Raw-density alpha profile
How the fit changes with α, for the thinned data and for the full data. Same method as in Alpha calibration, using Model 4a. LL(b) closer to 0 means a better fit.
| α | Downsampled LL(b) | Raw-density LL(b) |
|---|---|---|
| 1.0 | -0.6881 | -0.7127 |
| 2.4297 | -0.6701 | -0.6928 |
| 3.0 | -0.6707 | -0.6924 |
| 4.0 | -0.6725 | -0.6920 |
| 4.6159 | -0.6919 | |
| 6.0 | -0.6920 |
With thinned data, the best α is clearly between 2.4 and 2.6. With the full data, the best α is 4.6159. But between α = 3 and α = 6, the fit changes by less than 0.0009. So the full data does have a best value, but it is much less certain.
Part of Raw-density refit.
Linked from Alpha calibration · Complexity and transfer · Does downsampling matter? · Raw-density refit