I don't recommend cutting your variables up into categories for two reasons. First, you lose information. Second, trying to examine interaction effects when both your variables include multiple categories can be a real headache.

Hence, you should follow the advice of others and test multiplicative effects between your independent variables in the regressions equation.

You still can run the multiple regression, but with the interaction of Value Orientation (VO) with Awareness of Environmental Threads (AET). Therefore, in your data step you first need to calculate the product of the VO scores and the AET scores (VO*AET), and then run the regression: DV = VO + AET + VO*AET.

.

Hello,

If you want to replace your multible regression analyzes with anova, you will need to group your continuous variables. Often a good method is to create dichotomous groups according to the median or more or less 1 standard deviation to create more than two groups. Once the categorization of your continuous variables is done, you can then roll your multivariate anovas and add interactions effects if you deem it necessary. However, I would like to emphasize that it is possible to evaluate the statistical significance of intrection effects via multiple regression equally, although their interpretation often proves more complex. The advantage of using regressions to test interactions effects is that you will probably have more statistical power.

Here are some notorious references that suggest procedures for testing moderation effects from multiple hierarchical regression.

Tabachnick, B. G., & Fidell, L. S. (2007). Using Multivariate Statistics. Pearson Education. Boston, MA.

« Interactions between discrete IVs are common and are discussed in any standard ANOVA text. Interactions between continuous IVs are less common, but are of interest if we want to test whether the regression coefficient or importance of one IV (XI) varies over the range of another TV (X2) If so, X, - is said to moderate the relationship between XI and the DV. For example, is the importance of education in predicting occupational prestige the same over the range of income? If there is interaction, the regression coefficient (B) for education in the regression equation depends on income. A different regression coefficient for education is needed for different incomes. Think about what the interaction would look like if income and education were discrete variables. If income has three levels (low, middle, and high) and so does education (high school grad, college grad, postgrad degree), you could plot a separate line for education at each level of income (or vice versa) and each line would have a different slope. The same plot could be generated for continuous variables, except that distinct values for income must be used. » P. 157

« Recall from Chapter 4 that computational problems arise when IVs are highly correlated. If the IVs with interactions are not centered, their product (as well as higher-order polynomial terms such as x2 ) is highly correlated with the component IVs. That is, XIX2 ishighly correlated with both XI and with X2; X: is also highly correlated with XI. Note that the problem with multicollinearity in this case is strictly statistical; the logical problems sometimes associated multicollinearity among sup,posedly different predictors are not at issue. In the case of interaction between IVs or powers of an IV, multicollinearity is caused by the measurement scales of the component IVs and can be ameliorated by centering them. » P.157

Baron, R. M., & Kenny, D. A. (1986). The moderator–mediator variable distinction in social psychological research: Conceptual, strategic, and statistical considerations. Journal of personality and social psychology, 51(6), 1173.

« So if the independent variable is denoted as X, the moderator as Z, and the dependent variable as Y, Yis regressed on X, Z, and XZ. Moderator effects are indicated by the significant effect of XZ while X and Z are controlled. » P. 1176

As others have noted, you can examine continuous-by-continuous interactions without resorting to carving the variables into categories.  One way to explore and visualize the nature of a continuous-by-continuous interaction with SPSS is by using the model scoring features.  See the links below, for example.  The YouTube video shows how to score and apply a model in the context of logistic, not linear regression.  But it does a nice job of explaining the basic steps, which will be pretty similar for linear regression.  HTH.

https://www.ibm.com/support/knowledgecenter/en/SSLVMB_23.0.0/spss/tutorials/scoring_intro.html

https://developer.ibm.com/predictiveanalytics/2014/04/18/plotting-interactions-and-non-linear-predictions/

https://www.youtube.com/watch?v=eboTREc3uSk

I don't recommend cutting your variables up into categories for two reasons. First, you lose information. Second, trying to examine interaction effects when both your variables include multiple categories can be a real headache.

