Hello,
I doing a study with n = 60 and 100+ variables (both continuous and categorical). I want to investigate how different variables affect the possibility to get a specific bacteria.
I have done a shapiro test (for continuous variables to see if they a normal or nonnormal) and Fisher Exact (for categorical variables since I have a small sample size). By doing both I got some p-values.
I have been told that I need to work further with my p-values, by doing either Lasso test or Logistic regression. However, I'm in doubt as to which one to choose and which variables I should analyze by these two test? Is it both continuous and categorical? is it generally all my variables or should I pick some?
Thank you
What are you trying to determine by using Shapiro (Shapiro-Wilk ?) test ? What are you trying to determine by using Fisher Exact test ? These two tests aren't similar in their purposes.
What are you trying to determine where someone is telling you to use Lasso regression or logistic regression ?
Do you have a dependent variable and independent variables ?
What are you trying to do overall ?
Hello Anna,
I concur with Salvatore that some insight into your specific research aim(s) and the nature of the variables would be helpful in making recommendations.
A general comment: If your goal is model building/variable selection, and you have a small sample relative to the number of variables, then it would be important to think about a method that invokes suitable penalties to help avoid overfitting. Lasso is indeed one such method; the data mining community has come up with a number of possible methods, each with its own virtues and limitations.
Good luck with your work.
Choosing a sound statistical procedure rests on three key considerations: Study purpose and objectives, data type and distribution, and nature of observations. Without details on those considerations, selecting an appropriate statistical method is not possible. Here is a reader-friendly textbook.
Jasrai, L. (2020). Data analysis using SPSS. SAGE Publishing. https://us.sagepub.com/en-us/nam/data-analysis-using-spss/book274530
Good luck,
"I am trying to answer which variables affect the risk of getting a specific bacterium in the hospital."
What kind of variable is this? Are you looking at counts of cases? Maybe out of total patients? Or are the cases graded in some way (none, minor, moderate, major)? Or some other type of variable ?
Shapiro-Wilk test
It may not matter whether the individual variables are normally distributed or not, but it is useful information. It's probably more helpful to look at the histograms of individual continuous variables. Some may be bell-shaped or symmetric. But if they are notably skewed, sometimes it makes sense to transform them before inclusion in multiple regression.
The tests that should be used depend on your research hypotheses but in any case the dependent variable remains decisive in the choice of the test. for example if the dependent variable is binary (0,1) it is better to think about the binary multivariate logistic model and if it is nominal with several levels it is the multinomial multivariate logistic model and so on
- Badges
- Science topic
- Similar topics
- Psychology