I have a dataset of 5 variables of quantitative continuous type: 4 independent and 1 dependent (see attached). I tried using linear multiple regression for this (using the standard lm function in R), but no statistical significance was obtained. Then I decided to try to build a nonlinear model using the nls function, but I have relatively little experience in this. Could you help me, please: how to choose the right "equation" for a nonlinear model? Or maybe I'm doing everything wrong at all? So far I have used the standard linear model in the "non-linear" model.
I would be very grateful for your help.
If you do not have the opportunity to open the code and see the result, I copy it here:
------
library(XLConnect)
wk
Your DV is discrete with values between 1 and 8. It's distribution seems left-skewed. I don't think that the assumption of normal distribution is reasonable here. I'd recommend using a beta-regression. I also doubt that the data justifies a nonlinear model (unlessyou have a theory suggesting a model).
It sounds like you already tested your hypothesis with the linear model and used up your p value. So, you can't do more p-value tests.
As far as choosing a model, that isn't a question for the stats folks, but is related to the theory you have (stats folks can say how to implement or approximate the model).
hello Michael Zastrozhin
well, hope you are making all these efforts for the paper publication which might be useful for the thesis or any other recruitment.
Regarding research, you need to try a few more algorithms and see which one is giving logical results.
However, regarding lm function is not promissing these days in higher publisher. Nonetheless, you can find about nlm in followings
1) https://www.rdocumentation.org/packages/stats/versions/3.6.2/topics/nlm
2) https://stat.ethz.ch/R-manual/R-devel/library/stats/html/nlm.html
3) https://stackoverflow.com/questions/40620277/r-how-can-the-nlm-function-be-used-for-optimization-with-multiple-variables
Good Luck!
You should fit a model that makes sense to you as a researcher and then focus on the estimates.
If you're switching to a nonlinear model because your linear model was not satisfactory in terms of pvalues means that you are doing exploratory research
Small pvalues are not the outcome of statistical analyses, fit something reasonable and focus on the results other than pvalues, aka estimates, out of sample predictions etc
It seems you don't want to accept what the data is telling you, but the goal and information you want, "statistically significant!"
How did you know that the fit should be non linear? Any plot, or were you actually designing for best information about your research questions? If your "normality test failed DV" is greatly left skewed, non linear approach is not the way!
As a short answer :
1 - your DV is, as said before, severely skewed, 56% is below or equal to 6, the rest equal to 7. You should first transform the DV to binary (< or == to 7)
2 - a short graphical description will show
2.1 - that IV1 is independent from the others IV,
2.2 and IV2 to IV4 are more or less correlated.
3 - the box plots show that IV1 may be linked to the DV, the others IV are probably not
4 - this is confirmed by a mere logistic regression where the p-value is significant only for IV1
Happy New Year
PS : R code attached, the data were first translated to a *.text file, but that is idiosyncratic.
Dear Jean Maccario,
Thank you very much for the most complete answer and the attached R file. I am sure it will be useful not only to me, but also to other researchers who use regression models...
Happy New Year!
Kind regards,
Michael
- Badges
- Science topic
- Similar topics
- Correction