Table 3 Model performance, based on predictions of an out-of-sample subset of 150 site-weeks in NNM across all 13 study sites.

From: Next-generation visitation models using social media to estimate recreation on public lands

Model

Model performance

Mean RMSE

Mean r

R2

1

2.17

0.25

0.06

2

1.57

0.67

0.45

3

1.59

0.67

0.45

4

0.64

0.95

0.91

5

0.65

0.95

0.91

  1. Root mean square error (RMSE) and Pearson’s correlations (r) were calculated for each of 500 iterations, then averaged for an overall measure of predictive performance. The coefficient of determination (R2) was calculated by squaring the mean correlation, and can be interpreted as the proportion of variance in observed visitation predicted by the model.
Back to article page