What test needs to be done here?

[GreenShenna]

I'm helping a friend with the statistical portion of her proposal to
finish her BS degree in Psychology.

Her paper suggests that the levels of cortisol can be interpreted to
determine if a soldier is suffering from PTSD.

My part in this is to somehow show this information statistically. So,
we have decided to have the following groups:
A-1: having low levels of cortisol and claimed PTSD
A-2: having low levels of cortisol and did not claim PTSD
B-1: having normal levels of cortisol and claimed PTSD
B-2: having normal levels of cortisol and did not claim PTSD

Incr Decr Same Totals
A-1 number of soldiers that fit in the category
A-2
B-1
B-2
Totals ~400

Blood tests are taken before deployment, right after they return, and at 3 month intervals. The ideal example would be a soldier with a normal level
initially, then a low level after returning home, then finally coming back up to a normal level after counseling. The soldier who has a low level after deployment but does not claim PTSD, will continue to have low levels of cortisol. These measurements need to be statistically significant.

I had thought that a Chi Square Test of Independence would be best to show if the data had an inherent correlation between cortisol levels and PTSD. However, after submitting her draft with this suggestion, he said it should be MANOVA with a t test comparing the A groups and the B groups separately. I'm currently researching how to do that and it doesn't seem mathematically difficult. I'm having trouble putting these measurements in a normal distribution with the mean and standard deviation bit.

Are we completely off here and need to start from scratch? Where do we
need to go with this?

I've got an AA in Math and took a summer class in Stats, so it's not my
strongest area. But, my friend has no clue about stats and needed my
help. If you need more information, please let me know.

[jamesmartinn]

You can either run a repeated measures factorial anova (group x time); 4 levels of group as your between subjects factor and 3 levels of testing time as your within subjects factor. How you compare groups if you find a significant interaction depends on if you planned specific comparisons (i.e, you have a reason/theory predicting certain groups will differ) or if you want to explore the relationships between all possible comparisons.

Usually, if you have planned comparisons in mind, you can run multiple t-tests (controlling for an inflated type 1 error via bonferroni or holm procedures). If you don't know what to expect with your data, you can use a post hoc test (i.e, Tukey procedure) as an exploratory tool to see out of all the possible group comparisons, which are different. If you're using SPSS, theres also another kind of follow up procedure which I think might be more suitable; trend analysis. This tests linear, quadratic, etc relationships of your data - which I think is appropriate given your description of the data. Have a look into it.

With the repeated measures factorial anova approach, you may have problems regarding sphericity. Machauly's Test of Sphericity in SPSS will tell you if you have met this assumption. If you haven't you can go with one of the already laid out corrections in the SPSS Output (i.e, Greenhouse-Gieser for non sphericity).

If you find you can't meet the sphericity assumption even after correcting for non-sphericity, you might want to use a 1 way MANOVA - the MANOVA does not require sphericity. Here you would have one independent variable (group; 4 levels) and all the blood tests would serve as dependent variables (so you would have multiple DV's). You can do follow up comparisons for a MANOVA like how I stated above.

BTW, I'm a psych student as well. I read about a very similar study in my Psychopathology class. Very interesting stuff!

Best of Luck!

[STATS]

Hello,

I was hoping that someone would be able to help my daughter, she is working on a proposal for her 12th grade class, and she needs to determine what type of statistical test to run. My background is in law enforcement,and I can not began to answer her questions. My daughter wants to find out if providing increase health intervention in the nursing home will boost the morale of the residents who are currently not receiving the increase health intervention. She iwants to sample 200 people, male and female residents, and determine if the one that receive increase health intervention morale is better than those who are on receiving a non increased health intervention.

I appreciate any suggestions that you can provide.