What statistical analysis to use for my study?!!

[Isabelle williams]

Hi guys, I’m stuck as to what test to use for my research project. 

I am looking into the effects of socio economic status and  recurring negative life events on feelings of hopelessness. I am doing this by using three questionnaires, one measuring each variable. Socio economic status and negative life events are the independent variables. All participants will be answering all questions and it is hypothesized that lower socio economic status scores and greater levels of negative life events will significantly effect hopelessness scores. 

Now I am only an undergraduate so am very new to spss but am unsure whether to use ANOVA, then I can do paired comparisons of each variable and see interaction effects. Or correlation, then I can see a positive of negative association between the variables. 

At them moment I am considering both haha, but I would like to use the  correct test.

Many thanks!!!

Isabelle

[Stephen Gorard]

Two things first. Your design will not allow you to look at ‘effects’, only at covariation. You might speculate a causal model if you find a link, but it must be very tentative. And unless you have a complete random sample, don’t bother with any ‘tests’.

 

So. A correlation (like R) is fine as long as the measures crossplot into something like a straight line (and as long as you ignore the ** etc. and make your own judgements). If not, use the crossplot for ideas how to proceed – transformation, rho…. If the SES is a category really, then you could simply compare the mean scores of ‘hopelessness’ for each SES category. Whether the patterns are sufficient to pursue has to be your reasoned judgement.,

[Isabelle williams]

Thank you very much, I will pursue the correlational route, I just needed a stable place to start from 🙂 

[Dave Collingridge]

Isabelle,

Stephen’s statistical approach will work fine.

I agree that it is important to correctly interpret the whether your results are causal or correlational. Determining whether your results are causal or correlational depends on method. Causal interpretation requires, for example, randomly selecting participants from a population, randomly assigning to levels of a treatment variable that includes a control group, and controlling for extraneous and confounding variables. Most researchers never achieve these standards, and that is fine as long as they recognize that their conclusions are limited to associations between variables

But being limited to correlational relationships does not necessarily limit researchers to correlational analyses. ANOVA basically accomplishes the same thing as regression, regression with 2 variables is basically similar to correlation, and regression between a binary variable and a continuous variable basically does the same thing as the t-test. These tests are all related statistically speaking. So feel free to use ANOVA as long as you don’t overextend your conclusions by assuming causal relationships. Anyway, it may be easier to use correlation and/or regression if you aggregate scores from each survey and end up with continuous measures. 

[Isabelle williams]

Hi dave, 

yes I am aware of the issue of causation that is linked with correlation, but could I use correlation, then suggest in my discussion further tests to strengthen this relationship? I was unaware there was such a test that suggested causation but will definitely look into it. I was under the impression that ANOVA was only to be used under experimental design with two or more groups, which leads me to question what is my design? I have just said correlational so far.  If everyone is answering all questions I want to see if those total scores for each participant are associated with total scores for hopelessness. The more I think about it the more  confident that  I will use correlation. 

[Stephen Gorard]

Isabelle, you are right. Causation is assessed by the correct design, and using ANOVA inappropriately with non-randomised cases solves nothing  but creates huge problems. Stick to correlation.

[Author]

Posts