Evaluating distributional regression strategies for modelling self-reported sexual age-mixing
Figures

Observed means, variances, skewnesses, and kurtoses of partner age by 5-year age bin and sex in all three data sets.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for each probability distribution among women aged 35β39 in the AHRI data set.
Posterior predictive distributions come from fitting each age bin/sex combination independently.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for conventional regression and the most complex distributional model among men aged 16, 24, and 37 years in the AHRI data set.
Posterior predictive distributions come from regression models fit to the entire AHRI data set.

Estimated sinh-arcsinh distributional parameters from the conventional regression model, and distributional models 1 and 4 fit to the AHRI data.
βConventionalβ assumes no variation across age and sex, βDistributional 1β allows for independent age and sex effects, and βDistributional 4β includes sex-specific splines with respect to age.

Estimated sinh-arcsinh distributional parameters for Distributional Model 4 fit to the three main data sets.

Illustration of the effect of the deheaping algorithm on women aged exactly 24 years in the AHRI data.
Dark grey bars correspond to ages identified as potentially heaped (multiples of five away from 24). The red line is the expected count of observations estimated by excluding any potentially heaped ages.

Observed sexual partner age distributions among women in the AHRI data.
The left panel is original data, and the right panel is the same data set after deheaping age differences from multiples of five.

Overlaid quantile-quantile (QQ) plots for each probability distributionβs best fit to data in all three main data sets.
Presented quantiles range from 10th to 90th in increments of 10. Lines closer to the line of equality indicate better fit to empirical quantiles.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for each probability distribution among women in the AHRI data set.
Here, we plot the posterior predicitve distribution associated with each distributionβs highest-ELPD dependent variable.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for each probability distribution among men in the AHRI data set.
Here, we plot the posterior predicitve distribution associated with each distributionβs highest-ELPD dependent variable.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for each probability distribution among women in the AHRI Deheaped data set.
Here, we plot the posterior predicitve distribution associated with each distributionβs highest-ELPD dependent variable.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for each probability distribution among men in the AHRI Deheaped data set.
Here, we plot the posterior predicitve distribution associated with each distributionβs highest-ELPD dependent variable.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for each probability distribution among women in the Haiti 2016β17 DHS data set.
Here, we plot the posterior predicitve distribution associated with each distributionβs highest-ELPD dependent variable.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for each probability distribution among men in the Haiti 2016β17 DHS data set.
Here, we plot the posterior predicitve distribution associated with each distributionβs highest-ELPD dependent variable.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for each probability distribution among women in the Manicaland data set.
Here, we plot the posterior predicitve distribution associated with each distributionβs highest-ELPD dependent variable.

Observed partner age distributions (grey bars) and posterior predictive partner age distributions (lines) for each probability distribution among men in the Manicaland data set.
Here, we plot the posterior predicitve distribution associated with each distributionβs highest-ELPD dependent variable.
Tables
Details of the five distributions tested in this analysis.
We define , to be the standard normal PDF, to be the standard normal cumulative density function, , and .
Distribution | Parameters | Domain | |
---|---|---|---|
Normal | |||
Skew normal | |||
Gamma | |||
Beta | |||
Sinh-arcinh |
Summary of five models fit in this analysis.
Model | Distributional? | Location | Other parameters |
---|---|---|---|
Conventional | No | Age-sex interaction | Constant |
Distributional 1 | Yes | Age-sex interaction | Age and sex effects |
Distributional 2 | Yes | Age-sex interaction | Age-sex interaction |
Distributional 3 | Yes | Sex-specific splines | Age-sex interaction |
Distributional 4 | Yes | Sex-specific splines | Sex-specific splines |
Share of subsets in which each dependent variable yields the highest ELPD given each probability distribution (excluding deheaped AHRI data).
Variable | Normal | Skew normal | Sinh-arcsinh |
---|---|---|---|
Age difference | 22.2% | 25.0% | 16.7% |
Linear age | 8.3% | 5.6% | 16.7% |
Log-age | 19.4% | 41.7% | 30.6% |
Log-ratio | 50.0% | 27.8% | 36.1% |
Model comparison metrics averaged across all data subsets for all three data sets.
Higher ELPD values indicate better fit. Lower QQ RMSE values indicate more accurate prediction of empirical quantiles. Bolded rows are best across all three data sets.
