| . | All districts—poorest 50% HH . | All districts—all HH . |
|---|---|---|
| Rank correlations | ||
| Partial registry—10% sample (OOS + training) | 0.75 | 0.61 |
| Partial registry—10% sample (OOS) | 0.72 | 0.56 |
| PMT scores | (0.02) | 0.02 |
| IHS training sample | 0.09 | 0.19 |
| RWI | 0.20 | 0.14 |
| AUC | ||
| Partial registry—10% sample (OOS + training) | 0.89 | 0.81 |
| Partial registry—10% sample (OOS) | 0.84 | 0.77 |
| PMT scores | 0.45 | 0.50 |
| IHS training sample | 0.54 | 0.59 |
| RWI | 0.58 | 0.56 |
| R-squared | ||
| Partial registry—10% sample (OOS + training) | 0.57 | 0.35 |
| Partial registry—10% sample (OOS) | 0.52 | 0.28 |
| PMT scores | 0.00 | 0.00 |
| IHS training sample | 0.01 | 0.01 |
| RWI | 0.04 | 0.02 |
| . | All districts—poorest 50% HH . | All districts—all HH . |
|---|---|---|
| Rank correlations | ||
| Partial registry—10% sample (OOS + training) | 0.75 | 0.61 |
| Partial registry—10% sample (OOS) | 0.72 | 0.56 |
| PMT scores | (0.02) | 0.02 |
| IHS training sample | 0.09 | 0.19 |
| RWI | 0.20 | 0.14 |
| AUC | ||
| Partial registry—10% sample (OOS + training) | 0.89 | 0.81 |
| Partial registry—10% sample (OOS) | 0.84 | 0.77 |
| PMT scores | 0.45 | 0.50 |
| IHS training sample | 0.54 | 0.59 |
| RWI | 0.58 | 0.56 |
| R-squared | ||
| Partial registry—10% sample (OOS + training) | 0.57 | 0.35 |
| Partial registry—10% sample (OOS) | 0.52 | 0.28 |
| PMT scores | 0.00 | 0.00 |
| IHS training sample | 0.01 | 0.01 |
| RWI | 0.04 | 0.02 |
Source: Author's calculations using the fifth Integrated Household Survey (IHS), Unified Beneficiary Registry (UBR) 2017, Census (2018) and satellite data described in table 1.
Note: This table reports the results when using a different sample to construct the benchmark welfare measure, which consists of using all the households in all districts and comparing them to the preferred specifications, which uses the poorest 50 percent of households in all districts. For the partial registry method the analysis presents the metrics for out-of-sample predictions (OOS) and also for the preferred results, which include the OOS predictions plus the actual values of the imputed reference village welfare measure used to train the geospatial model.
| . | All districts—poorest 50% HH . | All districts—all HH . |
|---|---|---|
| Rank correlations | ||
| Partial registry—10% sample (OOS + training) | 0.75 | 0.61 |
| Partial registry—10% sample (OOS) | 0.72 | 0.56 |
| PMT scores | (0.02) | 0.02 |
| IHS training sample | 0.09 | 0.19 |
| RWI | 0.20 | 0.14 |
| AUC | ||
| Partial registry—10% sample (OOS + training) | 0.89 | 0.81 |
| Partial registry—10% sample (OOS) | 0.84 | 0.77 |
| PMT scores | 0.45 | 0.50 |
| IHS training sample | 0.54 | 0.59 |
| RWI | 0.58 | 0.56 |
| R-squared | ||
| Partial registry—10% sample (OOS + training) | 0.57 | 0.35 |
| Partial registry—10% sample (OOS) | 0.52 | 0.28 |
| PMT scores | 0.00 | 0.00 |
| IHS training sample | 0.01 | 0.01 |
| RWI | 0.04 | 0.02 |
| . | All districts—poorest 50% HH . | All districts—all HH . |
|---|---|---|
| Rank correlations | ||
| Partial registry—10% sample (OOS + training) | 0.75 | 0.61 |
| Partial registry—10% sample (OOS) | 0.72 | 0.56 |
| PMT scores | (0.02) | 0.02 |
| IHS training sample | 0.09 | 0.19 |
| RWI | 0.20 | 0.14 |
| AUC | ||
| Partial registry—10% sample (OOS + training) | 0.89 | 0.81 |
| Partial registry—10% sample (OOS) | 0.84 | 0.77 |
| PMT scores | 0.45 | 0.50 |
| IHS training sample | 0.54 | 0.59 |
| RWI | 0.58 | 0.56 |
| R-squared | ||
| Partial registry—10% sample (OOS + training) | 0.57 | 0.35 |
| Partial registry—10% sample (OOS) | 0.52 | 0.28 |
| PMT scores | 0.00 | 0.00 |
| IHS training sample | 0.01 | 0.01 |
| RWI | 0.04 | 0.02 |
Source: Author's calculations using the fifth Integrated Household Survey (IHS), Unified Beneficiary Registry (UBR) 2017, Census (2018) and satellite data described in table 1.
Note: This table reports the results when using a different sample to construct the benchmark welfare measure, which consists of using all the households in all districts and comparing them to the preferred specifications, which uses the poorest 50 percent of households in all districts. For the partial registry method the analysis presents the metrics for out-of-sample predictions (OOS) and also for the preferred results, which include the OOS predictions plus the actual values of the imputed reference village welfare measure used to train the geospatial model.
This PDF is available to Subscribers Only
View Article Abstract & Purchase OptionsFor full access to this pdf, sign in to an existing account, or purchase an annual subscription.
