Category: Uncategorized

If I were to continue working up the list of initial members in reverse alphabetical order, I’d hit Syria next. But as the last two have been tricky, and I strongly suspect Syria will be too, I’m going to cherry pick a few more first that I hope will be simpler! So let’s try New Zealand first…

• Date of joining: 1945-10-24
• Expected count from Wikipedia (Peter Fraser onwards): 19
• Number in Wikidata: 18
• PositionHolderHistory warnings: 3
• Score: 61/72 = 85% 174/190 = 92%

We know Turkey is going to be a little complicated, as our spreadsheet already lists the Head of Government changing from the Prime Minister to the President in 2018. So we’re going to need to score each of those separately, and them combine them:

• Date of joining: 1945-10-24

Prime Minister of Turkey

• Expected count from Wikipedia (Şükrü Saracoğlu onwards): 37
• Number in Wikidata: 14
• PositionHolderHistory warnings: 15
• Score: 37/144 110/370

President of Turkey

• Relevant date: 2018–
• Expected count from Wikipedia: 1
• Number in Wikidata: 1
• PositionHolderHistory warnings: 0
• Score: 2/2 10/10

(This exposes another flaw in our scoring system: where the first holder is also the last holder, we’re subtracting two points at each end, rather than only two in total. I suspect a few other places are going to have this, so we’ll need to address it more generally, but for now it’s easiest to override the total whist pondering the wider implications, and what score somewhere should have in such a case.)

Total score: 39/146 (27%) 120/380 = 32%

As we’re at the bottom of the list of the countries who were UN Member States since inception, let’s work upwards from there. This gives us Ukraine next.

We have a few new issues here. Firstly, when it joined it was the Ukrainian SSR (I have to admit I’m a little surprised it was a distinct member pre-independence). This gives us a bit of an anachronism, in that Q133356 is listed as having been a member from 1944–1991, but Q212 is also listed as having been a member since October 1945 (with a “subject has role: Ukrainian Soviet Socialist Republic” qualifier).

Resolving that inconsistency/duplication is out of scope for now, but for simplicity I’m going to treat the two as having been technically separate memberships, and only look at the post-1991 Head of Government data (a brief check suggests that Wikidata doesn’t actually know about anyone having held the Chairman of the
Council of People’s Commissars or Chairman of the Council of Ministers positions anyway).

Another issue is that the Wikipedia table also lists quite a few Acting Prime Ministers, some for reasonably long periods (e.g. Yukhym Zvyahilsky for almost 9 months in 1993/4). My understanding is that the normal practice is to add the person as having held the position (ideally with a “nature of statement: acting” qualifier), so generally we’ll want to count these in our tally. It’s possible that in some countries these will have been entered in Wikidata using a specific “acting Prime Minister” item, or some other way that would complicate things more for us, but we’ll work out how to deal with those if/when we encounter them.

We also have some problematic entries in Wikidata where people who were Prime Minister multiple times have a single “position held” entry, rather than one for each term (see, for example, Vasyl Durdynets). This greatly confuses PositionHolderHistory, but I think the most accurate way to score these is to include overlapping rows there as being only a single entry for the “how many people” count, but count each warning across all of them for the “how many warnings” count.

So this gives us:

• Date of joining: 1991 (exact date doesn’t affect anything)
• Expected count from Wikipedia (all): 24
• Number in Wikidata: 25 (!)
• PositionHolderHistory warnings: 35
• Score: 61/92 (66%)

I’m a little concerned that Wikidata knows about more position holders than Wikipedia (which seems to be down to an extra Yanukovych period), which implies that the scoring system needs adjusted slightly to be able to handle this. But where Wikidata is more correct than Wikipedia, we shouldn’t be flagging that as problematic either. So I think we need to wait to see what other problems come up before deciding what to do.

LATER UPDATE: Updating this post in light of the revised scoring system, I also discovered that Ukrainian Wikipedia has a more comprehensive list than English Wikipedia, so we can also switch to that:

• Date of joining: 1991 (exact date doesn’t affect anything)
• Expected count from Wikipedia (all): 26
• Number in Wikidata: 26
• PositionHolderHistory warnings: 36
• Score: 188/260 = 72%

I expect the US will also be pretty good…

• Date of joining: 1945-10-24
• Expected count from Wikipedia (Harry S. Truman onwards): 13
• Number in Wikidata: 13
• PositionHolderHistory warnings: 0
• Score: 130/130 = 100%

Eventually I’ll probably resort to going down the spreadsheet in order, but initially I’m going to cherry-pick some countries to get a good workflow going. I already know the UK data is good for about the last 150 years, so let’s start there:

• Date of joining: 1945-10-24
• Expected count from Wikipedia (Clement Atlee onwards): 16
• Number in Wikidata: 16
• PositionHolderHistory warnings: 0
• Score: 160/160 = 100%

As it’s possible we’ll need to change the scoring system once we discover more, it’s important to keep as much raw information as possible as we go. I’m also not sure yet what form the final reporting will take (though it’s likely to be more than one form), so I think a spreadsheet will be the easiest to manipulate now, and then convert to different formats at the end. Seeding a Google Sheet with the current results of our more recent SPARQL gives: https://docs.google.com/spreadsheets/d/167Du-ZU3I5NMCoQ92io1SAkQJvar3h2gMgKtQVYrBOA/edit?usp=sharing

I’ll also post the calculations for each country here as I go, but this spreadsheet will be the main working document.

