Table 1 Goodness-of-fit metrics.

From: High-resolution population estimation using household survey data and building footprints

Estimate

Prediction

Bias

Imprecision

Inaccuracy

R2

95% CI

Population totals

In-sample

13.81 (–0.02)

165.50 (0.41)

100.17 (0.27)

0.81

92.49%

Population totals

Out-of-sample

13.43 (–0.03)

173.28 (0.44)

105.61 (0.29)

0.79

90.50%

Population densities

In-sample

–13.96 (–0.02)

441.66 (0.41)

266.32 (0.27)

0.52

92.04%

Population densities

Out-of-sample

–15.20 (–0.03)

464.69 (0.44)

282.90 (0.29)

0.47

90.06%

Age and sex proportions

In-sample

0.00 (0.00)

0.00 (0.00)

0.00 (0.00)

1.00

100.00%

Age and sex proportions

Out-of-sample

0.00 (0.00)

0.00 (0.00)

0.00 (0.00)

1.00

100.00%

  1. Analysis of residuals for the estimated population totals (people), population densities (people/building footprint ha), and province-level age and sex proportions for in-sample and out-of-sample posterior predictions. Bias represents the mean of the residuals, imprecision the standard deviation of residuals, inaccuracy the mean of absolute residuals, R2 the squared Pearson correlation coefficient among the residuals, and the percentage of observations falling within the 95% credible intervals. Values in parentheses are computed using scaled residuals (residuals/predictions). Out-of-sample predictions are carried out using 10-fold cross-validation, where the model is fit ten times, each time withholding a random 10% of microcensus clusters until all is held out once.
Back to article page