Hi, Maaru,

I don't think you have any need to transform your data. The relationships between raw data and z-scores will have the same statistical result. Scale 1 looks to be ordinal with only 5 ranks, so Spearman rho is most likely to be the best test of the relationship with the other two. Your 7 subscales can be arguably treated as interval/ratio and Pearson would be somewhat appropriate to compare to the z-scores.

Make sure to check the factor structure of #3. It may be that you do not really have 7 subscales based on eigenvalues using exploratory factor analysis.

Dr. Maaru,

Your question is really in order.  The answer to your question should be  based on the known meaning and characteristics of the  z  statistics. In the first place,  z scores have no units. In the second place, their mean values are zero. So to run an explainable analysis with other variables, it is defensible to transform such other variables to z  score measures. I think that this is just a statistical issue and not particular to SPSD.

Dear Maru,

I agree with what Peter sated earlier. Yet, if the job performance to be predicted by the other two psychometric concepts, you may use Spearman rho, which facilitate comparing the predictability of the two concepts.

Good luck with your reearch

I agree with Profs Peter Kindle and Mohammad Al-Kilani on  the statistical technique that can be used in the analysis.  However, the question asked by the researcher is whether there will be need for transformation of some of  the data before the analysis.    The standard procedure in any statistical analysis is that all data  are first converted to correlations and displayed in multicorrelatoon matrix.  This matrix is now treated as input to any statistical analysis.  So the answer to the question is that any data configuration that can be converted to multicorrelatoon matrix will be appropriate for the analysis. In this case  z  scores may be used as variables of their own  just as the other variables in the equation.  In fact if Principal. Component Analysis were to applied to all the Likert scaled variables the output factor scores will result in an all interval scale study data set. Then linear regression  analysis will indeed mprove the quality of the result of this  research.        

I hope  my discourse is  useful.

Best wishes Mr Maru

Dear respondents,

Thank you very much for your inputs. I am really bad with statistics, that is why I am so confused. My main idea is to test two psychometric assessments for their predictive validity. So, I decided to test it either via correlation analysis or via regression analysis. I was reading in most of the literature, they say that predictive validity should be tested via correlation analysis. Then I was thinking to run correlation analysis, but the idea of doing that when using for example two types of variables - one zscores and the other one likert scale scores is basically ok. And as I understood there is no need to transform any of them? Sorry for such questions, I am just a newbie in statistics.

Thank you in advance

Dear Maru thank you for your  further  explanation.    Snce you had   decided to run a correlation analysis, you will.need to convert your data already  in  Likert scale  to interval  scale before your analysis.  In that case,   Principal Components Analysis will be handy to convert your Likert scale data to factor scores which are automatically interval  scale data also.  

A way of avoiding transformation of data is use of  logistic regression  which you had not planned for.

Technically likert-type scales probably do not meet the equal-interval assumption. However, most researchers just ignore that. An awful lot of self-report measures use likert-type scales, but item scores are summed to yield a total score anyway. So for what it's worth, I'm lining up with those who've said to just go ahead and run your correlations (or regression analyses, which amount to the same thing really).

The only reason to covert everything to the same scale would be if you wanted to compare scores across measures. For example, if you were conducting a clinical assessment of a person's cognitive functioning, you'd want all the tests to be scored on the same scale. But it won't make any difference to your correlation coefficients or beta weights.

Dear Maru,  the response from Prof  Stephen Joy  contains an admission that a   standard assumption of correlation analysis and regression analysis  is intentionally violated.   Both analyses  require  interval data input.  As already suggested,  an  alternative is to do a logistic regression with your Likert scale variables and  z  values.