Distribution | AHRI | Haiti 2016β17 DHS | Manicaland |
---|---|---|---|
ELPD | |||
Gamma | β14847.2 | β2917.9 | β13152.8 |
Beta | β14748.0 | β2896.5 | β13003.5 |
Normal | β14593.7 | β2868.4 | β12856.8 |
Skew normal | β14505.1 | β2854.0 | β12778.5 |
Sinh-arcsinh | β14312.5 | β2839.5 | β12625.8 |
QQ RMSE | |||
Gamma | 0.83 | 0.82 | 0.95 |
Beta | 0.99 | 0.82 | 1.11 |
Normal | 0.82 | 0.68 | 0.97 |
Skew normal | 0.77 | 0.65 | 0.85 |
Sinh-arcsinh | 0.36 | 0.37 | 0.44 |
ELPD and QQ RMSE values for all five distributional regression models fit to each data set.
The models increase in complexity from Conventional Regression to Distributional Model 4. Bolded ELPD values are more than two standard errors higher than the next best value in the column. Bolded QQ RMSE values are lowest in their column.
Model | AHRI | Haiti 2016β17 DHS | Manicaland |
---|---|---|---|
ELPD | |||
Conventional | 52689.2 | 4777.8 | 21011.3 |
Distributional 1 | 54335.2 | 5140.8 | 23192.5 |
Distributional 2 | 54794.8 | 5138.7 | 23472.1 |
Distributional 3 | 55534.2 | 5196.7 | 24313.7 |
Distributional 4 | 55841.9 | 5207.6 | 24516.1 |
QQ RMSE | |||
Conventional | 1.30 | 1.33 | 2.05 |
Distributional 1 | 1.15 | 0.98 | 1.89 |
Distributional 2 | 1.21 | 0.99 | 1.80 |
Distributional 3 | 0.93 | 0.91 | 1.34 |
Distributional 4 | 0.66 | 0.84 | 1.04 |
ELPD and QQ RMSE values for all five models fit to deheaped AHRI data The models increase in complexity from Conventional Regression to Distributional Model 4.
Bolded ELPD values are more than two standard errors higher than the next best value in the column. Bolded QQ RMSE values are lowest in their column.
Model | AHRI deheaped |
---|---|
ELPD | |
Conventional | 55296.2 |
Distributional 1 | 57097.4 |
Distributional 2 | 57503.7 |
Distributional 3 | 58219.2 |
Distributional 4 | 58504.0 |
QQ RMSE | |
Conventional | 1.26 |
Distributional 1 | 1.06 |
Distributional 2 | 1.14 |
Distributional 3 | 0.92 |
Distributional 4 | 0.62 |
Full ELPD and QQ RMSE table for women in the AHRI data set.
Higher ELPD values and lower QQ RMSE values are better.
Rank | Model | ELPD | ELPD Diff | SE of Diff | QQ RMSE |
---|---|---|---|---|---|
AHRI Female 20-24 | |||||
1 | Sinh-arcsinh | β31750.94 | 0.00 | 0.00 | 0.32 |
2 | Skew normal | β32056.39 | β305.46 | 48.63 | 0.47 |
3 | Normal | β32414.54 | β663.61 | 60.54 | 0.62 |
4 | Beta | β32953.92 | β1202.98 | 112.08 | 0.77 |
5 | Gamma | β33461.85 | β1710.92 | 148.15 | 0.80 |
AHRI Female 25-29 | |||||
1 | Sinh-arcsinh | β24647.65 | 0.00 | 0.00 | 0.28 |
2 | Skew normal | β24906.22 | β258.57 | 43.27 | 0.52 |
3 | Normal | β25238.71 | β591.06 | 54.82 | 0.68 |
4 | Beta | β25701.13 | β1053.48 | 114.84 | 0.89 |
5 | Gamma | β25995.81 | β1348.16 | 132.15 | 0.90 |
AHRI Female 30-34 | |||||
1 | Sinh-arcsinh | β19831.53 | 0.00 | 0.00 | 0.44 |
2 | Skew normal | β20200.44 | β368.91 | 69.40 | 0.51 |
