Testing different effects of an intervention on different outcomes

#1
Dear forum members,

we are planning to conduct a study with 2 groups (experimental vs. control) and random allocation to groups, and group sizes about 40-50 subjects each.

Outome measures will be interval scaled. Among them will be frequency of verbal communication with a partner (VCP), over a defined time interval (range possibly 0 to 25, median 15, in the control group,caccording to a pilot study), and display of certain nonverbal signs (NVS) (range maybe 0 to 80, median 50). In addition, we will measure attitudes towards topic A (6 items, 7-point answering scale), and attitudes towards topic B (5 items - other items than those used for topic B). The respective items measureing the attitudes towards a topic can be summed up.

According to our hypotheses, the experimental effect will be larger for VCP than for NVS, and larger for attitudes towards A than for attitudes towards B.

Using the terms from the header of this thread, and also several other terms, I tried to find out which statististical test or method could be used to test these hypotheses, but was unsuccessful. Repeated-measures ANOVA seemingly is not an option, since different variables should be compared, measured on different scales.

If someone could give me a hint where to direct my search for an appropriate method, I would certainly be thankful.
 
#2
The answer depends on the number of dependent variables and distributional properties of the dependent variables... If all of them are normally distributed, MANOVA is one path to follow.
 
#3
Thank you very much for your reply! As mentione above, we want to include k=2 variables at a time, and we want to test whether the impact (group effect) is larger on one of them than on the other.

As far as I know, MANOVA means to construct a composite out of the 2 variables - how do we then use MANOVA to answer the question whether one element of this composite affected more by the independent variable than its other element?

By the way, what do you actually mean by sitributional properties, and which properties should be present, preferably?
 
#4
If you want to test hypothesis

Ha: Expectation[ Y2(experimental) - Y2(control) ] > Expectation[ Y1(experimental) - Y1(control) ]

you can use difference in differences analysis.... On a different note, MANOVA does not "construct a composite out of the 2 variables". It tests whether the expectation of random vector (Y1, Y2, ..., Yp) in the experimental group is different from that in the control group.
 
#5
Thank you very much once more. As far as I know, in difference in differences analyses, the change of Y1 in condition A over time is compared with the change of the same variable Y1 in condition B over time. I am not sure how to adapt that approach to 2 different dependent variables. Would you suggest just to calculate the difference between rating scale 1 and rating scale 2 in group A and compare this with the according difference in group B?

By the way, even if this not contained in the problem at hand, I suppose the approach would not be feasable if outcomes were on different physical scales. For example to test whether the effect of an animal nutrition program on DV1 "weight gain" was larger than on DV2 "body size". So I wonder if there are additional solutions. The question whether an effect on Y1 is larger than on Y2 is not exotic, at least I thought so.
 
#6
1] Difference in differences analysis is not just about change in time. It is about comparing any type of change between two or more groups... Much attention is spent on the correlation structure of residuals after you do A - B - C + D. Non-trivial econometric methods are used. I suggest you read an extensive introduction to"panel data methods". Most suitable statistics tools will be there.

2] If some "outcomes were on different physical scales" then this wish of yours is ill-posed: "we want to test whether the impact (group effect) is larger on one of them than on the other". Please rephrase the problem and I will provide a solution... One potential way of looking at the matter is comparing standardized outcome variables, provided those outcomes are continuous. More steps must be taken with categorical outcomes.
 
#7
Please rephrase the problem and I will provide a solution...
Well...thank you... I really appreciate that you spend your efforts on my questions... but since
seemingly you either do not read carefully, or you do not clearly understand what is written,
and since in in addition you do not formulate clear answers, I am afraid that trying again
would be a waste of both our time. No harm intended. All the best!
 
Last edited: