Unit 11 of 12
Dependent Data and Time Series
- 3 lessons
- Videos 74 to 76
Repeated readings and values recorded over time are often not independent. This unit shows how that dependence changes standard errors, the choice of analysis and the way forecasts are checked.
Readings from the same cluster or person are often positively correlated, and treating each row as independent then understates the uncertainty. The intraclass correlation measures that similarity, the design effect measures how much it inflates the variance of an estimate, and the analysis should rest on the units that were sampled or assigned independently. Time series add trend, seasonality and autocorrelation. Moving averages, exponential smoothing and simple baseline forecasts are then evaluated forward in time, so that no forecast uses information from after the time it is made.
Every lesson will be one video. The videos are coming soon, and each lesson page already has the definitions and an example for every key term.
Clustered and Repeated Measurements
Repeated Measures, Clusters and Independence
Why readings that share a device, a person or a classroom are often not independent, and how the intraclass correlation, the design effect, the cluster bootstrap and a random-intercept mixed model account for clustering.
Video 74
Time Series and Forecasting
Time Series, Trend, Seasonality and Autocorrelation
What a time series is, how trend, seasonality, lags and autocorrelation describe ordered observations, what stationarity means, and why serial dependence changes the standard error of a mean.
Video 75
Smoothing, Forecasting and Time-Aware Validation
How naive, seasonal naive, moving average and exponential smoothing forecasts work, how rolling-origin evaluation at a stated forecast horizon avoids temporal leakage, and what a forecast interval needs.
Video 76