User:Doncram/Disambiguation-ChangeTWODABSpolicy

Disambiguation-ChangeTwoDabspolicy Quick Study

Purpose
To get some data to inform ongoing RFC at Wikipedia talk:Disambiguation Hypotheses of the researcher (Doncram):
 * H1: The deletion of a dab generally get reversed after other items added, IMO proving that the Delete decision was wrong and also that the nomination to delete at AFD was wrong.
 * H2: Everywhere else the dab should be kept, too. I.e., where a dab was deleted and not restored, and where a dab was kept, when I review what happened in the AFD I expect will still think all of the disambiguation pages should be kept, for other reasons (to keep edit history, because it is likely items will be added, etc.... see other reasons given in the RFC).

Sample and procedure
Ideal sample: a random sample of AFDs about disambiguation pages where wp:TWODABS was used to argue for deletion.

Sample achieved: A convenience sample which is a subset of AFDs about disambiguation pages whose names start with letter "D", out of AFD's about disambiguation pages with "(disambiguation)" in their title (i.e. usually where a primary topic has been determined). I didn't see how to find AFDs about disambiguation pages without "(disambiguation)" in their title.

Procedure:
 * 1. Using "subpages" tool, search on all subpages of "Articles for deletion/D": obtain
 * 2. Within those results, manually search for every occurrence of "(disambiguation)".
 * 3. Record the ones that are about TWODABS below. Discard the relatively few AFDs not about TWODABS (or segregate further below in case anyone wishes to review them).
 * 4. Count different types of results and summarize here.

Findings and summary of results
Results: Out of 28 cases found
 * 8 cases where Kept, where items had been added during AFD and that was convincing for at least some editors
 * 5 other cases where Kept
 * 8 cases where it was Deleted, but it was later recreated because in fact it was a valid dab topic.
 * 5 cases where Deleted, but the decision appears wrong. There were valid Keep arguments and/or I see other possible items that could be added so that all would agree they should be kept.
 * 2 cases where it was Deleted, where I see some merit in deleting. These turn out to be cases of "ZERODAB", i.e. no valid dab items.

Findings / Observations / Notes:
 * There were long-time participants consistently on the Delete side
 * There were longish-time participants consistently on Keep side (perhaps more editors in total, but not as long each?)
 * There may not be any long-time participants in the middle...it seems polarized, I think. I may be wrong about this.
 * No dabs where I (Doncram) participated happened to be included.
 * The sample turns out to have missed some types of AFDs that I know about:
 * No cases like Memorial Hall (disambiguation) (cited at Wikipedia talk:Disambiguation where it was incredibly totally obvious that more entries than just the Harvard one will be valid. (BTW, there are now 33 items plus a "See also" for that. And there was in fact no Articles for deletion/Memorial Hall or Articles for deletion/Memorial Hall (disambiguation).)
 * No cases like "Smith House" where common sense and sometimes local knowledge says that obviously "Jane Smith House" is sometimes called "Smith House", but such Jane Smith House items are disputed. (This has happened a number of times.)
 * No cases where an administrator was obviously miss-using administrative tools to delete linked articles in order "to win" deletion. (I have seen this.)
 * No cases where the nomination obviously lacked good faith effort to state reasons for deletion. (In the past, there were AFDs with nominator providing no statement.  These would be SpeedyKeep now, by some rule.)
 * No cases where a promotional/COI interest in pushing an item into Wikipedia was apparent.
 * At least one time there was an item deleted from the dab during the AFD, which would "win" the deletion. This is confusing for discussion, similar to wholesale deletions of material in other articles at AFD, by detractors of the topic.  Then the stripped-down article looks inadequate, and it makes it hard for all to consider the merit of the disputed material.  (IMO, as for other AFDs, guidance in AFD process should be given: add tags or otherwise call attention to items you dispute, but do not delete items off the dab during the AFD itself.  Perhaps one or more suitable tags for use during AFDs of dabs are needed.)
 * The TWODABS has been interpreted differently over time.
 * In AFDs in 2012, it was pointed out that hatnotes suffice, but having a two-item dab is harmless, and has some merit. (I recall that argument being made in multiple AFDs)
 * More recently, apparently, TWODABS has been interpreted by some as allowing Speedy Delete on basic of A6? housekeeping, although there's suggestion at the RFC that PRODS are allowed but not Speedy, now.
 * The phrase "harmless" was explicitly included in wp:TWODABS (verify? from when to when), but was removed (when).
 * There is at least one case where a newish contributor was confounded and frustrated by the bureaucracy, and the dab they created was deleted and they were effectively prevented from creating articles on topics they were trying to work on (What happened with the editor, did they quit?)
 * There is notable frustration expressed by Keep arguers, e.g. calls for a moratorium on AFDs of useful dab pages.
 * AFD #30 shows a problem is that hatnotes can be deleted at the primary-usage page, and with no two-item dab page, the other info can simply be lost. The hatnote deleted went to something like a DABMENTION, I think, it was not an exact match article.
 * I think the sample shows an evolution towards acceptance that redlink items are valid. MOS:DABRL is old, I believe, and applied during the entire period. Newer editors tend not to believe that redlinks are valid;  with fewer new editors these days and more experience in AFD commenters, the redlinks are better regarded.
 * I think DABMENTIONS (are these what I termed blacklinks?) have achieved greater acceptance as counting.
 * "See also" usages were present in some of the dabs but I think did not see arguments that they provided reason to keep a dab. OTOH in the RFC discussion there seems to be consensus that at least inbound "See also" links count as valid items.  And I think there may be consensus and/or it is a corollary that outbound "See also" links should count, too.

