MANOVA or something else?

#1
Hello everyone!

Please excuse my english, it is not my first language, so if something is unclear, let me know :)

For my master thesis, I'm using pre-collected data from Burn-Out patients to compare with a healthy control group in regards to figural creativity and subjective evaluation of the creative personality. The data for figural creativity was collected with the Torrace Test of Creative Thinking (Originality and Flexibility). The data for creative personality was aggregated via the Creative Personal Scale. Both groups have been matched by age and gender.

I'm investigating the following:
What differences can be found between patients who suffer from burn-out symptoms in comparison to healthy participants in figural creativity and subjective evaluation of the creative personality?

For this, I would chose the following variables:
* Independent variable: Burn-out vs. Healthy Group
* Dependent variable 1: TTCT (Originality vs. Flexibility)
* Dependent variable 2: CPS (Only sum-score?)

I think I'll have to use a 2x2x? MANOVA, but the CPS will only give me a Sum-Score, consisting of 12 Items for "not creative" and 18 Items for "creative", with a min/max of -12/18 (I think the not-creative-Item answers are coded with -1/0 and the creative-Item answers with +1/0, respectively). Unfortunately, I do not have the manuals for the tests, so I could not find the necessary Cut-Off values. The only information I found so far is, that the CPS result will be a "Sum". What I do not know is, how can I use a Sum when calculating a MANOVA? Would I not need two factors for the variable? Also, I think I'll need the Mean value, and I have no clue how to calculate a Mean from a Sum-Score...

My other investigation is regarding the relation between TTCT, CPS and Burn-out patients:
Is there a correlation between the Originality (or Flexibility, respectively) in figural creativity tests and the subjective evaluation of the creative personality in Burn-Out Patients?

As this is an undirected hypothesis, I think I need to use a Correlation. I'm not sure about that, tho, and have the same issue with the Sum-Score when it comes to calculating a Mean value...

Any help and suggestions are highly appreciated!
 
Last edited:

Karabiner

TS Contributor
#2
Not quite sure what your problem is. As it seems, you have an instrument with 2 (orthogonal, i.e. independent, ususally uncorrelated?) scales, and the other intrument with 1 scale (a score between -12 and +18). So why did you ask for MANOVA? Looks just like 3 t-tests (unfortunately, you did not give the crucial information of how large the sample sizes are).

Unfortunately, I do not have the manuals for the tests,
http://homepages.se.edu/cvonbergen/...ality-Scale-for-the-Adjuective-Check-List.pdf


Is there a correlation between the Originality (or Flexibility, respectively) in figural creativity tests and the subjective evaluation of the creative personality in Burn-Out Patients?
If this is your research question, then probably you can perform 2 correlations.

It would perhaps be more interesting to compare whether this association is of different strength in patients vs. controls. For such a question a regression analysis would be an option, e.g.
SubjectiveEvaluation = b0 + b1*Flexibility + b2*group + b3*Group*Flexibility + error

With kind regards

Karabiner
 
#3
First, thank you very much for answering my questins and the "manual". I know this paper, but there are no informaitons about a cut-off score. Would it be okay for you to write yon a PM in German? Would be much easier for me :)

For your information: There are 78 participants in each group. Sorry, I forgot that important information :(
 

Karabiner

TS Contributor
#4
I do not know what you mean by cut score here, or what you want to use it for. AFAICS there is no need for cut scores in the present analysis.

With kind regards

Karabiner
 
#5
Thank you again for your answer. Am I allowed to ask in german? I'm afraid a lot of mix-ups are happening because of the translation, and it appears you speak german. So if possible, can I switch to german in the thread or, if that is not allowed, write you a private message (in german) to explain my problems? That would be great :)

Thank you in advance!
 
#6
Would it be okay for you to write yon a PM in German? Would be much easier for me :)
Yes, but it would be a lot more difficult for the rest of us.

Here we live in the (possible) illusion that everybody can gain some knowledge from the others.

It seem to me that here it is just do an analysis of variance for each of the dependent variables with health/not healthy and age and gender as explanatory variables. And to estimate the correlation from the residuals, where the correlation was wanted.
 
#7
Thank you! Yes, I'm running a forum for IT issues for more than a decade, also fueled by "everybody can gain knowledge from others", so to say the same esprit de corps.

I will try to re-organize my thoughts and questions and post them tomorrow in a more structured way.

Any help is appreciated, as are suggestions, because frankly I'm currently not even sure how to treat the data and which statistical approach I best use.