To accurately assess how complete the Wikidata information is, we need something to compare against. PositionHolderHistory can certainly tell us that there are obvious gaps, but it can’t know how many people actually held the office during them. To keep things simple we’re going to use English Wikipedia as our point of comparison.

This will not involve taking any of the information from Wikipedia and entering it into Wikidata, or even checking that Wikidata has the same dates etc for each person: this is solely about getting a count of how many people have been the Head of Government of each country within the relevant time period. Wikipedia tends to provide a fairly straightforward table of officeholders for positions like this. The format of that table will differ significantly from country to country, but counting the number of rows since that country joined the UN should hopefully be relatively straightforward in the majority of cases.

Wikidata can even help direct us to where that information lives in Wikipedia in most cases: usually it will either be on a “List of of “-type of page, or else on the page for the office itself. So let’s try to find the most likely candidate in our master SPARQL query:

https://gist.github.com/tmtmtmtm/98ab6ad3e108bb6f45a33a4a7a742c17

All but four of our 202 entries now also has an associated Wikipedia page. The four currently missing (Supreme Protector of the Peru–Bolivian Confederation, Prime Minister of Botswana, Governor of Liechtenstein, and President of the Council of Ministers of the Russian SFSR) can be investigated when we get to each of those countries. There are almost certainly other cases where the relevant table lives somewhere different, but hopefully those won’t be particularly hard to find, and we can always update the Wikidata pointers as we come across them.

One of the key goals of this project is to give each country some sort of score for how good the information in Wikidata currently is, which we can then check again later to see if it has improved. This is always going to have some degree of subjectivity in at least some cases, and whatever we choose is likely to be contentious, especially as we need something that’s relatively easy to calculate. It’s also likely that whatever we choose will fail to reflect some special cases somewhere (some country will always do things very differently to everywhere else), but as a first stab, we’ll want to know:

1. How many people have been Head of Government in the relevant time frame?
2. How many of those does Wikidata know about?
3. How complete and consistent are the P39 (position held) statements in each case (do they all have start/end dates and predecessor/successor information)?†

Other than the current officeholder (who won’t have a successor or end date), and first officeholder (where any predecessor, or even this person’s period in office before joining the UN, is out of scope for this project), each of the expected P39 statements can therefore score 4 points (2 points for the first and last).

So each country will have an expected/maximum score of (four times the expected number of officeholders) minus four (for the first/last person), and an actual score of (four times the number present in Wikidata) minus four (for the first/last) minus the number of warnings PositionHolderHistory gives, which we can than express as a completeness percentage.

We might need to adjust something later based on whatever we find when we start to look at each country in turn, but let’s see where this approach takes us.

—

† For now we’ll ignore all the other common qualifiers found on the Position Held statements — series ordinal, end cause, etc

Ideally all the information for each country would be complete, consistent, and correct. But finding a trustworthy source for each country to check against, and then comparing each entry in Wikidata against that is a lot of work. So we’re going to take a few shortcuts here, primarily by focussing on the first two elements: completeness and consistency. It’s certainly possible that Wikidata could be complete and consistent, but still wrong, but as due to the “double-entry” nature of each P39 ‘position held’ statement pointing at the previous holder and next holder, looking for obvious gaps or mismatches — e.g. a ‘followed by’ statement with a mismatched or missing ‘follows’, or someone having a ‘start date’ earlier than their predecessor’s ‘end date’ — this much simpler check is likely to be Good Enough for an initial version.

Thankfully we already have a tool that will show us everyone who has held a given office, and provide exactly that sort of consistency checking: PositionHolderHistory.

So, for example, Wikidata’s current knowledge on historic Presidents of Argentina.

So now we need to update our SPARQL to see each country’s Head of Government position(s). Where a country has changed what office the Head of Government holds, we’ll need to know each of them, but only back as far as the date they joined the UN:

https://gist.github.com/tmtmtmtm/0479786d46ed103ad47013e19cde9a71

Currently that gives 202 results for the 193 countries, which I suspect is lower than it should be (i.e. Wikidata doesn’t yet know about some of the historic offices). I don’t know of any source that would tell what that figure should be, so we’ll just have to wait and see what happens when we look at each country in turn.

Ideally Wikidata would have great information for the entire history of every country, but for the sake of making this investigation a little bit more manageable, we’re going to reduce that scope a bit. Given that we’re already restricting this to UN member states, the obvious approach is to check how good the information is back to the point at which they became members.

So let’s update our SPARQL to also include the date countries joined:

https://gist.github.com/tmtmtmtm/b2de6e919d623cffa4e38704cc738ea4

That, rather surprisingly, gives us 194 results, rather than 193, however.

After a bit of investigation it seems that Indonesia joined, then left, then joined again, and so gets returned twice. So we probably also want to update the query to remove any memberships with an end date:

https://gist.github.com/tmtmtmtm/71f54ed7fa1f3e9a49026b6613d805f6

This time we get the 193 results we expect.