Study Notes on Excel Sampling, OLS Assumptions, and Statistical Testing

Introduction

• Overview of topics discussed in class, emphasizing the use of Excel for population simulation and OLS assumptions.

Excel and Random Sampling Functionality

• Mention of a gauge referred to during the discussion.
• Utilization of random sampling in Excel for simulating populations.
  • Introduction to the RANDBETWEEN function which generates random variables within a specified range.
  • Example: =RANDBETWEEN(lower, upper)
  • Explanation of how to set the number of samples and observations for better simulation results:
  • For instance, to create 50 samples, each with a size of 5, users can fill down from a sequence of 1 to 5.
• Importance of copying and pasting values to avoid regeneration of random numbers upon recalculation.

Statistical Concepts in OLS Assumptions

• Discussion on OLS (Ordinary Least Squares) assumptions for achieving the Best Linear Unbiased Estimator (BLUE).
  • Importance of imposing assumptions to minimize variance of coefficient estimates (beta hats).
• Variance of beta one is influenced by multiple terms related to assumptions regarding error terms.

Heteroskedasticity

• Definition: Heteroskedasticity refers to the condition in which the variance of the error terms is not constant across observations.
  • Example given of how underestimating standard errors due to incorrect assumptions leads to faulty hypothesis tests.
• Relation of standard errors to variance calculations:
  • Under heteroskedastic conditions, the standard error increases, causing issues in statistical inference.

Autocorrelation

• Definition: Autocorrelation occurs when errors are correlated with each other, typically seen in time-series data but can also manifest in cross-sectional datasets.
  • Discussed using residuals from previous observations (e.g., observation I and I-1) and their roles in regression analysis.
• Effect: Increased standard error estimation harming the validity of hypothesis tests if autocorrelation is present.

Testing for Assumptions Violations

• Methods of testing for heteroskedasticity and autocorrelation:
  • Regression of residuals on the variables, including clustered assessments if necessary.
  • Use Chi-square distribution in hypothesis testing for significance.

R-squared and Adjusted R-squared

• R-squared (R²): represents the proportion of variance in the dependent variable explained by the independent variable(s).
  • Formula: $R^2 = \frac{SS<em>{explained}}{SS</em>{total}}$
  • Criticism: A high R-squared does not imply a good model; can reflect added variables that may not contribute meaningfully to explanatory power.
• Adjusted R-squared: Corrects the R-squared value that accounts for the number of predictors in the model to avoid artificially inflating results.
  • Introduced for better assessment of models with multiple predictors.

Error Terms and Their Implications

• Significant errors in regression impact standard errors of coefficients:
  • Highlighted the unpredictability introduced by not controlling for certain variables.
  • Mentioned terms like "residuals" (u hats, epsilon hats) interchangeably, understanding that both denote estimation errors affecting hypothesis testing.
• Collective variables affecting performance, such as shared shocks influencing error terms in clustered data:
  • Example included discussions on school performance correlation based on common district expenditures and unobserved factors.

Conclusion Remarks

• Emphasized ongoing learning regarding OLS assumptions, heteroskedasticity, and autocorrelation through practical applications.
• Possible assistance from external resources like AI for data analysis issues.
• Exit ticket questions for review and application of learning:
  1. Which assumption violations can be fixed with robust standard errors and which cannot?
  2. Link to material on Blackboard for further resources.