Kind regards,
Sandra
 
#8
Hello again!

I'll try to give you as much information and background as possible :) There has been a study with Burn-out patients, and I must use that data for my study.
I'm supposed to test healthy controls with the same test battery. The initial study consisted of 78 patients, and I'll test enough people to get a good match (at least 50 per group).
The test battery consists of 18 tests, from which I can choose freely to formulate new research questions.
Mandatory is that I always must use the Figural Creativity Test (TTCT), as the study theme is "Burnout and Figural Creativity" (I was assigned the theme and did not choose it).
So far, I've initially decided for the following research questions and tests (Some are AFAIK german only, I'll try to translate):


1) Burnout-Specific
Test 1.1: EBF-W (Recovery-Stress-Questionnaire-Work, Erholungs-Belastungs-Fragebogen-Work) consisting of 2 parts. The first is 'Recovery-Life' (Erholungsleben), the second 'Exhaustionexperience (work-related)' (Beanspruchungserleben, arbeitsbezogener Teil). The Items are on a scale from 0 to 6.

Reserch Question 1.1:
How do Burn-out patients differ from healthy controls in regards of the score in the Figural Creativity Test (TTCT) and the Recovery- and Exhaustion experience (EBF-W)?


2) Mood and State of Mind
Test 2.1: BBS (Basler Mood Scale, Basler Befindlichkeits-Skala). 4 Dimensions of 4 Items each (7-step Likert-Scale, siebenstufige Likert-Skala): Validity, Intramental balance, social extroversy, vigility (Validität, Intraspychischer Gleichgewichtszustand, soziale Extravertiertheit, Vigilität).

Research Question 2.1:
How do Burn-out patients differ from healthy controls in regards of the score in the Figural Creativity Test (TTCT) and the Basler Mood-Scale (BBS)?


3) Sleep and Sleepiness
Test 3.1, Sleep: PSQUI (Pithburgh Sleep Quality Index). Sleep quality in 7 Dimensions, 18 Items, rated from 0-24 with a Cut-Off of 5 -> Will distinguish if sleep disorder or not.

Research Question 3.1:
How do Burn-out patients differ from healthy controls in regards of the score in the Figural Creativity Test (TTCT) and Sleep Quality (PSQUI)?

Test 3.2, Sleepiness: TSS (Tiredness Symptoms Scale). 14 dichotome Items (Yes/No), Score 0-7 equals low subjective sleepiness; score 8-14 equals high subjective sleepiness.

Research Question 3.2:
How do Burn-out patients differ from healthy controls in regards of the score in the Figural Creativity Test (TTCT) and Subjective Sleepiness (TSS)?


4) Creativity as a Trait
Test 4.1, CPS (Creative Personality Scale). 30 dichotome items (Yes/no), split into 18 positive and 12 negative. Score betweend -12 and +18 can be achieved.

Research Question 4.1:
How do Burn-out patients differ from healthy controls in regards of the score in the Figural Creativity Test (TTCT) and the subjective evaluation for a creative personality (CPS)?


5) Creativity as Performance
Test 5.1, TTCT (Torrance Test of Creative Thinking) - my fixed variable. Only the parts regarding Originality and Flexibility will be used and will be evaluated by categories. This is the mandatory main variable for my study.


6) Focus
Test 6.1, TEMEKKO (Test to evaluate executive control and focus, Test zur Messung exekutiver Kontrolle und Konzentration). Measures 3 scopes: Figural, Numeric, Verbal. Evaluation by performance index, number of correctly treated icons and percentage error score.

Research Question 6.1:
How do Burn-out patients differ from healthy controls in regards of the score in the Figural Creativity Test (TTCT) and executive control/focus (TEMEKKO)?


Initially, I also planned additional correlation Hypothesis as explorative research question, but since I'm already quite overwhelmed, I'll postpone that. I lost my synoptic view and am confused how to statistically check for my research questions, it's driving me to despair. All I got to work with is the pre-study from a colleague (the 78 burn-out patients data), and that's pretty much it - my professor is also not helping :( I've spent a lot of time on research and trying to find more information about all the available tests, but that only added to my confusion. Now I'm not even sure anymore if my research questions make any sense :/

Thank you all for your time, any help and suggestions are appreciated!

Kind regards,
Sandra
 
Last edited:

Karabiner

TS Contributor
#9
If dependent measures are interval scaled, then group comparisons can simply be carried out using the t-test, and either Pearson-correlation or Spearman rank correlations can be calculated.