Hence, you should follow the advice of others and test multiplicative effects between your independent variables in the regressions equation.

Following up on one of David's points, when you throw away information, you typically lose power.  The slides linked below are for a short conference talk I gave on that issue.  It uses a fairly simple simulation to make the point that the more information you throw away (by carving quantitative variables into categories, and by treating ordinal variables as nominal), the less power you'll have.  HTH.

Presentation Free Advice for Power-Hungry Researchers: Do Not Categorize ...

The best practice is to go for regression without any loss of information 

 I am actually advised that regression may not very much be reflective of my concept as regression will deem one of my variables as a mediator/ moderator? Thoughts on this would also be appreciated. I am (obviously) not very well versed in Statistics and may actually really have an anxiety against it.

But thank you everyone for your input! I am studying how to use factorial anova on my data and keep the scores, as I am told, it's OK.

Beyond fear and loathing: regression and ANOVA are interchangeable methods in statistics, i.e. they answer the same questions and yield the same results.  If there is a correlation between "Value orientation" and "Awareness of environmental stress", then you should not perform an ANOVA, because you then treat these "factors"  as independent (orthogonal).

As noted above ANOVA is a special case of multiple regression. There is no formal distinction between a moderator and moderated variable in a regression (it is a convenience of labelling or a designation from information outside the statistical model). It can be helpful to think of one predictor moderating another but it isn't a necessary distinction (and can be unhelpful).

In an interaction analysis there is no mediator, Mediation is a distinct concept from moderation (interaction) - though one can build up models in which moderation and mediation are combined this is not necessary.

As already noted, ANOVA and regression are statistically similar.

Also, when using regression, you avoid information loss (which would emerge from reducing the many values of a continuous variable to the few of a categorical).

Still, a practical reason for using regression instead of ANOVA would be that if you chose ANOVA and you proceeded in categorizing IVs you would probably have a hard time persuading reviewers and editors for this choice, because ANOVA with median-splits seems to be not very much liked in journals the last years (due to loss of information etc.).

Having said the above, I would mainly consider two criteria on what analysis to use. First, the loss of information matters most if the examined effect is at the borderline of being significant. In such case regression might be preferable (of course, irrespective of your hypotheses).

Second, when it comes to presentation of findings, for interaction effects (especially of higher order) means form an ANOVA might be more readable to than regression slopes.

Finally, let me refer to an issue that you did not ask for: One of your IVs, namely "values orientation" has three categories ("egoistic", etc..). Is that categories or dimensions? If it is categories, then you have there already a categorical and not continuous variable. But I may be missing something.

If you use an ANOVA you treat the factors as independent, as if you would have manipulated them as in a true experiment. Another reason to choose for regression

Hi, Thank you all again for your inputs. I meant dimension and must have typed "categories" because of all the debate in my head re: factorial ANOVA. 

I do have, as it turns out, 4 continuous independent variables and one continuous DV. I think all the advise is to use ANOVA, right? I will also try multiple regression with interaction among the variables. Am reading more on it now.

You should first check if the independent variables correlate. If so, you'd better perform a stepwise method (e.g. all-subsets) regression.

Thank you @Jos Feys, will try that

Before you decide to use step-wise methods, see the link below. 

http://www.stata.com/support/faqs/statistics/stepwise-regression-problems/

Stepwise regression is a rather out-dated technique. Correlations among the independent variables are to be expected, and are only a problem when they are quite high (i.e., multi-colinearity).

I advised 'all possible subsets' or 'best subsets' regression.

Hi Jos.  Some authors (e.g., Frank Harrell) believe that all possible subsets regression is subject to the same problems as other algorithmic variable selection methods.  The following is from the web-page I cited earlier.

Frank Harrell’s comments:

Here are some of the problems with stepwise variable selection.

“All possible subsets” regression solves none of these problems. (Emphasis added)

http://www.stata.com/support/faqs/statistics/stepwise-regression-problems/