nutriverse / ipctools Goto Github PK

• #14
• #15

@tomaszaba, any suggestion for what the hex sticker should be for this package?

@tomaszaba I noticed that the way IPC calculates prevalence based on what you have been showing in your approach is that no survey weights are being applied to account for the cluster design of the survey/s that produced the data. For example, in your South Sudan example, you basically just calculate prevalence by doing a simple number of SAM divided by total number of kids in sample.

Does IPC just automatically assume that the PPS done at the start was followed strictly as per design? From my experience, very rarely do survey data come up as designed and samples per cluster is never always what is expected. For S3M and RAM type surveys that Mark and I have developed, this doesn't matter too much as we always do weighting posteriorly (we don't do PPS). But, for a PPS design in which the weights are applied right from the sampling itself, having a sample that is different from your intended design has implications in the balance of your dataset such that a simple number of SAM cases divided by all 6-59 in sample will give biased results.

Do you guys don't do any weighting at all on the data to calculate prevalence?

I am still not fully understanding this but is this approach being used to avoid having to weight the data, that is, using muac-for-age z-score, cut-offs to determine sam and mam will be used and then use this to calculate prevalence? is this the next step?

@tomaszaba, should we consider datasets that have MUAC but no oedema data?

If so, then we probably need to re-factor the functions to accept no oedema data and still perform the checks and the calculations.

see IPC branding guidelines - https://www.ipcinfo.org/fileadmin/user_upload/ipcinfo/docs/communication_tools/IPC_BrandBook_2020_Draft.pdf

based on conversations with @tomaszaba, IPC uses the following criteria to classify p-value results of sex ratio and age ratio tests:

Functions that perform these classifications would be helpful. Note that for digit preference, such a classification already exists and is produced as part of the output of the digitPreference() function.

It might be logical to use the same approach for sex ratio and age ratio.

see https://www.ipcinfo.org/fileadmin/user_upload/ipcinfo/docs/IPC_Acute_Food_Insecurity_Mapping_Guidelines.pdf

https://www.ipcinfo.org/ipcinfo-website/communication-tools/icons-guidelines/

https://www.ipcinfo.org/fileadmin/user_upload/ipcinfo/docs/communication_tools/IPC_BrandBook_2020_Draft.pdf

• #6
• #7
• #8

to ensure that IPC folks won't feel insecure that this is encroaching on the way they analyse things. The functions here only help in processing data leading to whatever form of analysis and consensus-building IPC does to arrive at a final output

@tomaszaba as discussed in your repository, I will add a function in this package that you can use for IPC-specific purposes.

• #31
• #32