Additional fit statistics
Pairing and regression help
Choose two different datasets already added to the chart. Each dot pairs observations at the same date, within the main chart's selected date range. Each dataset uses its own Units transformation and Modify frequency settings from Edit Line. No interpolation or forward filling is used. To compare different frequencies, first set both datasets to a common frequency and choose their aggregation methods.
The selected curve is fitted automatically by least squares in the original Y units. Exponential requires positive Y; logarithmic requires positive X; power requires positive X and Y. Invalid domains are reported, not silently excluded. Linear needs 2 pairs, quadratic 3, cubic 4, and sigmoid at least 5, with enough distinct X values.
Fit quality (R²) measures how closely the curve fits these observations in the original Y units. Higher is better: 1 is a perfect in-sample fit, 0 is no better than predicting the average Y, and negative values are worse. The badge runs from red at 0 or below to green at 1; undefined values are neutral gray. Color is a visual guide, not a statistical significance test. Compare models on the same pairs, and remember that extra parameters can increase R² through overfitting.
Additional fit statistics are available separately. Adjusted R² penalizes extra parameters and needs more pairs than parameters. Pearson r measures the dots' linear correlation, not nonlinear fit quality. R² and r are undefined for constant Y. RMSE is in Y units.
Sigmoid uses a four-parameter logistic with fitted lower/upper plateaus, slope, and midpoint, and supports increasing or decreasing curves. Fits without identifiable parameters or numerical convergence are reported. Equations use a centered/scaled variable z where needed for numerical stability; displayed coefficients are rounded.
Compare fits on the same pairs. Flexible curves can overfit, and shared time trends can create misleading correlations. These descriptive fits are not evidence of causation or out-of-sample accuracy.
