Which tests? Clarification plz!

[Pippy]

Hi all!

I’m currently doing some research and I’m a bit of a novice so wanted to check if I was on the right lines with the statistical tests I’m thinking of using.

I am looking at the relationship between perceived stress scores (which are represented as low, moderate and high categories) and eating behaviour patterns in a single group of people (n=80). Three things are looked at in terms of eating behaviour patterns including emotional eating, cognitive restraint and uncontrolled eating. Each of those three things gave numerical scores which have been categorised into low moderate and high ranges again.

As this data would be classed as ordinal (I think) I first wanted to find out if there was a difference between gender in the items above (e.g. difference between male and females with emotional eating). I therefore was thinking a Mann Witney U test. Would I be right in this thought??? If so how can I work out the direction of significance? Do I use the mean rank differences (given on SPSS) between the male and female to work out a percentage difference?

Also as I am looking at finding a relationship between these things e.g. stress scores and emotional eating, stress scores and uncontrolled eating, would I be right in selecting Spearmans rank test and then working out the coefficient of determination?
If you could advise on any other tests that may be better or any further test I need to do to analyse my data more effectively I would very much appreciate it.

Thanks so much in advance!!

[noetsi]

If your DV is ordinal your might try ordered logistic regression. It is preferred to a non-parametric method such as Mann Witney if you meet the requirements of the method (you would have to check yours did).

Spearman's will give you bivariate correlations between ordinal variables. They don't tell you how the set of variables relate to each other other (that is how X effects Y that effects Z which...) There are "better" but far more complex methods to do that, but if you are new to statistics they probably are not the way to go.

[EliteResearch]

Can you explain why you took your numerical scores for emotional eating, cognitive restraint, and uncontrolled eating and rank ordered them? If you don't have any reason why you absolutely have to rank them, I'd recommend using the raw numerical scores. You'll have an easier time with your analyses.

IV- Perceived stress scores (3 levels) IV- Sex (M/F) DV- numerical values of emotional eating, cognitive restraint, and uncontrolled eating (continuous variables)

ANALYSES-
1. To examine gender differences in emotional eating, cognitive restraint and uncontrolled eating- 3 separate one-way ANOVAs (one for each DV) with Sex as the IV. If the three DVs are conceptually related, measured on the same scale, are moderately correlated with each other (but not exceeding .60 or .70), then you could run a MANOVA instead. Alternatively, you could create a composite variable of the three DVs (again, assuming it makes conceptual sense to do so) and run an ANOVA on that value.

2. To examine the relationship between stress scores and those three DVs. Repeat above steps but include a post-hoc to look for differences between the three stress levels.

In fact, if you are able to use the perceived stress scores as numerical scores instead of as ranked values, you could do a regression analysis (and your graphs would look prettier).

-Amanda
1925 E. Beltline Rd. Suite 200
Carrollton, TX 75006
(800) 806-5661 (972) 538-1374
consulting@eliteresearch.com