3 | Normal | β20314.79 | β483.26 | 52.24 | 0.80 |
4 | Beta | β20575.61 | β744.08 | 67.46 | 0.93 |
5 | Gamma | β20708.35 | β876.82 | 73.89 | 0.91 |
AHRI Female 35-39 | |||||
1 | Sinh-arcsinh | β15469.18 | 0.00 | 0.00 | 0.31 |
2 | Skew normal | β15749.79 | β280.61 | 53.04 | 0.77 |
3 | Normal | β15834.32 | β365.14 | 41.23 | 0.80 |
4 | Beta | β16026.51 | β557.33 | 53.99 | 1.18 |
5 | Gamma | β16087.40 | β618.22 | 57.06 | 1.04 |
AHRI Female 40-44 | |||||
1 | Sinh-arcsinh | β12556.61 | 0.00 | 0.00 | 0.45 |
2 | Skew normal | β12876.71 | β320.10 | 45.85 | 1.27 |
3 | Normal | β12935.34 | β378.73 | 52.38 | 0.92 |
4 | Beta | β13137.69 | β581.08 | 69.18 | 1.38 |
5 | Gamma | β13150.66 | β594.05 | 62.73 | 1.19 |
AHRI Female 45-49 | |||||
1 | Sinh-arcsinh | β10059.21 | 0.00 | 0.00 | 0.59 |
2 | Skew normal | β10391.95 | β332.74 | 42.75 | 1.36 |
3 | Normal | β10433.64 | β374.43 | 48.91 | 1.53 |
4 | Gamma | β10527.00 | β467.79 | 50.72 | 1.35 |
5 | Beta | β10545.33 | β486.12 | 56.02 | 1.58 |
Full ELPD and QQ RMSE table for men in the AHRI data set.
Higher ELPD values and lower QQ RMSE values are better.
Rank | Model | ELPD | ELPD Diff | SE of Diff | QQ RMSE |
---|---|---|---|---|---|
AHRI Male 20-24 | |||||
1 | Sinh-arcsinh | β20428.11 | 0.00 | 0.00 | 0.23 |
2 | Skew normal | β20499.86 | β71.75 | 17.12 | 0.25 |
3 | Normal | β20503.89 | β75.79 | 16.85 | 0.22 |
4 | Beta | β20545.59 | β117.49 | 23.21 | 0.22 |
5 | Gamma | β20700.24 | β272.13 | 43.53 | 0.29 |
AHRI Male 25-29 | |||||
1 | Sinh-arcsinh | β12664.21 | 0.00 | 0.00 | 0.26 |
2 | Skew normal | β12727.03 | β62.82 | 17.86 | 0.28 |
3 | Beta | β12739.03 | β74.82 | 18.65 | 0.31 |
4 | Normal | β12753.25 | β89.04 | 19.35 | 0.29 |
5 | Gamma | β12788.26 | β124.05 | 35.07 | 0.38 |
AHRI Male 30-34 | |||||
1 | Sinh-arcsinh | β9301.03 | 0.00 | 0.00 | 0.29 |
2 | Skew normal | β9357.18 | β56.15 | 14.08 | 0.43 |
3 | Beta | β9371.86 | β70.83 | 16.48 | 0.37 |
4 | Normal | β9385.63 | β84.60 | 14.67 | 0.46 |
5 | Gamma | β9419.34 | β118.31 | 35.11 | 0.27 |
AHRI Male 35-39 | |||||
1 | Sinh-arcsinh | β6746.89 | 0.00 | 0.00 | 0.30 |
2 | Skew normal | β6812.77 | β65.88 | 17.73 | 0.64 |
3 | Normal | β6817.86 | β70.97 | 23.24 | 0.70 |
4 | Beta | β6830.95 | β84.06 | 17.95 | 0.71 |
5 | Gamma | β6832.47 | β85.58 | 32.03 | 0.44 |
AHRI Male 40-44 | |||||
1 | Sinh-arcsinh | β4610.95 | 0.00 | 0.00 | 0.35 |
2 | Skew normal | β4711.78 | β100.84 | 18.66 | 0.92 |
3 | Normal | β4713.78 | β102.83 | 18.82 | 0.78 |
4 | Gamma | β4718.28 | β107.33 | 24.83 | 0.63 |
5 | Beta | β4742.70 | β131.75 | 17.28 | 1.07 |
AHRI Male 45-49 | |||||
1 | Sinh-arcsinh | β3683.47 | 0.00 | 0.00 | 0.34 |
2 | Skew normal | β3770.59 | β87.12 | 16.56 | 0.81 |
3 | Gamma | β3776.33 | β92.86 | 15.56 | 0.87 |
4 | Normal | β3778.84 | β95.37 | 14.50 | 1.17 |
5 | Beta | β3805.78 | β122.31 | 17.40 | 1.36 |
Full ELPD and QQ RMSE table for women in the AHRI Deheaped data set.
Higher ELPD values and lower QQ RMSE values are better.
Rank | Model | ELPD | ELPD Diff | SE of Diff | QQ RMSE |
---|---|---|---|---|---|
AHRI Deheaped Female 20-24 | |||||
1 | Sinh-arcsinh | β31411.24 | 0.00 | 0.00 | 0.26 |
2 | Skew normal | β31797.37 | β386.13 | 53.15 | 0.59 |
3 | Normal | β32179.29 | β768.05 | 65.50 | 0.56 |
4 | Beta | β32737.57 | β1326.32 | 118.47 | 0.76 |
5 | Gamma | β33254.17 | β1842.92 | 155.14 | 0.78 |
AHRI Deheaped Female 25-29 | |||||
1 | Sinh-arcsinh | β24439.47 | 0.00 | 0.00 | 0.27 |
2 | Skew normal | β24768.06 | β328.59 | 46.71 | 0.65 |
3 | Normal | β25104.46 | β664.99 | 58.32 | 0.82 |
4 | Beta | β25574.33 | β1134.86 | 119.65 | 1.03 |
5 | Gamma | β25870.30 | β1430.83 | 137.51 | 1.05 |
AHRI Deheaped Female 30-34 | |||||
1 | Sinh-arcsinh | β19680.77 | 0.00 | 0.00 | 0.41 |
2 | Skew normal | β20112.70 | β431.94 | 72.95 | 0.55 |
3 | Normal | β20228.52 | β547.76 | 56.19 | 0.81 |
4 | Beta | β20492.23 | β811.46 | 70.53 | 0.92 |
5 | Gamma | β20624.82 | β944.06 | 76.98 | 0.80 |
AHRI Deheaped Female 35-39 | |||||
1 | Sinh-arcsinh | β15381.68 | 0.00 | 0.00 | 0.26 |
2 | Skew normal | β15703.77 | β322.09 | 55.30 | 0.68 |
3 | Normal | β15788.73 | β407.05 | 43.67 | 0.82 |
4 | Beta | β15983.57 | β601.90 | 56.31 | 1.13 |
5 | Gamma | β16044.22 | β662.54 | 59.17 | 1.04 |
AHRI Deheaped Female 40-44 | |||||
1 | Sinh-arcsinh | β12491.91 | 0.00 | 0.00 | 0.25 |
2 | Skew normal | β12846.63 | β354.72 | 47.38 | 0.99 |
3 | Normal | β12905.04 | β413.12 | 54.14 | 0.89 |
4 | Beta | β13109.82 | β617.91 | 70.96 | 1.31 |
5 | Gamma | β13121.45 | β629.53 | 64.12 | 1.13 |
AHRI Deheaped Female 45-49 | |||||
1 | Sinh-arcsinh | β9981.83 | 0.00 | 0.00 | 0.53 |
2 | Skew normal | β10357.85 | β376.01 | 45.08 | 1.43 |
3 | Normal | β10401.64 | β419.80 | 51.57 | 1.46 |
4 | Gamma | β10493.73 | β511.90 | 52.90 | 1.37 |
5 | Beta | β10513.46 | β531.63 | 58.21 | 1.61 |
Full ELPD and QQ RMSE table for men in the AHRI Deheaped data set.
Higher ELPD values and lower QQ RMSE values are better.
