Which statistical test should I use for my assigned participants I am trying to analyse the interaction between level of class (either upper class or lower class) and level of prejudice attitudes. I have hypothesized that upper class individuals shall have higher levels of prejudice due to eliciting more essentialist beliefs. I am using the Essentialist Belief Scale, the Modern Racism Scale and the Modern Sexism Scale. 
What statistical test should I be running? 
I thought it could be a $t$-test or a regression but I am not entirely sure which one to use and why. 
 A: @KEA, you are clearly developing causal hypotheses, and I wonder if you have considered diagramming them using directed acyclic graphs (DAGs). For example, just reading your original post, and adding some imagination about what data you might collect, I'm led to sketch the following causal diagram:
 
Your own diagram of your hypotheses may be quite different from this. But I'll wager that no sooner will you have drawn your own causal diagram than you will see that your research question is genuinely nontrivial, and cannot possibly reduce to a handful of null hypothesis tests.
FYI: This diagram is made using the DOT language, like so:
digraph g {
  node [fontname="Helvetica"]
  SC [label="Social\nClass"]
  EB [label="Essentialist\nBeliefs"]
  PA [label="Prejudiced\nAttitudes"]
  EBS [label="EB\nscore"]
  MRS [label="Racism\nScore"]
  MSS [label="Sexism\nScore"]
  SRSC [label="Self-reported\nSocial Class"]
  SC -> EB -> EBS
  EB -> PA
  PA -> MRS
  SC -> MRS
  PA -> MSS
  PA -> SRSC [style="dotted", label="?"]
  SC -> SRSC
}

A: Are you measuring class as simply a dichotomous variable (upper or lower are the only possible values)?  If so, the simplest analysis would be three t-tests, one for each scale measure.
But class feels like it should be a continuous variable - in which case regressions would be appropriate.
At a higher level of analysis, this sounds like it would benefit from a multivariate analysis, in which the three outcome measures are considered simultaneously (MANOVA if you are working with dichotomous class variable).
