Hi guys,

I am having difficulties with choosing an appropriate statistical analysis for my research. I will do my best to describe it.

I have three DVs:

DV1: GPE is a continuous interval variable ranging from 10-50. This is measure pre and post an intervention.

DV2: GNE is a continuous interval variable ranging from 10-50. This is measured pre and post an intervention.

DV3: ATQ is a continuous interval variable with scores ranging from 30-150. This is only measured post manipulation.

I have two IVs:

IV1: Cond is a binary variable indicating the condition of the intervention. Levels are 0 and 1.

IV2: SCS is a continuous interval variable with scores ranging from 26-130. This score is measure prior to the intervention.

As some explanation of the background, research suggests that cond=0 may cause increase in GNE and decrease in GPE. New research suggests that cond=1 may have protective effects against the effects of cond=0 (same basic activity, with a different approach). Looking at it theoretically, one may suggest that these protective effects may be in the reduction of ATQ and the facilitation of SCS. Unfortunately, there is no state measure of SCS and therefore it cannot be included as a DV to support this proposition. We can either control for it or include it as a moderator.

I have three to four hypotheses that can come in multiple forms depending on the best statistical analyses.

Approach 1: Change score (could be percentage, std dev change, or raw change)

Hypothesis 1: Participants in cond=1 will demonstrate less decrease in gpe as measured by the change in GPE from pre to post compared to participants in cond=0, controlling for SCS.

Hypothesis 2: Participants in cond=1 will demonstrate less increase in gne as measured by the change in GNE from pre to post compared to participants in cond=0, controlling for SCS.

Hypothesis 3: Participants in cond=1 will show significantly lower ATQ scores following the intervention than will participants in cond=0, controlling for SCS.

To address these hypotheses I was planning on running three separate regression models with cond and SCS as IVs. These IVs would be used to predict the change scores on GPE and GNE (seperately), or ATQ score as DVs.

I had two concerns with this:

Firstly, change scores do not account for the scale of the original variables. However, this could be easily addressed by using a change score that is in STD DEVS or percentage

Secondly, if GPE and GNE are measured pre and post, therefore are within subjects variables, is it allowed to simply disguise this by using a change variable?

Approach 2: Not a change score, instead predicting post scores and controlling for pre scores.

Hypothesis 1: Participants in cond=1 will demonstrate significantly higher GPE post scores compared to participants in cond=0, controlling for pre GPE and SCS.

Hypothesis 2: Participants in cond=1 will demonstrate significantly lower GNE post scores compared to participants in cond=0, controlling for pre GNE and SCS.

Hypothesis 3: Participants in cond=1 will demonstrate significantly lower ATQ scores compared to participants in cond=0, controlling for SCS.

To address these hypotheses I planned on running three separate regression models with cond, pre scores (in first two) and scs as IVs. These will be used to predict post scores on GPE, GNE and ATQ (separately).

I have two concerns with this approach:

Firstly, this analysis would not allow me to have hypotheses that indicate the direction of change. That is, it does not recognise that we are NOT suggesting that cond=1 will result in an increase in GPE and a decrease in GNE.

Secondly, do I need to control for baseline scores if I run ttests prior and find no significant difference across groups?

Approach 4: Mixed Design ANCOVA

The last method I have considered was suggested to me by an advisor. They suggested I use a mixed design ANCOVA. I do not really know how to go about this. I know I need the data in long format which I have created but I am unsure if this is appropriate, or how to go about it.

Approach 3: SCS as a moderator

This approach could include any sets of hypotheses including SCS as a moderator.

I am unsure whether or not it makes more sense in the situation I have to include SCS as a moderator or to control for it.
__________________________________________________ __________________________________________________ ______________________________

Any feedback would be greatly appreciated.

​​​​​​​Thank you all!