Conclusions

 * H1 is borne out. Many (8 out of 13) deletions have already been reversed (one by me during this exercise).  I have so far found grounds that everyone would agree on to reverse at least 2 of the remaining 5.  I and some others see usefulness in the 2-item dab pages, whether or not additional items are immediately found to make everyone agree they have merit.  I personally think it is not right for my attention and/or the attention of other "curationist"-minded editors to have to fight to keep dabs, when the fight could be avoided by changing the interpretation of TWODABS back to "keeping is okay, if the dab has been created" (and strengthening that)
 * H2 is also borne out. I thought that I was finding I had to agree with some deletions, but on further consideration, these were not TWODAB cases, they were "ZERODAB" cases.  I disagree with all of the deletions where TWODABS was argued.
 * There were 2 cases where all entries were satisfactorily disputed and deleted (call this "ZERODABS"). If there is really not a single usage of any type, I have to concede that the disambiguation page should be deleted.
 * Although there were no cases of commercial/COI promotion in the sample, an "always keep" policy would lend itself to being gamed by promoter-types, who could get company names or brands or product names into Wikipedia by adding blacklink or redlink items, using actual names or nicknames of companies or their products. So I do want to allow rejection of some items on promotional/COI grounds, which could mean some DAB pages get brought down to one or zero items.
 * I realize I do want to extend the proposal about the following:
 * If there is just one usage, then redirect (do not delete). It may be labelled as a "Redirect with possibilities" if it seems likely a 2nd usage will be added eventually. If that doesn't seem likely, then simple Redirect.  A 2nd usage may nonetheless come up, and it saves edit history and history of disputes and references and so on, etc.
 * If there truly are zero usages, and it goes to AFD and no one finds any, then okay, delete.

Caveats: It is likely that I made one or more errors in categorizing or other, as people make errors, as there always are errors in any big analysis, and as categorizing is subjective. I also did this fairly quickly and it is incomplete right now, in terms of how I want to write it up. However, I swear this is an unbiased sample, in that I deliberately chose a sampling process where there was no a priori reason to expect a slant, and I didn't change results or selectively drop results. I also tried to be honest about what my hypotheses were beforehand. (A study is more valid if the hypotheses are stated in advance, and then sample is collected and tested, than if the hypotheses are formed partly from browsing the results.) The categories created here were not defined in advance, however, they are formed to describe the gathered data. And I acknowledge that the tone of my writing reflects my (informed) bias towards keeping all 2-item dab pages. With more time editing, I could make it read more objectively. However others have the data now and can assess how they would decide in the 32 cases (28 two-dabs ones, 4 unrelated). Note: In the sampling process early on I did not record some AFDs which seemed obviously unrelated. Should I go back and get them?

Further research hypotheses: Perhaps there is longer continuity of editors on the "Delete" side of these, and more intense frustration and turnover of the editors on the side of Keep. The total number of editors on "Keep" side may be larger. The collective view, the consensus, may in fact be more on the side of "Keep" for all of these, if all were considering each example. (But, as the frustrated Keepers will note, it is a waste of more editors time for all to be involved, again and again. Waste of time may not to be as much of a concern for those who vote "Delete" (i.e. they value it less, so are more inclined to vote "Delete".)

Classification of sample
AFDs in the sample are classified as to which of the following "best fit":
 * OldInterp:  Kept, with best outcome, applying the old interpretation of TWODAB items: Two-item dab kept, without editors being forced to make up stuff on a false, short deadline.  (5 examples / 28 = 18% for OldInterp + OtherKeep combined)
 * OtherKeep:  Kept for other reasons besides OldInterp and AddsSave.
 * AddsSave:   Kept, but AFD forced immediate dispute and refocus of editors to create articles and/or items to defend the dab. (8 examples / 28 = 29%)
 * ProvenWrong: Deleted, but that decision has been proven wrong since by identification of more than 2 items, and it has been recreated. However, the edit history has not been restored in any one of these, so Wikipedia's promise of attribution to contributors is not met.  Effort and goodwill was wasted, too.  (8 examples / 28 = 29%)
 * DelBut:     Deleted, but despite evidence of other likely items and other value of keeping dab.  Assuming growth generally, all would have eventually gotten to 3 or more items IMO.  Existing as 2 items would have been fine too IMO.  Edit history lost, work and discussion lost, and it remains work to recreate all of these and battle it all out again and again.  And/or it remains bad to continue to frustrate new and experienced editors trying in good faith to create decent navigation and help readers and make progress. (5 examples / 28 = 18%)
 * AllGone:    Deleted, and decision seems okay IMO.  All items deleted or never had articles.  (2 examples / 28 = 7%%)
 * Unrelated:  AFD did not involve argument to delete based on TWODABS.  (4 examples)
 * Total = 5+8+8+5+2+4= 32
 * Total where TWODABS-type argument made = 28
 * subtotal Kept or restored: 21/28 = 75%, where IMO all should agree that the AFD was costly and a waste of time.
 * subtotal Deleted, not yet restored = 5/28 = 18%, where some will agree AFD was costly and more work was created, is yet to be done.
 * subtotal Deleted validly = 2/28 = 7% where the community focus was productive. Really these were not TWODAB or even ONEDAB cases, they were ZERODAB cases, though.


 * OtherWrong: Extra code where something else went wrong IMO.  See the individual AFDs.

Sample: Table of AFDs
Note: In the table below "#" is merely a row number from 1-32, just reflecting original entry order. The table is sortable. The current presentation order is by Type category (OldInterp or OtherKeep; AddsSave; ProvenWrong; DelBut; AllGone). (Please comment and ask questions on sample and methods at Talk page.)

Unrelated
The following seem unrelated. Note also there were a few more disambiguation pages early alphabetically in the AFDs starting with "D", which I judged were unrelated, and I did not record them.