I have a dependent variable (resilience scores) that was devised by summing Likert scale (1-5) responses. I was advised to consider this as a continuous dependent variable. I have 6 independent variables, all of which are categorical (ordinal). I would like to understand how the independent variables impact resilience scores. My data is not normally distributed. What type of regression should I use? Or would I generate better results if I dichotomized the resilience scores (into "high" and "low" categories based on 50th percentile) and run a logistic regression? I should mention that I have a fairly small sample (n=84). Thank you!
You can adjust your data by evaluating the distance of mahalanobis and eliminate atypical data of .000, through the structural equation model (SEM) uses path analysis and multiple regression, which is robust for its analysis that has estimators that can adjust to the lack of normality of your data, and allow to strengthen the proposed theoretical-structural model.
Hi Lorian,
First, I would not use SEM as suggested by Christian as this approach would require significantly more power than you have with such a small sample size. I also would not dichotomize the outcome without a strong theoretical reason to do so. This is because a spit introduces additional error at the median and sacrifices meaningful information. It also makes the analysis a logistical regression, which also requires a larger sample size than you have.
I would rather prefer OLS multiple regression with a continuous outcome using the categorical variables as predictors. Regarding the normality of your outcome, the variable itself does not have to be normal. Rather, the residuals must be approximately normally distributed.
Hope this helps.
Cory
- Badges
- Science topic
- Similar topics
- Biological Science
- Kinesiology
- Running