Stepwise regression will focus on maximizing R^2 with significant predictors. Alternatively I would recommend dominance analysis or relative weights analysis. These procedures will provide an estimate of how much each variable contributes to the overall R^2. See:

Tonidandel, S., & LeBreton, J. M. (2011). Relative importance analysis: A useful supplement to regression analysis. Journal of Business and Psychology, 26(1), 1-9.

Thanks for your suggestions Gary. I will consider your suggestion about dominance analysis or relative weights analysis. However, I am not sure that SPSS which I commonly use provides these alternative procedures. I would rather prefer a type of regression analysis among multiple regression, hierarchical regression or stepwise mulitple regression in order to find the contribution of each variable to Rsq but I don't know which of them is more appropriate for the forementioned research purpose

I recommend stepwise MR when you just want to maximum R^2 but are not really concerned about what predictors are in your final equation. If you want to leave in as many variables as possible, then backwards, stewpwise regression.

How many variables are you working with? Both dominance analysis and relative weights are done with SPSS, but require additional work. With dominance analysis, which only works with a small number of predictors, you run the regression analyses in spss and put the information in an excel spreadsheet. With relative weights analysis you will be working with a spss syntax file to get the relative weights analysis.

Feel free to contact me at gary dot burns at wright dot edu.

Many thanks Gary. Currently, I am just at the first steps of my doctoral research and as you realize I have not totally designed all the procedures and methods which I am going to follow in the study. However, the predictors I will test perhaps will be six or seven (e.g. hours of practicum, years of training in counseling, number of clients in practicum, years of counseling-related experience, core self-evaluations scores, years of total education). Dependent variables are two: counseling self-efficacy and career counseling self-efficacy.

Me and my advisors came up with the idea to use mixed methods starting with the quantitative part of the study to find all possible predictors of counseling self-efficacy and career counseling self-efficacy testing relative bibliography and afterwards, running a qualitative research to explore deeper the factors shaping trainees' self-efficacy beliefs, using either a sequential explanatory mixed methods strategy or a concurrent embedded strategy.

Concerning the quantitative part my main concern refers to the limited sample. I consider that at most 120-150 trainees will participate in the study due to the fact that only career counseling trainees will be included in the survey so as to avoid statistical biases which may occur when comparing different groups of counseling psychology specialties (e.g. school counseling, multicultural counseling, family counseling etc).

As far as I am concerned I can use multiple regression procedures when n/k is at least 40/1 (see Wampold & Freund, 1987). This is one of the main reasons I asked this question on the site.

Many thanks for your time and useful recommendations.

Best regards,

Andronikos

Consider what you're asking for a moment. You want to find predictor variables of the most influence among your set in order to plan much further research and yet are trying to find the minimum number of subjects for that plan. That's equivalent to saying, what's the most variable estimate of my parameters I can get away with? I'd instead be looking around to find the way to get the least variable estimates I reasonably could.

Stepwise multiple regression should never be done. It finds biased relationships.

Perhaps see this page for a very simplified explanation.

http://stats.stackexchange.com/questions/10079/rules-of-thumb-for-minimum-sample-size-for-multiple-regression

thanks John..that was helpful.

The question is interesting and the answers most valuable. But I feel like you rely too much (too quickly) on statistic to answer your question. Statistic is just a tool and shouldn't replace well grounded hypothesis based on a review of the literature. What says the literature about the variables that are supposed to explain or influence your construct? If the aim is to explore and look for variables linked to a construct (that we don't know of), interviews seem to me the first method to use; that way you don't miss too many variables while exploring the subject.

@Sophie Berjot

Many thanks Sophie..I will carefully examine your view but it is likely that I will conduct a quantitative survey first and when survey has been completed I will then do a qualitative study attempting to strengthen and complement survey's results. Perhaps this procedure could enable me on the one hand to test hypotheses about relationships among the variables, and on the other hand find out new dimensions which have not been identified yet in the literature. In case I decided to do mixed methods research what would you suggest to start with, a quantitative study or a qualititative one (e.g. semistructured interviews)? You said you would prefer to use interviews first, however, literature has generated some well grounded hypotheses so far for the relationships among counseling self-efficacy and "independent" variables such as practicum length, experience in counseling, years of training etc.

..... so I suppose it would make sense to start with the quantitative survey, formulating a sequential explanatory design.. What' s your view for this?