In my recent study I had one independent variable (IV, 2 levels) and a number of different explanatory variables and one response. The aim was to assess the relationship between teamwork (response - continuous), presence (continuous), stress (3 continous variables - distress, engagement, worry) and immersion (the IV). This is really a sanity check for myself. I conducted partial correlations to ascertain the variance in teamwork accounted for by presence and stress without the shared variance of immersion (so the IV was used as a control in the partial correlation). Is the use of the IV "allowed" in partial correlations or is there a better way of achieving this? I am pretty new to stats so any help would be greatly appreciated. Thank you!
I suggest you follow the attached easy and comprehensive books for statistical data analysis.
Muhammad Ali the analysis I am double-checking is based on comprehensive reading of Andy Field's book. The books you have attached seem to be focused on data analysis for machine learning so I don't think that would be suitable for my needs. I only wish to double-check that the use of the independent variable as control in a partial correlation is "correct". Thank you!
Hi Beata Balint , I agree, the books and machine-learning approach suggested in another answer are not adequate for what you want to achieve.
About your analysis, if I understand you properly, I would say it is indeed appropriate to use your IV as control if you believe that the IV is exogenous from teamwork, i.e. if you expect that teamwork is affected by the IV, but the IV is not (or should not) be affected by teamwork.
I would say also that, given your problem at hand, Partial Least Squares/Structural Equation Modelling will be an interesting option to try. If you want more advanced and extremely nice methods, Bayesian networks is an interesting option, too.
Rolando Gonzales Thank you! Yes, the IV in this case is the set-up that people use (so there's 2 experimental groups in my case). Somebody calmed me out about using the IV as control since they've never seen that done before, however, from my understanding it was a legitimate decision (based on what I've read in books etc). Structural Equation Modelling would indeed be fantastic, however, my sample size is small (n = 24) and the observations are not independent (there is a hierarchy present which is why I decided to do correlational analysis and look for trends in the data to inform the hypotheses for a study with a bigger sample size that can accomodate multilevel modelling).
Beata Balint , you're welcome, happy to help :)
(One final comment: Bayesian networks are like SEM, but Bayesian methods are appropriate -work well- both in large sample and also in small samples like yours.)
"Is the use of the IV "allowed" in partial correlations ?" ..... this may be the wrong question in your context. It is "allowed", but you need to think about what the underlying models behind partial correlations are doing. These are linear model. You have two groups and you would be comparing two linear predictors, in one the existence of separate groups is ignored and in the other the existence of the two groups is represented by a constant shift between the two groups, but other coefficients remain the same, The most extreme alternative to this would be to have separate regressions for the two groups and then to combine the residual sum of squares.
I suggest that you should start by constructing a set of scatter plots of your variables in which you use different symbols to identify your two IV groups,
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics