I'm using SPSS for my data analyses. I'm not sure about which model to use for regression analysis. in version-20, IBM SPSS provide Automatic Linear modelling option, where it trims all data accordingly and generate results. Can I use it authentically for publication purpose? Or should I need to confine my model?
NB: I have attached a model output. Please look into that also.
Thanks in Advance.
Dear Duraisamy,
I have never used this automatic variable merging feature in SPSS and I personally would not recommend it. If you do you use it you will need to justify the decisions SPSS has made for you in merging your variables.
I also noticed that the automated regression model you selected is Forwards stepwise. I would recommend backwards stepwise, consistent with the advice given by Andy Field in his book Discovering Statistics using SPSS. Again, if you use an automated process for adding or removing variables you should explain and justify the choices it has made.
In summary, both automated features can be useful but should only be employed if you can explain what they are doing and why the final model obtained is in some sense the best (e.g. according to the principle of parsimony).
Hi, I would not recommend stepwise or all subsets regression because they are unstable (small changes in the data make large changes in the variables selected). See the first two attachments for details. What you can use are lasso or gradient boosting. See the third attachment for details. Hope this helps a bit. David
Dear all,
can somebody please give me a hint re analysis weighting in ALMs?
As far as I understand, it is a way to give more weight to observations with less variability.
My dataset includes a number of subjects assessed twice (e.g. repeated measure) within a few days. Thus, each subject in the dataset has a Time=1 and Time=2 assessment. Would it make sense to weight the analysis based on subject's ID and/or time, in order to give more importance to more stable variables? I've been looking around but I don't seem I can find the answer I'm looking for.
I know I can used mixed models with time as a repeated measure, but would like to use ALM too. Sort of to confirm the findings with two different analyses.
Thanks a lot in advance if you'll find some time to answer.
- Badges
- Science topic
- Similar topics
- Mathematics
- Algebra
- Matrix