User:Smallbones/Quality4by4

How does article quality change over time? Do articles in different topic categories increase in quality at different rates? Has the rate of quality increase changed over time? Are new articles getting better each year? This investigation proposes to examine these questions by following groups of random articles in 4 topic categories with 5 articles in each category for each of 5 years.

More specifically, using the random article data set from User:Smallbones/1000 random, I look at 4 categories of articles: with 5 articles chosen in each category in each year (using the first randomly selected in each year-category) for articles started in
 * Biography
 * Geography
 * Culture & Arts
 * Other (any other category)
 * 2008, 2009, 2010, 2011
 * recording the modified ORES score as a measure of quality (sub-stubs or level 1 assigned where p of being a stub > .70 according to the ORES output, level 2 for an ORES prediction of "stub", level 3 for an ORES prediction of "start", etc. )
 * comparing the entire sample of 80 series aligned as 1st year, 2nd year, 3rd year and 4th year will show how quality, on average, changes for the first 4 years of an article's life.

Dividing this data into the 4 topic category groups should show whether topics categories affect how quality changed.

Realigning the data into the 20 articles started in 2008, the 20 started in 2009, etc. should show where the rate of quality change has itself changed over time, for articles of the same age. In particular, we'll be able to see whether new created articles are increasing in quality each year.

Results
Overall article quality, as measured by the mORES scores, goes up on average 0.05 points each year from the end of an article's year of creation to the end of its fifth year. For each of the four article categories examined, over four years the average mORES score increases. The mORES score increases in eight of the possible 16 year*category combinations and decreases in only 2. While the scores increase fairly consistently each year, the size of the increase is small. A 0.05 annual increase translates to the increase by a single point (e.g. from stub to start) over 20 years.

Geographic articles have a much lower starting score on average, but the other three categories start at about the same level. The total average increase in scores is highest for the category Culture & Arts (0.35) and lowest for Biographies (0.10), with the other categories increasing 0.15.



When examined by the calendar year, articles for each calendar year on average increase in score over four years. In only 1 of the 16 possible calendar year*starting year combinations does an average score decrease, and the average score increases 9 times.

The scores for each year's newly created articles are much larger for the later 2 years, but the change is not uniformly positive. The average scores for newly created articles in 2011 (2.35) is 0.40 higher than the average score for newly created articles in 2008 (1.95), or an annual average increase of 0.13. While this tendency of the score to increase for newly created articles is not as consistent as for annual increases in previously created articles, it is more than twice the average size (0.13 to 0.05).

Looking at mORES score increases across articles sorted by page views shows that mORES score increase for all page view quartiles over 4 years. Average scores increase for 13 of the possible 16 quartile*year combinations, and decrease in only 2. The higher the page views, on average, the higher to scores increase. The lowest page view quartile average score increase by only 0.08 over four years, the 2nd quartile by 0.27, the 3rd by 0.32, and the highest by 0.20.

Summary
Quality as measured by mORES for previously created Wikipedia articles increases slowly, on average increasing only at a rate of 1 point (e.g. from stub to start) per 20 years. The tendency to increase is consistent however and holds for
 * articles in different categories
 * articles started in different years, and
 * articles in all page view quartiles.

The tendency for the quality of newly created articles to be better than the quality of the previous year's newly created articles is less consistent, but more than twice as large as the tendency for previously created articles to increase.