User:Smallbones/vital articles

Starting page to remind me of promises at [[1]]

This discussion has been closed. Please do not modify it.

The following discussion has been closed. Please do not modify it.

Thanks Smallbones - in your ever widening research you may also want to take a look at enWP;

(I do not know if those Icons are up to date) and see what the improvement history of those are. You could also compare those across projects or go deeper into the 1000 (which admittedly have a bit of serendipity in selection). Also maybe write it up for the internal notice. Alanscottwalker (talk) 16:37, 23 February 2016 (UTC)

That's an excellent suggestion Alan, though I'd go further than Level 1.♦ Dr. Blofeld 17:49, 23 February 2016 (UTC)

@Alanscottwalker: I am really tempted to take on the 10 Level 1 articles (the 100 Level 2 articles - not so much!) I mean, how much work can analyzing 10 articles be? (famous last words). Let me throw it back to you. What would you like to know about those 10 articles? I can (easily, in theory) collect data, say for the December 31 version every year, on size in bytes, the ORES score (article class prediction), number of words of text, and with some more difficulty page views (there's a new system that works great for the last 3 months only), other things that might be very difficult - number of edits in the last year, number of editors in the last year, even readability index! What would you like me to do with that data? What questions or theories do you have about that data? Can you think of any other data that might reflect on your questions? Can you help collect some of the data?

I'm not trying to give you a hard time, just seeing what is needed to be useful and to satisfy your curiosity. Here's something that I can promise within 2 weeks: I'll get bytes, text size, ORES, and maybe even readability index for 12-31-2010 and 12-31-2015. I might even get enough data to make a nice graph (ORES?). OK, but what would really float your boat? Smallbones_(smalltalk) 20:27, 23 February 2016 (UTC)

"Improvement" is however, you defined it - I don't really have an addition, although it is possible with those articles that the wikipedia "review scale" (as supposedly they are important) was applied decently over-time - and if no one reviewes it regularly that might say something - (also, if you want, you could randomize a selection of the 100). I also don't have a theory, but I am aware of claims that the important articles are stuck in a limbo of disinterest and lack of knowhow about how to generally and effectively cover a broad concept/thing (perhaps, in light of a tendency elsewhere for "smaller subjects" to sometimes go-on-and-on). Alanscottwalker (talk) 20:58, 23 February 2016 (UTC)

I'm probably going to talk myself into a lot of work here, so let me say anything that I agree to could take 6 months. Most of those series above could be gathered year-by-year for 5 or even 10 years. The 10 level 1 articles could be compared to 10 "randomly selected" Level X articles and maybe 10 other articles that were started about the same time (pretty early in most cases). The random article link in the left-hand column doesn't have a "in the vital articles" function but something could be more-or-less worked out. "Level X" might be the biggest broadest set of vital articles in order to empahsize the difference from the level 1 articles. The other articles - how would we select them to be comparable? There might be enough older articles in the data set of 1000, but there probably aren't enough in the Science, technology, and math category to match up.

Finally, there is a likely selection bias that would work against any clear cut results. It seems likely that the articles improve at a certain rate, not entirely because they are vital articles, but rather they are selected as vital articles because somebody thought they were important enough to improve to a large extent. In any case, this would only be an exploratory analysis.

Why don't I do the simplest stuff in my first answer, post some very basic stuff in 2-3 weeks? In the meantime you and others consider what you really want to be done and consider how much data you are willing to gather. Yes, that last part is important. Smallbones_(smalltalk) 23:49, 23 February 2016 (UTC)

Initial plan:

Collect data data on the above 10 articles, and
10 random articles from vital articles level 4 (get enough to match years started, throw out the rest)
10 random from Wikipedia sample of 1000 (earliest 10 non-vital to match years (almost))
data includes Dec 31 for first 3 years and last 5 years mORES score
(Dec 31 1st month, and 2015, mORES bytes, word count - maybe later)
pageviews Jan. 2015

questions
- do the three classes have equal starting quality?
- do they improve at the same rate?

mORES class (if stub probability > .70) classify articles as sub-stubs. The final quality classes I use range from 1 to 7:

1 (sub-stub); 2 (stub); 3 (start); 4 (C); 5 (B); 6 (GA); and 7 (FA).

Data

#	Article	year	mORES 1st 12/31	2nd 12/31	3rd	2011	2012	2013	2014	2015	Notes to self
1	Earth	2001	5	5	5	7	7	7	7	7
2	Life	2001	3	3	5	7	7	7	7	7
3	Human	2001	3	5	5	7	7	7	7	7
4	History of the world	2004	5	5	5	7	7	7	7	7
5	Culture	2001	1	3	5	7	7	7	4	4
6	Language	2001	2	3	5	4	7	7	7	7
7	The arts	2004	1	3	5	4	4	4	4	4
8	Science	2001	3	3	5	7	7	7	7	7
9	Technology	2001	3	3	3	7	7	7	7	7
10	Mathematics	2001	5	5	5	7	7	7	7	7
20	Afonso I of Portugal	2001	3	3	3	4	4	4	4	4
21	Iraq	2001	1	2	5	6	6	6	7	6
22	Korean War	2001	3	3	5	7	7	7	7	7
23	Phonetics	2001	1	3	3	4	4	5	5	5
24	James VI and I	2001	1	3	5	7	7	7	7	7
25	Computer architecture	2001	3	3	3	5	5	5	4	4
26	Meteorite	2001	1	3	3	7	7	7	7	7
27	Wars of the Roses	2001	1	3	5	4	7	7	7	7
28	Rainforest	2002	2	2	3	4	4	4	4	4
29	Hundred Years' War	2002	3	3	5	4	7	7	7	7
30	Chain rule	2001	3	3	3	4	4	4	4	5
31a	Clark, New Jersey	2002	3	3	3	5	6	6	6	6	replace Cy Young Award was a list
32	Ninth Fort	2001	1	1	1	3	4	4	4	4
33	East Machias, Maine	2002	3	3	3	3	3	3	3	4
34	Economy of Jordan	2002	3	3	3	4	5	7	7	7
35	Transport in Swaziland	2002	1	1	1	3	2	2	3	3
36	Mary River National Park	2002	1	1	1	1	1	3	3	3
37	Zero divisor	2002	1	3	3	3	3	3	3	3
38	Rockwood, Tennessee	2002	3	3	3	4	4	4	6	6
39	Morganville, New Jersey	2002	3	3	3	3	4	4	4	4

Discussion

How do Wikipedia articles improve over time? My interest in that question led me to do the exploratory investigation shown at User:Smallbones/1000 random results. Alanscottwalker asked whether the most important articles on Wikipedia, as listed at WP:Vital articles, might develop differently over time. This question was seconded by Dr Blofeld.

Their concern is that the most important articles may be too large or complex to develop in the usual manner that other articles develop. Perhaps a different method or a special effort should be used for these articles? I address the following questions:

Do vital articles start their lives as higher or lower quality than other articles?
Do these articles improve in quality at a faster or slower rate than other articles?
Do they currently have higher or lower quality?

The measure of quality used is a modified form of the ORES article quality rating system. ORES was programed by machine learning to predict how editors would rate articles, based on easily observed article attributes, such as text length, and internal and external links. It's strengths are its consistency and its ability to quickly update or guide a new quality rating. To my knowledge I'm the first person to use ORES to rate earlier versions of articles and this exercise might be useful in seeing whether it is also useful in this area. I also used the random in category function, which I had not used before, to randomly select articles in the Level 4 Vital articles category.

The ten Level 1 Vital articles are arguably the most important articles on Wikipedia. These articles form one of three groups of article that were examined. They are remarkable in that 8 of them were started in 2001, Wikipedia's first year. The two others History of the world and The arts were first edited in 2004.

Ten Level 4 Vital articles were selected by the "Random in category" function, out of 9,812 articles in the category. Actually only 8,762 articles (89%) were contained in the Category:All Wikipedia level-4 vital articles. Remarkably 8 of these articles were also started in 2001, with 2 started in 2002.

The third group of articles examined were randomly selected earlier as part of the User:Smallbones/1000 random dataset. To match the other two groups, the earliest 10 articles in that dataset were selected. Remarkably, two of the selected articles were also vital articles - Karl Marx, and Polymerase chain reaction. These articles were removed and replaced with the next earliest articles. The article Cy Young Award was also removed and replaced when it was discovered that it is classified as a list rather than as an article. This group consists of 2 articles started in 2001 and 8 started in 2002.

It is truly remarkable that these vital articles were almost exclusively started during the encyclopedia's first 2 years, and that articles from the encyclopedia's first 2 years tend to be classed as vital. This investigation may be limited by this focus on early articles.

A modified ORES score was calculated for each of the 30 articles for the December 31 version in each of the following years: the first year of the article's life, the second year, and the third year, as well as for the last 5 years, 2011-2015. The scores were then averaged for each group and are presented graphically below.

Group	Avg. 1st year	Avg 3rd year	Avg. annual improvement (yr. 1-3)	Avg. 2011	Avg. annual improvement (middle period)	Avg. 2015	Avg. annual improvement (2011-15)
Vital-L1	3.1	4.8	0.85	6.4	0.21	6.4	0
Vital-L4	1.9	4.0	1.05	5.2	0.15	5.8	0.15
Non-vital	2.2	2.4	0.10	3.3	0.12	4.5	0.30

Conclusions

While the 10 "most vital" articles in Wikipedia did start their lives at a fairly high quality level, the "Level 4 vital articles", a tiny subset of all Wikipedia articles, appear to have started their lives at about the same quality level - or at a slightly lower level - than a randomly selected sample of non-vital articles.

Both samples of the vital articles did increase in quality very quickly in their first three years of life, roughly increasing by one quality level each year. The randomly selected non-vital articles did not increase as quickly, averaging an annual increase of only about 0.10 level until 2011.

Once the vital articles reached the level of 5 or 6 (approximately "B" or "GA" class) the quality increase naturally slowed - the highest level possible is only 7. However, once the non-vital articles reached level 3 on average (roughly "Start" class), their annual quality increase almost tripled to 0.30 levels per year.

More than 50% of all Wikipedia articles are classed as "Stubs", so it is clear that the vital articles examined are well above average quality. Their early starts 2001-2004 are partly responsible for this quality - they have had a long time to mature and get better. But the non-vital article of a similar age did not increase in quality as quickly. Indeed they grew very slowly until on average they reached level 3. The early starts of all the articles examined here limit our ability to generalize about other Wikipedia articles, which typically are much younger.

The concerns about the quality of vital articles should not be dismissed because of the findings shown here. These are all exceptionally important topics - as shown by their exceptionally early start dates. The tools used here show that compared to typical articles they are high quality. But it should be recognized that, because of their exceptional importance, vital articles should all be of exceptional quality.