Multilateral financing for health?

[jm-rivera]

Can we have some visualisation that shows financing trends for MDBs, using OECD DAC data.

[jm-rivera]

@nupur-parikh

How is this? https://public.flourish.studio/visualisation/12959062/

We could also add the amounts as a share of total ODA from multilaterals. But before going further, I wanted to get your thoughts.

The data can also be cut by ODA grants, ODA Loans, other official flows (loans).

[nupur-parikh]

Thanks @jm-rivera I don't think absolutely necessary to show financing to different health sectors, instead it would be great to see how each MDB finances health in total if possible with the same cuts you have in the chart you shared above?

Is it possible to show how this breaks down by income group and region (Africa most importantly)? And maybe we could show using a sankey diagram so we're not using too many line charts? I think the MDBs of most interest would be these so it's consistent with the MDB data dive:

Honestly don't know if this is possible, please tell me if not as I'm not as familiar with this data and with the MDB campaign. Also please flag if you think this kind of analysis falls outside of the scope of the page as well, always helpful to hear your thoughts!

[jm-rivera]

Thanks @jm-rivera I don't think absolutely necessary to show financing to different health sectors, instead it would be great to see how each MDB finances health in total if possible with the same cuts you have in the chart you shared above?

which cuts do you mean @nupur-parikh ?

[nupur-parikh]

which cuts do you mean @nupur-parikh ?

Total (ODA + OOF), ODA Grants, ODA Loans, OOF, and Equity Investments

[jm-rivera]

New version of chart4_1, the Sankey.

Per our discussion: regions, with year filter. Financing for health from all the MDBs covered by our Data Dive.

https://public.flourish.studio/visualisation/13044897/

thoughts @nupur-parikh @TheebikaS

Also please feel free to comment on the regional breakdown. We could group things in another way if you want less categories, for example.

[nupur-parikh]

Thanks Jorge! This is really great! Two things:

1. I am still having year issues, the image you shared says 2021 and I'm seeing 2020. I'm assuming flourish hasn't fixed the issue yet, but that's fine! I really like the use of the timeline at the top, really interesting to see how the flows change overtime
2. I suggest grouping by continent instead

[jm-rivera]

Thanks @nupur-parikh!

1. indeed still an issue on their end but I've applied their workaround so you should be able to see it now.
2. here it is with a continent breakdown. Asia is really quite a broad mix so I've separated middle east an Asia. But if you'd prefer just 'Asia' i can group it. let me know

[nupur-parikh]

Thanks @jm-rivera that work around worked!

Noted on Asia and the Middle East I like this, can we do the same for America and show North and South America or is that not possible? I'm looking at the ODA topic page and it looks like we did not break these out so happy to keep as is for consistency

[jm-rivera]

would you include Central America & the Caribbean in North America? For clarity we may want to say 'North and Central America' and 'South America'. What do you think @nupur-parikh ?

[TheebikaS]

Looks good to me! Regional breakdown as suggested by you and Nupur sounds good to me, i.e. North & Central and South America.

[nupur-parikh]

That is perfect @jm-rivera!

[jm-rivera]

Thanks both. voila!

[nupur-parikh]

Amazing @jm-rivera thank you! One last thing, this is in USD constant 2021 prices, correct? any way we can add this to the chart or does that go into the page build process?

[jm-rivera]

aha! This is not in constant prices. But it should probably be. There's always a question of what the right price deflators are for multilaterals, but i'll keep it consistent with what we do for the ODA data.

Any notes, including on prices, would go directly on the chart object on the page (which includes the data download, the share button, the sources, etc)

[nupur-parikh]

thanks for the clarification!

[jm-rivera]

I've updated this: https://public.flourish.studio/visualisation/13044897/ to constant prices. The question is to which base year. The rest of the data will be in 2020 prices for now because we will show up to 2020 in most cases, right? But this source goes up to 2021, so usually we would want to make that the base year.... except for the fact that then it's not directly comparable to the other data on the page.

Based on that I've based it to 2020 prices. But let me know if you disagree @nupur-parikh.

[nupur-parikh]

Thank you for clearly explaining the reasoning behind this, I agree with the approach you've taken. If we note in the chart text or sources that 2020 prices are used for consistency and comparability, then should be fine.

