Base comparison analysis write up examples

From Wiki
Revision as of 16:12, 24 August 2020 by Ifsdatateam (Talk | contribs)

(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to: navigation, search

An example of final analysis by Mohammod:

  1. One of the things I have done to add to Barry’s analysis is to do some analyses between runs with “old” data and with “new” data for all series but one (for example, IFCF, or TFR) to figure out how much of the difference is from the particular series. I have done it first with IFCF (i.e., take all new data, but keep old IFCF, SeriesInvestGrCapForm%GDP), then with TFR (TFRMedUNPD) and then with both IFCF and TFR. (Barry also mentioned household consumption. That series seems to be same in my “old” 7.53 database and the one Yutang sent.)

For Timor, it looks like most of the GDP changes between runs with new and old data goes away when IGCF and TFR data do not change (and. Vice versa) . For Djibouti, most of the GDP changes are indeed from IGCF. (plots given below – second, and third from below)

This is what I have so far before we meet in 20 minutes. My conclusion is that we can take these data if the SeriesInvestGrCapForm%GDP data is sound.

An example of final analysis by Barry:

  1. I have now studied somewhat the runs using the old/7.54 IP1 and new databases.  Yes, the most surprising results are in big GDP changes in a few countries including Djibouti and Timor-Leste (in opposite directions).  I also drilled down somewhat into Angola.  My analysis suggests that In all three cases the explanation for the differences lies primarily in very different initial conditions for IGCF (gross capital formation).  The values in SeriesInvestGrCapForm%GDP have changed quite dramatically across the two data sets.  Obviously IGCF/GDP is a really important driver and should probably be added to the list of things we routinely check with very large data changes.  Also C/GDP coming from SeriesHouseCon%GDP, which interacts with IGCF in the preprocessor.

For instance, for Timor-Leste, the old data set led to a value of 21.9% for 2015 and the new generates 35.1% in the preprocessor.  The new dataset not only adds 2018 values, but changes those for 2015.  For Djibouti the value moved to 18% from 46%.  For Angola it moved from 34.2% to 25.3%.   There are really big changes and I assume also for some other countries.

This page was updated on 08/24/2020.