Rank | Model | ELPD | ELPD Diff | SE of Diff | QQ RMSE |
---|---|---|---|---|---|
AHRI Deheaped Male 20-24 | |||||
1 | Sinh-arcsinh | β20310.35 | 0.00 | 0.00 | 0.27 |
2 | Skew normal | β20429.90 | β119.55 | 27.09 | 0.22 |
3 | Normal | β20459.73 | β149.38 | 35.78 | 0.29 |
4 | Beta | β20574.15 | β263.80 | 75.13 | 0.22 |
5 | Gamma | β20899.52 | β589.17 | 175.99 | 0.27 |
AHRI Deheaped Male 25-29 | |||||
1 | Sinh-arcsinh | β12585.54 | 0.00 | 0.00 | 0.28 |
2 | Skew normal | β12680.59 | β95.05 | 21.53 | 0.44 |
3 | Beta | β12697.00 | β111.46 | 23.31 | 0.37 |
4 | Normal | β12701.76 | β116.23 | 22.96 | 0.41 |
5 | Gamma | β12763.81 | β178.27 | 41.24 | 0.39 |
AHRI Deheaped Male 30-34 | |||||
1 | Sinh-arcsinh | β9227.26 | 0.00 | 0.00 | 0.37 |
2 | Skew normal | β9302.42 | β75.16 | 16.15 | 0.41 |
3 | Beta | β9318.24 | β90.97 | 19.07 | 0.39 |
4 | Normal | β9327.58 | β100.31 | 16.18 | 0.41 |
5 | Gamma | β9372.32 | β145.06 | 38.27 | 0.27 |
AHRI Deheaped Male 35-39 | |||||
1 | Sinh-arcsinh | β6694.86 | 0.00 | 0.00 | 0.30 |
2 | Skew normal | β6774.11 | β79.26 | 19.32 | 0.61 |
3 | Normal | β6780.69 | β85.84 | 25.42 | 0.44 |
4 | Beta | β6791.95 | β97.10 | 19.81 | 0.69 |
5 | Gamma | β6796.41 | β101.55 | 34.45 | 0.40 |
AHRI Deheaped Male 40-44 | |||||
1 | Sinh-arcsinh | β4591.04 | 0.00 | 0.00 | 0.49 |
2 | Skew normal | β4700.54 | β109.51 | 19.38 | 1.16 |
3 | Normal | β4703.52 | β112.49 | 19.93 | 1.00 |
4 | Gamma | β4708.43 | β117.40 | 25.94 | 0.89 |
5 | Beta | β4731.41 | β140.37 | 17.84 | 1.30 |
AHRI Deheaped Male 45-49 | |||||
1 | Sinh-arcsinh | β3680.18 | 0.00 | 0.00 | 0.30 |
2 | Normal | β3796.06 | β115.88 | 19.24 | 1.15 |
3 | Skew normal | β3797.14 | β116.95 | 23.48 | 1.02 |
4 | Gamma | β3801.02 | β120.83 | 24.51 | 0.98 |
5 | Beta | β3817.97 | β137.79 | 19.37 | 1.39 |
Full ELPD and QQ RMSE table for women in the Haiti 2016β17 DHS data set.
Higher ELPD values and lower QQ RMSE values are better.
Rank | Model | ELPD | ELPD Diff | SE of Diff | QQ RMSE |
---|---|---|---|---|---|
Haiti 2016-17 DHS Female 20-24 | |||||
1 | Sinh-arcsinh | β3259.31 | 0.00 | 0.00 | 0.49 |
2 | Skew normal | β3263.46 | β4.15 | 4.95 | 0.53 |
3 | Normal | β3338.23 | β78.92 | 19.54 | 0.91 |
4 | Beta | β3441.91 | β182.60 | 45.77 | 1.24 |
5 | Gamma | β3504.85 | β245.54 | 53.90 | 1.29 |
Haiti 2016-17 DHS Female 25-29 | |||||
1 | Sinh-arcsinh | β4447.43 | 0.00 | 0.00 | 0.26 |
2 | Skew normal | β4471.22 | β23.78 | 8.41 | 0.57 |
3 | Normal | β4527.25 | β79.82 | 18.72 | 0.86 |
4 | Beta | β4625.97 | β178.54 | 40.88 | 1.23 |
5 | Gamma | β4678.20 | β230.77 | 45.81 | 1.22 |
Haiti 2016-17 DHS Female 30-34 | |||||
1 | Sinh-arcsinh | β4720.12 | 0.00 | 0.00 | 0.44 |
2 | Skew normal | β4749.57 | β29.45 | 9.06 | 0.68 |
3 | Normal | β4763.78 | β43.66 | 10.51 | 0.62 |
4 | Beta | β4809.11 | β88.99 | 17.19 | 0.85 |
5 | Gamma | β4836.82 | β116.70 | 20.32 | 0.83 |
Haiti 2016-17 DHS Female 35-39 | |||||
1 | Sinh-arcsinh | β4490.82 | 0.00 | 0.00 | 0.33 |
2 | Skew normal | β4518.58 | β27.75 | 8.14 | 0.57 |
3 | Normal | β4526.55 | β35.73 | 8.59 | 0.73 |
4 | Beta | β4561.27 | β70.45 | 13.60 | 0.94 |
5 | Gamma | β4577.84 | β87.01 | 15.27 | 0.86 |
Haiti 2016-17 DHS Female 40-44 | |||||
1 | Sinh-arcsinh | β3601.02 | 0.00 | 0.00 | 0.35 |
2 | Skew normal | β3629.45 | β28.43 | 7.51 | 0.83 |
3 | Normal | β3633.14 | β32.11 | 7.96 | 0.71 |
4 | Beta | β3641.61 | β40.59 | 9.76 | 0.73 |
5 | Gamma | β3644.89 | β43.86 | 10.47 | 0.64 |
Haiti 2016-17 DHS Female 45-49 | |||||
1 | Sinh-arcsinh | β3106.27 | 0.00 | 0.00 | 0.39 |
2 | Skew normal | β3133.10 | β26.82 | 7.68 | 0.88 |
3 | Gamma | β3133.61 | β27.33 | 7.50 | 0.68 |
4 | Normal | β3134.62 | β28.35 | 7.46 | 0.81 |
5 | Beta | β3136.89 | β30.62 | 8.61 | 0.88 |
Full ELPD and QQ RMSE table for men in the Haiti 2016β17 DHS data set.