[jm-rivera]

Thanks @nupur-parikh.

Here's the second chart (fyi @TheebikaS)

Is this what you had in mind? https://public.flourish.studio/visualisation/13044211/

Another option could be a stacked area graph:

https://public.flourish.studio/visualisation/13060855/

A line chart version of this remains an option of course:

https://public.flourish.studio/visualisation/13060949/

Thoughts?

[nupur-parikh]

@jm-rivera these are really great! So fascinating to see in all three versions just how little financing NCDs gets across each region, wow!

I think I prefer the area or line chart over the streamgraph, leaning more in favor of the area chart. But @TheebikaS let me know if you think otherwise. Can we add in the popup the share of MDB total financing each of the sectors makes up so it's clear in case that information is helpful for those visiting the page?

[jm-rivera]

NCDs get very little because it's a fairly new DAC code and I think it is still 'voluntary'. So that will be something to note on the description of the page.

I would need to sort the data differently so that NDCs are actually at the top so that it goes from most to least.

We can do quite a bit with the tooltips for this one, so absolutely can show shares or amounts or whatever else you find helpful.

[nupur-parikh]

Oh interesting, I had no idea that was the case, but that makes sense!

I think shares, amounts in USD, year is most useful, thank you!

[jm-rivera]

Updated with the tooltip. https://public.flourish.studio/visualisation/13060949/

Pending @TheebikaS feedback on the visual, anything else for this section?

[jm-rivera]

Just to add a thought there. I may have a slight preference for something like a steamgraph given the disparities in amounts. Given that Basic health and/or Health, General are orders of magnitude bigger, the stacked area/line makes it hard to see the smaller amounts, compared to the steamgraph.

What we lose with the steamgraph is a clear y-axis, but then it really depends on what use/point you want to make. So we should go with your preference in this case.

[nupur-parikh]

Just to add a thought there. I may have a slight preference for something like a steamgraph given the disparities in amounts. Given that Basic health and/or Health, General are orders of magnitude bigger, the stacked area/line makes it hard to see the smaller amounts, compared to the steamgraph.

What we lose with the steamgraph is a clear y-axis, but then it really depends on what use/point you want to make. So we should go with your preference in this case.

That makes sense and I agree with the area/line charts it's harder to see things like NCDs, do we have the same tooltip functionality for the steam graph as we do for the area/line? If so and pending @TheebikaS's thoughts, we could play around with that a little more to make it more clear. If not, then I prefer area/line so we can use the tooltip function.

Other than this though, nothing else on this section from me. Thank you!!!

[jm-rivera]

The tree versions (steamgraph, stacked area (%) and stacked area/line) are actually based on the same template. So we can do the same sort of thing with all three with tooltips or any other bit of interest.

[nupur-parikh]

The tree versions (steamgraph, stacked area (%) and stacked area/line) are actually based on the same template. So we can do the same sort of thing with all three with tooltips or any other bit of interest.

Okay got it, thanks. Can we add year back to the tooltips so it's clear?

Also is there a way to define basic health and health general or break these down into individual subsectors? Was showing this to Jenny and we realized it may be difficult to understand what the difference is between those two especially since they make up such large shares. I realize there are a lot of subsectors for basic health especially so it may not make sense in which case, @TheebikaS would love your thoughts on if the visualization as is will be useful for what you'd want to show.

[jm-rivera]

This is quite an important point @nupur-parikh and it makes me think we should carefully consider whether showing this or a more detailed breakdown is a good idea.

What we're showing right now are 'sectors'. The sub-sectors are called 'purpose codes' in the CRS. So an activity falls within a sector, and then has a purpose.

Purpose codes are quite detailed in some cases. Some of them are voluntary, some of them are mandatory. However, detailed reporting of purpose codes is voluntary for multilaterals, and not all of them do it to the same degree.

In theory, they could report all of this.

However, seeing the data, some MDBs (the larger) provide data for quite a few purpose codes. Others will have basically 1 purpose code for each sector code.

So we may be inadvertently showing reporting artefacts overtime rather than actual changes in priorties from these multilaterals.

Another thing to consider is the timeframe. In 2006, the start of our chart, there was data for 8 MDBs, in 2021 we had data for 18.

