Table 1 Random Forest Model Effect Sizes with uncorrected KDEs.

From: Climate and demography drive 7000 years of dietary change in the Central Andes

Isotope

Elevation category

Temp. EF‰ (%)

Temp. seasonality EF‰ (%)

Precip. EF‰ (%)

Precip. seasonality EF‰ (%)

Cumulative climate EF‰ (%)

Demography EF‰ (%)

δ15N‰

Coastal

1.4 (5.6%)

6.5 (25.7%)

10.2 (40.0%)

1.2 (4.8%)

19.4 (76.2%)

1.6 (6.6%)

Mid-elevation

0.7 (4.1%)

2.0 (11.8%)

5.3 (29.9%)

0.9 (5.5%)

9.0 (51.2%)

3.8 (22.1%)

Highland

0.4 (8.5%)

0.4 (9.3%)

0.5 (11.0%)

0.7 (15.4%)

2.1 (44.1%)

0.5 (11.2%)

δ13C‰

Coastal

0.7 (10.3%)

0.7 (10.2%)

1.5 (21.0%)

1.2 (16.7%)

4.1 (58.1%)

0.9 (13.3%)

Mid-elevation

1.0 (6.4%)

2.6 (15.9%)

1.8 (11.0%)

3.5 (21.2%)

8.9 (54.5%)

2.7 (16.4%)

Highland

2.2 (35.4%)

0.6 (9.6%)

1.1 (18.7%)

1.1 (17.1%)

4.9 (80.7%)

0.6 (10.4%)

  1. Model sum of squares effect sizes (amount of variation in δ15N‰ or δ13C‰ explained) for each variable within the elevation categories. These values are in per mil (‰). Percentages are the model sum of squares effect sizes as a percent of the total amount of variation. Effect sizes are directly comparable within each isotope, but not between them, given that there are different models for each stable isotope. Values here have been deducted by Friedman's H-statistic to remove the proportion of the effects resulting from interaction between variables. Here demography is generated via uncorrected KDEs. Note: the effect sizes are calculated through iterated simulation employing representative subsets of each key predictor variable, as they do not include every possible combination the percentages will not sum to 100.
Back to article page