If dependent measures are ordinal scaled, then group comparisons can be carried out using the Mann-Whitney U-test, and Spearman rank correlations can be calculated.

If dependent measures are categorical (such as sleep disorder present yes/no), then group comparisons can be carried out using Chi² tests.

In case that groups differ with regard to additional variables which also affect the dependent measures (e.g. age, education...), and you have measured such variables, you can integrate them into the analysis, if the dependent variable is interval scaled. E.g. by using multiple regression with "group" as 0/1-coded predictor variable and other characteristics as additional predictors.

HTH

K.
 
#10
Dear Karabiner,

thanks for practically saving me!

I have an additional question, if you do not mind :) As I'm not happy with the way I formulated my research questions, I want to change them to, e.g. for tiredness:
"What impact does tiredness have on the score of burn-out patients in the figural creativity test in regards of originality (flexibility), compared to healthy controls"
("Welchen Einfluss hat die Müdigkeit auf die Leistung von Burn-out Patienten im figuralen Kreativitätstest in Bezug auf Originalität (Flexibilität), verglichen mit gesunden Personen")

Do the changes to the research question have any impact on how to evaluate the scores? As you suggested, I plan to use t-tests for dependent samples and Chi² for the binary (yes/no) coded tests.

Thanks again and kind regards,
Sandra
 

Karabiner

TS Contributor
#11
"What impact does tiredness have on the score of burn-out patients in the figural creativity test in regards of originality (flexibility), compared to healthy controls"
That means: Is the correlation between tiredness and creativity in "burn-out" subjects different from the correlation between tiredness and creativity in healthy subjects. But maybe you wanted to ask something else.
I plan to use t-tests for dependent samples
Independent samples (2 independent groups), I suppose.

With kind regards

K.
 
#12
That means: Is the correlation between tiredness and creativity in "burn-out" subjects different from the correlation between tiredness and creativity in healthy subjects. But maybe you wanted to ask something else.
Yes, that would be what I mean to ask. As I understand, even tho the phrasing has changed, the outcome (research question) is the same and I can still use t-tests, correct?

Independent samples (2 independent groups), I suppose.
Independent, indeed!

Thanks again and kind regards,
Sandra
 

Karabiner

TS Contributor
#13
yes, that would be what I mean to ask. As I understand, even tho the phrasing has changed, the outcome (research question) is the same and I can still use t-tests, correct?
Independent t-tests are used for the comparison of means of a dependent variable between 2 groups.

But what you have asked was instead: Is the the impact of tiredness on creativity (measured e.g. by a correlation coefficient) in the "burn out"-group higher than the impact of tiredness on creativity in the healthy group. Is this really your research question?
 
#14
What still confuses me is that I have two variables in each group (e.g. creativity+tiredness, creativity+sleepiness, ...) and how to handle that. I have the feeling I can't see the woods because of all the trees -_-

I'm still pondering how I'll formulate the question (and what to exactly ask for). What I want to ask for a research question is to test the following hypothesis:
Burn-out patients are more tired than healthy controls and thus, have a lower mean score in the creativity test (lower originality/lower flexibility), hen compared to the mean score of healthy controls.

Kind regards,
Sandra
 

Karabiner

TS Contributor
#15
You hypothesize a causal relationship here, "Burn-out" => Tiredness => Creativity, i.e. the effect of burn-out on creativity is mediated by tiredness. This can be handled by mediation analysis, but you'll have to read a little bit about this first.
 
#16
Thank you again for your answer.

It appears I have problems to get the questions I planned in post #8 of this thread in sync with the hypothesis.

The overall topic is "Burn-out and figural creativity" and the task I was given was "Do something with it". So I'm trying to do something with it, using the little bit of pre-collected data available, and trying to get research questions that fit a hypothesis on which I can base my master study on, using the pre-specified test battery. I have a study partner who does not share their data and ideas and my professor is on vacation and won't be back before I have to hold my method presentation. Yay.

Thanks for your suggestion in regards of the mediation analysis, but since I've never used it (and frankly, can't remember even having heard of it) I'd be thankful for suggestions how I could change the hypothesis/research questions so I could rather use t-tests, variance analysis, regression, correlation... or other methods I at least have some knowledge of - even tho I've had my last statistics lectures in 2004, which is also not helping :/


Thanks,
Sandra
 
Last edited: