Николин блог

This post tries to assess how visitation of Wikipedias in various languages is dependent on Wikipedia size or quality.

Two days ago, I analysed Erik Zachte’s statistics about Wikipedia page views per country by replacing page views per person with page views per Internet user. In the meantime, Erik did the same in his original report, so in future it’s better to refer to it for that.

Now I am taking a look at another of Erik’s statistics, namely Wikipedia Page Views Per Country – Breakdown by also including in the data number of Wikipedias’ articles, Wikipedias’ depth, and Wikipedias’ score by sample of articles.

The results are at Wikipedia Page Views By Country – Breakdown with Wikipedia Size and Quality (HTML), but before you see them, you will want to know what they are and how are they interpreted. I will explain them on the example of the United States, with expanded column names:

Here,

• Wikipedia — is the name of the Wikipedia;
• Percentage of views — is the percentage of views from the original Erik’s data;
• Number of articles — is the number of articles on the Wikipedia (from the List of Wikipedias), a statistics I believe everyone is familiar with ;
• Score by number of articles — is the percentage of views divided by the number of articles (itself divided by 1,000), which gets a better explanation below;
• Depth — is the Depth from the List of Wikipedias, explained in detail on that page;
• Score by depth — is the percentage of views divided by the depth;
• Sample score — is the Score from the List of Wikipedias by sample of articles, explained in detail on that page; and,
• Score by sample score — confusingly named, is the percentage of views divided by the sample score.

What exactly is this score I am talking about? It’s a measure of popularity of Wikipedia by one of its parameters. For example, if in some country two Wikipedias would be read equally, but one of them would be twice smaller, it would have twice larger score (because readers will go to it despite the fact that they are half as likely to find what they need; and so, in a way, they like it twice as much). I hope this makes some sense to the reader; if not, I hope it will make sense by the end of the post.

Scores are a quick way for the readers to estimate „actual popularity“ a Wikipedia has regardless of its depth or breadth, as opposed to „measured popularity“ as measured by the number of visits (which is influenced by depth and breadth). Note that they shouldn’t be compared between countries.

Also note that scores by number of articles seem to roughly agree with scores by depth; but scores by the sample score seem not to. Not sure how to interpret this.

Now, in the example of the United States, scores by number of articles are 0.028 for English, 0.002 for Japanese and 0.002 for Spanish Wikipedia. What this means is that Japanese Wikipedia is as liked as Spanish, and is more visited only because it has more articles; if Spanish Wikipedia would be an exact translation of Japanese Wikipedia to Spanish language, it would be more visited than Japanese Wikipedia is now. However, Spanish Wikipedia has 14 times ( 0.028 / 0.002 ) smaller score than English Wikipedia; if it would be an exact translation of English Wikipedia to Spanish language, it still wouldn’t be read equally, but 14 times less (right now it is being read 81 times less).

My mind has just been blown by the discovery of the fact that this exactly matches the proportion of people in the United States who speak English only at home by people who speak Spanish only and ‘Speak English less than „very well“‘^[1] as the US Census Bureau would put it ( 227,365,509 / 16,156,584 = 14 ). In other words: in the US, the English Wikipedia is read exactly as much more than the Spanish Wikipedia as would be expected from the numbers of English-only and Spanish-only speakers in the US and the size difference between the two Wikipedias. It remains to be seen whether this will stay true for other cases.

In order to further explain what all these number means, I will use a better example, of Faroe Islands:

Assumption: if Danish Wikipedia would have the same number of articles as English Wikipedia, its readership would proportionally rise. Question: how much would it then be exactly?

The percentages are recalculated thus:

P = 3,159,745 / 122,013 = 26 (Danish Wikipedia would be read 26 times as much.)
X = P * 11.0 = 285 (We don’t know its actual readership but we assume it will rise 26 times in proportion, hence we have 285 „percents“.)
X_p = ( X + ( 100 – 11.0 ) ) / 100 = 3.7 (We add the remaining 89.0% and recalculate to 100% i.e. each „new percent“ has 3.7 „old percents“.)
E = E₀ / X_p = 19.5 (E₀ is old percentage of English Wikipedia etc.)
D = X / X_p = 76.1
F = F₀ / X_p = 0.7

I was not 100% sure of this math, but it was verified to be correct by Goran Obradovic. I would have felt much more comfortable if I had actual readership numbers instead of percents though The results are:

Note that the scores remained in proportion: 0.723 / 0.09 ≈ 0.18 / 0.024 and 0.09 / 0.024 ≈ 0.024 / 0.006

Meanwhile, if Faroese Wikipedia would have the same number of articles as Danish Wikipedia, the results would be:

And if it would have the same number of articles as English Wikipedia, The results would be:

Since percentages this high are usual in countries with large national Wikipedias, I pronounce the results correct

Finally: what if both Faroese and Danish Wikipedias would have the same number of articles as English Wikipedia?

It appears that the Statistics Faroe Islands does not have any data about languages spoken in Faroe Islands. If anyone could find that, it could serve to reaffirm (or not) these results.

By the way, you don’t have to compare Wikipedias only with each other. You can also ask the question like: if Faroese Wikipedia would be twice as large as it is now, how many readers it would have?

In case someone else would like to toy further with these data, here are the source code and source data I used to get it (as well as the previous post).