I am currently working on my dissertation and am trying to convert 3 separate count variables/multiple response variables for use in multiple regression. I am using SPSS to analyze the data. Any help would be appreciated. Thanks.
Why should you want to transform them? Use a Poisson, a negative-binomial or a quasi-Poisson model -- these are made to model count values. There is no need to transform anything.
Why should you want to transform them? Use a Poisson, a negative-binomial or a quasi-Poisson model -- these are made to model count values. There is no need to transform anything.
Dear Will,
Please, I think you have had it all. Follow the advice of Jochen and you will reap the best and robust findings. Do not try to recode or transform your count dependent variables into continuous ones to merit multiple linear regression. As Jochen put it, count dependent variables are perfectly handled by Poisson regression. Even if you have issues of under/overdispersion, please consider the alternatives modeling strategies of negative-binomial or a quasi-Poisson regression using Generalized Linear Models (GLM).
Please, have a quick look at the following link for a systematic approach:
https://statistics.laerd.com/spss-tutorials/poisson-regression-using-spss-statistics.php
Best wishes!
Razak, thank you. My dependent variable is not a count variable but instead continuous and my independent variables are the ones that are count variables. I don't believe a Poisson could be used.
If it is an independent variable then you might not transform it, except to linearize the relationship between this and the dpendent variable.
Jochen Wilhelm , Thank you. That is what I am having trouble with. How do I transform the data to see the linear relationship?
Again the best answer is to model it as a potentially non-linear relation, and you can do this for multiple predictors simultaneously (and include categorical predictors) ; see
Article “Moving Out of the Linear Rut: The Possibilities of Generali...
Article Generalized Additive Models, Graphical Diagnostics, and Logi...
https://www.researchgate.net/post/what_is_the_difference_between_generalised_additive_model_and_polynomial_regression
In the multivariate case the simple plot of Y on one X may not be revealing and that is why partial residuals are often used (as indeed they are in the GAM).
https://en.wikipedia.org/wiki/Partial_residual_plot
They may be issues if the X variables are highly (linearly) correlated.
Check out this video. Hope this helps
https://www.youtube.com/watch?v=ScV2FOlx41A
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS