Biostatistics 5 QA Exam Training

Biostatistics 5 QA Exam Training

• ChatGPT Acronym: Generative Pre-trained Transformer
• Python Library for ChatGPT Deployment: TensorFlow
• Assumption of Linearity: Ensures predictor and response variables have a linear relationship, residuals are normally distributed, predictors are independent of one another and variance is constant across observations.
• ChatGPT's Biostatistics Paper Summarization: Validates statistical results, produces concise summaries from abstracts, extracts graphs and charts, runs the code for analysis.
• Common Errors Using ChatGPT for Statistical Code Generation: Syntax errors, contextual misunderstandings, hardware dependencies, mathematical calculation errors.
• Key Ethical Consideration in Academia using ChatGPT: Properly acknowledging AI assistance, avoiding sensitive prompts, limiting the use of advanced models and testing the generated code for performance.
• Effective ChatGPT Prompt Features: Reduces model overfitting, avoids errors and vague responses, optimizes neural network performance.
• Durbin-Watson Test: Used to detect autocorrelation in regression model residuals.
• Durbin-Watson Test Values:
  • Value around 2 indicates no autocorrelation
  • Value close to 0 suggests negative autocorrelation
  • Value close to 4 indicates positive autocorrelation.
• ChatGPT Coding Advantages: Rapid prototyping, suggestions.
• ChatGPT Least Effective Tasks: Designing complete biostatistical studies.
• Validating ChatGPT Generated Code: Cross-check with documentation, test in software.

Further Biostatistics Concepts

• Dunnett's Test: Compares multiple treatments to each other or a single control group.
• ANCOVA (Analysis of Covariance): Adjusts for categorical group effects (covariates) to analyze continuous variables.
• Homogeneity of Regression Slopes (ANCOVA): Ensures identical mean values across groups regarding covariate and dependent variables.
• Cross-Validation: Evaluates model generalizability by splitting the data.
• LASSO Regression: Shrinks coefficients to zero for feature selection, uses quadratic penalties to avoid multicollinearity with predictors.
• Comparing Regression Models (AIC): Lower AIC value indicates a better-fitting model.
• VIF (Variance Inflation Factor): Measures multicollinearity.
• VIF Values and Multicollinearity:
  • VIF = 1: Suggest no correlation among predictors
  • VIF > 5: Suggests no perfect independence among predictors
  • VIF > 10: Suggests high multicollinearity among predictors
• Residual Plots: Identify non-linearity or heteroscedasticity in residuals.
• Outliers Identification: Large residual values, high VIF values.
• R-squared Value in Regression Models: Measures the percentage of variance explained by the model.
• Mixed-Effects Models (Random Effects): Account for variability specific to individual subjects or clusters.
• PCR (Principal Component Regression): Uses uncorrelated principal components as predictors for high multicollinearity datasets.
• Scheffé's Method: Compares all possible contrasts of treatment means with a control.
• Homoscedasticity: Constant variance of residuals across all predictor levels.
• Independence of Residuals: Ensures multicollinearity is minimised.
• Q-Q Plots: Evaluate normality of residuals.
• Multicollinearity Effects: Inflates standard errors of coefficients.
• AIC Value and Model Selection: Lower AIC value signifies a better-fitting model.
• Random Effects in Mixed-Effects Models: Accounts for variability and treatment-level effects in individual subjects or clusters.