I have a continuous data, but not counted data such as ( 1,2,3,4,5 etc), I have one dependent variable with many zero value, I can not delete the zero value, also I have a 3 independent variables, they are not normally distributed and they violate few assumption such as the linearity, Violations of independence , what is the best statistical test I can perform to get a prediction formula such as y=a+B1X1+B2X2+B3X3+B123X123+e
Count values should be analyzed with a Poisson- a quasi-Poisson or a negative binomial model. Independence can be acknowledged in hierarchical/multilevel models (by random effects).
Jochen Wilhelm Thanks for your answer but the dependent variable is not a count value, also has many zeros.
You do not need a statistical test, you need a fitting routine. After fitting you can examine if you have enough data and if the equation has any meaning for the data, including the zero values. The distribution of the zeros within the data are important if you cannot order the data by values. You can assume normality, linearity, etc and then test the importance of those assumptions.
It is difficult to recommend actions for data and effects that are vague or unknown. More information is necessary for a recommendation for your case.
Joseph L Alvarez Thank you for your inspiring answer.
i thought of what you said. the data can not be fitted into a count value, but it can be transformed to ranking value. as well independent variables can transform to a ranking value.
Thanks Paul Yarnold it is good to learn about new analysis in the field, any way I am going to use Cox regression. the model fit with my data and my assumptions.
Thank you I appropriate your contribution
Paul Yarnold I am trying to simplify my study to finish my dissertation, I have one dependent variable with many zero value with not normally distributed, I can not delete the zero value, also I have a 3 independent variables, they are not normally distributed and they violate few assumption such as the linearity, Violations of independence>
If i send you the data can you perform the test and then let me know if the model fit> any way I am willing to pay for the right analysis. so let me know if you can help me with this test, also it would be a good to use a new test in my dissertation and present this method in my oral defense.
Best regards
Paul Yarnold Thank you Prof. for all your contributions, and your big help, this method will increase the accuracy of my paper, as well will present better social change.
- Badges
- Science topic
- Similar topics
- Condensed Matter Physics
- Zero