Please look at the following below links which may help you in your analysis:

http://personal.cb.cityu.edu.hk/msawan/teaching/FB8916/FB8916Ch1.pdf

http://web.nchu.edu.tw/~numerical/course1012/ra/Applied_Regression_Analysis_A_Research_Tool.pdf

http://statlab.stat.yale.edu/workshops/IntroRegression/StatLab-IntroRegressionFa08.pdf

http://home.iitk.ac.in/~shalab/regression/Chapter2-Regression-SimpleLinearRegressionAnalysis.pdf

Thanks

This depends on the data. One could try linear regression, but you have to carefully check the fit and the assumptions. This is model building and needs some basic statistical skills. See e.g.

https://cran.r-project.org/doc/contrib/Faraway-PRA.pdf

Normally, in a clinical study, you should have planned that BEFORE you do the study. In hope the study was randomized as well. Otherwise the results might be confounded and then its just a observation study.

I didn't understand well, but if the dependent variable is categorical, you should try logistic regression.

Thank you for answers,

prof Nicoleta, the dependent variable is continuous not categorical, however I can shift the result to categorical, but the question here is 30 patient sufficient for doing regression for 6 independent variables with 1 dependent variable?

In theory, a sample of 30 observations is enough, but never in practice! My opinion.

Dear, can you give me a paper that confirm a sample of 30 is enough?

Because I already did a clinical trial of 2 arms (each arm is 30 patient) so I can't increase my sample size

Hi, Ehab,

Your question asks "what is the best regression analysis?".

This modern, exact, non-parametric alternative to regression analysis always provides the most accurate model, by definition:

• https://odajournal.com/2016/09/19/novometric-analysis-with-ordered-class-variables-the-optimal-alternative-to-linear-regression-analysis/

This methodology does NOT require large samples to achieve satisfactory (90%) statistical power:

• https://odajournal.com/2020/02/08/statistical-power-analysis-in-oda-cta-and-novometrics-invited/

And, while this approach always identifies the most accurate model which is possible for your data and hypothesis, it does NOT over-fit. That is, (1) legacy methods identify a less accurate model when an effect in reality exists; and (2) the legacy methods all find effects in random data while the modern approach does not find effects in random data:

• https://odajournal.com/2019/03/21/some-machine-learning-algorithms-find-relationships-between-variables-when-none-exist-cta-doesnt/
• https://odajournal.com/2019/04/11/effect-of-sample-size-on-discovery-of-relationships-in-random-data-by-classification-algorithms/
• https://odajournal.com/2019/04/22/multi-layer-perceptron-neural-net-model-identifies-effect-in-random-data/
• https://odajournal.com/2019/05/16/logistic-discriminant-analysis-and-structural-equation-modeling-both-identify-effects-in-random-data/

Dear all

If my sample is 35, and the data (some of independent variables are abnormal), can I do multiple linear regression (since my data is more than 30)? which assumption should I meet to do multiple regression in such a case (the dependent variable is normal)

In summary how can I use linear regression and avoid non-linear regression?

Your last comment stated that you wanted to "avoid non-linear regression".

Why?

If one has nonlinear trends in their data, then avoiding a nonlinear model is a type of error known as model misspecification. Sometimes a non-linear model is just what is needed! Here's how to tell when a non-linear model is appropriate (note that distributional assumptions speak to the validity of the p-value).

• https://odajournal.com/2019/01/19/when-to-evaluate-a-nonlinear-model/

linear regression is probably the first type you should consider. with special options available for linear regression.

You can find the minimum sample size for regression: https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0229345

I have my own statistical book, but it is written in Romanian.

See the section on mutivar. longitudinal models:

https://www.google.com/search?q=Spaghetti+plot&oq=Spaghetti+plot&aqs=chrome..69i57.8655j0j1&sourceid=chrome&ie=UTF-8

Davidian Notes

https://www4.stat.ncsu.edu/~davidian/st732/notes.html

Multivar. Long. Models:

https://www.researchgate.net/post/What_statistical_test_is_appropriate_for_pre_and_post_trails_with_1_before_treatment_value_and_4_after_treatment_values_assessed_weekly

Best D. Booth

Multiple linear regression. In interpretation, however, look at exponentiated regression coefficients.... Sample size concerns though with many predictors??

https://statisticsbyjim.com/regression/choosing-regression-analysis/

Cordial saludo, eso depende de la base de datos que tengas y cual sea el ajuste de los datos, debes modelar cada uno de ellos para revisar realmente la afinidad de las variables independientes respecto a la dependiente, puede ensayar con una logística a ver como le va. éxitos

Dear Ehab Mudher Mikhael ,

Hi,

You can try NTSYS-pc in this special case to get linear regression with confident ellipse curve.

Best,

Saeed

Dear Ehab,

Also, I refer you to the article link below

Article Gathering, Identification, Medicinal Utilization and Domesti...

Best,

Saeed

I think since, you told that the log transformation of the variables, provided you with the distribution being Normal. Hence, an option might be to just take the logarithmic values of the dependent variable, and then fit a linear regression as done usually.

Thank you all for your valuable answers

Linear or logistic regression depends on the type of outcome. If your outcome is quantitative without categorization then multiple linear regression is quite reasonable to do. Assumptions should be checked before any analysis, pl don't forget that too.