I assume you wish to predict things using the model that would be developed. If so run an adaptive lasso variable selection as your regression method. This produces a model that is optimal for the variables selected. Since you have 4 DVs i would select 4 models.For your ordinal dv I would use ordinal adaptive lasso regression or convert to a binary variable as is most common in risk studies. An easy to use R program for your continuous DVs can be found here:

https://www4.stat.ncsu.edu/~boos/var.select/lasso.adaptive.html

Ordinal logistic regression programs can be found here:

https://www.google.com/search?q=adaptive+lasso+ordinal+logistic+regression&rlz=1C1CHBF_enUS874US874&oq=adaptive+lasso+ordinal+logistic+regression&aqs=chrome..69i57.45525j0j4&sourceid=chrome&ie=UTF-8

I am attaching a complete (with protocol) study to show you what to do. Adaptive lasso is not only optimal but does not allow multicollinearity. Do NOT use BMI on both sides of your model for obvious reasons. If you have questions Contact me at Researchgate. Good luck, David Booth

Dear Dr. Booth,

Thank you very much for your detailed response and for sharing the valuable documents. I will read them carefully and let you know if I have further questions. I really appreciate your help!

Best Regards

Vu

Hi Vy,

The Spearman correlation just tells you about the trend of the association between studied variables -/+. The regression models could help you to answer the research questions despite the correlation level. The models are decided by the characteristic of the DV. In your study, WC can be dichotomized as 'normal weight' vs 'abdominal obesity'; BMI can be classified into 'normal weight', 'overweight', and 'obesity'; Disease risk level can be classified into e.g. low vs high. You should try the logistic regression model for those three outcomes. Regarding the diet quality index, if this outcome is normally distributed, you may try the linear regression model, otherwise, you should classify them into different quartiles or tertiles, or use the cut-off value from a previous study, the logistic regression model is suggested.

You should firstly run simple/bivariate regression models and then multiple/multivariate regression models.

There are different ways to select variables into the multivariate model: epidemiological or statistical approaches. Regarding statistical approach, some researchers select the variables with p< 0.2, others use p< 0.1 or p< 0.05 in the bivariate model.

Be referred to this article: Maldonado, G., & Greenland, S. (1993). Simulation Study of Confounder-Selection Strategies. American Journal of Epidemiology, 138(11), 923-936. doi: https://doi.org/10.1093/oxfordjournals.aje.a116813

Hope this could help you out.

Cheers,

Duong

Dear Dr. Duong,

This is a great answer. I understand the problem now. Thank you so much for your explanation and the recommended material.

Best Regards,

Vy

I would do the classification of dependent variables and run logistic regressions as recommended above or test the normality of the dependent variable. others may have different distribution pattern and may need transformation before running linear regression or some may need different type of regression e.g.. poisson or negative binomial models

Thank you very much for your advice.

Best Regards,

Vy