Higher ELPD values and lower QQ RMSE values are better.
Rank | Model | ELPD | ELPD Diff | SE of Diff | QQ RMSE |
---|---|---|---|---|---|
Haiti 2016-17 DHS Male 20-24 | |||||
1 | Skew normal | β468.98 | 0.00 | 0.00 | 0.43 |
2 | Sinh-arcsinh | β469.60 | β0.62 | 1.12 | 0.41 |
3 | Normal | β475.31 | β6.33 | 4.28 | 0.67 |
4 | Beta | β483.53 | β14.55 | 7.33 | 0.65 |
5 | Gamma | β500.53 | β31.55 | 13.35 | 0.94 |
Haiti 2016-17 DHS Male 25-29 | |||||
1 | Sinh-arcsinh | β1386.13 | 0.00 | 0.00 | 0.38 |
2 | Skew normal | β1390.54 | β4.41 | 3.19 | 0.49 |
3 | Normal | β1395.47 | β9.34 | 4.79 | 0.60 |
4 | Beta | β1407.46 | β21.32 | 7.31 | 0.62 |
5 | Gamma | β1434.18 | β48.04 | 11.75 | 0.79 |
Haiti 2016-17 DHS Male 30-34 | |||||
1 | Sinh-arcsinh | β2217.20 | 0.00 | 0.00 | 0.44 |
2 | Skew normal | β2222.10 | β4.89 | 3.42 | 0.69 |
3 | Normal | β2223.97 | β6.76 | 4.48 | 0.45 |
4 | Beta | β2240.58 | β23.37 | 9.32 | 0.52 |
5 | Gamma | β2281.18 | β63.98 | 17.91 | 0.73 |
Haiti 2016-17 DHS Male 35-39 | |||||
1 | Sinh-arcsinh | β2185.96 | 0.00 | 0.00 | 0.28 |
2 | Skew normal | β2189.87 | β3.91 | 2.67 | 0.69 |
3 | Beta | β2191.05 | β5.10 | 3.68 | 0.48 |
4 | Normal | β2191.11 | β5.16 | 3.55 | 0.49 |
5 | Gamma | β2205.69 | β19.73 | 9.57 | 0.52 |
Haiti 2016-17 DHS Male 40-44 | |||||
1 | Sinh-arcsinh | β2051.62 | 0.00 | 0.00 | 0.39 |
2 | Skew normal | β2060.16 | β8.54 | 4.21 | 0.72 |
3 | Normal | β2060.38 | β8.75 | 4.57 | 0.69 |
4 | Beta | β2062.00 | β10.37 | 4.87 | 0.70 |
5 | Gamma | β2063.79 | β12.17 | 5.73 | 0.47 |
Haiti 2016-17 DHS Male 45-49 | |||||
1 | Sinh-arcsinh | β2138.34 | 0.00 | 0.00 | 0.23 |
2 | Normal | β2150.53 | β12.19 | 6.38 | 0.35 |
3 | Skew normal | β2151.51 | β13.17 | 5.97 | 0.56 |
4 | Gamma | β2152.88 | β14.54 | 8.93 | 0.25 |
5 | Beta | β2156.13 | β17.79 | 6.14 | 0.48 |
Full ELPD and QQ RMSE table for women in the Manicaland data set.