Again another tricky thing here is trying to show total amounts for such a broad group. The increases in spending are likely, to a significant degree, improvements in reporting coverage.

This issue affects both charts, of course. So we should consider drastically reducing the number of years we're covering, and including only MDBs for which we have sufficient data.

[nupur-parikh]

Thanks @jm-rivera for explaining this so well. I'm also not sure if showing the sector and purpose codes is a good idea given the level of detail there is/isn't for many of the MDBs. I'll let @TheebikaS weigh in more on this and if showing sectors and purpose codes is helpful at this stage. Maybe an alternative is showing just the WB Group or IDA/IBRD similar to the MDB data dive.

As for the time frame and the improvements in reporting coverage what you say makes sense, maybe we just look at the last five years so 2016-2021, should historical data be needed users have the option to go directly to the source.

[TheebikaS]

Thanks @jm-rivera and @nupur-parikh. Sorry for holding this up with a late response.

A few questions:

• can we group purpose codes together as we see fit and show chart for that? Is that too much work? Any other issues that I'm forgetting?
• can we 'get around' the reporting bias, by clearly stating in any analysis that these are 'reported' findings. Or is it best to stay clear of analyzing trend data due to this reason (even for a shorter timeframe like 2016-2021)? For example, the chart that you shared above, makes me wonder about the link between outbreaks happening and 'surge'/'emergency' funding provided by MDBs - can we suggest this link or can we not due to reporting bias (also aware that this would be contextual assumption)?

If possible, would be good to have a chart for 2016/7-2021, doing our own grouping of purpose codes (if feasible) for all MDBs, or if not, as Nupur suggests for the WB. If not, given challenges, best to leave out this chart.

On a side note, I've been trailing through the World Bank's own project database, which uses its own purpose codes (under major themes) that seem to be better grouped together, but they do not match the ones from this OECD data : https://projects.worldbank.org/en/projects-operations/project-search

[nupur-parikh]

Thanks @jm-rivera and @nupur-parikh. Sorry for holding this up with a late response.

A few questions:

• can we group purpose codes together as we see fit and show chart for that? Is that too much work? Any other issues that I'm forgetting?
• can we 'get around' the reporting bias, by clearly stating in any analysis that these are 'reported' findings. Or is it best to stay clear of analyzing trend data due to this reason (even for a shorter timeframe like 2016-2021)? For example, the chart that you shared above, makes me wonder about the link between outbreaks happening and 'surge'/'emergency' funding provided by MDBs - can we suggest this link or can we not due to reporting bias (also aware that this would be contextual assumption)?

If possible, would be good to have a chart for 2016/7-2021, doing our own grouping of purpose codes (if feasible) for all MDBs, or if not, as Nupur suggests for the WB. If not, given challenges, best to leave out this chart.

Thanks Theebika, I'm leaning towards not including purpose codes/sector by sector analysis for now on the health financing page. We can however think about adding this bit of analysis in a future update once we're clear on what exactly we want to see and put together a clear methodology on grouping purpose codes. This will give us time to learn more about the data and determine what makes the most sense for the purpose of MDB financing conversations.

With that being said, I think just one chart (the sankey diagram) on MDB financing from 2016-2021 is necessary. @jm-rivera I realize this time frame also poses issues as it relates to increasing number of MDBs, I'm not 100% familiar with the data so correct me if I'm wrong here but are the 4 MDBs that were added between 2016 and 2021 the ones with the most incomplete data? If so, I say we do not include them as you suggested due to insufficient data.

[jm-rivera]

Thank you both! @TheebikaS,

can we group purpose codes together as we see fit and show chart for that? Is that too much work? Any other issues that I'm forgetting? We can group things based on sector or purpose codes. Our dashboard at data.one.org/aid-dashboard does the sort of grouping from the charts we were looking at in this conversation. But that is only possible in a methodologically-sound way for the DAC because they must report at purpose-code level.

For multilaterals, it isn't so much a question of how we group, but at the level at which they report. If we go more detailed than the least detailed reporting that means that we are presenting an incomplete/potentially inaccurate picture of what the data is saying. Since some of the data is grouped in a less detailed/more aggregated way, we cannot say whether or not part of an amount contributed to an activity that falls under a more detailed purpose code.

