I am studying the impacts of receiving a certain treatment (kind or mean) on the motivation. I expect that the treatment influences the motivation through three mediation variables. So my conceptual framework looks somewhat like this: X -> Z1, Z2, Z3 -> Y.
To explain it with an easier example: I want to know the effects your gender (male/female) has on your life expectancy, but I study it through the mediating variables height, weight and smoking (all on a likert scale). I expect that the gender affects the mediating variables, but they in their turn have an effect on the life expectancy.
(A little more detailed: I have an experimental design where I use vignettes to reboot the type of treatment someone will receive (the respondents either get the soft treatment or the hard treatment, not both and it is randomly assigned). Then I test my mediation and Y variables through survey questions with a 5 point Likert-scale.)
Which test do I use for this in SPSS or does anybody have an idea how to format my conceptual framework better?
To determine which statistical test to use, you need to know: whether your data meets certain assumptions.
Hello Mathilde,
The first concern is with respect to the outcome variable(s). You say you're using Likert-type response scale for items, but you don't indicate whether the relevant DV is based on a single item or on a sum/average of responses to a set of related items. In the former case, nobody would question that the DV is an ordinal variable; in the latter case, you'll quite often see the summated scores treated as if they are continuous/interval.
Options in SPSS:
1. Regression. Each of the relationships in your proposed model can be evaluated by the SPSS regression module. If DV is judged ordinal, then choose multinomial (logistic) regression. If DV is judged interval, then choose linear regression.
2. Andrew Hayes' PROCESS module (which semi-automates a lot of the analysis for you). You may download the module, and subsequently install it, by checking this link: http://processmacro.org/index.html For a mediation model, try his model #6.
3. If you consider all of your variables to be manifest (vs. factors or constructs), then a path analysis model can also handle your specific set of proposed relationships, as well as furnish indices reflecting how well the relationships in the data can be reproduced by the model in question. The add-on module, AMOS, allows you to do this quite handily. It's not free, however. As well, you could use SPSS's R hooks to run one of the freely available R library options, such as lavaan.
Good luck with your work.
- Badges
- Science method
- Similar topics
- Mathematics
- Statistics
- Statistical Modeling
- Analyses of Variance
- ANOVA