KEN 2016-20

[QLLZ]

Data pull: HASH: 10db6c83d6486774d3148785df24b3145f375bb2 config

[QLLZ]

reran the initial values data after the new updates on initial values HASH: 6772ea9da4f43ddc3da11c494a573af2ca3dd3c8

[QLLZ]

Model run: HASH: 6772ea9da4f43ddc3da11c494a573af2ca3dd3c8

[QLLZ]

Prestan report

Country data report

[eclee25]

Convergence (except sd_w), fits, and match to GAM input look good. Some Rhat issues at admin 1 & 2 levels.

Opinion: Approve

[QLLZ]

Config

Data pull(32G):

HASH: 211913e0eb7d570ed62c48bc1ae9ecd6671a6cbc

[QLLZ]

Model run:

HASH: 211913e0eb7d570ed62c48bc1ae9ecd6671a6cbc

[QLLZ]

Failed model run log file

[QLLZ]

Rerun on dev_u_combs_fix

HASH: e56580fa5ddab00293e31ff90139351f80ae3f6c

[QLLZ]

country data report

[eclee25]

rho and sd_w traceplots are not converged and they are having a mild effect on the w traceplots as well. The ID-45 model that was run on dev_dagar_fix seems to perform a bit better than this model and I think the only difference between the two is that the standard model run has drop_censored_adm0: yes and drop_censored_adm0_thresh: 2. However the national annual estimates seem to be quite between these two model runs.

Temp Approve - could consider changing this back to the ID-45 settings or changing to no-mixture sd_w prior to see if it helps settle the estimates into the lower range. Could also consider increasing the censoring threshold to reduce the number of lower tfrac full annual observations

[eclee25]

ID-61 showed that the no-mixture sd_w prior improved convergence in this model. Use the no-mixture sd_w prior with all other standard settings (even though this model has subnational data)

[QLLZ]

Data pull, model run:

HASH: 05f598b95b5d42ac0049963f8a0b7e04d6f1dbd7

[javierps]

Autocor in rho, std_w and ws OK. All Rhats OK.

Suggestion: Accept.

[eclee25]

Many conflicting annual obs - ok. Agree with J. Approve