Higher ELPD values and lower QQ RMSE values are better.
Rank | Model | ELPD | ELPD Diff | SE of Diff | QQ RMSE |
---|---|---|---|---|---|
Manicaland Female 20-24 | |||||
1 | Sinh-arcsinh | β16390.77 | 0.00 | 0.00 | 0.31 |
2 | Skew normal | β16502.01 | β111.25 | 21.22 | 0.44 |
3 | Normal | β16779.93 | β389.16 | 37.05 | 0.67 |
4 | Beta | β17111.57 | β720.80 | 62.02 | 0.86 |
5 | Gamma | β17387.38 | β996.61 | 76.80 | 1.02 |
Manicaland Female 25-29 | |||||
1 | Sinh-arcsinh | β18702.50 | 0.00 | 0.00 | 0.53 |
2 | Skew normal | β18923.04 | β220.53 | 25.27 | 0.94 |
3 | Normal | β19080.66 | β378.16 | 36.05 | 0.83 |
4 | Beta | β19405.80 | β703.30 | 64.97 | 1.05 |
5 | Gamma | β19615.53 | β913.03 | 76.38 | 1.09 |
Manicaland Female 30-34 | |||||
1 | Sinh-arcsinh | β16523.81 | 0.00 | 0.00 | 0.48 |
2 | Skew normal | β16877.96 | β354.15 | 40.36 | 0.87 |
3 | Normal | β16886.62 | β362.80 | 36.41 | 0.99 |
4 | Beta | β17021.26 | β497.44 | 43.60 | 1.12 |
5 | Gamma | β17094.58 | β570.76 | 49.53 | 0.93 |
Manicaland Female 35-39 | |||||
1 | Sinh-arcsinh | β14397.76 | 0.00 | 0.00 | 0.48 |
2 | Skew normal | β14736.64 | β338.88 | 28.35 | 1.25 |
3 | Normal | β14798.55 | β400.79 | 36.87 | 1.39 |
4 | Beta | β14824.80 | β427.04 | 33.02 | 1.47 |
5 | Gamma | β14835.11 | β437.35 | 34.49 | 1.14 |
Manicaland Female 40-44 | |||||
1 | Sinh-arcsinh | β12293.13 | 0.00 | 0.00 | 0.68 |
2 | Skew normal | β12488.28 | β195.15 | 21.36 | 1.03 |
3 | Gamma | β12500.93 | β207.80 | 22.18 | 1.03 |
4 | Normal | β12508.91 | β215.78 | 23.29 | 1.28 |
5 | Beta | β12537.14 | β244.01 | 25.41 | 1.22 |
Manicaland Female 45-49 | |||||
1 | Sinh-arcsinh | β9183.03 | 0.00 | 0.00 | 0.56 |
2 | Skew normal | β9455.87 | β272.83 | 23.57 | 1.68 |
3 | Normal | β9477.33 | β294.30 | 23.55 | 1.62 |
4 | Gamma | β9497.31 | β314.27 | 25.08 | 1.44 |
5 | Beta | β9576.44 | β393.40 | 32.07 | 1.94 |
Full ELPD and QQ RMSE table for men in the Manicaland data set.
Higher ELPD values and lower QQ RMSE values are better.
