New tool for Analyzing Unintended Disagreements caused by Thinning Parents

Committing taxon swaps and changing taxon ancestries often removes children from the parent of the the taxon. This thins what we mean by that parent taxon which can cause existing IDs of that parent to unintentionally disagree with existing IDs of the carved-off child.

For example, the 2021 Clements update carves Ruby-crowned kinglet (Regulus calendula) off from the Kinglet genus Regulus as Corthylio calendula in its own genus Corthylio.

Curators might be tempted to make this change by simply creating a taxon swap with Regulus calendula as the input and Corthylio calendula as the output. However, carving Ruby-crowned kinglet off from the Kinglet genus will thin what we mean by Genus Regulus. It’s therefore necessary to analyze the effect of this narrowing and the unintended disagreements that the thinning may cause.

We’ve launched some new tools that make this analysis of potential unintended disagreements easier when making an ancestry change or committing a taxon swap. We’ve also introduced a rule of thumb for when we recommend splitting the parent taxon to deal with these unintended disagreements:

if thinning a parent results in more than 10 unintended disagreements you should split the parent after to replace existing IDs of the parent with IDs that don’t disagree.

Unintended disagreements resulting from a taxon swap
Let’s use Ruby-crowned kinglet as an example. After making a draft taxon swap with the active taxon Regulus calendula as the input and a new inactive taxon Corthylio calendula as the output, there is now a yellow notice on the taxon swap page alerting that this swap may result in unintended disagreements and a link to the tool to analyze these.

The unintended disagreements analysis takes a sample of up to 1,000 identifications of the thinned parent (or each thinned ancestor in some cases) so it may take up to a minute or so to run if the thinned parent has lots of identifications or if multiple ancestors are thinned.

Here’s the unintended disagreement analysis for this taxon swap. In this case, only the parent, Genus Regulus, would be thinned. This analysis sampled 43% of all 1300 current identifications of Genus Regulus. The analysis then looks at observations containing these IDs and looks for examples where there is also an ID of Regulus calendula. These are examples of unintended disagreements. Based on the frequency of these unintended disagreements in the sample, the analysis estimates that 1195 unintended disagreements will result from this taxon change and a link to several such examples are displayed.

Clicking on the ‘View’ link to example observations with unintended disagreements displays 230 observations

Clicking on the first one, we can see how the existing IDs would result in an unintended disagreement. In this case, there is an ID of Genus Regulus followed by 2 IDs of Regulus calendula. The ID of Genus Regulus was made when it was broadly understood to include Ruby-crowned Kinglet. Therefore, the two subsequent IDs of Regulus calendula don’t disagree with it and the observation taxon is Ruby-crowned Kinglet.

After the swap is committed, the two IDs of Regulus calendula will be made not-current and new IDs of Corthylio calendula will be added. These new IDs will be interpreted by iNaturalist as disagreements with the ID of Genus Regulus and the observation taxon will be rolled back to the community taxon of Family Regulidae. Because Genus Regulus was broader when identifier blackbird6 added their ID and they weren’t intending to disagree with the species Ruby-crowned Kinglet, this is an unintended disagreement.

When narrowing a parent results in many unintended disagreements (1,290 in this example) a lot of observations may be rolled back in such a way that the number of identifications required to roll them forward can represent a real burden on identifiers. Our rule of thumb is that 10 or more unintended disagreements are enough to warrant splitting the parent.

In this case, after swapping Regulus calendula into Corthylio calendula, construct a split with Genus Regulus as the input and Genus Regulus and Genus Corthylio as the outputs. This split will replace the 1,290 existing IDs of Genus Regulus with the common ancestor Family Regulidae, or if atlases are used potentially with IDs of the outputs. In either case the IDs of Genus Regulus causing the unintended disagreements will be replaced.

If there are 10 or fewer unintended disagreements we recommend that you don’t bother splitting the parent. However, you still may want to comment on any linked observations to alert the IDers that what we mean by the taxon they used has been thinned.

Keep in mind that the analysis estimates unintended disagreements based on a sample so the number of estimated unintended disagreements might be wrong if the parent has a large number of identifications. Also, the analysis only catches observations where both the thinned parent and carved off child occur as IDs which may not be the case. For example, imagine an observation with a single ID of Genus Regulus and the remark “this is either Ruby-crowned Kinglet or Golden-crowned Kinglet”. Narrowing Genus Regulus would still impact what was meant by the identifier but wouldn’t register as an unintended disagreement by the analysis. The unintended disagreement analysis tool is a guide but you should also use your judgment whenever you’re making changes to the taxonomy about the proper set of operations to make.

Unintended disagreements resulting from changing a taxon’s ancestry
In addition to committing taxon swaps, Parents and other ancestors can also be thinned by changing ancestry. For example, imagine that the field cricket Genus Velarifictorus needs to be reshuffled from Tribe Modicogryllini to Tribe Gryllini. A curator might do this by editing the Genus Velarifictorus and using the “Classification” box to change the parent to Tribe Gryllini.

The notice to analyze unintended disagreement will also appear here if the parent or other ancestors are thinned by the ancestry change. In these cases, unintended disagreements should be analyzed before saving the taxon.

The unintended disagreement analysis for ancestry changes behaves identically to the unintended disagreement analysis for taxon swaps. In this example, 172 IDs were analyzed and 16 unintended disagreements were detected.

As per our rule of thumb, we recommend splitting Tribe Modicogryllini to replace the 16 IDs resulting in the unintended disagreements (in this case with IDs of Subfamily Gryllinae).

We hope this tool will make it easier to assess when changing the ancestry or committing a taxon swap will have other impacts that need to be dealt to help guide taxon curation.

13 Likes

Is this a similar case? I am maverick because I am using the new species name - but not all Plectranthus are now moved to Coleus.

https://www.inaturalist.org/observations/43880599

This sounds great, but wouldn’t this replace all IDs of Regulus with Regulidae? Or am I misunderstanding taxon splits…

That is correct. This is why curators who commit splits should be prepared to reID observations if necessary.

1 Like

Atlases could be used to assigned Old World IDs to Regulus (sensu stricto), but New World IDs would be rolled back since the atlases would overlap there. But regardless it takes fewer IDs to roll forward coarser IDs than it takes to outweigh unintended disagreements

2 Likes

Yes, since that change thinned Plectranthus by reshuffling Coleus madagascariensis over to Coleus, our rule of thumb is that if there’s more than 10 unintended disagreements like the ones in the example you linked, Plectranthus should be split.

Its possible to retroactively split taxa, but we don’t yet have tool to ‘analyze unintended disagreements’ after the taxon has been carved off as in this case.

Also the new ‘analyze unintended disagreements’ tool described here only shows up on taxon swaps so it wouldn’t have displayed on this change which is a taxon merge

1 Like

Ideally would the higher level split be done first? I would think it wouldn’t really matter, but there might be some complications I’m not thinking of.

It shouldn’t really matter whether you swap the child then split the parent or split the parent then swap the child

1 Like

I suppose this might be a feature request, but it would be nice if this considered subspecies IDs as unintended disagreements. Here is an example where there were no identified unintended disagreements, but two IDs became “disagreements” because of the complications with subspecies IDs.
https://www.inaturalist.org/observations/112492862
https://www.inaturalist.org/observations/24616869