Hi Natasha,

A couple of questions for clarification.

What are the I and the C? Do they mean anything there? If the idea is that these go from II to CC, or do the items just have 5 values and you just want to see what patterns arise?

Are the different groups known, in other words you want to compare people with different values on variable Y, or do you want to identify the groups on the basis of the responses to these items?

Dan

Hi Daniel,

So the scenarios and the options are designed to measure behaviors and intentions, and the different options have different combinations, for example CC means they picked the option that shows collectivist behavior with collectivist intention, IC means they chose individualist behavior but with collectivist intentions, and so on with 5 different categories (no ranking order).

We also have the participants' independence/interdependence scores from a different measure so we are looking to see whether for example highly interdependent group answered differently than highly independent group.

So the response variable is their independence/interdependence scores. Is this a continuous measure (is it one measure)? On the predictors, it sounds like all five of those options are available for each item, so there are a few approaches. One would be calculate the percentage of each of the options the individual makes (if they have to tick one and only one for each then this can be just 4 variables since the fifth is 1-these). Look at those. It may be clear (the so-called intraocular trauma test ... hits you between the eyes) that there are only a couple of patterns of responding. You could see how well these predict your response variable.

There are more complex procedures too, but this simple approach might be a good first step (and from your description the two should be measuring something similar, so the association should be clear). There are latent variable procedures that may be appropriate after exploratory data analysis. (and I'm sure other readers will suggest some)

Hi Natasha,

I see no one else has responded. How are you doing with this?  Dan

Hi Daniel,

Thank you very much for the advice and follow up. I calculated the percentages of responses for each option and was able to compare whether the two different groups answered them differently. My problem is I don't know how to compare the two different versions of responses to each other. Let me try to clarify the problem:

This multiple choice measure consists of 18 scenarios. The scenarios are loosely related to one another, but cover a variety of situations. Usually in the situation you are dealing with a family member, a friend, or a stranger/member of the greater community. I've focused my analysis by trying to see if there were differences in situation type, for example, do people who have individualist values act like an individualist toward everyone or just strangers, etc. and what is the intention? Same for collectivists, do they act more collectivist toward family and individualist toward strangers? However, the problem with this approach is that all my participants are in an individualist social climate (Canada). So, the participants filled out 2 versions of these scenarios one phrased in reference to the self ("What would you do in this situation?" and one phrased in reference to others "What would people typically do in this situation?"). So I'm not really sure how to compare the two different versions to see what the difference is. I can eyeball the difference but short of doing t-tests for each scenario I'm not sure how to statistically analyze it.

If it would help I can upload the measure for you to have a look.

Thank you for your time!

Okay, first, the eyeball test (called the Intra-Ocular Trauma test for the results hitting you between the eyes ... yea, stats folks have strange senses of humor) is good. Science did well (some argue better) before there were all the concerns with p-values and that kind of thing. I would avoid doing lots of t-tests, but plotting the data where you can see the differences among scenarios will be helpful.

The tricky bit is the actual model. If you had, say six family, six friend, and six stranger scenarios, and could categorize the response along a collectivist scale (either something like these three responses are collectivist, these two are not, or say these two get zero collectivist points, this one gets one, this one gets two, and this one gets three), then you can calculate the mean for the scenarios (both for answering for self and for other). This would give you a 3x2 within subject ANOVA format (assuming the assumptions of that are okay). Lots of people nowadays use multilevel/mixed models rather than repeated measure ANOVA models, but see what is the norm in your area. It may also be that you want to use values for what people said for how other people would act as covariates. This tests a slightly different model (something the statisticians refer to as Lord's paradox), but it is worth seeing what is the norm in your area (otherwise you have to battle with reviewers if they don't understand what you are doing, and even if you are right that can be difficult).

That approach has the problem of how to construct the scale. An alternative is to predict each response. The individual can choose 1-5 and as this is categorical (unless you can construct a scale), then the methods for predicting categorical (also called nominal, qualitative, etc.) data apply. These include multinomial logistic regression, discriminant analysis, nearest neighbor regression (this [http://www-bcf.usc.edu/~gareth/ISL/] is a good intro and its free). The problem is these are not really designed for repeated measures data. If it were me, after graphing the data, I would probably examine methods for multinomial multilevel models, but these can be complex.

My advise would be to talk with people in the area and see if it is sensible to construct a collectivist scale from those 5 responses. You might try different values.  You could even estimate values the maximize the correlation between that scale and a collectivist personality trait is you have one recorded.

Dan