Rank | Model | ELPD | ELPD Diff | SE of Diff | QQ RMSE |
---|---|---|---|---|---|
Manicaland Male 20-24 | |||||
1 | Sinh-arcsinh | β9770.00 | 0.00 | 0.00 | 0.30 |
2 | Skew normal | β9895.82 | β125.83 | 33.35 | 0.40 |
3 | Normal | β10139.11 | β369.11 | 79.13 | 0.49 |
4 | Beta | β10587.64 | β817.64 | 181.23 | 0.56 |
5 | Gamma | β11594.58 | β1824.59 | 388.26 | 1.15 |
Manicaland Male 25-29 | |||||
1 | Sinh-arcsinh | β13978.59 | 0.00 | 0.00 | 0.40 |
2 | Skew normal | β13990.39 | β11.80 | 8.51 | 0.48 |
3 | Normal | β14018.60 | β40.00 | 17.48 | 0.45 |
4 | Beta | β14152.35 | β173.76 | 48.77 | 0.40 |
5 | Gamma | β14500.47 | β521.87 | 117.58 | 0.55 |
Manicaland Male 30-34 | |||||
1 | Sinh-arcsinh | β12949.24 | 0.00 | 0.00 | 0.31 |
2 | Skew normal | β13016.44 | β67.21 | 25.01 | 0.37 |
3 | Normal | β13037.46 | β88.22 | 16.57 | 0.49 |
4 | Beta | β13070.31 | β121.07 | 54.74 | 0.41 |
5 | Gamma | β13285.47 | β336.23 | 171.92 | 0.42 |
Manicaland Male 35-39 | |||||
1 | Sinh-arcsinh | β11496.14 | 0.00 | 0.00 | 0.27 |
2 | Skew normal | β11528.36 | β32.22 | 9.83 | 0.39 |
3 | Normal | β11530.43 | β34.29 | 9.72 | 0.26 |
4 | Gamma | β11531.75 | β35.61 | 12.47 | 0.24 |
5 | Beta | β11582.63 | β86.49 | 12.97 | 0.48 |
Manicaland Male 40-44 | |||||
1 | Sinh-arcsinh | β8714.06 | 0.00 | 0.00 | 0.35 |
2 | Skew normal | β8749.78 | β35.72 | 10.11 | 0.51 |
3 | Gamma | β8777.08 | β63.02 | 10.38 | 0.55 |
4 | Normal | β8791.45 | β77.38 | 12.23 | 0.76 |
5 | Beta | β8860.22 | β146.16 | 18.15 | 0.93 |
Manicaland Male 45-49 | |||||
1 | Sinh-arcsinh | β7110.27 | 0.00 | 0.00 | 0.42 |
2 | Skew normal | β7177.03 | β66.75 | 25.08 | 0.76 |
3 | Gamma | β7213.99 | β103.72 | 13.02 | 1.07 |
4 | Normal | β7232.04 | β121.77 | 13.09 | 1.28 |
5 | Beta | β7312.35 | β202.08 | 18.61 | 1.61 |
LOO-CV estimated ELPD values, differences, and standard errors of differences, as well as QQ RMSE values, for all five regression models fit to all four data sets.
The βdifferenceβ value of a row is the difference between that rowβs ELPD value and data set-specific best ELPD value. Higher ELPD values and lower QQ RMSE values are better.
Rank | Model | ELPD | ELPD Diff | SE of Diff | QQ RMSE |
---|---|---|---|---|---|
AHRI | |||||
1 | Distributional 4 | 55841.91 | 0.00 | 0.00 | 0.66 |
2 | Distributional 3 | 55534.16 | β307.75 | 32.36 | 0.93 |
3 | Distributional 2 | 54794.79 | β1047.12 | 51.69 | 1.21 |
4 | Distributional 1 | 54335.19 | β1506.72 | 72.32 | 1.15 |
5 | Conventional | 52689.21 | β3152.70 | 100.59 | 1.30 |
AHRI Deaheaped | |||||
1 | Distributional 4 | 58503.98 | 0.00 | 0.00 | 0.62 |
2 | Distributional 3 | 58219.23 | β284.75 | 28.64 | 0.92 |
3 | Distributional 2 | 57503.68 | β1000.30 | 47.14 | 1.14 |
4 | Distributional 1 | 57097.39 | β1406.59 | 64.48 | 1.06 |
5 | Conventional | 55296.25 | β3207.73 | 99.42 | 1.26 |
Haiti 2016-17 DHS | |||||
1 | Distributional 4 | 5207.57 | 0.00 | 0.00 | 0.84 |
2 | Distributional 3 | 5196.69 | β10.89 | 6.54 | 0.91 |
3 | Distributional 1 | 5140.77 | β66.80 | 12.27 | 0.98 |
4 | Distributional 2 | 5138.75 | β68.83 | 12.24 | 0.99 |
5 | Conventional | 4777.78 | β429.80 | 30.54 | 1.33 |
Manicaland | |||||
1 | Distributional 4 | 24516.15 | 0.00 | 0.00 | 1.04 |
2 | Distributional 3 | 24313.74 | β202.40 | 20.52 | 1.34 |
3 | Distributional 2 | 23472.07 | β1044.08 | 47.77 | 1.80 |
4 | Distributional 1 | 23192.49 | β1323.66 | 54.97 | 1.89 |
5 | Conventional | 21011.29 | β3504.86 | 89.01 | 2.05 |