In other words, going more detailed than the 4-sector breakdown would not be sound. And as outlined previously, even that seems tricky because of the varied reporting practises amount the organisations we're looking at (keeping in mind that the details are not mandatory for them).

So I agree with this comment from @nupur-parikh

Thanks Theebika, I'm leaning towards not including purpose codes/sector by sector analysis for now on the health financing page.

But would add that it may not be feasible analysis if we want to capture 'MDBs' as a whole, instead of focusing on the MDBs which have really good data.

On

can we 'get around' the reporting bias, by clearly stating in any analysis that these are 'reported' findings. Or is it best to stay clear of analyzing trend data due to this reason (even for a shorter timeframe like 2016-2021)? For example, the chart that you shared above, makes me wonder about the link between outbreaks happening and 'surge'/'emergency' funding provided by MDBs - can we suggest this link or can we not due to reporting bias (also aware that this would be contextual assumption)?

Everything we're showing is as-reported. We can't get around that since that is the nature of OECD DAC statistics. But we are choosing a presentation of the data which is highlighting changes over time. It is therefore key that we are not highlighting changes that are not really present in the underlying data, but that are a product of our methodology (choosing to include a variable number of MDBs each year) and data coverage. In a chart which would show increasing MDB finance to health, we must be able to ensure that the 'increase' is not just caused by data from more organisations, but that it reflects the reality of the financing priorities/choices of the bulk of MDBs.

In my view the best alternative for this point would be to focus on a timeline and selection of MDBs that allows us to have a consistent selection of MDBs included in the period we're analysing. What this approach wouldn't let us do would be to claim that we're showing 'total MDB financing' to health. But we could very well do the financing of the biggest 5-10 (or whatever the magical number that meets the conditions above is).

And even then, if our ultimate goal is to analyse year-to-year changes, or changes in response to specific events, we should consider doing a moving average of a 2 or 3 year period. MDB financing can be spiky because of how their financing model works, and in some cases because of how it is reported. So smoothing it out may present the data in way in which reporting 'noise' is minimised and the trends over time are more clearly visible.

On a side note, I've been trailing through the World Bank's own project database, which uses its own purpose codes (under major themes) that seem to be better grouped together, but they do not match the ones from this OECD data : https://projects.worldbank.org/en/projects-operations/project-search

We have been scraping their database for the last few years as well. They provide more detail and different cuts from what I offered. We could get quite detailed for the WB specifically with OECD data as well (see the screenshot of the table with the purpose codes) as they report their data in quite a detailed fashion. But yes, as you point out, it's either their cut of their data and thus we focus only on their data, or we cut it in a way that makes it comparable with the rest of our ODA work, and with how other MDBs and bilateral donors present their official support to developing countries (ODA+OOFs).

On @nupur-parikh's last point

With that being said, I think just one chart (the sankey diagram) on MDB financing from 2016-2021 is necessary. @jm-rivera I realize this time frame also poses issues as it relates to increasing number of MDBs, I'm not 100% familiar with the data so correct me if I'm wrong here but are the 4 MDBs that were added between 2016 and 2021 the ones with the most incomplete data? If so, I say we do not include them as you suggested due to insufficient data.

I can put together the list of MDBs that would have data consistently over that period and share with you. We can then decide if that is the sort of presentation that you would want to have, and whether we can say something like the largest 10, 12 or 15 MDBs

[nupur-parikh]

Thank you as always @jm-rivera for your extremely thoughtful and informative responses. I truly appreciate this!

On this point:

We have been scraping their database for the last few years as well. They provide more detail and different cuts from what I offered. We could get quite detailed for the WB specifically with OECD data as well (see the screenshot of the table with the purpose codes) as they report their data in quite a detailed fashion. But yes, as you point out, it's either their cut of their data and thus we focus only on their data, or we cut it in a way that makes it comparable with the rest of our ODA work, and with how other MDBs and bilateral donors present their official support to developing countries (ODA+OOFs).

This is a really interesting analysis that we could do. I am of the mindset that we do this in a future iteration or MDB specific analysis as I'm conscious that the focus of the health financing page is not solely on MDBs and I think determining what exactly we want to show will lengthen the process of getting the health financing topic page finalized. @TheebikaS we can discuss next week or sometime in April how we would like to cut the data and clearly define what we want to show and revisit this for the health use case for MDBs analysis that is listed in the content pipeline . As always, please correct me if you feel strongly otherwise.

