Modeling Data Training Overview
This Modeling Data for Inference course teaches attendees how to use Python to perform causal inference on observational data. Participants learn how to work with inferential models, missing data, and experimental design.
Location and Pricing
Accelebrate offers instructor-led enterprise training for groups of 3 or more online or at your site. Most Accelebrate classes can be flexibly scheduled for your group, including delivery in half-day segments across a week or set of weeks. To receive a customized proposal and price quote for private corporate training on-site or online, please contact us.
In addition, some courses are available as live, instructor-led training from one of our partners.
Objectives
- Perform causal inference in observational data using Python
- Perform and interpret null hypothesis testing in Python
- Implement generalized linear models in statsmodels
- Understand missing data
- Impute missing data
- Generate accurate power calculations
- Implement non-parametric methods to test hypotheses.
- Use causal inference frameworks to identify causal effects from observational data
Prerequisites
Attendees must have a solid foundation in Python programming for descriptive analytics.
Outline
Expand All | Collapse All
Introduction
GLMs with Python using Stats Models
- Applying Statistical Models for Analysis in Python: The A/B test
- Explanation of statsmodels library of functions
- Inferential and descriptive statistics refresher
- Implementing A/B tests
Modeling Continuous Data (Linear models)
- Formulation of the simple linear model
- Application of the intercept only, null model
- Binary predictor
- Interpreting results
- Categorical predictor
- Continuous predictor
- Polynomial expansions
- Multiple linear regression
- Spline models
- Interaction terms
- Picking the “best” model
- Discussion of confounding, interaction terms, and model building approaches
- Modeling Binary Data (Logistic models)
- Discussion of the generalized linear model
- The Logit link function
- Binomial distribution
- Intercept only model
- Back transformation of coefficients
- Simple predictor
- Multiple predictors
- Odds ratio interpretations
- Generating a scoring data set
- Predicting from the model with new data
- Modeling Count Outcomes
- How are count outcomes different?
- Poisson models
- Over dispersed modeling options
- Log link functions
- Using offsets to model rates / uneven follow-up
Power Analyses/Study Design
- Understanding and estimating statistical power
- Type 1 and type 2 errors
- Using existing power estimators
- Simulating power through the data-generating process
Non-Parametric Analysis Methods
- Using bootstrapping/permutation tests
- Bootstrapping versus depending on asymptotic behavior to estimate confidence intervals
- How different/stable are my results?
- resampling a data set
- bias-corrected bootstrap interval
- Extending the bootstrap function to calculate more statistics
- Permutation tests for p-values
Missing data
- Quantifying
- Visualizing missing data
- MAR,MCAR,MNAR
- Sensitivity analysis
- Imputation
- MICE/trees pre-processing
Time to Event (Survival) Analysis
- Visualizing Hazards Across Time
- Understanding the Log Rank Test
- Cox Proportional Hazards Modeling
- Understanding and interpreting the Hazard Ratio
- Model diagnostics and assumptions
- Implementing Time Varying Covariates
- Parametric Survival Models
- Weibull Model
- Exponential Model
- Predicting Failure Times
Causal Inference: The Potential Outcomes Framework
- Defining treatment effects (ATT, ATE)
- Identifying populations of interest
- Defining your causal hypothesis
- Understanding the counterfactual
- Establishing the causal diagram for your problem
- Different methods for conditioning on variables:
- Propensity Scores
- Direct regression adjustment
- G-computation formulas
- Instrumental variable analysis
Conclusion
Training Materials
All Data Modeling training students receive comprehensive courseware.
Software Requirements
- Windows, Mac, or Linux
- A current version of Anaconda for Python 3.x, or a comparable Python installation with the necessary libraries (Accelebrate can provide a list)
