Hi all,
Just wonder if you may help with this: I have a research design that include 6 different creativity scores (verbal and figural creativity tests, each with three sub scales) all are of scale order. My assumption is that they are correlate with report on well being which are also quantitative and on scale order.
Should my analysis include simple correlations between the different creativity tests?
Also, in regard to my main analysis, should a stepwise regression would be suitable ( with 6 creativity scores as independent and well being score as dependent) ? I also want to control for age and sex on the same analysis, how should I do that?
You don't mention if you use well established instruments for creativity. If they are indeed validated you could go on with tha statistical analysis.
The statistical method you should use depends on data and which type of associations you are interested in. You write that data is "scale order", so I assume they could be analysed like any other numeric variable (like length and weight). The assumption is however rather strong, and if isn't explicityly stated for the instruments you use I wouldn't take it for granted. It might be sensible if there are many levels (10-20 levels) and your data set is large.
If they really are scale data you could use standard correlations between scores for wellbeing and creativity. This analyses is symmetric, in that both variables have a measurement error.
When controlling for sex and age you may use a regression model. Standard univariate multiple regression models impose an asymmetry between dependent and independent variables, assuming that independent variables are measured perfectly and dependent have a measurement error. So in this case you should choose which variable is you outcome and which is a predictor. That is, "how does wellbeing depend on creativity when adjusting for sex and age", or "how does creativity depend on wellbeing when adjusting for sex and age".
Given that decision you don't really need a stepwise procedure, you just add all three variables simultaneoulsy to you regression model.
Two comments:
1. Assuming you are interested in how creativity depends on wellbeing AND how that dependence change when adjusting for sex and age, you could first regress creativity on wellbeing (ressulting in regression coefficient b1) AND THEN regress creativity och wellbeing, sex and age (resulting in regression coefficient b2). The difference between b1 and b2 is a measure of how the dependence change when adjusting for sex and age.
2. Adding independent variables which are strongly dependent to the same regression model will give you problems, like if you regress wellbeing on more than one score for creativity.
3. All regression models come with assumptions. Regardless of the regression model you choose you should check so these assumption are reasonable. Some models are more robust than others to violations of assumptions.
I hope this will help you in your work.
Fredrik
The firs thing you should do is create scales, so that you can use the scores on those scales in your statistical tests. I've collected some useful resources on scale construction here:
https://www.researchgate.net/post/How_can_one_help_people_create_simple_scales_from_Likert-scored_items
You don't mention if you use well established instruments for creativity. If they are indeed validated you could go on with tha statistical analysis.
The statistical method you should use depends on data and which type of associations you are interested in. You write that data is "scale order", so I assume they could be analysed like any other numeric variable (like length and weight). The assumption is however rather strong, and if isn't explicityly stated for the instruments you use I wouldn't take it for granted. It might be sensible if there are many levels (10-20 levels) and your data set is large.
If they really are scale data you could use standard correlations between scores for wellbeing and creativity. This analyses is symmetric, in that both variables have a measurement error.
When controlling for sex and age you may use a regression model. Standard univariate multiple regression models impose an asymmetry between dependent and independent variables, assuming that independent variables are measured perfectly and dependent have a measurement error. So in this case you should choose which variable is you outcome and which is a predictor. That is, "how does wellbeing depend on creativity when adjusting for sex and age", or "how does creativity depend on wellbeing when adjusting for sex and age".
Given that decision you don't really need a stepwise procedure, you just add all three variables simultaneoulsy to you regression model.
Two comments:
1. Assuming you are interested in how creativity depends on wellbeing AND how that dependence change when adjusting for sex and age, you could first regress creativity on wellbeing (ressulting in regression coefficient b1) AND THEN regress creativity och wellbeing, sex and age (resulting in regression coefficient b2). The difference between b1 and b2 is a measure of how the dependence change when adjusting for sex and age.
2. Adding independent variables which are strongly dependent to the same regression model will give you problems, like if you regress wellbeing on more than one score for creativity.
3. All regression models come with assumptions. Regardless of the regression model you choose you should check so these assumption are reasonable. Some models are more robust than others to violations of assumptions.
I hope this will help you in your work.
Fredrik
David and Fredrik,
Thank you very much. I meant that my variables, all of them aside for sex, are on interval ratio scale. You suggest I would use multiple regression to test the significance of the correlation and the explained variability between creativity and wellbeing while controlling for possible confounding effect of age and sex?
- Badges
- Science topic