Research objectives:
1.To examine the nature and strength of relationship between:
a. Connectedness to nature and subjective well-being.
b. Connectedness to nature and environmental identity.
c. Connectedness to nature and pro-environmental behavior
d. Environmental identity and subjective well-being.
e. Environmental identity and pro-environmental behavior.
2. Connectedness to nature and environmental identity as predictors of subjective well-being and pro-environmental behavior.
All my variables are interval in nature (they are basically test scores)
For the first research objective I intend to use pearson correlation coefficient.
For the second I intend to use multivariate multiple regression ( because 2 predictors and 2 criterions ).
Why I am again doing regression after already finding out the strength and direction of relationship between the variables?
Because I want to know which is a better predictor of subjective well-being and pro-environmental behavior.
Am I being practical about this? Is the second research objective pointless? Is it simply making the problem more cumbersome and making me appear like an illogical researcher?
Do I just exclude the second research objective?
Any help from fellow researchgate members would be greatly appreciated!
Hi Shilpa,
correlation and regression analysis are basically the same. More precisely, if you assume a linear relationship between 2 variables and use the Pearson correlation coefficient R, the "explained variance" R2 of the regression analysis is simply the square of R. Therefore you may think that the regression analysis might be pointless.
But your second objective aims more precisely at a multivariate analysis! The results of this more complex analysis can not be derived from the other "simple" correlations and equations. Therefore, your research objective 2 is certainly not pointless (although I would prefer 2 multiple regression analyses of the 2 dependent variables for better understanding).
Please be also aware, that you can not find out the direction of relationships with a cross-sectional design. You have to assume or know it.
Best wishes and good luck,
Horst
It is quite common to begin by presenting a table of correlations, followed by multiple regressions. The point of the regressions is to show the effect of each variable in the equation, after controlling for all the other variables in the equation -- as opposed to correlations, which only examine the relationship between a pair variables at a time.
Bivariate correlation shows the relationship between 2 variables (called in statistics, Zero order correlation). Which means the strength and direction of relationship between the 2 variables regardless of the effect of any other variable. However, as in reality this is not the case, and multivariate effect occurs at the same time, thus the Multiple linear regression is one way to handle the pure effect/relationship for each IV with the DV {MLR handle this issue by counting the partial and semi partial correlation for the variables}.
I always interpret simultaneously corrélations and multiple corrélation. Corrélations give you the relation of each variable with the critéruium. Multiple correlation give you the betas weighs or partial correlations, the part of prédiction specific to each variable over the part predicted by all the other variables in the équation and the multiplication of the simple correlation by the beta weight give you the importance of the variable in the prédiction. The sum of these multiplications give you the R sure (V'B = Rsquare)
I think David Morgan has given you the clearest and most straightforward answer. I would present the tables of correlations as essentially descriptive information, so the reader can see the relationships between individual variables in the simplest way. The significance level associated with each correlation coefficient (which is produced automatically in most stats packages) can give you guidance on which sets of relationships to describe more comprehensively by using multiple regressions (both standard and stepwise), if you are adopting an exploratory approach. I would also include other multivariate techniques such as factor analysis, which shows which groups of variables cluster together, to explore the data. Moving into a hypothesis-testing mode, you can also use regression to test any specific multivariate hypotheses you have, by specifying which variables to include based on theory rather than relying on the correlations.
Later, if you have enough data, you might move on to modelling complex relationships using Structural Equation Modelling, though there is a rather steep "learning curve" in mastering such techniques - you have to understand how to interpret the output and modify your hypotheses before trying again, rather than just reporting what emerges from the computer.
- Badges
- Science topic
- Similar topics
- Psychology
- Psychometrics