search

HopkinsIDD / cholera-mapping-pipeline

Formerly part of cholera-taxonomy. The map creation scripts, packages, and file structure
1 stars 3 forks source link

SOM 2011-15 #441

Open eclee25 opened 1 year ago

QLLZ commented 1 year ago

Data pull: HASH: 6772ea9da4f43ddc3da11c494a573af2ca3dd3c8 config

QLLZ commented 1 year ago

Model run: HASH: 6772ea9da4f43ddc3da11c494a573af2ca3dd3c8

javierps commented 1 year ago

All checks look ok. Geometrical pattern in w although very similar values. Incidence rates are very high and homogenous, but consistent with gam input.

Suggestion: accept.

eclee25 commented 1 year ago

Most diagnostics ok with ~10% of observations and 11% and 8% of admin 1 and admin 2 genquant outputs with high Rhats. This may be because the subnational data are quite limited to one area of the country.

Opinion: Approve

eclee25 commented 1 year ago

On further investigation, the rates seem very high -- possibly because half the population of Mogadishu is missing so the overall population in the country is a bit low. This sfrac issue has been a problem in the past but we need to investigate further why sfrac is not handling this well...

QLLZ commented 1 year ago

Config

Data pull & Model run:

HASH: 211913e0eb7d570ed62c48bc1ae9ecd6671a6cbc

QLLZ commented 1 year ago

country data report

QLLZ commented 1 year ago

Super high rates, maybe due to underestimated population?

Slight convergency issue. Modeled cases look fine (there are some discrepancies in the observations, thus the modeled cases do not look very close to any of them for some years).

eclee25 commented 1 year ago

Mild convergence issue with sd_w but all other model diags look good. Population is >25% underestimated in every year which may contribute partially to the super high rates. Approve with possible pop postprocessing adjustment

javierps commented 12 months ago

Sep 2023 Production run: convergence issues with std_w and ws, 10% of Rhats above threshold.

Suggestion: no-mixture run.

eclee25 commented 11 months ago

Table 6 OC 5032 had 10,600 cases in 2013 in the standard setting model run and only 7168 cases in the new no-mixture sd_w model run… There don’t appear to be any changes in data processing or the database so it’s a little odd.

All other diags look good. After reviewing the cause of these data discrepancies, assuming no problem with the data, Approve

eclee25 commented 10 months ago

@QLLZ determined that the discrepancy is related to differences in selecting which observation gets aggregated in a given set when there is an overlapping TL, TR, and location in the same OC. @javierps will add a branch that first sorts observations descending by sCh and then aggregates. We will not rerun things at the moment.

Approve no-mixture