Statistical Modeling with SAS/STAT Cheng Lei [email protected] Department of Electrical and Computer...
-
Upload
emory-hampton -
Category
Documents
-
view
213 -
download
1
Transcript of Statistical Modeling with SAS/STAT Cheng Lei [email protected] Department of Electrical and Computer...
Statistical Modeling with SAS/STAT
Cheng [email protected]
Department of Electrical and Computer EngineeringUniversity of Victoria
April 9, 2015
Outline❖Overview of SAS/STAT❖Statistical Models❖Classes of Statistical Models❖Next step
SAS/STAT Overview• Over 70 procedures
• Most of the procedures dedicated to solving problems in statistical modeling
• The Output Delivery System (ODS) included to draw all kinds of graphs and produce different format result reports (PDF, HTML, CSV, and etc.)
Statistical Models• Deterministic and Stochastic Models
• Model-Based and Design-Based randomness
• Model Specification
Deterministic & Stochastic Models• Deterministic Models
• The relationship between inputs and outputs are theoretically in deterministic fashion
• Very important theoretical tools• But impractical for experimental data
Deterministic & Stochastic Models• Stochastic Models
• The outputs are uncertainly and affected by some parameters
• Parameters are unknown constants and needed to be estimated based on the assumptions about the model and the data
Model-Based & Design-Based Randomness• Model-Based Randomness
• Innate randomness• Source of random variation comes from the
unknown variation in the response
• Design-Based Randomness• Induced randomness• Random variation in the data is induced by
random selection
Model Specification
• Model selection• diagnosis• discrimination
Classes of Statistical Models• Linear & nonlinear Models
• Univariate & Multivariate Models
• Regression Models & Models with Classification Effects
• Fixed, Random, and Mixed Models• Generalized Linear Models• Latent Variable Models• Bayesian Models
Univariate & Multivariate Models• Univariate Models
• Each response variable is modeled separately
• i.e.:
• Mutilvariate Models• Multiple response variables are modeled
jointly• i.e.:
Latent Variable Models
• Involve variables not directly observed in the research
• Make a hypothesis the factors as the potential ones
• Apply regression models• Apply the evaluation methods to
validate the model
Next week’s work• Cluster Procedures• Classifier Procedures• Output Delivery System
• Ways to plot graphs• Ways to form result reports
Thank You!!!