On this point:

I can put together the list of MDBs that would have data consistently over that period and share with you. We can then decide if that is the sort of presentation that you would want to have, and whether we can say something like the largest 10, 12 or 15 MDBs

I like this approach, if not to time consuming to pull all that information, and think it makes sense to say largest 10 (or whatever number it may be) MDBs and how their regional health financing priorities change over time.

[jm-rivera]

Here are 18 of the MDBs ranked by the size of their disbursements during 2016-2021

rank	Bank
1	International Bank for Reconstruction and Development
2	International Development Association
3	Asian Development Bank
4	Inter-American Development Bank
5	European Investment Bank
6	European Bank for Reconstruction and Development
7	African Development Bank
8	Islamic Development Bank
9	Development Bank of Latin America
10	African Development Fund
11	Asian Infrastructure Investment Bank
12	New Development Bank
17	IDB Invest
13	Central American Bank for Economic Integration
14	Arab Fund
15	Council of Europe Development Bank
16	Caribbean Development Bank
18	Islamic Development Bank

Most of them have data for all years, except for

• Development Bank of Latin America ( only data for 2020, 2021)
• New Development Bank (only data for 2020, 2021)
• IDB Invest (only data for 2020,2021)
• Central American Bank for Economic Integration (only data for 2020, 2021)
• Islamic Development Bank (only data for 2019, 2020, 2021)

Only Arab Bank for Economic Development in Africa does not have enough data in the period for us to include (no data since 2015).

Given that we only show 1 year at a time with the Sankey, I would be fine with keeping this full list of 18.

Before we move forward - and sorry to be so annoying about it - but what are we hoping to show with this chart? What's the information we want to make available? What do we want users to clearly see with the chart? @nupur-parikh I just want to make sure that this is indeed the best way we have to do that.

[TheebikaS]

@jm-rivera thank you so much for explaining very clearly the scope and limitations of the database.

I don't want to hold up the health financing topic page with very specific MDB analysis, so happy to revisit at later stage to focus more specifically on health use case for MDBs @nupur-parikh.

Also, good to hear that we can get detailed for the WB specifically with OECD data @jm-rivera; think this is the data that Mattie pulled for me from the database?

To partly respond to your question @jm-rivera, the initial request to show specific health area flows to regions, was to understand whether there are any patterns, e.g., does MDB lending tend to flow to health infrastructure rather than disease specific programs? I was more interested in individual MDBs, like the WB or AfDB but didn't want another layer to complicate the chart. I wanted to then add a qualitative layer, by looking at these MDBs strategies that might be able to reinforce any trends, e.g. AfDB has a clear strategy that positions it as a health infrastructure financier (again this might not mean anything in terms of what any trend data shows, but thought worth a try).

This thread has been very helpful in clarifying some aspects of what we can and cannot do with the available database.

If possible, I would like to carve out some time at a later stage to understand this better. Happy to be specific about what we want users to see with charts and explain what I want to show, but sometimes it's the process of data mining that clarifies what we can and should be showing. This is a separate conversation, but wanted to flag here for reference.

[nupur-parikh]

Thanks @jm-rivera for doing that so quickly!

Given that we only show 1 year at a time with the Sankey, I would be fine with keeping this full list of 18. Showing only one sounds good to me.

To answer this question:

Before we move forward - and sorry to be so annoying about it - but what are we hoping to show with this chart? What's the information we want to make available? What do we want users to clearly see with the chart? @nupur-parikh I just want to make sure that this is indeed the best way we have to do that.

In addition to what @TheebikaS has explained, we also wanted to show the type of financing for health MDBs were giving and to where, trying to answer the question of do MDBs give more non-concessionary or concessionary loans to health, and where do these flows of financing go to the most. I think looking at a few years pre and during COVID would be helpful to see if this makeup changed as a result of the pandemic.

[jm-rivera]

Based on our discussions here and outside, here's the new version of a visualisation that we're considering: https://public.flourish.studio/visualisation/13148150/

@nupur-parikh, while we work on finalising this chart, here's the data: chart_4_1.csv