Detect obscure patterns in dataset

Moderators: statman, Analyst Techy, andris, Fierce, GerineL, Smash

Christian25
Posts: 7
Joined: Wed Dec 07, 2011 4:50 pm

Detect obscure patterns in dataset

Postby Christian25 » Wed Dec 07, 2011 5:12 pm

Hi dear board members,

I have got a dataset of a psychological study, in which several variables were collected.
As the study is meant to indicate some causality, I have classified the variables as independent and dependent variables:

Independent variables:
Personality trait factors 1 to 5 (a scale from 0 to 48 each), age, gender, current mood (scale from 1 to 7). I.e., there are eight independent variables in the dataset that may be used for the analysis.

Dependent variables:
Evaluation of several emotional stimuli (on a scale from 1 to 10).

The dataset comprises 86 cases to evaluate the influence of the indep. variables on the dependent ones.
Still, an initial correlation matrix yields only few significant results. I believe that particularly the interaction effects of the independent variables may be meaningful.

However, these interaction effects could be of such nature that they appear when one independent variable has a very low value while another shows a very high value (e.g., when one personality trait has a very low value while another is very pronounced).

Now I really need help in terms of which methods to employ in order to detect such relationships with SPSS 20. A simple computation of interaction variables by means of multiplication of two independent variables does not seem feasible if the interaction effects are of such nature as described above.

Are there any methods or techniques that may reveal something more about my data?

I'm very grateful for your support!

Kind regards,

Christian
statman
Administrator
Posts: 2756
Joined: Tue Jun 12, 2007 12:08 pm
Location: Florida, USA

Re: Detect obscure patterns in dataset

Postby statman » Wed Dec 07, 2011 8:18 pm

Check into the GLM model where you can specify interactions

From the correlations matrix, you say you see only few significant results, are these positive or negative and how significant
See the note below

NOTE: Please read the Posting Guidelines and always tell us your OS, the SPSS version and information about your study and data!

Statman
Statistical Services
Christian25
Posts: 7
Joined: Wed Dec 07, 2011 4:50 pm

Re: Detect obscure patterns in dataset

Postby Christian25 » Mon Dec 12, 2011 1:22 pm

Hi Statman,

thank you for your response. When I allow for an significance level of .1, then there are some significant positive and negative correlations between the independent and the dependent variables. For instance, age has a sign. negative effect on the evaluation of one stimulus and also some of the personality trait factors show sign. correlations - some even below .05.

However, there is one issue concerning the GLM as well as general regression analyses: The method of the experiment included a priming of the participants after the evaluation of the first half of the stimuli. Since every stimulus was supposed to be evaluated in the "neutral mood condition" (before the priming) as well as in the "sad mood condition" (after the priming), the experiment was conducted with two groups. The groups consist of 41 and 46 responses. To the best of my knowledge, these numbers are too small to conduct a regression analysis with many independent variables including interaction effects.

In order to find interaction effects, does it make sense to include the two independent variables plus an interaction variable (computed from those two) in a regression analysis with on of the stimulus rating variables as the dependent variable? Or does it make more sense to run a partial correlation of the interaction variable with the dependent (stimulus rating) variable while controlling for the two variables from which the interaction variable was computed?

I'd very much appreciate some help.

Thank you very much again.
statman
Administrator
Posts: 2756
Joined: Tue Jun 12, 2007 12:08 pm
Location: Florida, USA

Re: Detect obscure patterns in dataset

Postby statman » Mon Dec 12, 2011 3:20 pm

Perhaps you said this so just checking: What about correlations between the IVs? If have then their interaction would, mostlikely, cancel out versus enhance the IV / DV cause-effect

Also, you have use DV in a plural sense but your 1st post suggested only 1 DV - Which it right?
Christian25
Posts: 7
Joined: Wed Dec 07, 2011 4:50 pm

Re: Detect obscure patterns in dataset

Postby Christian25 » Mon Feb 06, 2012 9:19 pm

Hello again,

sorry I did not reply to this post for quite some time but I figured that I have to refine my theoretical concepts before I could progress with the statistical issues for my thesis.

I have several (i.e., six) DVs in my dataset on which the IVs may have an effect. In the meantime I have created a rather sound theoretical background that suggests several interaction effects among the IVs.

I must say that I do not fully understand the function of the general linear model. Thus, I'd rather like to check for the interaction effects "by hand" i.e., I'd rather calculate an interaction variable by multiplying two IVs with each other to test whether they have an interaction effect.

So my question: Let's say I want to test whether IV 1 and IV 2 have an interaction effect on DV 1. I multiply IV 1 * IV 2 and insert IV 1, IV 2, and IV1*IV2 in a linear regression model. When I do that, I obviously get very high VIF, indicating multicollinearity. However, when I center the IVs by subtracting each variable's mean, there are no multicollinearity issues indicated.

Can I just center variables to avoid multicollinearity or does this cause another problem?
Is my way of doing this analysis okay?

Thank you for your support!
Chris

Who is online

Users browsing this forum: No registered users and 1 guest

cron