How can I check based on residuals plots which model is suitable for my data?
What is the procedure to check?
How can I check it in SPSS and in STATA 12 Softwares?
This is a good question. Building statistical model does not only rely on the quality and quantity of your data, but also plays into the experience and expertise of the statistical analyst or whoever is conducting the model development. Good statistical model development depends on a combination of factors and not solely what you see when you click run to execute a particular statistical model whether that statistical test is a mixed regression, generalized regression, logistic model, etc.
To determine whether or not you have the right data to perform a specific set of statistical model, you first need to conduct series of exploratory data analysis and visualization. Doing this initial process once your data is manipulated and shaped in that way you want it to work relative to whether or not you are using R, SPSS, STATA, SAS, JMP Pro 12, etc allows you to identify links and relationships as well as those that are significant or not within your datasets. You could series of bi-variate analysis, multi-variate analysis, stepwise, relationships plots, etc to visualize your data and to identify salient relationships. Doing this also allows you to see if your data is normally distributed or not and whether or not you need to do some data transformation (data mining) in exponential, quadratic, cubic terms, etc to and how to control outlier as they might impact your model if no system is put in place to control the effects of outliers. Sometimes, leaving outlier in the datasets in relevant, but you can only do these once you've done preliminary data exploration and visualization even being starting the actual model development.
The residual plot is not the only sole determinant that explains a good model that fits your dataset. You could look at the results of ANOVA, R Square values of the training set (model itself) versus those of the model validation and test. The R Square value doesn't tell you whether or not your model variances is statistically significance. It only explains the portion of variability that is explain by the predictor variables of the response.
Similarly, there are other relevant statistical results once you run your model that explains whether or not your model fits well with the data. You might want to look at the prediction profile, you variable importance plot (VIP), the parameter estimates results to see, how each parameter used in the model correlates with others.
I hope this helps in someways. But in all, you choice of a statistical model given a set of datasets will depend on the data quality and quantity, your experience and expertise in data mining, exploration, visualization, and analysis as well as your technical understanding of statistical models development.
What is your data?
How many response variables do you have?
What type(s) of responses do you have?
How many predictor variables do you have?
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS