there are a plethora of methods that don't require transformations or such (e.g. glms, non-parametrics). Why don't you give it some thought, a method that doesn't require a transformation? For science?

Cheers,

TE
 

Jake

Cookie Scientist
#3
As far as completing the task itself is concerned (shifting the values up)... you could use scale(x, center=5), or equivalently, scale(x)+5.
 
#4
Hi!
Thanks for your replies; I tried using scale(x, center=5), but it didn't achieve what i wanted. there was no error message, but the resulting values where not shifted up; there were still a lot of negative values (which shouldn't have been the case).
Does anyone know the code?
I basically have a dataframe column of z-scores that go like:
0.149
-0.137
-0.8
0.23
...

I need to log these values. Of course, it's not possible to log negative values, so I'd have to shift the mean up in order to get scores that are above zero. If I add 5 or some other number to all scores (as suggested by Jake), how would I interpret the regression coefficients?

Sorry, I'm a beginner at these things. :/
Best wishes,

Amanda